Method and Substance in Macrocomparative Analysis (Research Methods Series)

Method and Substancein Macrocomparative

Analysis

Lane Kenworthy and Alexander Hicks

Edited by

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page i

Method and Substance in Macrocomparative Analysis

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page ii

Research Methods Series

General Editors: Bernhard Kittel, Professor of Social Science Methodology,Department of Social Sciences, Carl von Ossietzky Universität Oldenburg, Ger-many and Benoît Rihoux, Professor of Political Science, Université catholique deLouvain (UCL), Belgium.

In association with the European Consortium for Political Research (ECPR),Palgrave Macmillan is delighted to announce the launch of a new book seriesdedicated to producing cutting-edge titles in Research Methods. While politicalscience currently tends to import methods developed in neighbouring disciplines,the series contributes to developing a methodological apparatus focusing onthose methods which are appropriate in dealing with the specific research prob-lems of the discipline.

The series provides students and scholars with state-of-the-art scholarship onmethodology, methods and techniques. It comprises innovative and intellectu-ally rigorous monographs and edited collections which bridge schools of thoughtand cross the boundaries of conventional approaches. The series covers bothempirical-analytical and interpretive approaches, micro and macro studies, andquantitative and qualitative methods.

Titles include:

Audie Klotz and Deepa Prakash (editors)QUALITATIVE METHODS IN INTERNATIONAL RELATIONSA Pluralist Guide

Lane Kenworthy and Alexander Hicks (editors)METHOD AND SUBSTANCE IN MACROCOMPARATIVE ANALYSIS

Research Methods Series

Series Standing Order ISBN 978–0230–20679–3–hardcoverSeries Standing Order ISBN 978–0230–20680–9–paperback(outside North America only)

You can receive future titles in this series as they are published by placing a stand-ing order. Please contact your bookseller or, in case of difficulty, write to us at theaddress below with your name and address, the title of the series and one of theISBNs quoted above.

Customer Services Department, Macmillan Distribution Ltd, Houndmills,Basingstoke, Hampshire RG21 6XS, England

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page iii

Method and Substancein MacrocomparativeAnalysis

Edited by

Lane KenworthyProfessor of Sociology and Political ScienceUniversity of Arizona, USA

and

Alexander HicksProfessor of SociologyEmory University, USA

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page iv

Editorial matter, selection, and introduction © Lane Kenworthy andAlexander Hicks 2008 All remaining chapters © respective authors 2008

All rights reserved. No reproduction, copy or transmission of thispublication may be made without written permission.

No paragraph of this publication may be reproduced, copied or transmittedsave with written permission or in accordance with the provisions of theCopyright, Designs and Patents Act 1988, or under the terms of any licencepermitting limited copying issued by the Copyright Licensing Agency, 90Tottenham Court Road, LondonW1T 4LP.

Any person who does any unauthorized act in relation to this publicationmay be liable to criminal prosecution and civil claims for damages.

The authors have asserted their rights to be identifiedas the authors of this work in accordance with the Copyright, Designsand Patents Act 1988.

First published 2008 byPALGRAVE MACMILLANHoundmills, Basingstoke, Hampshire RG21 6XS and175 Fifth Avenue, NewYork, N.Y. 10010Companies and representatives throughout the world

PALGRAVE MACMILLAN is the global academic imprint of the PalgraveMacmillan division of St. Martin’s Press, LLC and of Palgrave Macmillan Ltd.Macmillan® is a registered trademark in the United States, United Kingdomand other countries. Palgrave is a registered trademark in the EuropeanUnion and other countries.

ISBN-13: 978 0 230 20257 3 hardbackISBN-10: 0 230 20257 8 hardback

This book is printed on paper suitable for recycling and made from fullymanaged and sustained forest sources. Logging, pulping and manufacturingprocesses are expected to conform to the environmental regulations of thecountry of origin.

A catalogue record for this book is available from the British Library.

Library of Congress Cataloging-in-Publication DataMethod and substance in macrocomparative analysis / edited by LaneKenworthy and Alexander Hicks.

p. cm. (Research methods series)Includes index.ISBN 0 230 20257 8 (alk. paper)

1. Social sciences Comparative methods. 2. Employment (Economictheory) I. Kenworthy, Lane. II. Hicks, Alexander M.H61.M49155 2008300.72 dc22 2008016151

10 9 8 7 6 5 4 3 2 117 16 15 14 13 12 11 10 09 08

Printed and bound in Great Britain byCPI Antony Rowe, Chippenham and Eastbourne

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page v

To the memory of Michael Wallerstein – friend, colleague,exemplary macrocomparativist

This page intentionally left blank

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page vii

Contents

List of Tables ixList of Figures xiiNotes on Contributors xiv

1 Introduction 1Lane Kenworthy and Alexander Hicks

2 Statistical Narratives and the Properties of Macro-LevelVariables: Labor Market Institutions and EmploymentPerformance in Macrocomparative Research 29Bernhard Kittel

3 Comparative Employment Performance: AFuzzy-Set Analysis 67Jessica Epstein, Daniel Duerr, Lane Kenworthy, andCharles Ragin

4 Do Family Policies Shape Women’s Employment?A Comparative Historical Analysis of France andthe Netherlands 91Joya Misra and Lucian Jude

5 The Welfare State, Family Policies, and Women’s LaborForce Participation: Combining Fuzzy-Set andStatistical Methods to Assess Causal Relations andEstimate Causal Effects 135Scott R. Eliason, Robin Stryker, and Eric Tranby

6 Family Policies and Women’s Employment:A Regression Analysis 196Alexander Hicks and Lane Kenworthy

7 Part-Time Work and the Legacy of BreadwinnerWelfare States: A Panel Study of Women’sEmployment Patterns in Germany, theUnited Kingdom, and the Netherlands, 1992–2002 221Jelle Visser and Mara Yerkes

vii

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page viii

viii Contents

8 Comparative Regime Analysis: Early Exit from Workin Europe, Japan, and the USA 260Bernhard Ebbinghaus

9 Identifying the Causal Effect of Political Regimeson Employment 290Adam Przeworski

Index 315

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page ix

List of Tables

1.1 Analytical strengths of the three methodological approaches 71.2 Summary of the contributions 192.1 Cross-sectional and time variance components of main

variables 402.2 Autoregression 402.3 Nonstationarity: Maddala and Wu’s Fisher test 412.4 Service employment and replacement rate: panel models 452.5 Employment in private sector consumer services,

cross-sectional analysis model 1: cross-section model, 1981 492.6 Controlling for other factors 532.7 Explaining variation in long-term change of private service

employment, 1981–91 552.8 Pooled model, first differences 582.9 Scores for replacement rate 592.10 Reanalysis using new data for replacement rate 603.1 Truth table from analysis of all six causal conditions 753.2 Examples of solution sets 783.3 Five causal pathways 823A.1 Fuzzy-set scores 874.1 Values regarding work and family, 1990 and 1999 1175.1 Decade means and standard deviations for cumulative

left cabinet incumbency 1485.2 Decade means and standard deviations for percentage

civilian government employment 1495.3 Decade means and standard deviations for maternity

leave index 1505.4 Decade means and standard deviations for public day care

index, children ages 0–2 1515.5 Decade means and standard deviations for public day care

index, children ages 3–School age 1525.6 Decade means and standard deviations for weeks of

extended leave 1535.7 Decade means and standard deviations for Cash/Tax

Family/Child Benefits Index 1545.8 Decade means and standard deviations for female labor

force participation rates 154

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page x

x List of Tables

5.9 Select five-way and four-way partition tests of goodness-of-fit F statistics for the outcome “High Female Labor ForceParticipation – Subsequent Year” 157

5.10 Goodness-of-fit for relationships between hypothesizedcausal condition “High Level of Cumulative LeftCabinet Incumbency” and select outcomes 160

5.11 Goodness-of-fit for relationships between outcome“High Female Labor Force Participation – Subsequent Year”and select hypothesized causal conditions 162

5.12 Bootstrapped EDF estimates of compliers average causaleffects on female labor force participation rates, with strongleft political tradition as the instrument in the intention-to-treat analysis 163

5.13 Bootstrapped EDF estimates of compliers average causaleffects on female labor force participation rates, with allother non-left political traditions as the instrument in theintention-to-treat analysis 165

6.1 Correlations between family policy measures 2046.2 Principal components analysis of the four family policy

measures 2046.3 Regression results 2066.4 Regression results: change in women’s employment on

level of family policy generosity 2177.1 Employment rates by presence of children, 2000 2227.2 Distribution of women aged 15–64 years by labor

market status in 1992–2002, by birth cohort 2367.3A Determinants of labor market status of women – inactive

(Ref: full-time) 2387.3B Determinants of labor market status of women – short

part-time, 1–19 hrs (Ref: full-time) 2397.3C Determinants of labor market status of women – long

part-time, 20–34 hrs (Ref: full-time) 2407.4A Determinants of labor market transitions into inactivity 2437.4B Determinants of labor market transitions to job from

inactive status 2457.5 Preferred and actual working hours of women with children

under six, living in a couple, 1998 2477.6 Preferences for more or less working hours and chances to

realize preferences, 1992–2002 2487.7 Determinants of women’s preferences for more or less

working hours 251

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page xi

List of Tables xi

8.1 Conceptual map of protection, production, andpartnership regimes 268

8.2 Relative exit rates, men and women aged 60–64, 1970–2003 2748.3 Index of pathways ranked by exit opportunities 2798.4 Early exit from work and regime configurations 2839.1 Different estimates of the average treatment effect 3089.2 Estimates of the effect of the treatment on the treated and on

the control group 308

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page xii

List of Figures

1.1 Employment levels and employment change, 1979to 2005 3

1.2 Women’s and men’s employment levels, 2005 51.3 Illustration of causal sufficiency and necessity 101.4 Illustration of “nearly always sufficient” 111.5 Illustration of small-N ordinal comparison 151.6 Mahoney’s argument for use of small-N analysis to

eliminate a hypothesized sufficient or necessarycondition 16

1.7 Summary of analysis in Schettkat (2005) 172.1 Main variables 392.2 Replacement rate and service sector employment 432.3 Annual changes in service employment 472.4 Long-term changes, 1982–91 482.5 Employment regulation and private service employment 502.6 Private service employment: replacement rate effect

conditional on employment regulations 512.7 Change in private service employment, 1981–91:

replacement rate effect conditional on employmentregulations 56

2.8 Change in private service employment, 1981–91:replacement rate effect conditional on employmentregulations, new data for replacement rate 61

3.1 Employment change in low-end private sector services,1979 to 1995 70

3.2 Employment change fuzzy-set scores by employmentchange raw values 71

3.3 Causal condition fuzzy-set scores by raw values 733.4 Consistency and coverage 793.5 Poor employment change performance by causal

configurations 1–4 from Table 3.3 833.6 Poor employment change performance by causal

configuration 5 from Table 3.3 844.1 Employment rate for women as a proportion of all

women, 15–64 974.2 Employment rate for women between 25 and 54, 1968–2004 98

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page xiii

List of Figures xiii

4.3 Full-time and part-time employment rates for women agedbetween 25 and 54, 1983–2004 99

4.4 Involuntary part-time work as a proportion of part-timework for women 25–54, 1992–2004 100

4.5 Percentage of respondents who believe that work is“very important” in their life 118

4.6 Proposed model explaining women’s employment 1205.1 Hypothesized causal chain, including variables used in the

empirical analysis 1556.1 Men’s and women’s employment, 2000–2005 1986.2 Women’s employment rates, 1960–2000 1996.3 Women’s employment by family policies, decade averages,

1960s–1990s 2006.4 Women’s employment by women’s preference for

employment and women’s education 2016.5 Women’s educational attainment by family policies,

Nordic countries, 1960s–1990s 2036.6 Public child care, age 0–2, 1960s–1990s 2106.7 Public child care, age 3–5, 1960s–1990s 2116.8 Maternity leave, 1960s–1990s 2126.9 Public employment, 1960s–1990s 2136.10 Family policy factor scores, 1960s–1990s 2146.11 Women’s employment by family policies and by

women’s education: over-time within-country patterns 2167.1 Incidence of part-time employment among women 2267.2 Employment population ratios and unemployment rates,

women, aged 15–64 years 2308.1 Employment rates for men aged 55–59/60–64/15–64,

1970–2000 2698.2 Employment rates for women aged 55–59/60–64/

15–64, 1970–2000 2718.3 Relative exit rates for men aged 60–64, 1970–2003 2738.4 Relative exit rates for women aged 60–64, 1970–2003 2759.1 Average rate of growth of employment in the world,

1950–1990 2929.2 Average difference between productivity and wage growths 2939.3 Labor share as a function of product per worker, by regime 2949.4 Growth of employment as a function of per capita

income, by regime 3029.5 Density of per capita income, by regime 3069.6 Density of average world employment growth, by regime 307

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page xiv

Notes on Contributors

Daniel Duerr is a PhD candidate in sociology at the University of Ari-zona. His research focuses on the development of the welfare state,stratification and income equality, and the causes of poverty in affluentcountries.

Bernhard Ebbinghaus is professor of sociology and director, MannheimCentre for European Social Research (MZES), University of Mannheim,Germany. His recent publications include Reforming Early Retirement inEurope, Japan and the USA (2006). His research interests are welfarestate analysis, industrial relations, institutional theory, and comparativemethods.

Scott Eliason is associate professor of sociology, and faculty affiliate ofthe Minnesota Population Center, at the University of Minnesota. He isalso a faculty fellow at the Center for the Study of Poverty and Inequal-ity at Stanford University. His research interests and publications spanthe areas of quantitative methodology and statistics, sociology of labormarkets, stratification, law and organizations, the welfare state, and thelife course.

Jessica Epstein is a PhD candidate in sociology at the University ofArizona. Her research interests are the political economy of food andagriculture, political ecology, and research methods.

Alexander Hicks is Winship Distinguished Research Professor of Sociol-ogy at Emory University. His publications include Social Democracy andWelfare Capitalism (1999) and articles in leading sociology and politicalscience journals. He has twice served as editorial board member for theAmerican Sociological Review and was inaugural co-editor of Socio-EconomicReview.

Lucian Jude is a former doctoral student in sociology at the Universityof Massachusetts – Amherst.

Lane Kenworthy is professor of sociology and political science at theUniversity of Arizona. He studies the causes and consequences of poverty,inequality, mobility, employment, economic growth, and social policy

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page xv

Notes on Contributors xv

in affluent countries. He is author of In Search of National Economic Success(1995), Egalitarian Capitalism (2004), and Jobs with Equality (2008). Hiscurrent book project is tentatively titled Rethinking Inequality and Poverty.

Bernhard Kittel is full professor of social science methodology atthe University of Oldenburg, Germany. His research interests includemethodology of cross-national research, in particular of the welfare stateand industrial relations, and the experimental study of political deci-sion making. Recent publications include “A Crazy Methodology? Onthe Limits of Macro-quantitative Social Science Research”, InternationalSociology (2006); “European Rigidity vs. American Flexibility? The Insti-tutional Adaptability of Collective Bargaining,” Work and Occupations(2005, with Bernhard Ebbinghaus).

Joya Misra is associate professor of sociology and public policy at theUniversity of Massachusetts, Amherst. Her research focuses on compar-ative welfare states and the intersection of race/ethnicity, nationality,gender, and class in labor markets and social policy. She has publishedarticles in journals such as American Sociological Review, American Journalof Sociology, Socio-Economic Review, and Social Problems.

Adam Przeworski is the Carroll and Milton Petrie Professor of Poli-tics at New York University. His books include The Logic of ComparativeSocial Inquiry (1970), Capitalism and Social Democracy (1985), Paper Stones(1986), Democracy and the Market (1991), Democracy and Development(2000), and States and Markets (2003). Recent relevant publications oncomparative work–family policy and women’s employment include arti-cles in the Journal of Comparative Policy Analysis (2007) and Gender &Society (2007).

Charles C. Ragin is professor of sociology and political science atthe University of Arizona. In 2000/01 he was a Fellow at the Centerfor Advanced Study in the Behavioral Sciences at Stanford University,and before that he was professor of sociology and political science atNorthwestern University. His main interests are methodology, politi-cal sociology, and comparative-historical research, with a special focuson such topics as the welfare state, ethnic political mobilization, andinternational political economy. His books include Redesigning SocialInquiry: Fuzzy Sets and Beyond (2008), Fuzzy-Set Social Science (2000),Constructing Social Research: The Unity and Diversity of Method (1994),What is a Case? Exploring the Foundations of Social Research (with Howard

9780230_202573_01_prexvi.tex 7/6/2008 16: 32 Page xvi

xvi Notes on Contributors

S. Becker, 1992), Issues and Alternatives in Comparative Social Research(1991), and The Comparative Method: Moving Beyond Qualitative andQuantitative Strategies (1987).

Robin Stryker is professor of sociology and affiliated professor of law atthe University of Minnesota. Among her recent publications are “HalfEmpty, Half Full, or Neither?: Law, Inequality and Social Change inCapitalist Democracies,” Annual Review of Law & Social Science (2007);“A Sociological Approach to Law and the Economy” (with Lauren Edel-man), in the Handbook of Economic Sociology (2005); “The Strength ofa Weak Agency: Title VII of the 1964 Civil Rights Act and the Expan-sion of State Capacity, 1965–1971” (with Nicholas Pedriana), AmericanJournal of Sociology (2004); and “Redefining Compassion to Reform Wel-fare: How Supporters of 1990s US Federal Welfare Reform Aimed for theMoral High Ground” (with Pamela Wald), Social Politics (2008). Her cur-rent research on the politics of social science in government regulationof equal employment opportunity is supported by the National ScienceFoundation.

Eric Tranby is a PhD candidate in sociology at the University of Min-nesota. His research interests include gender and racial inequality incontemporary social life, comparative public policy, and life courseresearch. His dissertation research examines the effect of family policieson women’s employment outcomes in the United States, Germany, andSweden. Ongoing projects include research on gendered labor markets,public policy, and female labor force participation; the diverse experi-ences of young adulthood; and understandings of racial and religiousdiversity in the United States. His work has been published in journalssuch as Social Problems and Research in Social Stratification and Mobility.

Jelle Visser is professor of sociology at the University of Amsterdam,where he directs the Amsterdam Institute of Advanced Labour Studies(AIAS). His main work is on labor relations, welfare states, social policies,employment, and trade unions. With Bernhard Ebbinghaus he edited theTrade Unions in the Societies of Europe series (published by Palgrave Macmil-lan) and with Anton Hemerijck he wrote “A Dutch Miracle” (1997).

Mara Yerkes is a post-doctoral research fellow in sociology at the ErasmusUniversity, Rotterdam. She is the author of What Women Want: IndividualPreferences, Heterogeneous Patterns? Her research interests include indus-trial relations, labor policy, work and care issues, and comparative welfarestate research.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 1

1IntroductionLane Kenworthy and Alexander Hicks

Macrocomparative researchers use a variety of methodologicalapproaches. This book features analyses of a single substantive topicusing several of the most common. The topic is comparative employ-ment performance in affluent countries. The chief methodologicalapproaches are pooled cross-section time-series regression, qualitativecomparative analysis (QCA), and small-N analysis.

The aim of the volume is to illustrate in a practical fashion the advan-tages and drawbacks of these analytical strategies. Instruction and adviceis available in numerous monographs, articles, and edited volumes (forexample, Greene, 2003; Ragin, 1987, 2000; King, Keohane, and Verba,1994; Mahoney and Rueschemeyer, 2003; Brady and Collier, 2004;George and Bennett, 2004). But often that advice is provided at a generallevel. Commonly, substantive illustrations are offered with reference to asingle methodological approach. A key question for researchers is whenand why to use one methodological approach rather than, or in addi-tion to, another, and what are the payoffs and sacrifices entailed by aparticular choice. Although general advice is helpful, the best way tounderstand the tradeoffs involved is via practical application.

This book was conceived partly as a follow-up to the 1994 vol-ume The Comparative Political Economy of the Welfare State, edited byThomas Janoski and Alexander Hicks. That book was aimed at com-parativists interested in the welfare state and in comparative politicaleconomy more generally. It included methodological and substantivechapters covering time-series regression, pooled cross-section time-seriesregression, event history analysis, and qualitative comparative analysis(“Boolean analysis”). This volume differs in three main respects. First,it covers a different set of methodological approaches, focusing exclu-sively on those that involve macro-comparison – that is, comparison

1

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 2

2 Method and Substance in Macrocomparative Analysis

across countries (or regions). The techniques explored in this book arepooled regression, qualitative comparative analysis, and small-N anal-ysis. Secondly, this book includes no strictly methodological chapters;the methodological techniques are described and discussed in substan-tive chapters. Thirdly, the substantive chapters in the Janoski–Hicks bookexamined disparate issues: economic growth, wage trends, active labormarket policy, and pension systems. The analyses here for the most partaddress the same substantive question: What are the determinants ofvariation in employment performance across affluent countries? Ourhope is that the common substantive focus helps to reveal as clearly aspossible the advantages and drawbacks of the methodological strategies.

The substantive issue: comparative employmentperformance

Macrocomparativists engage a wide array of substantive issues. We chosecomparative employment performance as the outcome for the analysesin this book. A country’s employment rate is measured as the numberof people with paying jobs divided by the population age 15 to 64 (theworking-age population).

Employment is a useful barometer of labor market performance in acountry. For most of the past half-century, unemployment rates havebeen considered the main indicator of labor market outcomes, but thefact that unemployment can be hidden in various ways – low labormarket participation, active labor market programs, and so on – hasencouraged a shift toward employment rates.

Employment has intrinsic merit (Jahoda, 1982; Wilson, 1996; Phelps,1997). With heightened geographical mobility, later marriage, andincreased divorce, neighborhood and family ties have dissipated some-what. As a result, work is an increasingly important site of socialinteraction. Employment imposes regularity and discipline on people’slives. It can be a source of mental stimulation. It helps to fulfill thewidespread desire to contribute to, and be integrated with, the larger soci-ety. For many individuals, work is inextricably bound up with identityand self-esteem.

In addition, an increasingly common view is that high employmentis critical to maintenance of low or moderate levels of income inequal-ity (Esping-Andersen, 1999; Ferrera et al., 2000; Scharpf and Schmidt,2000; Esping-Andersen et al., 2002; Kok et al., 2003; Kenworthy, 2004,2008; OECD, 2005, 2006). Meeting pension and health care commit-ments for an ageing population will require greater government funds

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 3

Lane Kenworthy and Alexander Hicks 3

0 20 40 60 80

Employment (%)

ItBelSpFr

GerIreFinAusAslUS

SweCanNthUKJaNZNorDenSwi

Employment levels, 1979 and 2005

19792005

10 20�10 0

Employment change, 2005 minus 1979 (%)

SweFinFr

BelGerDen

ItNorUKUSJa

AusAsl

CanSpIre

SwiNZNth

Employment change, 1979 to 2005

Figure 1.1 Employment levels and employment change, 1979 to 2005Note: Employment = employed persons as a share of the population age 15 to 64. Portugal isomitted due to lack of employment data for 1979.

in coming decades. Yet governments increasingly find it difficult to raisetax rates, due to capital mobility. This makes it difficult to maintaingenerous transfers for the working-age population and their children. Arising employment rate helps to increase tax revenues without raisingtax rates. And by bringing former benefit recipients into the workforce,it reduces government transfer payments.

Figure 1.1 shows employment rates in 1979 and 2005, and changesin employment during that period, for the group of affluent countriesexamined in this book: Australia, Austria, Belgium, Canada, Denmark,Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, NewZealand, Norway, Portugal, Spain, Sweden, Switzerland, the UnitedKingdom, and the United States. As of the mid-2000s, employmentrates ranged from less than 60 percent of the working-age populationin Italy to more than 80 percent in Switzerland. The variation in changebetween the late 1970s and the mid-2000s was equally large, with theemployment rate falling by more than 5 percentage points in Swedenand increasing by 20 points in the Netherlands.

How can countries achieve a high and/or rising employment rate?There are two principal debates around this issue. The first concernsthe determinants of overall employment performance and focuses on

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 4


the impact of labor market institutions and policies. These include wagelevels at the low end of the distribution, employment protection regula-tions, government benefits, and taxation. If wage levels at the low end ofthe labor market are high relative to productivity levels – due to a statu-tory minimum wage, a collectively bargained minimum, or a tight labormarket – employer demand for low-end workers may diminish, reduc-ing the employment rate. If it is difficult for employers to fire workerswhen the economy is bad or the firm’s sales are slumping – because theymust get the approval of a works council or provide a generous severancepackage or pay for extensive retraining and job placement – employersmay hire fewer workers when times are good. If government benefits –social assistance, unemployment insurance, sickness or disability com-pensation, pensions, and so on – are fairly generous, easy to access, andof lengthy duration, workers at the low end of the job market may beless likely to search for and accept employment. A high tax burden canreduce the net benefit to a worker from employment and/or increase thecost to an employer of hiring, thereby potentially producing less supplyof and demand for labor.

High low-end wages, strict employment protection regulations, gen-erous government benefits, and high taxes are sometimes referred to as“labor market rigidities.” The notion that such rigidities impede highand/or rising employment has been around for a long time, but it hasbeen especially prominent since publication of The OECD Jobs Studyin 1994. The Jobs Study was a clear and systematic statement of therigidities ➔ poor employment performance hypothesis, and it was pub-lished at a time when unemployment in a number of western Europeancountries had been high for roughly a decade and showed no signs ofimminent decline. Since 1994 dozens of comparative empirical stud-ies have examined the hypothesis. (Some recent studies, which includecitations to earlier ones, include Blau and Kahn, 2002; Kenworthy 2004,2008; Howell, 2005; Nickell, Nunziata, and Ochel, 2005; Bassanini andDuval, 2006; OECD, 2006; Baccaro and Rei, 2007.) Despite this extensiveresearch, there is nothing close to a consensus regarding the merit of thehypothesis.

The second debate is about the impact of so-called family policies (alsovariously referred to as work–family reconciliation policies and women-friendly policies) on female employment. In countries with employmentdeficits, the problem consists chiefly of a shortage of women’s employ-ment. This can be seen clearly in Figure 1.2, which shows employmentrates for men and women as of 2005. A critical task – perhaps the crit-ical task – for low-employment countries, therefore, is to identify and

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 5


0 20 40 60 80 100

Employment (%)

ItSpBelIreFr

GerJa

AusAslPorUKNthNZFinUS

CanDenSweNorSwi

WomenMen

Figure 1.2 Women’s and men’s employment levels, 2005Note: Employment = female or male employed persons as a share of the female or malepopulation age 15 to 64.

implement institutional or policy changes that can substantially increasefemale employment. A number of studies have suggested that the key isgenerous family policies (Winegarden and Bracy, 1995; Ruhm, 1998;Meyers, Gornick, and Ross, 1999; Plantenga and Hansen, 1999; Rubery,Smith, and Fagan, 1999; Sainsbury, 1999; Daly, 2000; Korpi, 2000;Dingeldey, 2002; OECD, 2001; Stier, Lewin-Epstein, and Braun, 2001;Esping-Andersen et al., 2002, ch. 3; Orloff, 2002; Pettit and Hook,2002; Ferrarini, 2003; Gornick and Meyers, 2003; Jaumotte, 2003;Morgan and Zippel, 2003; Mandel and Semyonov, 2006; Kenworthy,2008).

One such policy is public provision or financing of child care. Lackof affordable child care can pose a significant obstacle to employmentfor women with preschool-age children. A second is paid parental/careleave. The expectation is that if women know they can take a reasonablylong break from work without losing their job and without foregoing allof their earnings, more will choose to enter the labor market in the firstplace and more will return after having a child. A third is government

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 6


provision of public sector jobs, which may be more attractive to womenthan private sector jobs because they are more likely to be available atreduced hours (part-time), to be secure (governments are less likely thanprivate employers to fire employees during economic downturns), andto accommodate family needs such as illness. A fourth is promotion ofor support for part-time employment, which women may prefer becausethe shorter hours facilitate work–family balance. A fifth is the structureof the tax system. Of particular relevance is the degree to which a cou-ple with two earners is penalized relative to a couple with one earner;the greater the tax penalty, the stronger the disincentive for a womanwith an employed husband to get a job. A sixth is anti-discriminationand affirmative action laws. To the extent that women’s employment isimpeded by discriminatory action by employers, such policies are likelyto help.

The book’s chapters focus on these two substantive questions: Whathas been the impact of labor market institutions and policies on overallemployment performance? What has been the effect of family policieson women’s employment?

Methodological approaches

In attempting to answer a question such as what determines employ-ment performance, various analytical strategies can be pursued. One isto examine individual behavior. Another is to consider patterns acrossfirms or industries within countries. A third is to look at developmentsover time in a single country. A fourth approach is macrocomparative.Countries are the unit of analysis. The causal factors of interest (poli-cies and institutions) and the outcomes are measured at the level of thenation. Analytical leverage is gained at least in part, and often primarily,by comparison across countries.

Macrocomparative analysis can be conducted using a variety of tech-niques. In this book the focus is on three: regression, qualitativecomparative analysis, and small-N analysis. In this introductory chapterwe offer a brief outline of the main distinguishing features, advantages,and disadvantages of these three techniques. They are summarized inTable 1.1.

A key point we wish to stress at the outset is that these approaches arebest viewed as complementary rather than competing and overlappingrather than mutually exclusive (Lieberman, 2005; Ragin, 2005). Each iscapable of contributing to macro-level analysis in different ways.1 Over-laps between the three approaches are possible. For example, QCA is not

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 7


Table 1.1 Analytical strengths of the three methodological approaches

Pooled RegressionDesigned to assess tendential relationships, manifested in patterns of

co-variation among variablesUseful for assessing the net effect of a variable on the outcomeEnables variation across units (countries) and over time to be analyzed

togetherAllows formal estimation of the magnitude of impact of a cause

Qualitative Comparative AnalysisDesigned to assess deterministic causal relationships (logically conceived):

sufficiency and necessityUseful for exploring causal configurations (combinations of causes)Useful for examining multiple causal paths to the same outcomeAllows formal estimation of the magnitude of impact of a cause

Small-N AnalysisUseful for assessment of causal mechanisms via process tracingUseful for elimination of hypothesized (“always”) sufficient or

necessary conditionsOrdinal cross-country comparison can be used to assess hypothesized

tendential or quasi-deterministic relationships, but generalizationbeyond the studied cases is problematic

Possibility of considering variables that cannot be included in a large-Nanalysis because data are available for only a few countries

Possibility of better measurement of variables due to case knowledgePossibility of more nuanced attention to interaction among causal factors

than is possible with regression

uncommonly used for small Ns of 10 or fewer, and an analysis that treatsa few cross-sectionally differentiated units in the “small-N” style mightinclude a time-series regression of years encompassed by a cross-sectionalslice or QCA analysis of subunits (e.g., of states).

Pooled regression

Regression is the most commonly used analytical technique in macro-comparative analysis. It is a correlational technique, although causalinterpretation of regression slope estimates backed by special statisticalcare (for example, fastidiously specified lag structures) as well as the-oretical argument sometimes accompany regression analyses. The aimis to identify statistical associations between hypothesized causes andoutcomes. Such associations are based on co-variation. In regression, anindependent variable is associated with a dependent variable when levelsof the two variables correspond to one another.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 8


The relationships assessed in regression are “tendential” rather than“deterministic.” A particular level of a hypothesized cause is thought tomake it more likely that the outcome will have a certain level, but it isassumed that there may be exceptions. Regression predictions are madewith allowance for error (for example, residuals, confidence intervals).

For analysts interested in understanding differences across the richcountries, a fundamental analytical obstacle is the small number of cases.Depending upon one’s definition of “affluent,” there are approximately15 to 25 nations to study. This inhibits estimation of regressions withmore than a few independent variables. Another limitation of standardregression in macrocomparative analysis is that it frequently is confinedto cross-country variation, ignoring variation over time within countries.Time-series regression does the opposite.

A pooled cross-section time-series regression combines informationabout variation across countries with information about variation overtime within countries. The unit of analysis is the country-year or country-period, rather than either the country or the time period. This not onlycombines the two kinds of variation; it also substantially increases thenumber of observations, thereby helping to address the small-N prob-lem. For these reasons, pooled regression has, as Janoski and Hicks(1994) envisioned, become the dominant technique for large-N analysisin macrocomparative research.

Yet pooled regression has important limitations, to which practition-ers do not always pay sufficient heed. One is that the determinantsof cross-country variation may not correspond to those of over-timevariation within countries (Griffin et al., 1986; Kittel, 1999; Kenwor-thy, 2006, 2007; Shalev, 2007). Over a very long period, we wouldexpect such correspondence, but most analysts do not have lengthytime-series data that are comparable across more than a few countries.One way to partially address this is to pool periods of years ratherthan individual years (Hicks and Kenworthy, 1998, 2003, this volume;Barro, 2000).

A second limitation concerns time lags in causal effects. Manyhypotheses about determinants of change in macrocomparative ana-lysis either implicitly or explicitly refer to relatively long-term effects.Yet most pooled regression analyses use the country-year as the unit.This is likely a function of researchers’ desire to significantly boost thenumber of observations and thereby facilitate estimation of models witha large number of regressors. Sometimes analyses with annual data canpick up medium-term or long-term effects, but that hinges on getting thelag structure correct. More often than not, using annual data to examine

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 9


hypothesized medium-run or long-run associations will obscure ratherthan clarify.

Finally, successful estimation of pooled regressions requires meeting,or addressing, various technical requirements. Bernhard Kittel discussesthese issues extensively in chapter 2, so we will do no more than men-tion them here (see also Kittel, 1999; Ebbinghaus, 2005; Wilson andButler, 2007). They include independence of observations, non-trending(stationarity), and assessment of the cross-sectional or longitudinal dom-inance of the data array (the proportions of the variable variancesaccounted for by the longitudinal and cross-sectional dimensions).

Qualitative comparative analysis (QCA)

Qualitative comparative analysis is a technique for systematically explor-ing relations between explanatory factors and outcomes (Ragin, 1987,2000; Ragin and Rihoux, 2004). QCA was conceived originally with theaim of formalizing the analytical process often pursued by small-N quali-tative researchers, to enable the process to be applied more systematicallyand to a larger number of cases. This process is articulated in terms ofthe logical language of set theory.

There are two variants of QCA. One, crisp-set QCA (Boolean), usesdichotomous codings of causal conditions and outcomes. The other,fuzzy-set QCA, uses pseudo-continuous codings that vary between zeroand one. In both versions, the aim is to identify hypothesized causalfactors, or combinations of those factors, that are related to the outcomein a pattern consistent with that of a sufficient or necessary condition –the two main types of deterministic causal relationship. When plotted ina scattergram, a relationship of sufficiency is suggested if the data pointsall fall above and to the left of a 45-degree line running from the lower-left corner to the upper-right corner. When the hypothesized cause isabsent or low, the outcome may be absent/low or present/high; but whenthe cause is present/high, the outcome is always present/high. A relation-ship of necessity is indicated by the data points being located below andto the right of the 45-degree line. When the cause is present/high, theoutcome may be absent/low or present/high. But the outcome is neverpresent/high unless the cause is present/high. These two patterns areillustrated in Figure 1.3.

“Cause” is the preferred term for sufficient and necessary “conditions”in the QCA literature. Although some have cautioned against identifica-tion of the logical conditions of formal languages such as QCA’s Booleanlogic with real world causes (Ayer, 1956, pp. 170–5; Passmore, 1967,pp. 355–60; Manicas, 2006), we adopt the QCA usage here.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 10


Low

Out

com

eH

igh

High

Low

Out

com

e

Low High

Hig

h

Cause

Sufficiency

Low

Cause

Necessity

Figure 1.3 Illustration of causal sufficiency and necessityNote: The lines in the charts are 45-degree lines, not regression lines.

Charles Ragin and others have suggested that a deterministic relation-ship may have empirical exceptions, due to the randomness of socialprocesses, measurement error, or other reasons. A cause need not bealways sufficient or necessary; it may be “nearly always” or “usually”sufficient or necessary (Ragin, 1987, ch. 7; Ragin, 2000, pp. 107–16; seealso Goertz and Starr, 2003). For some, the idea of introducing a ten-dential element into an otherwise deterministic notion of causality isoxymoronic. Why not instead simply refer to the relationship as ten-dential? Consider the pattern in Figure 1.4. Except for one data point, itis consistent with a hypothesis of sufficiency: in cases where the causeis present/high, the outcome is also present/high. One could concep-tualize the relationship as tendential; the two variables correlate at .54.But “almost always sufficient” seems a more accurate description of theempirical pattern. Which interpretation is more sensible depends heavilyon what the substantive issue is and whether a tendential or deterministicunderstanding is more compelling on theoretical grounds.

For macrocomparative analysis, QCA has several potential advantagesrelative to regression. One is its focus on deterministic relationships.As noted earlier, regression is designed to assess tendential causal (or,more simply, explanatory) relationships. When a cause is hypothesizedto be sufficient or necessary for an outcome, QCA may thus be a moreappropriate method.

Secondly, QCA is adept at exploring causal configurations – situationsin which variables have an impact only in combination with a high or

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 11


Low

Out

com

e

Low High

Hig

h

Cause

Sufficiency

Figure 1.4 Illustration of “nearly always sufficient”Note: The line in the chart is a 45-degree line.

low degree of one or more other factors. In regression analysis, causalconfigurations are assessed via interaction terms. But with the small ormoderate number of cases that is common to macrocomparative ana-lysis of affluent countries, the number of interactions terms that can beincluded in a regression model tends to be limited. Pooling the cross-country and over-time variation can alleviate this problem, but thecollinearity produced by interaction terms that involve more than twoor three variables and the difficulty in interpreting the results makesmodeling complex interactions problematic. Moreover, while assessinginteractions in regression requires that variables demonstrate productterms, QCA treats any aspects of cases that appear together systematically– in any quantity – as potentially interdependent.

Thirdly, QCA facilitates identification of multiple pathways to an out-come. Many social phenomena have causes that are relevant to onlya fraction of the cases. With a correlational technique such as regres-sion, when the dependent variable is high/present but the independentvariable is low/absent, this weakens the estimate of the effect of the inde-pendent variable (or interacted combination of independent variables).QCA, by contrast, is designed to reveal patterns of association that dif-fer across subsets of cases. It thereby enables discovery of more complexcausal patterns than are generally recognizable via regression.

In the view of some, an additional advantage of QCA is that the deter-ministic relationships it identifies are more likely to be causal than the

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 12


tendential associations identified by regression (Mahoney and Goertz,2006). This view, however, has been contested and criticized – bylogical positivists such as Ayer (1956, pp. 170–5) as well as by realists likeManicas (2006) and interpretivists like Wittgenstein (Passmore, 1967,pp. 355–60) – for a conflation of logical and causal relations (but seePruss, 2006).

Like any technique, qualitative comparative analysis has limitations.First, if a causal factor is suspected to have a tendential relationship withthe outcomes, rather than a deterministic one, QCA is of little use.

Secondly, a QCA analysis is, in a key respect, bivariate. Multiple causalfactors are considered, and the “solution set” of causal configurationsyielded by a QCA analysis will vary with the set of causal factors enteredinto the analysis. Specification of additional explanatory conditions, likeentry of new control variables in regression, will often modify analyticalresults. Yet QCA examines the relationship between the outcome andeach single hypothesized cause or combination of causes without con-trolling for – that is, without “holding constant” – any other causes. AsAaron Katz, Matthias vom Hau, and James Mahoney (2005, p. 568; seealso Seawright, 2005) point out:

Fuzzy-set analysis is only a multivariate method in the sense that thetechnique can explore if combinations of variables represent sufficientcauses. However, since each combination is reduced to a single value,each combination is, in effect, treated as a single cause.

Jason Seawright (2002, p. 181) has argued that given the assumption thatthe causal relationship is a deterministic one, this is appropriate:

. . . claims of necessary and/or sufficient causation are fundamentallybivariate in nature. The hypothesis entailed in this idea of causation isthat no other variable or combination of variables can overcome theeffects of the necessary and/or sufficient cause. Therefore, control-ling for other variables cannot alter the conclusion of the bivariateanalysis, and a bivariate focus is fully appropriate.

However, positing that a causal relationship is deterministic and find-ing a bivariate pattern consistent with that hypothesis does not rule outthe possibility of spuriousness (omitted variable bias). Although regres-sion is scarcely immune to this concern, it is designed to estimate the“net” effect of each variable. Charles Ragin (2005, p. 35) sums up thispoint effectively: “Regression analysis is a preeminent tool for estimating

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 13


net effects; QCA’s primary analytic focus is on the different ways causalconditions combine.”

Thirdly, to this point there is no QCA counterpart to the combiningof cross-country and over-time variation that is possible with pooledregression.

Regression has several tools for assessing the magnitude of a vari-able’s causal effect. One is the variable’s coefficient, which estimates thechange in the dependent variable given a one-unit increase in an inde-pendent variable (net of other independent variables). Another is the R2

(coefficient of determination), which measures the precision or “good-ness of fit” of the coefficient estimates. Others include multiple partialcoefficients of determination for subsets of regressors, standardized coef-ficients, and various techniques of exogeneity and Granger causalityassessment. In QCA the principal tools for assessing the strength ofrelationships are “consistency” and “coverage.” Consistency refers tothe degree to which the empirical pattern corresponds to that of suffi-ciency or necessity. An “always” sufficient relationship can be consideredstronger than a “nearly always” or “usually” sufficient one. Coveragerefers to the share of cases having a particular outcome that feature aparticular causal factor or causal configuration. Consistency and cover-age are discussed in greater detail in chapter 3 of this volume and inRagin (2006).

Small-N analysis

By “small-N analysis” we refer to macrocomparative analyses in whichthe number of countries (cases) studied is ten or fewer. This is, of course,an arbitrary cutoff; there is no number of nations that objectively demar-cates “small” from “large.” The most common number of nations studiedin small-N analyses is one, two, or three.2

One of the most important contributions of small-N analysis is descrip-tive. Studying a small number of countries allows the researcher tolearn, and convey to the reader, a level of detailed knowledge that isbeyond the reach of an analyst committed to comparing a large num-ber of countries. The small-N researcher typically examines variables,events, actors, and other aspects of the national context in extensivedetail. This type of information is inherently interesting to compara-tivists. It also helps large-N researchers to check their coding of variables,to consider additional causal factors, to think about interactions amongcausal conditions, and to judge the general plausibility of their causalhypotheses.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 14


What can small-N analyses contribute in terms of theory testing(hypothesis testing)? One contribution is process tracing (Mahoney, 2000,2003; Hall, 2003; George and Bennett, 2004). Process tracing consists ofexamining theoretically-specified causal pathways (causal mechanisms)in the context of developments in a single country. This is particularlyuseful where a large-N analysis has suggested support for a particularcausal story but where, often due to lack of data, the large-N analysis isunable to examine the causal mechanisms. Of course, any finding of anassociation or “solution set” in a large-N analysis should be supportedby a plausible theoretical story. But investigating the causal paths empir-ically is equally critical, and small-N analysis is a useful way to do that.When the small-N analysis is of a single country, it is often referred toas a “case study.” Small-N analyses of multiple countries sometimes areactually multiple case studies, rather than cross-country comparisons.

A good example of a case study of employment performance is JelleVisser and Anton Hemerijck’s (1997) book-length analysis of develop-ments in the Netherlands, which was revisited in an article by Visser(2002). These two works consider a number of possible determinants ofthe Dutch “employment miracle” since the early 1980s (see Figure 1.1above). Visser and Hemerijck carefully trace developments in publicpolicy, economic institutions, and employment patterns in the 1980sand 1990s. They conclude that wage restraint and increased femaleeducational attainment played critical roles, but also that a variety ofconjunctural factors, such a growing preference for part-time workersamong public sector employers and reactions by employers’ associationsto an early-1980s agreement to reduce the standard work week, wereimportant. Family policy and union strategies are found to have playeddistinctly secondary roles.

Small-N analyses of more than one country sometimes are simplymultiple case studies. Frequently, however, they are comparative: theyattempt to gain analytical leverage from cross-country comparison. Howso? In an insightful paper, James Mahoney (2000, pp. 399–406; alsoMahoney, 2003) points out that small-N analyses often engage in a sim-ple form of correlational analysis, which he refers to as ordinal comparison.The countries are rank-ordered on the outcome and on a hypothesizedcausal variable, and the analyst draws inferences about causal impactbased on the consistency of the rankings for the two variables. Supposea researcher compares employment performance in Sweden, Italy, andthe United States during the 1990s. The analyst might argue that perfor-mance was strongest in the United States, followed by Sweden and thenItaly, and that non-market institutions and policies were weakest in the

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 15


United States and strongest in Italy, with Sweden in between. Amongthese three countries, the ranked positions on the two variables are con-sistent, which supports the notion that “rigidities” had an adverse effecton employment outcomes. The analysis is depicted in Figure 1.5.

The strength of such an analysis is likely to lie in the coding of the vari-ables and perhaps the attention to interactions among them. However,this type of implicit correlational analysis is problematic with a smallnumber of cases. It is difficult to take very many factors into account (tocontrol for them). And while an inference based on well-done analysis ofthis type can certainly be suggestive of a tendential causal relationship,it leaves open the possibility that the countries analyzed are atypical. Ofcourse, large-N correlational analyses are never definitive; they too cando no more than suggest a causal relationship. Still, all else equal, thelarger the number of countries analyzed, the less reason there is to worryabout representativeness or generality of findings (Geddes, 1990; King,Keohane, and Verba, 1994).

Mahoney (2000, pp. 391–8) argues that small-N comparison offers ana-lytical leverage chiefly via its ability to eliminate a hypothesized sufficientor necessary condition. A hypothesized sufficient condition, he suggests,can be tested using John Stuart Mill’s “method of difference.” Here casesare selected that differ on the outcome; in at least one country the out-come is present and in another it is absent. Since “sufficiency” impliesthat where the cause is present the outcome will occur, any cause thatis present in the country in which the outcome is absent can be ruledout as a sufficient condition. A hypothesized necessary condition can

It

Swe

US

Bad

Goo

dE

mpl

oym

ent p

erfo

rman

ce

Low HighLabor market “rigidities”

Figure 1.5 Illustration of small-N ordinal comparison

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 16


be assessed using Mill’s “method of agreement,” in which the outcomeis present in all countries selected for analysis. Any cause that is absentin any of the countries can be ruled out as a candidate for “necessity.”Findings in these types of analyses that are consistent with a hypothesisof sufficiency or necessity can be treated as supportive, but by no meansdefinitive, since there may be other nations – possibly many of them –for which the hypothesis is contradicted. Mahoney’s point is illustratedin Figure 1.6.

We find Mahoney’s observation illuminating. However, a causal fac-tor can be eliminated as sufficient or necessary based on a single case;there is no need for cross-country comparison. Indeed, if the aim isto eliminate a hypothesized sufficient or necessary condition, compar-ison across countries is irrelevant. Comparison implies variation, andin testing a hypothesized deterministic relationship there is no need forvariation. To test a hypothesized sufficient condition, one should selectonly cases in which the condition is present/high; countries in which it isabsent/low offer no analytical leverage. To test a hypothesized necessarycondition, a researcher should select only cases in which the outcome ispresent/high. For the purpose of eliminating a hypothesized sufficient ornecessary condition, then, the utility of studying more than one countryis not that it enables comparison across the countries. Rather, the gain issimply that examining multiple countries increases the opportunity forelimination.

Given this, does a small-N comparative analysis have any advantageover a large-N analysis for eliminating a hypothesized sufficient or neces-sary condition? After all, the larger the number of countries, the greaterthe opportunity for elimination – and the greater the confidence in thehypothesis if it is not eliminated. There are two potential advantages tokeeping the N small. One is that the likelihood of measurement error maybe reduced because of the researcher’s greater case knowledge. Anotheris that the researcher may be able to consider variables for which dataare not available for a larger set of countries.

A. Eliminating a hypothesized sufficient condition

Cause Outcome———————————

Country 1 Present Present

Country 2 Present Absent

B. Eliminating a hypothesized necessary condition

Cause Outcome———————————

Country 1 Present Present

Country 2 Absent Present

Figure 1.6 Mahoney’s argument for use of small-N analysis to eliminate ahypothesized sufficient or necessary condition

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 17


For a tendential or quasi-deterministic theory, small-N analysis is morelimited in its analytical utility. A single nonconforming case, and per-haps even two or three, cannot eliminate a hypothesized tendential,quasi-sufficient, or quasi-necessary cause. Then again, a nonconformingcase increase grounds for skepticism. This is particularly true if the caseis a “most favorable” one – one that there is reason to suspect will behighly likely to support the theory, or one that proponents of the theoryfrequently refer to as an illustration (Eckstein, 1975).

A recent small-N macrocomparative analysis of employment per-formance is Ronald Schettkat’s (2005) study of Germany and theNetherlands. Schettkat provides qualitative and quantitative informa-tion to suggest that these two countries have been similar in their degreeof institutional and policy “rigidities.” According to the tradeoff hypoth-esis, therefore, both should have had poor employment performanceduring the 1980s and 1990s. Yet the Netherlands arguably had very goodemployment performance. Schettkat provides extensive detail to supportthis coding decision.

Schettkat does not say whether he considers the hypothesis he isassessing to be deterministic, quasi-deterministic, or tendential – thatis, whether labor market “rigidities” are hypothesized to be a sufficientcondition for bad employment performance or to increase the likelihoodof bad employment performance. If the hypothesis is one of sufficiency,the Dutch case contradicts it. This is shown in Figure 1.7. But note thatthe German case (like country 1 in Figure 1.6), and therefore the cross-country comparison, is not needed to reach this conclusion. If Schettkatis treating the tradeoff view as a tendential hypothesis, then the cross-country comparison is helpful. But here the fact that the analysis includesonly two countries is problematic. After all, the Netherlands could simplybe an exception to the general tendency.

Like pooled regression and QCA, then, small-N analysis offers cer-tain advantages but also has important limitations. Its main assetsare descriptive detail, care in measurement, ability to consider causal

Hypothesizedsufficient cause:labor market“rigidities”

Outcome: bademploymentperformance

—————————————————————Germany Present Present

Netherlands Present Absent

Figure 1.7 Summary of analysis in Schettkat (2005)

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 18


variables for which data may not be available for a larger set of coun-tries, process tracing as a means of assessing causal mechanisms, andelimination of hypothesized sufficient or necessary conditions. Its prin-cipal drawbacks are limited generalizability and concern about omittedvariable bias.

Overview of the chapters

The book’s chapters are organized partly by substantive topic and partlyby methodological approach. Table 1.2 provides a summary.

Chapters 2 and 3 examine the rigidities ➔ poor employment per-formance hypothesis. These two chapters use the same data set.The employment data are for private sector employment in low-endservices – hotels, restaurants, wholesale and retail trade, and commu-nity/social/personal services (ISIC revision 2, sectors 6 and 9). Thesedata are available only through the mid-1990s, but they provide thetruest test of the rigidities hypothesis (see Iversen and Wren, 1998;Kenworthy, 2004, ch. 5). The analyses focus on six labor market institu-tions and policies: earnings inequality (a proxy for low-end wage levels),wage increases, payroll and consumption taxes, employment protectionregulations, unemployment benefit generosity, and public employment.

In chapter 2, Bernhard Kittel offers a clear and careful illustrationof some of the potential pitfalls of pooled regression. Perhaps mostimportant, he finds that the choice to pool observations adds little infor-mation and introduces significant estimation problems. Adding annualobservations does not add relevant variation to several key explanatoryvariables, as they are largely constant over time. And the dependent vari-able turns out to be nonstationary. This leads Kittel to prefer a small-Ncross-sectional design over a moderate-N pooled one. The lesson is notthat simple cross-sectional models are always or even usually preferable,but rather that analysts should use pooled regression where doing somakes theoretical and empirical sense, not simply because data avail-ability makes it possible to do so. Kittel’s substantive conclusion is thatgenerous unemployment benefits may reduce the employment growthof private consumer-service jobs, at least when employment protectionregulations are not stringent and when benefit generosity is measuredusing gross (pretax) rather than net benefits.

In chapter 3, Jessica Epstein, Daniel Duerr, Lane Kenworthy, andCharles Ragin use fuzzy-set qualitative comparative analysis to explorethe impact of the same institutions and policies on growth of employ-ment in private sector consumer-oriented services. They focus on paths

9780230_202573_02_cha01.tex

5/6/200815:33

Page19

Table 1.2 Summary of the contributions

Chapter Author(s) Dependent variable(s) Unit(s) of analysis Countries and years Method

2 Kittel Level and change in employment Country-years 14 countries, Regression: pooled and cross-sectionalin private-sector consumer-oriented and countries 1979 to 1995services

3 Epstein, Duerr, Change in employment in Countries 14 countries, Fuzzy-set QCAKenworthy, private-sector consumer-oriented 1979 to 1995and Ragin services

4 Misra and Jude Level and change in women’s Countries France and the Small-N analysis: within-countrytotal, full-time, and part-time Netherlands, 1960s process tracing and cross-countryemployment through 1990s ordinal comparison

5 Eliason, Stryker, Level of women’s labor force Country-years 14 countries, Fuzzy-set QCA with compliers averageand Tranby participation 1960 to 1999 causal effects (CACE) analysis

6 Hicks and Kenworthy Level and change in women’s Country-decades 14 countries, 1960s Regression: pooled and cross-sectionalemployment through 1990s

7 Visser and Yerkes Employment status and Individuals Germany, the Regression: multinomial logitemployment transitions among Netherlands, andwomen: full-time, long-hour the United Kingdom,part-time, short-hour part-time, 1992 to 2002non-employed

8 Ebbinghaus Early-exit rates and regimes Early-exit regimes 10 countries, Small-N analysis: cross-regimeamong men and women age and countries 1970 to 2003 ordinal comparison and55–59 and 60–64 within-regime process tracing

9 Przeworski Change in labor force Country-years 135 countries, Regression (pooled) withparticipation 1950 to 1990 selection bias estimators

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 20


to slow employment growth or employment loss (“poor employmentperformance”). One of the aims of the chapter is to carefully illustratethe mechanics of a fuzzy-set QCA analysis, which is far less commonlyused than regression in macrocomparative research. The chapter illus-trates the potential usefulness of QCA in situations where researcherswant to explore hypotheses of sufficiency and/or necessity and where theinterest is primarily cross-sectional. Although QCA is particularly adeptat examining multiple causal paths to the same outcome and at consid-ering combinations of causal factors, in this particular analysis it turnsout that there are only two causal paths consistent with a sufficiencyhypothesis and only one of them involves multiple causal factors. Theresults center on one simple causal configuration and another singularcausal factor: (1) low earnings inequality combined with high payrolland consumption taxes; (2) high unemployment benefit generosity.

Chapters 4–7 shift the focus to women’s labor force participation andemployment, with an emphasis on the impact of family policies.

In chapter 4, Joya Misra and Lucian Jude examine the effect of fam-ily policy on women’s employment in a small-N analysis of France andthe Netherlands. Part of their aim is to carefully trace over-time devel-opments in these two countries and thereby explore in a detailed andnuanced fashion the role of family policy, economic conditions, andcultural support. They also are interested in understanding two differ-ences between these countries: (1) higher full-time female employmentin France by the 1960s and 1970s; (2) dramatic growth in (mainly part-time) women’s employment in the Netherlands beginning in the 1980sversus stagnation in France. Based on their analysis, they argue thata combination of supportive family policy, greater economic need forwomen’s employment, and cultural support explains both the initialhigher levels of women’s employment in France as well as the dramaticgrowth of women’s employment in the Netherlands. However, culturaland policy differences in respect of caregiving for young children remain,helping explain the much higher levels of part-time employment amongDutch women.

One of the purposes of this volume is to highlight the advantages anddisadvantages of alternative methodological techniques. Equally impor-tant, however, is to move beyond these discussions to emphasize payoffsresulting from combining multiple methods. In chapter 5, Scott Eliason,Robin Stryker, and Eric Tranby combine fuzzy-set QCA methods with ananalysis of compliers average causal effects (CACE) to explore the impactof left government on family policies and of family policies on femalelabor force participation from the 1960s through the 1990s. The principal

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 21


family policies they consider are child care, maternity leave, publicemployment, and child benefits. They offer a methodological innova-tion in coupling the QCA analysis with a CACE analysis. The aim ofthe latter is to assess whether the effect resembles one that would havebeen observed of compliers had the treatment been randomly assigned,as would be the case in a typical experimental design framework.

They conclude that both demand-side and supply-side factors causallyinfluence female labor force participation. On the demand side, theyfind that an expanded public sector has a substantial impact on femalelabor force participation. On the supply side, they find that maternityleave and public day care programs also have non-negligible causal effectson female labor force participation, although in some cases modest incomparison to demand-side mechanisms.

In chapter 6, we (Hicks and Kenworthy) use Eliason, Stryker, andTranby’s family policy data to explore the relationship between familypolicy generosity and female employment via regression analysis. Weargue that even if the effect of family policy generosity on women’semployment is conceptualized in a deterministic fashion – as a sufficientcondition – there may be reason for concern about omitted variable bias.We examine the possibility that the association between family policyand female employment is spurious – a product of the fact that both areassociated with women’s educational attainment.

We examine unconstrained pooled models, pooled models with fixedeffects for time or country, and a cross-sectional model. The pooled mod-els without unit effects for countries suggest that both family policy andwomen’s educational attainment have tended to boost women’s employ-ment rates. In pooled models with country unit effects, however, thereis little or no indication of a family policy impact. In cross-sectionalmodels, we once again find support for effects of both family policiesand female education. Further exploration of the over-time develop-ments within countries confirms that support for the hypothesis thatgenerous family policies tend to increase female employment rates restslargely on the cross-sectional association. This does not mean familypolicies do not affect women’s employment, but it suggests less confi-dence than if there were supportive evidence both across countries andwithin countries over time.

In chapter 7, Jelle Visser and Mara Yerkes pursue a very promisinganalytical strategy in macrocomparative research: the use of individual-level panel data in a small, but deliberately chosen set of countries. Theyuse individual-level data to examine the effects of institutions, policies,and women’s preferences regarding employment and working hours in

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 22


Germany, the United Kingdom, and the Netherlands. These three coun-tries exhibit the highest incidence of women working part-time in theOECD. They also share a “breadwinner legacy”: each traditionally dis-couraged employment of mothers with young children, though theyhave in various ways moved away from that position in recent decades.Visser and Yerkes exploit the variation in welfare state and industrialrelations institutions and policies across the three countries and the avail-ability of individual panel data to explore how strongly the breadwinnerlegacies still affect the choice for and nature of part-time work of women.

Visser and Yerkes first estimate the effects of motherhood on the prob-ability of adult women to be full-time employed, part-time employed,or entirely outside the labor force (“inactive”). They focus on differencesacross birth cohorts, controlling for education and household status.Next they analyze transitions from part-time employment into inactiv-ity or full-time employment, focusing on the impact of motherhood.Comparison with the transitions from full-time jobs into long-hour orshort-hour part-time jobs or into inactivity can help answer the ques-tion of whether part-time employment encourages particular groups ofwomen to remain in the labor force. Finally, they examine the impactof “choice,” bringing into play working-time preferences of women andanalyzing whether or not they lead to transitions in the desired direction.

In chapter 8, Bernhard Ebbinghaus analyzes comparative patterns ofearly exit from the labor market. Affluent countries differ sharply in theiremployment rates among those aged 55 to 64 and in the degree to whichthose rates have shifted over the past several decades. Ebbinghaus exam-ines the impact of “early-exit regimes” (see also Ebbinghaus, 2006). Theseregimes are defined by three factors: social policy orientation (protectionsystems, which are hypothesized to differ in the degree to which theypull older workers out of the labor force), the organization of production(production systems, which are hypothesized to differ in the degree towhich they push older workers out of employment), and the organiza-tion of labor relations (partnership traditions, which are hypothesized todifferentially mediate pull and push factors). Ebbinghaus identifies fiveregimes. He selects ten countries (this is at the upper end of what we call“small-N” analysis; Ebbinghaus refers to it as “medium-N”) that enablecomparison both across and within these regimes. The cross-regimeanalysis is based on ordinal comparison as described above: regimesare ranked on the degree to which they are expected to promote earlyexit, and this ranking is correlated with a quantitative measure of earlyexit (relative exit rate). The within-regime analysis aims to account forthe unexplained variation among countries within regimes; it is akin

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 23


to attempting to account for the residuals in a large-N quantitativeanalysis.

Through his combination of within-case process-oriented and cross-case contextual analysis, Ebbinghaus finds that it is not only theincentives provided by welfare state pathways to early retirement thatexplain the cross-national variations but also the particular strategiesof employers and workplace representatives in coping with particulareconomic pressures caused by different production strategies. He usesoutliers as special cases to investigate the interaction between “pull” and“push” factors. For instance, the Swedish case shows that a generous wel-fare state does not always produce high early retirement, while Japanesefirms use mandatory retirement but also provide re-employment forolder workers, explaining their high employment rate in a coordinatedmarket economy.

In chapter 9, Adam Przeworski considers the crucial issue of selectionbias in macrocomparative analysis. The concern is that what appearsto be a causal effect of some institution or policy might rather be aneffect of whatever gave rise to that institution or policy. For exam-ple, generous family policies are found mainly in the Nordic countries.We observe a positive association between family policy generosity andfemale employment rates across countries, but the true cause of the lattermight be some feature of the Nordic societies or their policy-making pro-cesses that led them to adopt generous family policies, rather than thefamily policies themselves. As Przeworski puts it: “The generic problemin identifying causal effects is how to answer the counterfactual question:what would have occurred had the cause been absent?” In this example,the counterfactual hypothesis is that female employment rates would notbe comparatively high in the Nordic countries had those countries notimplemented generous family policies. Przeworski notes that “Whetherwe can successfully solve such problems is . . . largely a matter of luck,namely whether history has been kind enough to generate observationsthat can be used to inform us about the plausible counterfactuals.”

The substantive question Przeworski explores is the impact of politicalregime – conceptualized dichotomously as democracy or autocracy – onlabor force participation. Because the affluent countries are all democra-cies and because the approach Przeworski uses to estimate selection biasrequires a relatively large number of countries, he includes not only richnations but developing ones as well. His chapter nicely illustrates the useof appropriate techniques for addressing the selection bias worry.

In analyses confined to the rich countries, selection bias is, unfortu-nately, both more likely to be present and less amenable to statistical

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 24


estimation. The only recourse for the macrocomparative analyst iscareful, and explicit, counterfactual thinking (Fearon, 1991; Esping-Andersen and Przeworski, 2001). For the most part the contributions tothis book set this issue aside, but it ought to be an increasingly prominentconcern in such analyses.

Onward

The chapters in this volume attempt to highlight the advantages anddrawbacks of some prominent methodological approaches to macrocom-parative analysis. The principal aim is to help researchers – ourselvesincluded – to make more informed choices about which approach(es) touse in their research and to make better use of whichever one(s) theychoose. We hope the book succeeds in this endeavor.

Notes

1. Thus far, however, relatively few macrocomparative studies have made useof more than one of these methods in analyzing a particular research ques-tion. Regression and small-N analysis are combined in Boix (1998), Huber andStephens (2001), Swank (2002), and Kenworthy (2008). Ebbinghaus and Visser(1999) and Hicks (1999) couple regression with QCA.

2. It is sometimes thought that an analysis of a single country, usually referredto as a “case study,” is not comparative. But most such studies are comparative(Rueschemeyer, 2003; Gerring, 2005). The comparison is not across countriesbut rather over time and/or across sub-units (regions, localities) within thecountry. Although small-N analyses tend to be qualitative, they can be quanti-tative as well; the distinction between small-N and large-N analysis is not thesame as that between quantitative and qualitative analysis.

References

Ayer, A. J. 1956. The Problem of Knowledge. London: Penguin.Baccaro, Lucio and Diego Rei. 2007. “Institutional Determinants of Unem-

ployment in OECD Countries: Does the Deregulatory View Hold Water?”International Organization 61: 527–569.

Barro, Robert J. 2000. “Inequality and Growth in a Panel of Countries.” Journal ofEconomic Growth 5: 5–32.

Blau, Francine D. and Lawrence M. Kahn. 2002. At Home and Abroad: U.S.Labor Market Performance in International Perspective. New York: Russell SageFoundation.

Boix, Carles. 1998. Political Parties, Growth, and Equality. Cambridge: CambridgeUniversity Press.

Brady, Henry E. and David Collier, eds. 2004. Rethinking Social Theory. Lanham,MD: Rowman and Littlefield.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 25


Collier, David, Henry E. Brady, and Jason Seawright. 2004. “Sources of Leverage inCausal Inference: Toward an Alternative View of Methodology.” Pp. 229–66 inRethinking Social Inquiry, edited by Henry E. Brady and David Collier. Lanham,MD: Rowman and Littlefield.

Daly, Mary. 2000. “A Fine Balance: Women’s Labor Market Participation in Inter-national Comparison.” Pp. 467–510 in Welfare and Work in the Open Economy,Volume II: Diverse Responses to Common Challenges, edited by Fritz W. Scharpfand Vivien A. Schmidt. Oxford: Oxford University Press.

Dingeldey, Irene. 2002. “European Tax Systems and Their Impact on FamilyEmployment Patterns.” Journal of Social Policy 30: 653–72.

Ebbinghaus, Bernhard. 2005. “When Less Is More: Selection Problems in Large-Nand Small-N Cross-National Comparisons.” International Sociology 20: 133–52.

Ebbinghaus, Bernhard. 2006. Reforming Early Retirement in Europe, Japan and theUSA. Oxford: Oxford University Press.

Ebbinghaus, Bernhard and Jelle Visser. 1999. “When Institutions Matter: UnionGrowth and Decline in Western Europe, 1950–1995.” European SociologicalReview 15: 135–58.

Eckstein, Harry. 1975. “Case Study and Theory in Political Science.” In Strategiesof Inquiry: The Handbook of Political Science, vol. 7, edited by F. Greenstein andN. Polsby. Reading, MA: Addison-Wesley.

Esping-Andersen, Gøsta. 1999. Social Foundations of Postindustrial Economies.Oxford: Oxford University Press.

Esping-Andersen, Gøsta, with Duncan Gallie, Anton Hemerijck, and JohnMyles. 2002. Why We Need a New Welfare State. Oxford: Oxford UniversityPress.

Esping-Andersen, Gøsta and Adam Przeworski. 2001. “Quantitative Cross-National Research Methods.” Pp. 12649–55 in International Encyclopedia of theSocial and Behavioral Sciences. Amsterdam: Elsevier.

Fearon, James D. 1991. “Counterfactuals and Hypothesis Testing in PoliticalScience.” World Politics 43: 169–95.

Ferrarini, Tommy. 2003. Parental Leave Institutions in Eighteen Post-War WelfareStates. Swedish Institute for Social Research Dissertation Series, No. 58.

Ferrera, Maurizio, Anton Hemerijck, and Martin Rhodes. 2000. “The Future ofSocial Europe: Recasting Work and Welfare in the New Economy.” Reportprepared for the Portuguese Presidency of the European Union.

Geddes, Barbara. 1990. “How the Cases You Choose Affect the Answers You Get:Selection Bias in Comparative Politics.” Pp. 131–50 in Political Analysis, editedby James A. Stimson. Ann Arbor: University of Michigan Press.

George, Alexander L. and Andrew Bennett. 2004. Case Studies and TheoryDevelopment in the Social Sciences. Cambridge, MA: MIT Press.

Gerring, John. 2004. “What Is a Case Study and What Is It Good For?” AmericanPolitical Science Review 98: 341–54.

Goertz, Gary and Harvey Starr. 2003. “Introduction: Necessary Condition Logics,Research Design, and Theory.” Pp. 1–23 in Necessary Conditions, edited by GaryGoertz and Harvey Starr. Lanham, MD: Rowman and Littlefield.

Gornick, Janet C. and Marcia K. Meyers. 2003. Families That Work: Policies forReconciling Parenthood and Employment. New York: Russell Sage Foundation.

Greene, William H. 2003. Econometric Analysis, 5th edition. New York: PrenticeHall.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 26


Griffin, Larry J., Pamela Barnhouse Walters, Phillip O’Connell, and Edward Moor.1986. “Methodological Innovations in the Analysis of Welfare-State Devel-opment: Pooling Cross Sections and Time Series.” Pp. 101–38 in Futures forthe Welfare State, edited by Norman Furniss. Bloomington: Indiana UniversityPress.

Hall, Peter A. 2003. “Aligning Ontology and Methodology in ComparativeResearch.” Pp. 373–404 in Comparative Historical Analysis in the Social Sciences,edited by James Mahoney and Dietrich Rueschemeyer. Cambridge: CambridgeUniversity Press.

Hicks, Alexander. 1999. Social Democracy and Welfare Capitalism. Ithaca, NY:Cornell University Press.

Hicks, Alexander and Lane Kenworthy. 1998. “Cooperation and Political Eco-nomic Performance in Affluent Democratic Capitalism.” American Journal ofSociology 103: 1631–72.

Hicks, Alexander and Lane Kenworthy. 2003. “Varieties of Welfare Capitalism.”Socio-Economic Review 1: 27–61.

Huber, Evelyn and John D. Stephens. 2001. Development and Crisis of the WelfareState. Chicago: University of Chicago Press.

Iversen, Torben and Anne Wren. 1998. “Equality, Employment, and BudgetaryRestraint: The Trilemma of the Service Economy.” World Politics 50: 507–46.

Jahoda, Marie. 1982. Employment and Unemployment: A Social PsychologicalAnalysis. Cambridge: Cambridge University Press.

Janoski, Thomas and Alexander Hicks, eds. 1994. The Comparative PoliticalEconomy of the Welfare State. Cambridge: Cambridge University Press.

Jaumotte, Florence. 2003. “Female Labour Force Participation: Past Trends andMain Determinants in OECD Countries.” Working Paper 376, OECD EconomicsDepartment. Paris: OECD.

Katz, Aaron, Matthia vom Hau, and James Mahoney. 2005. “Explaining the GreatReversal in Spanish America: Fuzzy-Set Analysis versus Regression Analysis.”Sociological Methods and Research 33: 539–73.

Kenworthy, Lane. 2004. Egalitarian Capitalism. New York: Russell SageFoundation.

Kenworthy, Lane. 2006. “Reconsidering the Effect of Public Opinion on Social Pol-icy Generosity in Affluent Countries.” Unpublished. Department of Sociology,University of Arizona.

Kenworthy, Lane. 2007. “Toward Improved Use of Regression in Macrocompara-tive Analysis.” Comparative Social Research 24: 343–50.

Kenworthy, Lane. 2008 (forthcoming). Jobs with Equality. Oxford: OxfordUniversity Press.

King, Gary, Robert O. Keohane, and Sidney Verba. 1994. Designing Social Inquiry.Princeton, NJ: Princeton University Press.

Kittel, Bernhard. 1999. “Sense and Sensitivity in Pooled Analysis of Political Data.”European Journal of Political Research 35: 225–53.

Kok, Wim, Carlos Dell’Aringa, Federico Duran Lopez, Anna Eckström, Marta JoãoRodrigues, Christopher Pissarides, Annette Roux, and Günther Schmid. 2003.Jobs, Jobs, Jobs: Creating More Employment in Europe. Report of the EuropeanCommission’s Employment Taskforce. Brussels: European Commission.

Korpi, Walter. 2000. “Faces of Inequality: Gender, Class, and Patterns ofInequalities in Different Types of Welfare States.” Social Politics 7: 127–91.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 27


Lieberman, Evan S. 2005. “Nested Analysis as a Mixed-Method Strategy forComparative Research.” American Political Science Review 99: 435–52.

Lieberson, Stanley. 1991. “Small N’s and Big Conclusions: An Examination of theReasoning in Comparative Studies Based on a Small Number of Cases.” SocialForces 70: 307–20.

Mahoney, James. 2000. “Strategies of Causal Inference in Small-N Analysis.”Sociological Methods and Research 28: 387–424.

Mahoney, James. 2003. “Strategies of Causal Assessment in Comparative Histori-cal Analysis.” Pp. 337–72 in Comparative Historical Analysis in the Social Sciences,edited by James Mahoney and Dietrich Rueschemeyer. Cambridge: CambridgeUniversity Press.

Mahoney, James and Gary Goertz. 2006. “A Tale of Two Cultures: ContrastingQuantitative and Qualitative Research.” Political Analysis 14: 227–49.

Mahoney, James and Dietrich Rueschemeyer, eds. 2003. Comparative HistoricalAnalysis in the Social Sciences. Cambridge: Cambridge University Press.

Mandel, Hadas and Moshe Semyonov. 2006. “A Welfare State Paradox: State Inter-ventions and Women’s Employment Opportunities in 22 Countries.” AmericanJournal of Sociology. 111: 1910–49.

Manicas, Peter T. 2006. A Realist Philosophy of Science: Explanation and Understand-ing. Cambridge: Cambridge University Press.

Meyers, Marcia K., Janet C. Gornick, and Katherin E. Ross. 1999. “Public Child-care, Parental Leave, and Employment.” Pp. 117–46 in Gender and Welfare StateRegimes, edited by Diane Sainsbury. Oxford: Oxford University Press.

Morgan, Kimberly and Kathrina Zippel. 2003. “Paid to Care: The Origins andEffects of Care Leave Policies in Western Europe.” Social Politics 10: 49–85.

OECD (Organization for Economic Cooperation and Development). 1994. TheOECD Jobs Study. Paris: OECD.

OECD. 2001. “Balancing Work and Family Life: Helping Parents Into PaidEmployment.” Pp. 129–66 in OECD Employment Outlook. Paris: OECD.

OECD. 2005. Extending Opportunities: How Active Social Policy Can Benefit Us All.Paris: OECD.

OECD. 2006. OECD Employment Outlook: Boosting Jobs and Incomes. Paris: OECD.Orloff, Ann. 2002. “Women’s Employment and Welfare Regimes.” Social Policy

and Development Programme Paper 12. United Nations Research Institute forSocial Development.

Passmore, John. 1967. A Hundred Years of Philosophy. London: Macmillan.Pettit, Becky and Jennifer Hook. 2002. “The Structure of Women’s Employment

in Comparative Perspective.” Working Paper 330. Luxembourg Income Study.Available at: www.lisproject.org.

Phelps, Edmund S. 1997. Rewarding Work. Cambridge, MA: Harvard UniversityPress.

Plantenga, Janneke and Johan Hansen. 1999. “Assessing Equal Opportunities inthe European Union.” International Labour Review 138: 351–79.

Pruss, Alexander R. 2006. The Principle of Sufficient Reason. Cambridge: CambridgeUniversity Press.

Ragin, Charles C. 1987. The Comparative Method. Berkeley: University of CaliforniaPress.

Ragin, Charles C. 2000. Fuzzy-Set Social Science. Chicago: University of ChicagoPress.

9780230_202573_02_cha01.tex 5/6/2008 15: 33 Page 28


Ragin, Charles C. 2005. “Core versus Tangential Assumptions in ComparativeResearch.” Studies in Comparative International Development 40: 33–8.

Ragin, Charles C. 2006. “Set Relations in Social Research: Evaluating TheirConsistency and Coverage.” Political Analysis 14: 291–310.

Ragin, Charles C. and Benoît Rihoux. 2004. “Qualitative Comparative Analysis(QCA): State of the Art and Prospects.” Qualitative Methods Newsletter, Fall: 3–13.

Rubery, Jill, Mark Smith, and Colette Fagan. 1999. Women’s Employment in Europe.London: Routledge.

Rueschemeyer, Dietrich. 2003. “Can One or a Few Cases Yield Theoretical Gains?”Pp. 305–36 in Comparative Historical Analysis in the Social Sciences, edited byJames Mahoney and Dietrich Rueschemeyer. Cambridge: Cambridge UniversityPress.

Ruhm, Christopher J. 1998. “The Economic Consequences of Parental LeaveMandates: Lessons from Europe.” Quarterly Journal of Economics 113: 285–317.

Sainsbury, Diane. 1999. “Taxation, Family Responsibilities, and Employment.”Pp. 185–209 in Gender and Welfare State Regimes, edited by Diane Sainsbury.Oxford: Oxford University Press.

Scharpf, Fritz W. and Vivien A. Schmidt, eds. 2000. Welfare and Work in theOpen Economy. Volume 1: From Vulnerability to Competitiveness. Oxford: OxfordUniversity Press.

Schettkat, Ronald. 2005. “Is Labor Market Regulation at the Root of EuropeanUnemployment? The Case of Germany and the Netherlands.” Pp. 262–83 inFighting Unemployment: The Limits of Free Market Orthodoxy, edited by David R.Howell. Oxford: Oxford University Press.

Seawright, Jason. 2002. “Testing for Necessary and/or Sufficient Causation: WhichCases Are Relevant?” Political Analysis 10: 178–93.

Seawright, Jason. 2005. “Qualitative Comparative Analysis vis-à-vis Regression.”Studies in Comparative International Development 40: 3–26.

Shalev, Michael. 2007. “Limits and Alternatives to Multiple Regression inComparative Research.” Comparative Social Research 24: 261–308.

Skocpol, Theda and Margaret Somers. 1980. “The Uses of Comparative History inMacrosocial Inquiry.” Comparative Studies in Society and History 22: 174–97.

Stier, Haya, Noah Lewin-Epstein, and Michael Braun. 2001. “Welfare Regimes,Family-Supportive Policies, and Women’s Employment Along the Life-Course.”American Journal of Sociology 106: 1731–60.

Swank, Duane. 2002. Global Capital, Political Institutions, and Policy Change inDeveloped Welfare States. Cambridge: Cambridge University Press.

Visser, Jelle. 2002. “The First Part-Time Economy in the World: A Model To BeFollowed?” Journal of European Social Policy 12: 23–42.

Visser, Jelle and Anton Hemerijck. 1997. “A Dutch Miracle”: Job Growth, WelfareReform, and Corporatism in the Netherlands. Amsterdam: Amsterdam UniversityPress.

Wilson, Sven E. and Daniel M. Butler. 2007. “A Lot More to Do: The Sensitiv-ity of Time-Series Cross-Section Analyses to Simple Alternative Specifications.”Political Analysis 15: 101–23.

Wilson, William Julius. 1996. When Work Disappears. New York: Knopf.Winegarden, C. R. and Paula M. Bracy. 1995. “Demographic Consequences of

Maternal-Leave Programs in Industrial Countries: Evidence from Fixed-EffectsModels.” Southern Economic Journal 6: 1020–35.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 29

2Statistical Narratives and theProperties of Macro-Level Variables:Labor Market Institutions andEmployment Performance inMacrocomparative ResearchBernhard Kittel

2.1 Introduction

Panel data have become paramount in macrocomparative researchdesigns in many branches of the social sciences, but in particular in polit-ical economy and welfare state research. However, by including the timedimension into cross-sectionally posed research questions, much moreis added than just a few observations. Substantive issues like dynam-ics, variations in effect types, feedback, learning, and path dependencyimpact on the possibilities for making inferential statements. Technicalissues like nonstationarity and serial correlation affect the researcher’sability to draw conclusions from coefficients in regression models.

In this chapter, I attempt to contribute to the quest for more clarityabout the potentials and limits of panel data for comparative politicaleconomy by analysing a data set containing variables that clearly showthe issues at stake in the combination of cross-sectional and longitudi-nal data. In the first section, I discuss three types of concepts typical ofthis research area – economic outcomes, policies, and institutions – andfocus on their longitudinal properties. In the second section, I reanalyzea well-known study of the statistical associations between employment,the reservation wage, and labor market regulation, thereby including oneof each of the three types (Kenworthy, 2003). The third section focuseson the empirical evidence on a simple hypothesis linking institutional,policy, and economic variables. The basic theoretical argument is alsobased on Kenworthy (2003) and addresses the impact of labor marketinstitutions on employment opportunities. Kenworthy focuses on two

29

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 30


dimensions of labor market institutions in particular: pay equality andunemployment compensation. According to a well-established view, rel-atively high wages at the lower end of the wage scale will discourageemployers to create jobs. This is a demand-side argument, linking insti-tutions and policy variables to employment via an assumption aboutbehavior of the typical employer. By contrast, a high reservation wagewill negatively affect the willingness of displaced workers to take onemployment in the low-wage sector. This is a supply-side argument,because institutions and policies affect the typical potential employee’sbehavior. In his analysis, which is based on a pooled time-series cross-section regression model, Kenworthy finds evidence in favor of bothpropositions, but concludes that the effect is relatively weak. My reanal-ysis reveals a few inferential problems related to the use of institutionaland policy variables in cross-sectionally focused panel analyses and addsa few nuances to Kenworthy’s findings. Finally, in the fourth section Idraw some conclusions.

2.2 Time and the properties of macro-level variables

In comparative research, the implications of extending the researchdesign with the longitudinal dimension have not yet been fully appreci-ated (Pierson, 2004). Macrocomparative analysis is concerned with thesearch for empirical – mostly unobvious – regularities at the level of soci-etal organization. In comparative policy research, the bulk of researchidentifies nation states as units of analysis and primarily studies researchquestions about the variation between nation-states. For many questionsin this field, this is indeed an appropriate choice, given that both the phe-nomena to be explained and the phenomena proposed as explanatoryfactors are emergent at that level of societal organization. The typicalexplanandum is variation in some aggregate economic variable, whichis expected to vary systematically with some explanans. The typicalexplanans put forward in policy analysis is variation in policies or ininstitutions. Notwithstanding these modal positions in typical theoreti-cal models, which reflect some implicit assumptions about the nature ofthese phenomena, this mapping is not crisp. Since there is no clear uni-directional causal structure between these variables, one can find all ofthem both as explananda and explanans, although there is a clear prefer-ence by researchers to consider institutions and policies as independentand economic outcomes as dependent variables (examples of exceptionsare Ebbinghaus and Kittel, 2005; Polillo and Guillén, 2005).

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 31

Bernhard Kittel 31

Economic outcomes are usually economic aggregates, like economicgrowth, unemployment, inflation, etc. More recently, researchers haveincreasingly focused on more direct measures of state intervention,for example social expenditures. Policy differences are often mea-sured indirectly via indicators like the composition of government orthe coordination mode of wage bargaining. These operationalizationsbuild on assumptions about the likely behavior of particular collectiveactors. Sometimes, more direct policy measures are used, for examplethe percentage spent on active labor market policies, the replacementrate for unemployed workers, or the tax rate for certain populationsegments. Variation between institutions, in turn, is usually mea-sured by means of an indicator summarizing specific organizationalcharacteristics supposed to be relevant for the outcome. Examplesof such variables are indicators of democracy and political regimes,the independence of central banks, or the organizational structure ofassociations.

Although economic outcomes, policies, and institutions can be mea-sured at the same level of aggregation, the three types of variables differ intheir ontology in a way that is consequential for the manner they can bedealt with in macrocomparative analysis. Lazarsfeld and Menzel (1961)distinguish between analytical, structural, and global macrovariables.Analytical variables relate to properties of collectives that are derivedfrom properties of members of those collectivities by performing somemathematical operation. Structural variables refer to relations betweenmembers of a collectivity, which hence can only be captured at the levelof the collectivity. Global variables inform about emergent properties ofthe collectivity that are not based on information about the propertiesof individuals.

This classification, although useful for many sociological applications,must be adapted for the use in comparative policy research, because rela-tions between members of a collectivity are less relevant in the focusedtype of research questions, while global variables must be divided intotwo subclasses. Hence I propose a somewhat different classification.Economic macrovariables are the result of aggregated market behaviorof individuals, policy variables relate to collective action, and institu-tional variables to the framework of action. Here, I want to highlightdifferences in two dimensions and discuss the consequences of thesedifferences. First, I argue that the aggregate phenomena these variablesare meant to measure are constituted in different ways. Secondly, thesedifferences have implications for the time horizon of change in thesevariables and thus impact on the substantively relevant measurement

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 32


intervals. And thirdly, I discuss implications of these observations forthe use of econometric techniques in comparative policy analysis.

Macroeconomic variables are usually aggregated from individual actsor states. For example, economic growth is derived from individualbuying and selling acts by which these individuals create added value.Summing all of these transactions and comparing them to the samesum for the previous accounting period yields a measure of growth. Theunemployment rate is defined as the number of unemployed dividedby the number of people on the labor market. While the exact defini-tion depends upon various other definitions, which vary across spaceand time, like the difficult issue of who is to be considered part of thedependent workforce and who is assumed not to be on the job market,the macroeconomic unemployment rate is obtained by comparing thesums of individual states: individual workers are either employed, unem-ployed, or not available for employment and the aggregate measure givesan indication of the distribution of individuals among these sets. Thepoint to be highlighted is the aggregation mechanism underlying thesemeasures. It is a simple transformation rule – a sum or a ratio of sums.The aggregate measure says something about typical behavior or statesof individuals: If the economy grows, individuals have on average beenengaged in more economic transactions. The employment rate gives theprobability that any randomly selected individual out of a populationis employed. Both economic transactions of individuals and their statesare the result of individual decisions that are at the core autonomousand independent from each other. While there may be strong struc-tural forces inducing particular individual behavior, which may causean empirical regularity at the macro level, the point remains that theaggregate measure is the result of individual decisions.

Policy variables do not represent typical individual behavior or states.Instead, they relate to decisions of a collective actor and hence areemergent at the level of analysis. In macrocomparative policy analysisthis actor is usually the nation-state because the national governmentis responsible for the decision and implementation of the majority ofpolicies.1 In democracies, the decisions are taken by representatives inthe name of the individuals constituting the members of the collectiveactor. These collective decisions are the result of a deliberation processin which various segments of the population and other interest groupslobby for a formulation of the decision which favors their particular inter-ests. In contrast to economic variables, scores on policy variables are thusnot simply the sum of individual phenomena, but result from the tug-of-war of interests and thus crucially depend in their creation mode on the

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 33

Bernhard Kittel 33

prevalent power distribution. For example, the amount of money spenton active labor market policies depends only to a very limited extenton demand factors like individual applications for participation in suchprograms. More important are supply factors like the amount of publicspending earmarked for this aim, eligibility rules, and other regulations.These are not the result of a simple statistical aggregation of individualproperties or behavior, but are defined by collective decision. Even moreevident is this difference in the case of the tax rate or the replacement rateof unemployment benefits: both are more or less arbitrarily set and rede-finable guidelines for the conduct of administrative units. They are notaggregated via some simple transformation rule from individual behav-ior, but the result of state action and therefore emergent at the level ofthe nation state. Hence they lack a straightforward microfoundation.

Institutional variables are again different. With policy variables theyshare the emergent characteristic of not being directly traceable to indi-vidual action. Instead, while the result of past collective decisions, theyare system-level phenomena structuring individual behavior by definingrules and norms. They are meant to be the framework of rules in whichboth collective policies are convened and individual decisions are taken.Democratic regimes vary in terms of the mechanisms of collective deci-sion making, in the structure of representation, in the distribution ofrights and powers between various constitutional actors. Central banksvary in the extent to which they can set their policies independentlyfrom governmental preferences.

This ontology of macro-level variables has implications for the timehorizon of dynamics. Economic variables, which are constituted by thebehavior of individuals, obey regularity criteria to a considerable extent.Individuals may need some time to respond to external changes, somewill adjust faster and others slower, for some a change will have moreimpact than for others, but for large populations of individuals there willbe some modal response of mutually fairly independent decisions, whichin total result in some average behavior with a more or less symmet-ric, single-peaked distribution of behavior. Hence changes in aggregatebehavior at the macro level can be interpreted as shifts in the averageindividual response to some triggering factor. Given this individually-based responsiveness, usually the assumption of symmetry in behavioralso makes sense (Lieberson, 1985): if prices go up, supply increases whiledemand decreases, and vice versa. Moreover, adjustments will be smoothbecause of the individual-level foundation. At increasing prices, individ-uals will not stop buying all at once, but one after the other will retreatfrom the market, thereby causing the macro-level indicator to respond

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 34


according to a practically continuous function. These processes may notbe immediate (compared to the human metric of time), but with responseperiods of a few months they will be sufficiently fast to reveal adjustmentwithin the usual time metric in macrocomparative research, which is ayear by default of availability.

Policy variables, as against economic variables, are much less likely tobe subject to regularities and have a much longer time horizon of change.The typical time horizon of such changes can better be captured by sev-eral years, up to a decade or two, with periods of practically constantscores in between. Political decisions can be changed and even reversed,but they need to follow the formal rules of decision making and willbe the object of considerable deliberation. Lock-in, ratchet, threshold,and other effects inhibit regular and symmetric ups and downs. Oncea collective decision has been taken, actors will usually be reluctant toimmediately change a policy because of public credibility, considerationsof public expectations, resistance from groups profiting from a partic-ular policy, and other concerns. Hence changes in such variables canbe expected to occur but will be happenstance, non-incremental, anddiscontinuous. While this can be said for economic variables as well atthe individual level, at the aggregate level the large number of individualsaverages this out. This statistical foundation of regularity at the macrolevel is not available for policy variables based on collective decisionstaken at the system level. Since there is no micro-level justification formacro-level regularities, the notion of a typical response is inappropriate.

In contrast to policy variables, institutions – although not entirelyimmovable – cannot easily be changed by decision of the collective actorsince they are part of the formal or informal constitution of the politicalor social system. While policies are amenable to government discretion,political institutions usually are not. Procedures to change formal insti-tutions like the nation state’s constitution are protracted and require theconsent of many veto players. Moreover, since they define the rules of thegame, institutions increase behavioral reliability and consistency of col-lective actors. As a consequence, incentives for actors to change them arerather limited. Informal institutions cannot be changed even by deliber-ation because they are based in mores and customs. This, however, marksan important distinction between formal and informal institutions.While the former change in an even more haphazard and discontin-uous ways than policies, with – often several – decades of continuityin between shifts, the latter evolve gradually and progressively at a veryslow pace. Changes are hardly noticeable and practically irrelevant in theshort run, but may cumulate to a veritable upheaval over several decades.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 35

Bernhard Kittel 35

The upshot of these reflections is that the three types of variables dif-fer, first, in the periods in which substantively relevant variation overtime can be expected and, second, in the extent to which variables canbe expected to fluctuate around some long-term average value. Manyeconomic variables, because they are based on aggregated individualmarket behavior, do indeed more or less regularly fluctuate as a contin-uous function around some equilibrium or trend, and these dynamicsare usually subject to a periodization of a few months due to responsedelays. Policy variables, because they result from collective deliberations,are characterized by more extended periods of continuity lasting severalyears, interrupted by sudden shifts. There is no inherent tendency torevert to some long-run average. Formal institutions, because of beingpart of the structure of the society or the polity, tend to remain con-stant for long periods, mostly several decades, but the constant series aresometimes interrupted by sudden shifts reflecting institutional change.Finally, informal institutions tend to evolve over long periods, countedin decades or even centuries. For both formal and informal institutionsthe concept of a long-term equilibrium value for the scores does not bearany substantive meaning.

These characteristics have implications for the possibility to studythe various types of variables in a both longitudinal and cross-nationalcomparative research design. In comparative analysis, the use of paneldata is generally motivated by the small-N problem, which severely lim-its the ability of the researcher to control for potentially confoundingfactors. Adding observations in the time dimension is hence regardedas a way out of the impasse because the number of observations ofthe cross-sectional setup are multiplied by the number of periods. Inconsequence, the shortest periodization in which data are available iscommonly regarded as the best setup (Freeman, 1990). However, inview of the above considerations, this line of reasoning may be delu-sive, because adding observations does not necessarily add information(King et al., 1994, p. 48).

The use of annual data, which has become something of a mantrain quantitative comparative research, is questionable for many researchtopics. For example, in finance economics, it makes little sense to mea-sure stock prices at annual intervals given that much of the relevantaction occurs on a daily or even hourly basis. Using annual data for suchvariables will cancel out variation which is important for most researchquestions in that discipline. Employment, by contrast, may be more rea-sonably measured at monthly intervals since more fine-grained analyseswould contain too much noise and annual averages may remove too

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 36


much of the variation. Hence using availability as the main criterion forthe periodization of the observations in the data set often does not dojustice to the substantively motivated time horizons of change in thevariables. On the other hand, annual observations do not add any sub-stantively meaningful information to variables that remain constant orevolve at a very slow pace. This does not imply that annual data shouldnot be used at the outset. My argument is different: Taking into accountdivergent periodizations in the “real world” means that for any set ofvariables, the one with the slowest rate of change determines the paceat which observations can be meaningfully related. Hence, if one vari-able changes at a rate measured in several decades, and another variablevaries annually, the only variation in the latter that can be attributed tothe former is some summary measure, usually the mean or the variance,of the latter over the decades in question. This can be done explicitlyby taking long-term averages of levels or changes or implicitly by lettingthe regression software package sort out variance components.

The problems of including the longitudinal dimension are least foreconomic variables. Actually, economic theories usually focus on con-stant conjunctions between changes in variables. They primarily focuson longitudinal associations, and pooling adds cross-sectional hetero-geneity. Generally, macroeconomic studies pool time series; hence theycombine time series instead of repeating cross-sectional designs. In orderto remove any systematic cross-sectional variation, such studies usu-ally express the data in terms of deviations from the unit-specific mean(Maddala, 1999; Baltagi, 2001). Since macroeconomic variables tend toelastically respond to changes in external factors, one can explore theregularity conditions of their dynamics with statistical methods. Theremay be variation in impact strength, delay, and speed of adjustmentover time and across spaces, but these differences can in principle bemodeled, as long as they follow some systematic pattern and the timeseries are sufficiently long. The major concern is the possibility of non-stationarity in the data, but this can often be solved by means of thetoolbox developed in time series econometrics over the last two or threedecades. Detrending, period-demeaning, first differencing, or estimatinga cointegrating system are tools that help to obtain consistent estimatesunder various conditions. If the nonstationarity problem is solved, thetime dimension often adds serial correlation to the error term. All treat-ments for this property of time series involve some sort of accountingfor dynamics.

By contrast, both policy and institutional variables have time hori-zons of change beyond annual intervals. In this respect, and viewed

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 37

Bernhard Kittel 37

from a technical perspective, the main difference between the two is thatinstitutional variables change at an even slower pace than policy vari-ables. Hence for typical observation periods in comparative analysis –between about ten and forty years – these variables are either constantor reveal only very limited variation over time, for example one or twochanges in scores. Moreover, to the extent that there are changes dur-ing the observation period, the discontinuous nature of policy makingand institutional change causes possible regularities in the responsive-ness with regard to changes in other variables, as well as regularities ineffects, to be unlikely to show up in small-N situations. As a result, suchvariables capture only or predominantly cross-sectional variation in thedata and all short-term (e.g., annual) variation in the other variablesremains unaccounted for. The problem is not only that the estimates ofthe cross-sectional variance components are again based on the num-ber of cross-sections (represented by the averages), but also that theshort-term variation in the one variable creates too much “noise,” whichblurs the association between the variables. For example, assume paneldata with a time-constant, cross-sectionally varying institutional vari-able and an economic variable that fluctuates around some equilibriumvalue which varies across countries. Mapping this variable on the time-constant variable will allocate the averages of the former to the latter buttreat the deviations from the unit-specific means as errors to be allocatedto the standard error. If the time series overlap to a considerable degree,coefficients for the effect of the institutional variable on the economicvariable will be unlikely to become statistically significant because of thehuge standard errors in comparison to the cross-sectional variation inaverages captured by the institutional variable. Even in the case of a per-fect match of the cross-sectional variation, the overlap in variation mayyield insignificant results. If, in addition, the variable is nonstationaryfor a few countries, neither the differences in means of the economicvariable across countries nor the variance of the estimated coefficientcan be interpreted.

One could argue that for this reason control variables that deal withthe short-term variance components have to be added. This is true,but seldom solves the problem, since in many practical research sit-uations, adding substantive controls neither removes nonstationarityin annual data nor accounts for serial correlation in the error term,both of which invalidate regression results. Hence models need to betransformed, thereby changing the substantive research question froma cross-sectional one to a dynamic one. This implies that cross-sectionalvariation in levels cannot be assessed anymore.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 38


In the following, I discuss possible narratives about the associationbetween the replacement rate of unemployment and employment in thelow-wage sector (private-sector consumer services) that can be extractedfrom the data set. The data are a typical example of a macrocompara-tive setup, with a handful of countries (for the variables I use, data on14 countries are available) and another handful of annual observations(due to missing values, the number of years included varies considerablyacross countries, but there is a core of 11 years, 1981–91, for which practi-cally all relevant variables are available for all 14 countries). I first explorethe main variables (section 2.2) and then proceed with a specificationsearch, thereby following some leads of Leamer (1978) (section 2.3).

2.3 Longitudinal properties of political economic variables

Employment in the low-wage sector

Employment in the low-wage sector is operationalized as the percent-age of the population aged 15 to 64 that is employed in private-sectorconsumer-oriented services – ISIC6 (wholesale and retail trade, restau-rants, and hotels) and ISIC9 (community, social, and personal services).2

Employment in private consumer services is the most appropriate vari-able to study the effect of labor market institutions on employment,because the expected associations will be least blurred by interveningfactors (Iversen and Wren 1998; Scharpf 2000).

Panel (a) of Figure 2.1 shows the development of this variable overtime. For most countries we observe a gradually and smoothly growingprocess, although some appear more stationary. This is a typical patternof time series of macroeconomic data measured in levels: The series fordifferent countries enter the observation period at different levels andthe differences between the countries change only marginally each year,though over longer periods these minor differences in average growthadd up to more substantial variation. As Table 2.1 shows, the bulk of thevariation is between the countries, not within.

In the time dimension, panel (a) of Table 2.2 reveals that in a fixed(country) effects specification, the average autoregression coefficient inthe countries cannot be distinguished from unity, suggesting nonstation-arity of the series. Nonstationarity implies that the mean of the residualscannot assumed to be zero and that the variance of the residuals tendsto infinity instead of remaining constant over time, resulting in invalidparameter estimates.

This is indeed confirmed by Maddala and Wu’s Fisher test for nonsta-tionarity (Maddala and Wu 1999), irrespective of which of a set of

9780230_202573_03_cha02.tex

5/6/200815:34

Page39

39

(a) Employment (b) Reservation wage

510

1520

2530

Em

ploy

men

t in

priv

ate

sect

or c

onsu

mer

ser

vice

s

1980 1985 1900 1905Year

1980 1985 1900 1905Year

1980 1985 1900 1905Year

1980 1985 1900 1905Year

Employment in private sector consumer services,14 countries, 1980–95

0

Rep

lace

men

t rat

e

Replacement rate, 14 countries, 1980–95

(c) Pay equality (d) Employment regulation

.4.5

.6.7

.8

24

68

10

24

68

Pay

equ

ality

Pay equality, 14 countries, 1980–95

Em

ploy

men

t reg

ulat

ion

Employment regulation, 14 countries, 1980–95

Figure 2.1 Main variables

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 40


Table 2.1 Cross-sectional and time variance components of main variables

Variable Mean Std. Dev. Min Max Observations

epscs overall 16.31 5.44 8.6 28.9 N = 189between 5.28 9.61 25.83 n = 14within 1.32 12.39 20.29

replr overall .50 .23 .01 .92 N = 189between .23 .071 .90 n = 14within .05 .30 .74

payeq overall .62 .08 .41 .76 N = 189between .09 .42 .75 n = 14within .01 .58 .66

empreg overall 3.70 2.39 0 7 N = 189between 2.45 0 7 n = 14within 0 3.70 3.70

Table 2.2 Autoregression

1 2 3 4Coefficient Standard Error F-Test r(ui, Xb)

a) EPSCS 0.991 0.025 0.058 0.772b) REPLR 0.737 0.020 0.000 0.962c) PAYEQ 0.697 0.038 0.000 0.984

EPSCS: Employment in Private-sector Consumer ServicesREPLR: Replacement RatePAYEQ: Pay EqualityColumn 1: Coefficients are those of the lagged dependent variable in a fixed unit effectsspecification.Column 2: Associated standard errors.Column 3: F-Test that the fixed unit effects are irrelevant.Column 4: Correlation between lagged dependent variable and unit effects.

potentially plausible assumptions is imposed (Table 2.3). Most notably,neither controlling for a country-specific trend nor demeaning the series(subtracting the cross-country period mean from all observations in orderto eliminate a joint trend or drift) solves the problem. Hence, accordingto the tests, we cannot reject nonstationarity of the series.

This, however, is an implausible conclusion in substantive termsbecause the variable is measured as a percentage, which is bounded by0 and 100. There are four main substantive possibilities for a series to benonstationary – ever-growing (many economic aggregates for extendedperiods of time), drifting randomly (some macroeconomic series), by

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 41

Bernhard Kittel 41

Table 2.3 Nonstationarity: Maddala and Wu’s Fisher Test

EPSCS REPLR PAYEQ

Nonstationarity 5.12 53.55∗∗∗ 25.64Nonstationarity assuming 1 lagged difference 25.37 32.92 25.98Nonstationarity assuming group-specific trend 11.33 45.90∗∗ 76.10∗∗∗Nonstationarity assuming group-specific drift 29.45 84.23∗∗∗ 71.98∗∗∗Nonstationarity, period-demeaned data 7.55 51.12∗∗∗ 76.79∗∗∗

Entries are Chi-squared values (28 dF) of joint test H0: all series are nonstationary against H1:at least one series is stationary.∗∗∗p-value < 0.01, ∗∗p-value < 0.05. The test does not require a balanced panel. See Maddala& Wu (1999).Calculated with Stata module xtfisher.

construction (e.g., cumulative indices), and because of data limitations(the series is too short relative to the real-world process from which thedata are drawn to capture the stationarity of the process). Since the firstthree do not apply because the series can neither grow nor fall forever bydefinition, we have to conclude that we are confronted with a situationin which the observation window is too short to adequately observe upsand downs around some long-term average. I will come back to this issuebelow.

A final observation relates to the issue of persistence in between-country variation over a longer period of time. Pearson’s correlationcoefficient between employment in private-sector consumer services in1991 (the last year for which data for the whole 14-country sample is con-tained in the dataset) and its average 1974–79 is r = 0.95, and Spearman’srank-order correlation is ρ = 0.97. This confirms the visual impressionfrom Figure 2.1a that the major part of the between-country differenceswas already present before the observation period and did not change inrelevant ways during that period.

Reservation wage

The reservation wage is operationalized as the percentage of former earn-ings that is replaced by unemployment compensation for a worker at the33rd percentile in the first year after losing the job. Even more than foremployment, most of the variation is between the countries. As panel(b) in Figure 2.1 shows, these series are either fairly constant over timeor exhibit one or a few jumps resulting from policy changes. Given thecombination of longer constant periods with a few irregular jumps, the

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 42


autoregressive coefficient is reasonably far away from unity, and theFisher test tends to reject the null hypothesis of nonstationarity in favorof the alternative that at least one of the series is stationary. This conclu-sion is illustrated by the visual inspection of Figure 2.1b, in which someof the series trend into one direction while others tend to be constantover time. Hence the situation is indeterminate.

Pay equality

Pay equality, which is measured by the ratio of gross annual earningsof a full-time, year-round employed person at the 10th percentile of theearnings distribution to a person at the 50th percentile, again basicallyvaries between countries, with a ratio of within to between variation ofabout 1:7. The series appear fairly stable, although some have a slightlydeclining trend. Formal analysis suggests that nonstationarity can berejected, though – as can be expected from inspection of the graph –considerable autoregression is present.

Employment regulation

Finally, employment regulation is an index constructed from evaluationsof the strictness of legislation in the areas of working time, fixed-termcontracts, employment protection, minimum wages, and employee’srepresentation rights. Since the country scores on these dimensionsdid not change during the observation period, this is a time-constantvariable, and the analysis of its time series properties is moot.

2.4 Statistical narratives on employment in private sectorconsumer services

Pooled analysis

As noted in the previous section, the upper limit of employment in pri-vate sector consumer services will be much lower than 100 percent sincenot all workers can be employed in that sector. According to the reser-vation wage hypothesis, the extent to which this upper limit is reachedin a country is affected by the incentives for accepting a job at the wagelevel offered in this sector. Hence the existence of systematic variationbetween countries in this assumed upper limit is the core propositionof the reservation wage hypothesis: If workers prefer unemployment tolow-wage work because of a high replacement rate, two implicationsresult: (1) fewer workers will be searching for jobs in the low-wage sector,and (2) fewer jobs will be offered because of the higher wages employers

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 43

Bernhard Kittel 43

AUAUAUAUAU

AUAU

AU

AUAUAUAUAU

AUAUAU

BEBEBEBEBEBEBEBEBEBEBEBEBEBEBE

BE

CA

CACACACA

CA

CACACA

CACACA

CACACACA

DKDKDKDKDKDKDKDKDKDKDKDKDKDKDKDK

FIF IF IFIFI FIFI FIFIFIFIF I

F IFIFI FI

FRFRFRFRFRFRFRFRFRFRFRFRFRF RF RFR

D EDEDEDEDEDEDEDEDEDEDEDE

DEDE

ITITITITIT

ITITITITITIT

ITIT ITIT IT

JPJPJPJPJP

JPJPJPJPJPJPJPJPJPJPJP

N LN LN L NLNLNLNLNLNLNL

NLNLNLNLNLNL

NO NONONONONONONONONONONO

SESESESESESESESESESESESESESESE

UKU KU KUK

UKUKUK

UK

UKUKUKUK

UK

UKUS USUSUS

USUSUS

USUSUSUSUSUSUSUSUS

510

1520

2530

Em

ploy

men

t in

priv

ate

sect

or c

onsu

mer

ser

vice

s

0 .2 .4 .6 .8 1

Replacement rate

Employment in private consumer services

Figure 2.2 Replacement rate and service sector employment

would have to pay in order to exceed the reservation wage. Both mecha-nisms should result in a negative relationship between the replacementrate and employment in the low-wage sector.

Figure 2.2 plots this effect for the whole data set. It reveals that theexpectation is indeed confirmed. However, the graph also reveals somedisturbing elements. Firstly, the observations are clearly clustered bycountry, indicating that between-country variation in both variables isindeed much larger than within-country variation. Secondly, there isan important outlier in the lower left corner of the plot, Italy, whichcombines low replacement rates with low private services employment.Thirdly, the overall pattern of the plot can be interpreted in two ways. Onthe one hand, including Italy, there seems to be a wedge-shaped curveindicative of heteroskedasticity. On the other hand, if Italy is excludedfrom consideration the pattern suggests a convex shape indicative of anonlinear relationship.

Actually, since employment in private-sector consumer services isdefined as a proportion (population employed in the low-wage sectoras a percentage of the total population in working age), one of the basicassumptions of linear regression analysis, unboundedness of the depen-dent variable, is violated. Instead, we have to expect a curve in which thelower and upper limits are approached at an ever-decreasing rate, while

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 44


the slope is steeper at average values. But since the logit function trans-forms the bounded range into an unbounded one, the linear regressionmodel can still be used. Although the logit model is preferable for sub-stantive reasons, it is an empirical question whether the linear approx-imation, which comes close to the logit for values sufficiently close tothe mean, yields biased estimates for the range of values covered by theexplanatory variables. In the present case, the logit transformation doesnot discernibly differ from the linear specification, presumably becausethe scores do not get close enough to the extremes for causing a devia-tion. Hence, although strictly speaking a nonlinear model would be moreappropriate, I will stay with the crowd and remain in the linear world.

The grouping problem confirms the finding from univariate analysisthat autocorrelation is a serious issue. In substantive terms, the plot sug-gests that there is little information gained from adding observations forthe same countries. Because of the country clusters, new observationsfrom a country will be close to the existing ones and thereby simplyaffirm the stability of the two variables for that country without addingsubstantively new information. This can be seen by comparing the coef-ficient of the replacement rate in the pooled model (Table 2.4, model 1),which is a weighted average of the between and within variation, tothe one in the fixed-effects model (Table 2.4, model 2), which relatesexclusively to the within dimension: It drops from –10.8 to –2.6 andbecomes statistically insignificant. Moreover, the within component ofthe coefficient of determination is only 0.01, and in the dummy vari-ables specification of the fixed effects model (LSDV) this parameter soarsto 0.94 due to the contribution of the country dummies.

Durbin’s M test for autocorrelation, which consists of regressing theresiduals on the lagged residuals and all regressors of the initial model,indicates the potential presence of a unit root. If autocorrelation weredue to the clustered pattern of the observations, including fixed effectswould remove the problem. This is not the case here, however, sincethe autocorrelation coefficient drops only to just below 1.0, not beingsignificantly different from unity. If the country series reveal a joint pat-tern over time, nonstationarity can sometimes be removed by includingperiod effects (Table 2.4, model 3). In the present case, however, this is nosolution, given that the coefficient of the lagged residuals remains above1.0. Moreover, the period effects appear to be nonsignificant, suggestingthat there is no reason to include them.

The observation of nonstationarity in these specifications suggests todirectly shift to first difference models. However, as has been arguedabove, there is no substantive reason for nonstationarity in private

9780230_202573_03_cha02.tex

5/6/200815:34

Page45

45

Table 2.4 Service employment and replacement rate: panel models

1 2 3 4 5 6 7 8Pool FE(c) FE(t) AR LDV FD FD-LDV FD-LDV-FE(c)

Replacement Rate −10.77∗∗∗ −2.63 −10.58∗∗∗ −2.20∗ −0.44∗∗∗ −0.61∗∗∗ −0.36∗∗∗ −0.69(1.52) (1.96) (1.58) (1.21) (0.15) (0.14) (0.14) (0.72)

Lagged EPSCS (D.EPSCS) 1.02 0.43∗∗∗ 0.34∗∗∗(0.01) (0.07) (0.07)

Constant 21.77∗∗∗ 17.65∗∗∗ 17.39∗∗∗ 0.13 0.48∗∗∗ 0.30 0.47(0.85) (1.00) (1.21) (0.16) (0.08) (0.08) (0.37)

R2 0.21 0.21 0.22 0.66 0.99 0.10 0.27 0.25AR Test (Durbin’s M) 1.01∗∗∗ 0.98∗∗∗ 1.01∗∗∗ 1.02∗∗∗ 0.36∗∗∗ 0.39 −0.02 0.07

(0.008) (0.03) (0.01) (0.01) (0.07) (0.07) (0.17) (0.21)F-Test – 167.76∗∗∗ 0.14 – – – – 1.17

Dependent variable: Models 1–5: EPSCS, Models 6–8: first differences of EPSCS (D.EPSCS)Model 4: Panel Corrected Standard ErrorsF-Test: Joint test that fixed effects are zero.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 46


service employment and its indication seems to be more a problem ofdata shortage. Hence let us first explore approaches to controlling forautocorrelation. Applying the Prais-Winston transformation (Table 2.4,model 4) does not yield a satisfactory result: The estimated autocorre-lation coefficient is above unity and has to be forced to a value below1.0 (0.97 in the present case) and this procedure results in an estimateof residual autocorrelation above unity. An alternative approach is toinclude a lagged dependent variable (Table 2.4, model 5), the coefficientof which also turns out to be above unity. Thus, none of the conventionalapproaches to correcting for observation patterns typical of panel datayields a parameterization which removes nonstationarity and residualautocorrelation. Instead, both yield models that cannot be interpretedbecause of a meaningless estimate of the mean and variance of the depen-dent variable. In consequence, the next step is to take first differences.Since there are no indications that the replacement rate has a unit root,it is only necessary to transform employment in private sector consumerservices.

In substantive terms, the possibility to maintain the replacement ratein levels is a fortunate situation because the values this variable takes donot result from atomistic market outcomes, but are set by laws that arebased on political decisions. The difference between these two classes ofvariables in aggregate analysis relates to the expectable time horizon ofchange. Market outcomes, which are based on a multitude of individualdecisions, tend to respond flexibly and fairly quickly to changing con-ditions, while political decisions have a longer time horizon and thusremain constant at certain levels for longer periods. We can thus analyzethe effects of the levels of the replacement rate on growth of employmentin private-sector consumer services. Figure 2.3 presents the plot for thisassociation. We observe the country clusters again, as well as a slight neg-ative relationship, indicating that countries with lower replacement ratestend to have higher growth of employment in private sector consumerservices. Statistically we still observe a negative effect (Table 2.4, model6), which remains present after controlling for residual autocorrelation(Table 2.4, model 7). Including fixed effects (Table 2.4, model 8) destroysthis association, but since the fixed effects are not jointly significant, wecan remove them again and provisionally settle with model 7. Thus, forthe time being we conclude that there is a modest, but statistically sig-nificant association between the replacement rate and annual growth inemployment in private-sector consumer services.

Figure 2.4 reiterates this idea from a more narrative perspective. Itshows the long-run shifts on the two variables over the period 1982–91

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 47

Bernhard Kittel 47

AU

AU

AU

AU

AU

AUAU

AU

AU

AU

AUAU

AU

AUAU

BE

BE

BEBE

BEBE

BEBE

BE

BEBE

BE

BE

BE

BE

CA

CA

CA

CA

CA

CA

CA

CA

CA

CA

CA

CA

CACACA

DK

DKDKDK

DKDK

DKDKDKDKDKDK

DKDK

DK

FI

FI

FIFI FI

FI

FIFIFI

FI

FIFI

FI

FI

FI

FRFR

FRFRFR

FRFRFRFRFR

FRFRFR

FRFR

DE

DEDE

DE

DEDE

DE

DEDE

DE

DEDE

DEIT

IT

IT

ITIT

IT

ITIT

ITIT

IT

IT

ITIT IT

JPJP

JP

JPJP

JPJP

JP

JP

JPJP

JPJP

JPJP

NL

NL

NL

NL

NLNLNLNLNLNLNLNL

NLNL

NL

NONONONO

NONO

NO

NO

NO

NO

NOSE

SE

SESE

SESESESE

SE

SE

SE

SE

SE

SE

UK

UKUK

UKUK

UK

UK

UKUK

UK

UK

UK

UK

US

US

US

US

US

US

USUS

US

USUS

US

US

USUS

�2

�1

01

2

Ann

ual c

hang

e in

ser

vice

em

ploy

men

t

0 .2 .4 .6 .8 1

Replacement rate

Employment in private consumer services

Figure 2.3 Annual changes in service employment

and tells an even more nuanced story. If the hypothesized negative rela-tionship between the replacement rate and private service employmentwere a general regularity that is independent of space and time, we wouldobserve only movements from the upper left to the lower right or viceversa: a decline in the replacement rate would induce an increase inemployment and an increase in the replacement rate would lead to adecline in employment. This pattern is clearly confirmed for the UKand the US, while France and Belgium tend toward the same direction.However, Norway and Finland have increased the replacement rate with-out noticeable effect on employment; in Germany, Canada, Denmark,and Sweden employment slightly increased despite the absence of notice-able change in the replacement rate; and in Italy, the Netherlands, Japan,and Australia, an increase in the replacement rate coincided with anincrease in employment. This means that only four out of the 14 coun-tries confirm the assumed relationship with respect to changes occurringduring the observation period.

In terms of the original hypothesis, however, the need to control fornonstationarity in the dependent variable has induced a respecificationof the substantive model by focusing on the dynamics in employ-ment in private-sector consumer services. Under the conditions of thepresent data set, the time dimension does not add substantively relevant

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 48


AU

BE

CA

DKFI FR

DE

IT

JP

NL

NO

SE

UK

US

AU

BE

CA

DKFIFR

DE

IT

JP

NL

NO

SE

UK

US

1015

2025

30

Em

ploy

men

t in

priv

ate

sect

or c

onsu

mer

ser

vice

s

0 .2 .4 .6 .8 1

Replacement rate

Employment in private consumer services 1982–1991

Figure 2.4 Long-time changes, 1982–91Notes: Small fonts refer to 1982, large fonts to 1991.

information about differences in levels. The time horizon of changeof the focused regressor – decades – in relation to the measurementperiodization in the current example – years – explains the empiricalclustering of the annual observations on the horizontal axis of Figures2.2 and 2.3. The clustering in the vertical dimension is due to auto-correlation in both instances. The major part of the variation in thedependent variable was already present at the beginning of the observa-tion period and hence the observed dynamics cannot have affected thosedifferences.

Cross-sectional analysis

Going back to the initial hypothesis, we thus have to modestly stick toa cross-sectional design, as is appropriate for a cross-sectional researchquestion (Jackman, 1985). Models 1 and 2 of Table 2.5 report simplecross-sectional analyses for 1981 and 1991, the first and last periodsfor which data for all 14 countries are available. While the model for1981 is a simple cross-section, the regressors in the model for 1991 aredefined as averages for 1981–91. Note that the size of coefficients inthese snapshots is consistent with a steadily growing coefficient when

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 49

Bernhard Kittel 49

Table 2.5 Employment in private sector consumer services, cross-sectionalanalysisModel 1: Cross-section model, 1981.

1 2 3 4 51981 1991 1991 1991 1991

(excl. IT)

REPLR −4.01 −8.90 −18.99∗∗∗ −6.11 −24.26∗∗∗(6.79) (8.09) (4.84) (6.61) (7.32)

EMPREG −1.40∗∗ −2.96∗∗∗(0.51) (0.79)

REPLR × EMPREG 3.79∗∗∗(1.11)

Constant 16.77∗∗∗ 21.17∗∗∗ 27.45 25.08 32.47∗∗∗

(4.32) (5.24) (3.14) (3.49) (5.01)

R2 0.05 0.15 0.54 0.50 0.66N 14 14 13 14 14

Note: Robust standard errors in parentheses.

the cross-sectional regression is repeated for every year. This is causedby the increasing variation in the dependent variable (see Figure 2.1a).The negative association is present in both years, but does not attainstatistical significance. In order to keep the argument as crisp as possible,the further analysis will be done on the 1991 data, because the coefficientsize is largest of the two and hence effects will be the most pronounced.We use averages 1981–91 for the regressors in order to reduce the effect ofpossible idiosyncratic deviations in 1991 from the overall developments.We noted that Italy was an outlier and, indeed, removing this observa-tion doubles the size of the coefficient and yields a highly significanteffect, even for 13 observations (Table 2.5, model 3).

This leads back to the issue of substantive reasons for heteroscedastic-ity. One important potential reason is that another, intervening variablecauses variation on one side of a two-dimensional distribution to belarger than on the other side. In the present case, there is an obviouscandidate for such a conditional effect: employment regulation. In linewith the above reasoning, we should expect that the more regulatedemployment conditions are, the less will employers be inclined to createjobs in the low-wage sector. Model 4 in Table 2.5 adds this variable andreveals that it has a clear negative effect on employment. At the sametime, controlling for employment regulation does not cause the effect ofthe replacement rate to become significant.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 50


AUAUAUAUAUAUAUAU

AUAUAUAUAUAUAU

BEBEBEBEBEBEBEBEBEBEBEBEBEBEBE

CA

CACACACA

CA

CACACA

CACA

CACACACA

DKDKDKDKDKDKDKDKDKDKDKDKDKDKDK

FIFIFIFIFIFIFIFIFIFIFI

FIFIFIFI

FRFRFRFRFRFRFRFRFRFRFRFRFRFRFR

DEDEDEDEDEDEDEDEDEDEDEDEDE

ITITITITITITITITITITITITITITIT

JPJPJPJPJPJPJPJPJPJPJP

JPJPJPJP

NLNLNLNLNLNLNLNLNLNLNLNLNLNLNLNONONONONONONONONONONO

SESESESESESESESESESESESESESE

UKUKUKUKUKUKUKUKUKUKUKUK

UKUSUSUSUSUSUSUSUSUSUSUSUSUSUSUS

AU

BE

CA

DK FI FR

DE

IT

JP

NLNO

SE

UK

US

510

1520

2530

Em

ploy

men

t in

priv

ate

sect

or c

onsu

mer

ser

vice

s

0 2 4 6 8Employment regulation

Employment in private consumer services 1991

Figure 2.5 Employment regulation and private service employmentNote: Large fonts refer to 1991.

Figure 2.5 depicts the bivariate relationship between employment reg-ulation and private service employment. Comparing this graph to theone for the long-term developments in the replacement rate (Figure 2.4),note that the US and the UK, which both most clearly confirm theexpected relationship, score zero on the scale of employment regulation,while Italy scores highest. Hence these bits of information do indeedsuggest the potential presence of a conditional effect. If employmentregulation is high, the effect of the replacement rate on private ser-vice employment may be mitigated because strict regulations discourageemployers to offer jobs in that sector anyway. Thus while lowering thereplacement rate may push workers on the market and reduce the wagerate at which they are willing to take jobs, employers will still be reluc-tant to demand labor for reasons related to the strictness of regulationsand thereby cause employment to remain low. However, if employmentregulations are lax, the reservation wage will become the crucial factorstructuring the labor market. In consequence, labor supply will be theprime mover and changes in the replacement rate will have the expectedeffect on employment.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 51

Bernhard Kittel 51

AU

BE

CA

DK

JP

UK

US

FI FR

DE

IT

NLNO

SE

�10

010

2030

40

Em

ploy

men

t in

priv

ate

sect

or c

onsu

mer

ser

vice

s

0 .2 .4 .6 .8 1

Replacement rate

Figure 2.6 Private service employment: replacement rate effect conditional onemployment regulationsNotes: Thick line: Effect of Replacement Rate if Employment Regulation set to Maximum,with 95% CI.Thin line: Effect of Replacement Rate if Employment Regulation set to Minimum, with95% CI.Large country markers: Employment Regulation >4Small country markers: Employment Regulation < = 4.

In terms of the time horizon of change, employment regulations areeven less volatile than replacement rates because they form part of theinstitutional framework of the labor market, which is even more difficultto change than the proportion of previous wages that is paid in the firstyear of unemployment.

This conditional effect is indeed clearly present (Table 2.5, model 5).All coefficients are highly significant despite the small number of obser-vations and the specification accounts more than 60 percent of thevariation in private service employment. Figure 2.6 shows how well thewedge shape in the association between the replacement rate and privateservice employment is captured. The steeper regression slope representsthe effect of the replacement rate on employment if employment regu-lation scores the minimum, with the associated 95 percent confidenceinterval (dashed curves). The other, almost horizontal line represents

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 52


this effect if employment regulation scores the maximum, again with the95 percent confidence interval. The seven countries scoring below themedian on employment regulation (hence low labor market rigidity)are printed in small font; those scoring above the median (hence highrigidity) are printed in large font.

So far, we have not dealt with other factors that can potentially dis-turb the relationship. Let us introduce a few such variables now andtest their impact on the stability of the conditional association foundabove. For ease of comparison, Table 2.6 reproduces model 5 of Table2.5 in model 1. First, according to the incomes–jobs tradeoff hypothe-sis, pay equality is the core factor that makes labor at the low end ofthe distribution relatively more expensive and hence has a discouragingeffect on labor demand. Note, however, that this variable only affectsthe demand side: if at all, wage equality should have a positive effecton labor supply since fewer jobs will be offered at extremely low-wagelevels. The effect (Table 2.6, model 2) is quite impressive in the bivariatemodel – the estimate for the employment difference between minimumand maximum pay equality, based on the coefficient, is 11.6 percentagepoints. Model 3 in Table 2.6, however, clears the issue by showing thatthe pay equality effect drops to almost one-quarter which is statisticallyinsignificant when controlling for the joint effect of the replacement rateand employment regulations.

Secondly, public employment may be an important substitute for atleast some of the jobs offered in consumer services, in particular thosein community, personal, and social services (ISIC 9), and thereby crowdout the private sector. The crucial point of the argument relating to theimportance of public employment is that the reason for low levels ofemployment in private services is not labor market distortions, but sys-tem differences in the provision of such services. For example, ISIC 9contains health services and education. To the extent that such ser-vices are considered a public good to be provided by the state, lowprivate employment in cross-country perspective is due to the fact thatthere are no private employers by design of the system, not necessarilybecause of labor market distortions. An indicator of public employmentin consumer services can be constructed by subtracting private serviceemployment from total service employment (ISIC 6 + ISIC 9). There isindeed a substantial negative correlation between public and private ser-vice employment (r = −0.68), suggesting that there may indeed be asubstitution effect. Moreover, there is a positive correlation of the sameorder of magnitude (r = 0.73) between public employment in the servicesector and the replacement rate. These associations are much stronger

9780230_202573_03_cha02.tex

5/6/200815:34

Page53

53

Table 2.6 Controlling for other factors

1 2 3 4 5 6 7 8

REPLR −24.26∗∗∗ −22.55∗∗ −12.77 −15.96 −9.45(7.32) (7.95) (8.31) (11.01) (10.15)

EMPREG −2.96∗∗∗ −2.66∗∗ −2.46∗∗∗ −2.38∗∗ −2.23∗∗(0.79) (0.93) (0.67) (0.99) (0.94)

REPLR × EMPREG 3.79∗∗∗ 3.63∗∗∗ 3.03∗∗∗ 2.81∗ 2.58(1.11) (1.11) (1.22) (1.45) (1.59)

PAYEQ −37.45∗∗∗ −10.82 1.71(11.78) (9.40) (12.86)

PUBEMP −0.66∗∗∗ −0.51∗ −0.46(0.18) (0.25) (0.32)

GDP Growth 6.05∗∗∗ 2.85 1.57(1.71) (2.89) (2.56)

Constant 32.47∗∗∗ 40.21∗∗∗ 37.57∗∗∗ 27.28∗∗∗ 34.47∗∗∗ 1.31∗∗∗ 20.84 27.07∗(5.01) (7.86) (4.59) (3.34) (3.32) (3.58) (13.15) (12.51)

R2 0.66 0.37 0.68 0.44 0.79 0.43 0.71 0.80N 14 14 14 14 14 14 14 14

Dependent variable: Employment in Private-sector Consumer Services (EPSCS).Robust standard errors.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 54


than the one between the replacement rate and private service employ-ment (r = −0.43). Hence we may indeed be confronted with a potentiallydisturbing effect. In Table 2.6, model 4 gives the effect size of publicservice employment and model 5 confirms the confounding impact ofthis variable: While public service employment is marginally significant,the coefficient of the replacement rate is halved and no longer attainsstatistical significance.

Thirdly, low employment in private services may be due simply to alack of economic dynamism. As models 6 and 7 (Table 2.6) reveal, thereis indeed a clear positive cross-country association between GDP growthand private service employment, accounting for 42 percent of the varia-tion, but after controlling for the conditional effect of the replacementrate and employment regulation, this association is halved. However, itweakens the conditional effect considerably.

Finally, model 8 adds all controls jointly and thereby causes all effectsto become insignificant, although all coefficient signs except for payequality maintain their (expected) sign and the overall variation in pri-vate service employment captured by the model is 80 percent. Thediscrepancy between the coefficient of determination and the lack ofsignificant individual effects suggests the presence of multicollinearitywhich, in turn, must be mainly attributed to the small number of cases.

Thus far, the analysis has revealed a fairly stable relationship. How-ever, as noted above, we are dealing with levels, the variation of whichwas already present at the beginning of the observation period. Hencewe cannot attribute a causal status to the replacement rate when analyz-ing levels. But this issue can be remedied by focusing on the long-termdevelopment 1981–91 (see Figure 2.4). The change in private serviceemployment from 1981 to 1991 can be attributed to the conditionsprevalent during this period because they are not hampered by the per-sistence in the level differences. In contrast to the short-term variationanalysed in model 8 of Table 2.4, the relationship will be less affected bynoise. Hence, in the models presented in Table 2.7 the change in employ-ment in private sector consumer services 1981–91 is the dependentvariable. All regressors are again defined as the average 1981–91.

The setup of the table is basically identical to that of Table 2.6. What ismost striking about these models of long-term change is that they appearto show even more clearly the conditional relationship, which accountsfor over 80 percent of the variation in the private service employmentin the simple specification of model 1. All of the controls introduced inmodels 2 to 9, apart from the long-term change in public employmentin ISIC 6+9, appear to be insignificant if controlled for the conditional

9780230_202573_03_cha02.tex

5/6/200815:34

Page55

55

Table 2.7 Explaining variation in long-term change of private service employment (1981–91)

1 2 3 4 5 6 7 8 9 10 11

REPLR −7.52∗∗∗ −7.89∗∗∗ −6.04∗∗ −5.22∗∗∗ −8.40∗∗∗ −7.20∗∗ −5.82∗(1.67) (1.48) (2.33) (1.24) (2.53) (2.28) (2.58)

EMPREG −0.66∗∗∗ −0.73∗∗∗ −0.60∗∗∗ −0.41∗∗ −0.72∗∗∗ −0.80∗∗∗ −0.49∗(0.12) (0.16) (0.14) (0.14) (0.20) (0.19) (0.26)

REPLR × EMPREG 0.76∗∗∗ 0.79∗∗∗ 0.66 0.42∗ 0.86∗∗ 0.84∗ 0.49(0.27) (0.23) (0.39) (0.22) (0.35) (0.37) (0.37)

PAYEQ −8.76 −2.38 5.22∗ 1.61(5.91) (2.99) (2.24) (2.24)

PUBEMP −0.20∗∗∗ −0.07 −0.13∗(0.04) (0.05) (0.06)

D.PUBEMP −0.60∗∗∗ −0.29∗ −0.28(0.12) (0.14) (0.16)

GDP Growth 1.45∗∗ −0.30 −0.59 −0.090.55 (0.76) (0.61) (0.55)

Constant 6.68∗∗∗ 7.42∗ 5.56∗∗∗ 5.19∗∗∗ 6.94∗∗∗ 2.71∗∗∗ 5.64∗∗∗ −1.76 7.92∗∗ 7.13∗(0.77) (3.95) (1.41) (0.83) (0.83) (0.32) (0.70) (1.25) (2.93) (3.05)

R2 0.81 0.24 0.82 0.50 0.84 0.61 0.90 0.29 0.82 0.89 0.90N 14 14 14 14 14 14 14 14 14 14 14

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 56


AU

BE

CADK

JP

UKUS

FI

FR

DEIT

NL

NO SE

�5

05

10

Cha

nge

empl

oym

ent i

n pr

ivat

e se

ctor

con

sum

er s

ervi

ce

0 .2 .4 .6 .8 1

Replacement rate

Figure 2.7 Change in private service employment, 1981–91: replacement rateeffect conditional on employment regulationsNotes: Thick line: Effect of Replacement Rate if Employment Regulation set to Maximum,with 95% CI.Thin line: Effect of Replacement Rate if Employment Regulation set to Minimum, with95% CI.Large country markers: Employment Regulation >4Small country markers: Employment Regulation < = 4.

model of the replacement rate and employment regulation. Moreover,they do not affect this model since the basic specification remains statisti-cally significant in all of these alternative specifications. Note, however,that if one controls for all variables jointly (models 10 and 11), oneobtains statistically significant coefficients for all variables, except foreconomic growth, and the coefficient of determination almost reaches90 percent, a level seldom attained in cross-sectional models of this kind.Figure 2.7 visualizes the relationship and underlines the better fit of thelong-term change model in comparison to the levels model.

The cross-sectional models impressively support the hypothesis thatlabor market policies have an impact on employment in privateconsumer services. The results suggest that a clear-cut effect of thereplacement rate on employment in private consumer services exists as

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 57

Bernhard Kittel 57

soon as one controls for the dependence of this effect on the conditionthat employment regulations do not interfere with the labor market. Notonly cross-country differences in levels, but – more importantly – alsocross-country variations in the change 1981–91 are well predicted by theconditional model.

A brief return to the pooled model

Are we also able to observe this conditional effect in the pooled model?Since conventional comparative political economy would rely more onthe findings from that design than from the cross-sectional design dis-cussed above, this is a critical question. Model 8 in Table 2.4 has revealedthat first-differencing the dependent variable has resulted in insignifi-cant country effects. This result allows us to drop the fixed effects andto study the impact of employment regulation, which is a time-constantvariable, on the employment dynamics. Hence in this section, we analysepooled models of first differences in employment in private sector con-sumer services for the period covered by the data, using panel-correctedstandard errors which are the panel equivalent of the robust standarderrors used in the cross-sectional models.

In Table 2.8, model 1 is the baseline specification. There are noindications that the conditional model is valid, given that the interac-tive term is not significant. However, annual data contain much morenoise due to short-term deviations than a cross-sectional design aver-aging over a longer period. Controlling for the factors that Kenworthy(2003) introduced (model 2) indeed improves the situation considerably,although many of the controls fail to attain statistical significance. Analternative is model 3, which replaces the level of public employmentin consumer services by the first difference, which is significant. Therelevance and disturbing impact of this factor can be seen from a com-parison of model 4 and 6 on the one hand, and model 5 on the other.Including the change in public employment in ISIC 6 and 9 halvesthe focused coefficients of the conditional effect of the replacementrate and employment regulation and renders the conditional elementof the model statistically insignificant. Hence, one would either go fora non-conditional specification including the effect of public employ-ment or sacrifice 10 percentage point of the overall fit by excluding publicemployment in order to obtain a confirmation of the conditional model.Apart from this ambivalence, the pooled reanalysis controlling for someeconomic factors broadly confirms the findings from the cross-sectionaldesign.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 58

58

Table 2.8 Pooled model, first differences

1 2 3 4 5 6

REPLR −0.65 −0.96∗ −0.66∗ −1.20∗∗ −0.55∗∗ −1.05∗∗(0.43) (0.50) (0.36) (0.59) (0.28) (0.46)

EMPREG −0.06∗∗ −0.14∗∗ −0.10∗∗ −0.12∗∗ −0.06∗∗∗ −0.09∗∗∗(0.02) (0.06) (0.04) (0.06) (0.02) (0.03)

REPLR × EMPREG 0.08 0.22∗∗ 0.13 0.17∗ 0.07 0.15∗∗(0.07) (0.09) (0.09) (0.09) (0.04) (0.07)

PAYEQ 2.13 1.72 0.49(1.38) (1.15) (1.13)

GDPGROW 0.17∗∗∗ 0.19∗∗∗ 0.13∗∗∗ 0.16∗∗∗ 0.12∗∗∗(0.03) (0.03) (0.04) (0.02) (0.03)

TRADE −0.08 −0.04 0.13∗ 0.19∗∗ 0.16∗∗∗(0.09) (0.84) (0.07) (0.08) (0.06)

RLTIR 0.02 0.01 0.03∗ 0.03∗(0.02) (0.02) (0.02) (0.02)

ALMP 0.09 0.07(0.07) (0.07)

UNEMP BENEF −0.02 0.01(0.04) (0.03)

LEFTGOV −0.09 −0.07(0.10) (0.11)

WAGECOR −0.03 −0.01(0.04) (0.03)

UDENS −0.32 −0.01(0.35) (0.00)

PUBEMP −0.02(ISIC 6 + 9) (0.01)

D.PUBEMP −0.26∗∗ −0.26∗∗(ISIC 6 + 9) (0.11) (0.11)

D.EPSCS, lagged 0.39∗∗ 0.03 0.01 0.13 0.09 0.14(0.18) (0.12) (0.11) (0.21) (0.10) (0.21)

EPSCS7479 −0.03 −0.02 −0.01(0.02) (0.02) (0.01)

Constant 0.52 0.54 −0.01 0.13 0.05 0.19(0.19) (1.45) (1.35) (0.43) (0.11) (0.16)

R2 0.30 0.49 0.59 0.44 0.54 0.43AR Test 0.05 0.35∗∗∗ 0.38∗∗

(Durbin’s M)N(obs) 168 146 146 168 164 168

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 59

Bernhard Kittel 59

Table 2.9 Scores for replacement rate.

Country Replacement Rate (old) Replacement Rate (new)

AU .28 .28BE .58 .68CA .58 .64DK .90 .71FI .52 .52FR .65 .70DE .40 .64IT .04 .07JP .30 .61NL .69 .81NO .57 .67SE .91 .83UK .30 .25US .33 .64

Mean .51 .24Std. Dev. .58 .20

Note: Pearson Correlation coefficient: .82.

A digression: data considerations

Up to now, the analysis was based on the data used by Kenworthy (2003),which are taken from an unpublished OECD source and refer to grossreplacement rates. However, newer data for the net replacement rateare now available from the Scruggs welfare state entitlements data set(Allan and Scruggs, 2004; Scruggs, 2005). Table 2.9 compares the scoresfor the gross and net replacement rates for the year 1991. The correla-tion coefficient is sufficiently high (r = 0.82) to believe that the changesdo not matter much and the mean and standard deviation changedonly to a limited extent. However, a glance at the scores themselvesreveals considerable shifts. Most notably, some of the countries origi-nally scoring low, such as the United States and Japan, were recoded by30 percentage points to above-average values, and Germany shifted by 24percentage points from below-average to above-average. There are alsoobservable and substantive shifts at the upper end of the distribution:the Netherlands, Belgium, and Norway shifted upwards by about 10 per-centage points while the scores for Sweden and Denmark were correcteddownwards by 8 and 19 percentage points, respectively. These changesare definitely not immaterial. If the new scores are closer to the truththan the old ones, this would imply that the prototype of the liberal labor

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 60


Table 2.10 Reanalysis using new data for replacement rate.

1 2 3 4D10EPSCS DEPSCS

OLD NEW OLD NEW

REPLR −7.52∗∗∗ −2.72 −1.05∗∗ −0.55∗(1.67) (2.81) (0.46) (0.31)

EMPREG −0.66∗∗∗ −0.40∗∗ −0.09∗∗∗ −0.08∗∗∗(0.12) (0.15) (0.03) (0.02)

REPLR × EMPREG 0.76∗∗∗ 0.00 0.15∗∗ 0.09∗(0.27) (0.45) (0.07) (0.05)

GDPGROW 0.12∗∗∗ 0.13∗∗∗(0.03) (0.03)

TRADE 0.15∗∗∗ 0.16(0.06) (0.56)

RLTIR 0.03∗ 0.01(0.02) (0.02)

DEPSCS, lagged 0.14 0.18(0.21) (0.21)

Constant 6.68∗∗∗ 5.01∗∗∗ 0.19 0.19(0.77) (1.01) (0.16) (0.18)

R2 0.81 0.54 0.43 0.41N 14 14 168 166

Note: Col. 3, 4 (Panel models): PCSE.D10EPSCS: long-term change in private service employmentDEPSCS: annual change in private service employment

market, the United States, scores exactly the same on the core explana-tory variable as the currently most-favored scapegoat of so-called rigidity,Germany, as well as other examples of organized market economies.

Do these changes have an impact on the empirical findings? Table 2.10shows that this is indeed the case, for both the estimates of thecoefficients of this variable on long-term and short-term changes ofemployment in private sector consumer services. For ease of compari-son, I reprint model 1 of Table 2.7 and a reduced variant of model 4 ofTable 2.8 as model 1 and model 3 of Table 2.10. In the long-term anal-ysis (Table 2.10, model 2), nothing remains of the conditional effect ofthe replacement rate and only employment regulation retains statisticalsignificance. Figure 2.8 confirms this finding. In the short-term anal-ysis (Table 2.10, model 4), the effect size is almost halved, but retainsmarginal significance at the 10 percent level. Hence the conclusions wecan derive from the net replacement rate data are much less convincing

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 61

Bernhard Kittel 61

AU

BE

CADK

JP

UKUS

FI

FR

DEIT

NL

NO SE

�5

05

10

Cha

nge

empl

oym

ent i

n pr

ivat

e se

ctor

con

sum

er s

ervi

ce

0 .2 .4 .6 .8 1Replacement rate (Scruggs data)

Figure 2.8 Change in private service employment, 1981–91: replacement rateeffect conditional on employment regulations, new data for replacement rateNotes: Thick line: Effect of Replacement Rate if Employment Regulation set to Maximum,with 95% CI.Thin line: Effect of Replacement Rate if Employment Regulation set to Minimum, with95% CI.Large country markers: Employment Regulation > 4Small country markers: Employment Regulation < = 4.

than the ones based on the gross data. It remains an open questionwhether individuals use their gross or their net earnings as a yard-stick for employment decisions which can only be solved by a studyof individual-level data.

2.5 Implications and conclusion

Twenty years ago, Jackman (1985, pp. 173–5) suggested that it makeslittle sense to search for regularities in time-series data if there are goodreasons to expect the presence of the regularity only in the cross-sectionaldimension. More recently, Hall (2003) has questioned the appropriate-ness of regression analysis for the kind of questions comparative policyanalysts tend to ask. Przeworski (2004) has rejected the possibility todraw inferences about institutional effects across countries. And Kittel

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 62


(2006) has argued that policy variables lack the necessary microfounda-tion for finding regularities at the macrolevel. The present analysis, aftertwenty years of intensive usage of panel models, tends to confirm thesewarnings.

Expanding the research design by adding annual observations does notadd substantially relevant variation to the core explanatory variable, theretrenchment rate, which is a policy variable, and does not add any vari-ation to the core intervening variable, employment regulations, whichis an institutional variable. Moreover, the dependent variable, employ-ment in private sector consumer services, turns out to be nonstationaryby all available information, although this conclusion does not makesense in substantive terms because of its definition in proportions. Morespecifically, over 90 percent of the variation in employment in 1991is captured by variation in the 1970s. Hence all we can analyse withpanel data is the effect of levels differences in the replacement rate onemployment dynamics.

This is certainly not the worst of all worlds because such a refocus of theanalysis zooms in on the period analysed, instead of relegating the issueto factors present before the observations started. But even in the panelsituation, the variance component used for drawing inferences about theeffect of the replacement rate on employment growth is almost purelycross-sectional and hence based on the small number of observations.Thus, the model actually estimated is one of average growth over theobservation period. The annual variation in the dependent variable isjust a nuisance to this relationship, which must be controlled by othervariables exhibiting annual variation. As a result, as long as the modelcapturing the short-term variation is not perfect, the series will containnoise which blurs the cross-sectional pattern. On the other hand, aslong as the institutional and policy variables do not capture sufficientcross-sectional variation, the short-term coefficients for the economicmodel will be biased if no unit fixed effects are included. Includingthem, however, wipes out that variation completely. Since the coeffi-cient is based on the between variation, nothing is gained in fact forthe assessment of the core hypothesis, and much can be lost by poolingthe data. The most impressive model, in the end, appears to be a verysimple cross-sectional specification which focuses on long-term growthin employment in private-sector consumer services. This finding echoesHall and Franzese’s (1998, p. 520) greater trust in their cross-sectionalresults as compared to their time-series–cross-section analysis.

In the process of exploring the statistical relationship, several interest-ing findings emerged. First, the effect of the replacement rate appears

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 63

Bernhard Kittel 63

to be dependent on the strictness of employment regulations. Only ifemployment regulations are lax, a low replacement rate is associatedwith both above-average levels and growth of employment. Althoughthis result is premised on a single outlier, Italy, there is no case that con-tradicts the conditional effect. Secondly, the effect of public employmentin consumer services partly interferes with the replacement rate effect.This is suggestive of a second mechanism besides labor market distortionsimpacting on employment: system differences in the provision of publicgoods. Thirdly, the whole narrative appears to be contingent on the spe-cific scores on the replacement rate. Newer measures indicate that theones used in previous work differ by up to 100 percent and the detectedstatistical relationships partly collapse. This is a very discomforting find-ing, because it suggests that data quality and measurement problems mayloom larger than usually acknowledged and may invalidate considerableproportions of established results (De Deken and Kittel, 2007).

The least this analysis suggests is to be very cautious with any resultsobtained from macroquantitative research. Contrary to the initial hopesto rigorously test nomological theories independent of time and space(Przeworski and Teune, 1970; Lijphart, 1971), this chapter underlinesthe exploratory nature of macroquantitative research (Hoover, 2002)and highlights the possibility to use panel data for recounting statisticalnarratives.

Appendix: A Brief Overview of the Pooled Time-SeriesCross-Section Literature

Most econometrics textbooks now include a chapter on panels. A basicintroduction can be found in Gujarati (2003) and and more advancedtreatment in Greene (2003). Stimson (1985) and Hicks (1994) pro-vide early overviews of the state-of-the art in the late 1980 and early1990s in the context of political science. Hsiao (2003), Baltagi (2001),and Wooldridge (2002) are more advanced and technical overviews ofpanel analysis. Mátyás and Sevestre (1996) is a comprehensive hand-book of panel data econometrics at an advanced level. Beck (2001)summarizes the discussion on pooled analysis in political science forthe “reasonable”-T/small-N data structure typical of macro-level studies.Developments in dynamic panels and panel cointegration are discussedin Banerjee (1999) and Baltagi (2000). Arellano (2003) presents cur-rent econometric approaches for dynamic micro panel data, and Wawro(2002) gives a brief introduction into this literature for political scientists.Beck and Katz (2007) trace out some pathways for further development.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 64


Halaby (2004) criticizes the unreflective use of panel methods in manyapplications and discusses technical issues in typical data sets used inthe social sciences. Several conceptual problems are elaborated in Shalev(2007) and Kittel (1999). Kittel and Winner (2005) discuss specificationproblems and Wilson and Butler (2007) highlight deficiencies in currentapplications and provide some suggestions for improvement.

Notes

1. To the extent that national governments of EU member states simply imple-ment policy decisions taken in EU councils, the assumption of independencebetween observation units becomes even more problematic than the conceptof policy diffusion suggests. Nevertheless, national governments still make useof their sovereignty to implement their own variant of some joint decision.

2. Data sources for all variables used in this article are available in Kenworthy(2003).

References

Allan, James P. and Lyle Scruggs. 2004. “Political Partisanship and Welfare StateReform in Advanced Industrialized Societies.” American Journal of Political Science48: 496–512.

Arellano, Manuel. 2003. Panel Data Econometrics. Oxford: Oxford University Press.Baltagi, Badi H. 2000. Nonstationary Panels, Cointegration in Panels, and Dynamic

Panels (Advances in Econometrics Vol. 15). Oxford: Elsevier.Baltagi, Badi H. 2001. Econometric Analysis of Panel Data. 2nd Edition. New York:

Wiley.Banerjee, Anindya. 1999. “Panel Data Unit Roots and Cointegration: An

Overview.” Oxford Bulletin of Economics and Statistics 61: 607–29.Beck, Nathaniel. 2001. “Time-Series Cross-Section Data: What Have We Learned

in the Past Few Years?” Annual Review of Political Science 4: 271–93.Beck, Nathaniel and Jonathan Katz. 2007. “From Statistical Nuisances to Serious

Modeling: Changing How We Think About the Analysis of Time-Series Cross-Section Data.” Political Analysis 15: 97–100.

De Deken, Johan and Bernhard Kittel. 2007. “Social Expenditure under Scrutiny:The Problems of Using Aggregate Spending Data for Assessing Welfare StateDynamics.” Pp. 72–105 in Investigating Welfare State Change: The “DependentVariable Problem” in Comparative Analysis, edited by Jochen Clasen and NicoSiegel. London: Edward Elgar.

Ebbinghaus, Bernhard and Bernhard Kittel. 2005. “European Rigidity VersusAmerican Flexibility? The Institutional Adaptability of Collective Bargaining.”Work and Occupations 32: 163–95.

Freeman, John R. 1990. “Systematic Sampling, Temporal Aggregation, and theStudy of Political Relationships.” Pp. 61–98 in Political Analysis Vol. 1, 1989,edited by James A. Stimson. Ann Arbor: University of Michigan Press.

Greene, William. 2003. Econometric Analysis, 5th edition. Englewood Cliffs:Prentice Hall.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 65

Bernhard Kittel 65

Gujarati, Damodar N. 2003. Basic Econometrics, 4th edition. New York: McGrawHill.

Halaby, Charles N. 2004. “Panel Models in Sociological Research: Theory intoPractice.” Annual Review of Sociology 30: 507–44.

Hall, Peter A. 2003. “Aligning Ontology and Methodology in ComparativeResearch.” Pp. 373–404 in Comparative Historical Research in the Social Sciences,edited by James Mahoney and Dietrich Rueschemeyer. Cambridge: CambridgeUniversity Press.

Hall, Peter A. and Robert J. Franzese. 1998. “Mixed Signals: Central Bank Inde-pendence, Coordinated Wage Bargaining, and European Monetary Union.”International Organization 52: 505–35.

Hicks, Alexander M. 1994. “Introduction to Pooling.” Pp. 169–88 in The Com-parative Political Economy of the Welfare State, edited by Thomas Janoski andAlexander Hicks. Cambridge: Cambridge University Press.

Hoover, Kevin D. 2002. “Econometrics and Reality.” Pp. 152–77 in Fact and Fictionin Economics: Models, Realism, and Social Construction, edited by Uskali Mäki.Cambridge: Cambridge University Press.

Iversen, Torben and Anne Wren. 1998. “Equality, Employment, and Bud-getary Restraint: The Trilemma of the Service Economy.” World Politics 50:507–46.

Jackman, Robert W. 1985. “Cross-National Statistical Reseach and the Study ofComparative Politics.” American Journal of Political Science 29: 161–82.

Kenworthy, Lane. 2003. “Do Affluent Countries Face an Incomes–Jobs Trade-Off?”Comparative Political Studies 36: 1180–209.

King, Gary, Robert Keohane, and Sidney Verba. 1994. Designing Social Inquiry:Scientific Inference in Qualitative Research. Princeton: Princeton University Press.


Kittel, Bernhard. 2006. “A Crazy Methodology? On the Limits of Macroquantita-tive Social Science Research.” International Sociology 21: 647–77.

Kittel, Bernhard and Hannes Winner. 2005. “How Reliable Is Pooled Analysis inPolitical Economy? The Globalization-Welfare State Nexus Revisited.” EuropeanJournal of Political Research 44: 269–93.

Lazarsfeld, Paul F. and H. Menzel. 1961. “On the Relationship between Individualand Collective Properties.” Pp. 499–516 in Complex Organizations: A SociologicalReader, edited by Amitai Etzioni. New York: Holt, Rinehart, and Winston.

Leamer, Edward E. 1978. Specification Searches: Ad-Hoc Inference with Nonexperimen-tal Data. New York: Wiley.

Lieberson, Stanley. 1985. Making It Count: The Improvement of Social Research andTheory. Berkeley: University of California Press.

Lijphart, Arend. 1971. “Comparative Politics and the Comparative Method.”American Political Science Review 65: 682–93.

Maddala, G.S. 1999. “On the Use of Panel Data Methods with Cross-CountryData.” Annales d’économie et de statistique 55–6: 429–48.

Maddala, G. S. and S. Wu. 1999. “A Comparative Study of Unit Roots Tests withPanel Data and a New Simple Test.” Oxford Bulletin of Economics and Statistics61: 631–52.

Mátyás, Lásló and Patrick Sevestre, eds. 1996. The Econometrics of Panel Data: AHandbook of the Theory with Applications. Dordrecht: Kluwer.

9780230_202573_03_cha02.tex 5/6/2008 15: 34 Page 66


Pierson, Paul. 2004. Politics in Time: History, Institutions, and Social Analysis.Princeton: Princeton University Press.

Polillo, Simone and Mauro F. Guillén. 2005. “Globalization Pressures and theState: The Worldwide Spread of Central Bank Independence.” American Journalof Sociology 110: 1764–802.

Przeworski, Adam. 2004. “Institutions Matter?” Government and Opposition 39:527–40.

Przeworski, Adam and Henry Teune. 1970. The Logic of Comparative Social Inquiry.New York: Wiley.

Scharpf, Fritz W. 2000. “Economic Changes, Vulnerabilities, and InstitutionalCapabilities.” Pp. 21–125 in Welfare and Work in the Open Economy, edited byFritz W. Scharpf and Vivien A. Schmidt. Oxford: Oxford University Press.

Scruggs, Lyle. 2005. Comparative Welfare Entitlement Dateset, accessed April 2006.www.sp.uconn.edu/∼scruggs/wp.htm.

Shalev, Michael. 2007. “Limits of and Alternatives to Multiple Regression inMacro-Comparative Research.” Pp. 261–308 in Capitalisms Compared (Compar-ative Social Research, Vol. 24), edited by Lars Mjøset. Amsterdam: Elsevier.

Stimson, James A. 1985. “Regression in Time and Space: A Statistical Essay.”American Journal of Political Science 29: 914–47.

Wawro, Gregory. 2002. “Estimating Dynamic Panel Data Models in PoliticalScience.” Political Analysis 10: 25–48.

Wilson, Sven E. and Daniel M. Butler. 2007. “A Lot More to Do: The Sensitiv-ity of Time-Series Cross-Section Analysis to Alternative Specifications.” PoliticalAnalysis 15: 101–23.

Wooldridge, Jeffrey M. 2002. Econometric Analyses of Cross Section and Panel Data.Cambridge: MIT Press.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 67

3Comparative EmploymentPerformance: A Fuzzy-Set AnalysisJessica Epstein, Daniel Duerr, Lane Kenworthy, andCharles Ragin

Introduction

Much of the debate about comparative employment performance inrecent decades has focused on the impact of labor market institutionsand policies. A number of studies have found that institutions andpolicies that restrict or regulate market processes – for example, wagecompression, employment protection regulations, high taxes, generousunemployment benefits – have adverse effects on employment outcomes(OECD, 1994, 2006; Nickell, 1997; Scharpf, 1997, 2000; Siebert, 1997;Iversen and Wren, 1998; Blanchard and Wolfers, 2000; Blau and Kahn,2002; IMF, 2003; Kenworthy, 2004, 2008; Kemmerling, 2005; Nickell,Nunziata, and Ochel, 2005; Bassanini and Duval, 2006). Others questionthis conclusion (Glyn and Salverda, 2000; Esping-Andersen and Regini,2000; Martin, 2004; Baccaro and Rei, 2005; Baker et al., 2005; Schettkat,2005; Stephens and Bradley, 2005; Howell et al., 2006).

Almost without exception, quantitative macrocomparative studies onthis issue have used regression as the analytical technique. We insteaduse fuzzy-set qualitative comparative analysis (fuzzy-set QCA). Utilizingfuzzy-set QCA, we explore the determinants of poor employment perfor-mance in low-end private-sector services in 14 countries between 1979and 1995.

Why fuzzy-set QCA?

Fuzzy-set QCA offers several advantages. First, it is better-suited thanregression for exploring causal configurations – situations in which vari-ables have an impact only in combination with a high or low degreeof one or more other factors. In regression analysis, causal configura-tions are assessed via interaction terms. However, a small N limits thenumber of interactions terms that can be included in a regression model.

67

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 68


In addition, the difficulty of interpreting interaction terms with morethan two variables makes modeling complex interactions problematic.Moreover, while assessing interactions in regression requires that vari-ables demonstrate a multiplicative effect, QCA treats any case aspectsthat appear together systematically – in any quantity – as potentiallyinterdependent.

Secondly, fuzzy-set QCA allows us to identify multiple pathways to anoutcome. Correlational techniques such as regression treat the presenceof an outcome (dependent variable) without a given cause (independentvariable) as negative evidence for the strength of that causal explanation.Thus, a factor that has an impact in a subset – but only a subset – of casestends to become obscured in regression results with deflated coefficientsand inflated variance. In contrast, fuzzy-set QCA can reveal causal pat-terns that differ across subsets of cases. This method thereby allows usto examine relatively large datasets with more complex causal narrativesthan are generally possible with correlational techniques.

Thirdly, whereas regression is useful for examining tendential rela-tionships – the general tendency of a particular factor to influence anoutcome of interest – fuzzy-set QCA is helpful in exploring a differentkind of relationship: causal sufficiency. Fuzzy-set QCA assesses suffi-ciency via the logic of set-theoretic relations. Set theory is inherent(though often implicit rather than explicit) in much of social science(Ragin, 2000). Sets are simply conceptual categories like “generous gov-ernment benefits” or “low income inequality.” Much social scienceconcerns itself with the relative membership of cases in such categories,the theoretical validity of a set designation, or the ways one set mightsubsume another. The set–subset ordering of social phenomena is keyto understanding causal sufficiency. A causal factor is considered suffi-cient when its presence always (or nearly always) “produces” a particularoutcome. Assaulting one’s employer, for instance, is generally a suffi-cient condition for being fired. But it is not the only way to get fired:one could also stop coming to work, or embezzle, or perform poorly.In set-theoretic terms, the cause (assaulting one’s employer) is a sub-set of the outcome (being fired): it always produces the outcome, butit is not the only pathway to it. Because sufficient causes are alwayssubsets (or near subsets) of the outcomes they “produce,” discerningsubset relations points to potentially sufficient causal pathways. Thus,for instance, if all countries with strict employment protection regu-lations also have poor employment performance, we might considerstrict employment protection a sufficient condition for bad employmentperformance.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 69

Jessica Epstein, Daniel Duerr, Lane Kenworthy, and Charles Ragin 69

Data

Because it is concerned with ordering conceptual categories rather thanassessing statistical correlations, fuzzy-set QCA requires the use of set-theoretic variables. “Fuzzy sets” refer to the pseudo-continuous scale onwhich cases are coded. Like conventional quantitative indexes, fuzzysets range from zero to one. But because the interest is in how stronglycases conform to theoretical categories, fuzzy coding schemes are basedon connection to qualitative anchors rather than mathematical equiva-lence. Cases coded zero are considered “fully out” of a set, and thosecoded one are “fully in” the set. Membership along the continuumbetween zero and one can be determined qualitatively, on a case-by-casebasis, or via scaling using an arithmetic formula. For each of our vari-ables, we use three qualitative anchors of set membership: 0, 0.5, and 1.The 0.5 anchor serves as the “crossover point” from “more in than out”to “more out than in” the set. We then use a mathematical rescalingto distribute the rest of the cases between these points. The process isdescribed in greater detail below.

Outcome

The first step is to determine the relevant conceptual set or group. Shouldwe focus on good or bad employment performance? Because much ofthe scholarly and political debate has centered on understanding whysome European countries have struggled with job creation, we opt forbad employment performance as the outcome of interest. We set out toidentify causal configurations associated with slow or negative growthin employment rates.

The outcome we analyze is “poor” performance in low-end private sec-tor service jobs during the period 1979–95. (The chapter appendix pro-vides descriptions and data sources for all of our variables.) This includesrestaurants, hotels, retail and wholesale trade, and community-social-personal services. Because productivity in these jobs tends to be low anddifficult to increase, they are the most likely to be adversely affectedby institutional and policy “rigidities.” In addition, these jobs accountfor a relatively large share of the cross-country variation in both levelsof employment and change in employment in recent decades. Unfortu-nately, data are available for only 14 countries and only through 1995.

Because employment rates change only incrementally, an analysis ofemployment levels during or at the end of this period will be heav-ily influenced by employment levels at the beginning. We therefore

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 70


Asl

Bel

Can

Den

Fin

Fr

GerIt

Ja

Nth

Nor

Swe

UKUS

�2

0

2

4

6

8

Em

ploy

men

t cha

nge

in p

rivat

e-se

ctor

con

sum

er s

ervi

ces,

1995

min

us 1

979

(%)

Country

Figure 3.1 Employment change in low-end private sector services, 1979 to 1995

examine change in employment rates. We measure change in absoluteterms: employment rate in 1995 minus employment rate in 1979.

A potential concern about measuring employment performance interms of change over time is that there might be strong ceiling or catch-up effects. Countries that began the period with low employment ratesin low-end private sector services may have found it easier to increaseemployment, whereas those that began with high rates might havealready been near a ceiling and thus found it more difficult to increaseemployment. However, no such pattern is in evidence for these countriesduring this time period.

How do we translate absolute employment change into a qualitativelydefined fuzzy set? The first step is determination of “bad” employmentperformance, with both substantive knowledge and raw scores as ourguide. Figure 3.1 shows employment changes in all 14 countries. The

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 71


Asl

Bel

Can

Den

FinSwe

Fr

GerIt

Ja

Nth

Nor

UKUS

0

.25

.5

.75

1

Fuz

zy-s

et s

core

s

�2 0 2 4 6 8

Raw values

Figure 3.2 Employment change fuzzy-set scores by employment change rawvalues

sharpest declines were experienced in Finland and Sweden (−1.7 and−1.0, respectively). The next worst performer was Denmark, but itsemployment decline (−0.1) was far smaller than those of Sweden andFinland. We therefore draw the membership line around Finnish andSwedish levels, setting their fuzzy scores to “1” to signify full membershipin the set of poor employment performers. In a set-theoretic analysis,once the threshold for membership in the group is established, all casesthat meet the threshold are coded the same; any variation among caseswithin the group is treated as irrelevant.

Similarly, the sharpest break at the other end of the distribution clearlyfalls between Japan and the United States. Japan is therefore designatedfully out of the set of bad employment performers and is coded zero.

The other anchor that is often determined qualitatively in a fuzzy-setcoding is the “cross-over point”: 0.5. This separates cases that are “morein the set than out” from those that are “more out than in.” In thisinstance the obvious break point is between Germany and Australia. Asimple rescaling of the values for the countries neither fully in nor fullyout – using the formula (raw value – minimum value)/range – capturesthis break point and distributes these countries between the anchors.

Figure 3.2 shows the fuzzy-set scores for employment change plottedagainst raw values.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 72


Causal conditions

There are many hypothesized determinants of cross-country variation inemployment performance. We include six labor market institutions andpolicies that have been central in research and political debate on thisissue. We describe each causal condition in a way that corresponds tohow it is expected to contribute to bad employment performance.

1. Low earnings inequality. Because productivity tends to be low anddifficult to increase in low-end service jobs, high wages may deteremployment growth. The P50/P10 earnings ratio among the full-timeemployed is a fairly good indicator of the level of wages relative to thecountry median. Lower levels of this ratio, indicating greater pay com-pression, are expected to contribute to poorer employment changeperformance.

2. High wage increases. Rapid growth of overall wages – or, more precisely,wages adjusted for inflation and productivity growth, i.e. real unitlabor costs – is expected to deter employment growth in all types ofjobs.

3. High payroll and consumption taxes. Payroll taxes increase employ-ers’ nonwage labor costs and consumption taxes reduce consumerdemand for price-elastic goods and services. High levels of these typesof taxes are thereby expected to reduce employment growth. We usepayroll and consumption taxes as a share of GDP as our measure.

4. High employment protection regulations. If employers are less able to fireworkers during bad times, they may reduce hiring during expansion-ary periods. Stringent employment protection regulations are thusexpected to reduce employment growth.

5. High unemployment benefit generosity. Generous unemployment com-pensation programs are expected to deter employment growth fromthe supply side, by reducing the eagerness of benefit recipients toenter or reenter employment. To tap generosity, we utilize the per-centage of former earnings replaced by (gross) benefits, averaged overthe period 1980–95.

6. High public employment. A number of nations combat unemploymentin part via creation of jobs in the public sector. Government jobsmay supplant private-sector employment growth, particularly in low-end services. If so, high levels of public-sector employment maycontribute to slower growth of jobs in private sector services. The mea-sure we use is public employment as a percentage of the working-agepopulation.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 73

73

Asl

Bel

Can

DenFin

FrGer

It

Ja

Nth

NorSwe

UK

US0

.25

.5

.75

1

Fuz

zy-s

et s

core

s

0

.25

.5

.75

1

Fuz

zy-s

et s

core

s

0

.25

.5

.75

1

Fuz

zy-s

et s

core

s

0

.25

.5

.75

1

Fuz

zy-s

et s

core

s1.31.51.71.92.1

Raw values (reversed)

Low earnings inequality

Asl

Bel

Can

DenFin

Fr

Ge

It

Jar

NthNor

Swe

UK

US

�1.5 �1 �.5 0Raw values

Raw values Raw values

High wage increases

Asl

Bel

Can

Den

Fin

Fr

Ger

It

Ja

Nth

Nor

Swe

UK

US

10 15 20 25 30

High payroll and consumption taxes

Asl

Bel

Can

Den

Fin

Fr

Ger

It

JaNth

Nor

Swe

UK

US

0 .5 1.0 1.5 2.0

High employment protection regulations

Asl

BelCan

Den

Fin

Fr

Ger

It

Ja

Nth

Nor

Swe

UKUS

0

.25

.5

.75

1

Fuz

zy-s

et s

core

s

0

.25

.5

.75

1

Fuz

zy-s

et s

core

s

0 .2 .4 .6 .8 1

Raw values Raw values

High unemployment benefit generosity

Asl

Bel

Can

Den

FinFr

GerIt

Ja

Nth

Nor Swe

UK

US

5 10 15 20 25

High public employment

Figure 3.3 Causal condition fuzzy-set scores by raw valuesNote: Descriptions of the original variables and data sources are in the text.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 74


Figure 3.3 shows our scoring for these six causal conditions. Each chartplots the fuzzy-set scores we use for the condition by its raw values. Hereour fuzzy-set coding choices were determined largely empirically. Foreach variable we tried a variety of different codings and used the onethat fit best with the data.

Analysis

We conducted our fuzzy-set QCA analysis using the software pro-gram fs/QCA 2.0 (www.u.arizona.edu/∼cragin/fsQCA/software.shtml).The program simplifies data patterns to identify potentially “sufficient”causal associations. Its ultimate products are a set of logical statementsidentifying factors or combinations of factors that appear as subsets of anoutcome, along with mathematical measures with which to assess theirutility. The researcher has discretion at several stages of analysis to refineor alter the logical rules that produce these statements and a great dealof interpretive leeway in utilizing the results. We explain each stage ofthat process below.

As in regression analysis, fs/QCA requires that the researcher specify amodel – a set of causal conditions to be included in the analysis. To gen-erate the broadest range of solution sets, we employed multiple modelsutilizing various combinations of our six causal conditions.

After model specification, the next step is examination of a “truthtable.” The truth table is an analytic device that displays all logically pos-sible combinations of causal conditions and indicates cases’ distributionacross these combinations. Table 3.1 is a truth table produced to assessall six causal conditions: earnings inequality, wage changes, payrolland consumption taxes, employment protection regulations, unemploy-ment benefit generosity, and public employment. The fs/QCA programoriginally produced a 64-row table, representing all (26) logically possi-ble combinations of the causal conditions. To facilitate the presentation,we have removed those with no empirical instances.

The truth table is a simplified data map. The “number” column denoteshow many cases conform to the listed combination, but the readershould not consider each row to represent only those cases.1 Truth tablerows directly correspond to the logical possibility of a particular causalcombination. Graphically, imagine each row corresponding to a cornerof “vector space” – a multidimensional plot representing all possible com-binations of causally relevant case aspects. The 0 s and 1 s in each cell arean instruction about how to consider a case’s location – its membership ornonmembership (1 minus membership) in a particular fuzzy set. Because

9780230_202573_04_cha03.tex

5/6/200815:35

Page75

75

Table 3.1 Truth table from analysis of all six causal conditions.

Low High High High High High Number Outcome: poor Consistearnings wage payroll and employment unemployment public employmentinequality increases consumption protection benefit employment change

taxes regulations generosity performance

1 1 1 1 1 1 2 11 0 1 1 1 1 2 11 0 1 1 1 0 1 0.961 1 1 1 1 0 1 0.961 1 0 0 1 1 1 0.941 1 1 1 0 0 2 0.930 0 0 0 1 1 1 0.860 0 0 0 0 1 1 0.680 1 0 0 0 1 1 0.650 0 0 0 0 0 1 0.610 1 0 1 0 0 1 0.58

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 76


the lines and cells are theoretical configurations, few cases conformperfectly to the conditions they denote, and any given case has par-tial membership in multiple rows. For example, a country with a 0.4score on the employment protection fuzzy set also has a 0.6 member-ship in its negation, making it a partial member of both the last and next-to-last rows of table 1. The table thus maps case aspects rather than thecases themselves.

A primary purpose of the truth table is to guide the researcher in deter-mining standards for the consistency of causal relationships. This is donewith consistency scores. These are shown in the “Consist” column inTable 3.1. We are concerned in this analysis with causal sufficiency –the ability of certain configurations of policy and institutional factorsto consistently produce bad employment performance. In set theoreticterms, we are interested in the extent to which particular causal factors orconfigurations are subsets of the outcome. As in regression, increases inthe strength of set membership in a cause are expected to result in morecomplete membership in the outcome as well. If high payroll and con-sumption taxes are sufficient to produce bad employment performance,we should observe few or no cases with high payroll and consumptiontaxes (a fuzzy score of 1) and good or moderate employment perfor-mance (a fuzzy score of 0.5 or less). The consistency (“Consist”) scorefor a configuration is a measure of this subset relationship. It is thus ameasure of the extent to which membership strength in the outcome setis consistently equal to or greater than membership in the causal con-figuration. For each configuration (row in the truth table), minimummembership scores (causal combination versus outcome) are added forall cases. This number is divided by the sum of all minimum membershipscores in the causal combination. Formally, the calculation is: Consis-tency (Xi < Yi) = ∑

(min(Xi,Yi))/∑

(Xi). When membership in outcomeY is less than membership in causal configuration X, the numeratorwill be smaller than the denominator and the consistency score willdecrease. Consistency scores range from 0 to 1, with 0 indicating nosubset relationship and a score of 1 denoting a perfect subset relationship.

These consistency scores help the researcher decide which configura-tions should be considered reasonable subsets of the outcome. Once thisdecision is made for a particular configuration, the researcher enters a 1or 0 into the cell in the blank “outcome: poor employment change per-formance” column in the truth table, which tells the program whetheror not to treat that particular causal combination as an instance of bademployment performance. Minimum levels of set-theoretic consistencywould be achieved by setting a “Consist” threshold of at least .75 (Ragin,

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 77


2004), preferably higher. Between this level and full set-theoretic con-sistency (1), the analyst must choose a minimum threshold. For ouranalysis, we utilized various thresholds between .85 and .95 for eachmodel.

The next step of the analysis concerns the treatment of counterfactualcases (see Ragin and Sonnett, 2005). With six causal conditions, thereare 64 (26) logically possible combinations of conditions. But only 11of these combinations are actually represented empirically in our data.These are the 11 listed in Table 3.1. All other logically possible con-figurations are “remainders” – counterfactual configurations that lackempirical instances. Because remainders constitute neither positive nornegative evidence, the fs/QCA program allows the researcher to treatthem as either. In the first case, remainders are treated as potentially posi-tive evidence – cases that could have been – and are utilized by the programas logically simplifying assumptions. (In fs/QCA, this is referred to as the“don’t care” option.) But this option assumes that all non-instantiatedconfigurations are plausible. Where that is not the case, it is best to treatthem as negative instances of the outcome. (This is referred to in the pro-gram as the “false” option.) Doing so produces less parsimonious results.In our analyses, we examined both options with all models (Ragin andRihoux, 2004; Ragin and Sonnett, 2005).

Once these choices have been made, the program then utilizes set–subset logic to simplify the patterns of association displayed in the truthtable. Recall the first row of Table 3.1, where a configuration of allsix causal conditions was a perfect subset of the set of bad employ-ment performers. How do we discern which of these six conditionsreally matter, and which are superfluous? Note that in the followingsix rows, there are three cases of the outcome that do not include thehigh wage increases or high public employment causal conditions. Wecould reasonably conclude that both of these case aspects are superflu-ous “ingredients” in the causal pathways these rows express. Similarly,there are two cases of the outcome that do not include the high unem-ployment benefit generosity causal condition, and one that does notinclude high payroll and consumption taxes. Depending on the choicesdescribed above – about consistency thresholds and simplifying assump-tions – the fs/QCA program will offer several simplified formulations ofthese causal pathways by eliminating causal factors that appear super-fluous and identifying combinations that consistently appear sufficientto produce the outcome.

Table 3.2 shows two such simplified “solution sets.” The first modeledall six causal conditions using a 90 percent consistency threshold and the

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 78


Table 3.2 Examples of solution sets.

Coverage

Raw Unique Consistency

Solutions (no simplifying assumptions)Low earnings inequality ∗absence of high wage increases ∗High payroll and consumption taxes ∗High employment protection regulations ∗High unemployment benefit generosity 0.37 0.26 0.90

Low earnings inequality ∗High wage increases ∗High payroll and consumption taxes ∗High employment protection regulations ∗absence of high public employment 0.31 0.15 0.95

Low earnings inequality ∗High wage increases ∗absence of high payroll and consumption

taxes ∗absence of high employment protection

regulations ∗High unemployment benefit generosity ∗High public employment 0.18 0.11 0.94

Solution coverage: 0.67Solution consistency: 0.96

Solutions (simplifying assumptions)Low earnings inequality 0.84 0.84 0.83

Solution coverage: 0.84Solution consistency: 0.83

Note: ∗ = logical “and.” Coverage and consistency are explained in the text.

“false” option favoring empirical complexity. The second solution set dif-fers only in that it was derived using the “don’t care” option, which favorsparsimony. These two solutions, the parsimonious and the complex, canbe viewed as the two endpoints of a continuum. In between these twoendpoints are various intermediate solutions, which are also valid. Bydefinition, an intermediate solution must be a superset of the complexsolution (no simplifying assumptions used) and a subset of the parsimo-nious solution (all possible simplifying assumptions used, regardless oftheir plausibility).

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 79


Low consistency

Outcome Cause

High consistency, low coverage

Outcome

Cause

Figure 3.4 Consistency and coverage

The primary way to assess the value of these causal explanations isto examine their consistency scores. Like consistency scores in a truthtable, “consistency” in the results produced by fs/QCA refers to a causalcombination’s consistency as a subset of the outcome. In the simplestterms, low consistency means that there is no subset relation betweena combination of case aspects and the outcome. This relationship isdepicted in the first illustration in Figure 3.4. As with the assessmentof the consistency of truth table rows, scores closest to “1” representthe strongest connection, while those below .85 should be treated withcaution.

This assessment should also be guided by case knowledge. The fourthsolution in Table 3.2 – the presence of income inequality – is on thethreshold of acceptable consistency, with a .83 score. If we had reasonto believe that an outlying case – a country with very low inequality butvery good employment performance – affected this score without chal-lenging the causal story we want to tell, we could consider this solutionset as a strong contender. Note that this is the same logical process wewould follow in any comparative case study research. The mathematicalmarkers that QCA provides merely guide our logical comparisons.

The second tool for assessing results is the coverage score. Coveragerefers to the proportion of the sum of the membership scores in an out-come that a particular configuration explains. Very low coverage scoresindicate that even if a causal configuration is consistent with the out-come, it is substantively trivial. This is depicted in the second illustrationin Figure 3.4. Coverage and consistency often are inversely related to oneanother, because very particular or exact explanations (which may behighly consistent) tend to be less generalizable. In Table 3.2, “raw cover-age” scores refer to the proportion of the outcome scores covered by anexplanation by itself, while “unique coverage” refers to the proportion

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 80


of outcome scores covered, net of that solution’s coverage overlap withthe other solutions identified.

Utilizing coverage scores also entails the use of substantive and theoret-ical knowledge. Because QCA allows us to find “clusters of commonalitiesacross subsets of cases” (Ragin, 2000, p. 75), it also allows us to considerthe possibility that different causal stories work for distinct subsets ofcases. These stories can be pieced together into a larger narrative aboutthe differential factors driving a phenomenon across countries or regions.How to balance the complexity of that narrative with the mandate of par-simony is a matter of judgment. In the next section we compare coveragescores with scatterplots to match causal pathways to the countries bestexplained by them.

The next step in processing the fs/QCA program’s results involvesthe elimination of theoretically implausible terms using counterfac-tual analysis. Essentially, this process involves using only select subsetsof simplifying assumptions to produce solutions that are intermediatebetween the parsimonious solution (all simplifying assumptions permit-ted) and the complex solution (no simplifying assumptions permitted).The program analyzes casual conditions in both their present and absentstates when it considers all logically possible combinations of causal con-ditions. After all, the absence of a particular case aspect might be just asimportant in determining employment trends as its presence. In ouranalysis, however, all causal conditions are expected to contribute topoor employment performance only when they are present, and we havecoded them so that high membership scores should be linked to poorperformance. For this reason, we eliminated absent conditions from ourcomplex solutions in order to generate our intermediate solutions, whilerespecting the rule that intermediate solutions must be subsets of the par-simonious solutions. For instance, the first solution listed in Table 3.2is: high earnings inequality ∗ absence of high wage increases ∗ highpayroll and consumption taxes ∗ high employment protection regula-tions ∗ high unemployment benefits generosity. Following the procedurejust described (and Ragin and Sonnett, 2005), the intermediate solu-tion is: high earnings inequality ∗ high payroll and consumption taxes ∗

high employment protection regulations ∗ high unemployment benefitsgenerosity.

Findings

How did we utilize the tools and procedures described in the pre-vious section to reach and interpret results? First, we gathered the

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 81


multiple solutions generated by (1) multiple models (that is, vari-ous combinations of case aspects), (2) multiple consistency thresh-olds, and (3) different treatment of counterfactual cases (remainders).We then eliminated solutions with consistency scores below 0.85 andcoverage scores below 0.30. Next, we crafted intermediate solutionssuing the procedures described above. This process yielded five causalpathways.

Each of these five pathways was then coded as its own fuzzy set. Eachcountry’s score in each pathway is determined by its weakest membershipin the conditions that constitute the configuration. Why? To concludethat a particular causal factor generated an impact requires that factorto be present at least at the level ascribed to the entire combination. So,if there are three factors in the configuration and a country’s scores onthem are .50, .85, and .25, the country is coded .25 on the configurationfuzzy set.

Each of these configuration fuzzy sets was then run through the fs/QCAprogram to assess its specific consistency and coverage. The rest of ouranalysis is based on evidence from scatterplots and judgments aboutparsimony.

The five solutions and their consistency and coverage scores are shownin Table 3.3. The first four solutions point to one fairly simple expla-nation: a combination of low earnings inequality and high payrolland consumption taxes was a sufficient condition for generating poorlow-end private sector service employment performance. This solutionaccounts for 75 percent of the sum of the membership scores in pooremployment performance. It has slightly lower set-theoretic consistencythan the first three, but much higher coverage. It also is the logical super-set of the first three solutions, subsuming their more complex causalnarratives into a more parsimonious explanation.

Figure 3.5 shows four scatterplots, each with fuzzy employmentchange scores on the Y axis and one of the first four solutions fromTable 3.3 on the X axis. The higher consistency scores for the first twoconfigurations – #1 and #2 from Table 3.3 – are evidenced by the factthat more countries are located above main diagonal line in these twocharts than in the lower two charts. “Perfect” causal sufficiency wouldbe in evidence if every case were located above the main diagonal. Thatis not true for any of the four configurations, but only two countries liebelow the line for configurations 1 and 2, versus three below the line forconfiguration 3 and four below the line for configuration 4.

What cases do these explanations cover? Using rough cutoffs indi-cated by the dotted 45-degree lines, we determined three broad levels

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 82


Table 3.3 Five causal pathways.

Consistency Coverage

1. Low earnings inequality ∗High payroll and consumption taxes ∗High employment protection regulations ∗High unemployment benefit generosity 0.96 0.58

2. Low earnings inequality ∗High wage increases ∗High payroll and consumption taxes ∗High employment protection regulations 0.96 0.42

3. Low earnings inequality ∗High payroll and consumption taxes ∗High employment protection regulations 0.95 0.67

4. Low earnings inequality ∗High payroll and consumption taxes 0.93 0.75

5. High unemployment benefit generosity 0.87 0.85

Note: ∗ = logical “and.” Coverage and consistency are explained in the text.

of coverage: “clearly conforming” cases, “potentially conforming” cases,and “clearly not conforming” cases. Low earnings inequality combinedwith high payroll and consumption taxes, shown in the lower-rightchart, explains the most cases. Japan, the United Kingdom, Italy, Ger-many, the Netherlands, Norway, and Sweden all are on or very close tothe central line – the point of perfect correspondence between cause andoutcome. The United States, France, Finland, Belgium, and Australia areclose enough to the line that these cases are “potentially” covered by thisstory. Only Denmark and Canada clearly do not conform.

The fifth causal configuration in Table 3.3 consists of a single causalfactor: high unemployment benefit generosity. This solution has veryhigh coverage, at 85 percent, but a questionable consistency level of .87.Recall that consistency scores are calculated for a solution’s set-theoreticconsistency across all cases. Even a solution with relatively low consis-tency might explain some cases very well. The scatterplot in Figure 3.6indicates that high unemployment benefit generosity covers most of thecases in our data set reasonably well. Its consistency is relatively lowbecause so many cases have slightly higher levels of membership in thehigh unemployment benefit generosity fuzzy set than they do in the setof poor employment performers (they lie to the right of the central line

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 83


Asl

Bel

Can

Den

Fin

Fr

GerIt

Ja

Nth

Nor

Swe

UKUS

0 .25 .5 .75 1Causal configuration #1 from table 3

Earnings inequality * taxes * employmentprotection * unemployment benefits

Asl

Bel

Can

Den

Fin

Fr

GerIt

Ja

Nth

Nor

Swe

UKUS

0 .25 .5 .75 1Causal configuration #2 from table 3

Earnings inequality * wage changes * taxes *employment protection

Asl

Bel

Can

Den

Fin

Fr

GerIt

Ja

Nth

Nor

Swe

UKUS

0

.25

.5

.75

1

Poo

r em

ploy

men

t cha

nge

perf

orm

ance

0

.25

.5

.75

1

Poo

r em

ploy

men

t cha

nge

perf

orm

ance

0

.25

.5

.75

1

Poo

r em

ploy

men

t cha

nge

perf

orm

ance

0

.25

.5

.75

1

Poo

r em

ploy

men

t cha

nge

perf

orm

ance

0 .25 .5 .75 1

Causal configuration #3 from table 3

Earnings inequality * taxes * employmentprotection

Asl

Bel

Can

Den

Fin

Fr

GerIt

Ja

Nth

Nor

Swe

UKUS

0 .25 .5 .75 1


Earnings inequality * taxes

Figure 3.5 Poor employment change performance by causal configurations 1–4from Table 3.3Note: Lines are 45-degree lines.

in the scatterplot). This discrepancy is potentially the result of codingerror, but in any event it is fairly small. All cases except Japan, Finland,and Italy are at least potentially covered. Only Italy is clearly not con-forming. Denmark and Canada, which are not covered by any of thesolutions shown in Figure 3.5, are among the cases that clearly conformto the high unemployment benefit story.

Our analysis thus highlights two causal pathways in accounting forpoor employment performance among these 14 countries from 1979 to

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 84


Asl

Bel

Can

Den

Fin

Fr

GerIt

Ja

Nth

Nor

Swe

UKUS

0

.25

.5

.75

1P

oor

empl

oym

ent c

hang

e pe

rfor

man

ce

0 .25 .5 .75 1


Unemployment benefit generosity

Figure 3.6 Poor employment change performance by causal configuration 5 fromTable 3.3

1995. The first is low earnings inequality coupled with high payroll andconsumption taxes. This configuration appears likely to have been keyto poor employment growth in low-end private sector services in Swe-den, Norway, Finland, France, Italy, Germany, and the Netherlands. Thesecond is high unemployment benefit generosity. This appears to havebeen the main contributing factor for Denmark, Canada, and Belgium,but perhaps also for France, Germany, the Netherlands, Sweden, andNorway.

Conclusion

In the introductory section of this chapter we highlighted three advan-tages of using fuzzy-set QCA to analyze comparative employmentperformance. The first is the technique’s utility in exploring causalconfigurations. In the end, our results centered on one simple causalconfiguration and another singular causal factor: (1) low earningsinequality combined with high payroll and consumption taxes; (2) highunemployment benefit generosity. The inequality–taxes configurationcould have been captured reasonably well with the type of multiplicative

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 85


interaction term usually used in regression. Indeed, such a multiplica-tive term correlates at .98 with the fuzzy-set causal configuration scoresshown on the horizontal axis in the lower-right chart in Figure 3.5 above.Yet it is not always the case that the causal story yielded by a fuzzy-setQCA analysis will end up so simple. It is much less likely that an ana-lyst using regression would have identified one of the more complexcausal configurations from Table 3.3. Even our modest results, attainedwith relatively limited data, point to a combination that deserves fur-ther attention in regression-oriented work. Indeed, one potential use offuzzy-set QCA is to identify combinations of case aspects that deservefurther scrutiny in quantitative studies.

The second advantage of fuzzy-set QCA is that it can identify multi-ple paths to an outcome. While regression’s focus on average net effectstends to obscure causal relationships that do not operate across all ormost of a sample, QCA locates causal pathways that may be distinct tosmaller clusters of cases. In our analysis, two pathways emerged as mostimportant. Each helps to account for a number of countries, but not theentire group. Canada and Denmark clearly do not fit the low earningsinequality combined with high payroll and consumption taxes expla-nation. Similarly, Finland and Italy are not well accounted for by thehigh unemployment benefit generosity explanation. In other analyses alarger number of pathways may be needed to explain the variety of caseoutcomes.

The third advantage of QCA that we noted at the outset is its focus onsufficient causal conditions. For researchers interested solely in “tenden-tial” relationships, this is a drawback rather than an asset. But an interestin sufficiency is in fact implicit in a nontrivial amount of empirical workin the social sciences.

A fourth benefit of fuzzy-set QCA, unexplored here but fruitful forfurther analysis, is that unlike regression it does not assume causal sym-metry at the two “ends” of the dependent variable. That is, the causesof low values of the outcome are not seen as the inverse of the causes ofhigh values. Each is assumed to require a separate theoretical argumentand empirical analysis. Though we have not conducted an analysis of thedeterminants of good employment performance, such an investigationmight yield results that are not simply the inverse of the causes of pooremployment performance identified here.

Of course, like any analytical technique, fuzzy-set QCA has limita-tions. Some it shares with regression. Perhaps the most important is thatQCA too suffers from a small-N problem: with a relatively small num-ber of cases (14 here), only a limited number of causal conditions can

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 86


be included in an analysis. This leaves analyses open to concern about“omitted variable bias” (see the Hicks and Kenworthy chapter on fam-ily policies in this volume). As with regression, there also is a dangerthat researchers will mechanically utilize results produced by the fs/QCAprogram without drawing on theoretical and case knowledge. Thoughgrounded in the case study tradition, this method is no less susceptible tosuch an approach. QCA offers the greatest insight when paired with care-ful consideration of theory and cases. QCA also has limitations relativeto regression and other types of correlational analysis: it is not designedto examine net effects or tendential relationships. If a researcher’s inter-est is in identifying the tendential impact of a particular hypothesizedcause on an outcome, regression is a more appropriate technique.

Perhaps most important, we want to emphasize that the choice ofmethod in macrocomparative research is not an either/or decision. Eachtechnique has strengths and drawbacks. Which is most appropriate willdepend on the substantive question and the way in which the researcherwants to approach it. Sometimes this will call for regression, sometimesfor QCA, sometimes for other strategies or techniques, and sometimesfor the use of multiple methods.

Appendix

Poor employment change performance. Raw values: absolute change (1995value minus 1979 value) in employment in private sector consumer-oriented services – wholesale and retail trade, restaurants and hotels,and community/social/personal services (ISIC 6 and 9; ISIC revision350–2, 55, 90–3) – as a percentage of the population age 15 to 64.Unfortunately, private sector employment can be distinguished frompublic sector employment in these industries only through 1995,so the time series for this variable ends in that year. Source: TorbenIversen, Department of Government, Harvard University, calculatedfrom OECD data. For discussion see Iversen and Wren (1998).

High earnings inequality. Raw values: P50/P10 ratio for earnings amongfull-time employed individuals. Averaged over 1979–95. Source: OECDunpublished data set.

High wage increases. Raw values: year-to-year percentage change inemployee compensation, adjusted for changes in productivity andfor inflation (real unit labor costs). Averaged over 1979–95. Source:Authors’ calculations from data in OECD (2004a).

High payroll and consumption taxes. Raw values: government revenuesfrom social security contributions, payroll taxes, and taxes on goods

9780230_202573_04_cha03.tex

5/6/200815:35

Page87

87

Table 3A.1 Fuzzy-set scores.

Poor Low High High High High Highemployment earnings wage payroll and employment unemployment publicchange inequality increases consumption protection benefit employmentperformance taxes regulations generosity

Australia .27 .49 .45 .00 .22 .32 .76Belgium .61 1.00 .56 .88 .73 .71 .43Canada .70 .00 .47 .19 .11 .71 .57Denmark .90 .90 1.00 .39 .49 1.00 1.00Finland 1.00 .82 1.00 .82 .58 .67 .69France .76 .55 .66 1.00 .68 .81 .60Germany .57 .55 .92 .77 .80 .45 .36Italy .62 .87 .69 .68 1.00 .00 .25Japan .00 .46 .92 .09 .70 .33 .00Netherlands .67 .66 .00 1.00 .66 .85 .23Norway .87 .93 .00 .86 .76 .68 1.00Sweden 1.00 1.00 .18 1.00 .85 1.00 1.00United Kingdom .26 .30 .69 .42 .14 .33 .58United States .21 .00 .34 .00 .00 .35 .38

Note: Variable descriptions and data sources are listed below.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 88


and services (consumption) as a share of GDP. Averaged over 1979–95.Source: OECD (2004b, pp. 74 and 77, tables 14 and 20).

High employment protection regulations. Raw values: index capturing thestrictness of employment protection laws. Range is 0 to 2, withhigher scores indicating greater strictness. Averaged over 1979–95.Source: Baker et al. (2004) update of data in Nickell and Nunziata(2002).

High unemployment benefit generosity. Raw values: proportion of a worker’sformer earnings (pretax) that is replaced by unemployment compen-sation and related benefits – for a worker with earnings at two-thirdsof the national median (for example, the 33rd percentile) in thefirst year after losing the job. Averaged over 1979–95. Source: OECDunpublished data set; see Martin (1996) for discussion.

High public employment. Raw values: public employment as a share ofthe population age 15 to 64. Averaged over 1979–95. Source: Authors’calculations from data in OECD (2004a).

Note

1. The number column is useful in larger-N studies, where researchers ought toconsider eliminating rows from consideration because too few cases conformstrongly to their conditions. With an N of only 14, narrowing the data setwould be inappropriate.

References

Baccaro, Lucio and Diego Rei. 2005. “Institutional Determinants of Unemploy-ment in OECD Countries: A Time-Series Cross-Section Analysis (1960–1998).”Discussion Paper 160/2005. International Institute for Labour Studies. Geneva.Available at: www.ilo.org.

Baker, Dean, Andrew Glyn, David Howell, and John Schmitt. 2004. Data set for“Unemployment and Labor Market Institutions: The Failure of the EmpiricalCase for Deregulation.” Unpublished.

Baker, Dean, Andrew Glyn, David Howell, and John Schmitt. 2005. “Labor MarketInstitutions and Unemployment: Assessment of the Cross-Country Evidence.”Pp. 72–118 in Fighting Unemployment: The Limits of Free Market Orthodoxy, editedby David R. Howell. Oxford: Oxford University Press.

Bassanini, Andrea and Romain Duval. 2006. “Employment Patterns in OECDCountries: Reassessing the Role of Policies and Institutions.” OECD Social,Employment, and Migration Working Paper 35. Organization for EconomicCooperation and Development. Available at: www.oecd.org.

Blanchard, Olivier and Justin Wolfers. 2000. “Shocks and Institutions and the Riseof European Unemployment: The Aggregate Evidence.” Economic Journal 110:1–33.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 89



Esping-Andersen, Gøsta and Marino Regini, eds. 2000. Why Deregulate LabourMarkets? Oxford: Oxford University Press.

Glyn, Andrew and Wiemer Salverda. 2000. “Employment Inequalities.”Pp. 35–52 in Labour Market Inequalities, edited by Mary Gregory, WiemerSalverda, and Stephen Bazen. Oxford: Oxford University Press.

Howell, David R., Dean Baker, Andrew Glyn, and John Schmitt. 2006. “Are Protec-tive Labor Market Institutions Really at the Root of Unemployment? A CriticalPerspective on the Statistical Evidence.” Unpublished.

IMF (International Monetary Fund). 2003. “Unemployment and Labor MarketInstitutions: Why Reforms Pay Off.” Pp. 129–50 in World Economic Outlook.New York: IMF.

Iversen, Torben and Anne Wren. 1998. “Equality, Employment, and Bud-getary Restraint: The Trilemma of the Service Economy.” World Politics 50:507–46.

Kemmerling, Achim. 2005. “Tax Mixes, Welfare States, and Employment:Tracking Diverging Vulnerabilities.” Journal of European Public Policy 12:1–22.


Kenworthy, Lane. 2008. Jobs with Equality. Oxford: Oxford University Press.Martin, Andrew. 2004. “The EMU Macroeconomic Policy Regime and the Euro-

pean Social Model.” Pp. 20–50 in Euros and Europeans: Monetary Integrationand the European Model of Society, edited by Andrew Martin and George Ross.Cambridge: Cambridge University Press.

Martin, John P. 1996. “Measures of Replacement Rates for the Purpose ofInternational Comparisons: A Note.” OECD Economic Studies 26: 99–115.

Nickell, Stephen. 1997. “Unemployment and Labor Market Rigidities: Europeversus North America.” Journal of Economic Perspectives 11(3): 55–74.

Nickell, Stephen and Luca Nunziata. 2002. Data set for “The Beveridge Curve,Unemployment, and Wages in the OECD from the 1960s to the 1990s.”

Nickell, Stephen, Luca Nunziata, and Wolfgang Ochel. 2005. “Unemploy-ment in the OECD since the 1960s: What Do We Know?” Economic Journal115: 1–27.

OECD (Organization for Economic Cooperation and Development). 1994. TheOECD Jobs Study. Paris: OECD.

OECD. 2004a. OECD Statistical Compendium. Edition 01#2004. Paris: OECD.OECD. 2004b. Revenue Statistics, 1965–2003. Paris: OECD.OECD. 2005. “Earnings Database.” Version dated November 5, 2005. Data set.

Unpublished. Paris: OECD.OECD. 2006. OECD Employment Outlook. Paris: OECD.Ragin, Charles C. 1987. The Comparative Method. Berkeley: University of California

Press.Ragin, Charles C. 2000. Fuzzy-Set Social Science. Chicago: University of Chicago

Press.Ragin, Charles C. 2008 (forthcoming). Redesigning Social Inquiry: Fuzzy Sets and

Beyond. Chicago: University of Chicago Press.

9780230_202573_04_cha03.tex 5/6/2008 15: 35 Page 90


Ragin, Charles C. and Benoît Rihoux. 2004. “Qualitative Comparative Analysis(QCA): State of the Art and Prospects.” Qualitative Methods Fall, 3–13.

Ragin, Charles C. and John Sonnett. 2005. “Between Complexity and Parsimony:Limited Diversity, Counterfactual Cases, and Comparative Analysis.” Pp. 180–97 in Vergleichen in der Politikwissenschaft, edited by Sabine Kropp and MichaelMinkenberg. Wiesbaden: VS Verlag für Sozialwissenschaften.

Scharpf, Fritz W. 1997. “Employment and the Welfare State: A ContinentalDilemma.” Working Paper 97/7. Max Planck Institute for the Study of Societies.Available at: www.mpi-fg-koeln.mpg.de.

Scharpf, Fritz W. 2000. “Economic Changes, Vulnerabilities, and InstitutionalCapabilities.” Pp. 21–124 in Welfare and Work in the Open Economy. Volume 1:From Vulnerability to Competitiveness, edited by Fritz W. Scharpf and Vivien A.Schmidt. Oxford: Oxford University Press.

Schettkat, Ronald. 2005. “Is Labor Market Regulation at the Root of EuropeanUnemployment? The Case of Germany and the Netherlands.” Pp. 262–83 inFighting Unemployment: The Limits of Free Market Orthodoxy, edited by David R.Howell. Oxford: Oxford University Press.

Siebert, Horst. 1997. “Labor Market Rigidities: At the Root of Unemployment inEurope.” Journal of Economic Perspectives 11(3): 37–54.

Stephens, John and David Bradley. 2005. “Employment Performance in OECDCountries: A Test of the Deregulation Thesis.” Unpublished. Department ofPolitical Science, University of North Carolina.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 91

4Do Family Policies Shape Women’sEmployment? A ComparativeHistorical Analysis of France andthe NetherlandsJoya Misra and Lucian Jude∗

Do family policies influence women’s employment rates? Differencesin women’s employment rates, particularly for women of childbear-ing age, appear to be associated with the complex of work–familysupports available to families. In this chapter we explore differencesin women’s employment in France and the Netherlands over recentdecades, taking a comparative-historical approach to examine the fac-tors that shape women’s employment. We ask whether family policiesactually drive women’s employment, or whether they may be bet-ter understood as responses to women’s employment patterns. At thesame time, we explore alternative explanations for the variation inwomen’s employment – including the economic conditions that maydrive women’s employment and cultural differences regarding gender,care, and work. While quantitative approaches may identify associa-tions across a range of countries, we argue that comparative historicalmethods are best suited to exploring historically situated relationshipsamong policy, politics, economics, culture, and women’s labor marketparticipation.

In this volume, two other chapters examine macrocomparative evi-dence regarding the contention that family policy drives changes inwomen’s employment rates. In their examination of women’s labor forceparticipation across 14 countries over four decades, Eliason et al. (2008)provide support for the argument that public child care, maternity leave,and public sector employment play a central role in shaping women’semployment participation.1 Their sophisticated research design, usinga fuzzy-set analysis and a formulation of the intention-to-treat design,carefully assesses causal relations and effects. Hicks and Kenworthy(2008) similarly examine whether child care provision, maternity leave,

91

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 92


and public sector employment shape women’s employment, usingregression analysis for the same countries and time period. Yet, whiletheir bivariate analyses suggest a strong relationship between family pol-icy and women’s employment, their multivariate analyses suggest thatthis effect may be spurious – that differences in women’s employmentmay in fact be driven by women’s educational attainment. Clearly, thereare still questions about whether and how family policies may influencewomen’s employment.

In this chapter, we examine the historical trajectory of women’semployment in two countries, France and the Netherlands, over recentdecades, and attempt to identify how these patterns have been shaped by(or have shaped) family policy. In addition to differences in employmentrates, we consider differences in full-time versus part-time employmentrates, and in employment patterns among mothers of small children andamong single mothers. In the next section, we review the theoreticalarguments in support of family policies shaping women’s employment,as well as some counterarguments regarding economic and culturalfactors. Next, we discuss our case selection, and provide some simplecomparisons between France and the Netherlands regarding employ-ment. Finally, we trace changes in employment in both countries,considering carefully whether the evidence suggests that family policyplays a causal role in shaping women’s employment.

Factors shaping women’s employment

Given that women’s employment is highest in Scandinavian countries,which also provide generous family policies, family policies appear tosupport higher levels of women’s employment. For example, the pro-vision of free or subsidized childcare appears to be a strong predictorof women’s employment (Gustafsson and Stafford, 1992; Buchmannand Charles, 1995; Esping-Andersen, 2000; Gornick and Meyers, 2003;Eliason et al., 2008; Mandel and Semyonov, 2006), although it is a betterpredictor of full-time rather than part-time employment (Kremer, 2008).Effective school scheduling – with longer school hours and lunch breakstaken at school – has a similarly positive impact on women’s employ-ment (Buchmann and Charles, 1995; Gornick and Meyers, 2003). Shortparental leave policies, particularly when they guarantee women’s jobs,help maintain women’s continuity of employment after having children(Pettit and Hook, 2005; Eliason et al., 2008).

Yet family policies do not boost women’s employment unambiguously.For example, long periods of care leave may encourage women to return

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 93

Joya Misra and Lucian Jude 93

to work, but only after their children enter public pre-school (usually atthe age of three), and to work in part-time, less career-oriented jobs (Mor-gan and Zippel, 2003; Mahon, 2005; Pettit and Hook, 2005).2 Indeed, asOrloff (2002, p. iii) argues,

Social policy is also significant, if not so much for increasing women’semployment, then for shaping the patterns of women’s employment,especially the continuity of their participation over the life course,and the conditions under which they work – as well as for helpingconstitute the stakes in gendered social policies.

Particular groupings of policies may shape the patterns of women’semployment in terms of whether women work full time or part time,work after marriage or when their children are small, and so on. Thisresearch suggests that we should be examining not only the extent ofwomen’s employment, but the type of women’s employment and vari-ations among groups of women.3 Yet women’s employment continuityis certainly highest in countries with the greatest level of policy sup-port for working mothers (Stier et al., 2001; Gornick and Meyers, 2003;Kenworthy, 2008).

While countries with more generous family policies also have higherlevels of women’s employment, it may be that higher levels of women’semployment are driving family policy provision. Indeed, Huber andStephens (2000) show that women’s labor force participation is an impor-tant determinant of both public funding and the delivery of welfare stateservices, suggesting that as women enter the labor force they need relieffrom traditional caregiving duties, and “make demands on the state forbetter health, education, and welfare services, regardless of the institu-tional and political context” (Huber and Stephens, 2000, p. 334). Huberand Stephens (2000) also note that women pressure for the expansion ofwelfare state services in their roles as social service workers. Therefore,higher levels of women’s employment may produce generous familypolicy, generous family policy may generate higher levels of women’semployment, or both processes may be occurring simultaneously.

An added wrinkle is that family policies do not neatly map onto dif-ferences in women’s employment rates. Women’s employment is veryhigh in the United States, which provides very little family policy sup-port, suggesting that other factors clearly must play a role. Rather thanassuming that policies shape women’s employment decisions, a sec-ond explanation might highlight economic factors. Increased wagesand better benefits for women, as well as favorable taxation policies,may encourage women to enter employment (Kenworthy, 2008; Misra

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 94


and King, 2005). Indeed, where tax policies reward women’s employ-ment, women are more likely to enter and remain in the labor market(Kenworthy, 2008; Sainsbury, 1999).4 At the same time, where men’s realwages have dropped, women may also be more likely to enter the labormarket out of economic need.

Economic structure may also influence women’s employment. Forexample, where manufacturing wanes and the service sector grows,the economy may draw in more women workers. Indeed, womenare particularly likely to be drawn in as public sector workers (Huberand Stephens, 2000). Active labor market policies may draw womeninto employment, and may include education and training programs,or policies meant to create additional jobs through developing part-time employment.Indeed, the availability of part-time employment mayserve as a spur to women’s employment, depending on the context (Jen-son, 1996).5 At the same time, labor surpluses may discourage women’semployment, and programs may be put into place that reward womenfor caretaking during periods of slack labor markets. Therefore, eco-nomic factors may play a key role in explaining how and where womenare drawn into employment, and whether they are primarily viewed asfull-time or part-time workers.

Finally, a third explanation focuses on differences in women’s prefer-ence for employment (Hakim, 2000, 2003, 2004) or in cultural supportfor women’s employment (Pfau-Effinger, 1998, 2004; Kremer, 2008).6

This literature notes that while “women are treated [by many scholars]as rational individuals who orient their behaviour according to financialincentives” (Pfau-Effinger, 1998, p. 147) such as the costs of childcareand tax penalties for dual earner couples, long-lasting cultural traditions,values, and norms may play a role (Sackmann, 1998; Pfau-Effinger, 1998,1999, 2004; Kremer, 2005, 2008). Welfare states may promote different“ideals of care” which reflect different – and dynamic – notions of what isappropriate (and gendered) care (Kremer, 2008). As Kremer (2008) argues,structural supports and economic incentives must fit into moral idealsof care in order to effectively promote women’s employment. Therefore,this might suggest that presence of family and labor policies may notshape women’s employment patterns simply, and that larger culturaltraditions help explain variations in women’s employment.7

We examine Dutch and French women’s employment patterns toassess whether the historical evidence supports these explanations ofwomen’s employment. Of course, these explanations leave out otherimportant factors – party politics, women’s movements, and other socialmovements – that play a direct role in shaping policy choices. However,

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 95


we focus our attention on whether family policies, economic condi-tions, or gendered ideals of care best explain variations in women’semployment for these two cases.

Method and case selection

In order to explore the factors that shape employment patterns in greaterdetail, we take a comparative historical approach to explore two cases –France and the Netherlands. We rely on secondary data sources to con-struct contextualized comparisons of women’s employment in these twocountries. A focus on two carefully selected cases gains in depth what itlacks in breadth, giving it certain advantages over large-n quantitativestudies. First, the emphasis on processes over time highlights similaritiesand differences across cases on a range of different variables, rather thanassuming unit homogeneity. Also, this approach captures the duration,relative timing, and intersection of events, which is crucial for makingcausal assessments. Indeed, events occur both in time and over time, andhistorically grounded comparisons can detect these temporal nuances ina way large-n studies cannot.

Further, as Mahoney and Rueschemeyer (2003, p. 13) state,

Because comparative historical investigators usually know each oftheir cases well, they can measure variables in light of the broadercontext of each particular case, thereby achieving a higher level ofconceptual and measurement validity than is often possible when alarge number of cases are selected.

Social phenomena are typically characterized by “causal complexity”(Ragin, 1997), wherein the same variable often has different effects inheterogeneous contexts, and the same outcome is produced by differentcausal processes. Comparative historical analysis captures these con-textual particularities through process tracing, achieving an intensivedialogue between theory and evidence. The result is a more completepicture of the relevant social and political forces at play than is possiblewith more abstract theorizing.

While the direct goal of comparative historical research is not to gener-ate universally applicable knowledge, multiple observations both withinand across cases make this approach an effective means to “test” andrefine existing theories, formulate new concepts, and develop novelexplanations. These contributions are based upon critical historicaldetails that may be overlooked or difficult to measure in large-n studies.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 96


Yet large-n and small-n studies are complementary precisely because oftheir unique advantages. In constructing logical causal arguments aboutfactors impacting women’s employment, we draw on arguments fromlarge-n studies to legitimate our claims. We do not, however, select sup-porting data merely to bolster one particular analysis over another. Nordo we presume a deterministic set of forces. We construct a coherent pic-ture of large-scale social processes in specific contexts that make sensegiven the available evidence. Certainly, there is the risk that our sourcesdo not cover all of the important factors, which is similar to the risksurvey researchers face regarding missing variables when analyzing sec-ondary data. However, using a variety of different sources written byscholars in a range of disciplines, and attending to the available quantita-tive and qualitative data for each country, serve as a “check” on our data.Further testing and refinement of our arguments in other contexts andwith other methods will continue the ongoing process of constructing“better” social theories for the advancement of knowledge.

We have chosen France and the Netherlands as our two cases becausethey share some similarities, while they also differ in important ways.According to Esping-Andersen’s now-classic model (1990, 1999), inwhich he shows that nations tend to cluster in certain groups in termsof policy creation and outcomes, both France and the Netherlands fitamongst the conservative welfare regime (although the Netherlands issometimes typed as social democratic) (Esping-Andersen, 1990; Hobson,1994; Knijn, 1998). According to Esping-Andersen, in this regime thestate uses policy to uphold status differences and preserve the traditionalfamily. Of course, there are also important differences between the twocountries, including very different political cultures. While the Dutchhave subscribed to a notion of subsidiarity, which assumes that the centralauthority (state) should perform only those tasks that cannot be per-formed more effectively at a more immediate or local level, the Frenchhave a longer tradition of strong state intervention (and tolerance forstate intervention into the economy). Therefore, the Dutch and Frenchwelfare states have been structured differently, and reflect differingnotions about the relationship between the states, local communities,and families.

Yet, the puzzle we explore is the difference between women’s employ-ment in France and the Netherlands – and its recent convergence.Figure 4.1 describes differences in women’s employment rates in Franceand the Netherlands. While historically French women had higher lev-els of employment, the growth in Dutch women’s employment hasbeen unusually steep over the last two decades, and Dutch women’s

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 97


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002

FranceNetherlands

Figure 4.1 Employment rate for women as a proportion of all women, 15–64Source: ∗Data from OECDs Corporate Data Environment (CDE) (2006).

employment levels now surpass those of French women. Yet there areclearly differences in employment rates by age groups. For example,employment has been increasing for French women between the agesof 25 and 54 since 1970 and decreasing for women under 25 andover 54. In comparison, Dutch women in all groups have seen anincrease in employment rates since 1970, although women over 55 havesignificantly lower (though increasing) employment rates (Kenworthy,2008).

Given that our interest is in whether family policies may be shapingwomen’s employment, we focus our attention on women between 25and 54, women who are particularly likely to be facing care responsi-bilities that may be mitigated by family policies. Figure 4.2 describesemployment for this group for the two countries since about 1970. Thisgraph makes clearer the significant rise in women’s employment for bothcountries, along with the convergence in rates for the two countries.While the increase in Dutch women’s employment is steeper, giventhat French women’s employment rates were at much higher levels atthe beginning of this period, slightly more than 70 percent of womenbetween 25 and 54 were employed in both nations in 2004. This leavesus with two puzzles: (1) Why were French women’s employment rates

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 98


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004

FranceNetherlands

Figure 4.2 Employment rates for women between 25 and 54, 1968–2004Source: ∗Data from OECDs Corporate Data Environment (CDE).

initially so much higher than Dutch women’s employment rates? (2)What explains the remarkable increase in Dutch women’s employment?

One answer to the second part of this puzzle may be in understand-ing the different rates of full-time and part-time employment in thetwo countries. Figure 4.3 describes the difference in women’s full-timeand part-time employment rates across the two countries. Since 1983,women’s engagement in part-time employment has increased in bothFrance and the Netherlands. Yet the two countries are still more differentthan alike. Since about 1983, approximately half of French women haveworked full time, while a much smaller proportion (between 12 and 16percent) has worked part time. In the Netherlands, however, more womenwork part time than full time, and part-time employment rates havegrown more quickly than full-time. Indeed, in 2004, 43 percent of Dutchwomen worked part time, while 31 percent worked full time. Clearly,simple employment rates do not tell the entire story – Dutch women’semployment has grown in large part through their engagement in part-time employment. Indeed, in comparison to other European countries,part-time work is significantly higher than average in the Netherlands,and lower than average in France (OECD, 2001, p. 96).

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 99


0

0.1

0.2

0.3

0.4

0.5

0.6

1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

France P-TFrance F-TNeth P-TNeth F-T

Figure 4.3 Full-time and part-time employment rates for women aged 25–54,1983–2004Source: ∗Data from OECDs Corporate Data Environment (CDE) (2006).

Another way to consider the meaning of these numbers is to considerwhat proportion of employed women work part time. In 2004, whileapproximately 24 percent of employed French women worked part time,60 percent of employed Dutch women worked part time. If we com-pare the average weekly hours for employed women, French womenwork on average 34 hours a week, while Dutch women work 25 hoursa week (Hantrais and Letablier, 1997, p. 140). Even within part-timeworkers, we may see differences across the two countries. For example,among women working part time, approximately 25 percent of Dutchwomen work 25 hours a week or more, as compared to 40 percent ofFrench women (Hantrais and Letablier, 1997, p. 140). Interestingly, part-time women workers in France are also more likely to report that theyare working part time involuntarily and would prefer to work full time(see Figure 4.4). This suggests a significant difference between the waywomen’s work is conceptualized in France and the Netherlands.

Employment rates also vary across the two countries for mothers ofyoung children. In 1990, 61 percent of French mothers with childrenunder six were employed, compared to 37 percent of Dutch motherswith children under six. By 2002, the numbers for Dutch mothers had

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 100


0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004

France

Netherlands

Figure 4.4 Involuntary part-time work as a proportion of part-time work forwomen, 25–54, 1992–2004Source: Data from OECDs Corporate Data Environment (CDE) (2006).

changed dramatically, with 69 percent of Dutch women with childrenunder six employed, as compared to 62 percent of French women withchildren under six. However, high levels of part-time employment inthe Netherlands help explain this increase. Indeed, 81 percent of Dutchmothers with children under six are employed part time, in comparisonto only 26 percent of French mothers with children under six (OECD,2005, p. 60, table 4.1). Therefore, Dutch mothers may be somewhat morelikely to work, but they are primarily working part time.

Similarly, employment rates vary in interesting ways depending onfamily forms. Families with children under six are a good illustrative casehere as well. In 1999, 31 percent of French mothers in two-parent hetero-sexual families with children under six worked full time while 20 percentworked part time. In comparison, only 4 percent of Dutch mothers intwo-parent heterosexual families with children under six worked fulltime while 48 percent worked part time. Clearly, very different employ-ment patterns exist. Likewise, 35 percent of French lone mothers withchildren under six worked full time while 14 percent worked part time;in comparison, only 6 percent of Dutch lone mothers with childrenunder six worked full time while 32 percent worked part time (OECD,2001, p. 135, table 4.2; also see Chambaz, 2001). Over time, Dutch lonemothers with young children have entered the labor force in growing

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 101


numbers (Bussemaker et al., 1997; Pott-Buter, 1993). On the other hand,French lone mothers with children under six have actually reduced theiremployment since 1990 (OECD, 2001, p. 134, table 4.1).

In the following three sections, we explore these broad differencesin greater depth. What explains the stark initial differences in Frenchand Dutch employment rates? What explains the convergence inemployment rates for women between 25 and 54 over time? Whatexplains the difference in part-time employment rates across the twocountries? Do differences in family policy explain these variations?Do differences in economic factors and conditions? Or do culturaldifferences make the most persuasive argument?

Family policy

Family policy, aimed not only at increasing birth rates, but also at inte-grating women into the labor market, has a long history in France, datingback to the 1930s (Jenson 1990; Misra 1998).8 These policies supportedfamilies generously, although some financial incentives were premisedon the idea that women would wish to stay home with very youngchildren. While women’s employment had historically been fairly high,its steady increase since the 1970s – particularly for women of childbear-ing age – has been in part possible due to the high-quality child careavailable for working parents. High levels of women’s full-time employ-ment in France are premised on the availability of this child care. While(relative to other countries) fairly generous welfare benefits existed formothers (including a variety of measures directed at single mothers),family policy since the 1960s has primarily limited benefits to supportmothers staying home for care, instead encouraging mothers’ workforceparticipation and wage reliance.9 A range of additional work–family rec-onciliation policies put into place during the 1970s included reformsto family allowance and tax laws so that two-income families were nolonger penalized (Jenson, 1990; Hantrais, 1994). The benefit levels forthe “single salary allowance,” which explicitly supported a male bread-winner family, were reduced, and the allowance was finally eliminatedin 1978 (Letablier, 2003).

During the same period, Socialist President François Mitterrandexpanded the provision of child care for children under the age ofthree in order to promote gender equality in access to employment andencourage women’s employment (Daguerre and Taylor-Gooby, 2003;Letablier, 2003).10 Maternity leave was also expanded; by the 1980s,maternity leave was six weeks pre-birth and ten weeks after, at 90 percent

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 102


income replacement (Jenson, 1990). Official government policy statedthat women should have a “free choice” between working and stayingat home. By 1982, 37 percent of two-year-olds, 83 percent of three-year-olds, and 97 percent of five-year-olds were in child care in the publiccare system (Jenson, 1990, p. 118). The majority of the costs for childcare were borne by the state; even the crèche system for babies andtoddlers was heavily subsidized. Child care was also expanded to pro-vide wraparound care – afterschool care, lunch, and summer programs –prioritizing children with working parents (Jenson, 1990). Yet despitethis rhetoric of choice, single mothers have had fewer options than mar-ried mothers. While single mothers were eligible for benefits that supporttheir caretaking, policy encouraged them to work, giving them priorityin training and childcare. Indeed, single mothers were much more likelyto use the crèche system.

However, since the 1980s, some family policy reforms have workedagainst employment, particularly for mothers of small children. Forexample, rising unemployment led to policies that encouraged womento stay at home, including family supplements and a new allowancefor single parents (Jenson, 1990; Hantrais, 1993). The parental careleave (Allocation Parentale d’Education), created in 1985, expanded in1986, and further expanded and popularized in 1994, provides benefitsfor previously employed parents with children to care for their chil-dren at home.11 This policy may help explain why the rate of increasein women’s employment slowed during the late 1990s, since 99 per-cent of those who use the leave are women (Kenworthy, 2008; Heinenand Martiskainen de Koenigswarter, 2001).12 Women are more likely totake the leave due to its fairly low replacement rate, and larger culturalexpectations. While women do appear to eventually return to the labormarket, these longer leaves likely have negative effects on their long-termemployment trajectories.

Child care remains free and universally available from the age ofthree. Almost 100 percent of children between the ages of three andsix, as well as 75 percent of children between two-and-a-half andthree years of age, are enrolled in the ecole maternelle. Yet, in theearly and mid 1990s, the Right also developed new benefits and taxrelief to support private child care costs for young children, benefit-ing middle-class parents, while weakening the crèche system necessaryfor working-class parents (Hantrais and Letablier, 1997; Morgan, 2005;Daguerre and Taylor-Gooby, 2003). While the Socialist Party in 2001expanded the number of public child care places (Daguerre and Taylor-Gooby, 2003), it also developed a new benefit to pay for private child

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 103


care for women who are returning to work (Letablier, 2003).13 Thestate now provides less support to the crèche collective, the publiclysubsidized centers, and more young children are now in subsidizedand regulated family daycare (crèche families) than in crèche collective(Morgan, 2005). This subsidized care may encourage part-time employ-ment, as it covers the cost of only two to three days of child care(Morgan, 2005, p. 253). Clearly, French policy on child care has becomemore mixed.

Yet, as Leslie King (1998, p. 47) argues, “In general, French fam-ily policy has managed, compared to other states’ family policies, tohelp facilitate combining labor force participation and parenting; andwomen in France tend to have fewer interruptions in their work livesthan women in other countries.” While many French mothers exit thelabor force briefly when their children are infants, these interruptionsremain shorter, on average, than in many other European countries(Laufer, 2003).14 The short 2002 paid paternity leave has been very suc-cessful; though the replacement rate does not fully replace the salaryfor many fathers, 94 percent of fathers take the leave for the maximumperiod of 11 days in the first four months (18 days for multiple births)(Mazur, 2003; Laufer, 2003).

In comparison, Dutch family policy was much less developed duringthe first several decades of the twentieth century, in part due to the strongprinciple of subsidiarity, which encouraged care provision within thefamily, rather than through the state. Yet, in 1965, the General WelfareAct was inaugurated, and the state was delegated a stronger responsibil-ity for ensuring citizens’ well-being, or welzijn. Family policies promotedthe principle that all citizens should have equal access to welfare benefitsand services, yet policy also reflected the primacy given to maternal care(Korteweg, 2005). For example, welfare policy provided generous socialbenefits for caregiving by single mothers, who were included in the gen-eral cultural ideal of the caring housewife (Pfau-Effinger, 1999). Socialrights were tied to the principle of family maintenance, and tax bene-fits were given to the (usually male) family provider (Sainsbury, 1994).When married women did work for pay, couples were jointly taxed athigh rates without the option of separate taxation.15

In addition, child care policy was grossly underdeveloped. Duringthe 1950s, the only support for public childcare came from the Dutchwomen’s movement (Morgan, 2006). Until 1965, child care was pro-vided primarily by church organizations as a form of poor relief only.Church leaders discouraged mothers’ labor market involvement, whichthey believed undermined traditional family structures (Bussemaker,

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 104


1998; Gustafsson and Stafford, 1995). In the 1960s and 1970s, var-ious types of child care developed, including pre-school playgroups,child day care centers, and out-of-school child care centers (OECD,2000, 2002). However, these arrangements were oriented toward chil-dren’s social development rather than support for working parentsand women’s autonomy, and were available on a mostly part-timebasis (Bussemaker, 1998; Morgan, 2006). As Morgan (2006, p. 85)notes, even Dutch Social Democratic and Communist parties assumedthat mothers “should devote themselves full-time to the care of theirchildren.”

In the mid-1970s, a left-wing coalition government sparked a newdiscourse on formal gender-equality policies: “Left-wing parties andsome liberal politicians placed increasingly more emphasis on women’sinterests as a part of the childcare debate. Various politicians remarkedthat equality policies had no chance of success without accessibleand affordable childcare” (Bussemaker, 1998, p. 82). That discursiveopening, though, was short-lived as more conservative governmentsreturned to power in the 1980s and Christian-democratic ideology againrose to prominence, which branded women’s labor market involve-ment as problematic. Child care provision thus remained fairly limitedthroughout the 1980s even though women’s labor market participationincreased dramatically.16

At the end of the 1980s, a booming economy, a shortage of labor, anincrease in the number of elderly and single households, and a decreasein the number of taxpayers led the government to expand family policy.With the goal being to increase the labor market participation of bothsingle and partnered women, policy provision shifted toward the needsof mothers (and the labor market) (Gustafsson and Stafford, 1995; Knijnand Selten, 2002; OECD, 2000; Plemper, 1996). A 1990 tax reformreduced the basic tax allowance for breadwinners, thus lowering disin-centives for second-earners to take up paid work (Visser, 2002, p. 33).That same year, a coalition cabinet made up of Christian Democratsand Social Democrats passed the Stimulation Measure on Childcare toexpand childcare facilities (Bussemaker 1998).17 The policy emphasizedchild care provision as a responsibility that must be shared by the gov-ernment, parents, and employers.18 Parental contributions are calculatedaccording to income, yet the majority of Dutch families who use formalchild care provisions have an average or above-average income (OECD,2000).19 Child care is subject to collective corporate arrangements. Only80 percent of employers fall under Collective Agreements; by 2003, 76percent of collective agreements included childcare provisions (Morgan,

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 105


2006, p. 172). However, many employees end up paying full price forchild care (Kremer, 2005). Child care policy thus benefits mostly edu-cated, middle- and upper-class men and women in high-level positions(Grünell, 1999). Parents, though, commonly face long waiting listsas they compete for a limited (though growing) number of childcareplaces.20

Despite serious shortcomings, child care in the Netherlands hasimproved dramatically over the years. The percentage of children frombirth to three years of age in state-subsidized child care was 2 percent in1988, 8 percent in the mid-1990s, and almost 22 percent around 2001;for children aged three to six years the rate was only 6 percent in themid-1990s, which is partly explained by the fact that most children startschool early (Knijn and Kremer, 1997; Kremer, 2008; Morgan, 2006;OECD, 2000). Yet there remain few options for very young children,with enough child care places for only around 13 percent of childrenunder four years old, which constrains the options available to Dutchmothers (OECD, 2002, p. 97). Since 2005, parents have been reimbursedfor a portion of child care costs through the taxation system, while thecenters no longer receive state subsidies. These changes were made inorder to stimulate supply through demand subsidies, but have led tounusually high childcare costs for Dutch parents.

The Netherlands legislated its first leave arrangements in 1990. Sincethat year, all mothers are entitled to 16 weeks of paid maternity leave,with the government making payments up to a bit over the average wagerate in the economy. In 1991 the Dutch government introduced parentalleave, providing leave part time on an unpaid basis. For a six-monthperiod until the child is eight, parents are entitled to work part time(at least 20 hours per week) while caring for their child(ren) (Jaumotte,2003; Knijn and Kremer, 1997; Morgan and Zippel, 2003; OECD, 2000,2002; Plantenga and Hansen, 1999).21 Dutch parents in two-adult fam-ilies can choose to take parental leave simultaneously, yet it is rarefor fathers to make use of prolonged child-related leave.22 In addition,the number of mothers using parental leave is significantly lower thanthe number of mothers using maternity leave. Strict eligibility criteriafor parental leave, which maintain that employees must work for anemployer for at least 12 months, partly explain this discrepancy (OECD,2002, p. 130). Paternity leave only became available as a legal right in2001 with the introduction of the Work and Care Act, and then providedfathers a mere two-day absence from work.23 This combination of leavearrangements seems to have had a significant effect on employment,as the employment rate of Dutch women with a child under age three

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 106


increased from 42 percent in 1992 to 70 percent in 2003 (Kremer, 2005,p. 3). The design of leave programs on a part-time basis further explainsthe popularity of part-time employment in the Netherlands, especiallyamong mothers with children.

Dutch family policy also promotes the ideal of “parental sharing,” theequal participation of men and women in both paid and unpaid work.Billboards depicting prominent Dutch men engaged in care tasks are partof a government-sponsored media campaign. Fathers’ roles as carers areemphasized in public discourse, law, and social policy (Knijn and Selten,2002). Nevertheless, while Dutch men are more involved in caregivingand more likely to take parental leave than in many other countries(Kremer, 2008), women still perform three-quarters of the care work infamilies with children (Knijn and Selten, 2002).

Does family policy explain the similarities and differences we havenoted? Clearly, French family policy has historically been more gen-erous, and more oriented toward supporting women’s employment,which may help explain the initial difference in employment rates.Dutch family policy has also expanded over the years, which may helpexplain the greater convergence. Yet, as Figure 4.2 suggests, increasesin Dutch women’s employment actually appear to have begun beforethe institution of these policies – suggesting that family policy hasresponded to increased women’s employment, rather than creating it.And there remain fairly important differences in the generosity andavailability of family-work reconciliation policies. These differences mayhelp explain the significant divergence, illustrated in Figure 4.3, in full-time employment rates for the two countries. As Lane (1995, p. 13)notes, work–family measures “explain why a relatively large proportionof French women are able to have continuous working life and whythey are not forced to switch to part-time employment after childbirth.”In the absence of high-quality, affordable child care, full-time women’semployment in the Netherlands remains difficult to achieve.

Economic factors and labor market policies

Economic factors have substantially shaped French women’s incorpora-tion into the labor market. Before World War II, French women’s laborforce participation was high as a repercussion of the high death rateof men during World War I ( Jenson, 1990). Because industrializationwas fairly gradual, women were primarily employed in family-basedforms of production, including agriculture, small businesses, and crafts(Duane-Richard, 1995). During the postwar period, married women’s

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 107


employment dipped, in part because of the decline of the agriculturalsector ( Jenson, 1988; Tilly and Scott, 1978; Duane-Richard, 1995). How-ever, the shortage in skilled labor in the postwar period meant thatwomen were not formally excluded from employment. By the 1960s,changes in the economy meant that women were increasingly beingdrawn into manufacturing, as well as the expanding service sector, andwomen’s employment increased rapidly during the 1960s and 1970s.24

Many women were employed in low-wage service and industrial sectorjobs, although women also entered more protected jobs in the publicsector (Jenson, 1988; Beechey, 1989). By 1974, women accounted forthree-quarters of semi-skilled and unskilled manufacturing jobs (Jenson,1988, p. 158).

Labor market policies helped reinforce and remake women’s employ-ment during this period. The French constitutions of 1946 (4th Republic)and 1958 (5th Republic) affirmed the principle of women’s right towork, and workplace equality for women.25 Women were given accessto civil service jobs, where pay scales were integrated, while the Frenchminimum wage law helped ensure a wage floor for women. Research per-formed by the Mouvement Démocratique Féminin (established in 1962 byprominent Left feminists) and the Comité du Travail Féminin (establishedin 1965, as an advisory committee on women’s employment within theDepartment of Labor) as well as others indicated that wage inequalitywas due to job segregation, unequal access to education, training, anddiscrimination (Laufer, 2003; Revillard, 2006). As a result, the legislaturepassed laws banning wage discrimination in 1972, requiring equal payfor equal work in 1973, and banning discrimination (including preg-nancy discrimination) in hiring and firing in 1975. In the 1970s, thegovernment also introduced professional training programs meant toequalize women’s opportunities, provide women greater opportunitiesfor promotion, and facilitate re-entry for women who took time out ofthe labor market (Jenson, 1990).26 However, as Revillard (2006) notes,these issues were on the political agenda due to the larger context of alabor force shortage.

Following the employment crisis of the mid-1970s, men’s labor mar-ket participation dropped, but that of women continued to rise, sincewomen served as a more tractable labor force, accepting jobs withlower wages, fewer benefits, and fewer opportunities for advancement(Duane-Richard, 1995, p. 43; Jenson, 1988, 1990, p. 109). By the1980s, men continued to earn higher wages and be more likely to bepromoted, particularly in the private sector and among professionaloccupations, while women were more likely to be “stuck in dead-end

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 108


jobs and paid poorly throughout their working lives” (Jenson, 1990,p. 109).

Until the 1970s, there was little part-time employment in France, par-ticularly in comparison to other European nations, and most womenworked full time. In the 1970s, due to the employment crisis andshortage of full-time jobs (Duane-Richard, 1995), the state began empha-sizing active labor market policies and employment creation, includingthrough part-time work. From 1973 to 1981, there was a 42 percentincrease in part-time work for women (Bakker, 1988). While the Socialistsinitially discouraged part-time work, they eventually put incentives inplace to create part-time employment, by allowing employers to reducetheir social security contributions for part-time jobs (Malo et al., 2000).As a result, more than two-thirds of new jobs taken by women in the1980s were part-time ( Jenson, 1990).27 However, the Socialists also putprotections in place for part-time workers regarding rights, wages, andbenefits similar to those held by full-time workers, and part-time workerswork relatively long hours ( Jenson, 1996; Orloff, 2002). Since part-timework is thus more “expensive” for employers, levels of part-time workremain lower in France than in many other countries (Laufer, 1998;Hantrais and Letablier, 1997).28,29

During the 1980s, a number of reforms, led by Yvette Roudy of theMinistry of the Rights of Women (created in 1981 by the Socialists),helped protect and clarify the rights of women workers. These reformsincluded the 1983 comprehensive anti-discrimination law, the loi Roudy,which filled in the gaps of earlier legislation and was “designed totackle three problem areas: access to training and promotion, pay, andunemployment” (Hantrais, 1994, p. 95).30 The law equalized opportuni-ties, strengthened equal pay legislation, gave trade unions more powerin addressing gender equity issues, and required employers with morethan 50 employees to report annually the relative situation of men andwomen regarding both wages and positions and develop plans regard-ing recruitment, training, and promotions (under the rubric of “positiveaction”) (Hantrais, 1993, 1994; Laufer, 2003).31 This law reflects com-parable worth policy, by basing pay specifically on “a comparable levelof professional knowledge, of equivalent qualifications, experience, andresponsibilities” (Laufer 2003, p. 430). Feminists have continued to pushfor and win expansions in training programs and, since 2001, havehelped create a detailed process for negotiations around gender equalitywithin firms, including real sanctions for firms that do not conform tothe reporting requirements of the loi Roudy (Mazur, 2003; Laufer, 2003).

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 109


In an effort to increase women’s entrance into non-traditional fields,the state also established a contract that subsidized wages and train-ing costs for women in training programs for jobs dominated by men,and worked with the Ministry of Education to create scholarshipsdirected at bringing women into non-traditional careers (Laufer, 1998,2003; Jenson, 1988). As a result, France has a higher proportion ofwell-educated women, as well as higher levels of women working in“professional and technical” employment than in comparable countries(Laufer, 1998; Lane, 1995). Yet gender segregation remains a real issue forFrench women (Blau and Duncan, 2002).32 While wage inequality hasgenerally been decreasing, fairly high levels of labor market inequalityremain, both in terms of occupational gender segregation, and in termsof wages – particularly at the top of the wage distribution (Pfefferkorn,1996; Laufer, 2003).33

Finally, as noted previously, rising unemployment rates have also ledto changes in French family policies, such as parental care leave, thatencourage women to stay at home. Because the leaves have been con-ceptualized as a job-creation measure (as workers go on leave), theyhave been accepted by both the Left and Right, even though the leaveshave clearly reduced the number of women with two children in thelabor force and reinforced a more traditional gender division of laborin the home (Fagnani, 1995; Silvera, 2000; Laufer, 2003; Heinen andMartiskainen de Koenigswarter, 2001; Morgan and Zippel, 2003).

For many years Dutch labor market policies were discriminatory towomen, encouraging or coercing women to remain at home so thatmen could secure needed jobs. For example, in 1924 married womenwere banned from taking public service jobs, including jobs as teachers(Gustafsson and Stafford, 1995; Pott-Buter, 1993).34 This ban, supportedby most unions, was withdrawn in 1957 but persisted in many munici-palities even ten years later. Employers could legally fire a woman becauseof pregnancy, childbirth, or marriage until new legislation was passed in1973; Swedish women, by contrast, secured such protection in 1939(Pott-Buter, 1993). Further, as Visser (2002, p. 28) notes, “Under theregime of centrally guided wage policy, in force between 1945 and 1962,women were paid less than men in the same jobs.”35 A statutory mini-mum wage was instituted in 1969 that applied equal rates to men andwomen, but until 1971 employers received a dispensation to pay lowerwages to women. Until 1993, the minimum wage did not apply to the(mostly female) employees who worked less than one-third of the usualworking week (Visser, 2002).

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 110


The relative affluence of the Netherlands enabled most people to sus-tain relatively high living standards on a single income. But during aperiod of rapid modernization after World War II, women’s opportuni-ties for paid employment began to change. A shortage of workers ledto increased demand for labor while the public service sector grew, com-pelling some companies to try to attract female personnel (Karsten, 1995;Pfau-Effinger, 1998, 1999; Plemper, 1996). The preferred strategy was toattract Spanish, Turkish, and Moroccan men to compensate for the laborshortage (Bussemaker, 1998; Karsten, 1995). However, as early as 1956,some companies began offering childcare for working mothers withyoung children. By 1973 corporations provided about 10 percent of thenation’s negligible supply of child care. However, when the labor short-age disappeared during the 1970s, firms closed their childcare centers(Bussemaker, 1998).

From the oil crisis of 1973 until the mid-1980s, the Netherlands suf-fered a severe economic crisis, which resulted in high unemployment.The government initially responded with efforts to decrease labor supply,using such strategies as early retirement, repatriating foreign workers,and encouraging women to stay at home.36 However, pressed by eco-nomic necessity, the government restructured the welfare state and putinto place active labor market policies directed at women. Activity bywomen’s movements contributed to this shift. In the 1970s, feministsput a call for change in the national policy on women on the politicalagenda. In 1974 a commission for the promotion of gender equality wasestablished at the highest political level. In the 1980s the governmentembraced gender equality and the possibility of women to lead an inde-pendent existence, as well as a just division of care and housework, asnew political aims (Outshoorn, 1995; Pfau-Effinger, 1999).

Welfare restructuring during this period led to the dismantling of someelements of the welfare state, a freezing of wages, and the promotion ofpart-time jobs.37 The government presented part-time work as a way forwomen to reconcile work and family, and also as an instrument for facil-itating more flexibility for employers while decreasing unemployment(Plantenga, 1996). These measures worked to turn the Dutch economyaround, and by the end of the 1980s a booming economy and a shortageof labor helped support the expansion of family policies in order to bringwomen workers into the labor market (Knijn and Selten, 2002). In thetight labor market of the 1990s, the promotion of part-time work helpedmatch labor demand with labor supply by giving employers flexibility,expanding the pool of potential workers, and helping employees meettheir goals for work and family balance (Berg et al., 2004).

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 111


Two laws regulate part-time work in the Netherlands. The Equal Treat-ment Act, effective since 1994, mandates that part-time work must betreated the same as full-time work in terms of wages, benefits, and train-ing. This act also prohibits discrimination on the grounds of gender,marital status, race, nationality, religion, belief, political opinion, andsexual orientation. Havinga (2002) reports, however, that this legislationhas limited effectiveness in regulating gender-based discrimination dueto deficient knowledge of the law, incompatibility with existing rou-tines and traditional ways of doing things, expected costs of compliance,the passive approach by the Equal Treatment Commission, and inade-quate support and legal assistance for complainants. Indeed, comparedto men, women are concentrated in a relatively small segment of thelabor market.38 In addition, as in other countries, both women and menearn lower wages if employed in female-dominated occupations, indi-cating a gender bias independent of education, skills, and responsibility(De Ruijter et al., 2003). Further, 68 percent of Dutch women in the labormarket work part time, compared to 18 percent of men. Of women part-time workers, 32 percent work less than 20 hours per week (Berg et al.,2004; Kremer, 2001; OECD, 2002).

The second law that regulates part-time work is the Adjustment ofHours Act, which came into effect in July 2000. This act provides work-ers with the legal right to periodically request reductions or increasesin weekly working time, for whatever reason. This option, though,mainly benefits full-time workers – rarely mothers – who want to reducetheir working hours. Employers can refuse workers’ requests if they candemonstrate that fulfilling the request will create a hardship for the com-pany, but most requests are granted as the burden of proof lies withemployers (Berg et al., 2004; OECD, 2000, 2002).39

Since 1996, welfare reform has required lone mothers to find part-time work – defined as 12 hours per week – once their youngest childreaches age five.40 Beyond 12 years old, mothers are required, like every-one else, to find full-time work. However, the 1996 law has been ratherunsuccessful in getting lone mothers off social assistance. Lone parentson social assistance generally have low educational attainment and poorjob prospects, so part-time work is unlikely to make a material differencein their lives. Further, full-time work conflicts with the belief held bymany mothers that they alone are best suited to care for their children.Indeed, a lack of access and uncertainty about the quality of public childcare, and a lack of after-school care facilities makes work difficult forthese mothers.41 Since local authorities wield significant discretionarypower, they often grant exemptions to lone mothers based upon lenient

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 112


interpretations of paid work rules (Korteweg, 2005; OECD, 2002). Asurvey conducted by Knijn and van Wel (2001) suggests that 60 per-cent of lone mothers with children above age five are exempted from“normal” job search requirements. Nevertheless, over the years Dutchlone mothers with young children have entered the labor force in grow-ing numbers (Bussemaker et al., 1997; Pott-Buter, 1993), perhaps due toaltered cultural perceptions and increased levels of educational attain-ment (de Jong Gierveld and Liefbroer, 1995; Jaumotte, 2003). Yet mostlone mothers who are employed work only part time, which lowers theirchances of achieving economic independence.42

Do economic conditions and labor market policies explain the simi-larities and differences we have noted? Clearly, French and Dutch policymakers have responded to economic conditions – bringing women intothe labor market when labor supply has been low, and encouragingpart-time employment as an active labor market strategy. At the sametime, family policy has been used as a labor market policy – pushingwomen out of the labor market and into care when unemployment hasbeen problematic. Yet these strategies do not fully explain the initialdifferences in Dutch and French women’s employment, or the recentconvergence. Indeed, as Figure 4.2 makes clear, throughout this periodwe see an increase in women’s employment in both countries. Thesestrategies do a better job of explaining the increase in Dutch women’spart-time employment over the last two decades than the fairly stablelevel of French women’s part-time employment (these trends are evidentin Figure 4.3). As explained in OECD (2002, p. 134), the nature of childcare provision, leave legislation, and work-time policy fosters part-timework in the Netherlands:

The high costs and childcare capacity constraints in the Netherlandsmean that formal childcare is used on a part-time basis, if at all, and bytheir nature informal care arrangements are often used on a part-timebasis. In addition Dutch legislation of different sorts facilitates the useof part-time employment solutions. Equal rights to part-time workersand other workers on flexible contracts, the Adjustment of Hours Actand the design of leave programmes on a part-time basis illustrate therole of part-time work in the Dutch policy model. These policy signalscontribute to explaining the popularity of part-time employment inthe Netherlands, especially among mothers with children.

Labor market policies may provide more insight. French women clearlyhad a higher level of workers’ protection in earlier decades, which might

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 113


explain French women’s higher levels of early participation in the labormarket. Yet the evidence for this is far from definitive, as these pro-tections may also reflect women’s earlier engagement in employment,Dutch labor market policies, in comparison, offered few protections for(mostly part-time) women workers until the early 1990s, when women’semployment rates had already increased substantially. Thus the timingof reforms in that nation also suggests their responsive nature.

Perhaps labor market policies must be read in the context of the largereconomic situation. In the French case, reforms in labor market policyappear to be in response to feminist movements and research – yet thesepolicy changes may not have happened if not for a labor force shortage(Revillard, 2006). Similarly, reforms in the Netherlands can be partlyattributed to feminist activity, but the promotion of part-time work andincreasing emphasis on mothers’ roles as workers were driven mostly bygovernmental efforts to create jobs, reduce unemployment, and resolveeconomic strain created by a shrinking tax base.

Cultural factors

Cultural traditions, values, and norms may play a role in determiningwomen’s employment. France has a long history of ideological supportfor women’s labor force participation. For example, Jenson (1990, p.153) notes that political discourse includes competing representationsabout gender identities, including the proper roles for women in society,and that even in the early twentieth century “. . .France, such identi-ties included the possibility – and indeed at times the assumption – ofthe validity and importance of women’s paid work, both for single andmarried women.”43

This strong cultural support for women’s roles as mothers and work-ers has to do with the historical threat of dépopulation and dénatalité(the French population dropped precipitously during the nineteenth andearly twentieth centuries) as well as the strategic approaches of Frenchfeminists. During the early twentieth century, pronatalists encouragedsupport for women’s roles as mothers, while feminists used pronatalistrhetoric to gain support for women’s employment and for family poli-cies (Cova, 1991; Offen, 1991; Misra, 1998). French feminists used theexisting ideological discourse of all citizens in society as both citizensand producers to emphasize women’s duties to the nation as both moth-ers and workers (Jenson, 1990). Indeed, when pronatalists attemptedto restrict women’s roles to motherhood (for example, evicting womenfrom the workforce in order to solve problems of low birth rates and

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 114


men’s unemployment), feminists strategically insisted that women’swork was economically necessary in order to support their families(Offen, 1991; Pedersen, 1993).44

Despite these cultural frames of women as both workers and moth-ers, women continued to be viewed as “different” rather than “equal.”For example, during the 1970s, the ideology of “free choice” for womenbecame more powerful. Laufer (1998, p. 63) argues:

During the 1970s, state policies sought to encompass both the ideal of“free choice” for women, that is that they could choose between paidwork and unpaid caring work in the family, and the principle thatwomen should not have to choose, but should be able to have both,achieving a full career and simultaneously being a wife and mother,an achievement which involved women being economically activeeven when their children were small.

Yet at the same time there was little change in the gender division oflabor in the household, although women working part time did sig-nificantly more domestic labor relative to men than women workingfull time (Duane-Richard, 1995). French cultural ideals emphasized theimportance of supporting women’s responsibility to care, without alsoensuring – or requiring – men’s caregiving responsibilities.

As part-time work expanded during the 1980s, women were muchmore likely to be viewed as marginal and part-time workers, even whenmost women preferred full-time work ( Jenson, 1988). As a result, “Iron-ically, despite the fact that women as a category were becoming moreessential to the production process, individual women’s work situationbecame increasingly precarious as they were offered temporary or part-time work, less protected by collective agreements and state regulations”( Jenson, 1988, p. 160). Jenson argues that the effect of the govern-ment’s acceptance of labor market segmentation in the form of part-timework is that a “profoundly non-egalitarian discourse emerged about the’difference’ of women workers: their lower training, their supposed fond-ness for part-time work, their childcare responsibilities” ( Jenson, 1990,p. 121). Therefore, French culture provides support for women’s roles asworkers as well as caregivers. At the same time, there appears to be lesssupport for changing men’s roles regarding care, and women continueto be seen as different from men in their work ambitions and interests.

Dutch culture provides a different set of expectations about women’sroles. Historically, religious, community, and political norms promoteda common ideal for all Dutch citizens: a traditional family with a male

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 115


breadwinner, a wife, and several children (Pott-Buter, 1993; Karsten,1995; Knijn, 1994). This ideal, coupled with the accepted principle ofstate subsidiarity emphasizing family as the first line of care, contributedto the devaluation of women’s paid work outside the home. Prevailinggender ideals assigned women responsibility for upholding the moralstate of the family. Women were thus valued as wives and mothers, andthe idea of women working for wages was seen as neither desirable nornecessary (Pott-Buter, 1993).

There was thus little support for women as workers, or for policiesthat would enable women to balance work and home. In addition, ColdWar fears dominated the political climate, and led to the association ofpublic child care either with conditions of poverty where mothers wereforced to work for money, or with contemporary totalitarian politicalsystems (Bussemaker, 1998; Gustafsson and Stafford, 1995).45 Traditionalwomen’s organizations also upheld the ideal of separate roles for men andwomen.46 Therefore, there was substantial consensus that women hada moral obligation to care for their children in the home (Bussemaker,1998). Bussemaker (1998, p. 71) notes:

The commonly held belief that the best way to care for and raise chil-dren was in a family environment presided over by a mother wasthe majority view well into the 1970s. Roman Catholics, Protestants,social democrats, and liberals (i.e., the most important political move-ments in the Netherlands) all agreed that a stable and tranquil familylife was the best guarantee for social prosperity.

Given the emphasis on the traditional family, single women and work-ing mothers remained anomalies in Dutch society. Yet Dutch womenbegan to question the traditional housewife arrangement, and a grow-ing number of often highly educated women wanted to sustain paidemployment (Bussemaker, 1998). In the 1960s, feminist action groupsemerged as part of the feminist movement to challenge women’s seclu-sion in the home, their dependence upon men, and the impossibilityof combining motherhood with work outside the home (Bussemaker,1998; Outshoorn, 1986, 1995; Plemper, 1996). Responding to feministdemands, in 1974, the Social Democrat-headed government created anexternal advisory committee called the Emancipatiekommissie (Emanci-pation Commission [EK]). Yet the commission’s first report, adopted in1977, simply reinforced women’s being valued as wives and mothers(Outshoorn and Swiebel, 1998).47

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 116


During the 1980s, this understanding of women’s roles changed torecognize that women’s equality may also be contingent upon employ-ment opportunities. Feminists supported this emphasis through theBreed Platform (Broad Platform), which focused on the promotion ofwomen’s economic independence. Attitudes and norms about womenworking for pay underwent large shifts. For example, in 1965 84 percentof the Dutch population expressed reservations about mothers of school-age children working outside the home; this figure fell to 44 percent in1970, and 18 percent in 1997 (Social and Cultural Planning Office, 1998,p. 141). Throughout the 1980s women’s labor market participation con-tinued to climb, and by the 1990s, there was a heightened emphasis onlabor force participation for all in order to enhance both social cohesionand economic competitiveness, as well as a changing ideology of mother-hood (Bussemaker et al., 1997). Yet still today a strong ideology of womenas “natural” carers persists, and paid work is considered a choice openmainly to higher-educated women (Grünell, 1999; Karsten, 1995).48 AsMorgan (2005, p. 177) notes, “many parents continue to regard childcare with suspicion and prefer to maximize parental caring time whileminimizing that of outside caregivers.” Employment preferences reflectthat culture; of couples with young children, 70 percent prefer the manworking full time and the woman working part time (Jaumotte 2003).49

Further, the government continues to espouse an advanced care ethosthat values motherhood and caretaking, yet has cut back the welfarestate, making it difficult for women outside the labor market to sus-tain a reasonable standard of living for themselves and their families(Kremer, 2001).

Clearly, Dutch culture has emphasized women’s roles as mothers morepowerfully than French culture, while there is stronger support in Francefor women’s roles as workers. In France, however, women’s work rolesremain somewhat marginalized, as women are viewed as “different”rather than “equal.” Attitudinal data may help show the different cul-tural values in these nations. Table 4.1 presents recent data from theWorld Values Survey regarding ideologies about work and care. Respon-dents were asked whether they agreed or disagreed with the statementslisted in the table.50

These data, presented for 1990 and 1999, suggest that there have beensignificant changes in values regarding women’s workforce participation,but also that French and Dutch values are significantly different. Forexample, in both 1990 and 1999 approximately 80 percent of Frenchmen and women agree that both husbands and wives should contributeto the household income – as compared to 31 percent of Dutch men and

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 117


Table 4.1 Values regarding work and family, 1990 and 1999

Percentage agreeing French men French women Dutch men Dutch womenwith the followingstatements . . . 1990 1999 1990 1999 1990 1999 1990 1999

Both the husband and 80 79 80 80 26 31 30 40wife should contributeto household income

Having a job is the best 77 79 80 85 51 61 55 63way for a woman to bean independent person

In general, fathers are – 77 – 83 – 69 – 83as well-suited to lookafter their children asmothers

A pre-school child is 66 61 63 51 70 54 54 35likely to suffer if hisor her mother works

A job is alright but 70 67 67 65 36 36 42 29what most womenreally want is a homeand children

Being a housewife is 62 64 57 60 55 55 55 46just as fulfilling asworking for pay

A working mother can 72 72 72 79 65 76 76 86establish just as warmand secure a relationshipwith her children as amother who doesnot work

When jobs are scarce, 35 21 31 23 22 12 27 13men should havemore right to a jobthan women

Note: ∗Data from World Values Survey (http://www.worldvaluessurvey.org/).

40 percent of Dutch women in 1999 (these numbers are even lower in1990). Similarly, 85 percent of French women suggest that having a job isthe best way for women to be independent, as compared to 63 percent ofDutch women. Yet these two measures may tell us not only about valuesabout women’s work, but also more generally about values about work.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 118


60.5 61.7

69

57.5 60.3

69.2

53.6 5349.2

43.4

0

10

20

30

40

50

60

70

80

1981 1990 1999

French menFrench womenDutch menDutch women

Figure 4.5 Percentage of respondents who believe that work is “very important”in their life . . .

Source: ∗Data from World Values Survey (http://www.worldvaluessurvey.org/).

Figure 4.4 shows the percentage of French and Dutch men and womenwho view work as “very important” to their lives: while Dutch womenhave the lowest levels, Dutch men also view work as less important thando either French men or women (who almost equally appear to see workas very important to their lives). Therefore, these values about women’swork may reflect wider values regarding the importance and primacy ofwork.

However, French men appear to be more likely than Dutch men tobelieve that men are as well suited as women for caring for children.The Dutch government and trade unions have recently sponsored mediacampaigns to promote fathers’ roles as carers, which may help creategreater convergence in men’s values and roles.51 The table does sug-gest other important changes, particularly for the Dutch, over time. Forexample, in 1990, 70 percent of Dutch men and 54 percent of Dutchwomen thought pre-school children would likely suffer if their motherswere employed, compared to 54 percent of Dutch men and 35 percentof Dutch women in 1999. Clearly, there has been a significant change inhow the Dutch view work impacting motherhood.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 119


Yet Table 4.1 also suggests that Dutch men and women are less likelyto agree than French men and women on statements suggesting thatwomen “really” want a home and children, or being a housewife is morefulfilling than working for pay, or that pre-school children are likely tosuffer if their mothers are employed. The Dutch are also slightly morelikely to agree that employed mothers can develop as strong a relation-ship with their children as mothers who primarily provide care, whilethey are less likely to agree that men should have greater priorities toemployment when jobs are scarce.

Clearly, the countries have different cultural resonances around theseideas of women’s equality, men’s involvement in caregiving, and theimportance of women’s involvement in the labor market. The heavyconcentration of employed Dutch women in part-time jobs may explainthese somewhat perplexing findings. Since restructuring of the Dutchwelfare state during the 1990s, public and political discourse has empha-sized labor force participation for all in order to enhance social cohesionand economic competitiveness. This emphasis has contributed to achanging ideology of motherhood wherein (part-time) labor market par-ticipation for mothers is more socially acceptable (Bussemaker et al.,1997). Indeed, 70 percent of couples with young children prefer a manworking full time and a woman working part time. In France, by con-trast, 52.4 percent of couples with young children prefer a man workingfull time and a woman working full time ( Jaumotte, 2003).

Do cultural framings explain the similarities and differences we havenoted? Clearly, there has been more cultural support for women’s rolesas workers in France, and there remains a stronger sense of the impor-tance of women’s engagement in the labor force. This may help explainboth French women’s initially higher levels of employment, and theirhigher levels of full-time employment (see Figures 4.2 and 4.3). At thesame time, ideologies about women’s work have clearly changed dramat-ically in the Netherlands over the past several decades, providing greatersupport for women to enter the labor market, albeit primarily throughpart-time employment (see Figure 4.3). Yet, it remains somewhat diffi-cult to identify whether cultural norms changed in response to patternsof women’s employment, or whether patterns of women’s employmentchanged in response to cultural norms.

Discussion and conclusion

Through our discussion of the Dutch and French cases, we have triedto examine whether women’s employment has been shaped by family

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 120


Women’semployment

Family &labormarketpolicies

Culturalcontext

Economicconditions

Figure 4.6 Proposed model explaining women’s employment

policy, or whether economic and cultural factors may play a role. As ourfindings suggest, we find some evidence in support of all three factors,but also find that there appear to be significant “feedback” loops. His-torically, it appears that family and labor market policies have shapedwomen’s employment, but have also responded to increases in women’slevels of employment. Similarly, cultural norms appear both to respondto and to influence women’s employment.

Figure 4.6 illustrates the theoretical argument that emerges from ouranalysis. We argue that economic conditions shape women’s employ-ment both directly and indirectly through labor market policies andfamily policies. At the same time, cultural norms shape policies andwomen’s employment. Changes in policy must fit dominant ideals ofcare in order to be effective (Kremer, 2008).52 The policies themselves –and also the prevailing economic conditions – influence cultural norms,both directly and also through their impact on women’s labor marketparticipation. Further, women’s employment impacts economic condi-tions both directly and indirectly through policies and cultural norms.Economic conditions, policies, and cultural norms, then, do not operatein isolation, but intersect and overlap to produce a situation of ‘causalcomplexity’ (Ragin, 1997). These factors do not have any predeterminedeffect on women’s employment, but rather operate as important fac-tors, among many others, in unique and dynamic social and politicalcontexts.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 121


To better illustrate our argument, we offer two examples. In France,the system of parental care leave (Allocation Parentale d’Education) wascreated in 1985 with the express intention of encouraging mothers toleave the labor force, with the added bonus of expanding family sizeto three or more children. This policy was constructed, in part, as anactive labor market policy in response to economic conditions, yet alsoreflected Christian Democratic pronatalist values. Yet the policy was notinitially successful: take-up was extremely low, in part because of therequirement that a parent must have three or more children, and havesignificant work experience. The policy, in itself, was not enough toencourage women to leave the labor market. The next revision expandedthe leave, making it available to anyone employed for two out of the pre-vious ten years, therefore making the policy more available to womenwho considered themselves to be primarily homemakers (Morgan andZippel, 2003). Yet, the level of take-up still remained fairly low. It wasonly when the program shifted eligibility to parents of two or morechildren that the program became popular. Perhaps the policy did notprovide enough of an incentive for families to choose to have a thirdchild; yet once parents of two children could take advantage of the pol-icy, increasing numbers of parents did so. Culturally, the policy also fitsinto the “free choice” ideology, allowing women the opportunity to stayat home and caretake, but not requiring them to do so. Although the pol-icy is gender neutral, 99 percent of those taking it are women, reflectingclearly gendered understandings of women’s roles as mothers.

The Dutch government legislated its first leave arrangements in theearly 1990s. As in France, this legislation was motivated by economicconditions, but the impact here was to promote rather than discouragewomen’s employment. Dutch leave arrangements provide a short periodof full-time paid leave, followed by a period of subsidized adjustment tolimited hours. This set-up stimulates part-time work, which conforms tothe labor market strategy of reducing unemployment while providingmore flexibility for employers, and also to persistent gendered idealsof care that deem mothers to be the most appropriate caretakers fortheir own children. Part-time work enables mothers to reconcile theirwork-based and family-based identities in a way not required of fathers.Most leave-takers, then, are women, whose labor market participationhas increased markedly. Yet while traditional ideals of caretaking persist,a changed ideology of motherhood (and fatherhood) accompanies theincrease in women’s employment. The 2001 Work and Care Act providedmore expansive provisions for care, reflecting a heightened emphasisin public and political discourse on women’s (and men’s) rights and

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 122


responsibilities as workers as well as carers. Ultimately, workers’ abilityto balance work and family life within the social and political contextinfluences the extent of their engagement in the labor market, and thuslarge-scale economic conditions.

Another less explicit element to our argument regards the role of poli-tics in shaping women’s employment. While in this chapter we empha-sized policies rather than politics, it is critical to recognize that politics,as well as economic and cultural factors, do drive policy decisions. Inboth countries, shifts between governments have clearly impacted poli-cies that affect women’s labor market involvement. Differences in thecomplexion of the governing party play a major role in determininghow women’s labor force participation has been framed and supported.In France, since the 1960s, the Left has been associated with women’sworkplace rights. When in power, it has developed supportive policies,although it has also accepted certain policies (such as the home careallowance or the growth of part-time employment) that work againstwomen’s interests. However, the Center-Right has much more consis-tently pushed against women’s interests as workers, conceptualizingwomen primarily in terms of their roles as mothers. In the Netherlands,Social Democrats have been the biggest proponents of women’s rightsas workers, although their policies have certainly not been consistentlysupportive. Christian Democrats, however, have held a central positionin Dutch politics since 1945, which partly explains why the Dutch gov-ernment has demonstrated reluctance to create conditions favorable towomen’s labor market participation (de Jong Gierveld and Liefbroer,1995). Dutch political discourse still emphasizes women’s roles as car-ers, but cultural and economic developments mean that political powersnow recognize the important role women play as workers.

Social movements also play a central role in shaping the politicalagenda around these issues. In both France and the Netherlands, feministmovements of the late 1960s and 1970s helped create the conditions forsignificant change in labor market policy. For example, major successesfor the Dutch women’s movement include the creation of the Eman-cipatiekommissie (Emancipation Commission) in 1974 and the DirectieCoordinatie Emancipatiebeleid (DCE) in 1978; similarly, major successesfor the French women’s movements include the creation of the Comitédu Travail Féminin (Committee on Women’s Employment) in 1965 andthe creation of the Ministere des Droits de la Femme (Ministry of the Rightsof Women) in 1981. These movements clearly changed the political land-scape and, particularly through their work within the Left, helped creategreater employment opportunities for women.53

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 123


As our cases should make clear, a variety of structural conditions –including labor market and family policies – provide the frameworkfor women’s employment opportunities. However, cultural frames helpinform those structures and infuse them with meaning. Family policies,in particular, reflect cultural notions of men’s and women’s appropriateroles in society. Yet these cultural ideals are not static, and both labormarket policies and family policies help to both reinforce and changethese ideals.

While our argument does not identify the one most crucial determi-nant of women’s employment, these cases suggest that such a strategyis simply not useful. Clearly family policy does matter in explainingemployment. Family policies can provide structural support (or in thecase of care leaves, disincentives) for women’s employment, while alsohelping provide cultural legitimacy for women’s employment. At thesame time, however, the economic conditions and the cultural contextsalso shape women’s patterns of employment.

We began this analysis with two puzzles: (1) Why were French women’semployment rates initially so much higher than Dutch women’s employ-ment rates? And (2) What explains the remarkable increase in Dutchwomen’s employment in recent decades? Based on our historical anal-ysis, we argue that French employment rates were initially higher dueto the economic need for women workers, as well as more effectivefamily policies and a far more supportive cultural context. In respectof the increase in Dutch women’s employment, greater economic needhas combined with changing cultural norms, leading to greater familypolicy support for women’s employment. In the final analysis, it is theentanglement of multiple factors that help explain variations in women’semployment, reflecting the multifaceted complexity of the social world.

Notes

∗We are grateful to Kathleen Boggs and Karen Mason for excellent researchassistance, to the Social and Demographic Research Institute and the Centerfor Public Policy and Administration at the University of Massachusetts forsupport, to Lane Kenworthy, Monique Kremer, Ann Orloff, and Anne Revil-lard for their helpful assistance, and to all of the participants of the “Methodand Substance in Macrocomparative Analysis” conference (Amsterdam, April2006) for their thoughtful comments.

1. Interestingly, child care for children less than three years of age is stronglylinked to increased women’s labor force participation, while child care forolder children is also linked to increased women’s labor force participation,but only when the Netherlands was excluded from the analysis.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 124


2. “Though often couched in the language of choice and helping parents bal-ance work and family, the underlying goal of care leave policies has beento value and support full-time female caregivers” (Morgan and Zippel, 2003,p. 51). Mahon (2005, p. 4) similarly suggests that these types of care leaves,meant to give women a “choice” between labor market participation andcaretaking, “does little for gender equality, and a typically low rate of reim-bursement means that it operates primarily as an incentive for working class,not professional women, to withdraw from the labour market.”

3. Indeed, Mandel and Semyonov (2006) argue that family policy and welfarestate institutions boost women’s employment, but that women in these coun-tries tend to work in more feminized occupations and are less likely to berepresented in more powerful managerial positions. Charles (2005) similarlynotes that welfare state policies may shape women’s labor force participation,but have less of an effect on occupational gender segregation.

4. Sainsbury (1999b) argues that a taxation system that is tailored to dual-breadwinner couples may increase women’s labor market participation byas much as 20 percent. Yet tax systems do not neatly map onto differences inwomen’s employment rates. Women’s employment may rise even when taxrates penalize second earners (Kremer, 2008).

5. Yet, part-time employment still reinforces traditional gender arrangements,though if part-time work was the norm for men as well as women, it neednot reinforce traditional gender arrangements (Jenson, 1996; Gornick andMeyers, 2003).

6. Of course, there are important differences between preference theory, as laidout by Hakim, and the cultural arguments of scholars like Pfau-Effinger andKremer. Hakim (2003) suggests that family policies are put into place to helpsupport childbearing among working women. Yet, counter to many otherscholars, she argues that policies should instead focus on supporting home-centered women, who are much more likely to have more than one child.

7. Hakim (2000) makes slightly different arguments, but agrees that womendo not consistently respond the same ways to social policies encouragingemployment.

8. Indeed, paid maternity leave was adopted in the first decade of the twentiethcentury (Jenson, 1986, 1990; Heinen and Martiskainen de Koenigeswarter,2001), significantly earlier than in most of the rest of the world.

9. In the 1960s and 1970s, while Christian Democrats tried to use certain ben-efits as a wage to encourage women to stay at home, the Left worked todecrease benefit levels and withdraw access to these benefits from womenwithout young children (Jenson, 1990).

10. These changes were driven by a labor shortage, which created a demand fromemployers for women workers.

11. The 1985 law allowed a parent with three or more children and who hadworked for 24 out of the last 30 months to take a short period of paid leave,and had a fairly low take-up. However, the 1986 law expanded the leave tothree years, and made it available to anyone employed for two out of thelast ten years, making the policy more available to women who thoughtof themselves primarily as homemakers (Morgan and Zippel, 2003). Eventhen, levels of take-up remained fairly low. However, after the 1994 reformexpanded eligibility to parents of two or more children who had worked for

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 125


two out of the previous five years, the program became very popular, eventhough the benefits remained fairly low (paid at 39 percent of the averagewage) (Morgan and Zippel, 2003, p. 55, table 1).

12. “From 175,000 in 1994, the number of beneficiaries has risen to 536,000, ofwhom 99 percent are women” (Heinen and Martiskainen de Koenigswarter,2001, p. 173).

13. “The result is that the current policies favor dual earner families to thedetriment of lower-income households, who are less able to afford privatechildcare and use the new benefits” (Daguerre and Taylor-Gooby, 2003,p. 634).

14. Married women, women with more than two children, and immigrantsare more likely to interrupt their working career for longer periods. Morerecent cohorts have become less likely to interrupt their working career for asignificant period (Grimm and Bonneuil, 2001; Mattioli, 2003).

15. Further, married women were ineligible for the basic old age pension,extended unemployment benefits, and later general disablement benefitswhen they were introduced in the mid-1970s (Sainsbury, 1994).

16. An economic recession led to the decentralization of responsibilities andderegulation, along with major budget cuts. Since 1987, municipalities havebeen responsible for organizing sufficient, quality child care, which hasresulted in a considerable variation in the level and quality of services.

17. The government passed this measure in response to a 1990 report by theScientific Council for Government Policy (WRR) that definitively recognizedchild care as necessary for women’s labor market participation. At the timethe motivation was economic rather than an interest in women’s autonomyper se (Bussemaker, 1998, p. 87).

18. One of the government’s goals through this policy was to extend child daycare via the extension of child care places financed by employers. By 1997, thegovernment’s contribution to the total costs of formal child care had fallen to35 percent, with companies paying 21 percent and parents 44 percent (OECD,2000). Indeed, half of all child care provision is now linked to companies(Bussemaker, 1998).

19. Under certain conditions, parents may deduct some child care expenses fromtheir income tax, with exemptions being higher for single parents than fortwo-parent households. All parents with one or more children living at homeare also entitled to a child allowance, which ranges from 41 percent of thecosts of bringing up a child for low-income families, to 17 percent for high-income families. In addition, the Ministry of Social Affairs has a separateSubsidy Scheme for Childcare that helps to single parents on social securityto meet the costs of child care. However, using formal child care may con-flict with some parents’ values concerning mothers’ role in caring for youngchildren (OECD, 2000).

20. Child care policy stipulates that day care centers reserve a certain percentageof their places for companies that buy them. However, companies often havelimited interest, leaving many places unused even as demand by parents ishigh (Bussemaker, 1998, p. 88).

21. Some collective agreements include continuation of part of the wages. Pub-lic sector employees, for example, receive 75 percent of their wages duringparental leave, which explains why nearly two-thirds of workers who use

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 126


parental leave are employed in the public sector. Private sector employersrarely pay parental leave (OECD, 2002).

22. Traditional gendered ideals persist that uphold men’s role as breadwinnersand women’s role as caretakers. Also, “Men returning to their career after leaveare more likely to encounter a prejudice that they do not ‘take work moreseriously’ than women in similar circumstances.” Thus, even when leave ben-efits are fully paid to either parent, the long-run household opportunity costswill be higher if a man takes leave because of the harm it may do to hiscareer-prospects (OECD, 2002, p. 138).

23. Aside from specifically child-related leave benefits, the Work and Care Act of2001 included other provisions to care for family and household members.Emergency leave provides paid leave of short duration at full wages to coverfor unforeseen situations at home, such as the death of a family member.Short-term carers’ leave provides a maximum of 10 days per year to care forsick children or the employee’s partner, paid at minimum wage or 70 percentof full wages, whichever is highest. Career break leave provides 70 percent ofthe minimum wage for six months, though a longer period is possible withemployer agreement (OECD, 2002, p. 135).

24. This sustained increase primarily occurred in the intermediate age groups,particularly women of childbearing age (Grimm and Bonneuil, 2001). Indeed,among two-parent families with children, one-third of women were in thelabor force in 1968, compared to almost two-thirds by 1987 (Jenson, 1990,p. 109).

25. France also ratified international legislation during the 1950s, requiring equalpay for equal work.

26. However, women continued to be segregated in particular occupations andjobs, and by and large continued to earn lower pay than men (Jenson,1988).The laws allowed for significant wiggle room, in that they did notspecify the meaning of “equal work,” and allowed employers to present a“legitimate motive” for treating men and women differently (Laufer, 2003).

27. Between 1982 and 1986, 130,000 women lost a full-time job while 450,000part-time positions were created (Jenson, 1990, p. 113).

28. However, women do the vast majority of part-time work (Laufer, 1998; OECD,2002), and one-third of all women working part time would prefer to workmore hours (Laufer, 2003; Hantrais and Letablier, 1997). Malo et al. (2000,p. 256) note that in 1997, only 5 percent of men worked part time, ascompared to 31 percent of women.

29. Yet part-time jobs may be in less rewarding sectors (such as the service sector).Laufer (2003, p. 438) argues, “Part-time work as it is used in France today isnot egalitarian. It is deeply unequal – unequal between men and women,but also unequal between different socioeconomic groups of women. Part-time jobs often lead to discrimination against women in pay, training, andcareer development. Outside of public sector employment, part-time jobs areassociated with poor qualifications and with an image of a weak commitmentto paid work with fewer career opportunities and, obviously, lower pay thanfull-time work.”

30. Other reforms include a 1982 reform which clarified the role of spouses ofartisans and merchants, ensuring that they received status as workers, and aremoval of almost all gender-based restrictions on civil service job categories.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 127


31. Yet few firms actually developed and implemented equality plans, in part dueto a lack of information, political will, and sanctions, and the weakness oftrade unions to enforce the policy (Laufer, 1998, 2003).

32. Almost half of all women are employed in only 20 professions. Women aremost evident in administrative and managerial fields, childcare and teaching,and sales (Laufer, 1998).

33. For women managers, few work part time or take extended leave, and, asLaufer (2003, p. 426) notes, “In most firms, the implicit and explicit norms ofmanagerial culture are that part-time work and effective career developmentare generally incompatible, given the demands of the job.” Laufer and Fou-quet (2001) show that while women make up almost one-third of managersand professionals, they compose only 7 percent of top managers.

34. The Depression of the 1930s reinforced the practice of excluding women fromthe labor market (Gustafsson and Stafford, 1995).

35. Visser (2002, p. 28): “The Netherlands waited till 1971 before ratifying ILOConvention 100 on equal pay of men and women, which became the basisfor Article 119 in the Treaty of Rome (1957).” The Equal Pay Act, the firstDutch anti-discrimination act, came into force in 1975 (Havinga, 2002).

36. The nation’s poorly managed disability scheme, characterized by a lowthreshold for entitlement and generous benefits, became a major tool forindustrial restructuring. Employers used the scheme to reduce the supplyof labor through early retirement rather than firing redundant workers andcreating social friction. Workers, employers, unions, industrial boards, andlocal governments all misused the scheme, putting tremendous strain onwelfare resources. The scheme was originally meant to support no more than200,000 people, but by 1989 the number of recipients was near one million.Costs exploded as the number of taxpayers shrank in relation to the numberof beneficiaries, necessitating welfare state restructuring (Hemerijck and vanKersbergen, 1997).

37. The Agreements of the Wassenaar Deal of 1982 formed the basis of the gov-ernment’s efforts to restructure the labor market and reduce the costs of thewelfare state. Unemployment benefits became more difficult to obtain, dis-ability benefits decreased, elder care was cut back, and child care becamemore marketized and expensive. Women were hit especially hard, since theywere assigned primary responsibility for care for the disabled, children andelders (Kremer, 2001).

38. Popular fields for women include teaching, child care, administrative work,and health care. Havinga (2002) explains that traditional job segregation isreinforced by the preferences within organizations and the application behav-ior of women and men seeking employment. With few exceptions, womentend not to apply for a “man’s job,” and men tend not to apply for a “woman’sjob.” Further, whereas Dutch employers are legally required to submit annualreports on the proportion of ethnic minorities they employ in order toimprove their labor market participation, no such legal requirement existsto improve women’s situation in the labor market.

39. Employers with ten or fewer employees are exempt from this law (OECD,2002, p. 134).

40. Lone parents make up around 60 percent of social assistance recipients, whichstems from the social practice – in place since the 1970s – of excluding single

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 128


mothers from the work obligation (Knijn and van Wel, 2001, p. 239; OECD,2002, p. 169). The 1996 reform is an attempt to reduce the financial strainthese beneficiaries put on the national budget, and indicates greater culturalacceptance of mothers’ labor force involvement.

41. Further, parents must manage unpredictable school hours, as teacher short-ages frequently force schools to close for all or part of a day at short notice(OECD, 2002, p. 22).

42. Misra and Moller (forthcoming) show that Dutch lone mothers face a22 percent poverty rate despite a 69 percent employment rate (40 percentpart time), which reveals the particularly adverse effect policy reforms havehad on their lives.

43. Misra (1998, p. 394) further argues that gender identities in France reflect“not only the importance of women’s paid work, but also the validity ofwomen’s unpaid work as mothers.” Although the ideological discourses ofmany nations assume women will be responsible for caregiving, in mostcontexts society accords limited value to the role of mother. Women areexpected to be mothers, but are not valued for their contributions to societyas mothers.

44. When social Catholics succeeded in getting a care subsidy passed to supportmaternal caregiving in 1938, feminists recast the allowance “as an allowancefor families with dependent children living on a single wage [which] couldbenefit not only families with male earners and dependent wives, but equallydivorced, single, or widowed parents – thus providing, ironically, somethingof a safety net for mothers seeking to survive without men” (Pedersen, 1993,p. 268).

45. As Bussemaker (1998, p. 76) states, “Childcare provisions were not seen as partof new social welfare arrangements, but rather, the absence of such facilitieswas proof of the achievement of the welfare state.”

46. Activities promoted women’s involvement in “feminine” pursuits includingchildcare, handicrafts, health and hygiene problems, and volunteer work.Politically neutral by statute, the groups served essentially “to defuse thenotion of a battle between the sexes” (Outshoorn and Swiebel, 1998, p. 148).

47. The EK also called for the creation of a special unit for women’s policywithin Dutch bureaucracy. In 1978 the government responded by creatingthe Directie Coordinatie Emancipatiebeleid (Department for the Coordinationof Equality Policy [DCE]) within the Ministry of Culture, Recreation, andWelfare, meant to raise consciousness and change attitudes (Outshoorn,1995).

48. Karsten (1995, p. 192) notes: “The great majority of full-time housewivesbelong to the lower socioeconomic classes of Dutch society. Until recently,economic necessity had little impact on participation rates.”

49. As Pfau-Effinger (1998) notes, the majority of Dutch women still embrace thecultural ideal that children are most appropriately cared for by their mothersin the household as long as possible. Part-time work thus represents a solutionto mothers’ “moral dilemma” between individual self-fulfillment and care forchildren. Yet the employment orientations of women remain unfulfilled insocial practice, as the proportion of women who want to be employed ismuch higher than that of women who are in fact employed (Pfau-Effinger,1999, p. 143).

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 129


50. For purposes of clarity, we combine “strongly agree” and “agree” in thesetables, although this means that we lose some interesting variation.

51. Yet whereas 68 percent of Dutch men agree that men and women should sharecarework equally, only 8 percent of fathers with children under 18 want toreduce their working hours (Knijn and Selten, 2002). Plantenga (1996) reportsthat whereas half of women part-timers say they work part time to “managetheir household,” only 5 percent of men part-timers give caregiving as theirreason for working part time.

52. While the expansion of the Allocation Parentale d’Education in France serves asa good example of how successful policy changes must draw on ideals of care,the failure of Dutch welfare reform provides a similar example. While theDutch 1996 welfare reform dramatically increased labor market requirementsfor mothers (in less drastic, but similar ways to the American welfare reform),a disproportionate number of Dutch single mothers continue to provide carefull time, or are employed only very few hours a week (Bussemaker et al.,1997). In addition, most Dutch men and women believe that men shouldwork full time and women should work part time when children are young(Jaumotte, 2003).

53. Yet, women’s movements in both countries were also somewhat co-opted, asprominent feminists were brought into the state bureaucracy.Linda Hantrais(1993, p. 124) notes that “by taking over control of women’s rights andby introducing equal opportunities legislation, the state has preempted thewomen’s movement and appropriated its cause.”

References

Bakker, Isabella. 1988. “Women’s Employment in Comparative Perspective.”Pp. 17–44 in Feminization of the Labor Force: Paradoxes and Promises, editedby Jane Jenson, Elisabeth Hagan, and Ceallaigh Reddy. New York: OxfordUniversity Press.

Beechey, Veronica. 1989. “Women’s Employment in France and Britain: SomeProblems in Comparison.” Work, Employment, and Society 3: 369–78.

Berg, Peter, Eileen Appelbaum, Tom Bailey, and Arne L. Kalleberg. 2004. “Contest-ing Time: International Comparisons of Employee Control of Working Time.”Industrial and Labor Relations Review 57: 331–49.


Buchmann, Marlis and Maria Charles. 1995. “Organizational and InstitutionalFactors in the Process of Gender Stratification.” International Journal of Sociology25(2): 66–95.

Bussemaker, Jet. 1998. “Rationales of Care in Contemporary Welfare States: TheCase of Childcare in the Netherlands.” Social Politics 5: 70–96.

Bussemaker, Jet, Annemieke van Drenth, Trudie Knijn, and Janneke Plantenga.1997. “Lone Mothers in the Netherlands.” Pp. 96–119 in Lone Mothers in Euro-pean Welfare Regimes: Shifting Policy Logics, edited by Jane Lewis. London: JessicaKingsley Publishers.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 130


Chambaz, Christine. 2001. “Lone Parent Families in Europe: A Variety of Eco-nomic and Social Circumstances.” Social Policy and Administration 35: 658–71.

Charles, Maria. 2005. “National Skill Regimes, Postindustrialism, and SexSegregation.” Social Politics 12: 289–316.

Cova, Anne. 1991. “French Feminism and Maternity: Theories and Policies, 1890–1918.” Pp. 119–37 in Maternity and Gender Policies, edited by Gisela Bock andPat Thane. London: Routledge.

Cox, Robert H. 1993. The Development of the Dutch Welfare State: From Workers’Insurance to Universal Entitlement. Pittsburgh: University of Pittsburgh Press.

Daguerre, Anne and Peter Taylor-Gooby. 2003. “Adaptation to Labour MarketChange in France and the UK: Convergent or Parallel Tracks?” Social Policy andAdministration 37: 625–38.

de Jong Gierveld, Jenny and Aart C. Liefbroer. 1995 “The Netherlands.” Pp. 102–25in The New Role of Women: Family Formation in Modern Societies, edited byH. P. Blossfeld. Boulder, CO: Westview Press.

de Ruijter, Judith M. P., Anneke van Doorne-Huiskes, and Joop J. Schippers. 2003.“Size and Causes of the Occupational Gender Wage-Gap in the Netherlands.”European Sociological Review 19: 345–60.

Duane-Richard, Anne-Marie. 1995. “Women’s Employment and Different SocietalEffects in France, Sweden, and the United Kingdom.” International Journal ofSociology 25: 39–65.

Eliason, Scott, Robin, Stryker, and Eric Tranby. 2008. “The Welfare State, Fam-ily Policies, and Women’s Labor Market Participation: Combining Fuzzy Setand Statistical Methods to Assess Causal Relations and Estimate Causal Effects.”Chapter 5 in this volume.

Esping-Andersen, Gøsta. 1990. The Three Worlds of Welfare Capitalism. Princeton,NJ: Princeton University Press.


Esping-Andersen, Gøsta. 2000. “Regulation and Context: Reconsidering theCorrelates of Unemployment.” Pp. 99–112 in Why Deregulate Labor Markets?,edited by Gøsta Esping-Andersen and Marino Regini. New York: OxfordUniversity Press.

Fagnani, Jeanne. 1995. “L’Allocation Parentale d’Education: Effets Pevers etAmbiguities d’une Presentation.” Droit Social 3: 287–95.

Gornick, Janet C. and Marcia K. Meyers. 2003. Families That Work: Policies forReconciling Parenthood and Employment. New York: Russell Sage Foundation.

Grimm, Michael and Noël Bonneuil. 2001. “Labour Market Participation ofFrench Women over the Life Cycle, 1935–1990.” European Journal of Population17: 235–60.

Grünell, Marianne. 1999. “Employers on Labour and Care in the Future LabourMarket Structure.” Netherlands Journal of Social Sciences 35: 140–64.

Gustafsson, Siv. 1994. “Childcare and Types of Welfare States.” PP. 45–62 inGendering Welfare States, edited by Diane Sainsbury. London: Sage.

Gustafsson, Siv and Frank Stafford. 1992. “Daycare Subsidies and Labor Supply inSweden.” Journal of Human Resources 27: 204–30.

Gustafsson, Siv and Frank Stafford. 1995. “Links between Early Childhood Pro-grams and Maternal Employment in Three Countries.” The Future of Children 5:161–74.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 131


Hakim, Catherine. 2000. Work–Lifestyle Choices in the 21st Century: Lifestyle Theory.Oxford: Oxford University Press.

Hakim, Catherine. 2003. “A New Approach to Explaining Fertility Patterns:Preference Theory.” Population and Development Review 29: 349–74.

Hakim, Catherine. 2004. “Lifestyle Preferences versus Patriarchal Values: Causaland Non-Causal Attitudes.” Advances in Life Course Research 8: 69–91.

Hantrais, Linda. 1993. “Women, Work, and Welfare in France.” Pp. 116–37in Women and Social Policies in Europe, edited by Jane Lewis. Brookfield, VT:Elgar.

Hantrais, Linda. 1994. “Comparing Family Policy in Britain, France, andGermany.” Journal of Social Policy 23: 135–60.

Hantrais, Linda and Marie-Therese Letablier. 1997. “The Gender of Paid andUnpaid Work Time: A European Problem.” Time and Society 6: 131–49.

Havinga, Tetty. 2002. “The Effects and Limits of Anti-Discrimination Law in theNetherlands.” International Journal of the Sociology of Law 30: 75–90.

Heinen, Jacqueline and Heini Martiskainen de Koenigswarter. 2001. “FramingCitizenship in France and Finland in the 1990s: Restructuring Motherhood,Work, and Care.” Social Politics 8: 170–81.

Hemirijck, Anton and Kees van Kersbergen. 1997. “A Miraculous Model? Explain-ing the New Politics of the Welfare State in the Netherlands.” Acta Politica 32:258–80.

Hicks, Alexander and Lane Kenworthy. 2008. “Family Policies and Women’sEmployment: A Regression Analysis.” Chapter 6 in this volume.

Hobson, Barbara. 1994. “Solo Mothers, Social Policy Regimes, and the Logics ofGender.” Pp. 170–88 in Gendering Welfare States, edited by Diane Sainsbury.London: Sage.

Huber, Evelyne and John Stephens. 2000. “Partisan Governance, Women’sEmployment, and the Social Democratic Service State.” American SociologicalReview 65: 323–42.

Jaumotte, Florence. 2003. “Female Labour Force Participation: Past Trends andMain Determinants in OECD Countries.” Economics Department WorkingPapers No. 376. OECD.

Jenson, Jane. 1986. “Gender and Reproduction: Or, Babies and the State.” Studiesin Political Economy 26: 9–46.

Jenson, Jane. 1988. “The Limits of ‘and the’ Discourse: French Women as MarginalWorkers.” Pp. 155–72 in Feminization of the Labor Force: Paradoxes and Promises,edited by Jane Jenson, Elisabeth Hagan, and Ceallaigh Reddy. New York: OxfordUniversity Press.

Jenson, Jane. 1990. “Labor Market and Family Policy in France: An IntersectingComplex for Dealing with Poverty.” Pp. 107–55 in The Feminization of Poverty:Only in America?, edited by Gertrude Schaffner Goldberg and Eleanor Kremen.New York: Greenwood Press.

Jenson, Jane. 1996. “Part-Time Employment and Women: A Range of Strategies.”Pp. 92–108 in Rethinking Restructuring: Gender and Change in Canada, edited byIsabella Barker. Toronto: University of Toronto Press.

Karsten, Lia. 1995. “Women’s Leisure: Divergence, Reconceptualization andChange—The Case of the Netherlands.” Leisure Studies 14: 186–201.

Kenworthy, Lane. 2008. "Women-Friendly Policies." In Jobs with Equality.New York: Oxford University Press.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 132


King, Leslie. 1998. “‘France Needs Children’: Pronatalism, Nationalism, andWomen’s Equity.” The Sociological Quarterly 39: 33–52.

Knijn, Trudie. 1994. “Fish without Bikes: Revision of the Dutch Welfare State andIts Consequences for the (In)dependence of Single Mothers.” Social Politics 1:83–105.

Knijn, Trudie. 1998. “Social Care in the Netherlands.” Pp. 85–110 in Gender, SocialCare, and Welfare State Restructuring in Europe, edited by J. Lewis. Aldershot:Ashgate.

Knijn, Trudie and Monique Kremer. 1997. “Gender and the Caring Dimension ofWelfare States: Toward Inclusive Citizenship.” Social Politics 4: 328–61.

Knijn, Trudie and Peter Selten. 2002. “Transformations of Fatherhood: TheNetherlands.” Pp. 168–87 in Making Men into Fathers: Men, Masculinities and theSocial Politics of Fatherhood, edited by Barbara Hobson. New York: CambridgeUniversity Press.

Knijn, Trudie and Frits van Wel. 2001. “Careful or Lenient: Welfare Reform forLone Mothers in the Netherlands.” Journal of European Social Policy 11: 235–51.

Korteweg, Anna. 2005. “Between Punishment and Well-Being: The Role of Ideas inthe Work of Welfare in the United States and the Netherlands.” Paper presentedat the Research Committee on Poverty, Social Welfare and Social Policy RC19,Northwestern University, Chicago, IL.

Kremer, Monique. 2001. “A Dutch Miracle for Women?” Social Politics, 8: 182–5.Kremer, Monique. 2005. “The Cultural Dimension of Welfare States: Ideals of Care

and Women’s Work.” Paper presented at the Research Committee on Poverty,Social Welfare and Social Policy RC19, Northwestern University, Chicago, IL.

Kremer, Monique. How Welfare States Care: Culture, Gender and Parenting in Europe.Amsterdam: Amsterdam University Press.

Lane, Christel. 1995. “Gender and the Labor Market in Europe: Britain, Germany,and France Compared.” International Journal of Sociology 25: 8–38.

Laufer, Jacqueline. 1998. “Equal Opportunity between Men and Women: TheCase of France.” Feminist Economics 4: 53–69.

Laufer, Jacqueline. 2003. “Equal Opportunity Policy in France: Symbolic Supportand a Mixed Record.” Review of Policy Research 20: 423–42.

Laufer, Jacqueline and A. Fouquet. 2001. “Les Cadres à l’Épreuve de la Fémini-sation.” Pp. 243–67 in Cadres: La Grande Rupture, edited by Paul Bouffartigue.Paris: Le Découverte.

Letablier, Marie-Therese. 2003. “Fertility and Family Policies in France.” AccessedFebruary 10, 2006: www.ier.hit-u.ac.jp/pie/Japanese/discussionpaper/dp2003/dp160/text.pdf.

Mahon, Rianne. 2005. “The OECD and the Reconciliation Agenda: CompetingBlueprints.” Paper presented at the Research Committee on Poverty, SocialWelfare and Social Policy RC19, Northwestern University, Chicago, IL.

Mahoney, James and Deitrich Rueschemeyer, eds. 2003. Comparative HistoricalAnalysis in the Social Sciences. New York: Cambridge University Press.

Malo, Migeul A., Luis Toharia, and Jerôme Gautié. 2000. “France: The Deregula-tion that Never Existed.” Pp. 245–70 in Why Deregulate Labor Markets?, edited byGosta Esping-Andersen and Marino Regini. New York: Oxford University Press.

Mandel, Hadas and Moshe Semyonov. 2006. “A Welfare State Paradox: State Inter-ventions and Women’s Employment Opportunities in 22 Countries.” AmericanJournal of Sociology 111: 1910–49.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 133


Mattioli, Marie-Annick. 2003. “Chomage en France et en Grande-Bretagne 1990–2000: La Situation Specifique des Femmes.” Revue Française de CivilisationBritannique 12: 29–38.

Mazur, Amy G. 2003. “Drawing Comparative Lessons from France and Germany.”Review of Policy Research 20: 491–523.

Misra, Joya. 1998. “Mothers or Workers? The Value of Women’s Labor: Womenand the Emergence of Family Allowance Policy.” Gender and Society 12: 376–99.

Misra, Joya and Leslie King. 2005. “Women, Gender, and State Policies.” Pp. 526–45 in A Handbook of Political Sociology: States, Civil Societies, and Globalization,edited by Thomas Janeoski, Robert Alford, Alexander Hicks, and MildredSchwartz. Cambridge: Cambridge University Press.

Misra, Joya, Stephanie Moller, and Michelle Budig. 2007. “Work–Family Policiesand Poverty for Partnered and Single Women in Europe and North America.”Gender and Society, 21: 804–27.

Morgan, Kimberly. 2005. “The ‘Production’ of Child Care: How Labor MarketsShape Social Policy and Vice Versa.” Social Politics 12: 243–63.

Morgan, Kimberly. 2006. Working Mothers and the Welfare State: Religion and thePolitics of Work-Family Policies in Western Europe and the United States. Stanford:Stanford University Press.

Morgan, Kimberly and Kathrin Zippel. 2003. “Paid to Care: The Origins and Effectsof Care Leave Policies in Western Europe.” Social Politics 10: 49–85.

Myles, John. 1991. “Post-Industrialism and the Service Economy.” Pp. 350–66 inThe New Era of Global Competition: State Policy and Market Power, edited by DanielDrache and Meric S. Gertler. Montreal: McGill-Queen’s University Press.

OECD. 2000. “Early Childhood Education and Care Policy in the Netherlands.”Background report to the OECD project: Thematic Review of Early ChildhoodEducation and Care Policy. The Hague: Ministry of Health, Welfare and Sport.

OECD. 2001. OECD Employment Outlook. Paris: OECD.OECD. 2002. Babies and Bosses: Reconciling Work and Family Life. Volume 1:

Australia, Denmark, and the Netherlands. Paris: OECD.Offen, Karen. 1991. “Body Politics: Women, Work, and the Politics of Motherhood

in France, 1920–1950.” Pp. 138–59 in Maternity and Gender Policies, edited byGisela Bock and Pat Thane.New York: Routledge.

Orloff, Ann Shola. 2002. “Women’s Employment and Welfare Regimes: Glob-alization, Export Orientation, and Social Policy in Europe and North Amer-ica.” United Nations Research Institute for Social Development, ProgrammePaper Number 12.

Outshoorn, Joyce. 1986. “The Feminist Movement and Abortion Politics in theNetherlands.” Pp. 64–84 in The New Women’s Movement: Feminism and Polit-ical Power in Europe and the USA, edited by Drude Dahlerup. Beverly Hills,CA: Sage.

Outshoorn, Joyce. 1995. “Administrative Accommodation in the Netherlands:The Department for the Coordination of Equality Policy.” PP. 168–85 in Com-parative State Feminism, edited by Dorothy McBride Stetson and Amy G. Mazur.Thousand Oaks, CA: Sage.

Outshoorn, Joyce and Joke Swiebel. 1998. “Feminism and the State in The Nether-lands.” Pp. 143–65 in Women’s Movements and Public Policy in Europe, LatinAmerica, and the Caribbean, edited by Geertje Lycklama à Nijeholt, VirginiaVergas, and Saskia Wieringa. New York: Garland Publishing.

9780230_202573_05_cha04.tex 5/6/2008 15: 36 Page 134


Pedersen, Susan. 1993. Family, Dependence, and the Origins of the Welfare State:Britain and France, 1914–1945. Cambridge: Cambridge University Press.

Pettit, Becky and Jennifer Hook. 2005. “The Structure of Women’s Employmentin Comparative Perspective.” Social Forces 84: 779–801.

Pfau-Effinger, Birgit. 1998. “Gender Cultures and the Gender Arrangement:A Theoretical Framework for Cross-National Gender Research.” Innovations: TheEuropean Journal of Social Sciences 11: 147–66.

Pfau-Effinger, Birgit. 1999. “Change in Family Policies in the Socio-CulturalContext of European Societies.” Comparative Social Research 19: 135–59.

Pfau-Effinger, Birgit. 2004. “Socio-Historical Paths of the Male BreadwinnerModel – An Explanation of Cross-National Differences.” The British Journal ofSociology 55: 377–99.

Pfefferkorn, Roland. 1996. “Métiers ‘Masculins’ et Métiers ‘Féminins’: DesInégalités Persistantes.” Revue des Sciences Sociales de la France de l’Est 23: 143–52.

Plantenga, Janneke. 1996. “For Women Only? The Rise of Part-Time Work in theNetherlands.” Social Politics, Spring: 57–71.

Plantenga, Janneke and Johan Hansen. 1999. “Assessing Equal Opportunities inthe European Union.” International Labour Review 138: 351–79.

Plemper, Esther. 1996. “Women’s Strategies in Dutch Philanthropy.” Voluntas 7:365–82.

Pott-Buter, Hettie A. 1993. Facts and Fairy Tales about Female Labor, Family,and Fertility: A Seven-Country Comparison, 1850–1990. Amsterdam: AmsterdamUniversity Press.

Ragin, Charles C. 1997. “Turning the Tables: How Case-Oriented ResearchChallenges Variable-Oriented Research.” Comparative Social Research 16: 27–42.

Revillard, Anne. 2006. “Bringing the Movement within the State: The Comitédu Travail Féminin (1965–1981), or The Unknown Origins of French State Femi-nism.” Paper to be presented at the European Social Science History Associationmeetings.

Sackmann, Rosemarie. 1998. “European Gender Roles: Public Discourses andRegional Practices.” Innovation: European Journal of Social Sciences 11: 167–90.

Sainsbury, Diane, ed. 1994. Gendering Welfare States. London: Sage.Sainsbury, Diane, ed. 1999. Gender and Welfare State Regimes. Oxford: Oxford

University Press.Silvera, Rachel. 2000. “Gender Impact Assessment and the Employment Strategy

in France.” Report for the European Commission’s Expert Group on Gender andEmployment.

Social and Cultural Planning Office. 1998. Social and Cultural Report 1998.Rijswijk: SCP.

Stier, Haya, Noah Lewin-Epstein, and Michael Braun. 2001. “Welfare Regimes,Family Supportive Policies, and Women’s Employment along the Life Course.”American Journal of Sociology 106: 1731–60.

Tilly, Louise, and Joan W. Scott. 1978. Women, Work, and Family. New York:Routledge.

Visser, Jelle. 2002. “The First Part-Time Economy in the World: A Model to beFollowed?” Journal of European Social Policy 12: 23–42.

Visser, Jelle and Anton Hemerijck. 1997. “A Dutch Miracle”: Job Growth, WelfareReform and Corporatism in the Netherlands. Amsterdam: Amsterdam UniversityPress.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 135

5The Welfare State, Family Policies,and Women’s Labor ForceParticipation: Combining Fuzzy-Setand Statistical Methods to AssessCausal Relations and EstimateCausal EffectsScott R. Eliason, Robin Stryker, and Eric Tranby

In Why We Need a New Welfare State (2002), four long-time male scholarsof the twentieth-century welfare state – Gøsta Esping-Andersen, JohnMyles, Anton Hemerijck, and Duncan Gallie – argue that the welfarestate of the twenty-first century requires “comprehensive redesign.” Thetwenty-first-century welfare state must be redesigned around not justgovernment–market relations and the life-course patterns of men, butalso work–family interactions and the life-course patterns of women.Similarly, as Myles and Quadagno (2002) argue in their recent reviewof literature on social policy and the welfare state, gender relations, fam-ily forms, and women’s employment are central to contemporary welfarestate restructuring in a way that they were not during the “golden age”of welfare expansion.

Feminist scholars and themes such as “female friendliness,” “socialcare,” and “work–family conflicts” have contributed much to placingwomen and children on center stage along with men when it comes towelfare state redesign.1 But likewise, labor market and demographic real-ities loom large (Huber and Stephens, 2000, 2001; Scharpf and Schmidt,2000; Myles and Quadagno, 2002; Esping-Andersen et al., 2002). Inmuch of Europe, life-span is increasing as are the costs of supportingretirement, at the same time as fertility rates have plummeted. Especiallywhen the cost of socializing retirement is paid for in large part by payrolltaxes on those working, fiscal crisis looms and additional taxable workersare required. Under these conditions, even the core ‘male-breadwinner’

135

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 136


countries of continental Europe – such as Germany – will find it seduc-tive to contemplate increased female labor force participation. Enter thefemale worker and, with her, new twists on scholarly and policy concernwith social care (Lewis, 1998; Daly and Lewis, 2000; Wennemo, 1994;Gauthier, 1996; Randall, 2000; Mahon, 2002; Morgan and Zippel, 2003;Morgan, 2004; Misra, Budig and Moller, 2005).

Clearly, facilitating women’s labor force participation on the supplyside through the public provision of child care or otherwise will not besufficient to put the welfare state on a sound fiscal footing. This is espe-cially true in those countries like Germany, in which job growth has beenstagnant in part because substantial employer payroll contributions tofunding pensions and unemployment may decrease employers’ enthusi-asm for job creation, especially when strong unions do not countenancelower wage-lower benefit jobs (Scharpf, 1999; Scharpf and Schmidt,2000; Streeck, 2001). Nonetheless, encouraging female labor force par-ticipation on both the supply and demand sides is now widely viewed asa central pillar of the twenty-first-century welfare state (O’Connor, 2003;Esping-Andersen et al., 2002). Whether the motivation for this has beenproactive gender egalitarianism, reactive adaptation to new family struc-tures and labor force participation patterns of both women and menacross the life course, public concern with child poverty and develop-ment of children’s human capital, or some combination of these, thepolicy emphasis on women’s labor force participation calls for increasedresearch on the avenues through which this is – or can be – achieved.

Key among these are the demand-side mechanism of gendered jobcreation and the supply-side mechanism of parental leaves and publiclysubsidized or provided child care (Rubery et al., 1999; Esping-Andersenet al., 2002; Stryker and Eliason, 2004). This chapter contributes to ongo-ing debate about design of the “new welfare state” by examining thecausal links between these and other welfare state policies and insti-tutions that may facilitate female labor force participation in fourteenadvanced capitalist democracies from 1960 to 1999. Countries includedin our analyses are Austria, Belgium, Canada, Denmark, Finland, France,Germany, Ireland, Italy, the Netherlands, Norway, Sweden, Britain, andthe United States.

We assess causal relations and estimate causal effects by bringingtogether two very diverse methodological tools – fuzzy-set analysisand a formulation of the intention-to-treat design. Specifically, we usegoodness-of-fit tests developed by Eliason and Stryker (2007) for assess-ing causal relations and conjunctions in fuzzy-set analyses (Ragin, 2000).We then estimate causal effects suggested by the fuzzy-set analysis using

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 137

Scott R. Eliason, Robin Stryker, and Eric Tranby 137

Imbens and Rubin’s (1997) and Hirano, Imbens, Rubin, and Zhou’s(2000) formulation of the intention-to-treat design for non-experimentaldata. Combining these methods provides a powerful lens with whichto study the causal influence that social policies have had, and can beexpected to have, on outcomes of interest.

Theoretical and empirical background

There is substantial literature on family policies, including child carepolicies and parental leaves, in the European Union and other advancedindustrial democracies.2 Until recently, much of that literature waspredominantly historical and descriptive. Scholars who attempt causalanalyses often focus on explaining what has led to institutional variationin day care provision, maternity and paternity leaves, family allowancesor other family policies, including child and family cash and tax benefitsacross time and place among advanced industrial democracies (Wen-nemo, 1994; Gauthier, 1996; Daly, 2000; Randall, 2000; Misra, 2003).Research attempting to examine empirical links between variation infamily policies and outcomes – whether quantity and quality of women’slabor force participation, child and family poverty and health, child edu-cational outcomes, or class- and gender-related income inequality still isin its infancy. But it is increasing quickly, given the centrality of thesequestions to current policy making.3

In this chapter, we are interested in both aggregate-level causes of varia-tion in welfare state institutions and family policies and, in consequenceof these, variations for gendered labor markets. Because we are interestedin both demand-side and supply-side incentives for women’s labor mar-ket participation, we examine causes and consequences of cross-countryand over time variation in expansion of public sector employment andalso the causes and consequences of variation in day care and parentalleave policies.

Public sector expansion and female labor force participation

It is well established that long-term incumbency of social democraticparties promotes development of generous, redistributive welfare statescharacterized above all by a public service orientation to welfare stateprovision. Meanwhile, although long-term Christian democracy like-wise promotes welfare state generosity, it does so predominantly throughtransfer payments that are not especially redistributive (Huber, Raginand Stephens, 1993; Huber and Stephens, 2001). Although Huber andStephens (2000) treat public sector expansion as a consequence of rising

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 138


female labor force participation when experienced during social demo-cratic governance, we have specified and examined empirically a reversecausal argument (Stryker and Eliason, 2004).4

Our reverse causal argument suggests that cumulative social demo-cratic governance expands the public sector, in turn promoting femalelabor force participation. Publicly-provided social services allow womento enter the labor force and provide employment for them (Huberand Stephens, 2001). As we elaborate further below, because publicly-provided care services remove family-oriented time constraints, theseservices should increase female labor supply (O’Connor, Orloff, andShaver, 1999, pp. 78–88). At the same time, however, publicly pro-vided care services should increase demand for female – as opposedto male – labor insofar as public sector jobs require tasks consideredfemale-oriented.

Welfare state and labor market research has argued consistently thatthe growth of public sector service occupations expands women’s jobopportunities (OECD, 1982; Esping-Andersen, 1990, pp. 206–29; Rosen-feld and Kalleberg, 1990; Myles and Turegin, 1994; Gornick and Jacobs,1998; Huber and Stephens, 2000; Rubery, Smith, and Fagan, 1999, pp.44–5; Myles and Quadagno 2002). Likewise, in liberal, market-orientedwelfare states, private sector service expansion has increased job oppor-tunities for women, especially in low wage service, including care, jobsthat are disproportionately feminized (England, Budig, and Folbre, 2002;Daly, 2000; O’Connor, Orloff, and Shaver, 1999, pp. 97–8; Rubery et al.,1999, p. 22). Underlying these observations is an extensive literatureexamining how gender-segregated labor markets affect gender inequal-ities in income, as a function of: (1) the gender of the job task; (2) the(mostly-transcultural) expectations that women should be matched tojobs requiring (mostly) female-oriented tasks and men to jobs requir-ing (mostly) male-oriented tasks: and 3) the systematic devaluation ofskills involved in female-oriented tasks (Rosenfeld and Kalleberg, 1990;Steinberg, 1990; Reskin, 1993; Ridgeway, 1997; Ridgeway and SmithLovin, 1999; Grusky and Charles, 2001; Bonstead-Bruns and Eliason,2002; England et al., 2002; Charles and Grusky, 2004; Pettit, 2006).5

In short, incorporating gendered tasks and gendered matching pro-cess into welfare state arguments clarifies why public sector expansion –creating jobs requiring female-specific skills – should enhance demandfor, as well as supply of, female labor. Consistent with this reasoning,our prior fuzzy-set research on 1977–94 Germany, France, Denmark,Britain, Italy, and Belgium showed that high cumulative left cabinetincumbency was causally necessary for high public sector size. In turn,

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 139


high public sector size was causally sufficient for high female labor forceparticipation. These results conformed to our theoretical expectations.However, only further data collection and analysis could show whetherthe results would generalize to a broader, more representative group ofadvanced capitalist democracies. As well, further data collection andanalysis were required to explore whether our prior findings would holdfor a longer time period that captured cross-country variation in bothearly trends toward public sector expansion and later trends toward pub-lic sector retrenchment (see, for example, OECD, 1982; Castles, 2001;Clayton and Pontusson, 1998).

Social democratic governance, public sector expansion,and family policies

Given that a substantial amount of public sector expansion from the1960s forward was fueled by growth in publicly provided caring ser-vices, public sector expansion itself incorporates supply-side as well asdemand-side promotion of women’s labor market participation (Strykerand Eliason, 2004). But supply-side incentives can best be examined bydirectly measuring the family policies that can be expected to shapeinstrumental incentives and the cognitive and normative context forwomen’s labor force participation. Here, expectations about causal linksfrom variation in cumulative governance patterns to publicly providedday care, and from day care to female labor market participation are rea-sonably straightforward. Both theoretically and empirically, however,the situation is murkier with respect to maternity leave (Misra, Budigand Moller, 2005; Pettit, 2006).

With respect to day care first, Huber and Stephens (2000, 2001)have pointed out that women already in the labor market are likelyto demand government support – including publicly provided or sub-sidized child care – for managing work–family conflicts. Consistent withsocial democratic governments’ emphasis on universalistic, public provi-sion of services in general, these governments should be especially likelyto respond to such demands. Thus, we would expect cumulative socialdemocratic governance, relative to Christian democratic or center-rightsecular governance, to be positively related to public provision of childcare. From the early 1960s, the Scandinavian countries were leaders inquickly expanding publicly provided day care (Gauthier, 1996; Randall,2000). Thus, historical descriptions of the advent and trajectories of daycare provision across advanced capitalist democracies lend support to thehypothesis of a general positive association between cumulative socialdemocratic governance and public provision of child care.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 140


Even here, however, deeper historical investigation reminds us thatmotivation for initiating a policy and its subsequent effects or functionscan be very different. For example, Gauthier (1996) notes that, contraryto other European countries, Sweden highlighted public day care as anissue early in the twentieth century. But at this time, Swedish day careformed part of a policy package seen as responding to fertility and pop-ulation concerns rather than those of gender and employment. Randall(2000) points out that, more recently, there has been an internationalconvergence of attention on issues of female labor force participationand related child care concerns. Likewise, there has been substantialrecent convergence in public provision of child care for children from 3to school age – although not for day care for younger children (Randall,2000).

In our earlier fuzzy-set analyses, we found that high cumulative socialdemocratic governance was causally necessary for high public provisionof day care for children 0–2 and also for children 3 to school age (Strykerand Eliason, 2004). Again, however, we can not be sure that this find-ing will hold up for a more representative group of advanced capitalistdemocracies and over a longer time frame. This is especially so becausesuch diverse countries as Austria, France, Germany, Ireland, and Britain –concerned more about pre-primary child socialization and educationthan about women’s labor force participation – nonetheless in recentyears have accelerated their child care provisions. In short, althoughwe continue to hypothesize a positive causal link between cumulativesocial democratic governance and public provision of day care, we nowmobilize more extensive data to examine whether there are differencesin causal patterns involving day care for younger and older children.

With respect to maternity leave, our earlier fuzzy-set analyses foundno causal relationship between high levels of cumulative social demo-cratic governance and high levels of maternity leave in Denmark, Britain,Germany, France, Belgium, and Italy, 1977–94. Perhaps this is not sur-prising, given that gender ideologies from the relatively egalitarian “dualearner” and “earner–carer” models to the gender distinctive “separatespheres,” “male-breadwinner” or “primary caregiver/secondary earner”models have motivated the origins and expansion of maternity leaves,as they have other family support policies depending on the countryand historical period in question (Gauthier, 1996, 2000; Wennemo,1994; Misra, 2003; Moss and Deven, 1999; Randall, 2000; Daly andLewis, 2000; Misra, Budig, and Moller, 2005). In addition, maternityleaves have been motivated by health and fertility-related concerns aswell as labor market-related goals. Indeed, Gauthier (1996) argues that

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 141


maternity leave in Europe arose predominantly from concerns aboutthe health of working mothers and their children, rather than from adesire to promote women’s employment. Partial exceptions to this gen-eralization included Sweden and France, where pre-World War II genderideologies were more accepting of women’s right to employment (Gau-thier, 1996; Jenson, 1990). Diverse historical motivations for adoptingand expanding maternity leave policies help us make sense of our priornull findings. But these must be re-examined for a more representativeset of advanced capitalist democracies.

Public sector expansion, family policies, and femalelabor market participation

As Gauthier (1996) reminds us, “family” policies encompass an amalgamof legislation and benefits that ordinarily are not part of any coherentframework. They are often as much or more responsive to population,health and education concerns as they are to labor market issues. Forexample, already in the 1960s, a goal of supporting female employmentand working mothers directly motivated day care policies in Scandinavia.These policies also were motivated by post-World War II Scandinavianstates’ desires to support early socialization and learning. During a similartime frame, the French government also highlighted public provision ofday care, but almost exclusively because of beliefs in early education andsocialization, and not because of a preoccupation with working mothers.After 1975, as changes in family structure and life-course patterns of menand women accelerated, Nordic countries continued to espouse genderequality as a major goal for family policies. Meanwhile liberal countriessuch as the US espoused gender equality without endorsing governmentresponsibility to provide day care. At the same time as Nordic socialdemocracies and Anglo-American liberal governments were professingcommitment to gender equality, the German government assumed someresponsibility for child care, but continued to hew ideologically to the“male breadwinner” model of work–family division of labor.

Whatever the divergent motivations and causes for different policiesand policy packages across countries and over time, current debatesabout welfare state restructuring emphasize the function or impact ofthese policies for female labor market participation. Among researcherswho have tried to assess the causal import of family policies for women’semployment, findings have been mixed. Though results have been moreconsistent across diverse samples and measures for day care policies thanmaternity or parental leave, our capacity to interpret the labor market

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 142


impact of family policies is diminished when researchers fail to disag-gregate composite indices of women-friendly policies that may embodyconflicting sets of incentives or norms.

For example, in a cross-national study of women’s employment pat-terns during childrearing in 12 advanced capitalist democracies, includ-ing Sweden, Norway, Australia, New Zealand, the United Kingdom,Israel, Italy, Austria, Germany, New Zealand, Canada and the Nether-lands, Stier, Lewin-Epstein, and Braun (2001) found that there wascross-country variation in the percentage of women who, in 1994,reported working full-time prior to the birth of children – from 44 per-cent in the Netherlands to 81 percent in Canada. But there was muchgreater cross-country variation in the employment patterns of womenwith pre-school children. In 1994 Sweden and Norway, categories of con-tinuous full-time employment and continuous non-employment weresmall. Most women worked part time when they had children, or theyalternated between non-employment and part-time work or part-timework and full-time work. In 1994 Israel, Italy, and to a lesser extentAustria, there were large percentages of women in both continuousfull-time employment and continuous non-employment. Overall, Stieret al.’s results (2001) suggested that observed cross-national variation inwomen’s employment patterns during childrearing could be accountedfor in part by a framework combining the impact of broad welfare stateregimes and country-specific patterns of family policies. Most relevantto our research, within all Esping-Andersen regime types – social demo-cratic, liberal and conservative-corporatist – employment continuity wasgreater in countries that provided support for working mothers.

However, Stier et al. (2001) measured support for working mothersas a composite categorical variable – coded high, medium or low basedupon Gornick et al.’s (1997) composite index of family policies presumedto support female labor force participation. As adapted by Stier et al.(2001), this composite measure combined multiple indicators of bothchild care and parental leave. Child care indicators took into accounttax relief for child care, the existence of legislation guaranteeing childcare for children ages 0–2, and children 3–school age, child care expen-ditures, the percentages of children ages 0–2 and children 3–school agein publicly funded child care, the percentage of children aged 5 in pre-primary school and the percentage of children in public after schoolprograms. Meanwhile, measures of parental leave schemes took intoaccount legislated job protections, weeks of paid maternity leave, thewage replacement rate, policy coverage, and the availability of extendedleave and paternity benefits. Thus, although Stier et al.’s (2001) findings

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 143


do suggest that family policies positively affect female labor market par-ticipation, it is impossible to trace labor market effects to the specific,potentially diverse norms and incentives embodied in specific child careor leave policies.

With respect to day care, extant economic and sociological theo-ries of the labor market predict that readily available, affordable daycare should increase the level of women’s labor market participation.Available, affordable day care is predicted to loosen women’s family-related budget and time constraints, thus decreasing their reservationwages and altering their preference formation (Blau and Ferber, 1992;Rønsen and Sundstrom, 2002; Stryker and Eliason, 2004; Pettit, 2006).At the same time, widespread use of publicly provided child care facilitiesmay reshape cognitive expectations and normative evaluations about theacceptability or desirability of child care provided outside the home andby someone other than the mother (Leira, 2002; Stryker and Eliason,2004). However, expectations about the import of parental leaves of var-ious types may not be so straightforward (Bruning and Plantenga, 1999;Rønsen and Sundstrom, 2002; Stryker and Eliason, 2004; Misra, Budig,and Moller, 2005; Pettit, 2006).

Bruning and Plantenga (1999) evaluate how variation in regulationsgoverning parental leave produce country variation in leave take-uprates by gender in the Netherlands, Finland, Norway, Sweden, Denmark,Austria, Germany, and France. Although the authors do not examinehow these differences in take-up rates affect labor force participation bywomen, their study does suggest the importance of separating parentalleave variables from public provision and public subsidy of child care.As well, it suggests measuring aspects of leaves such as their target(s),generosity, duration, and flexibility both individually and as a totalpackage.

Research that has disaggregated parental leave from day care policies,and looked at the impact of one or the other individually yields mixedresults. Based on regression analyses, Esping-Andersen (2002) argued thatthe paramount barrier to full time employment among women is lack ofaffordable child care. Somewhat consistent with this conclusion are ourown prior fuzzy-set analyses for 1977–94 Denmark, France, Germany,Britain, Belgium, and Italy (Stryker and Eliason, 2004). These showedthat both high aggregate levels of publicly available day care for chil-dren from 0–2 and for children from 3 to school age were individuallycausally sufficient, albeit not necessary, for high aggregate labor force par-ticipation among women 15–64. Meanwhile, Rønsen and Sundstrom’s(2002) micro level econometric analyses showed that increases in the

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 144


local supply of public day care encouraged entry into employment forfirst- and second-time mothers in Finland (1989), Norway (1988), andSweden (1992). The impact was stronger in Sweden than in Norway.In Finland, the availability of an allowance for home care seemed todiscourage full-time work, although not employment in general.

On the other hand, based on data from West Germany, Kreyenfeld andHank (2000) specified a multinomial logit model in which the probabilityof an individual mother with one or more children under 12 not workingversus working part-time versus working full-time was a function of theavailability of day care and diverse other factors. These other factorsincluded the unemployment rate, whether the mother was native orforeign born, whether she was a lone mother, number of children, ageof youngest child, mother’s education, partner’s wage and whether ornot the child’s grandparents lived in the same town. The authors foundthat variation in the availability of publicly funded child care did notinfluence mother’s employment in West Germany. They presumed thattheir null finding resulted because day care centers in Germany are openduring limited hours and because primary school hours are irregular.Thus, they concluded that the structure of German child care fails tofacilitate mother’s labor force participation.

With respect to parental leave policies, Winegarden and Bracy (1995)provided econometric analysis of the impact of paid maternity leaveson female labor force participation of women, ages 20–34, for four timepoints between 1959 and 1989 in 17 countries. The analysis includedAustria, Canada, Denmark, France, Greece, West Germany, Italy, theNetherlands, New Zealand, Norway, Portugal, Sweden, Spain, the UnitedKingdom, and the United States. The authors found that, as the durationof paid leave increased, female labor participation rates for those betweenages 20 and 34 likewise increased.

Conversely, measuring maternity leave provision as a function of bothduration and generosity, our own prior fuzzy-set investigation of therelationship between maternity leave schemes and female labor forceparticipation in 1977–94 Denmark, Britain, Italy, France, Germany, andBelgium found no relationship between high aggregate levels of mater-nity leave provision and high aggregate levels of women’s labor marketparticipation. In their investigation of women-friendly policies, Gor-nick et al. (1997) and Korpi (2000) had assumed that maternity leavefunctioned as part of a package of dual-earner support. In contrast, wespeculated that for women entering the labor force to meet financialneeds, high maternity benefits might substitute for subsequent laborforce participation rather than encourage it (Stryker and Eliason, 2004).

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 145


Alternatively, our failure to find the expected impact of maternity leaveson aggregate female labor force participation could have been due to thelimited number of countries, time frames, and measures that we wereable to investigate prior to our current study.

Consistent with our assumption that maternity leave provisionsembody conflicting labor force incentives depending on the type of pol-icy and the demographic profile of woman we are considering, Rønsenand Sundstrom’s (2002) in-depth analysis of Finnish, Norwegian, andSwedish mothers found a highly nuanced pattern of labor market effects.The authors assessed the impact of parental leave and child care policiesby comparing the full-time and part-time (re) entry into the labor forceof all first- and second-time mothers, regardless of prior work history, in1988 Norway, 1989 Finland, and 1992 Sweden. Testing statistical mod-els motivated by microeconomic theories of female labor supply, Rønsenand Sundstrom (2000) concluded that generous, flexible parental leaveprograms, such as the one found in Sweden, contain incentives encour-aging more mothers to work prior to the birth of a child. The Swedishprogram also encouraged more mothers to remain in the labor marketthroughout the childbearing years. In Norway, a shorter entitlementperiod created lower pre-birth employment incentives, so that Norwe-gian mothers were somewhat less likely to work prior to the birth of achild than were their Swedish counterparts. And, although women eli-gible for maternity leave tended to resume employment more quickly inNorway than in either Sweden or Finland, a larger number of Norwegianmothers remained out-of-the-labor force, because an inflexible entitle-ment system made it harder for women to reconcile work and familyresponsibilities. Finland is characterized by generous, extended leavescombined with moderate flexibility and the availability of an allowanceto compensate home care. Here, women who had worked pre-birth andthus were eligible for maternity leave tended to remain out of the laborforce for a longer time period after the birth of a child than did theirNorwegian counterparts.

Similarly suggestive of the potentially competing incentives embod-ied in parental leaves is Pettit’s (2006) multi-level comparative analysisof women’s employment in the United States, continental and east-ern Europe, and the Nordic countries. Pettit found that the impactof parental leave depended both on the number of children in thehousehold and on the length of the leave. Parental leaves up to twoyears were associated with an increased probability of employment forwomen with young children, but the positive impact diminished asthe number of children increased. In addition, leaves of more than

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 146


two years were associated with a decreased probability of women’semployment.

In the following analyses, we build on our prior fuzzy-set methodology,so that we can continue to distinguish not just between the absence andpresence of evidence for some causal relationship among welfare stateinstitutions, family policies and female labor market participation, butalso among empirical support for relations of causal necessity, causalsufficiency, and causal necessity and sufficiency combined. Empiricalsupport for causal relations of sufficiency versus those of necessity withrespect to various aspects of family policy has quite different implicationsfor policy makers concerned to enhance female labor force participation.In particular, finding causal sufficiency in the absence of necessity tells usthat there are alternative routes to achieving the desired policy outcome.Finding causal necessity in the absence of sufficiency, however, tells usthat, although a particular welfare state or family policy is essential forachieving a particular labor market outcome, its presence alone will notensure that outcome.

Following the general thrust of our literature review, our currentanalyses build on our own prior research to continue examining bothsupply-side and demand-side incentives for increased women’s laborforce participation, this time for a larger, more representative groupof advanced capitalist democracies over a longer time period. We like-wise continue to examine disaggregated policy measures, rather thancomposite indexes of female-friendly or family-friendly policies.

Our analyses focus first on replicating precisely our prior six-country,1977–94 analyses with a 14-country fuzzy-set analysis, from 1960 to1999. We then incorporate our fuzzy-set measures into a Rubin-styleanalysis of compliers average causal effects. This allows us to comple-ment our findings of necessity and sufficiency with respect to causalrelations with a more rigorous test of causal effects, and of the relativestrength of demand- and supply-side causal effects. We consider this tobe a first ‘cut’ on the evidence we can supply to scholarly and policy ques-tions involved in redesigning the welfare state, given the richness of datawe are gathering and the combination of methods we employ relative tothe cues we take from the totality of prior theory and research.

Data and variables

We compiled our data set from the diverse array of sources presentedin Appendix 5.1. In total, we currently have gathered information fromapproximately 50 different data sources, including OECD documents

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 147


and data sets, country monographs, comparative studies, and data setscollected by other researchers. Whenever possible, we used multiplesources for each measure, so that we could verify consistency amongsources, and – when sources were inconsistent – locate the source of theinconsistency and create the most complete and accurate measurementpossible for all our variables.

In order to maintain data completeness over the full 1960–99 timeperiod, we were constrained to drop Japan, Australia, and New Zealandfrom our analyses. We also eliminated Switzerland, because family poli-cies are administered at the cantonal, rather than national level. Thoughwe also faced particularly challenging issues in locating data on childcare policies for Norway and Ireland, we were able to overcome thesesufficiently to maintain these two countries in our data base.

Thus, our analyses include data from 1960 to 1999 for Austria, Belgium,Canada, Denmark, Finland, France, (West) Germany, Ireland, Italy, theNetherlands, Norway, Sweden, Britain, and the United States. Althoughwe had hoped to represent the universe of advanced capitalist democra-cies, we can only do so within the limits of available, comparable data. Inaddition to eliminating Japan, Australia, New Zealand, and Switzerland,because of the data difficulties mentioned above, we cannot analyze datafor countries outside of the OECD or for countries that have joined OECD(or EU) data collection efforts very late relative to our time frame.6

The current analysis extends to these countries and time frame theset of policy measures developed in our earlier research (Stryker andEliason, 2003, 2004). More precisely, as the foundation for construct-ing our fuzzy-set membership scores for the fuzzy-set and CACE analyses(discussed below and in Appendix 5.2), we use information on: (1) mater-nity leaves including wage replacement rates, number of weeks of paidleave, proportion of employed women covered, and public expendi-tures for paid leaves, (2) weeks of extended leave, (3) public day careincluding, separately, the proportions of children 0 to 2 years old, and3 to school-age, in public daycare and public expenditures for daycare,and (4) family/child cash and tax benefits including child care tax relief,family allowances for children, and family support benefits.

Except for extended leaves, fuzzy-set membership scores are thenobtained from indexes on this empirical information originally devel-oped in Stryker and Eliason (2003). More precisely, membership scoresare constructed for high levels of government support for (1) publicday care for younger children, (2) public day care for older children,(3) maternity leaves, (4) extended leaves, and (5) family/child cash andtax benefits. We also construct membership scores for high levels of

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 148


Table 5.1 Decade means and standard deviations for cumulative left cabinetincumbency

Decade means Decade standard deviations

1960s 1970s 1980 1990s 1960s 1970s 1980s 1990s

Austria 8.97 14.95 24.43 30.67 0.97 3.03 2.44 1.61Belgium 7.01 10.33 12.45 16.11 1.01 0.84 0.31 1.57Canada 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00Denmark 13.32 18.91 25.16 27.50 2.50 2.25 0.55 1.99Finland 6.39 11.09 16.61 19.71 0.85 1.59 1.49 0.66France 3.09 3.09 6.44 12.75 0.00 0.00 2.13 1.40Germany 0.29 6.22 12.11 12.33 0.52 2.59 0.47 0.06Ireland 1.99 2.57 3.79 5.20 0.00 0.50 0.43 0.87Italy 1.13 2.37 3.45 5.85 0.48 0.15 0.60 0.93Netherlands 5.43 7.10 8.62 11.20 0.34 1.09 0.11 1.26Norway 18.32 23.29 30.05 36.78 1.76 2.65 1.37 2.28Sweden 18.36 27.66 32.61 40.04 3.03 2.18 2.68 1.98United 6.93 12.93 16.16 16.76 1.96 2.09 0.00 1.07

KingdomUnited States 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Note: The minimum value of 0 includes 1960–99 United States, 1960–99 Canada, and1960–65 Germany; the maximum value of 43.61 includes 1999 Sweden.

cumulative left governance, high levels of civilian public sector employ-ment, and high levels of female labor force participation. See Stryker andEliason (2003) for details on the indexes (except for extended leaves,which was constructed solely on the number of weeks of extendedleave7). See Appendix 5.2 for details on constructing the fuzzy-setmembership scores from these indexes.

Finally, to give some idea of the general across-country differencesand within-country change over time in this empirical information,Tables 5.1–5.8 present the decade means and standard deviations forthe indexes and other empirical information used in forming the fuzzy-set scores in the following analysis. For each index, higher values referto higher levels of governmental support. Minimums and maximums,including the corresponding country-times, are noted with each table.

Empirical methods

We combine two methodological approaches to examine our hypotheses.First, to assess causal relations, including the possibility of conjunctural

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 149


Table 5.2 Decade means and standard deviations for percentage civilian govern-ment employment


1960s 1970s 1980 1990s 1960s 1970s 1980s 1990s

Austria 7.11 9.52 11.33 13.39 0.43 1.03 0.57 0.70Belgium 5.80 7.80 9.56 9.65 0.54 0.85 0.16 0.12Canada 8.92 12.44 13.59 14.29 1.00 0.64 0.52 0.29Denmark 7.95 15.38 21.14 21.40 1.52 2.48 0.57 0.21Finland 5.46 8.95 12.78 13.85 0.81 1.33 1.01 0.61France 9.10 10.94 12.19 12.92 0.53 0.45 0.44 0.36Germany 5.61 7.57 8.63 8.60 0.51 0.62 0.23 0.18Ireland 5.93 7.62 8.72 8.36 0.43 0.56 0.18 0.06Italy 4.88 6.70 7.83 8.37 0.43 0.70 0.17 0.13Netherlands 5.06 6.06 6.50 6.24 0.12 0.29 0.12 0.07Norway 8.20 12.34 17.58 20.48 0.71 1.97 0.95 0.78Sweden 9.70 18.03 24.31 22.92 1.67 2.87 0.61 1.15United 10.23 13.23 13.53 10.45 0.73 0.89 0.19 1.76

KingdomUnited States 7.22 8.67 9.13 10.04 0.51 0.65 0.25 0.18

Note: The minimum value of 4.28 includes 1961 Italy; the maximum value of 24.97 includes1985 Sweden.

relations, we use Eliason and Styker’s (2007) goodness-of-fit extensionsof Ragin’s (2000) fuzzy-set methodology. To estimate causal effects aris-ing from the relations suggested in the fuzzy-set analysis, we use Imbensand Rubin’s (1997) and Hirano, Imbens, Rubin, and Zhou’s (2000) for-mulation of the intention-to-treat design to obtain what are known ascompliers average causal effects. We describe both these approaches andhow we combine them here.

Assessing causal relations

In the short time since Ragin (2000) initially developed fuzzy-set method-ology as a tool to assess causal relations, there have been increasingapplications of these methods (for example, Mahoney, 2003, Goertz andMahoney, 2005, and Stryker and Eliason, 2004), as well as a series ofmethodological advances (see, for example, Smithson 2005; Goertz andMahoney 2005). As initially developed by Ragin (2000),8 set-theoreticprinciples in general, and the subset principle specifically, are used toassess causal relations in a fuzzy-set framework. In its simplest form, the

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 150


Table 5.3 Decade means and standard deviations for maternity leave index(dashed entries indicate missing data for corresponding country/decade)


1960s 1970s 1980 1990s 1960s 1970s 1980s 1990s

Austria 0.52 0.50 0.51 0.51 0.01 0.01 0.01 0.02Belgium – – 0.36 0.44 – – 0.01 0.01Canada 0.00 0.00 0.35 0.40 0.00 0.00 0.01 0.02Denmark 0.00 0.42 0.63 0.75 0.00 0.22 0.05 0.04Finland 0.33 0.60 0.63 0.69 0.29 0.03 0.06 0.05France 0.48 0.53 0.53 0.54 0.01 0.02 0.02 0.01Germany 0.41 0.43 0.42 0.51 0.01 0.02 0.04 0.01Ireland 0.26 0.28 0.34 0.34 0.01 0.01 0.03 0.01Italy 0.39 0.45 0.45 0.43 0.01 0.03 0.02 0.01Netherlands 0.12 0.33 0.32 0.00 0.16 0.01 0.01 0.00Norway 0.00 0.41 0.52 0.80 0.00 0.04 0.05 0.07Sweden 0.59 0.81 0.87 0.90 0.04 0.12 0.01 0.07United Kingdom 0.00 0.15 0.15 – 0.00 0.19 0.19 –United States 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Note: The minimum value of 0.00 includes many countries (see 0.00 means and standard devi-ations indicating 0’s for the decade for some country); the maximum value of 0.98 includes1992 Sweden.

subset principle operates to establish the relation between a hypothe-sized cause (or conjunction of hypothesized causes) and some outcome.Fuzzy-set membership scores, reflecting the degree to which empiri-cal cases belong to some set, are used to assess the subset principle.9

Empirical evidence for causal necessity is obtained to the extent thatthe outcome can be established as a subset of the hypothesized cause.Evidence for causal sufficiency, on the other hand, is obtained to theextent that the hypothesized cause can be established as a subset ofthe outcome. Evidence for causal necessity and sufficiency combinedis obtained to the extent that the fuzzy scores on the hypothesized cause(or conjunction of causes) and outcome can be established as equal. SeeRagin (2000), Ragin and Pennings (2005), and Eliason and Stryker (2007)for details on the logic of fuzzy-set analysis. (See Appendix 5.2 for detailsdescribing the calculation of fuzzy-set scores used in our analysis.)

For the fuzzy-set portion of our analysis, we use Eliason and Stryker’s(2007) goodness-of-fit tests to identify potential factors, and conjunc-tions of factors, expected to produce high levels of female labor forceparticipation. Eliason and Stryker’s (2007) goodness-of-fit tests pro-vide an inferential framework within which to empirically assess subset

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 151


Table 5.4 Decade means and standard deviations for public day care index,children ages 0–2 (Dashed entries indicate missing data for correspondingcountry/decade)


1960s 1970s 1980 1990s 1960s 1970s 1980s 1990s

Austria – – 0.13 0.17 – – 0.02 0.02Belgium – – 0.14 0.13 – – 0.00 0.01Canada 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00Denmark 0.05 0.22 0.80 0.96 0.07 0.04 0.06 0.02Finland 0.00 0.29 0.46 0.52 0.00 0.08 0.06 0.03France – – 0.20 0.40 – – 0.02 0.01Germany – – 0.04 0.12 – – 0.00 0.04Ireland – – 0.03 0.04 – – 0.00 0.00Italy – – 0.08 – – – 0.00 –Netherlands 0.00 0.00 0.08 0.13 0.00 0.00 0.01 0.04Norway 0.00 – 0.27 0.53 0.00 – 0.00 0.16Sweden – – 0.89 0.84 – – 0.08 0.09United Kingdom – – 0.00 0.00 – – 0.00 0.00United States 0.00 0.01 0.01 0.01 0.00 0.00 0.00 0.00

Note: The minimum value of 0.00 includes many countries; the maximum value of 1.00includes 1998 Denmark.

relations. That inferential framework is based on the idea that fuzzy-setmembership scores are subject to measurement error and that measure-ment error, in turn, influences the assessment of the subset principle.The goodness-of-fit tests thus indicate whether the data are consistentwith some causal hypothesis up to a specified degree of measurementerror. While the logic of this approach is derived from fairly standardand widely used ideas in statistics, the empirical relationship assessed bythe goodness-of-fit test derives fully from the logic of the subset princi-ple described above. See Appendix 5.3 and Eliason and Stryker (2007) fordetails.

Recall that our general hypothesis, depicted in Figure 5.1, as well asour prior empirical work, suggests a causal chain in which high levelsof cumulative left cabinet incumbency give rise to an expanded pub-lic sector and increased family support policies. In turn, through themechanisms described above, public sector expansion and increasedfamily support policies heighten the demand for and facilitate the sup-ply of female labor. This results in higher rates of female labor forceparticipation. Our first objective for the fuzzy-set portion of the anal-ysis, therefore, is to test an alternative conjunctural hypothesis, in

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 152


Table 5.5 Decade means and standard deviations for Public Daycare Index, Chil-dren ages 3–school age (dashed entries indicate missing data for correspondingcountry/decade)


1960s 1970s 1980s 1990s 1960s 1970s 1980s 1990s

Austria – – 0.52 0.78 – – 0.06 0.10Belgium – – 0.31 0.26 – – 0.01 0.03Canada 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00Denmark 0.09 0.34 1.07 1.28 0.12 0.05 0.08 0.03Finland 0.00 0.40 0.64 0.81 0.00 0.11 0.08 0.02France – – 0.45 0.80 – – 0.04 0.00Germany – – 0.36 0.59 – – 0.02 0.01Ireland – – 0.14 0.21 – – 0.01 0.05Italy – – 0.31 – – – 0.01 –Netherlands 0.31 0.42 0.44 0.45 0.03 0.04 0.02 0.02Norway 0.00 – 0.49 0.75 0.00 – 0.00 0.01Sweden – – 1.15 1.09 – – 0.10 0.12United Kingdom – – 0.00 0.00 – – 0.00 0.00United States 0.01 0.03 0.03 0.03 0.01 0.00 0.00 0.00

Note: The minimum value of 0.00 includes many countries; the maximum value of 1.34includes 1998 Denmark.

which these factors act in a combinatoric contextual manner, ratherthan in the chain-like manner suggested by our hypothesis. To do this,we use Eliason and Stryker’s (2007) partitioning of the goodness-of-fitstatistic which tests whether a conjunction provides a better fit to thedata than does any component of the conjunction, including lower-order conjunctions and each of the individual factors making up theconjunction.

As we show below, the partitioning of the goodness-of-fit statisticsuggests that, for our data, we can reject the conjunctural hypothesis.Therefore, our second objective for the fuzzy-set portion of our analysisis to test how well the data conform to each portion of the hypothesizedcausal chain. To do this, we use Eliason and Stryker’s (2007) goodness-of-fit statistics to test causal hypotheses involving cumulative left cabinetincumbency and, in turn, civilian public employment, public day carefor children ages 0–2, public day care for children ages 3 to school age,maternity leave, extended leave, and family/child cash and tax benefits.We next use these goodness-of-fit tests to test causal hypotheses involv-ing female labor force participation as an outcome of each of the welfarestate policy measures and of civilian public employment.10

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 153


Table 5.6 Decade means and standard deviations for weeks of extended leave


1960s 1970s 1980s 1990s 1960s 1970s 1980s 1990s

Austria 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00Belgium 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00Canada 0.00 0.00 0.00 14.00 0.00 0.00 0.00 2.11Denmark 10.00 10.00 10.00 10.00 0.00 0.00 0.00 0.00Finland 9.96 16.60 24.36 26.30 8.57 0.00 4.09 0.00France 88.00 88.00 119.20 140.00 0.00 0.00 26.85 0.00Germany 0.00 0.00 29.60 129.60 0.00 0.00 26.04 30.36Ireland 0.00 0.00 9.60 12.00 0.00 0.00 5.06 0.00Italy 26.00 26.00 26.00 26.00 0.00 0.00 0.00 0.00Netherlands 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00Norway 0.00 52.00 36.40 0.00 0.00 0.00 25.12 0.00Sweden 26.00 26.00 26.00 26.00 0.00 0.00 0.00 0.00United Kingdom 0.00 16.00 40.00 40.00 0.00 20.66 0.00 0.00United States 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Note: The minimum value of 0.00 includes many countries; the maximum of 144 includes1992–1999 Germany.

Estimating causal effects

While we use fuzzy-set methods to assess potential causal factors giv-ing rise to high female labor force participation, we use Imbens andRubin’s (1997) and Hirano, Imbens, Rubin, and Zhou’s (2000) intention-to-treat (ITT) design to estimate the causal effects on female labor forceparticipation rates deriving from each of these factors.11 In the generalITT design, counterfactual causal treatment effects are obtained in partthrough some mechanism encouraging subjects to obtain some treat-ment. For example, Imbens and Rubin (1997) examine the causal effectof vitamin supplements (the treatments) on children’s survival rates invarious Indonesian communities in the 1990s. In that case, the encour-agement (or intention) to treat was given by the assignment of childrento receive the supplements, while the treatment itself was given by theactual receipt of the supplements. In the Hirano et al. (2000) study,the effectiveness of a flu vaccine is assessed, where there is imperfectcompliance with the encouragement to be vaccinated.

We apply this logic to estimate the causal effect that specific policiesmay have on female labor force participation. Consistent with results

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 154

154

Table 5.7 Decade means and standard deviations for Cash/Tax Family/ChildBenefits Index


1960s 1970s 1980s 1990s 1960s 1970s 1980s 1990s


Note: The minimum value of 0.04 includes 1960 Denmark; the maximum value of 0.63includes 1963 France.

Table 5.8 Decade means and standard deviations for female labor force partici-pation rates


1960s 1970s 1980s 1990s 1960s 1970s 1980s 1990s


Note: The minimum value of 0.26 includes 1960–65, 69 Netherlands; the maximum value of0.81 includes 1989 Sweden.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 155


Long-term leftincumbency

Variable:

Cumulative leftCabinetIncumbency

Public sector expansion

Variables:

Civilian GovernmentEmployment, Maternity leave,Public day care for ages 0–2,Public day care for ages3-School age, and family/childcash and tax benefits

Female laborforce

participation

Variable:

Female LFParticipationRate

Figure 5.1 Hypothesized causal chain, including variables used in the empiricalanalysis

from the fuzzy-set portion of our analysis (described below), we considerhigh levels of cumulative left cabinet incumbency as the mechanismthrough which specific policies are implemented and maintained. Thatis, high levels of cumulative left cabinet incumbency act as the encour-agement portion in the ITT design, while the specific policy itself (forexample, public daycare for young children) acts as the treatment.

To be more precise, our ITT analysis requires only the followingassumptions.

1. Mechanisms exist, though need not be observed, such that high levelsof cumulative left cabinet incumbency acts as an encouragement forgovernments to enact left-specific policies.

2. There are tendencies for governments to comply with these mech-anisms and tendencies for governments to not comply with thesemechanisms.

3. The tendency to not comply with these mechanisms is considered amixture of

a. the tendency to always enact a specific policy regardless of the levelof cumulative left cabinet incumbency and

b. the tendency to never enact a specific policy regardless of the levelof cumulative left cabinet incumbency.

4. The only non-negligible way that governments can have an impact onaggregate female labor force participation at time t is through policiesor related mechanisms operating at times prior to t.

5. Policies at time t and female labor force participation at time t + 1cannot affect cumulated left cabinet incumbency up to time t.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 156


All of these assumptions are consistent with our empirical findings fromthe fuzzy-set portion of our analysis (described below). Importantly,no additional assumptions – save for the parametric form of the log-likelihood – are necessary to generate intention-to-treat (ITT) effects(Hirano et al., 2000) or compliers average causal effects (CACE) (Imbensand Rubin, 1997).12 Below we adopt Imbens and Rubin’s (1997) approachand terminology because it is more widely adopted in the literature.

In general, the CACE identifies the expected counterfactual causaleffect on some outcome were a complier to change treatment statuses.This is the effect that would have been observed in the populationof compliers had the treatment been randomly assigned, as would bethe case in a typical experimental design framework. In our case, theCACE identifies, for the population of compliers as described in assump-tions 2 and 3 above, the counterfactual causal effect on female laborforce participation rates due to specific policies as enacted through leftgovernments. Thus, the estimated CACE gives the expected changein female labor force participation rates due to specific policies imple-mented through left governments.

Given that, to our knowledge, compliers average causal effects havenever been estimated in the context of fuzzy-set measures, we detailour CACE estimation in Appendix 5.4. To estimate the CACE, we use abootstrapped estimator on 1000 replications derived from the likelihoodgiven in Appendix 5.4. Using bootstrapped estimators alleviates the needto impose any distributional assumptions (such as normality) on theCACE itself. We thus present the median and inter-90% percentile rangefrom the bootstrapped empirical distribution function.13

Results

Fuzzy-set analysis: testing conjunctural relations

In this section we present results on tests for all possible conjunctionsas given by the full set of hypothesized causal factors. We are especiallyinterested in conjunctions involving cumulative left cabinet. If any ofthese conjunctions provide a better fit to the data when compared to eachindividual factor, then the causal chain hypothesis would necessarily beinconsistent with the data. Instead, the data would then be consistentwith a more complex conjunctural set of relations. Here we use Eliasonand Stryker’s (2007) partitioning of the F statistic to test whether eachhigher-order conjunction provides a better fit to the data than the set oflower-order conjunctions or the single factors nested within each higher-order configuration.14

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 157


Table 5.9 Select five-way and four-way partition tests of goodness-of-fit F statis-tics for the outcome “High Female Labor Force Participation – Subsequent Year.”Hypothesized causal factors, and corresponding letter representations, includeA = High Level Cumulative Left Cabinet IncumbencyC = High Level Public Sector EmploymentD = High Level Public Daycare, Ages 0–2E = High Level Public Daycare, Ages 3 to School AgeF = High Level Maternity LeaveG = High Level Family/Child Cash and Tax Benefits.

Conjunctions Causal hypotheses SD DF MSD F Pand partitions

Select Six-Way ConjunctionACDEFG Necessity 746.5978 248 3.0105 11.7258 0.0000

Sufficiency 0.7181 15 0.0479 0.1865 0.9997Necessity & Sufficiency 747.3159 263 2.8415 11.0677 0.0000ACDEF v ACDEFG – – – 0.7569 0.9879ACDEG v ACDEFG – – – 0.9759 0.5784ACDFG v ACDEFG – – – 1.0000 0.5000ACEFG v ACDEFG – – – 0.9341 0.7095ADEFG v ACDEFG – – – 1.0000 0.5000CDEFG v ACDEFG – – – 0.9887 0.5368

Select Five-Way ConjunctionACDEF Necessity 561.8962 248 2.2657 8.8250 0.0000

Sufficiency 3.762 15 0.2508 0.9769 0.4800Necessity & Sufficiency 565.6582 263 2.1508 8.3773 0.0000ACDE v ACDEF – – – 0.9627 0.6209ACDF v ACDEF – – – 0.9986 0.5046ACEF v ACDEF – – – 0.9113 0.7741ADEF v ACDEF – – – 1.0000 0.5000CDEF v ACDEF – – – 0.9527 0.6526

Note: See Appendix 5.5 for complete set.

Table 5.9 gives goodness-of-fit statistics and partitions for select five-way and four-way conjunctions derived from the full set of factorsdiscussed above, with respect to their causal relation with the set highfemale labor force participation in the subsequent year. The full set ofpartitions and tests can be found in Appendix 5.5. Note that the par-tition tests include only those cases (country-times) for which we havevalid empirical information on all factors. This reduces the total numberof cases to 263 for these tests.

Since no conjunction fits the causal necessity and sufficiency hypoth-esis, we begin with the set of six-way conjunctions that fit the causal

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 158


sufficiency hypothesis, and then test whether lower order conjunctionsfit the data as well as the higher order conjunction. If all tests indicatethat lower-order conjunctions fit just as well as the parent higher-orderconjunction (tests with p-values greater than 0.05), we then continue tothe next set of conjunctions. This approach will lead to the lowest-order,most general, sufficient conjunctural or single-factor condition that fitsthe data. See Eliason and Stryker (2007) for details.

The first panel of Table 5.9 gives select six-way conjunctions, andAppendix 5.5 provides test statistics for the remaining six-way con-junctions. All six-way conjunctions provide a strong fit between thedata and the causal sufficiency hypothesis (all p-values are greater than0.05). In other words, these data are consistent with all possible six-way conjunctions of the causal factors considered in our analysis – highlevel of cumulative left cabinet incumbency; high public sector employ-ment; high levels of daycare, maternity, and extended leave policies;and high levels of family/child cash and tax benefits – combining fora causally sufficient relation with high female labor force participation.The question then becomes, do any or all of the five-way conjunctionsnested within these six-way conjunctions provide equally good fits to thehypothesis of causal sufficiency? If so, there is redundant empirical infor-mation in the six-way conjunction used to explain female labor forceparticipation.

Partitioning the F statistic from this set of six-way conjunctions intoeach nested five-way conjunction reveals that none of the six-way con-junctions provides a better fit to these data than each of the nestedfive-way conjunctions. This is given by a p-value greater than 0.05 foreach of the partition test statistics, and indicates that the six-way con-junction contains redundant empirical information in explaining femalelabor force participation. Similarly, the partitions on each of the pos-sible five-way conjunctions suggest that each of the nested four-wayconjunctions fit these data just as well as the parent five-way conjunc-tion. Continuing on in this manner, the information in Table 5.9 and inAppendix 5.5 reveals that no higher-order conjunction provides a betterfit to these data for any causal hypothesis than the fit achieved by eachfactor alone.

In sum, these data do not support the idea that conjunctions providea better understanding of female labor force participation beyond thatgained by examining each potential causal factor separately. In the nextsection we analyze each factor separately, to examine further the ideaof a causal chain in which, once left political parties accumulate power,they better implement and subsequently maintain family policies. These,

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 159


along with left governments’ expansion of the public sector, create thecondition for high female labor force participation.

Fuzzy-set analysis: assessing the impact of left governancelevels on policy

Having rejected the set of conjunctural hypotheses, in this section weturn to replicating our prior analyses on the larger, more represen-tative 14 country data set and for the substantially longer 1960–99time frame.15 We also consider extended leaves as an additional policyoutcome. Overall, there is substantial similarity, but there also are someimportant differences between the findings of our current analysis andthose of our earlier six-country, 1977–94 study. Given that the com-bination of an expanded set of countries and the longer time framesubstantially shifts the maximum and minimum values on most ofthe variables used to construct membership scores, such differences inresults should come as no surprise. (See Appendix 5.2 for details on theconstruction of membership scores.)

For each causal hypothesis, Table 5.10 gives the goodness-of-fit statis-tics for the relationship between female labor force participation and allthe factors examined in our past work, as well as for extended leave(Stryker and Eliason, 2004).16 Specifically, Table 5.10 gives the sumsof squared distances (SSD) from that expected under each hypothesis,the corresponding degrees of freedom (DF) and mean squared distance(MSD). Table 5.10 also provides the goodness-of-fit F statistic (F) and cor-responding descriptive level of significance (P). The SSC, DF, MSD, F, P,and therefore the goodness-of-fit tests are all unaffected by the omissionof any third factor. Because there are no parameters estimated from thesedata to calculate any of the relevant quantities that enter the goodness-of-fit tests, any concern about biased estimation or results due to, e.g.,spuriousness, is misplaced (see also Appendix 5.3, Eliason and Stryker,2007 for details). Those who are used to employing regression methods,in which we must always concern ourselves with the possibility of modelmisspecification, may find this refreshing.

As shown in Table 5.10, a high level of cumulative left governancecontinues to exhibit a strong causal relationship with high levels ofpublic sector expansion for data representing 1960–99 Austria, Belgium,Canada, Denmark, Finland, France, Germany, Ireland, Italy, the Nether-lands, Norway, Sweden, Britain, and the United States. However, whereasour earlier analyses for 1977–94 Denmark, Britain, France, Germany,Italy, and Belgium supported only a relationship of causal necessity, forthe larger, more representative group of advanced capitalist democracies

9780230_202573_06_cha05.tex

5/6/200815:37

Page160

160

Table 5.10 Goodness-of-fit for relationships between hypothesized causal condition “High Level of Cumulative Left Cabinet Incum-bency” and select outcomes. Bold-faced and italicized entries indicate a good fit between the hypothesized causal relation and thedata at the .05 and .01 levels respectively

Condition Outcome Causal hypothesis SSD DF MSD F P

High Level of High Level of Public Causal Necessity 171.62 291 0.5898 2.30 0.0000Cumulative Left Sector Employment Causal Sufficiency 37.29 193 0.1932 0.75 0.9891Cabinet Incumbency Causal Nec & Suf 208.91 490 0.4264 1.66 0.0000

High Level of Public Causal Necessity 24.11 71 0.3395 1.32 0.0607Daycare Ages 0–2 Causal Sufficiency 95.57 121 0.7898 3.08 0.0000

Causal Nec & Suf 119.68 262 0.4568 1.78 0.0000

High Level of Causal Necessity 41.47 137 0.3027 1.18 0.1270Public Daycare Causal Sufficiency 46.07 71 0.6489 2.53 0.0000Ages 3–School Age Causal Nec & Suf 87.54 278 0.3149 1.23 0.0446

High Level of Causal Necessity 267.84 320 0.8370 3.26 0.0000Maternity Leave Causal Sufficiency 62.97 87 0.7237 2.82 0.0000

Causal Nec & Suf 330.81 450 0.7351 2.86 0.0000

High Level of Causal Necessity 184.80 90 2.0533 8.00 0.0000Extended Leave Causal Sufficiency 435.39 321 1.3564 5.28 0.0000

Causal Nec & Suf 620.19 490 1.2657 4.93 0.0000

High Level of Causal Necessity 390.12 298 1.3091 5.10 0.0000Family/Child Cash & Causal Sufficiency 172.50 185 0.9324 3.63 0.0000Tax Benefits Causal Nec & Suf 562.62 484 1.1624 4.53 0.0000

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 161


and longer time frame, we find instead that a high level of cumulativeleft cabinet incumbency is causally sufficient for a high level of publicsector employment. Results of the goodness-of-fit F tests indicate thatthese data fit the causal sufficiency hypothesis extremely well (F = 0.75and P = 0.9891).

Goodness-of-fit tests given in Table 5.10 also show that the data areconsistent with a causally necessary relationship between high levels ofcumulative left governance and high levels of public day care for chil-dren 0–2 (F = 1.32, P = 0.0607) and children 3 to school-age (F = 1.18,P = 0.1270). Perhaps most interesting in Table 5.10 is that, using a slightlymore liberal Type I error rate of 0.01 for the goodness-of-fit test, thesedata are consistent with a causally necessary and sufficient relationshipbetween high levels of cumulative left governance and high levels ofpublic day care for children 3 to school-age (F = 1.23, P = 0.0446). Finally,as we found in our prior analyses, the current analysis fails to supportany causal relation between high levels of cumulative left governanceand either high levels of maternity leave or high levels of family/childcash and tax benefits. In addition to these null results, there is no supportfor any causal relation between high levels of cumulative left governanceand high levels of extended leave.

Fuzzy-set analysis: causal relations with high female laborforce participation

Table 5.11 examines causal hypotheses linking public sector expansionand family policies to female labor force participation measured in thesubsequent year. Goodness-of-fit tests in Table 5.11 reveal that a highlevel of public sector employment is causally sufficient for high femalelabor force participation (F = 0.78, P = 0.8996), as are a high level of pub-lic day care for children ages 0–2 (F = 0.82, P = 0.4828) and a high levelof maternity leave (F = 0.97, P = 0.6041). Perhaps surprisingly, results inTable 5.12 also show no support for any causal hypothesis linking highlevels of public day care for older children with high levels of femalelabor force participation. However, inspection of the fuzzy-set graph forthis relation (not shown) suggests that the entire lack of fit in these datafor this relation is attributable to the Netherlands. This is confirmedin that, once the Netherlands data are removed from consideration,the goodness-of-fit F test reveals a strong fit to the causal sufficiencyhypothesis for the remaining countries (F = 0.23, P = 0.9193; not shownin Table 5.11). Finally, there is no support in these data for a causalrelation between high levels of extended benefits and female labor force

9780230_202573_06_cha05.tex

5/6/200815:37

Page162

162Table 5.11 Goodness-of-fit for relationships between outcome “High Female Labor Force Participation – Subsequent Year” and selecthypothesized causal conditions. Bold-faced and italicized entries indicate a good fit between the hypothesized causal relation andthe data at the .05 and .01 levels respectively

Outcome Condition Causal hypothesis SSD DF MSD F P

High Female LaborForce Participation – High Level of PublicSubsequent Year Sector Employment Causal Necessity 301.37 460 0.6551 2.55 0.0000

Causal Sufficiency 14.05 70 0.2007 0.78 0.8996Causal Nec & Suf 315.42 530 0.5951 2.32 0.0000

High Level of PublicDaycare Ages 0–2 Causal Necessity 892.88 276 3.2351 12.60 0.0000


High Level ofPublic DaycareAges 3–School Age Causal Necessity 735.22 267 2.7536 10.73 0.0000


High Level ofMaternity Leave Causal Necessity 642.98 304 2.1151 8.24 0.0000


High Level ofExtended Leave Causal Necessity 1879.53 471 3.9905 15.54 0.0000


High Level ofFamily/Child Cash &Tax Benefits Causal Necessity 582.98 339 1.7197 6.70 0.0000


9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 163


Table 5.12 Bootstrapped EDF estimates of compliers average causal effects onfemale labor force participation rates, with strong left political tradition as theinstrument in the intention-to-treat analysis∗

Bootstrapped EDF estimates

5th percentile Median 95th percentile

Demand-Side FactorPublic Sector Employment 0.35 0.38 0.40

Supply-Side PoliciesPublic Daycare – Ages 0–2 0.08 0.11 0.12Public Daycare – Ages 3 to School Age 0.07 0.09 0.11Maternity Leave 0.16 0.18 0.20Extended Leave −0.07 0.00 0.03Family/Child Cash & Tax Benefits −0.25 −0.23 −0.21

∗Bootstrapped estimates are based on 1000 replications. Country-times with high likelihoodsof strong left political traditions include Norway after 1992 and Sweden after 1987. Like-lihoods of strong left political traditions are based on cumulative left cabinet incumbencyfuzzy-set scores. See text for details.

participation, nor for high levels of family/child cash and tax benefitsand female labor force participation.

Summarizing the results in Tables 5.10–5.11, the fuzzy-set goodness-of-fit F tests and partitions indicate that these data are consistent withthe following:

1. High levels of cumulative left governance are (a) causally sufficientfor high levels of public sector employment and (b) causally neces-sary for high levels of public daycare for younger and older children.Somewhat weaker, though detectable, evidence exists suggesting acausally necessary and sufficient relationship between high levels ofcumulative left governance and high levels of public daycare for olderchildren.

2. High levels of public sector expansion, public daycare for youngerchildren, and maternity leave are, separately, causally sufficient forhigh levels of female labor force participation. Excluding the Nether-lands, a high level of public day care for older children is also causallysufficient for high levels of female labor force participation.

3. No hypothesis involving two-way or higher-order conjunctions pro-vides a better fit to these data than the series of hypotheses involvingsingle factors.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 164


CACE estimates of supply and demand factors on femalelabor force participation

For each policy and for public sector expansion, Table 5.12 gives themedian, the lower 5th percentile, and the upper 95th percentile fromthe bootstrapped empirical distribution estimate of the CACE. These per-centiles are similar to confidence intervals. However, no distributionalassumptions are made because the percentiles derive from the observedboostrapped empirical distribution function (EDF), which was obtainedon 1000 replicates. For useful overviews and details on various boot-strapped estimators, see Efron and Tibshirani (1994) and DiCiccio andEfron (1996). Importantly, the CACE, as derived by Hirano et al. (2000)and Imbens and Rubin (1997), is unaffected by traditional notions ofspuriousness.17

High levels of public sector employment, the single demand-side fac-tor, have a substantial effect on female labor force participation rates. Themedian causal effect for that factor is an increase of 38 percentage points,with lower and upper bounds on that effect of 35 and 40 percentagepoints respectively.18 Thus, the expansion of public sector employmentas generated through cumulative left governance is responsible for a sub-stantial increase in female labor force participation. It is, in fact, thesingle most important factor in our analysis in accounting for high levelsof female labor force participation.

Of the supply-side policies, maternity leave has the largest causal effecton female labor force participation rates, with a median effect of 18 per-centage points, and lower and upper bounds of 16 and 20 percentagepoints respectively. The CACE estimate for public day care for youngerchildren is in the range of 8 to 12 percentage points with a median effectof 11 points. For public day care for older children, the range of the effectis from 7 to 11 percentage points, with a median effect of 9 points. Forextended leave, the median of 0 and lower and upper bounds of −7 and+3 suggest no effect. For family/child cash and tax benefits, the medianof −23 and bounds of −25 and −21 suggest a substantial negative causaleffect.19

To understand better the causal effects highlighted by the CACE,recall that the CACE identifies the counterfactual causal effect onfemale labor force participation rates due to specific factors as operatingthrough cumulative left governance. Thus, the estimated CACE givesthe expected change in female labor force participation rates due to fac-tors and policies that are an outgrowth of left government legacies.20

Importantly, these CACE estimates do not give the effects due to some

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 165


Table 5.13 Bootstrapped EDF estimates of compliers average causal effects onfemale labor force participation rates, with all other non-left political traditionsas the instrument in the intention-to-treat analysis∗

Bootstrapped EDF estimates

5th percentile Median 95th percentile

Demand-Side FactorPublic Sector Employment 0.45 0.46 0.48

Supply-Side PoliciesPublic Daycare – Ages 0–2 0.14 0.14 0.15Public Daycare – Ages 3 to School Age 0.46 0.48 0.51Maternity Leave 0.36 0.39 0.41Extended Leave −0.18 −0.16 −0.14Family/Child Cash & Tax Benefits −0.04 −0.02 −0.01

∗Bootstrapped estimates are based on 1,000 replications. Country-times with high likelihoodsof other non-left political traditions include France prior to 1982, Germany prior to 1973,Ireland prior to 1993, Italy prior to 1989, and Canada and the US over the entire time period1960–99. Likelihoods of other non-left political traditions are based on the complement ofthe cumulative left cabinet incumbency fuzzy-set membership scores. See text for details.

specific left government’s policy implementation and/or maintenance.Rather, these CACE estimates give the effects due to each factor or policyas implemented by any government where strong prior left governmentlegacies are observed.

But what about other political legacies? To answer that question weestimate CACE where, now, low levels of cumulative left governance actsas the instrument in the intention-to-treat analysis. Given in Table 5.13,this set of CACE estimates tell us the expected change in female laborforce participation rates due to factors and policies implemented by gov-ernments in country-times with no strong left government legacy. Suchcountries in our data include current-day Canada and the United States.

Table 5.13 shows once again that expanded public sector employ-ment has a substantial impact on female labor force participation,with a median CACE of 46 percentage points and bounds of 45 and48. Thus, it appears that an expanded public sector would have asignificant effect on female labor force participation regardless of thepolitical tradition. This, in turn, gives strong credence to the demand-side component of our theoretical model, and suggests that the narrowcreation of public sector jobs with traditionally female-tasked skills

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 166


drives a large portion of any observed increase in female labor forceparticipation.

Also showing very strong positive effects in Table 5.13 are public daycare for older children and maternity leave programs, with median effectsof 48 and 39 percentage point increases respectively. These effects are sub-stantially higher for these factors than those found in Table 5.12 with thestrong left government legacy instrument. This suggests that, were theseprograms implemented in countries with non-left political legacies (suchas Canada and the United States), substantially higher levels of femalelabor force participation than are currently observed might be achieved.Although considerably smaller in magnitude, results in Table 5.13 alsoshow that public day care programs for younger children are likely to netan increase in female labor force participation, with a median CACE of14 percentage point increase.

By way of contrast, both extended leaves and family/child cash and taxbenefit programs have negative impacts on female labor force participa-tion rates, with median CACEs of −16 and −2 respectively and with bothupper bounds negative. This suggests that both extended leaves and fam-ily/child cash and tax benefit programs would likely decrease female laborforce participation were they implemented in countries with non-leftpolitical legacies.

Discussion and conclusions

Both the fuzzy-set and CACE analyses and results have important schol-arly and policy implications. For these analyses, we have compiled andused data for 14 countries, including Austria, Belgium, Britain, Canada,Denmark, Finland, France, Germany, Ireland, Italy, the Netherlands,Norway, Sweden, and the United States. We likewise have compileddata for a substantially longer time frame – 1960–99 – than has beenmobilized in prior aggregate level cross-national research on the rela-tionship between diverse family policies and female participation inpaid employment. Consistent with our own theoretical understandingand predictions, we have shown that both supply-side and demand-side mechanisms must be considered by scholars and policy makerspromoting female labor force participation as part of a redesign forthe twenty-first-century welfare state. We have further shown that, interms of fit to the empirical information, it is unnecessary to considerthese mechanisms as operating in some complex conjunctural fashion.Instead, the data are more consistent with understanding these mecha-nisms as operating in an individual and serial fashion. This is consistent

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 167


with conceptualizing the process as the causal chain described above andin Figure 5.1.

Consistent with the general thrust of prior research, the fuzzy-setportion of our analysis revealed that high levels of cumulative left gover-nance are causally sufficient, albeit not necessary, to create an expandedpublic sector. In turn, an expanded public sector is sufficient, albeit notnecessary, for high female labor force participation. Our CACE anal-ysis further showed a causal effect of 35 to 40 percentage points onfemale labor force participation rates, due to an expanded public sectoras resulting from high levels of cumulative left governance.

Consistent with our fuzzy-set findings suggesting a relationship ofsufficiency but not necessity between cumulative left governance andpublic sector expansion, we were led to ask whether a large public sec-tor achieved in the absence of high cumulative left governance wouldnonetheless increase female labor force participation. This would allowus to estimate the causal effect of public sector expansion in countries likeFrance, which have relatively large public sectors but also high cumula-tive experience with secular right governments. As scholars have noted,France’s long-standing “strong state” tradition and its across the politicalspectrum commitment or willingness to tolerate government interven-tion in the economy has produced relatively high levels of civiliangovernment employment (see Stryker and Eliason, 2004).

When we added to fuzzy-set results by conducting the CACE analysisusing non-left political traditions as the intention-to treat component,we found an effect of 45–48 percentage points (Table 5.13). This showsthat an expanded public sector would indeed have a significant effecton female labor force participation regardless of the political tradition.Thus, although cumulative left governance does provide a route to publicsector expansion, and public sector expansion does provide a route torather substantial gains in female labor force participation rates, this isnot the only route to this end.

Similarly, whereas the demand-side mechanism of public sectorexpansion promotes increased female labor force participation, such anincrease can also be achieved by alternative, supply-side routes. Ourfuzzy-set analysis showed that high levels of public day care for youngerchildren and high levels of maternity leave were individually sufficientfor high female labor force participation. Removing the Netherlandsfrom the analysis further revealed that high levels of public day care forolder children are sufficient for high female labor force participation. Asfurther elaborated in the Misra and Topp chapter in this volume, publicopinion in the Netherlands continues to place a particularly high value

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 168


on mothers rearing children in the ho,e relative to support levels for thisvalue in other European countries. It is possible that the strength of thisvalue in the Netherlands encourages mothers there to remain outsidethe labor force even when the public day care that elsewhere facilitateslabor force participation of such mothers is provided.

Our CACE analyses further shows that the causal effects of maternityleave and public day care programs on female labor force participationare non-negligible, although in some cases they are modest in compari-son to demand-side mechanisms. For countries with strong left politicallegacies (Table 5.12), we see effects in the neighborhood of 10 percentagepoints deriving from both types of public day care and an upper boundof 20 percentage points due to maternity leave. For countries with otherpolitical legacies (Table 5.13), we see substantially higher effects due topublic day care for older children (in the range of 46 to 51 percentagepoints) and due to maternity leave (in the range of 36 to 41 percentagepoints).

Finally, our analysis shows that there is no evidence in these data tosuggest that either extended leaves or family/child cash and tax benefitsoperate to increase female labor force participation. On the contrary, theCACE analysis reveals that both of these programs would do harm in thisregard. For countries with strong left political legacies, extended leaveshave no effect, while introducing family/child cash and tax benefits pro-duces negative consequences for female labor force participation. Forcountries with other political legacies, family/child cash and tax bene-fits have negligible negative effects, whereas extended leave policies havemore substantial negative effects on female labor force participation.Our findings for extended leaves complement and further nuance find-ings from regression analyses suggesting that leaves will have a positiveimpact up to some threshold point of duration, but that very lengthyleaves are likely to have a negative impact (Pettit, 2006).

Because none of the factors in our fuzzy-set analysis exhibited acausally necessary relationship with high levels of female labor forceparticipation, we cannot infer that the absence of publicly provided daycare or maternity leave policies invariably is a barrier to women’s entryinto paid employment (cf. Esping-Andersen et al., 2002). On the con-trary, countries with liberal welfare state regimes, such as the UnitedStates, have managed to achieve relatively high rates of female laborforce participation, albeit much of it through the low-wage service-sectorroute.

Thus, on the one hand, welfare state policy makers would be correct toinfer from our analysis that achieving high levels of female labor force

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 169


participation does not require them to invest heavily in publicly pro-vided or subsidized day care. But this is not to suggest that all is well.Government commitment to the provision of affordable day care ona large scale may well be causally related to women’s capacity to workfull-time and/or in higher-quality jobs with better wages, benefits, andworking conditions. Thus, government commitment to providing avail-able, affordable child care may well help diminish gender inequalitiesin earnings and other employment conditions. In future research, weplan to pursue these key questions for scholarship and policy making,by investigating how sectoral employment patterns and diverse typesand aspects of family policies are related to variations in the type andquality of women’s employment across countries and over time.

From a methodological standpoint, we have shown how fuzzy-setmethods to assess causal relations can be fruitfully combined with sta-tistical methods to assess causal effects for non-experimental data. Wehave shown how goodness-of-fit tests developed by Eliason and Stryker(2007) may be employed in fuzzy-set analysis to assess hypotheses aboutcausal relations, including hypotheses that include higher-order con-junctions. Likewise, we have shown how an intention-to-treat formof analysis as developed by Imbens and Rubin’s (1997) and Hirano,Imbens, Rubin, and Zhou (2000), along with the bootstrapped estima-tion of the compliers average causal effect, can be leveraged to obtaina distribution on specific causal effects. Using all these tools togetherallows us to better answer the scholarly and policy questions with whichwe began.

We close by noting that, for our data set of 14 advanced capitalistdemocracies from 1960 to 1999, our fuzzy-set analysis showed that highlevels of cumulative social democratic governance were necessary, albeitnot in themselves sufficient, for high levels of maternity leave and forhigh levels of public day care both for younger and for older children.This is an important finding, because it suggests that, historically speak-ing, without high levels of cumulative social democratic governance, keyfemale-friendly and family-friendly policy measures are not likely to beachieved. Citizens in advanced industrial democracies could usefullyconsider this finding when they decide whether and how to partici-pate in electoral politics, including deciding whether – and for whichparty – they will vote. Our future research will try to establish what otherfactors, in combination with cumulative social democratic governance,might produce a causal relationship of necessity and sufficiency to pro-vision of high levels of publicly provided day care and other family andfemale-friendly policies.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 170


Appendix 5.1: Data Sources

Baker, Maureen. 1995. Canadian Family Policies: Cross-National Compar-isons. Toronto: University of Toronto Press.

Bastelaer, Alois van, Georges Lemaitre, and Pascal Marianna. 1997. “TheDefinition of Part-Time Work for the Purpose of International Com-parisons.” Labour Market and Social Policy Occasional Papers. Paris:OECD 22.

Bruning, C. and J. Plantega. 1999. “Parental Leave and Equal Opportu-nities: Experiences in Eight European Countries.” Journal of EuropeanSocial Policy 9(3).

Caramani, Daniele. forthcoming. The Formulation of National Electionsand Party Structure in Europe: A Comparative and Historical Study.

Census Bureau. 2000. Statistical Abstract of the United States. Washington:US Government Printing Office.

Cochran, Moncrieff. 1993. International Handbook of Child Care Policiesand Programs. London: Greenwood Press.

Council of Europe. 1982–98. Comparative Tables of Social Security Schemes:In Council of Europe Member States and Other Countries: General Scheme.1st–9th Editions Strasbourg, Germany: Council of Europe Press.

Commission of the European Communities. 1970–1984. ComparativeTables of Social Security Schemes: In Member States: General Scheme.4th–13th Editions Strasbourg, Germany: Council of Europe Press.

Commission of the European Communities. 1990. Childcare in the Euro-pean European Communities. Women of Europe Supplements No. 31.Brussels: Commission of the European Communities.

Esping-Andersen, Gøsta, et al. 2002. Why We Need a New Welfare State.Oxford: Oxford University Press.

EURYDICE and CEDEFOP. 1990. Structures of Education and Initial Train-ing Systems in the Member States of the European Community. Brussels:EURYDICE and CEDEFOP.

European Commission Network on Childcare and Other Measures toReconcile Employment and Family Responsibilities. 1995. A Reviewof Services for Young Children in the European Union. Brussels: EuropeanCommission.

European Commission. 1998. Care in Europe: Joint Report of the “Genderand Employment” and the “Gender and Law” Groups of Experts.

European Commission, Eurydice, and Eurostat. 2000. Key Data onEducation in Europe: Primary Education 1999/2000.

European Journal of Political Research. 1995–2000. “Issues in NationalPolitics.” Various Authors.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 171


Gauthier, Anne H. 1996. The State and the Family: A Comparative Analysisof Family Policies in Industrialized Countries. Oxford: Clarendon Press.

Gornick, Janet C., Marcia K. Meyers, and Katherine E. Ross. 1996.“Supporting the Employment of Mothers: Policy Variation AcrossFourteen Welfare States.” Journal of European Social Policy 7(1): 45–70.

Hofferth, Sandra J. et al. 1991. National Child Care Survey, 1990.Washington: Urban Institute Press.

Huber, Evelyne, Charles Ragin, and John D. Stephens. 1997. ComparativeWelfare States Data Set. Available at: www.lisproject.org/publications/welfaredata/welfareaccess.htm.

International Labour Office. 1985. Maternity Benefits in the Eighties: AnILO Global Survey (1964–84). Geneva: International Labour Office.

International Labour Office. 1994. “Maternity and Work.” Conditions ofWork Digest Vol. 13. Geneva: International Labour Office.

International Social Security Association. 1999. Social Security ProgramsThroughout The World. Available at: www.issa.int/ssw.

Kamerman, Sheila. 1980. “Maternity and Parental Benefits and Leaves:An International Review.” Impact on Policy Series Monograph. New York:Columbia University.

Kamerman, Sheila. 2000. “Early Childhood Education and Care: AnOverview of Developments in the OECD Countries.” InternationalJournal of Educational Research 33: 7–29.

Kamerman, Sheila and Alfred J. Kahn. 1975. Child Care Programs in NineCountries: A Report Prepared for the OECD Working Party on the Role ofWomen in the Economy. Washington, DC: US Department of Health,Education, and Welfare.

Kamerman, Sheila and Alfred J. Kahn. 1978. Family Policy: Governmentand Families in Fourteen Countries. New York: Columbia UniversityPress.

Kamerman, Sheila and Alfred J. Kahn. 1981. Child Care, Family Benefitsand Working Parents: A Study in Comparative Policy. New York: ColumbiaUniversity Press.

Kamerman, Sheila and Alfred J. Kahn. 1983. Income Transfers for Familieswith Children: An Eight-Country Study. Philadelphia: Temple UniversityPress.

Kamerman, Sheila and Alfred J. Kahn. 1991a. Child Care, Parental Leave,and the Under 3s: Policy Innovation in Europe. New York: Auburn House.

Kamerman, Sheila and Alfred J. Kahn. 1991b. “Government Expen-ditures for Children and Their Families in Advanced IndustrializedCountries, 1960–1985.” Innocenti Occasional Papers: Economic Pol-icy Series No. 20.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 172


Kamerman, Sheila, Alfred J. Kahn, and Paul Kingston. 1983. MaternityPolicies and Working Women. New York: Columbia University Press.

Kreyenfeld, Michaels and Karsten Hank. 2000. “Does the Availabil-ity of Childcare Influence the Employment of Mothers? Findingsfrom Western Germany.” Population Research and Policy Review 19 (4)317–37.

MZES/Eurodata. 2002. European Family Policy Database. Available fororder online at: www.mzes.uni-mannheim.de/fs_daten_e.html.

O’Connor, Julia S., Ann Shola Orloff, and Sheila Shaver. 1999. States, Mar-kets, Families: Gender, Liberalism, and Social Policy in Australia, Canada,Great Britain, and the United States. Cambridge: Cambridge UniversityPress.

OECD. 1974. Educational Statistics Yearbook, Vol. 1: International Tables.Paris: OECD.

OECD. 1975. Educational Statistics Yearbook, Vol. 2: Country Tables. Paris:OECD.

OECD. 1981. Educational Statistics in OECD Countries. Paris: OECD.OECD. 1982. Employment in the Public Sector. Paris: OECD.OECD. 1984a. Tax Expenditures: A Review of the Issues and Country Practices.

Paris: OECD.OECD. 1984b. “Social Expenditure: Erosion or Evolution.” OECD

Observer 126: 3–6.OECD. 1989, 1990, 1993a. Education in OECD Countries: A Compendium

of Statistical Information 86–87, 87–88, 88–89 and 89–90. Paris: OECD.OECD. 1993b. Private Pay for Public Work: Performance-related Pay for Public

Sector Managers. Paris: OECD.OECD. 1994a. Women and Structural Change: New Perspectives. Paris:

OECD.OECD. 1994b. Educational Statistics in OECD Countries. Social Policy

Studies, No. 12. Paris: OECD.OECD. 1999. Labour Force Statistics: 1978–1999. Paris: OECD. Available

at: www.sourceoecd.org.OECD. 2001. Social Expenditure Database. Paris: OECD. Available at:

www.sourceoecd.org.OECD. 2002. Economic Outlook Database. Paris: OECD. Available at:

www.sourceoecd.org.Olmsted, Patricia P and David Weikart. 1989. How Nations Serve Young

Children: Profiles of Child Care and Education in 14 Countries. Michigan:High/Scope Press.

Randall, Vicky. 2000. “Childcare Policy in the European States: Limits toConvergence.” Journal of European Public Policy 7: 346–68.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 173


Rønsen, Marit and Marianne Sundstrom. 2002. “Family Policy and after-Birth Employment Among New Mothers: A Comparison of Finland,Norway and Sweden.” European Journal of Population 18: 121–52.

UN Department of Economic and Social Affairs. 1995–1998. DemographicYearbook Nos 47–50. New York: United Nations.

US Department of Education. 1999. “Preprimary Education Enrollment.”Statistics in Brief: National Center for Education Statistics. Washington,DC: US Department of Education.

US Department of Education. 1995. Project Head Start Statistical Fact Sheet.Varley, Rita. 1986. “The Government Household Transfer Data Base.”

OECD Department of Economics and Statistics Working Papers No. 36.Paris: OECD

Wennemo, Irene. 1994. Sharing the Costs of Children: Studies on the Devel-opment of Family Support in the OECD Countries. Swedish Institute forSocial Research No. 25.

West, Jerry. 1995. “Child Care and Early Education Program Participa-tion of Infants, Toddlers, and Preschoolers.” Statistics in Brief: NationalCenter for Education Statistics US Department of Education.

Appendix 5.2: Coding Fuzzy-Set Membership Scores

A fuzzy-set membership score reflects the degree to which a case belongsto a set as defined on some hypothesized cause or on some outcome.Because the subset principle is critically sensitive to choice of measure-ment, great care is required in coding membership scores. Researcherswill necessarily be confronted with two primary considerations. Thefirst is the logic used to establish minimum and maximum membershipscores. The second is the logic underlying the distribution of empiri-cal cases between that minimum and maximum. For each, there are anumber of informed choices.

For the current analysis, we use what we have elsewhere termed aMin/Max Uniform Distribution coding (Stryker and Eliason, 2004).21 Thismeasurement logic depends on a continuous distribution of informationon the attribute of interest.22 Membership scores are then constructedbased on that information, uniformly distributed across a range definedon the observed minimums and maximums in the data itself. That is,the fuzzy-set membership score si is defined as

si = xi − min {X}max {X} − min {X}

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 174


where X gives the measurement on the attribute of interest and xi is theobservation for the ith case.

For this type of coding, a case with the maximum score on the attributeof interest will have a membership score of 1, indicating certainty (upto measurement error) that that case belongs to the corresponding set.A case with the minimum score on the attribute of interest will have amembership score of 0, indicating certainty (up to measurement error)that that case does not belong to the corresponding set. All other degreesof belonging to the set are measured on a uniform distribution relativeto the range given by [min {X}, max {X}] on the original information. Forexample, the case in our data that has the maximum score on “Cumula-tive Left Cabinet Incumbency” (1999 Sweden) is measured as belongingto the set “High Cumulative Left Cabinet Incumbency” with member-ship score 1. The cases in our data that have the minimum score onthe measure “Cumulative Left Cabinet Incumbency” (United States andCanada, 1960–1999) are measured as belonging to the set “High Cumula-tive Left Cabinet Incumbency” with membership score 0. All other caseshave membership scores relative to that maximum and minimum.

Distributions other than the uniform, or none for that matter, couldbe used to distribute cases across the range given by the min and max.Using no reference distribution exposes coding, and thus the analysis,too much to known and unknown vagaries of the individual researcher.This, in turn, leaves the validity of the analysis in question. Using a morecomplex reference distribution can produce empirical results that areartifacts of the complexity of the distribution chosen. Using the uniformdistribution, which gives equal weight to cases across the range of theempirical information, avoids both these potential problems.

Additionally, researchers may wish to consider minimums and maxi-mums outside the range of the observed data when constructing mem-bership scores. This is not advisable, however, as moving off the datarange will necessarily bias the subset assessment up to known multi-plicative and additive factors, and will thus adversely affect internalvalidity.

To understand this more precisely, let min∗ {X} = min {X} + a andmax∗ {X} = max {X} + b be, respectively, the minimum and maximumobtained outside the data range (say, for example, based on theo-retical minimums and maximums not present in the data), wherea and b are arbitrary constants and where min {X} and max {X} are,respectively, the minimum and maximum for the data. Further, letR∗ = max∗ {X} − min∗ {X} and R = max {X} − min {X} define the respec-tive ranges. From this, the fuzzy-set membership score s∗

i based on

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 175


min∗ {X} and max∗ {X} can be shown to be a simple function of si,

s∗i = xi − min∗ {X}

max∗ {X} − min∗ {X} = si

(RR∗

)− a

R∗

Thus, relative to si, the subset assessment using s∗i will be artifactually

influenced multiplicatively as a function of the ratio of the two rangesand additively as a function of the ratio of a to R∗.

In general, fuzzy-set analysis based on the Min/Max Uniform Distributioncoding provides results that are certain to have internal validity. That is,results hold for the data we use. External validity – moving off the supportof the data to draw inferences to, in our case, other countries and timepoints – depends on the degree to which our data (countries at specifictime points) represent other data not used in our analysis (other countriesat specific time points) with respect to the variables used in our analysis.

Additionally, assessment of the subset relationship based on ourgoodness-of-fit tests for any two factors – say for example high levelsof cumulative left incumbency and high levels of support for public daycare – is unaffected by consideration of a third factor – say, for example,aggregate education levels. This is because: (1) any one subset assess-ment is completely and solely defined on a nonparametric function ofthe observed membership scores si, (2) the observed membership scoresdefined above are solely a function of xi and independent of any informa-tion on some third factor, and (3) for the Eliason–Stryker goodness-of-fittest statistics there are no parameters estimated from the data that may, inturn, be influenced by the omission of some third factor. Thus, nowherecan a third factor contaminate the subset assessment. Moreover, therecan be no concern over biased estimation of parameters due to omit-ted variables simply because there are no parameters estimated in theEliason–Stryker goodness-of-fit statistics.

From a practical standpoint, therefore, results from our fuzzy-set anal-ysis may be extended to those country-periods not used in the analysis,insofar as these country-periods exhibit the same (or highly similar) dis-tributions on the observed factors used in the analysis. Second, movingoff the support of the data to infer results to country-periods not used inour analysis does not depend on unobserved factors. Other countries andtime periods we do not study will differ on a wide range of factors. Forinferences to be drawn reasonably to country-periods not in our analysis,such country-periods must resemble the country-periods in our analysisonly on the factors and outcomes used in our analysis.

We underscore these features of fuzzy-set analytic methods becausetheir scholarly and policy implications are so striking. On the one hand,

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 176


in contrast to inferences made from many econometric models, fuzzy-set analyses are not prone to specification bias from failure to includeall appropriate covariates. That we do not (yet) analyze the relationshipbetween, say, global or international factors and country-level familypolicies does not invalidate the results we do obtain for relationshipsamong the factors we do analyze. It means only that we can not (yet)speak about additional paths to over time or cross-country variation inspecific family policy outcomes that might involve such internationalfactors.

On the other hand, caution is required when scholars and policy mak-ers consider generalizing fuzzy-set analytic findings off the data set thatproduced the findings. When we replicate a fuzzy-set analysis on anexpanded data set that includes more countries and time points, wecan expect to see different results to the extent that the additional datachanges the empirical distribution on the factors and outcomes we haveused in prior analyses and in the replication on the more extensive dataset, especially with respect to the minimum and maximum anchoringfuzzy-set membership scores of 1 and 0 respectively.

Appendix 5.3: Goodness-of-Fit Tests for Fuzzy-Set Data

To assess fuzzy-set relations based on the subset principle articulated inRagin (2000), we use the goodness-of-fit strategy developed by Eliasonand Stryker (2007). Here we provide a brief description of this approach,and refer the reader to Eliason and Stryker (2007) for details. This strategyis based on comparing, in the context of fuzzy-set relations and the subsetprinciple (Ragin, 2000), the observed distance of cases from that expectedunder some causal hypothesis (necessity, sufficiency, and necessity andsufficiency) with the distance that would be expected given the truth ofthe causal hypothesis while accounting for measurement error.

To articulate this more precisely, define xi and yi as the fuzzy-setmembership scores for an hypothesized causal condition and outcome,respectively, for case i. To ensure distributional properties hold forthe goodness-of-fit tests, Eliason and Stryker (2007) work with stan-dardized normal scores, zx(i) = �−1{xi} and zy(i) = �−1{yi}, where �−1{·}is the inverse cumulative distribution function of the standard unitnormal distribution. Let zy(i) = zt

y(i) + εi and zx(i) = ztx(i) + ηi be equations

linking the observed membership scores to the standardized membershipscores measured without error, zt

y(i) and ztx(i), and errors in measure-

ment, εi and ηi.23 Finally, define observed and expected distances

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 177


D(nec & suf) =N∑

i=1(zy(i) − zx(i))2 and E{D(nec & suf)} =

N∑i=1

(εi − ηi)2. Eliason and

Stryker (2007) show that these distances are, respectively, the observeddistance of the data from, and the expected distance under the truth of,the causal necessity and sufficiency hypothesis.

With this, they then show that the ratio

F∗(nec & suf) = D(nec & suf)/N

E{D(nec & suf)|causal necessity & sufficiency}/N

will be distributed as an F random variable on (N, N) degrees of freedomif the causal necessity and sufficiency hypothesis is indeed true given thedata. Similar goodness-of-fit F ratios are derived for causal necessity andcausal sufficiency hypotheses separately. In addition, Eliason and Stryker(2007) show that partitioning of the F ratio for conjunctions providesinference tests for conjunctions as compared to components making upthe conjunction. A computer program to conduct the above tests andpartitions can be found at www.soc.umn.edu/∼eliason/.

Importantly, these goodness-of-fit F tests and partitions are appropri-ate regardless of whether the data constitute some sample (random ornot) from a population or the entire population. Additionally, thesegoodness-of-fit F tests and partitions do not require any assumptionsabout the functional relationship – beyond that expected under the sub-set principle – between fuzzy-set membership scores on the outcome andfuzzy-set membership scores on the hypothesized cause (or conjunctionof causes). No type of statistical model (linear or otherwise) is assumedto fit the fuzzy-set data, nor is any type of statistical model assumed tofit the empirical information on which the fuzzy-set scores are based.Thus, concerns over model specification, distributional assumptions (onerrors or outcomes), and undesirable statistical properties (bias, ineffi-ciency, etc.) in estimators (least squares, maximum likelihood, etc.) ofmodel parameters are not applicable to these goodness-of-fit tests. Thatis, none of the baggage that comes along with estimating a statisticalmodel is applicable to these goodness-of-fit tests.

Appendix 5.4: Estimating Compliers Average Causal Effectsusing Fuzzy-Set Membership Scores on Intention andTreatment Variables

In this appendix we detail our compliers-average-causal-effect (CACE)analysis using fuzzy-set membership scores. To formalize ideas, let πc be

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 178


the probability of compliance with the cumulative left cabinet incum-bency mechanism in enacting or maintaining a specific policy. Let πa

and πn be the probabilities of noncompliance with these mechanisms asgiven in assumptions 3a and 3b (described above) respectively. That is,πa gives the probability of always enacting a specific policy, and πn givesthe probability of never enacting a specific policy, regardless of the levelof cumulative left cabinet incumbency.

Next, define Y1it as the female labor force participation rate for countryi at time t under high levels of policy P’s implementation in country-time(i, t). Similarly, define Y2it as the female labor force participation rate forcountry i at time t under non-high levels of policy P’s implementationin country-time (i, t). Finally, define sx

it as the fuzzy score measuring thedegree to which country-time (i, t) belongs to the set “high level of policyP’s implementation” and sz

it as the fuzzy score measuring the degree towhich country-time (i, t) belongs to the set “high level of cumulative leftcabinet incumbency.”

From this, the likelihood function described in Imbens and Rubin(1997) and Hirano et al. (2000) adapted for use with fuzzy measures sx

itand sz

it , is given by

�CACE =I∏

i=1

T∏t=1

(πcfc1{Y1it } + πafa1{Y1it })sxit s

zit

×I∏

i=1

T∏t=1

(πcfc2{Y2i} + πnfn2{Y2i})(1−sxit )(1−sz

it )

×I∏

i=1

T∏t=1

(πafa2{Y1i})sxit (1−sz

it )

×I∏

i=1

T∏t=1

(πnfn1{Y2i})(1−sxit )s

zit (5A.1)

Here, fjk{·}, with j = c, a, n and k = 1, 2, refers to the density functions forfemale labor force participation rates for compliers, always-takers, andnever-takers for outcomes Y1it and Y2it . From this likelihood function,the CACE can be defined as

CACE = E{Y1it − Y2it |Compliance} =∫

yfc1{Y1}dy −∫

yfc2{Y2}dy (5A.2)

(See Imbens and Rubin’s (1997) or Hirano et al. (2000) for details.) Asdescribed above, this gives the expected change in female labor force

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 179


participation rates due to specific policies implemented through leftgovernments.

Importantly, for observations on Yit , within-country i serial correla-tion is likely across time points t and within-time t spatial correlationis likely across some countries i. However, for the CACE to be free ofbias due to these sources, we need only assume, for the population ofcompliers, that the correlation structures are the same for the errors onthe paired random variables (Y1it , Y2it ) prior to t. Given that the processgenerating pre-treatment serial and spatial correlation in the errors ofrandom variables Y1it and Y2it derives from the same process generatingserial and spatial correlations for those in Yit , this assumption appearsreasonable for these data. Moreover, no assumptions are necessary fornoncompliers.

To estimate the CACE, we use a bootstrapped estimator derived fromthe above likelihood on 1000 replications. Using bootstrapped estimatorsalleviates the need to impose any distributional assumptions, such asnormality, on the CACE itself. We thus present the median and inter-90percent percentile range from the bootstrapped empirical distributionfunction.

Appendix 5.5: Goodness-of-fit F Tests and Partitions onconjunctions, for outcome “High Female Labor ForceParticipation – Subsequent Year”

Hypothesized causal factors, and corresponding letter representations,include

A = High Level Prior Year Cumulative Left Cabinet IncumbencyC = High Level Public Sector EmploymentD = High Level Public Daycare, Ages 0–2E = High Level Public Daycare, Ages 3 to School AgeF = High Level Maternity LeaveG = High Level Family/Child Cash and Tax Benefits

----------------------------------------------------------------------HYPOTHESES

CONJUNCTIONS & TESTS SSD DF MSD F P----------------------------------------------------------------------

AC NECESSARY 431.7389 243 1.7767 6.9203 0.0000SUFFICIENT 5.5584 19 0.2925 1.1395 0.3115

NEC AND SUFF 437.2973 263 1.6627 6.4763 0.0000A v AC ----- ----- ----- 0.9279 0.7277C v AC ----- ----- ----- 0.4083 1.0000

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 180


AD NECESSARY 537.0686 248 2.1656 8.4350 0.0000SUFFICIENT 3.7620 1 3.7620 14.6530 0.0002

NEC AND SUFF 540.8306 263 2.0564 8.0097 0.0000A v AD ----- ----- ----- 0.7503 0.9899D v AD ----- ----- ----- 0.9375 0.6995

AE NECESSARY 457.2793 238 1.9213 7.4836 0.0000SUFFICIENT 7.4913 23 0.3257 1.2686 0.1885

NEC AND SUFF 464.7706 263 1.7672 6.8832 0.0000A v AE ----- ----- ----- 0.8731 0.8641E v AE ----- ----- ----- 0.8977 0.8089

AF NECESSARY 465.9570 237 1.9661 7.6578 0.0000SUFFICIENT 6.2466 18 0.3470 1.3517 0.1561

NEC AND SUFF 472.2036 263 1.7955 6.9933 0.0000A v AF ----- ----- ----- 0.8593 0.8902F v AF ----- ----- ----- 0.6768 0.9992

AG NECESSARY 605.5539 230 2.6328 10.2549 0.0000SUFFICIENT 4.6652 31 0.1505 0.5862 0.9624

NEC AND SUFF 610.2191 263 2.3202 9.0373 0.0000A v AG ----- ----- ----- 0.6650 0.9995G v AG ----- ----- ----- 0.6683 0.9994

CD NECESSARY 503.0120 242 2.0786 8.0960 0.0000SUFFICIENT 5.5812 3 1.8604 7.2463 0.0001

NEC AND SUFF 508.5932 263 1.9338 7.5322 0.0000C v CD ----- ----- ----- 0.3511 1.0000D v CD ----- ----- ----- 0.9969 0.5101

CE NECESSARY 436.7079 245 1.7825 6.9428 0.0000SUFFICIENT 7.2128 16 0.4508 1.7559 0.0373

NEC AND SUFF 443.9207 263 1.6879 6.5744 0.0000C v CE ----- ----- ----- 0.4022 1.0000E v CE ----- ----- ----- 0.9399 0.6923

CF NECESSARY 354.3491 240 1.4765 5.7508 0.0000SUFFICIENT 6.4369 14 0.4598 1.7908 0.0400

NEC AND SUFF 360.7860 263 1.3718 5.3432 0.0000C v CF ----- ----- ----- 0.4949 1.0000F v CF ----- ----- ----- 0.8858 0.8369

CG NECESSARY 422.3340 237 1.7820 6.9409 0.0000SUFFICIENT 4.1499 25 0.1660 0.6466 0.9036

NEC AND SUFF 426.4839 263 1.6216 6.3162 0.0000C v CG ----- ----- ----- 0.4186 1.0000G v CG ----- ----- ----- 0.9562 0.6417

DE NECESSARY 497.7304 246 2.0233 7.8807 0.0000SUFFICIENT 11.0621 3 3.6874 14.3623 0.0000

NEC AND SUFF 508.7924 263 1.9346 7.5352 0.0000D v DE ----- ----- ----- 0.9965 0.5113E v DE ----- ----- ----- 0.8200 0.9459

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 181


DF NECESSARY 531.1132 245 2.1678 8.4436 0.0000SUFFICIENT 6.6225 3 2.2075 8.5982 0.0000

NEC AND SUFF 537.7357 263 2.0446 7.9638 0.0000D v DF ----- ----- ----- 0.9429 0.6832F v DF ----- ----- ----- 0.5943 1.0000

DG NECESSARY 717.0342 248 2.8913 11.2615 0.0000SUFFICIENT 0.7181 1 0.7181 2.7971 0.0956

NEC AND SUFF 717.7523 263 2.7291 10.6298 0.0000D v DG ----- ----- ----- 0.7064 0.9975G v DG ----- ----- ----- 0.5682 1.0000

EF NECESSARY 443.1947 233 1.9021 7.4088 0.0000SUFFICIENT 17.8073 22 0.8094 3.1527 0.0000

NEC AND SUFF 461.0020 263 1.7529 6.8274 0.0000E v EF ----- ----- ----- 0.9050 0.7905F v EF ----- ----- ----- 0.6933 0.9985

EG NECESSARY 626.1874 232 2.6991 10.5129 0.0000SUFFICIENT 16.5087 29 0.5693 2.2173 0.0006

NEC AND SUFF 642.6960 263 2.4437 9.5183 0.0000E v EG ----- ----- ----- 0.6492 0.9998G v EG ----- ----- ----- 0.6345 0.9999

FG NECESSARY 543.1012 225 2.4138 9.4017 0.0000SUFFICIENT 14.8448 30 0.4948 1.9274 0.0036

NEC AND SUFF 557.9460 263 2.1215 8.2631 0.0000F v FG ----- ----- ----- 0.5728 1.0000G v FG ----- ----- ----- 0.7309 0.9944

ACD NECESSARY 539.9968 248 2.1774 8.4810 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 543.7588 263 2.0675 8.0530 0.0000AC v ACD ----- ----- ----- 0.8042 0.9611AD v ACD ----- ----- ----- 0.9946 0.5174CD v ACD ----- ----- ----- 0.9353 0.7059

ACE NECESSARY 486.2627 246 1.9767 7.6992 0.0000SUFFICIENT 5.4128 17 0.3184 1.2402 0.2330

NEC AND SUFF 491.6755 263 1.8695 7.2817 0.0000AC v ACE ----- ----- ----- 0.8894 0.8287AE v ACE ----- ----- ----- 0.9453 0.6758CE v ACE ----- ----- ----- 0.9029 0.7960

ACF NECESSARY 485.6026 245 1.9821 7.7201 0.0000SUFFICIENT 4.4649 18 0.2481 0.9662 0.4994

NEC AND SUFF 490.0675 263 1.8634 7.2579 0.0000AC v ACF ----- ----- ----- 0.8923 0.8219AF v ACF ----- ----- ----- 0.9635 0.6182CF v ACF ----- ----- ----- 0.7362 0.9934

ACG NECESSARY 636.6543 246 2.5880 10.0804 0.0000SUFFICIENT 2.3690 17 0.1394 0.5428 0.9295

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 182


NEC AND SUFF 639.0233 263 2.4297 9.4639 0.0000AC v ACG ----- ----- ----- 0.6843 0.9989AG v ACG ----- ----- ----- 0.9549 0.6457CG v ACG ----- ----- ----- 0.6674 0.9995

ADE NECESSARY 537.8769 248 2.1689 8.4477 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 541.6389 263 2.0595 8.0216 0.0000AD v ADE ----- ----- ----- 0.9985 0.5048AE v ADE ----- ----- ----- 0.8581 0.8924DE v ADE ----- ----- ----- 0.9394 0.6938

ADF NECESSARY 561.0879 248 2.2625 8.8123 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 564.8499 263 2.1477 8.3654 0.0000AD v ADF ----- ----- ----- 0.9575 0.6376AF v ADF ----- ----- ----- 0.8360 0.9265DF v ADF ----- ----- ----- 0.9520 0.6549

ADG NECESSARY 725.8544 248 2.9268 11.4000 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 726.5725 263 2.7626 10.7605 0.0000AD v ADG ----- ----- ----- 0.7444 0.9915AG v ADG ----- ----- ----- 0.8399 0.9212DG v ADG ----- ----- ----- 0.9879 0.5394

AEF NECESSARY 495.0917 240 2.0629 8.0349 0.0000SUFFICIENT 6.2060 23 0.2698 1.0510 0.4019

NEC AND SUFF 501.2977 263 1.9061 7.4242 0.0000AE v AEF ----- ----- ----- 0.9271 0.7300AF v AEF ----- ----- ----- 0.9420 0.6859EF v AEF ----- ----- ----- 0.9196 0.7513

AEG NECESSARY 651.7791 238 2.7386 10.6667 0.0000SUFFICIENT 4.4475 25 0.1779 0.6929 0.8629

NEC AND SUFF 656.2266 263 2.4952 9.7187 0.0000AE v AEG ----- ----- ----- 0.7082 0.9973AG v AEG ----- ----- ----- 0.9299 0.7220EG v AEG ----- ----- ----- 0.9794 0.5670

AFG NECESSARY 659.3546 238 2.7704 10.7907 0.0000SUFFICIENT 3.1846 25 0.1274 0.4962 0.9806

NEC AND SUFF 662.5393 263 2.5192 9.8121 0.0000AF v AFG ----- ----- ----- 0.7127 0.9969AG v AFG ----- ----- ----- 0.9210 0.7474FG v AFG ----- ----- ----- 0.8421 0.9179

CDE NECESSARY 504.8016 247 2.0437 7.9603 0.0000SUFFICIENT 5.5620 16 0.3476 1.3540 0.1649

NEC AND SUFF 510.3636 263 1.9405 7.5584 0.0000CD v CDE ----- ----- ----- 0.9965 0.5112CE v CDE ----- ----- ----- 0.8698 0.8706DE v CDE ----- ----- ----- 0.9969 0.5100

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 183


CDF NECESSARY 531.9591 246 2.1624 8.4227 0.0000SUFFICIENT 5.5812 17 0.3283 1.2788 0.2057

NEC AND SUFF 537.5403 263 2.0439 7.9609 0.0000CD v CDF ----- ----- ----- 0.9461 0.6731CF v CDF ----- ----- ----- 0.6712 0.9994DF v CDF ----- ----- ----- 1.0004 0.4988

CDG NECESSARY 720.0992 248 2.9036 11.3096 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 720.8173 263 2.7408 10.6752 0.0000CD v CDG ----- ----- ----- 0.7056 0.9976CG v CDG ----- ----- ----- 0.5917 1.0000DG v CDG ----- ----- ----- 0.9957 0.5138

CEF NECESSARY 469.1479 245 1.9149 7.4585 0.0000SUFFICIENT 6.2646 18 0.3480 1.3556 0.1539

NEC AND SUFF 475.4124 263 1.8077 7.0408 0.0000CE v CEF ----- ----- ----- 0.9338 0.7106CF v CEF ----- ----- ----- 0.7589 0.9872EF v CEF ----- ----- ----- 0.9697 0.5985

CEG NECESSARY 657.9967 246 2.6748 10.4183 0.0000SUFFICIENT 2.3690 17 0.1394 0.5428 0.9295

NEC AND SUFF 660.3656 263 2.5109 9.7800 0.0000CE v CEG ----- ----- ----- 0.6722 0.9993CG v CEG ----- ----- ----- 0.6458 0.9998EG v CEG ----- ----- ----- 0.9732 0.5870

CFG NECESSARY 574.8020 246 2.3366 9.1010 0.0000SUFFICIENT 1.4207 17 0.0836 0.3255 0.9953

NEC AND SUFF 576.2227 263 2.1910 8.5338 0.0000CF v CFG ----- ----- ----- 0.6261 0.9999CG v CFG ----- ----- ----- 0.7401 0.9925FG v CFG ----- ----- ----- 0.9683 0.6030

DEF NECESSARY 532.5317 247 2.1560 8.3976 0.0000SUFFICIENT 6.5671 16 0.4104 1.5987 0.0689

NEC AND SUFF 539.0988 263 2.0498 7.9840 0.0000DE v DEF ----- ----- ----- 0.9438 0.6803DF v DEF ----- ----- ----- 0.9975 0.5082EF v DEF ----- ----- ----- 0.8551 0.8974

DEG NECESSARY 717.0342 248 2.8913 11.2615 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 717.7523 263 2.7291 10.6298 0.0000DE v DEG ----- ----- ----- 0.7089 0.9973DG v DEG ----- ----- ----- 1.0000 0.5000EG v DEG ----- ----- ----- 0.8954 0.8145

DFG NECESSARY 737.7776 248 2.9749 11.5873 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 738.4957 263 2.8080 10.9371 0.0000DF v DFG ----- ----- ----- 0.7282 0.9948

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 184


DG v DFG ----- ----- ----- 0.9719 0.5913FG v DFG ----- ----- ----- 0.7555 0.9883

EFG NECESSARY 656.3900 234 2.8051 10.9258 0.0000SUFFICIENT 11.9584 29 0.4124 1.6061 0.0292

NEC AND SUFF 668.3483 263 2.5412 9.8982 0.0000EF v EFG ----- ----- ----- 0.6898 0.9987EG v EFG ----- ----- ----- 0.9616 0.6244FG v EFG ----- ----- ----- 0.8348 0.9281

ACDE NECESSARY 540.8051 248 2.1807 8.4937 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 544.5671 263 2.0706 8.0650 0.0000ACD v ACDE ----- ----- ----- 0.9985 0.5048ACE v ACDE ----- ----- ----- 0.9029 0.7960ADE v ACDE ----- ----- ----- 0.9946 0.5174CDE v ACDE ----- ----- ----- 0.9372 0.7004

ACDF NECESSARY 561.0879 248 2.2625 8.8123 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 564.8499 263 2.1477 8.3654 0.0000ACD v ACDF ----- ----- ----- 0.9627 0.6211ACF v ACDF ----- ----- ----- 0.8676 0.8749ADF v ACDF ----- ----- ----- 1.0000 0.5000CDF v ACDF ----- ----- ----- 0.9517 0.6559

ACDG NECESSARY 728.5613 248 2.9377 11.4425 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 729.2795 263 2.7729 10.8006 0.0000ACD v ACDG ----- ----- ----- 0.7456 0.9912ACG v ACDG ----- ----- ----- 0.8762 0.8577ADG v ACDG ----- ----- ----- 0.9963 0.5120CDG v ACDG ----- ----- ----- 0.9884 0.5377

ACEF NECESSARY 511.0231 246 2.0773 8.0912 0.0000SUFFICIENT 4.4646 17 0.2626 1.0229 0.4333

NEC AND SUFF 515.4877 263 1.9600 7.6343 0.0000ACE v ACEF ----- ----- ----- 0.9538 0.6492ACF v ACEF ----- ----- ----- 0.9507 0.6590AEF v ACEF ----- ----- ----- 0.9725 0.5894CEF v ACEF ----- ----- ----- 0.9223 0.7439

ACEG NECESSARY 677.1361 246 2.7526 10.7213 0.0000SUFFICIENT 2.3690 17 0.1394 0.5428 0.9295

NEC AND SUFF 679.5050 263 2.5837 10.0634 0.0000ACE v ACEG ----- ----- ----- 0.7236 0.9955ACG v ACEG ----- ----- ----- 0.9404 0.6906AEG v ACEG ----- ----- ----- 0.9657 0.6112CEG v ACEG ----- ----- ----- 0.9718 0.5915

ACFG NECESSARY 677.9242 246 2.7558 10.7338 0.0000SUFFICIENT 1.4207 17 0.0836 0.3255 0.9953

NEC AND SUFF 679.3450 263 2.5831 10.0610 0.0000ACF v ACFG ----- ----- ----- 0.7214 0.9959

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 185


ACG v ACFG ----- ----- ----- 0.9406 0.6899AFG v ACFG ----- ----- ----- 0.9753 0.5804CFG v ACFG ----- ----- ----- 0.8482 0.9087

ADEF NECESSARY 561.8962 248 2.2657 8.8250 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 565.6582 263 2.1508 8.3773 0.0000ADE v ADEF ----- ----- ----- 0.9575 0.6374ADF v ADEF ----- ----- ----- 0.9986 0.5046AEF v ADEF ----- ----- ----- 0.8862 0.8360DEF v ADEF ----- ----- ----- 0.9530 0.6516

ADEG NECESSARY 725.8544 248 2.9268 11.4000 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 726.5725 263 2.7626 10.7605 0.0000ADE v ADEG ----- ----- ----- 0.7455 0.9912ADG v ADEG ----- ----- ----- 1.0000 0.5000AEG v ADEG ----- ----- ----- 0.9032 0.7952DEG v ADEG ----- ----- ----- 0.9879 0.5394

ADFG NECESSARY 746.5978 248 3.0105 11.7258 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 747.3159 263 2.8415 11.0677 0.0000ADF v ADFG ----- ----- ----- 0.7558 0.9882ADG v ADFG ----- ----- ----- 0.9722 0.5902AFG v ADFG ----- ----- ----- 0.8866 0.8353DFG v ADFG ----- ----- ----- 0.9882 0.5383

AEFG NECESSARY 681.3751 240 2.8391 11.0582 0.0000SUFFICIENT 3.1621 23 0.1375 0.5355 0.9618

NEC AND SUFF 684.5372 263 2.6028 10.1379 0.0000AEF v AEFG ----- ----- ----- 0.7323 0.9941AEG v AEFG ----- ----- ----- 0.9586 0.6339AFG v AEFG ----- ----- ----- 0.9679 0.6043EFG v AEFG ----- ----- ----- 0.9764 0.5769

CDEF NECESSARY 533.3574 247 2.1593 8.4106 0.0000SUFFICIENT 5.5620 16 0.3476 1.3540 0.1649

NEC AND SUFF 538.9194 263 2.0491 7.9813 0.0000CDE v CDEF ----- ----- ----- 0.9470 0.6704CDF v CDEF ----- ----- ----- 0.9974 0.5083CEF v CDEF ----- ----- ----- 0.8822 0.8450DEF v CDEF ----- ----- ----- 1.0003 0.4989

CDEG NECESSARY 720.0992 248 2.9036 11.3096 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 720.8173 263 2.7408 10.6752 0.0000CDE v CDEG ----- ----- ----- 0.7080 0.9974CDG v CDEG ----- ----- ----- 1.0000 0.5000CEG v CDEG ----- ----- ----- 0.9161 0.7610DEG v CDEG ----- ----- ----- 0.9957 0.5138

CDFG NECESSARY 738.1357 248 2.9764 11.5929 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 186


NEC AND SUFF 738.8538 263 2.8093 10.9424 0.0000CDF v CDFG ----- ----- ----- 0.7275 0.9949CDG v CDFG ----- ----- ----- 0.9756 0.5793CFG v CDFG ----- ----- ----- 0.7799 0.9779DFG v CDFG ----- ----- ----- 0.9995 0.5016

CEFG NECESSARY 677.5220 246 2.7542 10.7274 0.0000SUFFICIENT 1.4207 17 0.0836 0.3255 0.9953

NEC AND SUFF 678.9427 263 2.5815 10.0551 0.0000CEF v CEFG ----- ----- ----- 0.7002 0.9980CEG v CEFG ----- ----- ----- 0.9726 0.5889CFG v CEFG ----- ----- ----- 0.8487 0.9079EFG v CEFG ----- ----- ----- 0.9844 0.5507

DEFG NECESSARY 737.7776 248 2.9749 11.5873 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 738.4957 263 2.8080 10.9371 0.0000DEF v DEFG ----- ----- ----- 0.7300 0.9945DEG v DEFG ----- ----- ----- 0.9719 0.5913DFG v DEFG ----- ----- ----- 1.0000 0.5000EFG v DEFG ----- ----- ----- 0.9050 0.7906

ACDEF NECESSARY 561.8962 248 2.2657 8.8250 0.0000SUFFICIENT 3.7620 15 0.2508 0.9769 0.4800

NEC AND SUFF 565.6582 263 2.1508 8.3773 0.0000ACDE v ACDEF ----- ----- ----- 0.9627 0.6209ACDF v ACDEF ----- ----- ----- 0.9986 0.5046ACEF v ACDEF ----- ----- ----- 0.9113 0.7741ADEF v ACDEF ----- ----- ----- 1.0000 0.5000CDEF v ACDEF ----- ----- ----- 0.9527 0.6526

ACDEG NECESSARY 728.5613 248 2.9377 11.4425 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 729.2795 263 2.7729 10.8006 0.0000ACDE v ACDEG ----- ----- ----- 0.7467 0.9909ACDG v ACDEG ----- ----- ----- 1.0000 0.5000ACEG v ACDEG ----- ----- ----- 0.9317 0.7165ADEG v ACDEG ----- ----- ----- 0.9963 0.5120CDEG v ACDEG ----- ----- ----- 0.9884 0.5377

ACDFG NECESSARY 746.5978 248 3.0105 11.7258 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 747.3159 263 2.8415 11.0677 0.0000ACDF v ACDFG ----- ----- ----- 0.7558 0.9882ACDG v ACDFG ----- ----- ----- 0.9759 0.5784ACFG v ACDFG ----- ----- ----- 0.9090 0.7800ADFG v ACDFG ----- ----- ----- 1.0000 0.5000CDFG v ACDFG ----- ----- ----- 0.9887 0.5368

ACEFG NECESSARY 696.6613 246 2.8320 11.0305 0.0000SUFFICIENT 1.4207 17 0.0836 0.3255 0.9953

NEC AND SUFF 698.0821 263 2.6543 10.3385 0.0000ACEF v ACEFG ----- ----- ----- 0.7384 0.9929ACEG v ACEFG ----- ----- ----- 0.9734 0.5865

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 187


ACFG v ACEFG ----- ----- ----- 0.9732 0.5872AEFG v ACEFG ----- ----- ----- 0.9806 0.5631CEFG v ACEFG ----- ----- ----- 0.9726 0.5891

ADEFG NECESSARY 746.5978 248 3.0105 11.7258 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 747.3159 263 2.8415 11.0677 0.0000ADEF v ADEFG ----- ----- ----- 0.7569 0.9879ADEG v ADEFG ----- ----- ----- 0.9722 0.5902ADFG v ADEFG ----- ----- ----- 1.0000 0.5000AEFG v ADEFG ----- ----- ----- 0.9160 0.7614DEFG v ADEFG ----- ----- ----- 0.9882 0.5383

CDEFG NECESSARY 738.1357 248 2.9764 11.5929 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 738.8538 263 2.8093 10.9424 0.0000CDEF v CDEFG ----- ----- ----- 0.7294 0.9946CDEG v CDEFG ----- ----- ----- 0.9756 0.5793CDFG v CDEFG ----- ----- ----- 1.0000 0.5000CEFG v CDEFG ----- ----- ----- 0.9189 0.7533DEFG v CDEFG ----- ----- ----- 0.9995 0.5016

ACDEFG NECESSARY 746.5978 248 3.0105 11.7258 0.0000SUFFICIENT 0.7181 15 0.0479 0.1865 0.9997

NEC AND SUFF 747.3159 263 2.8415 11.0677 0.0000ACDEF v ACDEFG ----- ----- ----- 0.7569 0.9879ACDEG v ACDEFG ----- ----- ----- 0.9759 0.5784ACDFG v ACDEFG ----- ----- ----- 1.0000 0.5000ACEFG v ACDEFG ----- ----- ----- 0.9341 0.7095ADEFG v ACDEFG ----- ----- ----- 1.0000 0.5000CDEFG v ACDEFG ----- ----- ----- 0.9887 0.5368

----------------------------------------------------------------------

Notes

1. See, for example, Siim 1988; Hernes 1987; Lewis 1992, 1998; Orloff 1993;Gauthier 1996; Gornick, Meyers and Ross 1997; Daly 1997; O’Connor, Orloffand Shaver 1999; Sainsbury 1999; Rubery, Smith and Fagan 1999; Crompton1999; Daly and Lewis 2000; Montanari 2000; Korpi 2000; Stier, Lewin-Epsteinand Braun 2001; Williams 2001; Brush 2002; Michel and Mahon 2002, Leira2002; Morgan and Zippel 2003; Gornick and Meyers 2003; Morgan 2004;Orloff 2004; Misra, Budig and Moller 2005.

2. See, for example, OECD 1990, 2001; European Commission Network onChildcare and Other Measures to Reconcile 1995; Kammerman and Kahn1991a, 1991b; Wennemo 1994; Ruhm 1995; Ruhm and Teague 1995;Gauthier 1996; Gornick, Meyers and Ross 1997; Leira 1992, 1998, 2002;Lewis 1998; Bruning and Plantenga 1999; Moss and Deven 1999; Rubery et al.1999, pp. 157–164; Roch 1999; Daly 2000; Daly and Lewis 2000; Montanari2000; Randall 2000; Kammerman 2000; Meyer 2000; Meyers and Gornick2000; Vlemicks and Smeeding 2001; Michel and Mahon 2002; Mahon 2002;

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 188


Sundstrom and Duvander 2002; Waldfogel 2002; Ferrarini 2003; Henderseonand White 2003; Morgan and Zippel 2003; Morgan 2004; Misra, Budig andMoller 2005.

3. See especially, Kammerman and Kahn 1991a; Winegarden and Bracy 1995;Ruhm and Teague 1995; Gornick, Meyers and Ross 1997; Korpi 2000; Brewsterand Rindfuss 2000; Stier, Lewin-Epstein and Braun 2001; Esping-Andersenet al. 2002; Waldfogel 2002; Ferrarini 2003; Mandel and Semyonov 2003;Morgan and Zippel 2003; Stryker and Eliason 2004; Misra, Budig and Moller2005; Pettit 2006. These scholars have used a variety of analytic techniques,each of which has strengths and weaknesses in assessing potential causallinks. The totality of prior work teaches us much about empirical associa-tions and potential causal links, but almost nothing about causal effects ofpolicies per se.

4. See Stryker and Eliason 2004, for a more complete elaboration of how andwhy our arguments are similar in some respects, yet different in others, fromthose of Huber and Stephens (2000, 2001). We discuss empirical support forboth points of view (see also Myles and Quadagno 2002).

5. Matching theory takes on different empirical specifications and leads tosomewhat different understandings of the relationship between gender segre-gated labor markets and the gender pay gap depending on how the genderingof tasks is conceived and measured. Bonstead-Bruns and Eliason (2002) fur-ther elaborate on Alice Eagly’s agentic versus communal characterization ofmale-typed versus female-typed tasks to investigate how gender of the personand the matching of both men and women to gender-typed jobs affects thegender pay gap in the United States. Theoretically framing their cross nationalanalyses of gender segregated labor markets, Charles and Grusky (2004)suggest two key processes underlying that segregation. The first is a “ver-tical” dimension privileging men over women, such that especially withinnon-manual occupations, men are matched to more prestigious, higherpaying jobs. The second is a “horizontal” dimension of gender essential-ism encouraging the matching of women to non-manual jobs. Where maleprivilege increases both segregation and gender inequality, gender essential-ism increases segregation while undermining gender inequality. Charles andGrusky’s (2004) approach provides leverage to investigate and explain theapparent paradox of greater gender segregation associated with lesser gender-pay gaps in Scandinavia (see also Rosenfeld and Kalleberg 1990). However, theequation of non-manual with female typed jobs is extremely crude. It doesnot capture gender stereotyped expectations that caring (versus non-caring)work is female work, that community oriented (versus self-actualizing) workis female work, that working with people (as opposed to working with things)is female work, or that coordinating (versus managing) is female work. Allthese distinctions have been important to the sex-typing of jobs (Steinberg1990; Bonstead-Bruns and Eliason 2002; England et al. 2002). In short, “gen-der essentialism” and the matching of women to female-typed jobs maybe entering Charles and Grusky’s (2004) vertical dimension as well as theirhorizontal dimension underlying gender segregated labor markets.

6. We were especially disappointed about having to drop Japan, Australia andNew Zealand from our analyses, because these countries provide at least someinsight into, on the one hand, a non-western political-cultural context and,

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 189


on the other hand, the two “wage-earner” welfare state regimes identified byHuber and Stephens (2001). However, after months of data collection efforts,we were convinced that this was the most reasonable decision.

7. Available empirical information on public expenditures for paid maternityleave, wage replacement rates, one-time lump sum payouts, and coveragestatistics tend to be related to maternity and/or parental leaves per se, andtypically not including extended leaves. Thus, we use only weeks of extendedleave to construct our fuzzy-set membership scores on extended leaves.

8. Ragin’s (1987, 2000) logic is similar to and draws on the constant-conjunctioncausal logic established by Mill (1967 [1843] in the indirect method ofdifference.

9. There is ongoing debate about the nature of fuzzy scores and fuzzy theory,and their relation to probabilities and probability theory. For informativebackground and discussion, see Puri and Ralescu (1985), Kandel, Martins,and Pacheco (1995), Laviolette, Seaman, Barrett, and Woodall (1995), andZadeh (1995).

10. We plan more nuanced analyses of the impact of family policies on both thenature and quantity of women’s labor force participation across countriesand over time. However, our current analyses do not distinguish betweenemployment in the public sector and the private sector, nor between part-time and full-time employment, nor between employment in higher wageversus lower wage jobs.

11. See Przeworski’s contribution to this volume for a similar methodologicalapproach to establishing causal effects on employment outcomes.

12. Note that assumptions 3a and 3b encapsulate in our case what is moregenerally called the monotonicity assumption. Also note that assumption5 is related to what is often called the ignorable assignment-to-treatmentassumption. Here, however, we are establishing the exogeneity of cumu-lated left cabinet incumbency up to time t compared to policies at time tand female labor force participation rates at time t + 1. So, this assumptionis better thought of as an exogenous assignment-to-treatment assumption,which is weaker than the ignorable assignment-to-treatment assumption. SeeImbens and Rubin (1997) or Hirano et al. (2000) for details.

13. See Efron and Tibshirani (1994) and DiCiccio and Efron (1996) for usefuldescription and discussion of bootstrapped estimators.

14. 14 For all tests we use a measurement error factor such that the maximummeasurement error of .05 is found at fuzzy-set membership scores of 0.5, withmeasurement error diminishing smoothly and symmetrically toward 0 at theboundary membership scores of 0 and 1. See Eliason and Stryker (2007) forrelevant details.

15. See Stryker and Eliason (2004) for details of the prior analysis.16. Eliason and Stryker (2007) show that the goodness-of-fit F statistics on the

fuzzy scores are unaffected by correlated data (either temporal or spatial)on the original measures used to construct fuzzy scores or on the fuzzy scoresthemselves. However, the degrees of freedom for the test may be affected tothe extent that the correlated structure is also present in the fuzzy scores.Thus, the degrees of freedom may require downward adjustments propor-tional to the order of the correlated structure in those scores. Given the num-ber of cases in our analysis, unless the correlated structure in the fuzzy scores

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 190


is beyond an order of three (which is very unlikely), the effect on the assess-ment of fit will be negligible. This, combined with the fact that we achievedvery good fit levels in our analyses (with p values often greater than 0.05,suggests that the results from our fuzzy-set analysis are not likely affected bythe longitudinal and spatially connected nature of these aggregate data.

17. To better understand this, see Hirano et al. (2000) and Imbens and Rubin(1997). Also see Heckman et al. (1999) for a valuable more general discussionof related causal estimators and estimands.

18. Given this large CACE effect on the additive percentage point change infemale labor force participation rates, we also estimated multiplicative CACE’sgiving the percentage change in those rates. The multiplicative CACE resultsreveal similarly large effects. Thus, we continue to present the additivepercentage point change effects throughout.

19. This finding may appear, at first glance, inconsistent with results from thefuzzy-set analysis. Recall, however, that the fuzzy-set analysis examinedwhether high levels of family/child cash and tax benefits were necessaryand/or sufficient for high levels of female labor force participation. Finding,as we did, that high levels of family/child cash and tax benefits were neithernecessary nor sufficient for high levels of female labor force participation isentirely consistent with the CACE analysis findings presented here. It does,however, highlight that the fuzzy-set analysis was focused on those factorsthat may have an impact on high, and not low, levels of female labor forceparticipation.

20. In the language of the classic intention-to-treat design, cumulative left gov-ernance acts as the intention-to-treat instrument for the results discussed intable 12, while specific policies act as the treatments. Similarly, other cumu-lative governance patterns act as the intention-to-treat instrument for theresults discussed in Table 5.13, with specific policies again acting as treat-ments. In both cases it is the specific policy in question, and not cumulativegovernance, that acts as the treatment, and thus carries causal status, ininfluencing female labor force participation.

21. See also Verkuilen (2005) for a fairly comprehensive and useful discussion ofmeasurement strategies for fuzzy-set methods.

22. With minor modifications, this logic and coding extends to an ordinaldistribution on the original information.

23. It is important to note that, while these are equations, they do not have anyparameters to be estimated from the data. Thus, issues of bias, inconsistency,inefficiency, and the like, are not applicable here.

References

Blau, Francine D. and Mariane A. Ferber. 1992. The Economics of Women, Men andWork. Englewood Cliffs, NJ: Prentice Hall.

Bonstead-Bruns, Melissa and Scott Eliason. 2002. “Reconsidering the Role of Gen-der in Labor Market Attainments and Evaluating the Wage Gap.” Unpublishedmanuscript.

Brewster, K. L. and R. R. Rindfuss. 2000. “Fertility and Women’s Employment inIndustrialized Nations.” Annual Review of Sociology 26: 271–96.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 191


Brush, Lisa D. 2002. “Changing the Subject: Gender and Welfare Regime Studies.”Social Politics 9: 161–86.

Bruning, Gwennaële and Plantenga. 1999. “Parental Leave and Equal Opportuni-ties Experiences in Eight European Countries.” Journal of European Social Policy9: 195–209.

Castles, Francis G. 2001. “On the Political Economy of Recent Public SectorDevelopment.” Journal of European Social Policy 11: 195–211.

Charles, Maria and David B. Grusky. 2004. Occupational Ghettos: The WorldwideSegregation of Women and Men. Stanford: Stanford University Press.

Clayton, Richard and Jonas Pontusson. 1998. “Welfare State Retrenchment Revis-ited: Entitlement Cuts, Public Sector Restructuring and Inegalitarian Trends inAdvanced Capitalist Societies.” World Politics 51: 567–98.

Crompton, Rosemary, editor. 1999. Restructuring Gender Relations and Employment:The Decline of the Male Breadwinner. Oxford: Oxford University Press.

Daly, Mary. 1997. The Gender Division of Welfare: The Impact of the British andGerman Welfare States. Cambridge: Cambridge University Press.

Daly, Mary. 2000. “A Fine Balance: Women’s Labor Market Participation inInternational Comparison.” In Welfare and Work in the Open Economy: DiverseResponses to Common Challenges, volume 2, edited by F. Scharpf and V. Schmidt.Oxford: Oxford University Press.

Daly, Mary and Jane Lewis. 2000. “The Concept of Social Care and the Analysisof Contemporary Welfare States.” British Journal of Sociology 51: 281–98.

Diciccio, Thomas J. and Bradley Efron. 1996. “Bootstrap Confidence Intervals.”Statistical Science 11: 189–212.

Efron, Bradley and Robert J Tibshirani. 1994. An Introduction to the Bootstrap. BocaRaton: CRC Press.

Eliason, Scott R. and Robin Stryker. 2007. “Goodness-of-Fit Tests and Descrip-tive Measures in Fuzzy-Set Analysis.” Unpublished manuscript. Available atwww.soc.umn.edu/∼eliason.

England, Paula, Michelle Budig and Nancy Folbre. 2002. “Wages of Virtue: TheRelative Pay of Care Work.” Social Problems 49: 455–73.

Esping-Andersen, Gøsta. 1990. The Three Worlds of Welfare Capitalism. Princeton:Princeton University Press.

Esping-Andersen, Gøsta, John Myles, Anton Hemerijck and Duncan Gallie. 2002.Why We Need a New Welfare State. Oxford: Oxford University Press.

Ferrarini, Tommy. 2003. “Paid Parental Leave and Poverty: Analyzing thePolicy Outcome Link Among Families with Infants in Western Countries.”Paper presented at RC 19 Annual Meetings, Toronto, August 21–4.

Gauthier, Anne. 1996. The State and the Family: A Comparative Analysis of FamilyPolicies in Industrialized Countries. Oxford: Clarendon Press.

Gauthier, Anne. 2000. “Public Policies Affecting Fertility and Families in Europe:A Survey of the 15 Member States.” Paper prepared for the European Observatoryon Family Matters: Annual Seminar, 2000: “Low Fertility, Families and PublicPolicies,”’ Sevilla, Spain, September 15–16.

Goertz, Gary and James Mahoney. 2005. “Two-Level Theories and Fuzzy-SetAnalysis.” Sociological Methods and Research 33: 497–538.

Gornick, Janet, Marcia Meyers and Katharin Ross. 1997. “Supporting the Employ-ment of Mothers: Policy Variation Across Fourteen Welfare States.” Journal ofEuropean Social Policy 7: 45–70.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 192


Gornick, Janet and Jerry Jacobs. 1998. “Gender, the Welfare State and Pub-lic Employment: A Comparative Study of Seven Industrialized Countries.”American Sociological Review 63: 688–710.

Gornick, Janet and Marcia K. Meyers. 2003. Families That Work: Policies forReconciling Parenthood and Employment. New York: Russell Sage.

Grusky, David B. and Maria Charles. 2001. “Is There a Worldwide Sex SegregationRegime?” Pp. 689–703 in Social Stratification in Comparative Perspective: Class,Race and Gender, edited by D. Grusky. Second edition. Boulder, CO: Westport.

Heckman, James J., Robert J. LaLonde, and Jeffrey A. Smith. 1999. “The Economicsand Econometrics of Active Labor Market Programs.” In Handbook of Labor Eco-nomics, volume 3, edited by Orley Ashenfelter and David Card. Amsterdam:Elsevier.

Henderson, Ailsa and Linda A. White. 2003. “Shrinking Welfare States? Com-paring Maternity Leave Benefits and Child Care Programs in European Unionand North American Welfare States, 1985–2000.” Paper presented at RC 119Meetings, Toronto, August 21–4.

Hernes, Helga. 1987. The Welfare State and Women Power. Oslo: Univeritetsforlaget.Hirano, Keisuke, Guido W. Imbens, Donald B. Rubin, and Xiao-Hua Zhou. 2000

“Assessing The Effect Of An Influenza Vaccine In An Encouragement Design.”Biostatistics 1: 69–88.

Huber, Evelyne and John Stephens. 2000. “Partisan Governance, Women’sEmployment and the Social Democratic Service State.” American SociologicalReview 45: 323–42.

Huber, Evelyne and John Stephens. 2001. Development and Crisis of the WelfareState: Parties and Policies in Global Markets. Chicago: University of Chicago Press.

Huber, Evelyne, Charles Ragin, and John Stephens. 1993. “Social Democracy,Christian Democracy, Constitutional Structure and the Welfare State.” AmericanJournal of Sociology 99: 711–49.

Imbens, Guido W. and Donald B. Rubin. 1997. “Bayesian Inference for CausalEffects in Randomized Experiments with Noncompliance.” The Annals ofStatistics 25: 305–27.

Jenson, Jane. 1990. “Representations of Gender Politics to ‘Protect’ Women: Work-ers and Infants in France and the United States Before 1914.” Pp. 152–77 inWomen, the State and Welfare, edited by Linda Gordon: Madison: University ofWisconsin Press.

Kammerman, Sheila. 2000. “Early Childhood Education and Care: An Overviewof Developments in the OECD Countries.” International Journal of EducationalResearch 33: 7–29.

Kammerman, Sheila and Alfred J. Kahn. 1991a. Child Care, Parental Leave and theUnder 3s: Policy Innovation in Europe. New York: Auburn House.

Kammerman, Sheila and Alfred J. Kahn. 1991b. “Government Expenditures forChildren and Their Families in Advanced Industrialized Countries, 1960–1985.”Innocenti Occasional Papers: Economic. Policy Series No. 20.

Kandel, Abraham, Alejandro Martins, and Roberto Pacheco. 1995. “On the VeryReal Distinction between Fuzzy and Statistical Methods.” Technometrics 37:276–81.

Korpi, Walter. 2000. “Faces of Inequality: Gender, Class and patterns ofInequalities in Different Types of Welfare States.” Social Politics Summer:127–91.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 193


Kreyenfeld, Michael and Karsten Hank. 2000. “Does the Availability of Child-care Influence the Employment of Mothers? Findings from Western Germany.”Population Research and Policy Review 19: 317–37.

Laviolette, Michael, John W. Seaman, Jr., J. Douglas Barrett, and WilliamH. Woodall. 1995. “A Probabilistic and Statistical View of Fuzzy Methods.”Technometrics 37: 249–61.

Leira, Arnlaug. 1992. Welfare States and Working Mothers: The ScandinavianExperience. New York: Cambridge University Press.

Leira, Arnlaug. 1998. “Caring as a Social Right: Cash for Child Care and DaddyLeave.” Social Politics 5: 362–79.

Leira, Arnlaug. 2002. Working Parents and the Welfare State: Family Change andPolicy Reform in Scandinavia. Cambridge: Cambridge University Press.

Lewis, Jane. 1992. “Gender and the Development of Welfare State Regimes.”Journal of European Social Policy 2: 159–73.

Lewis, Jane, ed. 1998. Gender, Social Care and Welfare State Restructuring in Europe.Aldershot: Ashgate.

Mahon, Rianne. 2002. “Child Care: Toward What Kind of ‘Social Europe’?” SocialPolitics 9: 343–79.

Mahoney, James. 2003. “Long-Run Development and the Legacy of Colonialismin Spanish America.” American Journal of Sociology 109: 50–106.

Mandel, Hadas. Moshe Semyonov. 2003. “The Prevalence of Welfare State Policiesand Gender Socio-Economic Inequality: A Comparative Analysis.” Paper pre-sented at Meeting of the ISA Research Committee on Stratification and Mobility(March, 2003), Tokyo.

Meyer, Madonna Harrington, ed. 2000. Care Work, Gender, Labor and the WelfareState. New York: Routledge.

Meyers, Marie K. and Janet C. Gornick. 2000. “Early Childhood Education andCare (ECEC: Cross National Variation in Services Organization and Financing).”Paper presented at the Consultative Meeting on International Developments inEarly Childhood Education and Care. New York, May 11–12.

Michel, Sonya and Rianne Mahon, eds. 2002. Child Care Policy at the Crossroads:Gender and Welfare State Restructuring. New York: Routledge.

Mill, John Stuart. 1967 [1843]. A System of Logic: Ratiocinative and Inductive.Toronto: University of Toronto Press.

Misra, Joya. 2003. “Women as Agents in Welfare State Development: A Cross-National Analysis of Family Allowance Adoption.” Socio-Economic Review 1:185–214.

Misra, Joya, Michelle Budig, and Stephanie Moller. 2005. “Employment, Wagesand Poverty: Family Policies and Gender Equity.” Paper presented at AnnualMeetings of Research Committee 19, Sociological Association, Washington, DC,September 8–10.

Montanari, Ingalill. 2000. “From Family Wage to Marriage Subsidy and ChildBenefits: Controversy and Consensus in the Development of Family Support.”Journal of European Social Policy 10: 307–33.

Morgan, Kimberly J. and Kathrin Zippel. 2003. “Paid to Care: The Origins andEffects of Care Leave Policies in Western Europe.” Social Politics 10: 49–85.

Morgan, Kimberly J. 2004. “Caring in Western Europe: Trends and Implications.”Paper presented at annual meetings of Research Committee 19, Social Welfareand Social Policy, Paris, September.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 194


Moss, Peter and Fred. F Deven, eds. 1999. Parental Leave: Progress or Pitfall? TheHague/Brussels: NIDI:CBGS Publications.

Myles, John and Jill Quadagno. 2002. “Political Theories of the Welfare State.”Social Service Review March: 34–66.

Myles, John anmd Adrian Turegin. 1994. “Comparative Studies in Class Struc-ture.” Annual Review of Sociology 20: 103–24.

O’Connor, Julia. 2003. “Measuring Progress in the European Social Model: PolicyCoordination, Social Indicators and the Social Policy Agency in the EuropeanUnion.” Paper presented at RC-19 Annual Meetings, Toronto. August.

O’Connor, Julia, Ann Shola Orloff and Sheila Shaver. 1999. States, Markets, Fam-ilies: Gender, Liberalism, and Social Policy in Australia, Canada, Great Britain, andthe United States. Cambridge: Cambridge University Press.

OECD. 1982. Employment in the Public Sector. Paris, OECD.OECD. 1990. “Child Care in OECD Countries.” Employment Outlook 1990.OECD. 2001. “Starting Strong: Early Childhood Education and Care.” Paris,

OECD.Orloff, Ann Shola. 1993. “Gender and the Social Rights of Citizenship: The Com-

parative Analysis of Gender Relations and Welfare States.” American SociologicalReview 58: 303–28.

Orloff, Ann Shola. 2004. “Farewell to Maternalism (and Hello to Employment forAll?): The State and Employed Women.” Paper presented at annual meetings ofResearch Committee 19, Social Welfare and Social Policy, Paris, September.

Pettit, Elizabeth M. 2006. Presentation at Department of Sociology, University ofMinnesota, February.

Puri, Madan L. and Dan A. Ralescu. 1985. “The Concept of Normality for FuzzyRandom Variables.” The Annals of Probability, 13: 1373–9.

Ragin, Charles. 1987. The Comparative Method. Berkeley: University of CaliforniaPress.

Ragin, Charles. 2000. Fuzzy-Set Social Science. Chicago and London: University ofChicago Press.

Ragin, Charles and Paul Pennings. 2005. “Fuzzy Sets and Social Research.”Sociological Methods and Research 33: 423–30.

Randall, Vicky. 2000. “Childcare Policy in the European States: Limits toConvergence.” Journal of European Public Policy 7: 346–68.

Reskin, Barbara. 1993. “Sex Segregation in the Workplace.” Annual Review ofSociology 19: 241–70.

Ridgeway, Cecilia. 1997. “Interaction and the Conservation of Gender Inequality:Considering Employment.” American Sociological Review 62: 218–35.

Ridgeway, Cecilia and Lynne Smith-Lovin. 1999. “The Gender System andInteraction.” Annual Review of Sociology 25: 191–216.

Roch, Maria Teresa Soler, ed. 1999. Family Taxation in Europe. London: Kluwer LawInternational (Academic Committee on European Tax Law).

Rønsen, Marit and Marianne Sundstro′′m. 2002. “Family Policy and After-BirthEmployment Among New Mothers: A Comparison of Finland, Norway andSweden.” European Journal of Population 18: 121–52.

Rosenfeld, Rachel A. and Arne L. Kalleberg. 1990. “A Cross-National Comparisonof the Gender Gap in Income.” American Journal of Sociology 96: 69–106.

Rubery, Jill, Mark Smith, and Colette Fagan. 1999. Women’s Employment in Europe:Trends and Prospects. London and New York: Routledge.

9780230_202573_06_cha05.tex 5/6/2008 15: 37 Page 195


Ruhm, Christoper. 1995. “Parental Leave Policies in Europe and North America.”Working Paper 5065. Cambridge, MA: National Bureau of Economic Research.

Ruhm, C. J. and J. L. Teague. 1995. “Parental Leave Policies in Europe and NorthAmerica.” Working Paper 5065. National Bureau of Economic Research.

Sainsbury, Diane. 1999. Gender and Welfare State Regimes, Oxford: OxfordUniversity Press.

Scharpf, Fritz. 1999. Governing in Europe: Effective and Democratic? Oxford: OxfordUniversity Press.

Scharpf, Fritz and Vivien A. Schmidt, eds. 2000. Welfare and Work in the OpenEconomy. Oxford: Oxford University Press.

Siim, Berte. 1988. “Towards a Feminist Rethinking of the Welfare State.” In ThePolitical Interests of Gender, edited by K. B. Jones and A. G. Jonasdottir, London:Sage.

Smithson, Michael. 2005. “Fuzzy Set Inclusion: Linking Fuzzy-Set Methods withMainstream Techniques.” Sociological Methods and Research 33: 431–61.

Steinberg, Ronnie J. 1990. “The Social Construction of Skill: Gender, Power andComparable Worth.” Work and Occupations 17: 449–82.

Stier, Haya, Noah Lewis-Epstein, and Michael Braun. 2001. “Welfare Regimes,Family-Supportive Policies and Women’s Employment Along the Life-Course.”American Journal of Sociology 106: 1731–60.

Streeck, Wolfgang. 2001. “High Equality, Low Activity: The Contribution of theSocial Welfare System to the Stability of the German Collective BargainingRegime.” Industrial and Labor Relations Review 54: 698–706.

Stryker, Robin and Scott Eliason. 2003. “The Welfare State, Gendered LaborMarkets and Aggregate Political Orientations in France, Belgium, Germany,Italy, Denmark and Britain, 1977–1994.” European University Institute. RobertSchuman Centre Advanced Studies Working Paper [RSC2003/20].

Stryker, Robin and Scott Eliason. 2004. “The Welfare State, Gendered LaborMarkets and Political Orientations in France, Belgium, Germany, Italy, Den-mark and Britain, 1977–1994.” Paper circulated at Annual Meetings of RC-19,Toronto, August 21–4.

Sundstrom, Marie and Ann-Zofie Duvander. 2002. “Gender Division of Childcareand the Sharing of Parental Leave Among New Parents in Sweden.” EuropeanSociological Review 18: 433–47.

Verkuilen, Jay. 2005. “Assessing Membership in Fuzzy Set Analysis.” SociologicalMethods and Research 33: 462–96.

Vleminks, K. and Timothy Smeeding, eds. 2001. Child Wellbeing, Child Poverty,and Child Policy in Modern Nations: What Do We Know? Bristol: Policy Press.

Waldfogel, Jane 2002. “Child Care, Women’s Employment, and Child Outcomes.”Journal of Population Economics 15: 527–48.

Wennemo, Irene. 1994. Sharing the Cost of Children. Doctoral Dissertation Series#25. Swedish Institute for Social Research. Stockholm: Stockholm University.

Williams, Joan. 2001. Unbending Gender: Why Family and Work Conflict and Whatto Do About It. Oxford: Oxford University Press.

Winegarden, C. R. and Paula M. Bracy. 1995. “Demographic Consequences ofMaternal-Leave Programs in Industrial Countries: Evidence from Fixed-EffectsModels.” Southern Economic Journal 6: 1020–35.

Zadeh, Lotfi. 1995. “Probability Theory and Fuzzy Logic are ComplimentaryRather than Competitive.” Technometrics 37: 271–6.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 196

6Family Policies and Women’sEmployment: A RegressionAnalysisAlexander Hicks and Lane Kenworthy

There is good reason to suspect that “family-friendly” or “women-friendly” policies such the government provision or subsidization ofchild care, paid maternity leave, and extensive public employment willincrease levels of female employment. There is supportive evidencefrom studies of individual behavior (Gustafsson and Stafford, 1992;Leibowitz, Klerman, and Waite, 1992; Barrow, 1996; Ondrich, Spiess, andYang, 1996; Ilmakunnas, 1997; Joesch, 1997; Fagnani, 1998; Kimmel,1998; Powell, 1998; Anderson and Levine, 1999; Ondrich et al., 1999;Michalopoulos and Robins, 2000; Smith, Downs, and O’Connell, 2001;Chevalier and Viitanen, 2002; Del Boca, 2002; Pylkkänen and Smith,2003; Rønsen and Sundstrom, 2002; Gottschall and Bird, 2003; Hofferthand Curtin, 2003). But at the macro (country) level, the association hasbeen largely assumed rather than demonstrated.

This assumption is based principally on cross-country differencesbetween affluent nations. Most notable is the fact that the Nordic coun-tries have been at the forefront in introducing and expanding these typesof policies and are also the countries with the highest rates of femaleemployment. But this apparent cross-sectional association has been moreoften the subject of casual observation than of careful analysis. And sel-dom have researchers examined the relationship between these policiesand over-time changes within countries.

In their chapter in this volume, Scott Eliason, Robin Stryker, and ErikTranby (2008) attempt to do just this. They create new measures of theextensiveness of these types of family policy, and they use qualitativecomparative analysis (QCA) to examine the impact of such policies onthe extent of women’s labor force participation in 14 OECD countriessince 1960. In this chapter we use regression analysis to explore this issue.We examine the same set of countries during the same time periods, and

196

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 197

Alexander Hicks and Lane Kenworthy 197

we use Eliason, Stryker, and Tranby’s family policy measures. The coun-tries are Austria, Belgium, Canada, Denmark, Finland, France, Germany,Ireland, Italy, the Netherlands, Norway, Sweden, the United Kingdom,and the United States. The time periods are decades: the 1960s, 1970s,1980s, and 1990s. Our findings are in some respects complimentary toand in other respects inconsistent with those of Eliason, Stryker, andTranby.

Eliason, Stryker, and Tranby use women’s labor force participationrates as their dependent variable. For consistency with the other chaptersin this volume, we instead examine women’s employment rates. This isunlikely to affect the findings, as the two correlate very strongly acrosscountries and over time.

Patterns of female employment

For those seeking to understand differences among affluent coun-tries in employment performance, getting a handle on gender-specificemployment patterns is important. Figure 6.1 shows employment rates(employed as a share of the working-age population) for men and forwomen as of the period 2000–05. We organize the countries into threegroups familiar to macrocomparative researchers: Nordic, continental,and Anglo. These groups have been found to differentiate sets of casesthat vary in socioeconomic characteristics and processes of policy deter-mination but are relatively homogeneous internally (Esping-Andersen,1990, 1999; Goodin et al., 1999; Huber and Stephens, 2001). However,we use the grouping simply for heuristic purposes; no causal importanceof group membership is implied. Much of the cross-country variation inFigure 6.1 is in female employment. The coefficient of variation (stan-dard deviation divided by the mean) for women’s employment is .14,compared to .07 for men’s employment.

Figure 6.2 shows trends in women’s employment over time in the 14countries. Several things are worth noting. First, there is considerablecross-country variation in levels of women’s employment (already appar-ent in Figure 6.1). The Nordic countries have tended to have the highestfemale employment rates, followed by the Anglo countries, with the con-tinental countries lagging behind. Secondly, women’s employment hasincreased in all of the countries. Thirdly, the countries vary markedlyin degree and timing of this over-time increase. The Nordic countriestended to experience growth in the 1960s, 1970s, and 1980s, but thenstagnation or decline in the 1990s. The Anglo countries experiencedsteady growth throughout the four decades. Three of the continental

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 198


0 20 40 60 80 100

Share of male or female population age 15–64 (%)

IreUK

CanUS

ItBelFr

GerAusNth

FinDenSweNor

Men’s employment rateWomen’s employment rate

Figure 6.1 Men’s and women’s employment, 2000–2005Note: For data definitions and sources, see the chapter appendix.

countries – France, Germany, and Italy – experienced no significantchange in women’s employment rates in the 1960s, 1970s, and 1980s,but then some increase in the 1990s. Another continental country, theNetherlands, experienced by far the most dramatic change, with anincrease in the female employment rate of nearly 30 percentage pointssince the mid-1980s.

What role have family policies played in generating this cross-countryand over-time variation in female employment? We follow Eliason,Stryker, and Tranby in focusing on the impact of three types of pol-icy: public child care (separated into two age groups: 0 to 2 and 3 to 5),maternity leave, and public employment. They also examine the impactof child allowances/benefits, but we do not because there is no reasonto expect this type of policy to increase women’s employment (if any-thing, the reverse is true, as a child allowance provides income that isnot conditional on prior or current employment).

Family policies as causally sufficient?

Eliason, Stryker, and Tranby’s conclusion with respect to the impact offamily policies is that “High levels of public sector expansion, public day-care for younger children, and maternity leave are, separately, causallysufficient for high levels of female labor force participation” (p. 163).These inferences are based on the patterns shown in Figure 6.3.

9780230_202573_07_cha06.tex

7/6/200816:32

Page199

199

0

20

40

60

80

Wom

en’s

em

ploy

men

t rat

e (%

)

0

20

40

60

80

Wom

en’s

em

ploy

men

t rat

e (%

)0

20

40

60

80

Wom

en’s

em

ploy

men

t rat

e (%

)

1960 1970 1980 1990 2000

Year

Nordic Countries

1960 1970 1980 1990 2000

Year

Continental Countries

1960 1970 1980 1990 2000

Year

Anglo Countries

Den FinNor Swe

AusFrIt

BelGerNth

CanUK

IreUS

Figure 6.2 Women’s employment rates, 1960–2000Note: Missing data: 1960–67 for Austria; 1960–75 for Belgium; 1960–65 for Canada; 1961–64, 1996, 1968 for Denmark; 1960–62 for Finland; 1960–65for Ireland; 1972–76 for Italy; 1960–74 for the Netherlands; 1960–71 for Norway; 1960–62 for Sweden; For data definitions and sources, see the chapterappendix.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 200


Aus80s

Aus90s

Bel80s

Bel90s

Can70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr80sFr90sGer80s

Ger90s

Ire80s

Ire90s

It80sNth80s

Nth90s

Nor60s

Nor90s

Swe80s

Swe90s

UK80s

UK90s

US60s

US70s

US80s

US90s

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)0 .2 .4 .6 .8 1.0

Public child care, age 0–2 (index)

Public child care, age 0–2

Aus80s

Aus90s

Bel80s

Bel90s

Can70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr80sFr90sGer80s

Ger90s

Ire80s

Ire90s

It80sNth80s

Nth90s

Nor60s

Nor90s

Swe80s

Swe90s

UK80s

UK90s

US60s

US70s

US80s

US90s

0 .2 .4 .6 .8 1.0 1.2 1.4


Public child care, age 3–5

Aus70sAus80s

Aus90s

Bel80s

Bel90s

Can70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr70sFr80sFr90sGer60sGer70sGer80s

Ger90s

Ire70sIre80s

Ire90s

It60sIt80s

It90sNth80s

Nth90s

Nor60s

Nor90s

Swe70s

Swe80s

Swe90s

UK60s

UK70sUK80s

US60s

US70s

US80s

US90s

0 .2 .4 .6 .8 1.0

Maternity leave (index)

Maternity leave

Aus70sAus80s

Aus90s

Bel80s

Bel90s

Can70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr70sFr80s

Fr90sGer60sGer70sGer80s

Ger90s

Ire70sIre80s

Ire90s

It60sIt80s

It90sNth80s

Nth90s

Nor60s

Nor90s

Swe70s

Swe80s

Swe90s

UK60s

UK70sUK80s

UK90s

US60s

US70s

US80s

US90s

5 10 15 20 25

Public employment (%)

Public employment

Figure 6.3 Women’s employment by family policies, decade averages,1960s–1990sNote: For data definitions and sources, see the chapter appendix.

The figure includes four scatterplots. Each chart has women’s employ-ment rates on the vertical axis and one of the four family policy measureson the horizontal axis. The data points are decade averages for each of thetwelve countries (so each country appears up to four times, dependingon missing data).

The patterns in the charts in Figure 6.3 are consistent with an infer-ence of “causal sufficiency.” Causal sufficiency means that “if X, thenY.” Here this can be read as “if family policy is generous, then the femaleemployment rate will be high.” In Figure 6.3, in all observed instances

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 201


Fin

Ger

Nor

Swe

US

55

60

65

70

75

Wom

en’s

em

ploy

men

t rat

e (%

), 2

000

15 40 65

Share of women age 25–59 strongly agreeingthat both husband and wife should contribute

to household income (%), 1995–97

Women’s preference for employment

Asl70s

Asl80s

Asl90s

Aus70sAus80s

Aus90s

Bel80s

Bel90sCan70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr70sFr80s


Ger90s

Ire70sIre80s

Ire90s

It60sIt80s

It90s Nth80s

Nth90s

Nor60s

Nor90s

Swe70s

Swe80s

Swe90s

UK60s

UK70sUK80s

UK90s

US60s

US70s

US80s

US90s

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)4 6 8 10 12

Average years of schooling completed

Women’s education

Figure 6.4 Women’s employment by women’s preference for employment andwomen’s educationNote: For data definitions and sources, see the chapter appendix.

of generous family policy (to the right on the charts’ horizontal axes),women’s employment is indeed high.

But are these patterns enough to warrant a reading of causal suffi-ciency? No. As is almost always the case in analyses of non-experimentaldata, there is reason to worry about spuriousness. The “sufficiency” pat-terns in Figure 6.3 hinge on the position of the Nordic countries, andthere are factors other than generous family policies that might be thetrue cause(s) of those countries’ high female employment rates.

One is women’s preferences for employment (Hakim, 2000; Bielenski,Bosch, and Wagner, 2002). Perhaps more women in the Nordic countriesprefer employment over staying home than is the case in other countries.The first chart in Figure 6.4 shows a positive association between theshare of women aged 25 to 59 strongly agreeing that both husband andwife should contribute to household income and the female employmentrate. Unfortunately, we are unable to include women’s preferences in ouranalyses because the number of countries for which data are available istoo small. Moreover, reliable longitudinal data on women’s preferencesare altogether absent. This is particularly problematic because women’spreferences might be endogenous; if family policies (or some other fac-tors) boost the share of women in employment, this may become thenorm and generate a preference in favor of it.

A second potential source of spuriousness is women’s educationalattainment. Within countries there is a positive association across

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 202


individuals between educational attainment and likelihood of employ-ment (OECD, 2004, p. 147). Across countries, too, the average level ofeducational attainment among women and the female employment rateare positively correlated. This can be seen in the second chart in Fig-ure 6.4, which uses the average number of years of schooling completedamong women age 25 and over as a measure of women’s educationalattainment. The data points in this chart are decade averages.

Eliason, Stryker, and Tranby’s sufficiency argument is that when fam-ily policy is generous, female employment will be high, regardless ofother institutions, policies, and conditions in the country. As notedin the introductory chapter to this volume, cases that score high onthe hypothesized causal factor are the key in assessing a hypothesis ofcausal sufficiency; cases scoring low on the causal factor are analyticallyirrelevant. For a hypothesis of causal sufficiency, then, spuriousness is aconcern if there is a causal factor that is plausibly related to the outcomeon theoretical grounds and that is similar to the hypothesized causallysufficient factor(s) for the cases that score high on the hypothesized factor(s).Women’s educational attainment fits the bill. There is good reason tothink that high levels of female educational attainment increase women’semployment, by changing women’s preferences and by increasing theirearnings capacity. And the Nordic countries, which are characterized bygenerous family policies, have high levels of female educational attain-ment. This can be seen in the charts in Figure 6.5, which plot women’seducation by the family policy measures for the Nordic countries. To feelconfident that generous family policy is a sufficient condition for highwomen’s employment, we need a case with generous family policy butlow women’s educational attainment. Unfortunately, no such cases exist.

A regression approach

If we shift from a deterministic hypothesis of causal sufficiency to a ten-dential hypothesis, we can get more leverage on the question of whetherthe apparent impact of generous family policies on women’s employ-ment is spurious. A tendential hypothesis would be that generous familypolicies tend to increase women’s employment, rather than that they willalways yield high women’s employment. For a tendential hypothesis, thefact that family policy generosity correlates strongly with women’s edu-cational attainment among the Nordic countries is not an obstacle aslong as they do not correlate too strongly among the full set of coun-tries. As it turns out, they do not. Women’s education correlates at .17with public child care for age 0–2, at .04 with public child care for age

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 203


Den60sDen70s Den80s

Den90s

Fin60s

Fin70s

Fin80s

Fin90s

Nor60s

Nor80s

Nor90s

Swe80s

Swe90s

4

6

8

10

12

Wom

en’s

ave

rage

yea

rs o

f sch

oolin

g co

mpl

eted

4

6

8

10

12

Wom

en’s

ave

rage

yea

rs o

f sch

oolin

g co

mpl

eted

4

6

8

10

12

Wom

en’s

ave

rage

yea

rs o

f sch

oolin

g co

mpl

eted

4

6

8

10

12

Wom

en’s

ave

rage

yea

rs o

f sch

oolin

g co

mpl

eted

0 .2 .4 .6 .8 1.0

0 .2 .4 .6 .8 1.0


Den60sDen70s Den80s

Den90s

Fin60s

Fin70s

Fin80s

Fin90s

Nor60s

Nor80s

Nor90s

Swe80s

Swe90s

0 .2 .4 .6 .8 1.0 1.2 1.4


Den60sDen70s Den80s

Den90s

Fin60s

Fin70s

Fin80s

Fin90s

Nor60s

Nor70s

Nor80s

Nor90s

Swe60sSwe70s

Swe80s

Swe90s

Maternity leave (index)

Den60sDen70s Den80s

Den90s

Fin60s

Fin70s

Fin80s

Fin90s

Nor60s

Nor70s

Nor80s

Nor90s

Swe60sSwe70s

Swe80s

Swe90s

5 10 15 20 25

Public employment (%)

Figure 6.5 Women’s educational attainment by family policies, Nordic countries,1960s–1990sNote: For data definitions and sources, see the chapter appendix.

3–5, at −.06 with maternity leave, and at .47 with public employment.These correlations are weak or moderate in strength because the Anglocountries tend to have moderate-to-high levels of female educationalattainment but low levels of family policy generosity, while the reverseis true for a number of the continental countries.

A standard approach for assessing a tendential hypothesis about theimpact of a variable net of one or more other variables is multivariateregression. Because we are interested in the independent effects of thethree types of family policies, our first inclination would be to include allfour of the family policy measures – two of public child care and one eachof maternity leave and public employment – together in a regression with

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 204


Table 6.1 Correlations between family policy measures

Public child Public child Maternitycare, age 0–2 care, age 3–5 leave

Public child care, age 3–5 0.89Maternity leave 0.79 0.80Public employment 0.84 0.70 0.70

Note: N = 37. For data definitions and sources, see the chapter appendix.

Table 6.2 Principal components analysis of the four familypolicy measures

Factor loadings

Factor 1 Uniqueness

Public child care, age 0–2 0.96 0.07Public child care, age 3–5 0.93 0.14Maternity leave 0.90 0.20Public employment 0.88 0.22

Note: N = 35. One factor retained. Eigenvalue: 3.37. Proportion of variancein items explained: 0.84. For data definitions and sources, see the chapterappendix.

women’s educational attainment. However, the family policy measuresare too closely correlated with one another to permit this. Table 6.1 showsa correlation matrix for the four measures. For the 37 country-decadeobservations, the correlations among the measures range between .70and .89. The inclusion of variables this highly intercorrelated in the sameregression would create a severe multicollinearity problem.

One option, therefore, is to estimate separate regressions, each includ-ing one of the family policy measures along with women’s educationalattainment. Another option is to combine the family policy variablesinto a single measure. A “principal components” analysis of the fourfamily policy measures yields a component with four high loadings (thesmallest is .88) which explains 85 percent of the variance in the vari-ables. A “principal factor” analysis yields a first factor that correlates at.99 with this first principal component “factor” (Table 6.2). We use thisfirst factor as a composite measure of family policy.

We estimate a series of pooled cross-section time-series regressions offemale employment rates on each of the family policy variables – firston the family policy measures alone and then controlling for women’s

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 205


educational attainment. To avoid multicollinearity, we enter the familypolicy variables in separate regressions. A pooled regression approach isuseful because we are interested both in the cross-country variation atparticular points in time and in the over-time variation within coun-tries. We estimate these regressions using ordinary least squares, with a“robust cluster” option that adjusts the standard errors for the noninde-pendence of observations within countries (clusters) and is appropriatefor “unbalanced” panels in which the number of observations differsacross countries (Bradley et al., 2003; Moller et al., 2003).

The regression results are presented in Table 6.3. For each regressionwe list the unstandardized coefficients, t-statistics (in parentheses), andadjusted R2. The odd-numbered columns show the results of bivariateregressions. The even-numbered columns show results controlling forwomen’s educational attainment. We describe the former first and thenturn to the latter.

Bivariate results

In the bivariate regressions in the table’s first set of rows, each of the fam-ily policy variables as well as the composite (factor score) measure has theexpected positive sign and a substantively strong coefficient. For exam-ple, the coefficient for the public child care age 0–2 variable suggests thata country scoring 1 on the index had, on average, a female employmentrate about 25 percentage points higher than a country scoring 0 on theindex (column 1). The estimated magnitude of the difference betweenthe low and high end is very similar for each of the other three familypolicy measures.

To focus on the cross-country variation, we can add a set of period(decade) dummy variables to these regressions (we use the 1980s as thereference group). The results are shown in the second set of rows inTable 6.3. There is no noteworthy reduction in the magnitude of thefamily policy coefficients.

A pooled model with time (period) dummies essentially duplicates thefindings of a simple cross-sectional model; it expunges over-time within-country variation and estimates an average set of results based on purelycross-country variation. What this type of pooled model does, in effect,is to stack cross-sections together. Where there is little change in cross-sectional variation across time, the chief advantage is to increase degreesof freedom. That is the case here: between 61 and 68 percent of thevariance in the independent variables, and 70 percent of the variance infemale employment, is between countries. In the third set of rows

9780230_202573_07_cha06.tex

7/6/200816:32

Page206

206Table 6.3 Regression results

Public child care, Public child care, Family policyage 0–2 age 3–5 Maternity leave Public employment factor scores

1 2 3 4 5 6 7 8 9 10

All countries; no period or countrydummiesFamily policy variable 26.9 24.5 16.6 17.6 22.1 23.7 1.9 1.5 8.1 7.5

(5.6) (8.1) (3.2) (5.2) (2.3) (5.2) (8.0) (9.4) (5.1) (12.1)Women’s education 4.2 4.9 4.8 2.9 4.3

(9.5) (11.5) (10.6) (7.6) (9.0)Adjusted R2 0.43 0.73 0.28 0.69 0.21 0.70 0.62 0.76 0.48 0.79

Time period dummies a

Family policy variable 24.2 17.3 14.1 23.2 17.5 26.0 1.8 1.5 7.5 7.6(4.2) (6.7) (2.0) (3.9) (1.5) (4.8) (6.6) (8.5) (3.5) (8.4)

Women’s education 3.9 4.8 5.2 2.7 4.3(7.1) (7.8) (8.4) (5.5) (9.0)

Adjusted R2 0.47 0.71 0.29 0.66 0.22 0.69 0.63 0.75 0.47 0.77Cross-section of 1980s and 1990s averages



Adjusted R2 0.38 0.81 0.24 0.78 0.20 0.77 0.47 0.84 0.65 0.82

(Continued)

9780230_202573_07_cha06.tex

7/6/200816:32

Page207

207

Table 6.3 (Continued)

Public child care, Public child care, Family policyage 0–2 age 3–5 Maternity leave Public employment factor scores

1 2 3 4 5 6 7 8 9 10

Country dummies (full set) a



Adjusted R2 0.55 0.83 0.56 0.83 0.73 0.87 0.78 0.88 0.64 0.83Denmark and Sweden excluded b



Adjusted R2 0.17 0.63 0.02 0.59 0.02 0.63 0.42 0.65 0.20 0.72N 31 31 31 31 40 40 41 41 27 27

Notes: Unstandardized coefficients and absolute t-statistics (in parentheses) from ordinary least squares (OLS) regressions with “robust cluster” option(country is the clustered variable). Unit of analysis is the country-decade. Dependent variable is women’s employment rate. For data definitions andsources, see the chapter appendix.a Results for dummy variables are not shown to save space.b No time period or country dummies included. Excluding Denmark reduces the number of observations by three. Excluding Sweden reduces thenumber of observations by two for the public child care regressions and by three for the maternity leave and public employment regressions.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 208


in Table 6.3, we show the results from a model that simply estimatesthe 1980s and 1990s means of variables’ values for the 1980s and 1990spanels alone, as these are virtually free of missing data for any vari-able. (Actually, 1990s data are missing for child care in Italy and formaternity leave in the United Kingdom and for the composite familypolicy measure for these two countries; we substitute 1980s values forthese variables.) The results for these strictly cross-sectional estimatesof family policy effects are always similar to those in most of the othermodels.

To focus on the over-time variation within countries, we can add aset of country dummies. This seems particularly likely to reduce theestimates of the impact of family policies because there are more coun-tries than time periods (14 versus four), there is more variation in familypolicies across countries than over time within countries, and country-specific features such as culture may influence both family policy andwomen’s employment. The results with country dummies are in thefourth set of rows in Table 6.3. Perhaps surprisingly, the coefficients donot decrease, at least in models without a control for female educationalattainment (odd-numbered columns).

The scatterplots in Figure 6.3 above suggest a key role for Denmarkand Sweden. These two countries have tended to have by far the mostgenerous family policies and also the highest rates of women’s employ-ment. To what degree does the association between family policy andwomen’s employment hinge on these two nations? Regression resultswith Denmark and Sweden omitted are shown in the fifth set of rows inTable 6.3. Here we do observe a noteworthy decline in the magnitude ofthe estimated effect for two of the family policy measures – public daycare for children age 3–5 and maternity leave. The coefficients for thesetwo measures decrease by more than half, the t-statistics are only around1.0, and the adjusted R2s fall to nearly zero.

Multivariate results controlling for women’s education

What happens when we control for women’s educational attainment(average years of schooling completed among women aged 25 and over)?The findings are in the even-numbered columns in Table 6.3. Consistentwith the bivariate pattern shown in the second chart in Figure 6.4 above,substantively strong and statistically significant effects are evident forfemale educational attainment in all of the models.

In many of the models the estimates for the family policy measuresremain similar to those in the bivariate regressions. However, in themodels with country fixed effects (country dummies) the coefficients

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 209


and t-statistics for the family policy variables drop dramatically. Thecoefficient for female education, by contrast, does not decline.

What should we make of this? The common interpretation is that theassociations between the family policy measures and women’s employ-ment are probably spurious; they are artifacts of unmeasured nation-specific differences, rather than truly causal. That might be the correctinference. But we should be careful about settling on this conclusion tooquickly.

The presumed spuriousness may be mistaken, either because (a) thereare no such unmeasured variables with the stipulated traits and inclusionof units simply truncates variance in X or (b) the unmeasured correlatesof the units affecting Y and attenuating the estimate for X are con-sequences of X and thus mere technical causes of bias in a structuralcoefficient but not causes of any spuriousness. It is also possible that theunits correlate with unmeasured interaction terms which would, couldthey be measured, reveal notable effects for a subset of cases. But thatdoes not mean such a variable has no causal impact. Suppose the eco-nomic or social or political environment in countries changes in waysconducive to women’s employment. Perhaps, for instance, women’s(and maybe men’s) attitudes toward female employment become moreencouraging. But suppose the degree to which this yields an actualincrease in female employment depends on the degree to which a coun-try has supportive family policies. Ideally, the researcher would modelthis via an interaction between women’s attitudes and family policies.But suppose we lack good data on women’s attitudes; perhaps cross-nationally comparable survey data are not available, or those data donot effectively capture attitudes. Without such an interaction, it wouldbe impossible to capture the effect of unchanging family policies in apooled regression with country fixed effects. Instead, an appropriatemodel might be a cross-sectional design with women’s employment mea-sured in change scores and family policies measured in (average) levels(see, for example, Kenworthy, 2004, 2008).

Yet this is not what accounts for the family policy variables’ results inthe “country dummies” regressions in Table 6.3 here. Figures 6.6 through6.9 show the over-time data for the four family policy measures: publicchild care for age 0–2, public child care for age 3–5, maternity leave, andpublic employment. Figure 6.10 shows over-time trends in the factorscores derived from these four policy measures and used in the regres-sions in Table 6.3. While some of the trend lines indicate little or nochange over time – most notably, child care support in the Anglo coun-tries – in most countries family policy generosity has increased. Thus,lack of over-time variation does not appear to be the problem for thefamily policy variables.

9780230_202573_07_cha06.tex

7/6/200816:32

Page210

210

0

.2

.4

.6

.8

1

Inde

x

1960 1970 1980 1990 2000

Year

Den FinNor Swe

Nordic Countries

0

.2

.4

.6

.8

1

Inde

x

1960 1970 1980 1990 2000

Year

CanUK

IreUS

Anglo Countries

0

.2

.4

.6

.8

1

Inde

x

1960 1970 1980 1990 2000

Year

AusFrIt

BelGerNth


Figure 6.6 Public child care, age 0–2, 1960s–1990sNote: Data are decade averages. Missing data: 1960s and 1970s for Austria, Belgium, France, Germany, Ireland, Sweden, and the United Kingdom; 1960s,1970s, and 1990s for Italy; 1970s for Norway. For data definitions and sources, see the chapter appendix.

9780230_202573_07_cha06.tex

7/6/200816:32

Page211

211

0

.2

.4

.6

.8

1

1.2

1.4

Inde

x

1960 1970 1980 1990 2000

Year

Nordic Countries

Den FinNor Swe

0

.2

.4

.6

.8

1

1.2

1.4

Inde

x

1960 1970 1980 1990 2000

Year


AusFrIt

BelGerNth

0

.2

.4

.6

.8

1

1.2

1.4

Inde

x

1960 1970 1980 1990 2000

Year

Anglo Countries

CanUK

IreUS

Figure 6.7 Public child care, age 3–5, 1960s–1990sNote: Data are decade averages. Missing data: 1960s and 1970s for Austria, Belgium, France, Germany, Ireland, Sweden, and the United Kingdom; 1960s,1970s, and 1990s for Italy; 1970s for Norway. For data definitions and sources, see the chapter appendix.

9780230_202573_07_cha06.tex

7/6/200816:32

Page212

212

0

.2

.4

.6

.8

1

Inde

x

1960 1970 1980 1990 2000

Year

Nordic Countries

Den FinNor Swe

1960 1970 1980 1990 2000

Year


AusFrIt

BelGerNth

0

.2

.4

.6

.8

1

Inde

x

0

.2

.4

.6

.8

1

Inde

x

1960 1970 1980 1990 2000

Year

Anglo Countries

CanUK

IreUS

Figure 6.8 Maternity leave, 1960s–1990sNote: Data are decade averages. Missing data: none. For data definitions and sources, see the chapter appendix.

9780230_202573_07_cha06.tex

7/6/200816:32

Page213

213

0

5

10

15

20

25

% o

f wor

king

-age

pop

ulat

ion

1960 1970 1980 1990 2000Year

Nordic Countries

Den FinNor Swe

% o

f wor

king

-age

pop

ulat

ion

0

5

10

15

20

25

1960 1970 1980 1990 2000Year


AusFrIt

BelGerNth

% o

f wor

king

-age

pop

ulat

ion

0

5

10

15

20

25

1960 1970 1980 1990 2000Year

Anglo Countries

CanUK

IreUS

Figure 6.9 Public employment, 1960s–1990sNote: Data are decade averages. Missing data: 1960s and 1970s for Belgium. For data definitions and sources, see the chapter appendix.

9780230_202573_07_cha06.tex

7/6/200816:32

Page214

214

1960 1970 1980 1990 2000

Year

Nordic Countries

�1

�.5

0

.5

1

1.5

2

2.5

Fact

or s

core

s

Den FinNor Swe

�1

�.5

0

.5

1

1.5

2

2.5

Fact

or s

core

s

1960 1970 1980 1990 2000

Year


AusFrIt

BelGerNth

�1

�.5

0

.5

1

1.5

2

2.5

Fact

or s

core

s

1960 1970 1980 1990 2000

Year

Anglo Countries

CanUK

IreUS

Figure 6.10 Family policy factor scores, 1960s–1990sNote: Data are decade averages. Missing data: 1960s and 1970s for Austria, Belgium, France, Germany, Ireland, and Sweden; 1960s, 1970s, and 1990sfor Italy and the United Kingdom; 1970s for Norway. For data definitions and sources, see the chapter appendix.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 215


Moreover, the coefficients for the family policy variables shrink notwhen country fixed effects are added to the regressions, but rather whenthe country fixed effects and women’s educational attainment are added.This suggests that over time within countries, female employment ismore closely associated with women’s educational attainment than withfamily policy generosity.

Careful inspection of the scatterplots and time plots in Figures 6.2–6.10suggests that this is due to developments in a variety of countries. Thegenerosity/extensiveness of family policies increased in the Nordic coun-tries, particularly during the 1960s, 1970s, and 1980s. This correspondsto the steady increase in female employment rates in these countriesduring those three decades. But other countries are more problematic.

Women’s employment rates in the Anglo countries rose steadily duringthe four decades, and the extent of this rise was on par with that in theNordic countries. Trends in public child care are of no help in explainingthis development. Trends in maternity leave and public employment areconsistent with the rise in women’s employment, but they seeminglywere not large enough to be of much explanatory relevance – certainlynot in the United States, at least.

A second problematic pattern is the rise in the level of women’semployment in a number of the continental countries in the 1990s aftereffectively no change in the 1960s, 1970s, and 1980s. For the familypolicies for which early data are available, the data suggest increasinggenerosity in these countries throughout the period, not just in the1990s. Perhaps the policies produced a slow release of cultural pressureson women and employers that inhibit female labor force participation.

The Netherlands is also a problematic case in terms of over-time trends.From the mid-1980s the female employment rate in the Netherlands rosedramatically. None of the country’s family policies were significantlyaltered prior to or during this period. Indeed, studies of the Dutch caseseldom assert a substantial role for government policy in precipitatingthe rise in female employment (for example, Visser, 2002; Misra andJude, this volume).

Finally, a fourth problematic trend is the stagnation (and, in somecases, the decline) in female employment rates in the Nordic countriesin the 1990s. Part of this owes to the economic crises experienced byFinland and Sweden during the first half of the decade, and part of it maybe a product of a “ceiling” effect (female employment may have nearedits maximum achievable level by the early 1990s). Still, the pattern isnot what we would expect given the continued increase in family policygenerosity in most of these countries during that decade.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 216


Aus80s

Aus90s

Bel80s

Bel90sCan70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr80sFr90sGer80s

Ger90s

Ire80s

Ire90s

It80sNth80s

Nth90s

Nor60s

Nor90s

Swe80s

Swe90s

UK80s

US60s

US70s

US80s

US90s

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)

30

40

50

60

70

80

Wom

en’s

em

ploy

men

t rat

e (%

)�1.5 �.5 .5 1.5 2.5

Composite family policy (factor scores)

Family policies

Aus70sAus80s

Aus90s

Bel80s

Bel90sCan70s

Can80s

Can90sDen70s

Den80sDen90sFin80s

Fin90s

Fr70sFr80s


Ger90s

Ire70sIre80s

Ire90s

It60sIt80s

It90s Nth80s

Nth90s

Nor60s

Nor90s

Swe70s

Swe80s

Swe90s

UK60s

UK70sUK80s

UK90s

US60s

US70s

US80s

US90s

4 6 8 10 12

Women's educational attainment (years)

Women’s education

Figure 6.11 Women’s employment by family policies and by women’s education:over-time within-country patternsNote: For data definitions and sources, see the chapter appendix.

Over-time trends in female educational attainment cannot fullyaccount for these developments, but the association between women’sschooling and female employment is much stronger than is the casefor family policies. A useful way to see this is via the scatterplots inFigure 6.11. The first shows women’s employment by the composite fam-ily policy variable; the second shows women’s employment by women’seducation. Both plots feature pooled data, and in each we show within-country regression lines. One issue is that the data are more completefor women’s educational attainment than for the family policy variable,but the key thing to note is that several of the regression lines in thefamily policy plot are vertical or nearly so. This indicates a rise in femaleemployment despite little or no increase in the generosity of family pol-icy. By contrast, virtually all of the lines in the women’s education plothave the expected positive slope.

What of the possibility that it is, in fact, levels of family policy gen-erosity that have influenced over-time trends in women’s employment?We can test this with an OLS regression of women’s employment change(measured as 1990s average minus 1970s average; 1960s data are notavailable for many countries) on the average level of family policy gen-erosity over the four decades. Table 6.4 shows the results of such aregression for each of the family policy measures, both with and withouta control for change in female educational attainment. These findingsoffer little support for the hypothesis. Virtually all of the family policy

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 217


Table 6.4 Regression results: change in women’s employment on level of familypolicy generosity

No control Control Controls for initial levelfor change for change of women’s employmentin women’s in women’s and for change in women’seducational educational educational attainmentattainment attainment

Public child care, age 0–2 −10.0 −17.9 −3.3Public child care, age 3–5 −8.4 −11.7 −4.7Maternity leave −23.0 −31.4 −23.2Public employment 0.04 −0.42 1.12Family policy factor scores −3.4 −5.8 −2.3

Note: Unstandardized coefficients. N = 14. For data definitions and sources, see the chapterappendix.

coefficients are negative, even without controlling for women’s school-ing. The one exception is the coefficient for public employment in thethird column. In the regressions reported in that column, the initiallevel of women’s employment is controlled for. This takes into accountthe possibility of a ceiling effect, whereby countries beginning (in 1979)with a high level of female employment find it more difficult to realizefurther increases. Here the coefficient is positive, as expected. It also isstatistically significant (at the .10 level), and the standardized coefficientis larger than that for the change in women’s education variable. Thissuggests some indication of a positive effect of family policy on over-timetrends in female employment rates.

On the other hand, public employment is less directly a family policythan are child care and maternity leave, in the sense that the level ofpublic employment is aimed less at promoting women’s employment.Arguments for generous family-friendly policies tend to focus on childcare and parental leave.

Conclusion

We have suggested reasons to question Eliason, Stryker, and Tranby’sconclusion that generous family policy is a sufficient condition for highlevels of women’s employment. That conclusion hinges on the gener-ous family policies and high female employment rates in the Nordiccountries, and it could be that high levels of women’s education are thetrue cause of high female employment in those countries. To examinethe net effect of family policies, we turn to a tendential hypothesis and

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 218


use regression analysis. When we control for female educational attain-ment and country fixed effects, much of the association between familypolicy generosity and women’s employment disappears. Our best guessis that generous family policies have helped to boost women’s employ-ment, but the macrocomparative evidence is less than overwhelming insupport of this conclusion.

Appendix: Variable Definitions and Data Sources

The data used in this chapter are available at www.u.arizona.edu/∼lkenwor.Employment: men’s. Employed men as a share of the male population age

15 to 64. Source: Author’s calculations from data in OECD (2006).Employment: women’s. Employed women as a share of the female popu-

lation age 15 to 64. Source: Author’s calculations from data in OECD(2006).

Family policy factor scores. See the text for discussion. Source: Author’scalculations.

Maternity leave. Index of the generosity of maternity leave. Source: Strykeret al. (2008, table 3).

Public child care, age 0–2. Index of the generosity of government provisionand subsidization of child care for children age zero to two. Source:Stryker et al. (2008, table 4).

Public child care, age 3–5. Index of the generosity of government provisionand subsidization of child care for children age zero to two. Source:Stryker et al. (2008, table 5).

Public employment. Persons employed in the public sector as a share ofthe population age 15 to 64. Source: Stryker et al. (2008, table 2).

Women’s education. Average years of schooling completed among womenage 25 and over. Source: Barro and Lee (n.d.).

Women’s preference for employment. Share of women age 25 to 59 stronglyagreeing that both husband and wife should contribute to house-hold income. Source: Author’s calculations from data in World ValuesSurvey (1995–97).

References

Anderson, Patricia M. and Andrew Levine. 1999. “Child Care and Mother’sEmployment Decisions.” Working Paper 7058. Cambridge, MA: NationalBureau of Economic Research.

Barro, Robert J. and Jong-Wha Lee. n.d. “International Data on EducationalAttainment.” Available at: www.cid.harvard.edu/ciddata/ciddata.html.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 219


Barrow, Lisa. 1996. “An Analysis of Women’s Labor Force Participation Follow-ing First Birth.” Working Paper 363. Industrial Relations Section, PrincetonUniversity.

Bielenski, Harald, Gerhard Bosch, and Alexandra Wagner. 2002. Working TimePreferences in Sixteen European Countries. Dublin: European Foundation for theImprovement of Living and Working Conditions.

Bradley, David, Evelyne Huber, Stephanie Moller, François Nielsen, and JohnStephens. 2003. “Distribution and Redistribution in Postindustrial Democra-cies.” World Politics 55: 193–228.

Chevalier, A. and T. Viitanen. 2002. “The Causality Between Female Labour ForceParticipation and the Availability of Childcare.” Applied Economics Letters 9:915–18.

Del Boca, Daniela. 2002. “Low Fertility and Labour Force Participation of ItalianWomen: Evidence and Interpretations.” Labour Market and Social Policy Occa-sional Paper 61. Organization for Economic Cooperation and Development.Available at: www.oecd.org.

Esping-Andersen, Gøsta. 1990. The Three Worlds of Welfare Capitalism. Princeton,NJ: Princeton University Press.


Fagnani, J. 1998. “Recent Changes in Family Policy in France: Political Trade-offs and Economic Constraints.” Pp. 58–65 in Women, Work, and Family inEurope, edited by Eileen P. Drew, Ruth Emerek, and Evelyn Mahon. London:Routledge.

Goodin, Robert E., Bruce Headey, Ruud Muffels, and Henk-Jan Dirven. 1999. TheReal Worlds of Welfare Capitalism. Cambridge: Cambridge University Press.

Gottschall, Karin and Katherine Bird. 2003. “Family Leave Policies and Labor Mar-ket Segregation in Germany: Reinvention or Reform of the Male BreadwinnerModel?” Review of Policy Research 20: 115–34.

Gustafsson, Siv and Frank Stafford. 1992. “Child Care Subsidies and Labor Supplyin Sweden.” Journal of Human Resources 27: 204–30.

Hakim, Catherine. 2000. Work–Lifestyle Choices in the 21st Century: PreferenceTheory. Oxford: Oxford University Press.

Hofferth, Sandra L. and Sally C. Curtin. 2003. “The Impact of Parental Leaveon Maternal Return to Work after Childbirth in the United States.” OECDSocial, Employment, and Migration Working Paper 7. Organization for Eco-nomic Cooperation and Development. Available at: www.oecd.org/dataoecd/26/45/2955849.pdf.

Huber, Evelyne and John D. Stephens. 2001. Development and Crisis of the WelfareState. Chicago: University of Chicago Press.

Ilmakunnas, S. 1997. “Public Policy and Child Care Choice.” Pp. 178–93 in Eco-nomics of Family and Family Policy, edited by I. Persson and C. Jonung. London:Routledge.

Joesch, Jutta M. 1997. “Paid Leave and the Timing of Women’s Employ-ment Before and After Childbirth.” Journal of Marriage and the Family 59:1008–21.


Kenworthy, Lane. 2008. Jobs with Equality. Oxford: Oxford University Press.

9780230_202573_07_cha06.tex 7/6/2008 16: 32 Page 220


Kimmel, Jean. 1998. “Child Care Costs as a Barrier to Employment for Single andMarried Mothers.” Review of Economics and Statistics 2: 287–99.

Leibowitz, A., J. A. Klerman, and L. J. Waite. 1992. “Employment of New Mothersand Child Care Choice.” Journal of Human Resources 27: 112–33.

Michalopoulos, Charles and Philip K. Robins. 2000. “Employment and ChildCare Choices in Canada and the United States.” Canadian Journal of Economics33: 435–70.

Moller, Stephanie, David Bradley, Evelyne Huber, François Nielsen, and JohnD. Stephens. 2003. “Determinants of Relative Poverty in Advanced CapitalistDemocracies.” American Sociological Review 68: 22–51.

OECD (Organization for Economic Cooperation and Development). 2004. Educa-tion at a Glance. Paris: OECD.

OECD. 2006. OECD Labour Force Statistics Database. Available at: www.oecd.org/scripts/cde/members/lfsdataauthenticate.asp.

Ondrich, Jan, C. Katherina Spiess, and Qing Yang. 1996. “Barefoot in a GermanKitchen: Federal Parental Leave and Benefit Policy and the Return to Work AfterChildbirth in Germany.” Journal of Population Economics 9: 247–66.

Ondrich, Jan, C. Katherina Spiess, Qing Yang, and G. G. Wagner. 1999. “FullTime or Part Time? German Parental Leave Policy and the Return to Work AfterChildbirth in Germany.” Research in Labor Economics 18: 41–74.

Powell, Lisa M. 1998. “Part-Time versus Full-Time Work and Child Care Costs:Evidence for Married Mothers.” Applied Economics 30: 503–11.

Pylkkänen, Elina and Nina Smith. 2003. “Career Interruptions Due to ParentalLeave: A Comparative Study of Denmark and Sweden.” Social, Employment,and Migration Working Paper 1. Organization for Economic Cooperation andDevelopment. Available at: www.oecd.org.

Ragin, Charles 2000. Fuzzy-Set Social Science. Chicago: University of Chicago Press.Rønsen, Marit and Marianne Sundstrom. 2002. “Family Policy and After-Birth

Employment Among New Mothers: A Comparison of Finland, Norway, andSweden.” European Journal of Population 18: 121–52.

Smith, Kristen, Barbara Downs, and Martin O’Connell. 2001. “Maternity Leaveand Employment Patterns: 1961–1995.” Household Economic Studies ReportP70–9. Washington, DC: US Bureau of the Census.

Stryker, Robin, Scott Eliason, and Eric Tranby. 2008. “The Welfare State, Fam-ily Policies, and Women’s Labor Force Participation: Combining Fuzzy-Set andStatistical Methods to Assess Causal Relations and Estimate Causal Effects.”Chapter 5 in this volume.

Swank, Duane. 2002. Global Capital, Political Institutions, and Policy Change inDeveloped Welfare States. Cambridge: Cambridge University Press.

Visser, Jelle. 2002. “The First Part-Time Economy in the World: A Model To BeFollowed?” Journal of European Social Policy 12: 23–42.

World Values Survey. 1995–97. Available at: www.icpsr.org.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 221

7Part-Time Work and the Legacy ofBreadwinner Welfare States: APanel Study of Women’sEmployment Patterns in Germany,the United Kingdom, and theNetherlands, 1992–2002Jelle Visser and Mara Yerkes

7.1 Introduction

If there is one non-controversial stylized fact about the development ofemployment in the western world, it is the feminization of labor mar-kets. On average, calculated across twenty OECD countries, the femaleemployment rate rose from 49.2 to 59.0 percent between 1983 and 2003,whereas the male employment rate decreased from 77.7 to 73.6 percent.In these twenty years the male–female employment gap halved from 28.5to 14.5 percent.

The gender employment gap widens in all countries when we takethe presence of children into account (Table 7.1, left panel). Women arestill predominantly responsible for looking after children: the presenceof children in the household reduces the level of female employmentwhereas it increases the level of male employment. On average, cal-culated for the 17 countries shown in Table 7.1, the employment rateof women aged 25 to 54 years decreases from 72.0 percent for womenwithout children to 62.7 percent for women with two or more chil-dren, and the gender gap increases from 12.5 percent to 30.5 percent.This effect is found in all countries, though it is smallest in the welfarestates of Northern Europe (Denmark, Sweden, Finland, and Norway).The largest contrast is with the three Mediterranean countries (Greece,Italy, and Spain), which have the lowest female employment rates to startwith, experience the largest fall in employment due to the presence of

221

9780230_202573_08_cha07.tex

5/6/200815:39

Page222

222Table 7.1 Employment rates by presence of children, 2000

Women aged 25 to 54 years Persons aged 25 to 54 years

No children Two or more children No children Two or more children

Employment Gender Employment Gender Incidence part-time Incidence part-timerate gap rate gap

Female Male Female Male

Denmark 78.5 −7.7 77.2 −12.9 18.5 . . 16.2 3.7Finland 79.2 −9.1 73.5 −19.7 7.5 . . 13.6 3.7Norway 82.9 −5.9 78.0 . . 24.7 5.0 41.1 5.0Sweden 81.9 −0.4 81.8 −9.4 14.6 5.2 22.2 4.3

Austria 76.0 −10.5 65.7 −29.0 17.4 2.1 43.7 1.9Belgium 65.6 −17.4 69.3 −24.7 29.2 6.5 46.1 5.9France 73.5 −9.6 58.8 −32.9 20.0 5.2 31.8 4.4Germany 77.3 −7.2 56.3 −35.6 24.0 4.2 60.2 3.4Netherlands 75.3 −15.6 63.3 −30.8 38.3 6.2 82.7 5.5

Greece 53.1 −31.1 50.3 −45.4 8.4 2.8 11.2 2.7Italy 52.8 −26.2 42.4 −49.9 20.0 5.5 34.4 5.1Spain 54.6 −26.0 43.3 −48.6 13.7 2.6 18.6 1.9Portugal 72.6 −13.4 70.3 −24.8 11.5 2.7 11.3 2.0

Ireland 65.8 −14.1 40.8 −43.2 16.6 4.3 46.4 4.0United 79.9 −5.4 62.3 −28.2 23.7 4.1 62.8 3.7Kingdom

Canada 76.5 −6.0 68.2 −23.6 17.0 5.2 30.7 4.3United 78.6 −7.2 64.7 −29.0 10.1 3.5 23.6 2.7States

Average 72.0 −12.5 62.7 −30.5 18.5 4.3 35.1 3.8

Source: based on figures from OECD Secretariat, derived from labour force surveys (national data); . . = no data.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 223

Jelle Visser and Mara Yerkes 223

children and have the largest gender gaps for women with and with-out children. However, the presence of children also lowers the level ofwomen’s employment and widens the gender employment gap in thewelfare states of mainland Western Europe, the UK and the US.

Women tend to work fewer paid hours than men, however, so focus-ing on the gender gap in employment–population rates understates thedifferences in employment levels between men and women. Generally,the incidence of part-time work is largest among students (especiallyin countries where student grants do not match the costs of highereducation), among mothers of young children, and among men andwomen close to retirement. Focusing on persons aged 25 to 54 years,we see in Table 7.1 (right panel) that the incidence of part-time workamong employed women increases with the presence of children, froman average of 18.5 percent for women without children to 35.1 percentfor women with two or more children. The presence of children tendsto increase the share of part-time jobs among women in employment inall countries except Denmark and Portugal. In contrast, the presence ofchildren decreases the incidence of part-time work among men, from 5.2percent to 4.3 percent on average, with remarkably little cross-nationalvariation.1

The effect of children on the incidence of part-time employmentamong women is small or absent in Finland, Greece, Spain, and Portu-gal, countries in which part-time work has been relatively undevelopedand the choice for most women is between a full-time job or no job,resulting in traditionally high female employment rates in Finland andPortugal, and low rates in Greece and Spain (Italy belongs to the samegroup, but more women have taken to part-time employment in recentyears).2 The presence of children also has only a modest or negligibleeffect in two high employment countries in which the level of part-timejobs among women used to be common but has decreased since the1980s: Sweden and Denmark. The pattern in the two Scandinavian wel-fare states suggests that part-time employment may be a lifestyle choiceunconnected to motherhood, whereas women with children who wantto continue working full-time can do so thanks to the public provision ofchild care and employment relationships, which include extensive leaverights for parents.3 The largest impact of children on the take-up of part-time jobs among women aged 25 to 54 years is found in the Netherlands,the United Kingdom and Germany – the three countries where the inci-dence of part-time working women with two or more children exceeded60 percent in 2000 (Table 7.1, right panel).

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 224


Part-time employment is a complex phenomenon. It may mean any-thing from working one hour to working 34 hours a week.4 Part-timejobs may be offered the same protection as full-time jobs or they mayfall under a separate regime; they may be involuntary, because full-timejobs are not available, or voluntary, as it best suits the personal situationand preferences. Employers may use part-time work to increase work-force flexibility and part-time employment may come with a penaltyto wages, training and career opportunities, but it can also be used bywomen or parents as a means to increase family income while organiz-ing the job around other priorities, such as caring for young children.Scholars working within the framework of a “transitional labor market”speculate about the potential of the part-time model as a new form of“full” employment over the life cycle, based on working weeks of 30hours on average (Rogowski and Wilthagen, 2002; Schmid and Gazier,2002). In the Netherlands, the government endorsed a so-called “com-bination model” of work and care during the 1990s, in which the choiceof part-time employment is expected to bring about a less gendered divi-sion of paid work and household tasks (Plantenga, 2002; Visser et al.,2004). Bollé (2001), writing for the ILO, concludes his overview that, ifchosen freely and protected by high standards, part-time employmentmay be an attractive option and offer a means of striking a balancebetween time to earn a living and time for other activities. Since its incep-tion in 1998, the European Employment Strategy has tried to promotehigher employment levels as a foremost policy goal for EU member states.Increasing the popularity of part-time work is seen as an instrument foreasing the transition into the labor market and reaching higher ratesof female employment, especially in Southern and continental Europe(Taskforce, 2003). In recent years, ten member states, mostly in Southernand Eastern Europe, received specific recommendations from the Euro-pean Council to modernize labor law in order to raise the attractivenessof part-time work.

Citing the title of a well-known collection concerning women work-ing part-time in Europe and the US, part-time employment can beeither an equalizing force by bringing more women into the labormarket or preventing them from withdrawing, or a force marginaliz-ing female employment to unrewarding, dead-end jobs (Blossfeld andHakim, 1997). Which of the two tendencies is dominant is a key issue forresearch and policy. In the collection cited above, Catherine Hakim raisesthe question of whether part-time work is a qualitatively different typeof workforce involvement from full-time employment, one which givespriority to some other non-market activities around which the part-time

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 225


job must be fitted (Hakim, 1997). This is consistent with the notionthat most part-time employment is highly gendered and concentratedamong mothers with young children. This may be sufficient cause formarginalization, though this need not be so. Much will depend on insti-tutions and policies shaping female labor supply and employer demand.Even if organized around other, possibly dominant non-market activi-ties such as child care, part-time workers may gain pay levels, workingconditions and protection on a par with full-time workers. Marginaliza-tion of women and part-time workers can have its main causes in factorsnot relating to the characteristics of supply, but to labor demand andemployer behavior. We will argue that with regard to part-time worktwo sets of institutions are especially important: (1) welfare state provi-sions and interventions, in particular relating to child care and leaverights; and (2) employment relations, in particular labor law, union behav-ior and collective bargaining, since many issues related to working timeand employee rights are settled by law and collective agreement or unionpressure.5

7.2 Choice of countries, research questions and data

For our empirical study we have chosen the three countries with the high-est incidence of part-time employment among women: Germany, theNetherlands, and the United Kingdom. As can be seen from Figure 7.1,part-time employment remained at a high level (UK) and increased to ahigh (Germany) or very high level (Netherlands), quite distinct from thepatterns shown elsewhere. In Denmark, for example, women’s part-timeemployment peaked in 1978 and has halved since this time. This decreasereflects various changes in taxation and unemployment insurance, mak-ing part-time work less attractive, but the availability of full-time publicchild care for young children of pre-schooling age is probably decisive(Rasmussen et al., 2005). There was also a decline in the incidence ofpart-time work in Sweden and the US, whereas it has remained at lowlevels in France and Southern Europe.

The UK, (West) Germany6 and the Netherlands are welfare states inwhich the “male breadwinner/female carer” model has strong rootsand paid employment of mothers of young children was long discour-aged (Lewis, 1992; Knijn, 1994). Childcare provisions were lacking orunaffordable and opening times of schools and kindergartens limited,making part-time employment the dominant choice of mothers withyoung children (Killmann and Klein, 1997; Pfau-Effinger, 1998; Blossfeldand Rohwer, 1997; Visser, 2002). Furthermore, spouse-based splitting

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 226


0,0

10,0

20,0

30,0

40,0

50,0

60,0

70,0

1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005Wom

en in

par

t-tim

e em

ploy

men

t as

% o

f tot

al fe

mal

e em

ploy

men

t

DK FR GE IT

NL SW UK US

Figure 7.1 Incidence of part-time employment among womenSource: OECD data.

joint tax systems supported high levels of sole earner family income(Dingeldey, 2001). Under these conditions, part-time employmentbecame an inroad into the labor market for many women in each of thesecountries.

Beyond this common legacy, there are major institutional and policydifferences in how the welfare state and industrial relations operate. TheUK is usually classified as a liberal welfare state and uncoordinated marketeconomy (Hall and Soskice, 2001; Esping-Andersen, 1990). Work–familylife choices are deemed a private matter not to be interfered with bythe state, and the role of unions and collective bargaining has beenmuch diminished in the past two decades (Milward et al., 2000; Kersleyet al., 2006). Germany is commonly classified as a coordinated mar-ket economy and conservative welfare state with the status rights ofworking families shielded against market forces. Unions and collectivebargaining, although in decline, have remained important in shapingemployment, wage and benefit patterns throughout the 1990s (Hassel,

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 227


2006). The Netherlands is a mixed case, with an increasingly liberal-ized and flexible economy and labor market, and a revived tradition ofunion and employer cooperation and involvement in government pol-icy making (Visser and Hemerijck, 1997). How these differences play outin shaping female employment – and especially part-time employment –is the key question of this chapter.

In the next section, we specify the institutional conditions in rela-tion to our hypotheses concerning patterns of female employment andpart-time work. Our strategy is to test these hypotheses in three steps,using individual panel data covering the period 1992–2002. First we esti-mate the effects of motherhood on the probability of adult women tobe employed full-time or part-time, or remain outside the labor force.In particular, we are interested in the differences across birth cohorts,controlling for education and household status. Decline of the breadwin-ner state legacy should show in a weakening of the association betweenmotherhood and particular employment patterns such as inactivity andsmall part-time jobs. Our next step is to analyze the transitions frompart-time employment into inactivity or full-time employment, againfocusing on the impact of motherhood upon relative risk ratios. The con-tinued strength of a breadwinner legacy is expected to show in a higherrisk to move from full-time to small part-time jobs and from part-timeemployment to inactivity. Following, a comparison with the transitionsfrom full-time jobs into long or small part-time jobs or into inactivitycan help answer the question whether part-time employment does infact help particular groups of women to remain in the labor force. In thefinal part, we address the question of choice, bringing into play workingtime preferences of women and whether or not they lead to transitionsin the desired direction.

We use three separate panel data sets covering the years 1992–2002: the British Household Panel Survey (BHPS), the German Socio-Economic Panel (GSOEP) and the OSA Labor Supply Panel (LSP) for theNetherlands.7 These surveys allow us to follow the same individuals andhouseholds over several consecutive years. Rather than using the Euro-pean Community Household Panel (ECHP, 1995–2001), we have chosento use national data in order to cover a longer period (1992–2002) andinclude questions relating to the working time preferences of women.The main comparability issue is that the British and German panelsinterview the same respondents on a yearly basis, while the Dutch panelinterviews respondents every two years.8 Samples are restricted to per-sons aged between 15 and 64 (working-age population) at the time of theinterview. Full-time students are excluded from the analyses, as they are

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 228


likely to increase the amount of inactivity found in the sample or createa bias towards small hours part-time jobs.

The use of panel surveys has some well-known advantages and disad-vantages. Panel data give researchers the opportunity to follow behav-ioral choices through time for the same individuals, increasing our abilityto identify causal factors and allowing for a robust statistical analysis(Halpin, 2004). The main drawback is that a panel based on a rep-resentative sample of the population in its initial year may have lostits representativeness due to changes in the underlying population andpanel attrition. Attrition is a particular problem if the respondents whodrop out of the panel share certain characteristics systematically relatedto the subject of study (for instance, mothers of large families or womenretiring from the labor market). The three panels use several methods tocontrol for such biases, among others by renewing the panel on an incre-mental basis and the use of weights for descriptive statistics. We believethat the advantages outweigh the disadvantages and that using longi-tudinal panel data gives us a better hold on identifying the changingrelationship between marital status, motherhood and employment thancross-sectional analysis or time-series studies based on annual surveys(see Bielenski et al., 2002; van der Lippe and van Dijk, 2001).

In this study, we adopt a multinomial logit approach to measuringwomen’s employment. Multinomial logit models use maximum like-lihood estimation for models with a polytomous dependent variable.These models assume a choice of outcomes. In our analysis, the depen-dent variable is represented by a four-point dummy variable indicatingthe respondent’s allocation to one of four possible labor market states:a full-time job of 35 hours per week or more, a substantial part-timejob (20–34 hours), a small part-time job (1–19 hours), or a positionoutside the labor force (inactive status in the labor market).9 Althoughvarious classifications of part-time working hours are feasible (Gustafsonet al., 2001; Lemaitre et al., 1997), most standard full-time workweeks inEurope vary between 36 and 41 hours a week. We therefore place thedistinction between full-time and part-time employment at 35 hoursa week. The distinction between “small” and “substantial” part-timejobs is commonly drawn at jobs involving half a working week or day,reflecting the threshold for employment rights and involvement in theworkplace in many firms and countries and hence serving as indicator ofthe marginality of the part-time worker (Ellingsaeter, 1992; Rubery et al.,1999).

The propensity of one individual to be in a particular labor marketstate versus a reference category can be expressed by using relative risk

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 229


ratios. The relative risk ratio is calculated by taking the exponential ofthe log risk ratio. In this study, if we are interested in the characteristicsof the “risk” that women will find themselves in any state (m) other thanfull-time employment (=1), i.e. to be out of the labor force or work in asmall or substantial part-time job, the equation

log(

Pr(Yi = m)Pr(Yi = l)

)=

R∑r=ft

βmrχir = Zim, (m = 2, . . . , M)

expresses the logarithm of the risk-ratio of the probability of that out-come versus its alternative of full-time employment, where β is thecoefficient associated with the rth characteristic (r = 1, . . ., R) for the mthalternative pattern of labor market participation (m = 2,. . ., M), where Xis the value of the rth characteristic for individual i. Since we are inter-ested in analyzing whether part-time work is used as a stepping stoneto get a full-time job or as a pathway to exit the labor market, we willalso apply this model to a selected sample of part-time working womenwhose “risk” to move to full-time employment or exit from the laborforce is estimated relative to those remaining in a part-time job. We areespecially interested in the differences between small and substantialpart-time positions.

Finally, we will attempt to analyse women’s choices by taking intoaccount not only actual transitions but also preferences of working moreor less hours and how they affect women’s actual working hours. For thatpart of the analysis, we have used a fixed effects model with pooled datafor women who stay in employment.

7.3 Different pathways moving away from thebreadwinner state

In this section we delineate our expectations regarding the developmentof women’s part-time work based on what we know about economic,political and institutional differences between the three countries. Theseexpectations will help us orient our empirical analysis and interpretthe results in a meaningful way. Each of the three countries is mov-ing away from the breadwinner model and married women and mothersof young children have turned increasingly to paid employment. Part-time employment has played a big role in this development, but thenature of part-time work is quite different across the three countries. TheNetherlands, coming from a lower level, experienced the fastest growthin female employment of all OECD countries during the 1990s, and

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 230


50

52

54

56

58

60

62

64

66

68

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

EP

ratio

2

3

4

5

6

7

8

9

10

11

12

13

UE

rate

Germany EPratio Netherlands EPratio

United Kingdom EPratio Germany UErate

Netherlands UErate United Kingdom UErate

Figure 7.2 Employment population ratios and unemployment rates, women,aged 15–64 yearsSource: Employment in Europe, 2005 (based on European Labour Force Survey).

the biggest contribution came from women working part-time. Some-thing similar – accelerating growth in female employment associatedwith more women accepting part-time jobs – happened in Britain dur-ing the 1980s. But there the growth of part-time employment halted inthe 1990s, in contrast to the Netherlands, where part-time employmentcontinued its ascent. In Germany, the rise of part-time employment isa more recent phenomenon, mostly concentrated in (former) West Ger-many and situated in a context of overall stagnation of employmentgrowth and slowly growing female employment rates. Limiting ourselvesto the period 1992–2002, we observe that conditions – as indicated byfemale unemployment – seemed more benign in the UK and the Nether-lands compared to Germany (Figure 7.2). We expect that in Germany alarger share of women working part-time do so involuntarily; and thatthe gap between preferred and actual working hours is larger.

The main factor behind discrediting the male breadwinner model inthese countries is probably the stagnation in wage growth, the rise in(male and female) unemployment and increased job insecurity since the1980s (Western and Healy, 1999). This happened earlier in Britain than

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 231


in Germany, with the Netherlands in between. Moreover, the British“liberal” welfare state did less to compensate the stagnation in earn-ings and rise in unemployment of male breadwinners than the two“conservative-corporatist” welfare states, though the Netherlands startedto revise its tax system and social security programs towards greater “indi-vidualization” in various steps throughout the 1980s and 1990s.10 Inaddition to the “push” of married women into employment, there wasthe “pull” of rising relative female wages, related to increased partici-pation in education and women’s investment in human capital, as wellas female-biased technological change with the rise of services (Hartogand Theeuwes, 1985). The later and more limited change to a serviceeconomy in West and East Germany, combined with stricter legislationon business operating hours compared to the UK and the Netherlands,is likely to have dampened employer demand for part-time work andlimited opportunities for women wanting to work part-time.

Women’s decisions to seek part-time jobs, or to switch from a full-timeto a part-time job, are often understood as an adjustment of womenseeking to balance their roles as mothers and workers.11 Part-time workis then seen as a coping strategy for women at a life stage when fam-ily obligations are most comprehensive, i.e. when mothering youngchildren, especially during the pre-schooling years. Consequently, post-natalist state and employer policies and services are very important forthe employment patterns of women (Beechy and Perkins, 1987; Pettitand Hook, 2005). The provision of good quality public child care is prob-ably the single most important political measure enabling mothers tocontinue working and to choose full-time rather than part-time employ-ment. In this respect our three countries are very different from Denmarkor Sweden, and also from France and Belgium, countries where there isan existing tradition of day schools and kindergartens for children ofpre-schooling age (Daly, 2000). Until the 1990s, the UK, Germany andthe Netherlands had negligible coverage rates of child care for childrenless than four years of age, compared to 48 percent in Denmark (CECdata, cited in Elligsæter, 1992). In the Netherlands and Germany, mostday care centers only offer child care places on a part-time basis, requir-ing one or both parents to work part-time or flexible hours (Plantenga,2007). With regard to children aged three to compulsory schooling age,differences have narrowed.

According to Brannen and Moss (1990), policies directed at supportingthe double role of parents and mothers have been met by a mixture ofindifference and hostility by postwar UK governments. In keeping withthe philosophy of the liberal welfare state and the sanctity of markets,

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 232


the provision and financing of child care in the UK has remained a matterfor parents and employers. In the Netherlands and Germany, child carewas seen as a shared responsibility between families, firms, social part-ners and the government. Characteristically, in the Dutch “tripartite”approach, child care expenses have been a matter of subsidies, tax deduc-tions and exhortations to make employers pick up part of the bill. Thishas met with limited success: research in 2005 showed that one-third ofall employees receives no childcare support from employers (Plantenga,2007).12 In short, the limited availability of childcare provision, althougheasing over time, has remained a constraint on women wanting to workfull-time in all three countries. The high cost of child care probably hasthe strongest impact on the choices of women with low education and,we may assume, low earnings, and this effect should be largest in theUK, other conditions held equal.

The lack of sufficient and affordable care facilities and services hasled the governments in the three countries to enact legislation tofacilitate the change from full-time to part-time employment duringparenthood. The Netherlands has been a forerunner in this regard. Asearly as 1982, the Wassenaar agreement between unions and employ-ers recommended part-time work as a work-sharing solution to risingunemployment and in 1993, the peak federations of unions and employ-ers recommended that local negotiators include a “part-time clause” incollective agreements in order to make sure that employers “recognizea qualified right for full-time employees to work reduced hours, unlessthis cannot reasonably be granted on grounds of conflicting businessinterests.”13 At the same time, the conditions and benefits of part-time workers were brought on par with full-time workers, clauses whichdenied (small) part-time workers the benefits granted to full-time work-ers were removed from law and collective agreements, and in 1996,all remaining discrimination on the basis of working hours was madeunlawful (Yerkes, 2006). During the 1990s, the percentage of firms witha so-called “part-time clause,” allowing a reduction in the working hoursof full-time workers, increased from 23 to 70 percent. Yet, the unionscriticized that too few firms cooperated and in 2000, Parliament votedin a new law granting a statutory right to work part-time for all employ-ees (Visser et al., 2004). The same law also introduced a qualified rightfor part-time working employees to demand more hours.

It was not until 1996 that, faced by rising unemployment and stag-nating employment, the German government, inspired by the Dutchjob “miracle,” began to promote part-time employment (Killman andKlein, 1997). In its bid to reduce unemployment, the Center-Left, which

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 233


gained office in 1998, began to promote mini jobs, allowing employeesto earn up to 400 euros monthly, exempt from payroll taxes and socialsecurity. These jobs are mostly found in the service sector, part-timeand are typically performed by women. This development is likely tohave reinforced a negative image, associating part-time work with sub-standard jobs in terms of social security and employment protection inthe eyes of workers, unions and employers. Recently, and in responseto European legislation that “forbids” discrimination on the groundsof working hours (Directive 97/81/EC),14 there have been attempts tore-regulate “mini-jobs” and part-time work. In 2001, the new law onpart-time and fixed-term work came into effect – this gives employeesthe right to request a reduction in working time in companies with morethan 15 employees and places restrictions on the use of fixed-term con-tracts (Yerkes, 2006). German collective bargainers seem to have been lessprominent in their attempt to raise the profile of part-time work. Untilrecently, many collective agreements contained clauses against smallpart-time work. Eligibility thresholds, usually at around 15–18 hours perweek, tend to exclude small part-time workers from employment protec-tion or social insurance. Only in the case of part-time work of substantiallength are conditions similar to those of full-time workers.

The development of part-time work in the United Kingdom differsfrom the pattern observed both in Germany and the Netherlands mainlybecause it is shaped in a much less regulated labor market. The collectivestatutes that existed had been weakened significantly during the 1980sand relevant EU legislation did not apply until the Labour Party returnedto power in 1997, after 18 years of opposition. The expansion of part-time employment that did take place was market-driven and before 1998hardly constrained by legal norms. In the UK, part-time work is moremarginalized and organized around small jobs than in the other twocountries. Like part-time workers in mini-jobs in Germany, employeeswith low incomes are not covered by occupational social insurance, butthresholds are much higher than in Germany and were raised severaltimes.15

Working hours are much more polarized in the UK than on the Euro-pean continent and part-time work often consists of short hours forwomen and very long hours for their partners or husbands.16 British tradeunions often remain quite supportive of overtime and unsocial workinghours because their members need the extra pay (Barnard et al., 2003;Dickens and Hall, 2005). British governments, including “New Labour,”maintain a policy of allowing firms to choose their own working timearrangements, even where it contradicts avowed policy objectives such

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 234


as creating a better work–family balance and eradicating an unproductive“long hours” culture. As is argued by Berg and colleagues (2003), incountries with little labor market regulation, as in the UK, employershave more power to shape when and how long people work, and lev-els of employee autonomy are rather low. In the case of the UK, thisis reflected in higher levels of conflict and reported levels of stress overwork-family issues compared to Germany and, especially, the Nether-lands according to ISSP data from 2002. Fagan (2004) concurs that –apart from the length of working hours – it is the control workers haveover their individual working hours that is crucial. We expect, therefore,that the differences between actual and preferred working hours will belargest in the UK and smallest in the Netherlands.

The second pillar of a post-natalist state and employer policy is basedon guaranteeing that non-work issues such as parenthood and caringresponsibilities can be pursued within the employment relationship.Key to this is the granting of leave of absence rights relating to parent-hood, sickness or caring duties. European legislation introduced in 1995establishes minimum leave of absence rights related to maternity andparenthood, but entitlement to these rights, compensation for the lossof earnings, and the right to return to the same job or career, are issuesdealt with by national states. The UK leaves most of these issues to thefree play of negotiations between employers and employees (or unions,as the case may be), though under Labour this tendency was partiallyreversed. Each of the three countries now guarantees a 16–18 weeks baseparental leave at a 100 percent replacement rate (90 percent in the caseof the UK) and optional leave rights for up to 24 weeks in the case ofthe UK and the Netherlands, and up to 136 weeks in Germany (untilrecent legal changes, effective in 2007). Long-term leave arrangements,up to three years after childbirth, tend to encourage withdrawal fromthe labor market (Buddelmeyer et al., 2004; EC, 2004). In Germany weexpect therefore that the presence of young children will be associatedwith withdrawal while in Britain and the Netherlands there will be a ten-dency to work part-time, with a pronounced tendency to small part-timejobs in the UK.

In the 1990s, as part of its strategy to “normalize” part-time jobs(Visser, 2002) and propagate a combination-model of equal sharing ofwork and care (Plantenga, 2002), the Netherlands tried to develop achoice model, based on ample opportunities for both employees andemployers to choose individualized (part-time) working hours whileminimizing the inequalities deriving from such choices. Preceded byten years of experiment by means of collective bargaining between

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 235


unions and employers, covering nearly 80 percent of all employees,this development culminated in broad parliamentary support for leg-islation guaranteeing employees the right to request either an increaseor decrease in their working hours. The Working Hours Adjustment Act,passed in 2000, is quite unique in so far it gives employees an almostunilateral right to change contractual working hours. Its framers believedthat this would strengthen the resolve of men to demand shorter workinghours and of women to demand longer working hours (in line with con-sistent survey findings of the 1990s) and, hence, bring about a more equalgender division of work and family tasks. In 2001, Germany enacteda weaker version of this idea, giving employees a conditional right todemand part-time working hours. Under Labour’s “Fairness at Work”program, since 2003, the UK has granted parents with children under theage of six to ask for changes in working hours and request to be allowedto work at home. German and British legislation came too late to affectthe choices of women before 2002, the final year of our study. In theNetherlands we do expect the effect of collective agreements with part-time clauses to show in an alignment of actual and preferred workinghours among employed women in the second half of the 1990s.

7.4 The “breadwinner” determinants of part-timeemployment

In this section we analyze panel data for the period 1992–2002 in orderto identify the key determinants affecting women’s decisions to accepta part-time job. The probability of being “inactive” in the labor market,working full-time, a “small” or “substantial part-time job” is estimatedconditional on a vector of explanatory variables. Since we want to verifythe continuing strength of the “breadwinner” model regarding women’schoice of working pattern, “marital status” and “motherhood” are thetwo main variables of interest. The definition of marital status is basedupon the respondents’ own definition and unmarried women living in“couples” or civil partnership are classified as married women, since theyare in recent years treated almost identically in terms of income tax,family allowances, child care benefits and leave rights in the three coun-tries studied here. Single women are the reference category. Motherhoodis measured using two variables: the presence and age of children inthe household. We distinguish between children of pre-schooling age(0–5 years) and older children (6–15 years); the reference category is ahousehold without children.17 The control variables are level of educa-tion and birth cohort. For educational level we use the internationally

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 236


Table 7.2 Distribution of women aged 15–64 years by labor market status in1992–2002, by birth cohort

Birth No job Short part- Long Part- Full-timecohorts time job time job job (35 hours

(1–19 hours) (20–34 hours) or more)

Germany 1932–44 62.5 5.1 11.4 21.0 100.01945–54 26.8 9.7 21.1 42.3 100.01955–64 28.0 12.0 20.9 39.3 100.01965–74 36.3 8.0 10.5 45.2 100.01975–86 41.0 3.3 4.7 51.1 100.0

All 38.1 8.3 14.7 39.0 100.0

Netherlands 1932–44 75.0 9.0 11.2 4.9 100.01945–54 43.3 20.2 25.0 11.5 100.01955–64 31.8 23.1 31.2 14.0 100.01965–74 20.5 14.2 30.4 34.9 100.01975–86 5.1 6.8 35.4 52.7 100.0

All 40.1 16.9 25.5 17.5 100.0

UK 1932–44 55.4 12.1 15.3 17.2 100.01945–54 25.8 16.0 22.9 35.4 100.01955–64 25.5 17.0 21.8 35.8 100.01965–74 25.2 12.6 14.2 48.1 100.01975–86 5.7 9.0 13.0 59.9 100.0

All 30.2 14.1 18.2 37.6 100.0

Source: Germany: GSOEP; UK: BHPS; Netherlands: OSA LSP 1992–2002.

comparable CASMIN scale (Müller, 2005; Kerckhoff et al., 2002; Bernardiet al., 2004) to determine three levels of education: high (university),intermediate (secondary) or low (less than secondary), corresponding toCASMIN levels 1, 2 and 3. We use higher education as the referencecategory. Finally, we distinguish five birth cohorts: 1932–44; 1945–54; 1955–64; 1965–74 and 1975–86. Our reference category is the firstcohort – that is, women who entered the labor market when the bread-winner model reached its postwar apex, before the expansion of masseducation and before the general diffusion of oral contraceptives in thelate 1960s.

Table 7.2 shows that the largest differences in patterns of labor marketparticipation are those between the oldest and the other four cohorts.This also reflects an age effect, since labor market participation in nearlyall European countries decreases sharply after reaching 55,18 and the

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 237


women in this cohort were in their fifties and sixties between 1992 and2002.19 Remarkably, of the German women who do work in this agecohort, most are employed in full-time jobs, a pattern barely evident inthe Netherlands among women of this cohort. UK women in this cohortare just as likely to work part-time or full-time.

A full-time job is the dominant choice only for women in the youngestcohort in the Netherlands and the two youngest cohorts in the UK andGermany. These are women who entered the labor market in the 1980sor later, and it is likely that the working hours choice of the youngestwomen among them has not yet been compromised by the choice ofmotherhood.20 We observe that for these younger women the choice towork full-time dominates the part-time option, although in the Nether-lands this tendency is less pronounced and long part-time jobs appear tobe an alternative to full-time jobs even for these younger women. Finally,we note that part-time jobs are less common in Germany among womenof all ages and that, in sharp contrast to the other two countries, the “nojob” option is particularly widespread among German women in the twoyoungest birth cohorts.21 The data suggest that for them the dominantchoice is the one between a full-time job and no job, whereas in theNetherlands and in the UK, part-time jobs, often short-hours part-timejobs, have offered a third option. Considering the period of observation,that is, 1992–2002, it seems plausible to see this difference as reflectingthe poor labor market conditions, especially for women, in Germany,as well as the stronger “welfare-to-work” pressures building up in theseyears in both the UK and the Netherlands.22

Moving from description to analysis, we present the results of ourmultinomial logit model in Table 7.3. In all three countries there is a cleartendency of both motherhood and marital status to lower the probabil-ity that women work full-time and to increase the probability of womenremaining or becoming inactive. The effect of the presence of youngchildren is particularly pronounced and as a rule, motherhood tends tohave a stronger impact than marital status.23 Marital status effects arestrongest in the Netherlands, especially in respect of the expressed pref-erence for short-hours part-time jobs. They are least pronounced in theUK. The effects of education are also unequivocal: a lower level of educa-tion decreases the probability of full-time work and increases the risk ofinactivity. This tendency has become stronger in most countries duringthe period under investigation and is most pronounced in the UK. In theNetherlands and Germany intermediate-level education is no (longer) asignificant predictor of the choice between full-time and long part-timeworking hours.

9780230_202573_08_cha07.tex

5/6/200815:39

Page238

238Table 7.3A Determinants of labor market status of women – inactive (REF: FULL-TIME) (multinomial logits for probabilities)

Netherlands UK Germany

2002 1998 1992 2002 1998 1992 2002 1998 1992Main Main Main

Interaction Interaction InteractionRRR & (SE) RRR & (SE) RRR & (SE)

Marital status (ref: single)Married/cohabitating 2.520∗∗∗ 1.676∗ 2.219∗∗ 1.156∼ .940 .957 2.703∗∗∗ .671∗∗∗ .573∗∗∗

(.464) (.416) (.574) (.101) (.114) (.135) (.284) (.073) (.072)One or more children (ref: no children)0–5 years old 9.250∗∗∗ 1.125 .688 16.316∗∗∗ 1.226∼ 2.170∗ 20.733∗∗∗ 1.700∗ .581∗∗

(4.994) (.876) (.511) (1.983) (.220) (.442) (2.864) (.350) (.114)6–15 years old 4.144∗∗∗ 1.032 1.190 3.270∗∗∗ 1.328 1.248 3.160∗∗∗ .820 .683∗∗

(1.654) (.566) (.641) (.359) (.084) (.236) (.304) (.108) (.099)Educational level (ref: high)Medium 3.053∗∗∗ 1.275 .972 2.067∗∗∗ .780∗ .714∗ 1.540∗∗∗ .700∗ .694

(.610) (.359) (.313) (.183) (.097) (.110) (.167) (.122) (.143)Low 9.126∗∗∗ 1.437 .959 7.520∗∗∗ .723∗ .565∗∗∗ 3.069∗∗∗ .758 .859

(2.009) (.448) (.319) (.758) (.096) (.089) (.345) (.131) (.177)Cohort (ref: born 1932–1944)born 1945–1954 .109∗∗∗ 1.527 2.704∗ .123∗∗∗ 1.109 2.117∗∗∗ .081∗∗∗ 2.052∗∗∗ 3.843∗∗∗

(.045) (.735) (1.279) (.019) (.196) (.425) (.011) (.295) (.665)born 1955–1964 .040∗∗∗ 2.668∗ 3.215∗ .041∗∗∗ 1.586∗ 3.259∗∗∗ .037∗∗∗ 2.323∗∗∗ 5.805∗∗∗

(.167) (2.332) (1.532) (.007) (.321) (.740) (.005) (.384) (1.087)born 1965–1974 .037∗∗∗ 1.137 .772 .032∗∗∗ 1.488 2.957∗∗∗ .037∗∗∗ 2.697∗∗∗ 11.536∗∗∗

(.154) (.571) (.369) (.006) (.302) (.679) (.006) (.444) (2.293)born 1975–1986 0.007∗∗∗ .602 8.181∗∗∗ .024∗∗∗ 1.441 .535 .081∗∗∗ 3.172∗∗∗ 35.560∗∗∗

(.003) (.471) (7.320) (.004) (.350) (.565) (.012) (.613) (14.043)

Year effects (ref: 2002) .306∗ .166∗∗∗ .633∗ .210∗∗∗ .690∗ .221∗∗∗(.113) (.045) (.140) (0.052)

∼p < .10, ∗p < .05, ∗∗p < .01, ∗∗∗p < .001.Sources: British Household Panel Study, German Socio-Economic Panel, OSA Labour Supply Panel 1992–2002.

9780230_202573_08_cha07.tex

5/6/200815:39

Page239

239

Table 7.3B Determinants of labor market status of women – short part-time. 1–19 hrs (REF: FULL-TIME) (multinomial logits forprobabilities)


2002 1998 1992 2002 1998 1992 2002 1998 1992Main Main Main


Marital status (ref: single)Married/cohabitating 10.197∗∗∗ .493∗ .684 1.623∗∗∗ .986 1.567∗ 3.287∗∗∗ .688∗ .675

(2.720) (.167) (.257) (.172) (.152) (.290) (.400) (.128) (.153)One or more children (ref: no children)0–5 years old 10.342∗∗∗ .875 .422 10.687∗∗∗ .931 1.121 14.847∗∗∗ 1.313 .554∗

(5.580) (.682) (.326) (1.413) (.184) (.264) (2.550) (.364) (163)6-15 years old 6.107∗∗∗ .844 .809 4.137∗∗∗ 1.448∗ 1.278 5.914∗∗∗ .937 .549∗∗

−2337 (.457) (.437) (.490) (.261) (.248) (.712) (.182) (127)Educational level (ref: high)Medium 2.290∗∗∗ .735 .256∗∗∗ 1.432∗∗∗ .984 .989 1.143 .717 .508∗

(.535) (.206) (.084) (.145) (.143) (.170) (.154) (.173) (.150)Low 5.106∗∗∗ .861 .361∗∗ 2.892∗∗∗ 1.038 .807 1.712∗∗∗ .929 1.016

(1.153) (.276) (.122) (.343) (.168) (.147) (.247) (.227) (.299)Cohort (ref: born 1932–1944)born 1945–1954 .431 1.810 1.916 .126∗∗∗ 1.450∼ 1.408 .534∗∗ 1.303 1.113

(204) (1.007) (1.036) (.058) (.326) (.336) (.115) (.366) (.332)born 1955–1964 .366∗ 1.894 .887 .164∗∗∗ 1.544∼ 1.496 .329∗∗∗ 1.338 1.1072

(.171) (1.056) (.480) (.033) (.377) (.372) (.072) (.400) (.342)born 1965–1974 .252∗∗ .766 .415 .127∗∗∗ 1.365 .818 .248∗∗∗ 1.305 .982

(.118) (.431) (.232) (.025) (.336) (.230) (.057) (.413) (.375)born 1975–1986 .047∗∗∗ 2.631 4.940∗∗∗ .024∗∗∗ 1.487 4.191∗ .180∗∗∗ 1.289 3.165

(.027) (1.926) (4.770) (.018) (.475) (2.803) (.049) (.573) (3.435)Year effects (ref: 2002) 1.198 1.461 .590∗ .360∗∗∗ .690 .582

(.718) (.894) (.143) (.099) (.245) (.234)

∼p < .10, ∗p < .05, ∗∗p < .01, ∗∗∗p < .001.Source: British Household Panel Study, German Socio-Economic Panel, OSA Labour Supply Panel 1992–2002.

9780230_202573_08_cha07.tex

5/6/200815:39

Page240

240Table 7.3C Determinants of labor market status of women – long part-time, 20–34 hrs (REF: FULL-TIME)(multinomial logits for probabilities)


2002 1998 1992 2002 1998 1992 2002 1998 1992Main Main Main


Marital status (ref: single)Married/cohabitating 2.586∗∗∗ .972 1.264 1.540∗∗∗ .888 1.083 1.822∗∗∗ .845 .998

(.218) (.301) (.137) (.117) (.169) (.169) (.113) (.166)One or more children (ref: no children)0–5 years old 5.576∗∗ 1.196 .855 4.669∗∗∗ .977 .977 4.812∗∗∗ .906 1.107

(2.979) (.909) (.636) (.558) (.183) (.216) (.759) (.233) (.273)6–15 years old 2.882∗∗ 1.088 .942 2.810∗∗∗ 1.404∗ 1.488∗ 2.724∗∗∗ 1.1017 1.117

(1.093) (.576) (.507) (.280) (.221) (.258) (.268) (.148) (.194)Educational level (ref: high)Medium 1.192 1.006 .845 1.393∗∗∗ .907 .697∗ 1.202 .594∗∗ .560∗∗

(.191) (.235) (.237) (.118) (.114) (.107) (.135) (.105) (.121)Low 1.962∗∗ .865 .798 2.017∗∗∗ .911 .721∗ 1.467∗∗ .661∗∗ .716

(.380) (.239) (.238) (.216) (.135) (.119) (.176) (.112) (.158)Cohort (ref: born 1932–1944)born 1945–1954 .586 1.392 1.286 .589∗∗ .805 .963 .640∗∗ 1.177 .874

(.259) (.715) (.706) (.108) (.165) (.212) (109) (.226) (.196)born 1955–1964 .606 1.233 .582 .402∗∗∗ .705 .718 .461∗∗∗ .916 .752

(.264) (.634) (.294) (.075) (.155) (.166) (.080) (.194) (.181)born 1965–1974 .438∼ .840 .449 .258∗∗∗ .707 .648∼ .323∗∗∗ .804 .730

(.191) (.433) (.227) (.048) (.157) (.170) (.059) (.180) (.201)born 1975–1986 .243∗∗ 1.550 4.097 .166∗∗∗ .906 1.534 .139∗∗∗ .984 1.594

(.109) (.863) (4.210) (.034) (.247) (.915) (.030) (.323) (1.712)Year effects (ref: 2002) .653 .490 1.116 .699 1.194 .618

(.328) (.248) (.237) (.163) (.287) (.177)Pseudo R2 .163 .147 .156N 5853 13424 14777


9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 241


Is the breadwinner condition retreating? Controlling for the presenceof children, marital status and education, the cohort effects clearly sug-gest that in each successive cohort a position outside the labor forcehas become less common for women and also that the relative riskof part-time compared to full-time work is lower in the youngest agecohorts. However, during the 1990s, cohort effects concerning the choicebetween (long) part-time and full-time working hours are absent orweaker in the Netherlands compared to the other two countries, reflect-ing the rapid diffusion of part-time work. Interacting our main variableswith the results of earlier years, we may conclude that motherhood, espe-cially when it involves young children, has become less important as apredictor of inactivity in the UK, but stronger in Germany. In the UK,but not in the other countries, caring for older children is also less asso-ciated with the choice of part-time hours than in the recent past. InGermany, especially, the opposite development is found. Finally, onlyin the Netherlands has marital status, or living as a couple, become aless strong predictor of inactivity during the 1990s, again with signsof an opposite trend in Germany. However, in the Netherlands thisdevelopment seems to have involved a displacement of inactivity bysmall part-time jobs, possibly as a result of the new Benefit ClaimantAct of 1995, which required women claiming benefits and mothers ofyoung children to seek paid work on at least a part-time basis (Visser andHemerijck, 1997).

7.5 Part-time work as a coping strategy and bridgehead outof inactivity

Our next interest is in part-time employment as a way for women to fore-stall exit from the labor market and in part-time jobs as a stepping-stonetowards full-time jobs. For this purpose, we proceed with a multino-mial logit estimation based on selected samples of women in a particularlabor market state and analyze the relative risk of their transition intoother states within a two-year time span. For instance, when analyzingthe risk of inactivity of women in full-time jobs, our dependent vari-able will be a four-point variable that takes the value 0 if the individualremains in a full-time job, 1 if she becomes inactive, 2 if she shifts intoa small part-time job, 3 if she adjusts her hours into a long part-timejob. Taking respondents who remain employed on a full-time basis asthe omitted category, three sets of coefficients can be estimated. As inour earlier tables, we present them as relative risk ratios to make it easierto interpret the results.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 242


In Table 7.4A, we compare the impact of motherhood and maritalstatus upon the relative risks of a labor market exit for three differentpopulations: women with full-time jobs and women working in largeand small part-time jobs. As before, we keep education as a control vari-able and we control for birth cohort. However, to avoid problems withvery small Ns in transition analysis, we take the 1945–54 cohort, agedbetween 38 (youngest) and 57 (oldest) between 1992 and 2002 and thuswith a significant presence in the labor market, as our reference group.We are especially interested in the differences between that cohort, ofwomen who have their marital and motherhood choices behind them,and the two younger birth cohorts.

The main finding is that in each country motherhood stronglyincreases the risk of inactivity for full-time working women, no matterthe age of the child. The risk of a labor market exit associated with moth-erhood is smaller for women working part-time. The risk of a change intoinactivity still exists, especially in the case of young children, for womenworking in small and long part-time jobs, but has disappeared in theNetherlands. In other words, for Dutch women who work in part-timejobs, motherhood does not seem a significant determinant of discon-tinuing their labor market careers. In Germany and the UK, holding asmall part-time job lowers the risk associated with the presence of chil-dren, of exiting the labor market, but only when children are older and ofschool age. Of further interest is that in each country, but most clearly inthe Netherlands, being married, or living in couples, strongly increasesthe risk of a labor market exit for full-time working women, but notfor women working part-time. In fact, being married or living together,while increasing the probability that full-time working women exit fromthe labor market, lowers the risk that women with small part-time jobsstop working. For women working in long-hours part-time jobs we findno effect either way.

Lower levels of educational attainment strongly increase the risk ofleaving the labor market and nearly the same pattern is found amongwomen working in full-time or substantial part-time jobs. In Germany,an intermediate-level education appears to guarantee more continu-ity, whereas in the Netherlands and in the UK this is only the caseamong women working in small part-time jobs. Possibly, these are “everywomen’s jobs” in services, which require only a low level of qualifica-tion and no specific vocational training. Hence, educational attainmentappears to matter little for these jobs. Finally, against our expectations,younger cohorts of women in Germany and the UK are more likely thantheir mothers to exit the labor market, no matter what their prior status

9780230_202573_08_cha07.tex

5/6/200815:39

Page243

243

Table 7.4A Determinants of labor market transitions into inactivity (multinomial logits for probabilities)


From: full-time small PT long PT full-time small PT long PT full-time small PT long PTN = 1769 584 1027 11909 2955 351 13126 1838 3098

RRR & (SE) RRR & (SE) RRR & (SE)

Marital status (ref: single)Married/cohabitating 4.963∗∗∗ .299∗ .889 1.533∗∗∗ .636∗ .873 2.005∗∗∗ .506∗∗ .752

(1.376) (.170) (.349) (.177) (.119) (.160) (.174) (.114) (.130)One or more children(ref: no children)0–5 years old 13.490∗∗∗ 1.183 1.389 25.480∗∗∗ 2.057∗∗∗ 3.623∗∗∗ 50.715∗∗∗ 2.645∗∗∗ 12.733∗∗∗

(6.280) (.507) (.547) (2.655) (.405) (.638) (5.739) (.650) (2.979)6–15 years old 11.586∗∗∗ .862 .990 4.352∗∗∗ .679 .828 2.641∗∗∗ .491∗∗ .900

(4.361) (.343) (.426) (.616) (.151) (.166) (.237) (.114) (.173)Educational level (ref: high)Medium 2.368∗∗ 1.188 4.976∗ 1.975∗∗∗ 1.010 1.640∗∗ 1.247 1.337 1.277

(.628) (.819) (3.112) (.197) (.161) (.269) (.167) (.380) (.309)Low 8.484∗∗∗ 2.926 15.169∗∗∗ 4.874∗∗∗ 1.754∗∗ 3.059∗∗∗ 2.977∗∗∗ 1.172 1.572

(2.408) (1.896) (9.308) (.629) (.316) (.577) (.434) (.357) (.432)Cohort (ref: born 1945–1964)born 1965–1974 .554∗∗ 1.713 1.497 1.232 1.354∗ 1.683∗∗ 1.625∗∗∗ 1.963∗∗∗ 2.145∗∗∗

(.108) (.557) (.415) (.139) (.211) (.271) (.160) (.354) (.390)born 1975–1986 .241∗∗∗ 2.646 .455 1.663∗∗∗ .2.099∗∗ 3.059∗∗∗ 5.957∗∗∗ 9.420∗∗∗ 37.614∗∗∗

(.241) (2.528) (.359) (.237) (.573) (.755) (.724) (3.012) (11.187)


9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 244


of employment in terms of working hours. It is impossible to say whetherthis reflects choice (in combination with marriage and motherhood), ora lack of opportunity and fewer job security rights compared with theolder generation. In the Netherlands, instead, we see – at least for theyoungest birth cohort – a move in the opposite direction. They tend tohave a much lower risk of a transition into inactivity compared to womenborn two decades before them, especially when they held full-time jobsor part-time jobs of substantial length. We think that indeed part of theexplanation is historical and is related to the low levels of participation ofthe Dutch women in our reference category, not only much lower thanlater cohorts, but also much lower than British and German women ofthat generation (see De Graaf and Vermeulen, 1997).

The preceding analysis naturally leads to the next question: to whatextent is part-time employment a bridgehead out of inactivity? Again,we compare three types of movement: into full-time jobs, short and longpart-time jobs (Table 7.4B). The first thing to note is that marital statusand having children always diminishes the probability of moving out ofinactivity into a full-time job. Further analysis, not shown in Table 7.4B,reveals that this is also true for women working part-time. This barrierto full-time jobs for married women with children is rather similar in allthree countries and the fact that we still find signs of it in the 1990s istestimony of the strength of the breadwinner legacy.

Our analysis also shows that it is very important to make a distinctionbetween short- and long-hours part-time jobs. The latter behave morelike full-time jobs, while the barriers between employment and inactiv-ity, as well as the effects of marital status, children and education arerather slight in the case of small part-time jobs. The chances of changingfrom inactivity into small part-time jobs are even increasing for marriedor cohabitating women in the UK, compared to those who are single,and the same is true for women with older children, compared to child-less women, in both Germany and the UK. Lower levels of education,however, tend to depress women’s chances of changing from inactivityinto any job, no matter its size. The inhibiting effects of low educationare strongest in the UK and weakest in the Netherlands.

Finally, the youngest birth cohort has better chances of moving outof inactivity into full-time or substantial part-time jobs, except perhapsin Germany, a finding that probably reflects the depressed state of theGerman labor market during this period. Only in the youngest genera-tion and only in Germany, is the probability of making a transition frominactivity to small part-time jobs significantly lower than the probabilityof remaining inactive. This may indicate the undesirability of these jobs,

9780230_202573_08_cha07.tex

5/6/200815:39

Page245

245

Table 7.4B Determinants of labor market transitions to job from inactive status (multinomial logits for probabilities)


N = 1064 5804 7928

To small PT long PT Full-time small PT long PT Full-time small PT long PT Full-timeRRR & (SE) RRR & (SE) RRR & (SE)

Marital status (ref: single)Married/cohabitating .640 .277∗∗∗ .099∗∗∗ 1.372∗∗ .916 .570∗∗∗ 1.025 .689∗∗ .328∗∗∗

(.148) (.086) (.030) (.226) (.111) (.069) (.133) (.101) (.038)

One or more children(ref: no children)0–5 years old .783 .153∗∗ .076∗∗ .943 .449∗∗∗ .086∗∗∗ 1.142 .905 .249∗∗∗

(.240) (.090) (.081) (.133) (.062) (.012) (.161) (.141) (.031)

6–15 years old 1.369 .494∗ .332∗∗ 1.480∗∗ .702∗ .316∗∗∗ 1.421∗ 1.129 .564∗∗∗(.328) (.179) (.174) (.227) (.114) (.044) (.202) (.183) (.063)

Educational level (ref: high)medium .897 .634 .314∗∗ .760∗ .681∗∗ .537∗∗∗ .956 .686∗ .593∗∗∗

(.370) (.251) (.123) (.091) (.092) (.072) (.165) (.120) (.088)low .979 .393∗ .106∗∗∗ .422∗∗∗ .280∗∗∗ .154∗∗∗ .591∗∗ .381∗∗∗ .382∗∗∗

(.386) (.163) (.047) (.055) (.044) (.026) (.108) (.066) (.060)

Cohort (ref: born 1945–1964)born 1965–1974 .951 1.668∗ 3.882∗∗∗ 1.270∗ 1.203 1.650∗∗∗ 1.141 .998 1.421∗∗

(.218) (.422) (1.176) (.133) (.154) (.223) (.127) (.127) (.167)born 1975–1986 1.823 11.753∗∗∗ 37.020∗∗∗ 1.098 1.787∗∗ 3.329∗∗∗ .475∗∗∗ (.381)∗∗∗ 1.313

(1.665) (.7336) (25.141) (.196) (.331) (.590) (.080) (.076) (.181)

∼p < .10, ∗p < .05, ∗∗p < .01, ∗∗∗p < .001.Sources: British Household Panel Study, German Socio-Economic Panel, OSA Labour Supply Panel 1992–2002.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 246


though it is more plausible to hypothesize that, compared to British andDutch women, German women faced fewer pressures to exit inactivity.

7.6 The preferences of employed women

To what extent do women realize their preferences with regard topaid employment? In the previous sections we saw that the choicesof women are still strongly related to motherhood and marital status.We may indeed be struck by the similarity between the three countriesand the continued strength of the breadwinner legacy. Comparison ofthese rather similar countries also allowed us to make some fine dis-tinctions – showing how part-time work in the Netherlands, far fromoffering women an assured path to careers and economic independence,nonetheless may help combine different demands on women’s time andreduce the risk of inactivity. It would seem plausible to see this as a resultof a choice model based on increased opportunities for both employeesand employers to choose individualized (part-time) working hours, withwhich the Netherlands experimented during the 1990s before it becamediffused to the other two countries.

Since 2000, as part of the European Employment Strategy (EES) and asan effect of the Part-Time Directive of the European Union, discrimina-tion on the basis of working hours has been outlawed and the right ofparents to switch from full-time to part-time hours has been establishedas an optional right in all three countries. Before the statutory changes of2000, such choice rights did exist as part of collective agreements in theNetherlands and perhaps on an individual basis in all three countries.It was, of course, always possible to change working hours by changingemployer, but making this a choice within continuing employment rela-tionships lowered the costs of such choices. Finally, in order to improvethe status of part-time employment and enhance the quality of work,the EES also promotes the conditional right of women and parents toincrease working hours, or switch back from part-time to full-time work,business conditions permitting. The Dutch law of 2000 contains a condi-tional provision to this effect (Visser et al., 2004). The other two countrieshave moved in the same direction, but on a more limited base and theydid not have the long period of experimentation and learning of theNetherlands.

In the previous section we saw that transitions from part-time to full-time employment – and vice versa – are strongly associated with andconditioned by motherhood, marital status and educational attainment.Our next step is to try and relate these transitions to working preferences

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 247


Table 7.5 Preferred and actual working hours of women with children under six,living in a couple, 1998

Full-time job Part-time job No job

Preferred Actual Gap Preferred Actual Gap Preferred Actual Gap

Finland 80 49 31 9 6 2 11 44 −33Sweden 67 51 15 22 13 9 11 36 −25

Austria 36 19 17 40 28 12 24 53 −28Belgium 55 46 9 29 19 10 16 35 −18France 52 39 14 22 14 8 26 47 −21Germany 32 16 16 43 23 20 25 61 −36Netherlands 6 5 1 70 55 15 24 40 −16

Italy 50 35 16 28 12 16 22 53 −31Spain 60 26 34 12 6 6 29 68 −39Portugal 84 75 10 8 5 3 8 21 −13Greece 66 42 23 11 8 3 24 50 −26

UK 21 25 −4 42 32 10 37 43 −6Ireland 31 31 0 42 19 24 27 50 −24

Source: Calculated from Employment Options for the Future survey of the European Foundationfor the Improvement of Living and Working Conditions.

research conducted by the European Foundation for the Improvementof Living and Working Conditions. It compares the preferred and actualworking hours of women who have one or more child(ren) under the ageof six and live in a couple (married or cohabitating), i.e. the subcategorythat is the focus of our investigation throughout this chapter.

The survey reveals, firstly, that on average some 20–25 percent of thesewomen in Europe prefer not to work or seek paid employment, with avariation that runs from 11 percent in Scandinavia to 29 percent in Spainand 39 percent in the UK. One might interpret this as an indication ofthe “burden” of motherhood, in Spain worsened by the lack of publicchild care and in the UK by the tendency of men to work long hoursand the absence of state and employer support. Note that these figuresrefer to 1998, before New Labour’s moderate attempts at reconciling thedemands of work and family, discussed in section 7.2.

Secondly, we note that there is a huge gap between actual and pre-ferred working hours among women with young children and livingwith a partner: only a third to one-half of the women who do not havea job actually prefer not to have one. They are jobless either because of

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 248


Table 7.6 Preferences for more or less working hours and chances to realizepreferences, 1992–2002

Preferences (at t-2) Change in desired direction (at t)

Netherlands UK Germany Netherlands UK Germany% odds ratios∗

Less hours 20.2 30.8 44.7 2.2 1.6 1.2Same hours 66.4 61.1 30.2 – – –More hours 13.4 8.2 25.1 4.2 2.8 2.5

∗ compared to those preferring the same hours.Source: British Household Panel Study, German Socio-Economic Panel, OSA Labour SupplyPanel 1992–2002.

failing labor demand or because available jobs do not match their timepreferences (that is, the number of hours of work and the specific time ofwork) and other (unobserved) qualities of female labor supply. In 1998,most European countries suffered from high unemployment and femaleunemployment rates were generally much higher than male unemploy-ment rates, the UK being an exception. Unsurprisingly, the “no job”syndrome among these mothers is strongest in Germany, next to Spain,whereas it is lowest in Britain and also quite low in the Netherlands,probably thanks to strong female job growth in these years (Visser andHemerijck, 1997).

Thirdly, there is a considerable shortage of jobs, but also a largemismatch in the working hours that these women want to work andthe working hours they actually work when they do have a paid job.The cross-national variation in the preferences for part-time or full-timehours is striking, with the Netherlands leading the pack in favor of part-time hours, at some distance followed by the UK and Germany, Irelandand Austria – all with “breadwinner” traditions (Daly, 2000). The factthat, according to this survey, only 6 percent of Dutch mothers withyoung children and living with a partner wanted to work full-time andonly 5 percent actually work full-time, is quite remarkable. In Britain, sec-ond to the Netherlands in this regard, a quarter of these mothers workedfull-time, though fewer said that they preferred to work full-time. In theother countries, including Germany, many more mothers would haveliked to work full-time than was actually the case. This would suggestthat for some of them, part-time work is involuntary.24 Part-time jobs,too, are in short supply, and again, especially in Germany, according to

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 249


the Federal Agency of Employment (IAB), there were many more womenwho would have preferred to work part-time in the 1990s had these jobsbeen available (IAB, 1999).

In order to sort out whether preferences for more or less hours lead toadjustments in the desired direction, we now return to the three paneldata sets used in this study. Women were asked to consider their cur-rent weekly working hours and whether they would prefer a workweekof a different length. Unfortunately, the BHPS questions allow only acategorical answer of more, fewer or the same hours, rather than indi-cating the actual number of hours per week that respondents want towork. This limits our possibilities and we have therefore used the simplecategorization in the British study for the other two countries as well.

To begin with, Dutch and British women tend to be much more con-tent with their existing working hours than German women. Only 33.6percent of Dutch and 39.0 percent of British women want to changetheir working hours, whereas 69.0 percent of German women expressa wish to do so. The direction of change is towards fewer hours in allthree countries, though especially in Germany there is also a large groupof women, one-quarter of the total, which would prefer to work longerhours. Unsurprisingly, in each country the preference for fewer hours isstrongest among women working full-time and the preference for morehours strongest among women working in short-hours part-time jobs.However, it is remarkable that three-quarters of the Dutch and British,but only one-quarter of the German women working in jobs of less than20 hours per week do not want a change in hours. It is also interestingthat only 27.2 percent of full-time working women in Germany expresssatisfaction with existing working hours, compared to nearly 68.8 per-cent in the Netherlands and 52.1 percent in the UK, and that as many as25.7 percent of these German women want to work longer, a proportionthat drops to 0.6 percent in the Netherlands and 1.6 percent in the UK.This might be an indication of economic hardship (perhaps related toconditions in Eastern Germany) or reflect dissatisfaction with workinghours reduction for full-time workers set by collective agreement as partof the campaign for the 35 hours workweek of German trade unions.Reflecting the opposite conditions – long working hours and few legal orcontractual limits – British women, especially when working full-time,but also those working in large part-time jobs, tend to want fewer hours.

Our panel data allow us to check whether or not women have realized achange in working hours in the desired direction two years later. We canexpress this in terms of either conditional probabilities or odds ratios,comparing the changes in the working hours of women who expressed

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 250


a desire to make that change with women who did not want a change.This leads to two conclusions.

First, it appears that women who want to work more hours are moreeffective in realizing their preferences than women who want to workless. The odds ratios are nearly twice as high. It needs to be added thatwomen who want to work more hours are much less numerous, espe-cially in the UK and the Netherlands, than women who want fewerhours. Moreover, women who want to change to fewer hours becauseof motherhood more often leave employment. If we control for mother-hood, marital status, education and age, and apply a pooled regressionover all waves of the panel, with fixed effects, we find that a preferencefor more hours leads to a small, but significant increase in working hoursin all three countries, but that the preference for less hours has no sig-nificant effect, or, in the case of Germany, even an opposite effect. Thisis largely explained by the fact that a large number of mothers wantingto work fewer hours (temporarily) leave employment.

Secondly, both women who want to work more and women who wantto work less hours stand a better chance of realizing their wishes in theNetherlands than in Germany, with the UK somewhere in between. Thismay simply reflect economic opportunity – there was much more slackin the labor market for women in Germany than in the other two coun-tries – but it would seem plausible to argue that institutional differencesexplain the differences between the Netherlands and the UK.

We finish this chapter with a brief analysis of the relative risk ratiosof women preferring more or less hours for two groups of women, thoseworking full-time and those in small part-time jobs, relative to women inpart-time jobs of considerable length (Table 7.7). Our assumption is thatthe latter group (our omitted category) has best realized its preferenceand that our questions of interest – how are preferences influenced bymotherhood and marital status, controlling for education – concern theother two groups.25 It appears that, especially in Germany, marriage orliving as a couple reduces the probability to want more working hours.We also note that in the Netherlands women with young children and inBritain women with the lowest level of educational attainment prefer towork more hours, all other things held constant. With regard to the wishto work fewer hours, this is most clearly expressed by women workingfull-time, compared to those in part-time jobs of substantial length, andonly in the case of Germany is this very clearly related to motherhood.This suggests that in the Netherlands and the UK women with youngchildren have already, to a much greater extent, realized their preferencefor part-time work, though our analysis suggests that many, especially

9780230_202573_08_cha07.tex

5/6/200815:39

Page251

251

Table 7.7 Determinants of women’s preferences for more or less working hours (multinomial logits for probabilities)

More hours Fewer hours

Germany UK Netherlands Germany UK Netherlands

Current working hours (ref: long PT)Small part-time 2.094∗∗∗ 1.670∗∗ 1.797∗∗ 1.220 .544∗∗∗ .470∗∗

(.239) (.262) (.346) (.176) (.062) (.105)Full-time .809∗ .727∗ .964 2.954∗∗∗ 2.012∗∗∗ 1.365∗

(.069) (.109) (.174) (.269) (.156) (.184)Marital status (ref: single)Married/cohabitating .646∗∗∗ .641∗ .779 1.033 .958 .796

Children in household (ref: no children) (.068) (.117) (.236) (.107) (.087) (.169)0–5 years old .814 .832 1.875∗ .437∗∗∗ 1.226 .911

(.152) (.214) (.700) (.098) (.141) (.301)6–15 years old .793∗ .865 1.675 .764∗ .863 .641

(.088) (.183) (.549) (.088) (.090) (.222)Educational level (ref: high)Medium .856 .783 .830 .736∗ .701∗∗∗ .550∗∗

(.110) (.142) (.234) (.093) (.059) (.105)Low .840 1.515∗ 1.457 .634 .720∗∗ .472∗∗

(.117) (.304) (.420) (.088) (.080) (.105)


9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 252


those with the lowest levels of education, regret to hold steadily ontoonly a small part-time job. Marital status does not seem to matter, buta lower level of education, which, as we saw before, is strongly associ-ated with working in small part-time jobs, does decrease the preferencefor shorter hours, especially in the Netherlands and the UK. Combinedwith the finding that in both countries women in short-hours part-timejobs, compared to those in longer part-time jobs, clearly want no furtherreduction of their working hours, this is an indicator of dissatisfaction,perhaps economic hardship, lack of independence or full realization ofone’s possibilities.

7.7 Conclusions and further work

We began our analysis by considering women’s labor market participa-tion patterns in a context of individual constraints. Given the variationin institutional contexts and developments in employment prior to 1992described in section 7.1, we expected cross-country variation in thissmall-N analysis to be evident. Data for the Netherlands, Germany, andthe UK demonstrate various points of convergence as well as divergence.Our main variables of interest are motherhood and marital status, as indi-cators of the breadwinner legacy in these countries. They show in eachcountry a clear negative effect on women’s labor market participationpatterns, in terms of both the chances of holding a job and the length ofthat job. Motherhood also offers an example of divergence, where Ger-man mothers are most strongly affected by having a young child underthe age of three in the household and the number of children mattersmore in the other countries (recent legal changes may be expected tomove Germany to more convergence). Marital status effects, irrespec-tive of the presence of children in the household, are more in evidencein the Netherlands and Germany than in the UK. At the same time, alack of (continued and tertiary) education leads to a decreased probabil-ity of employment – and of full-time employment in particular – in allthree countries, and this effect grew stronger across time, most strikinglyin the UK. Finally, cohort effects are present in each country, showinga decreased probability of inactivity for younger cohorts, especially inthe Netherlands. Finally, the difference between short- and long-hourspart-time jobs is most clearly visible when analyzing the transition ofwomen out of a prior state of inactivity. While long part-time jobs seemto present the same barriers to women with children than full-time jobs,short-hours part-time jobs seem to present fewer obstacles. This does not,

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 253


however, imply that women do not try to avoid these mini-jobs if theycan (Germany) or will be satisfied with these jobs as they are (UK, theNetherlands).

The effect of the conditions of motherhood and marital status changeslittle after taking into account individual desires in the form of pref-erences. After controlling for individual preferences, the conditions ofmotherhood, marital status and education are the strongest determiningfactors of women’s employment patterns. Holding the other conditionsconstant, we find a clear trend that a majority of mothers want to reducetheir weekly working hours across time. Our analysis of the transitionsand preferences, the issue of the two final sections of this chapter, sug-gests that Dutch women are more able or likely to realize their preferencesthan women in the other two countries, with German women in a lessenviable position than those in the UK. But Dutch and British womenwho have ended up in short-hours part-time jobs do want more, even orrather especially when they have little education and (we may speculatewith confidence) low wages.

In conclusion, women are heterogeneous according to conditions ofmotherhood, marital status, and education. Their preferences tend to betied, if not determined, by these differences, but that does not mean thatthey do not matter. The comparison between these three rather similarcountries tells us that women’s preferences stand a better chance of beingrealized when labor market conditions are not too constraining and labormarket institutions are facilitating – a situation that, during the 1990s,was closer to being realized in the Netherlands than in Germany, withthe UK in between. Still, many women find themselves in conditionsthat are far from what they would desire ideally.

Notes

1. The high incidence of part-time employment among men reported in somecountries (the Netherlands, Australia, Denmark, to mention the three withthe highest incidence in 2006 according to the OECD) is mostly due to part-time working high school and university students. We must therefore becautious to interpret these figures as indicating a shift in the traditional gen-der division of household tasks and paid employment in families with youngchildren.

2. The extent to which low employment rates in Greece, Spain and (Southern)Italy reflect the existence of a large informal labor market and family-basedactivities is an issue we cannot deal with here.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 254


3. Norway, the third Scandinavian welfare state, suggests a different pattern,closer to that of mainland Western Europe (Ellingsæter, 1992).

4. In this study we use 35 hours per week as the cut-off point between part-time work (1–34 hours per week) and full-time work (35 and more). Thischoice is arbitrary but common, since 35 or 36 hours is the lowest numberof weekly hours set for full-time workers by collective agreement or law inEurope (EIROnline, Working Hours in Europe, 2003). The OECD proposes 30hours as the cut-off point on the ground that shift workers sometimes work32–4 hours a week and might erroneously be counted as part-timers (Lemaitreet al., 1997), but this nearly exclusively affects male employment and istherefore not relevant for this study, which focuses on female employmentpatterns.

5. While the importance of welfare state interventions for female employmentis generally acknowledged, the role of industrial relations is less commonlyrecognized. For early studies which did stress the importance of both sets ofinstitutions: see Beechy and Perkins, 1987; Ellingsæter, 1992).

6. In the former German Democratic Republic, both men and women wereexpected, and needed, to be in paid employment, but reunification after 1989meant that West German institutions were applied in the East, and work-ing patterns of women have converged since, albeit with some significantdifferences (Rosenfeld, Trappe and Gornick, 2004).

7. The data have been made available to us by the Institute for Social and Eco-nomic Research in Colchester (BHPS); the German Institute for EconomicResearch (DIW) in Berlin (GSOEP); and the Organisation for Strategic LabourMarket Research at the University of Tilburg (OSA-LSP).

8. This has forced us to calculate transition rates on a two-year basis for Germanyand also the UK.

9. Since we are interested in analyzing choice, we do not consider “unem-ployment” as one of the possible labor market “states” women “choose” tobe in.

10. The 1990 tax reform reduced the basic tax allowance for breadwinners andpeople with care responsibilities, and integrated social security contributions,thus lowering disincentives for second earners to take up more hours. The2001 tax reform removed the remaining shared taxation components forwage earners (though not for all benefits).

11. The Danish case reminds us that part-time work may also be a lifestyle choice,unconnected to motherhood.

12. Meanwhile, the government has introduced legislation making employers’contributions obligatory. In Germany, too, recent legislation (2006) hasmoved the system closer to the Scandinavian model.

13. Foundation of Labour (Stichting van de Arbeid), Overwegingen en aanbevelin-gen ter beoordeling van deeltijdarbeid en differentiatie van arbeidsduurpatronen,The Hague, 1993.

14. This Directive implemented a Framework Agreement between the Europeanunions and employers and had to be transposed in national law by 2000(Sciarra et al., 2005).

15. The first attempt to introduce European-wide legislation in the EuropeanUnion providing income protection for part-time workers goes back to 1982,proposing a rather low but common threshold, but defending the notion

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 255


that part-time workers should in principle not be excluded from occupa-tional social insurance schemes. Acceptance of the 1982 Voluntary Part-timeDirective was, at the time, subject to unanimity among EU Member States.The Directive was in effect vetoed by the British government (Rubery andTarling, 1988; Hakim, 1989).

16. According to the European Labour Force Survey of 2001 (Eurostat, Luxembourg),about one-third of all men in the UK and close to 10 percent of all womenworked 48 hours (the EU maximum) or more; in Germany the percentagesdrop to 15 and 5 percent, in the Netherlands to below 10 and 5 percent.

17. We have also experimented with a further distinction between the very young(0–2 years) and young (3–5 years old), and with the number of children,distinguishing between 1, 2 and 3 or more child(ren) in the household. Theresults do not differ from those shown here and are available from the authorsupon request.

18. Our three countries are remarkably similar in this respect. According to theEuropean Labour Force Survey, in 2002 the employment rate of women aged25–54 was 71.6 percent in Germany and 73.6 percent in both the UK and theNetherlands, dropping to just around 30 percent in the 55–64 age group inall three countries.

19. Note, however, that only women under the “working age” of 65 are includedin our sample.

20. In 1995, the average age at which women have their first child had increasedto 29 years in all three countries.

21. We recall that students have been excluded from our sample.22. The Harz IV reforms in Germany might change all that, but these reforms

only started to be felt in 2004 and beyond.23. Further analysis shows that labor market choices of German mothers are,

as expected, particularly affected by having a young child under the age ofthree, greatly increasing the risk of inactivity, whereas in the other two coun-tries the number of children in the household matters most (available fromauthors).

24. The European Labor Force Survey of 2005 does include a question about the“reasons why you work part-time” with pre-coded response categories thatare not really mutually exclusive: “care,” “education,” “disability,” “could notfind full-time job,” “did not want full-time job,” and “other.” Non-responseon this item is very high (around 50 percent). German women working part-time respond three times as often “could not find full-time job” than Britishwomen and about four times as often as Dutch women. Oddly, less than 3percent of Dutch women working part-time mention “care” as a reason forworking part-time, against a quarter of German and a fifth of British women.Half of Dutch (compared with 6.4 percent of German and 7.2 percent ofBritish) women working part-time mark the answer “did not want full-timejob” and this is perhaps how they see it.

25. In this case, controlling for birth cohorts does not reveal any extra informa-tion, except that in younger cohorts both the wish to work more and thewish to work less is more pronounced. The explanation might be that olderwomen have had more time to match their working time preferences to jobsof different length, with many of them disappearing from the labor market.This analysis is available from the authors.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 256


References

Beechy, V. and T. Perkins. 1987. A Matter of Hours: Women, Part-time Work,and the Labour Market. Cambridge and Oxford: Polity Press and BasilBlackwell.

Berg, P., E. Appelbaum, T. Bailey, and A.L. Kalleberg. 2003. “Contesting Time:International Comparisons of Employee Control of Working Time.” Industrialand Labor Relations Review 56: 331–49.

Bernardi, F., M. Gangl, and H.G. van de Werfhorst. 2004. “The From School-to-Work Dynamics: Timing of Work and Quality of Work in Italy, the Netherlands,and the United States, 1980–1998.” Working Paper 2004/201. Madrid: InstitutoJuan March de Estudios e Investigaciones.

Bielenski, H., G. Bosch, and A. Wagner. 2002. “Working Time Preferences in Six-teen European Countries.” Working Paper. Dublin: European Foundation forthe Improvement of Living and Working Conditions.

Blossfeld, H.-P. and C. Hakim. 1997. “Introduction: A Comparative Perspective onPart-Time Work.” Pp. 1–21 in Between Equalization and Marginalization: WomenWorking Part-Time in Europe and the United States of America, edited by H. P.Blossfeld and C. Hakim. Oxford: Oxford University Press.

Blossfeld, H.-P. and G. Rohwer. 1997. “Part-Time Work in West-Germany.” Pp.164–90 in Between Equalization and Marginalization: Women Working Part-Timein Europe and the United States of America, edited by H. P. Blossfeld and C. Hakim.Oxford: Oxford University Press.

Bollé, P. 2001. “Part-time Work: Solution or Trap?” Pp. 215–38 in Women, Gender,and Work: What Is Equality and How Do We Get There?, edited by M.F. Loutfi.Geneva: International Labour Office.

Brannen, J. and P. Moss. 1990. Managing Mothers: Dual Earner Households afterMaternity Leave. London: Unwin Hyman.

Buddelmeyer, H., G. Mourre, and M. Ward. 2004. “The Determinants of Part-time Work in EU Countries: Empirical Investigations with Macro-Panel Data.”Economic Papers, no. 213. Brussels: European Commission, Directorate-Generalfor Economic and Financial Affairs.

Calmfors, L., A. Booth, M. Burda, D. Checci, R. Naylor, and J. Visser. 2001. “TheFuture of Collective Bargaining in Europe.” Pp. 1–156 in The Role of Unions in theTwenty-First Century, edited by T. Boeri, A. Brugiavini, and L. Calmfors. Oxford:Oxford University Press.

Daly, M. 2000. “A Fine Balance: Women’s Labour Market Participation in Inter-national Comparison.” Pp. 467–511 in Welfare and Work in the Open Economy.Volume II: Diverse Responses to Common Challenges, edited by F.W. Scharpf andV.A. Smith. Oxford: Oxford University Press.

Dickens, L. and M. Hall. 2005. “The Impact of Employment Legislation: Reviewingthe research.” London: Department of Trade and Industry, Employment RelationsResearch Series, no. 45: 7–72.

Dingeldey, I. 2001. “European Tax Systems and their Impact on Family Employ-ment Patterns.” Journal of Social Policy 30: 653–72.

European Commission (EC). 2004. Industrial Relations in Europe in 2004. Direc-torate General of Social Affairs and Employment. Luxembourg: Office of OfficialPublications of the European Community.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 257


Ellingsaeter, A. 1992. Part-time Work in European Welfare States: Denmark, Ger-many, Norway and the United Kingdom Compared. Oslo: Institute for SocialResearch.

Esping-Andersen, G. 1990. The Three Worlds of Welfare Capitalism. Cambridge:Polity Press.

Fagan, C. 2004. “Gender and Working Time in Industrialized Countries.” In Work-ing Time and Workers’ Preferences in Industrialized Countries: Finding the Balance,edited by J. Messenger. London: Routledge.

Fagan, C. 2000. Employment Options of the Future: Actual and Preferred Work-ing Hours. National Working Paper for the United Kingdom. Dublin: EuropeanFoundation for the Improvement of Living and Working Conditions.

Gustafson, S., E. Kenjoh, and C. Wetzels. 2001. “Does Part-Time and IntermittentWork during Early Motherhood Lead to Regular Work Later? A Comparison ofLabor Market Behavior of Mothers with Young Children in Germany, Britain,the Netherlands, and Sweden.” Vierteljahrshefte zur Wirtschaftsforschung 70(1):15–23.

Hakim, C. 1997. “A Sociological Perspective on Part-Time Work.” Pp. 22–70 inBetween Equalization and Marginalization: Women Working Part-Time in Europeand the United States of America, edited by H.-P. Blossfeld and C. Hakim. Oxford:Oxford University Press.

Hall, P. and D. Soskice, eds. 2001. Varieties of Capitalism. New York: OxfordUniversity Press.

Halpin, B. 2004.Introduction to the British Household Panel Study. Course Notes.University of Essex, Colchester, England.

Hartog, J. and J. Theeuwes. 1985. “The Emergence of the Working Wife inHolland.” Journal of Labor Economics 3: 235–56.

Hassel, A. 2006. “What Does Business Want? Labour Market Reforms in CMEs andIts Problems.” Pp. 223–52 in Beyond Varieties of Capitalism, edited by B. Hancké,M. Rhodes, and M. Thatcher. Oxford: Oxford University Press.

IAB (Institut für Arbeitsmarkt- und Berufsforschung). 1999. “Teilzeitarbeit.”Mitteilungen aus der Arbeitsmarkt- und Berufsforschung 1: 5–30.

Kerckhoff, A.C., E.D. Ezell, and J.S. Brown. 2002. “Toward an Improved Measure ofEducational Attainment in Social Stratification Research.” Social Science Research31: 99–123.

Kersley, B., C. Alpin, J. Forth, A. Bryson, H. Bewley, G. Dix, and S. Oxenbridge.2006. Inside the Workplace: Findings from the 2004 Workplace Employment RelationsSurvey. London: Routledge.

Killmann, C. and M. Klein. 1997. “Part-Time Work in Germany: Gender-SpecificStructures of Working Hours.” Pp. 81–94 in Part-time Work in Europe: Gender,Jobs, and Opportunities, edited by M. Klein. Frankfurt/Main: Campus Verlag.

Klein, M., ed. 1997. Part-Time Work in Europe: Gender, Jobs, and Opportunities.Frankfurt/Main: Campus Verlag.

Knijn, T. 1994. “Social Dilemmas in Images of Motherhood in the Netherlands.”European Journal of Women’s Studies 1: 183–206.

Lemaitre, G., P. Marianna, and A. Van Bastelaer. 1997. “International Compar-isons of Part-Time Work.” OECD Economic Studies 29: 139–52.

Lewis, J. 1992. “Gender and the Development of Welfare Regimes.” Journal ofEuropean Social Policy 2: 159–73.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 258


Lippe, T. van der and L. van Dijk, eds. 2001. Women’s Employment in ComparativePerspective. New York: Aldine de Gruyter.

Milward, N., A. Bryson, and J. Forth. 2000. All Change at Work: British Employ-ment Relations 1980–1998, as Portrayed by the Workplace Industrial Relations SurveySeries. London: Routledge.

Müller, W. 2005. Casmin Education Classification. www.nuff.ox.ac.uk/Users/Yaish/NPSM/Casmin%20Education.htm. Accessed 10 February 2005, Oxford,Internet.

O’Reilly, J. and C. Fagan, eds. 1998. Part-Time Prospects: An International Compar-ison of Part-Time Work in Europe, North America, and the Pacific Rim. London:Routledge.

Orloff, A. S. 1993. “Gender and the Social Rights of Citizenship: The ComparativeAnalysis of Gender Relations and Welfare States.” American Sociological Review58: 303–28.

Pfau-Effinger, B. 1998. “Culture or Structure as Explanations for Differences inPart-Time Work in Germany, Finland, and the Netherlands?” Pp. 177–98 inPart-Time Prospects: An International Comparison of Part-Time Work in Europe,North America, and the Pacific Rim, edited by J. O’Reilly and C. Fagan. London:Routledge.

Petit, B. and J. Hook. 2005. “The Structure of Women’s Employment in Compar-ative Perspective.” Social Forces 84: 779–800.

Plantenga, J. 2002. “Combining Work and Care in the Polder Model: AnAssessment of the Dutch Part-Time Strategy.” Critical Social Policy 22: 53–71.

Plantenga, J. 2007. “Arbeid en zorg in een participatiemaatschappij.” Beleid enMaatschappij 34: 47–56.

Rasmussen, E., J. Lind, and J. Visser. 2005. “Divergence in Part-Time Work in NewZealand, the Netherlands, and Denmark.” British Journal of Industrial Relations42: 637–58.

Rogowski, R. and A.C. Wilthagen. 2002. “Legal Regulation of Transitional LabourMarkets.” Pp. 233–73 in The Dynamics of Full Employment: Social Integra-tion through Transitional Labour Markets, edited by B. Gazier and G. Schmid.Cheltenham: Edward Elgar.

Rubery, J., M. Smith, C. Fagan, and D. Grimshaw. 1998. Women and EuropeanEmployment. London: Routledge.

Schmid, G. and B. Gazier, eds. 2002. The Dynamics of Full Employment: SocialIntegration through Transitional Labour Markets. Cheltenham: Edward Elgar.

Sciarra, S., P. Davies, and M. Freedland, eds. 2005. Employment Policy and the Regula-tion of Part-time Work in the European Union: A Comparative Analysis. Cambridge:Cambridge University Press.

Taskforce on Employment. 2003. Jobs, Jobs, Jobs – Creating More Employment inEurope. Report of the Employment Taskforce, chaired by Wim Kok. Brussels.

Tijdens, K. 2000. Employment Options of the Future: Actual and Preferred WorkingHours. National Working Paper for the Netherlands. Dublin: European Foundationfor the Improvement of Living and Working Conditions.

Visser, J. 2002. “The First Part-time Economy in the World: A Model To BeFollowed?” European Journal of Social Policy 12: 23–42.

Visser J. and A.C. Hemerijck. 1997. A Dutch Miracle: Job Growth, Welfare Reform,and Corporatism in the Netherlands. Amsterdam: Amsterdam University Press.

9780230_202573_08_cha07.tex 5/6/2008 15: 39 Page 259


Visser, J., A.C. Wilthagen, R. Beltzer, and E. Koot-van der Putte. 2004. “TheNetherlands: From Atypicality to Typicality.” Pp. 190–222 in Employment Policyand the Regulation of Part-Time Work in the European Union: A Comparative Anal-ysis, edited by S. Sciarra, P. Davies, and M. Freedland. Cambridge: CambridgeUniversity Press.

Yerkes, M., and J. Visser. 2006. “Women’s Preferences or Delineated Policies:The Development of Part-Time Work in the Netherlands, Germany, and theUK.” Pp. 235–61 in Decent Working Time: New Trends, New Issues, edited byJ.Y. Boulin, M. Lallement, J. Messenger, and F. Michon. Geneva: InternationalLabour Office.

Yerkes, M. 2006. What Women Want: Individual Preferences, Heterogeneous Pat-terns? Women’s Labour Market Participation Patterns in Comparative Perspective.PhD thesis. Amsterdam School for Social Science Research/AIAS. University ofAmsterdam.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 260

8Comparative Regime Analysis:Early Exit from Work in Europe,Japan, and the USABernhard Ebbinghaus

Introduction: early exit from work

One of the fundamental policy problems of contemporary welfarestates, particularly in Continental Europe, has been the extensive useof early retirement as a labor-shedding and passive labor market policysince the mid-1970s. Early exit from work – that is, before the age of65 – has not only played a major role in lowering the overall employ-ment rate for the working-age population, thereby reducing the socialcontribution and income tax base for welfare state financing; it has alsopushed up the levels of social expenditure through the increased take-upof various social transfer programs. The labor-shedding problem hasbeen criticized by many scholars as a fundamental part of the “Con-tinental dilemma” (Scharpf, 2001) and the “welfare state without work”syndrome (Esping-Andersen, 1996). Using a comparative regime per-spective, I will analyze the specific institutional configurations of thesepolitical economies that can explain the general rise in early exit fromwork, but also its cross-national variations. In recent years, there havebeen some policy reversals, such as the European Union’s commitmentto increase the activity rate of older workers, yet in order to evaluatethese reforms we need to understand the initial causes of and particularpatterns of early exit from work.

Early retirement commonly has two meanings: withdrawal fromemployment prior to age 65 and the drawing of pre-retirement benefitsuntil a statutory pension becomes available. It is thus a social practicethat is related to and entails consequences for both social security pro-grams and the labor market. During the first three postwar decades,retirement at statutory pension age (in most countries at age 65, albeit

260

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 261

Bernhard Ebbinghaus 261

with some significant exceptions) became a social institution of mod-ern welfare states that regulated the later part of the life course (Atchley,1982). Since the 1970s, early withdrawal from work before age 65 hasbecome increasingly widespread, leading to a destandardization of theretirement age (Kohli and Rein, 1991): Increasing numbers of older peo-ple have expected to leave work early, while the age at exit is becomingless predictable. Although the decision to retire early depends upon theindividual circumstances of older workers, the strategies of firms andother co-fellow workers, these micro-level decisions occur in the con-text of particular institutional environments. The welfare state regime,the production system and labor relations all play important roles inexplaining the specific trajectory and cross-national differences in therise and reversal of early exit from work.

Early exit from work has been studied from two often-competing per-spectives: protection-oriented analyses of pull factors that impact laborsupply or production-oriented studies of push factors that affect labordemand (Kohli and Rein, 1991). Therefore we expect that both cross-national regime differences in welfare states and also varieties of capitalistsystems should have an impact in shaping pull and push factors affect-ing the early exit from work. Arguments about the role of social partnersin both perspectives remain implicit, although I will argue that employ-ers and workers’ organizations play a crucial role in mediating betweenwelfare incentives (pull) and economic contingencies (push). The socialpartners are involved in the arena of social policy and collective bargain-ing, and management–labor relations shape the ways in which firm-levelactors abstain from or utilize early exit under the given constraintsand opportunities. Informed by these three complementary perspectivesof protection-related pull, production-related push, and partnership-related mediation, I adopt here an encompassing regime approach thatcombines insights from comparative welfare states research, politicaleconomy and labor relations analysis.

In addition to providing a cross-national analysis of a widespreadtrend and specific practice with major repercussions for welfare statesand employment systems, this contribution highlights two particularmethodological approaches. First, I will present a comparative analysisthat highlights differences between as well as within regimes, discussingthree typologies of welfare states, production systems, and labor rela-tions that have attracted much debate in comparative political economy.Early exit from work is a prime case to analyze the potential impactof macro-institutional contexts on micro-level social processes as wellas the interaction and even “institutional complementarities” (Hall

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 262


and Soskice, 2001) between different societal subsystems. The secondcontribution to the dialogue between methods and substance of thisvolume is the adoption of a dynamic perspective and the developmentof process-oriented indicators for macrocomparative analysis. Cross-national analyses of employment indicators, such as the level of activityamong older workers, are often static aggregate comparisons of contem-porary levels without taking into account the long-term developmentaltrajectory and the often cohort-specific nature of social change.

I will first describe the comparative regime approach applied here,in particular the case-oriented analysis of institutional configurationsand the selection of a medium number of ten cases, carefully selectedfrom different regime configurations. In the next section, I will describethe main decline in employment rates for men and women aged 55to 64. The following section turns to cohort-adjusted early exit ratesthat allow a better analysis of early retirement trends over time andacross countries. The next section summarizes the main variations inearly exit regimes. The protection-related pull thesis – that is, the cross-national variations in welfare state regimes and the different multiplepathways – will be then discussed. The subsequent section then turns tothe production-related push thesis and the role of social partners, dis-cussing the different economic governance modes and their impact onlabor-shedding strategy.

Comparative regime analysis

Adopting a variation-finding comparative design (Tilly, 1984), I seekto explain cross-national variations in early exit patterns with partic-ular regime constellations. The leading research question is thus: Underwhich production, protection, and partnership regimes do we find earlyexit from work to be most common? The inter-regime comparison, usingthe most-dissimilar-country design (Przeworski and Teune, 1970), helpsto single out the impact of specific institutional configurations on thedevelopment of particular early exit trajectories. While the inter-regimecomparison helps to account for the path-dependent trajectories ofearly exit from work, it cannot unravel path departures through policyand institutional change. Embracing an additional most-similar-countrydesign (Dogan and Pelassy, 1990), the intra-regime comparison enablesanalysis of more subtle variations from the dominant model as well as thepotential for institutional change within similar regime configurations.Outliers that do not fit prior expectations given a regime comparison

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 263


can then be used to explore the particular circumstances in within-casestudies.

As my dependent variable, I use longitudinal but cohort-adjusted dataon employment changes as indicators of early exit from work before age65 for men and women. Aggregate labor force data was collected mainlyfrom OECD sources (with additions from EU and national sources) tocompare early exit patterns over time (1970–2004) and across ten selectedOECD countries. This study applies cohort-adjusted early exit rates forboth men and women that are informed by the life course approach (Set-tersten and Mayer, 1997). Especially regarding women, exit indicatorsthat are not cohort-adjusted would be misleading due to the often-substantial increases in female labor force participation from cohort tocohort. With cohort-adjusted exit rates, we can see that not only amongolder men but also among their female peers, early withdrawals fromwork before age 65 had increased during the three decades since theearly 1970s.

For practical reasons, the empirical analysis was limited to macro-leveldata due to the focus on longitudinal developments across ten OECDcountries and the scarcity of cross-nationally comparable micro-leveldata (for recent micro-level studies using national surveys see Gru-ber and Wise (1999) and Blossfeld, Buchholz, and Hofäcker (2006) orEuropean household panel data see Schils (2005)). The analysis adoptsa qualitative case-oriented comparison instead of using cross-nationalpooled time-series analysis since this method exhibits serious limita-tions (Ebbinghaus, 2005; Kittel, 1999), given the highly auto-correlatedbut nonstationary processes due to long-term diffusion processes andpath-dependent trajectories of early exit from work. This study uses lon-gitudinal macro-indicators to describe the specific national trajectoriesin early exit from work that will be then explained by different macro-institutional configurations. Since the independent variables are difficultto quantify, therefore institutional regime typologies as qualitative holis-tic classifications are used and their configurations are related to themore or less extensive early exit patterns. The comparative strategy thusresembles what Mahoney (2003) calls “ordinal comparison.”

In order to allow systematic inter- and intra-regime comparisons,I chose a selected group of ten OECD countries that share not onlyenough commonalities to be comparable, but also sufficient differenceson both dependent and independent variables. The selected countriesrepresent cases from the main configurations along the conceptuallydefined dimensions of protection, production, and partnership regimes(see chapter 3 in Ebbinghaus, 2006). For pragmatic reasons, the study

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 264


is limited to a manageable number of countries for which crucial quan-titative indicators were available and additional qualitative case studiesalready existed and could be used for secondary analysis. I opt for asmaller set of countries than usual in OECD studies, yet this mediumN set of ten countries allowed the combination of intensive within-casestudies based on secondary analyses (not reported here) with enoughcross-national variations to systematically explore differences in institu-tional configurations. My comparative method thus follows Ragin (1987)in analyzing countries as theory-relevant cases that allow us to explorecontext-dependent, process-oriented analyses and to check for alterna-tive hypotheses (Mahoney, 2003). They are not observations of a largercountry sample to gain statistical leverage, thus the aim is not to extendbeyond the cases at hand but to systematically analyse the findings fromintensive within-case studies in a cross-case comparison.

This study applies an institutionalist approach. The decisions of actorsat the workplace level or in the social policy and bargaining arenas areembedded in institutional environments that shape actors’ orientationsand interests as well as the opportunity structures for the actor constel-lations. Following Esping-Andersen (1990), I use “regime” to refer to theways in which institutions hang together and interact in a systemic way,using the term as an analytical construct of the interrelations betweeninstitutions and their interaction with their environment. It is a heuris-tic tool to conceptualize complex institutional arrangements as a holisticsystem. A regime approach can help us understand how social protec-tion is institutionalized, production systems are organized, and laborrelations are governed. The regime approach is particularly useful in com-parative analyses in order to conceptualize distinct regime typologies inwhich to classify empirical similarities and differences. While regimesshould be theoretically grounded, representing ideal-types, much com-parative work nevertheless seeks to use typologies to classify empiricalcases or real-types.

Seen from the protection-oriented pull perspective, social transferprograms provide incentives and opportunities to retire from employ-ment before the statutory pension age (around age 65). There areoften multiple exit pathways or “a combination of different institutionalarrangements that are sequentially linked to manage the transition pro-cess, that is, the period between exit from work and entry into thenormal old-age pension system” (Kohli and Rein, 1991, p. 6). Dif-ferent social policy programs provide alternatives that facilitate earlywithdrawal from work: Flexible old age pensions, disability pensions,special pre-retirement schemes, long-term unemployment as well as

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 265


partial pension benefits (Kohli et al., 1991). Concentrating on only onesocial security program is thus insufficient to fully grasp pull factorsbecause there are several alternative pathways, even when they are notnecessarily complete functional equivalents. The welfare regime anal-ysis cannot confine itself to the public programs of the welfare state,but must acknowledge the importance of the public–private mix (Esping-Andersen, 1999; Rein and Rainwater, 1986). Not only public policy, butalso private occupational benefits – provided by employers or negotiatedby the collective bargaining partners – may offer additional opportuni-ties or supplement insufficient public benefits for preretirement. Earlyretirement thus presents a prime case of the regime approach’s utility(Esping-Andersen, 1990), as it emphasizes the systemically interwovenweb of institutions and their interaction within particular environ-ments. In comparative perspective, we can expect that different welfareregimes provide varying opportunities of income support for early exitfrom work.

On the other hand, in order to fully understand early retirement, weshould take into account the reasons why firms shed older workers –the production-related push factors. In addition to institutionalized exitpathways and their incentive structures that affect labor supply, thereare economic forces at work that influence the labor demand side. Mostimportantly, labor shedding or retaining of older workers will dependon firms’ age-related hiring, training, and firing policies. Early retire-ment is one major socially acceptable response to these pressures andconstraints (Naschold and de Vroom, 1994). As institutional complemen-tarities, pre-retirement benefits help socially buffer firms’ labor-sheddingstrategies. Thus the public–private welfare mix is particularly relevantfor the study of early exit from work (Casey, 1992). What seem to be –in the protection-oriented “pull” perspective – income transfers to indi-viduals that provide strong incentives to withdraw from work are, whenseen from the firm-oriented “push” perspective, economically-motivatedlabor-shedding measures. In this case, public or private welfare poli-cies are not politics against markets, enforcing redistributive socialrights on free market economies, but politics for markets, enhancingthe adaptability of social market economies (Ebbinghaus and Manow,2001).

However, neither protection-oriented nor production-orientedaccounts sufficiently explain early exit. The labor supply perspectiveassumes that incentives determine the decision of older workers to retire,while the labor demand perspective perceives early exit as the out-come of firms’ human resource strategies. Protection systems provide

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 266


the pathways and the incentives for early retirement; the productionsystems induce pressure to shed older workers. Yet the partnership insti-tutions are crucial in shaping the ways in which the social actors reactto push and pull factors. At workplace level, the main actors – man-agement, worker representatives, and the older workers – interact tofind adequate responses given constraints and opportunities providedby the protection and production systems. Organized labor and capital,together with the government, influence and implement policies affect-ing early exit from work. The social partners not only regulate wages,employment conditions, and workplace relations through collective bar-gaining, they also influence social policy making and implementationthrough their political channels and their involvement in social insur-ance or occupational welfare programs. We should thus expect that thenational traditions of partnership, that is, the institutionalized relationsbetween the state, organized labor, and employer associations as well asthe workplace relations between management, workers, and their repre-sentatives will have a major influence on national exit policies and onthe everyday social practice of early retirement.

In this study, I use regime typologies from three different fields tomap the main differences across countries with respect to protection,production, and partnership institutions (Ebbinghaus, 2006). Althoughthese regime typologies were developed largely independently and dealprimarily with different social systems, they share a similar systemic viewof institutions.

• First, for a regime typology of protection systems that provide thepull towards early exit, I rely on Esping-Andersen’s well-knownwelfare regime typology (1990, 1999). Three regime clusters are dis-tinguished: Social-democratic universalist welfare states in Nordiccountries, Christian-democratic conservative social insurance states inContinental Europe, and liberal-residual basic social security systemswith substantial private pensions in the United Kingdom, Ireland, theUnited States, and Japan.

• Secondly, for a classification of production systems and their economicgovernance, I borrow from similar typologies (Hall and Soskice, 2001)that juxtapose two political economy models: Anglo-American liberal(uncoordinated) market economies and coordinated (non-liberal) mar-ket economies (Germany and its Rhenish neighbors, Scandinavia, andJapan).

• Finally, for the analysis of partnership traditions, I draw on compara-tive neo-corporatist studies of labor relations and organized interests

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 267


(Crouch, 1993). These delineate three different management–labormodes: voluntarist (or “give-and-take”) bargaining traditions in Anglo-American labor relations, contentious labor relations in Latin Europe,and cooperative labor relations in the remaining countries.

There is no clear one-to-one relationship between these institutionalconfigurations; nevertheless, there are some intriguing institutionalaffinities between welfare regimes, production systems, and labor rela-tions. The Liberal Market Economies (the UK, the US, and Ireland) allcombine residual welfare and voluntarist labor relations. Among theCoordinated Market Economies (CMEs), there are several combinationsbetween welfare regimes and labor relations. In fact, we may needmore subtle distinctions for these CMEs: Central coordination in Nordiccountries, sectoral coordination for Germany and the Netherlands, andstate-coordination for Latin Europe (Kitschelt et al., 1999).

The purpose of locating countries in these analytical typologies is toprovide a conceptual map (Rokkan, 1999), which explains particular out-comes based on the institutional configurations. Thus, no one “mastervariable” is sufficient on its own to explain the divergent early exittrajectories; rather, it is the particular interaction of protection (pull)and production (push), mediated by specific partnership relations. Thecomparative regime typologies also help in delineating the institutionalobstacles and opportunities for policy reversal as they provide the basisfor evaluating the degree of path departure from regime-specific trajec-tories. Thus, the purpose of this encompassing comparison (Tilly, 1984)is to generate hypotheses about institutional macro-configurations andconfront these with more process-oriented within-case studies.

For this study, I selected eight member states of the European Union,representing four regime configurations, and added two major non-European OECD countries, the United States and Japan (see Table 8.1).These two global players are not only major economic competitorsto these European economies, but also prime examples of opposingpolitical economy models: The United States is a case of an (uncoor-dinated) Liberal Market Economy, Japan of a (non-liberal) CoordinatedMarket Economy (Hall and Soskice, 2001). Explaining early exit froma production-oriented view, these two juxtaposed political economymodels pose a puzzle. Because these countries with opposing produc-tion systems both have low exit rates, the differences in productionregime (that is, the push factor) alone cannot explain such a similar out-come. Conversely, taking into account the pull factor, the liberal-residualwelfare regimes cannot be the sufficient cause for low early exit since

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 268


Table 8.1 Conceptual map of protection, production, and partnership regimes

Cluster countries Protection (“pull”) Production (“push”) Partnership (mediation)

Center Conservative Coordinated CooperativeGERMANY

NETHERLANDS

Latin Conservative Coordinated ContentiousFRANCE

ITALY

Nordic Universalist Coordinated CooperativeSWEDEN

DENMARK

Anglo-American Liberal-residual Liberal VoluntaristUKIRELAND

USA

Asian Liberal-residual Coordinated CooperativeJAPAN

Note: see Ebbinghaus 2006: chap. 3 for details on these three typologies.

Sweden has also had relatively low early exit from work, despite sharinga relatively generous social security system with high exit ContinentalEuropean welfare states.

Declining employment rates

A first indicator of the rise in early retirement is the decline in employ-ment; we should use employment rates which exclude those that are notactive since unemployment benefits for older workers allow de facto earlyretirement. A drop in the employment rate indicates that fewer olderpeople remain in gainful employment, while the others – non-workingpeople – are most likely to be dependent on pension, unemployment,other welfare benefits, or – especially in the case of housewives – theirspouse. For more detailed analysis we also need to disaggregate theearly exit trends into at least two age groups: early exit (age 60–64), forwhich many pre-retirement options exist, and very early exit (age 55–59),for which few pre-retirement opportunities, other than unemploymentbenefits, exist.

The most significant decline in employment rates occurred amongmen in the age group 60–64. With the exception of Italy, which already

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 269


0

10

20

30

40

50

60

70

80

90

100G

erm

an

y (W

est)

Net

herl

ands

Fra

nce

Italy

Sw

eden

Den

ma

rk UK

Irel

and

US

A

Japa

n

Ge

rma

ny

(Wes

t)

Net

herl

ands

Fra

nce

Italy

Sw

eden

Den

ma

rk UK

Irel

and

US

A

Japa

n

Men 60–64Men 55–59

1970 1985 2000 15–64

Figure 8.1 Employment rates for men aged 55–59/60–64/15–64, 1970–2000Source: Ebbinghaus 2006: Table 4.4; OECD Labour Force Statistics, 1970, 1985, 2000.

had low employment rates (above 50 percent), all other countries expe-rienced a long-term decline from the employment level of the 1960s(above 70 percent). The drop was most pronounced in the cases of Ger-many, France, and the Netherlands during the 1970s; more gradualthereafter, but plunging in the 1980s – to even below Italy’s tradition-ally low level. All four Continental welfare states stand out as having thelowest levels of active employment among men aged 60–64: less thanone in three West Germans or Italians, less than one in four Dutchman,and one in every six Frenchmen or East German.

Early retirement among men aged 60–64 is less common in the Nordicand Anglo-American countries (see Figure 8.1). Nevertheless, the UnitedKingdom, the United States, and Ireland have seen a drop to mediumemployment levels: Only every second man aged 60–64 is in work,despite a later statutory retirement age of 65. Sweden maintained a higherlevel until the surge in unemployment during the early 1990s that led toa drop from 60 percent to 50 percent in employment rates. Neighboring

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 270


Denmark saw an even more severe drop – to an employment rate of lessthan 40 percent in 2000, coming close to Germany around the turn ofthe century (30 percent). In terms of the timing of labor shedding, Swe-den and Ireland experienced their major declines in the 1990s; Denmarkand Britain in the 1980s; and the United States even earlier. Finally, Japanstands out with a more gradual decline during the mid-1980s and againsince the mid-1990s. More than 70 percent of Japanese men still workafter the age of 60 when they can draw the second-tier contributory pen-sion, which they may combine with a part-time job. While most of thelarger Japanese companies enforce mandatory retirement by the age of60 (Kimura et al., 1994), these workers seek re-employment in order tosupplement their pension and severance pay income.

During the 1960s, nearly all societies had high levels of employmentamong men aged 55–59 (above 90 percent), the exceptions being France(80 percent), the United States (85 percent) and, most notably, Italy (onlyaround 75 percent). During the 1970s, all countries experienced a declinein employment rates in this age group, but the Nordic, Anglo-Americancountries, and Japan went through the decline more slowly than Con-tinental Europe. After relatively slow decline throughout the 1970s and1980s, with high unemployment in the 1990s, Swedish employmentamong men 55–59 also dropped rapidly from 86 percent in 1990 to 76percent in 1995. The downward trend in Denmark already started inthe 1980s and continued until the mid-1990s. Britain and Ireland havehad a more rapid decline since the late 1970s, due to high and fluctuat-ing unemployment in this age group: every third man aged 55–59 wasinactive by the mid-1990s. In comparison, the United States performedsomewhat better, starting from a lower level and experiencing a slowerdecline: one in every four American men aged 55–59 was not (or nolonger) employed in the 1990s.

The Continental European welfare states are distinct in their signifi-cant drops in employment rates in this early age group. Italy always had alow employment level: Until the mid-1970s only three out of four Italianmen aged 55–59 were working and by the 1990s it was only every secondman. France and the Netherlands had higher employment rates in theearly 1970s, but thereafter witnessed a major drop to only 60 percentby the mid-1990s. West Germany’s decline was somewhat less dramaticthan in the Netherlands but by the late 1990s both countries had reachedthe same level of inactivity: Every third man aged 55–59 was not working(in East Germany it was 40 percent).

Analysing early retirement among women with the help of employ-ment rates produces the same shortcoming as previously discussed for

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 271


0

10

20

30

40

50

60

70

80

90

100G

erm

any

(Wes

t)

Net

herla

nds

Fran

ce

Italy

Sw

eden

Den

mar

k

UK

Irel

and

US

A

Japa

n

Ger

man

y (W

est)

Net

herla

nds

Fran

ce

Italy

Sw

eden

Den

mar

k

UK

Irel

and

US

A

Japa

n

Women 55–59 Women 60–64

1970 1985 2000 15–64

Figure 8.2 Employment rates for women aged 55–59/60–64/15–64, 1970–2000Sources: Ebbinghaus 2006: Table 4.5; based on OECD Labour Force Statistics, 1970, 1985,2000.

participation rates. Nonetheless, the disaggregation into two age groupsprovides some additional information (see Figure 8.2). In 1970, nearlythe majority (around 40–50 percent) of women aged 55–59 was workingin all countries, except Germany (37 percent), Ireland (below 20 percent),the Netherlands (18 percent) and Italy (14 percent). The most dramaticincrease in employment occurred in Sweden, reaching a level beyond70 percent in the mid-1980s, followed by gradual increases in Denmark,the United Kingdom, the United States, and Japan; all these countrieshad reached levels between 50 percent and 60 percent by the 1990s. TheContinental European countries and Catholic Ireland showed stagnat-ing levels until the 1990s. During this time, female employment ratesparticularly improved in the Netherlands, Ireland and Italy.

A comparison with the employment rates for the later age group (thoseaged 60–64) shows considerably lower levels and often falling employ-ment rates, this indicates that early retirement supersedes the trend ofincreasing female participation in this age group. In particular, where

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 272


women can draw on pensions earlier, employment rates are considerablylower: British women receive the basic pension at the age of 60, whileGerman, Danish, Italian, and Japanese women can, under some condi-tions, draw pensions earlier than men, though they will often receivelower benefits than if they worked their remaining years until statutoryretirement. In Sweden, employment levels still increase from cohort tocohort; at least they did before the unemployment crisis of the 1990s.In Japan, with the second highest employment rates (short under 40percent), there is no visible change over time. The United States and, ata lower level, the United Kingdom and Ireland have experienced someincrease in recent years, while Denmark has a much lower (and unstable)employment level among women aged 60–64 due to the use of unem-ployment insurance as a bridging pension. In all Continental Europeancountries, the level of employment has been falling, as in Germany andFrance, or remains very low, as in the Netherlands and Italy. However,falling employment rates are an incomplete means to detect the scopeof early retirement, particularly among women since each female cohorttends to have a higher likelihood to be working at the age of 55. In orderto take cohort-specific effects into account, we need to adopt a perspec-tive that studies employment patterns over the life course and acrosscohorts.

Early exit from work

Seen from a life course perspective, we should measure exit from workby cohort-adjusted employment rates, particularly in the case of womenwith rising participation levels and in countries where early retirementhas lowered participation levels for several age groups. Following a lifecourse perspective, this study compares the impact of cohort and histori-cal changes on early retirement patterns with the help of cohort-adjustedexit rates or “net withdraw rate” (Blöndal and Scarpetta 1998). It uses therelative exit rate (percentage change) to measure cohort-adjusted with-drawal rate as a proportion of the population “at risk” of exiting work,that is, those in the same birth-cohort who were employed five yearsearlier (for example, aged 55–59 in 1990). The relative exit indicator isbetter than measuring absolute changes in employment rate in the caseswhere employment rates in previous age groups are relatively low, thisholds particularly for women or countries with very early exit.

Early exit among men aged 60–64 follows similar cross-national differ-ences as those discussed for the decline in employment rates, with someminor differences. The Continental European countries show the same

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 273


Continental Europe

0

10

20

30

40

50

60

70

80

0

10

20

30

40

50

60

70

80

1970 1975 1980 1985 1990 1995 2000 2005 1970 1975 1980 1985 1990 1995 2000 2005

D (West)D (East)NLFI*

Other countries

S DKUK IRLUSA JAP

Figure 8.3 Relative exit rates for men aged 60–64, 1970–2003Notes: 5-year moving average of relative exit rates (see Ebbinghaus 2006); I∗: Italy 1965–72:55–59 partly estimated.Sources: OECD Labour Force Statistics 1966–2004, except IRL: Ireland 1983-, S: Sweden 1997-:Eurostat Labour Force Surveys, and own calculations.

pronounced trend of rapidly increasing early exit and high levels of earlyretirement, while the trajectories in the other countries are more gradualand remain at a lower exit level (see Figure 8.3). Given declining or lowemployment levels for the previous age group, the two withdrawal mea-sures diverge over time: absolute exit rates, the share of the age group60-64 leaving work, fall behind the relative exit measure, which showsthe propensity of those previously (age 55–59) employed to leave workupon reaching ages 60–64.

During the first growth period (from 1970 to the first peak in 1985),the Continental European countries witnessed acceleration (with annualgrowth rates of 8–9 percent) in relative early exit from work (see Table 8.2).An exception is Italy, which had a more gradual annual increase (2.6percent) due to an already high level of early retirement in 1970 (30percent). By the 1980s, more than 40 percent of the age group 60–64retired early and more than half of those initially employed before 60stopped working within the next five years. During the second period,beginning in the mid-1980s, we also see some stagnation and short-term

9780230_202573_09_cha08.tex

5/6/200818:29

Page274

274

Table 8.2 Relative exit rates, men and women aged 60–64, 1970–2003

Men aged 60–64 Women aged 60–64

1970 1985 2003 1970–1985 1985–2003 1970 1985 2003 1970–1985 1985–2003

Germany – – 51.5 – c −0.90 – – 63.5 – c −1.20(West) 18.2 57.9 49.9 +8.00 −0.81 38.4 71.2 60.6 +4.20 −0.89(East) – – 58.6 – c−0.83 – – 74.7 – c −1.07Netherlands a 19.8 63.4 52.1 +8.67 −0.47 a 27.0 66.1 52.8 +6.15 −1.24France 19.7 62.0 76.3 +7.95 +1.11 19.2 60.8 72.8 +7.97 +1.01Italy 31.0 45.2 40.8 +2.74 +0.35 22.4 40.6 54.2 +4.04 +1.62Sweden 15.8 28.8 24.2 +4.10 +0.74 24.4 36.7 27.9 +2.75 −1.50Denmark a 13.3 45.2 33.9 +8.49 +0.69 a 34.4 44.6 53.9 +1.74 +1.06UK 15.9 41.3 23.4 +6.55 −2.07 37.3 66.1 48.2 +3.89 −1.74Ireland a 6.8 27.6 19.0 +5.10 −1.81 a 7.1 31.1 17.4 +10.31 −3.18USA 18.4 32.6 28.5 +3.89 −0.81 24.4 31.3 27.1 +1.68 −0.80Japan 11.8 23.7 29.0 +5.14 +1.13 b 23.3 24.1 34.6 b +0.23 +2.04

Notes: relative exit rate: decline in cohort-specific employment rate in % previous employment rate,; a 1970/75 estimated based on participation rates(ILO); b 1971, c1991 (estimated), d1996; 1970–85, 1985–2003: annual natural growth rate.Sources: OECD Labour Force Statistics 1965–2004, except: Germany: 1991–2003 German Statistical Office internal data; Italy 55–59 (1966–72 estimatedbased on 50–59, 1973–), Sweden 1997–2003, Denmark 1975–83, and Ireland: Eurostat Labour Force Surveys 1973–2004; and own calculations.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 275


Continental Europe

0

10

20

30

40

50

60

70

80

0

10

20

30

40

50

60

70

80

1970 1975 1980 1985 1990 1995 2000 2005 1970 1975 1980 1985 1990 1995 2000 2005

D (West)D (East)NLFI*

Other countries

S DKUK IRLUSA JAP

Figure 8.4 Relative exit rates for women aged 60–64, 1970–2003Notes: 5-year moving average of relative exit rates (see Ebbinghaus 2006); I∗: Italy 1965–72:55–59 estimated.Sources: OECD Labour Force Statistics 1966–2004, except IRL: Ireland 1983-, S: Sweden 1997–:Eurostat Labour Force Surveys, and own calculations.

fluctuations, with the exception of France, where early retirement con-tinues to increase. Given the considerable labor shedding among oldermale workers occurring throughout Continental Europe, one can refer tothese countries as “welfare states without work” (Esping-Andersen 1996).They all show a long-term growth trajectory that follows an S-curve forboth men (see Figure 8.3) and women (see Figure 8.4), suggesting a dif-fusion process that first excels during the 1970s and since the 1980s hasreached saturation at a very high level of “penetration”: early retirementsometime between 60 and 64 has become the dominant social norm inContinental Europe.

In contrast, the other countries on average show less dramatic increasesand reach lower levels of early exit among men aged 60–64, though theyalso go through more pronounced cyclical fluctuations (see Figure 8.3).During the first period (1970–85), growth in early exit was slower (4–5percent) and reached a much lower level (below 33 percent in 1985) thanon the Continent (above 45 percent) with two exceptions. Denmark andBritain show rapid growth rates (7–8 percent), albeit starting from a some-what lower level (about 15 percent in 1970) but (nearly) reaching the

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 276


Italian level in net withdrawal (45 percent in 1985). Sweden, the UnitedStates, Ireland, and Japan had relatively low early exit rates; betweenone-fifth and one-quarter of men aged 60–64 left work early. While earlyretirement declined somewhat during the late 1980s, the early 1990sbrought another cyclical upturn, most dramatically in Sweden and Den-mark. Interestingly, despite high levels of old age employment, morethan a quarter of Japanese men aged 60–64 had left employment inthe 1990s. Differences between the two measures are relatively minimal,given the high employment rates among men aged 55–59.

We draw a much clearer picture of early retirement when we ana-lyze early exit rates and adjust for cohort effects than if we look onlyat employment changes in aggregate. We find three distinct groups ofcountries for early retirement among men aged 60–64: the ContinentalEuropean high early exit countries, the in-between cases of the UnitedKingdom and Denmark, with medium-level early retirement, and theremaining countries (Sweden, Ireland, the US, Japan) with oscillating,but overall lower levels (see Figure 8.3). In contrast to employment rates,the analysis of cohort-adjusted exit rates indicate a very similar increasein early retirement for older women aged 60–64 in Continental Europe,with the exception of Italy, which maintained a medium level until theearly 1990s (see Figure 8.4). In general, all four countries show paral-lel trends for women and men with only a few exceptions: Germany’sexit rates among women exceed those for men; Italy’s gender gap hasincreased since the 1980s; there is no large difference in the Netherlands;and, in France women are less prone to retire early in the age group60–64. Yet Germany and France have higher levels of female employ-ment compared to the Netherlands and Ireland and thus nearly everyfourth woman aged 60–64 withdraws from work. The high relative exitrates in the Netherlands and Italy are less important in absolute num-bers because fewer women were employed at all before age 60. The othercountries show lower early exit rates for women too, again with thepartial exception of Denmark and Britain.

Such different societies as Sweden, the United States, and Japan showvery similar early retirement patterns for women aged 60–64, with cycli-cal ups-and-downs around a low level of early retirement (between 20percent and 35 percent) in the 1980s. Yet, in the early 1990s, female exitrates increased in Sweden and Japan as a result of more severe labor mar-ket conditions. Another exception is Catholic Ireland, where, despitelow overall levels of female participation, a substantial share of olderwomen withdrew from work during the unemployment crisis of the1980s, a trend that was reversed with improved labor market conditions

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 277


in the 1990s. The two major outliers are Denmark and the United King-dom. Given the statutory female pension age of 60, British women tendto withdraw at rates comparable to Continental Europe (66 percent in1985, declining to 52 percent in 2000). Since the late 1970s, Danishwomen aged 60–64 withdrew from work at an increasing rate, such thatby 1995 two-thirds had left employment, this has been largely possi-ble through disability pensions that allowed retirement long before thenormal pension age of 67.

Early exit regimes

The analysis of long-term trends in early exit from work for the agegroup 60–64 revealed two different early exit trajectories within the over-all trend: (1) Continental European countries showed an S-curve-likediffusion process from low to high early retirement over the past threedecades (with the partial exception of Italian men, who maintained analready high level of early exit); whereas (2) in the other countries earlyexit from work grew less rapidly and continued fluctuating during the1980s and 1990s with the exception of early exit among British andDanish women that came close to Continental levels during the 1980sand 1990s, respectively. By the mid-1990s, we can distinguish three dif-ferent worlds of early exit: (1) the Continental European high early exitcountries (Germany, the Netherlands, France, and Italy); (2) the Britishand Danish medium exit levels (particularly among women); and (3) theother countries (Sweden, Ireland, the United States, and Japan) withlower levels of early exit.

Would the inclusion of early exit before age 60 alter these findings?Not significantly. Indeed, the difference between the Continent and theother countries might even be enhanced (see Ebbinghaus, 2006). All fourContinental European countries have high exit rates that are rising, whilethe other countries show more cyclical patterns due to unemploymentwaves that remain below the Continental European level. Very earlyexit (age 55–59) follows similar patterns as exit after age 60, though thegroups of very early exit clearly stand out: Italians, French, and EasternGermans. Moreover, earlier exit before age 55 occurs occasionally (withthe exception of Italy), but remains limited to some occupational groups(such as miners) and to cyclical unemployment due to the absence ofother institutionalized public exit pathways for this age group. The threeworlds of exit regimes – high exit Continental Europe, medium-level out-liers Denmark and Britain, and low exit countries – are still observablein the 1990s.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 278


Cohort-adjusted early exit rates provide a better tool to study earlywithdrawal from work, particularly for women. Continental Europe’smain trend in early exit for the age group 60–64 shows an S-shapeddiffusion curve with fast growth in the 1970s and early 1980s and a lev-eling off thereafter. In the other countries, however, early exit from workincreased more gradually and cyclically, with the exception of relativelysignificant trends in early exit among women in the United Kingdomand Denmark. Although the level of very early exit before age 60 remainsconsiderably lower, it increased rapidly during the 1980s and remainedsubstantial in the 1990s in Continental Europe, while the other coun-tries show a lower but cyclical trend. Premature early exit before age 55remains rare, with Italy a major exception.

Based on the early exit trajectories in the age groups 55–64 (see Ebbing-haus, 2006), there are clear country clusters that partly overlap with theregime typologies discussed thus far (see Table 8.3). The four Continen-tal European welfare states show highest levels of early exit from workfor both men and women. While Italy traditionally has had very earlyexit (age 50–59) and France showed increasing exit from work in theage group 55–59, very early exit before age 60 by and large remains lim-ited to unemployment in Germany and the Netherlands (see Figures 8.3and 8.4). With the exception of France, the rise in early exit has leveledoff in recent years, and there has even been some improvement, par-ticularly in the Netherlands. The remaining countries – universalist andresidual-liberal welfare states – have moderate or lower levels of earlyexit. Denmark and the United Kingdom are outliers among the non-Conservative welfare states with relatively high levels of early exit amongolder women (age 60–64). They also stand out due to relatively high lev-els of male early exit during the 1980s, continuing in Denmark intothe 1990s. Sweden and the United States have moderate levels of earlyexit for men and women, while Japan and Ireland have comparativelylow levels of early retirement. The trend among the non-Continentalcountries is less marked by a general diffusion curve, following a cyclicalpattern instead (see Figures 8.3 and 8.4), indicating that early retirementresults more from changing labor market situations than solely becauseit has become a socially accepted role. This holds also for Denmark andSweden where exit increased during periods of high unemployment inthe 1990s, but then receded. While British and Irish early exit trendsdeclined with labor market improvements during the 1990s, Japan, afterlong periods of low early exit, experienced an unusual increase with theeconomic downturn in the late 1990s. Can the “push” and “pull” thesesexplain these early exit patterns?

9780230_202573_09_cha08.tex

5/6/200818:29

Page279

279

Table 8.3 Index of pathways ranked by exit opportunities

Statutory pension Flexible pension Special scheme Unemployment Disability Index of pathways Exit trend

Italy ∗∗∗ ∗∗ ∗∗ ∗∗ ∗ {∗∗} 10 {12} early, highFrance ∗∗∗ ∗∗∗ ∗∗ ∗ 9 early, highGermany ∗ w: ∗∗ ∗∗ {∗∗} ∗∗ ∗∗ 7 {9} w: 8 highNetherlands ∗∗∗ ∗∗ ∗∗∗ 8 highDenmark ∗ ∗∗ {∗∗∗} ∗∗ w: ∗∗∗ 5 {6} w: 6 {7} m: mod.w: highUK w: ∗∗∗ ∗∗ {m: ∗∗} ∗ m: 3{5} w: 6 m: mod.w: highSweden ∗ ∗ ∗ {∗∗} 3 {5} moderateUSA ∗∗ ∗ 3 moderateIreland ∗ ∗ ∗ 3 lowJapan ∗∗ ∗ 3 low

Note: opportunities for early exit (see Ebbinghaus 2006: Table 5.5 for details): #: with labor market consideration; Index of pathways: ∗∗∗ major pathway(3 points); ∗∗ conditional pathway (dismissal, unemployment, disability); (2 points); ∗ limited pathway (means-tested, medical-test only, partial pension,actuarial reduction) (1 point); exit trend for men (m) and women (w) (see Ebbinghaus 2006: Table 4.8).

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 280


The protection-pull thesis

Seen from the pull perspective, the combination of multiple pathways –whether explicitly intended for early exit or not – shapes the overallpull towards early exit from work. By the mid-1980s, early exit trendshad peaked in most countries. Different early exit regimes emerged thatreflect considerable cross-national differences in the availability of earlyexit pathways (see Table 8.3). The Continental European welfare statesprovide the most generous and largest set of pre-retirement options,though there are intra-regime differences. Most importantly, the Frenchand Italian public pension schemes provide relatively early pension ben-efits for older workers (at age 60 or even earlier) and except for disabilitypensions provide relatively open exit pathways. The German systemgrants some early “normal” pensions (largely to women aged 60 withfew working years), while the Dutch basic pensions are paid out onlyfrom age 65. Although all Continental European welfare states have hadhigh levels of early exit from work since the late 1970s, the Latin welfarestates show particularly high early exit rates (before age 60) in compari-son with Germany and the Netherlands, this is also reflected in the indexof pathway availability.

The two “outlier” countries, Denmark and Britain, only partially fol-low the patterns of the Scandinavian universalist and Anglo-Americanliberal models, respectively (see Table 8.3). Denmark provides more mul-tiple pathways (5–6 index points for men and 6–7 points for womendepending on the period) than does Sweden, which provided mainlythe disability-unemployment pathway until the early 1990s. In the Dan-ish case, pre-retirement pay, in particular, turned out to be a major exitroute, as did disability pensions, especially for women. Early exit for bothmen and women around age 60 became increasingly important duringthe 1990s. Sweden has, at least thus far, relied on part-time work andpartial benefits, increasing the share of older working men who combineincome from work with pension benefits.

In Britain during the 1980s, early retirement was common for mendue to the high level of unemployment, the Job Release Scheme (JRS),and employer policies, but it is no longer common before age 65. Due toan earlier standard retirement age, British women do retire earlier thanmen, though insufficient means lead many to continue working, oftenpart-time up to age 65. Indeed, an increasing share of older workingmen combine work and public or private pension benefits (OECD 2001,p. 36). In respect to availability of pathways, British women have moreopportunities to retire early then men (not least because of the earlier

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 281


normal pension age), yet the available pathways has been larger for menduring the 1980s (Table 8.3). In comparison, Ireland has less institution-alized pathways than Britain. During the 1990s, old age unemploymentbefore 65 was common for Irish men, while for Irish women 65 was the“norm” (Blöndal and Scarpetta, 1998). Thus relatively similar welfarestates, Denmark and Sweden but also the United Kingdom and Ireland,show significant intra-regime differences in early retirement pathwaysand actual exit from work patterns, at least with respect to particulartime windows and gender variations. These differences are not merelythe outcome of dissimilar government policies, but the consequence ofdifferent strategies by the social partners and variations in the productionregimes.

Finally, the United States and Japan are two welfare regimes with aliberal-residual orientation that allow companies significant scope indetermining occupational welfare, while both provide limited accessto disability and unemployment benefits. In addition, pensions canbe drawn earlier but these are insufficient, requiring additional occu-pational benefits. Moreover, in both countries, employers have usedmandatory retirement rules in the past (Kimura et al., 1994). How-ever, there are still marked differences in employment rates betweenboth countries that reflect very different employer strategies and par-ticular private–public mixes: American workers may be induced toretire early through employer-provided plans or may be let go throughgeneral downsizing, while Japanese employers provide reemploymentopportunities for their “retired” workers. The United States has less devel-oped public pathways and leaves a larger role for private pathways (seeTable 8.3). In Japan, early retirement increases slowly up to age 60, whensome Japanese men and women exit, but most continue working untilat least 65. Japan and the United States are also the two countries inwhich older workers (aged 60–69) tend to combine work and some formof public or private pension benefits (OECD, 2001, p. 36). Together withIreland, the United States and Japan rank lowest on the index of pathwayavailability (see Table 8.3), and indeed range also among the countrieswith the lowest or medium level of early exit.

The production-push thesis

The protection-oriented pull thesis does not explain why the socialpartners played such an active role in bringing about and using earlyexit; we thus need to turn to the production-related push factors.Employers were thus far willing to co-finance early exit from work

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 282


in all countries (see Ebbinghaus, 2006), but particularly in liberal-residual welfare regimes that lack generous public pathways. Most ofthe larger British, Irish, American, and Japanese employers (in both pri-vate and public sectors) provide favorable early occupational pensions,topping-up of flexible pensions or special “window” plans. However,these employer-led pathways remain limited to larger (often unionized)companies, and firms have increasingly retreated from defined benefitpensions and long-term employment commitments. Private occupa-tional pensions play less of a role than the publicly provided early exitpathways in the Continental European and Scandinavian welfare states,but also because social partners run schemes beyond the firm.

Among the push factors that explain the trend toward early exit fromwork are structural changes: deindustrialization and the growth to lim-its of public employment. However, these general shifts cannot explainthe cross-national variations in early exit trajectories in a systematic way,although the particular expansion and stagnation of the public sector inDenmark and Sweden provides one reason for the specific Nordic pat-tern. Although some special early retirement schemes were introducedto facilitate industrial restructuring, early exit from work has become amuch broader social right, common across nearly all private industryand public service sectors. However, the analysis of age-related skill lev-els shows that older workforces tend to be less skilled, and that thosewith lower skills have been shed at a higher rate than others.

Although all firms seem to have an interest in using early exit, thereare differences according to the particular partnership traditions, labormarket regulations, production strategies, and economic governancestructures (see Table 8.4). Taking the ideal-typical production models,we can derive two different sets of push forces. Under Liberal MarketEconomies (the United Kingdom, Ireland, the United States), Fordistmass production tends to rely more heavily on general skills, while hir-ing and firing is largely unregulated. As labor turnover is more common,employers use occupational pensions to retain skilled workers. Further-more, in unionized firms, unions defend seniority rights (“last in–firstout” rules) and employers thus have to buy out older workers with“golden handshakes.” Given liberal corporate and financial governance,short-termism puts additional pressure on these firms to be numericallyflexible during downturns. Therefore, early exit patterns in LMEs tendto be cyclical, as some companies used firm-sponsored early retirementplans to downsize during economic downturns, while others shed olderworkers in largely unregulated labor markets, pushing the risks of findingwork (or income support) onto individuals.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 283


Table 8.4 Early exit from work and regime configurations

Women: High and High Moderate LowMen: very early

High andvery early

France*Italy*

Germany,Netherlands

Denmark Sweden

UK** USA**

Ireland**

Japan

High

Moderate

Low

Notes: welfare regimes: conservative; universalist; liberal-residual; production system(underlined): liberal market economy; non-liberal market economy; labor relations (type face):cooperative; ∗contentious; ∗∗voluntarist.

In sharp contrast, Coordinated Market Economies tend to be character-ized by firms with specialized or functionally flexible production meth-ods, requiring industry or company-specific skills and well-developedvocational training (Estevez-Abe et al., 2001). Labor rights are more insti-tutionalized: Employment is highly regulated, collectively negotiatedwages tend to be high and more egalitarian, and workers have co-decisionrights, particularly in employment matters. Corporate governance fol-lows a stakeholder model that gives some co-decision rights to workerrepresentatives at company level. Furthermore, financial and corporategovernance has thus far consistently provided long-term patient capital.Therefore, during economic downturns, companies could hoard skilledworkers, instead of downsizing immediately. However, given the senior-ity wage system and the need for costly skill upgrading, employers seekto induce timely retirement in order to enhance efficiency and maintaininternal labor markets. Given the institutionalized workers’ rights andmore favorable public benefits, early retirement is a socially acceptablemeans for restructuring; it is likely to receive the support of workers andworkplace representatives.

Nevertheless, both early exit practices are not in equilibrium. They fol-low either a downward or an upward spiral. In the case of the shareholdermodel in LMEs, financial market pressure has increasingly pushed com-panies to withdraw from their commitment to defined benefit pensions.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 284


Moreover, unions have lost in bargaining power and can hardly enforceseniority rights. Thus, the burden of continued downsizing is placed onindividuals and this cost has been amplified by retrenchment in liberalwelfare states. On the other hand, the stakeholder employment modelin CMEs has led to a self-defeating, self-reinforcing spiral of early exitfrom work. Unions are still strong enough to defend social rights, wel-fare states still provide public exit pathways that allow externalizationof costs, and employers themselves find labor shedding of older workersthe easiest solution to maintain the internal labor market. Employers’belief in the diminished productivity of older workers leads to the self-fulfilling prophecy of ever shorter employment tenure: Older workers areless productive; therefore, they are shed earlier, but because employersexpect workers to retire early, they stop investing in training at earlierstages, and thus older workers have obsolete skills, which indeed makethem less productive.

However, the country case studies show two exceptions to the viciouscircle of the internal labor market model: Sweden and Japan maintainedhigher levels of activity in older age groups, at least until the 1990s. InSweden, the combination of partial pensions and part-time work allowedfirms to retain older workers and their expertise. The Swedish work inte-gration strategy has been highly contingent on the generosity of thepublic benefits, the willingness of employers to reorganize work, the sup-port of local unions at workplace level, and activation policies by publicemployment agencies. In fact, this model came under severe pressurewith the end of full employment in the 1990s and the fiscal crisis ofthe Swedish welfare state, which cut back on partial pension benefits.In recent years, the gradually phased-in pension reforms and future skillshortage may once again lead to a return to the prolonged working lifemodel (Wadensjö, 2002).

A very different social practice is at the heart of the Japanese partialexit model. Larger firms enforce mandatory retirement around age 60,but they provide reemployment at a lower wage or “secondment” tosmaller supplier firms. In addition, firm-sponsored benefits are providedas income supplements to the lower earnings, until public pension age.This model also came under pressure in the 1990s, as Japanese employerswere more reluctant to fulfill their “social responsibility” and the pensionreforms gradually postponed pensions to cope with the country’s rapidlyaging population. The importance of a long working life remains, but itmay entail further segmentation into a primary core workforce (“com-pany men”) and a secondary labor market made up of women, oldermen, and the unemployed (Brinton, 1998).

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 285


The analysis of push factors complements and fine-tunes the analy-sis of protection-induced pull factors. The basic claim of the varietiesof capitalism (Hall and Soskice, 2001) approach seems to be valid: Theinteractions of the different institutional arrangements pose specificincentive structures and follow different institutional logics at firm-level.Firms in Liberal Market Economies are more likely to suffer cyclical down-sizing pressure and early exit from work remains a partly internalizedfirm strategy or individualized risk of unemployment and poverty in oldage. In contrast, firms in Coordinated Market Economies face the contin-gencies of seniority wage, high skill, highly regulated, and cooperativeworkplace relations. They seek to buy out older workers in a sociallyacceptable way, supported by multiple possibilities to externalize theserestructuring costs (Naschold and de Vroom, 1994). The analysis of eco-nomic push factors also suggests that a reversal in early exit trends cannotbe achieved merely by cutting back on public pathways. As long as firmsare compelled to downsize or restructure, labor shedding will continue –a particular challenge to public policy that seeks to reverse the trends ofearly exit from work.

Conclusion

This comparative historical analysis of early exit from work trajecto-ries has shown the interaction of specific institutional configurations inrespect to protection, production and partnership. In an “ordinal com-parison” (Mahoney, 2003) we found a rank order of early exit from workpatterns ranging from labor-shedding Continental Europe (France andItaly with high and very early exit, Germany and the Netherlands withhigh early exit) to intermediate levels in Liberal and Scandinavian welfarestates with particularly higher exit levels for women in Denmark and UK,somewhat lower levels in Sweden and the USA, and relatively low levelsin Ireland and Japan. In this cross-case comparison presented here, thesevariations in early exit from work can be explained as a combination ofboth “pull” and “push” factors, welfare state regimes and productionssystems, while from an actor constellation perspective process-orientedwithin-case analysis would add the mediating factor of employers andworker representatives.

The particular welfare states provided different numbers of pathwaysto early retirement; the “pull” argument holds for the major differencebetween Continental and Liberal welfare regimes, but this perspectivefails to account for the intra-regime differences and the late or mediumlevel in early exit patterns of the generous Scandinavian welfare regimes.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 286


The integrative policies of the Swedish welfare state do not followthe incentive-pull argument, but relies on the cooperation by work-place actors in using gradual retirement and reintegration strategies topostpone exit from work. The Danish and British case also deviated attimes from the respective regime, requiring a more detailed intra-regimeanalysis of the general unions in promoting early exit for unskilled work-ers in Denmark and gendered social policies, facilitating earlier exit forwomen in both countries.

The “push” thesis has also proven a partial explanatory factor for themore market-driven cyclical exit patterns in the LME countries, and theuse of early retirement to meet the internal labor market problems ofCME firms. Yet the Varieties of Capitalism thesis needs amendment toaccount for the high Japanese and medium Scandinavian activity ratesamong older workers. Japanese and Scandinavian firms face similar prob-lems as their German counterpart due to internal labor markets with highseniority wages, but they rely much less on an externalization of costsonto public early retirement policies than in Continental Europe. TheJapanese re-employment of older workers is a particular arrangementthat is consistent with both early retirement from career-jobs and a highemployment level among older workers.

The role of employers and social partners in providing additional path-ways also requires the consideration of labor relations at national andworkplace level. More detailed case studies (see Ebbinghaus, 2006) indi-cate the crucial role of employers and workers’ organizations in bringingabout and maintaining early retirement practices at both national andworkplace levels. Moreover, policy reversal is hardly possible without theconsent of trade unions and workplace representatives.

Comparing different institutional configurations thus allowed an anal-ysis of “pull” and “push” factors across countries and over time, showingthat both perspectives can be utilized to explain early exit patterns. Infact, I have argued that they are often complementary, thus while we canfind evidence for the support of a “pull” of more or less generous wel-fare states on early exit, it would be shortsighted to ignore that publicpathways are used by employers and workplace actors to mediate the eco-nomic “push.” Thus, these inter- and intra-regime comparisons allow amore systematic analysis that summarizes the findings from within-caseanalysis and sheds new insights on the interpretation of cases in light ofthe cross-case analysis. The case-oriented analysis requires taking “out-liers” more seriously than in quantitative large-N studies, thus it pointsat the need to study deviations from the expected regime logic by moreintensive within-case analysis.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 287


An analysis of reform processes (see Ebbinghaus, 2006) reveals theimportance of intra-regime analysis, highlighting interesting casesof path departure of some reform processes from the “impasse” ofpath-dependent inertia. Although past practices provide major obstaclesfor reform as actors at various levels have grown accustomed to earlyretirement, recent reform efforts have led to a slow change in employ-ment rates. Some countries have accelerated their way out of the impasse(most notably the Netherlands and Denmark), yet some still remain stuckin an undecided switch of direction. A major reason for this difficulty inreversing the course of early exit from work is the institutionalization ofearly retirement practices in welfare state and production systems as wellas the interest coalitions of workers and employers supporting these.

References

Atchley, Robert C. 1982. “Retirement as a Social Institution.” Annual Review ofSociology 8: 263–87.

Blöndal, Sveinbjörn and Stefano Scarpetta. 1998. “The Retirement Decision inOECD Countries.” OECD Economic Working Papers 202.

Blossfeld, Hans-Peter, Sandra Buchholz, and Dirk Hofäcker, eds. 2006. Globaliza-tion, Uncertainty and Late Careers in Society. London: Routledge.

Brinton, Mary C. 1998. “Institutional Embeddedness in Japanese Labor Markets.”Pp. 181–207 in The New Institutionalism in Sociology, edited by Mary C. Brintonand Victor Nee. New York: Russell Sage Foundation.

Casey, Bernard. 1992. “Redundancy and Early Retirement: The Interaction ofPublic and Private Policy in Britain, Germany and the USA.” British Journal ofIndustrial Relations 30: 425–43.

Crouch, Colin. 1993. Industrial Relations and European State Traditions. Oxford:Clarendon Press.

Dogan, Mattei and Dominique Pelassy. 1990. How to Compare Nations: Strategiesin Comparative Politics. London: Chatham House.

Ebbinghaus, Bernhard. 2005. “When Less is More: Selection Problems in Large-Nand Small-N Cross-National Comparison.” International Sociology 20: 133–52.

Ebbinghaus, Bernhard. 2006. Reforming Early Retirement in Europe, Japan and theUSA. Oxford: Oxford University Press.

Ebbinghaus, Bernhard and Philip Manow, eds. 2001. Comparing Welfare Capital-ism: Social Policy and Political Economy in Europe, Japan, and the USA. London:Routledge.

Esping-Andersen, Gøsta. 1990. Three Worlds of Welfare Capitalism. Princeton, NJ:Princeton University Press.

Esping-Andersen, Gøsta. 1996. “Welfare States without Work: The Impasse ofLabour Shedding and Familialism in Continental European Social Policy.”Pp. 66–87 in Welfare States in Transition: National Adaptations in Global Economies,edited by Gøsta Esping-Andersen. London: Sage.


9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 288


Estevez-Abe, Margarita, Torben Iversen, and David Soskice. 2001. “Social Pro-tection and the Formation of Skills: A Reinterpretation of the Welfare State.”Pp. 145–83 in Varieties of Capitalism: The Institutional Foundations of ComparativeAdvantage, edited by Peter A. Hall and David Soskice. New York: OxfordUniversity Press.

Gruber, Jonathan and David A. Wise, eds. 1999. Social Security and Retirementaround the World. Chicago: University of Chicago Press.

Hall, Peter A. and David Soskice. 2001. “An Introduction to Varieties of Cap-italism.” Pp. 1–68 in Varieties of Capitalism: The Institutional Foundations ofComparative Advantage, edited by Peter A. Hall and David Soskice. New York:Oxford University Press.

Kimura, Takeshi, Ikuro Takagi, Masato Oka, and Maki Omori. 1994. “Japan:Shukko, Teinen and Re-employment.” Pp. 247–307 in Regulating Employmentand Welfare: Company and National Policies of Labour Force Participation at the Endof Worklife in Industrial Countries, edited by Frieder Naschold and Bert de Vroom.Berlin: W. de Gruyter.

Kitschelt, Herbert, Peter Lange, Gary Marks, and John Stephens. 1999. “Con-vergence and Divergence in Advanced Capitalist Democracies.” Pp. 427–60 inContinuity and Change in Contemporary Capitalism, edited by Herbert Kitschelt,Peter Lange, Gary Marks, and John Stephens. New York: Cambridge UniversityPress.


Kohli, Martin and Martin Rein. 1991. “The Changing Balance of Work and Retire-ment.” Pp. 1–35 in Time for Retirement: Comparative Studies on Early Exit from theLabor Force, edited by Martin Kohli, Martin Rein, Anne-Marie Guillemard, andHerman van Gunsteren. New York: Cambridge University Press.

Kohli, Martin, Martin Rein, Anne-Marie Guillemard, and Herman van Gunsteren,eds. 1991. Time for Retirement: Comparative Studies on Early Exit from the LaborForce. New York: Cambridge University Press.

Mahoney, James. 2003. “Strategies of Causal Assessment in Comparative Histori-cal Analysis.” Pp. 337–72 in Comparative Historical Analysis in the Social Sciences,edited by James Mahoney and Dietrich Rueschemeyer. New York: CambridgeUniversity Press.

Mahoney, James and Dietrich Rueschemeyer. 2003. “Comparative HistoricalAnalysis: Achievements and Agendas.” Pp. 3–38 in Comparative Historical Anal-ysis in the Social Sciences, edited by James Mahoney and Dietrich Rueschemeyer.New York: Cambridge University Press.

Naschold, Frieder and Bert de Vroom, eds. 1994. Regulating Employment and Wel-fare: Company and National Policies of Labour Force Participation at the End ofWorklife in Industrial Countries. Berlin: W. de Gruyter.

OECD. 2001. Ageing and Income: Financial Resources and Retirement in 9 OECDCountries. Paris: OECD.

Przeworski, Adam and Henry Teune. 1970. The Logic of Comparative Social Inquiry.New York: Wiley.

Ragin, Charles C. 1987. The Comparative Method: Moving Beyond Qualitative andQuantitative Strategies. Berkeley, CA: University of California Press.

Rein, Martin and Lee Rainwater, eds. 1986. Public/Private Interplay in SocialProtection: A Comparative Study. Armonk, NY: M.E. Sharpe.

9780230_202573_09_cha08.tex 5/6/2008 18: 29 Page 289


Rokkan, Stein. 1999. State Formation, Nation-Building and Mass Politics in Europe:The Theory of Stein Rokkan. Oxford: Oxford University Press.

Scharpf, Fritz W. 2001. “Employment and the Welfare State: A ContinentalDilemma.” Pp. 270–83 in Comparing Welfare Capitalism: Social Policy and PoliticalEconomy in Europe, Japan, and the USA, edited by Bernhard Ebbinghaus and PhilipManow. London: Routledge.

Schils, Trudie. 2005. Early Retirement Patterns in Europe: A Comparative Panel Study.Amsterdam: Dutch University Press.

Settersten, Richard A. and Karl Ulrich Mayer. 1997. “The Measurement of Age,Age Structure, and the Life Course.” Annual Review of Sociology 23: 233–61.

Tilly, Charles. 1984. Big Structures, Large Processes, Huge Comparisons. New York:Russell Sage.

Wadensjö, Eskil. 2002. “Active Strategies for Older Workers in Sweden.” Pp. 381–402 in Active Strategies for Older Workers, edited by Maria Jespen, David Foden,and Martin Hutsebaut. Brussels: ETUI.

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 290

9Identifying the Causal Effect ofPolitical Regimes on EmploymentAdam Przeworski

9.1 Introduction

The question studied here is whether political regimes, dichotomized asdemocracies and autocracies, affect the rate of growth of employment.But broader issues are at stake.

The central claim of “new institutionalism” is that institutions are theprimary cause of economic development. The theoretical program hasbeen laid out by North (1997, p. 224, italics supplied): “To make senseout of historical and contemporary evidence, we must rethink the wholeprocess of economic growth . . . The primary source of economic growthis the institutional/organizational structure of a political economy . . .”(For similar assertions, see Rodrik, Subramanian, and Trebbi, 2002 andAcemoglu, 2003.) Yet the new institutionalism also recognizes that insti-tutions are endogenous. As already North and Thomas (1973, p. 6)observed, “new institutional arrangements will not be set up unless theprivate benefits of their creation promise to exceed the costs.” The embar-rassingly obvious thought is that if endogeneity is sufficiently strong,causal effects of institutions cannot be identified. Imagine that only someparticular institutions exist under the given conditions. Then the effectsof institutions cannot be distinguished from the effects of the conditionsunder which these institutions are found.

Consider the substantive question posed above in the context ofthe OECD countries. Since almost all of them had democratic regimesbetween 1950 and 1990 – the period studied here – it is not possible todetermine whether the slow rate of growth of labor force in these coun-tries, on the average 0.97 per annum as contrasted with 2.32 in the rest ofthe world, is the effect of democracy or of the high productivity of theirlabor force, on the average $19,257 per worker as opposed to $5,931 inother countries,1 or perhaps of the already higher levels of participation,on the average 44 percent as compared to 40 percent elsewhere.

290

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 291

Adam Przeworski 291

Whether the effect of political regimes can be identified in the worldas a whole is the question pursued below. It is important, however, tokeep in mind that identification is not a matter of sample size but ofendogeneity. The reason causal effects of political regimes are next-to-impossible to identify among the OECD countries is not a small numberof observations, but the fact that history has mischievously eliminatedautocracy in developed countries. The logic entailed in identifying causaleffects does not depend on N (see Fearon, 1991). Even if we are analyzinga single observation, we need to distinguish the effect of a cause fromthe effect of the conditions that activated this cause. Did the French Rev-olution generate little social change, as Tocqueville (1964 [1856]) wouldhave it, because revolutions result in little change or because they occuronly in countries resistant to change?

The generic problem in identifying causal effects is how to answer thecounterfactual question: what would have occurred had the cause beenabsent? But to engage in counterfactual inferences we need some system-atic criteria to choose among several plausible candidates (Hawthorn,1991). For example, the argument that colonialism had a positive effecton economic development of the colonies is based on the counterfac-tual hypothesis that these colonies would not have developed withoutforeign penetration, while claims that colonialism had a pernicious eco-nomic effect are based on the premise that they would have developedhad they been left alone.2 Whether we can successfully solve such prob-lems is, in my view, largely a matter of luck, namely whether history hasbeen kind enough to generate observations that can be used to informus about the plausible counterfactuals. Hence, some causal effects maybe identifiable, while others may not be.3

Since this is mainly a methodological chapter, the theory is intro-duced rather briefly in section 9.2, only to motivate the statistical modelto be estimated. Section 9.3 emphasizes that to identify causal effectsit is necessary to make assumptions about counterfactuals. Section 9.4presents different biases that may be present due to non-random assign-ment (“selection”) of causes to exogenous conditions. Section 9.5 is areview of estimators designed to avoid some of these biases. In section 9.6these estimators are applied to the substantive problem at hand. Finally,section 9.7 focuses on the effects of globalization.

9.2 Growth in the labor force

Assume a Cobb-Douglas economy with constant returns to scale, of theform

Yt = AtF(Kt , Lt ) = AtKαt L1−α

t . (9.1)

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 292


1.5

22.

5

Rat

e of

gro

wth

of l

abor

forc

e

1950 1960 1970 1980 1990

Year

Lowess smooth. Source: PWT5.6

Figure 9.1 Average rate of growth of employment in the world, 1950–1990

The demand for labor in this economy is

L∗t =

((1 − α)A

w

)1/α

K, (9.2)

where w is the wage rate per unit of L. The rate of growth in the laborforce is thus given by

LL

= KK

+ 1α

(AA

− ww

), (9.3)

where the dots indicate time derivatives. One way to read this expressionis that labor force grows at the same rate as the capital stock as long asincreases in wages follow exactly the increases in Hicks-neutral produc-tivity, Å/A. In turn, if wages grow slower than productivity, the growthof employment is faster than the growth of capital stock.

The average rate of growth of employment in the world between 1950and 1990 is presented in Figure 9.1.4

Given (9.3), we can use the series for labor force and capital stock tocompare the rate of growth of productivity to that of wages. Figure 9.2shows that until the mid-1970s average increases of wages largely out-paced average increases of productivity, but this difference was rapidlyreduced so that by 1990 wages and productivity grew at the same pace.

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 293

Adam Przeworski 293

�8

�6

�4

�2

0

Diff

eren

ce

1950 1960 1970 1980 1990

Year

Lowess smooth. Source: PWT5.6

Figure 9.2 Average difference between productivity and wage growths

To introduce the effect of political regimes, assume that autocraciespay lower wages than democracies. The prima facie evidence for thisassumption is based on Rodrik (1998) as well as Przeworski et al. (2000),who found that labor shares are lower in autocracies than in democraciesat the same income levels. These data are reproduced in Figure 9.3.

The labor share data, however, are scarce, cover only the manufactur-ing sector, and are highly unreliable.5 Hence, I will think in reduced formterms, assuming that the growth of wages is higher in democracies:

ww

= θ ∗ REG, (9.4)

θ > 0, where REG = 1 if the political regime at time t is a democracy andREG = 0 otherwise. Letting L

L ≡ γL, KK ≡ γK , and A

A ≡ γA, we get

γL = γK + 1α

γA − θ

αREG, (9.5)

where θ/α is the causal effect of political regimes on the growth ofemployment.

We seek to identify this causal effect using data from 135 countriesbetween 1950 and 1990.6 Assume that the rate of technical progress is

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 294


Autocracies

Democracies

United States

2030

4050

60

Labo

r sh

are

0 10000 20000 30000 40000

Product per worker

fpfit. Shaded areas are 95% confidence intervals.

Figure 9.3 Labor share as a function of product per worker, by regime

constant over time (but it may vary across countries or across regimes).Then we can write (9.5) as

γL(it) = β0 + β1γK(it) + β2REG(it) + e(it), (9.6)

where β0 = 1αγA (or 1

αγA(i)), β1 = 1/α, and β2 = −θ/α.

The question is whether it is possible to identify β2 when regimes areendogenous.

9.3 The problem

First, we need some notation. Let T stand for the (potential) cause,where T = 1 indicates “treatment” and T = 0 “control” (or a differenttreatment).7 Without a loss of generality, we will think of democracy(REG = 1) as the treatment and of autocracy (REG =) as the control.X and V are “covariates,” that is, traits of an individual unit prior tothe application of the treatment. X is the vector of covariates observedby the researcher, V are covariates not observed. Y = {Y0, Y1} is the vari-able subject to the potential effect of the cause, where Y0 stands for statesof the units not exposed to treatment and Y1 of those exposed to it, so

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 295

Adam Przeworski 295

that for each unit i we observe either Y1 or Y0:

Yi = TiY1i + (1 − Ti)Y0i. (9.7)

A “unit” is an opportunity for the cause to operate. It may be an indi-vidual, a country, or what not. Moreover, it may be the same individualor a country in a different state: say Sweden in 1950 and in 1951. Hence,the “unit” is a full set of observable and unobservable covariates: i iscoextensive with the vector of “background conditions” (xi, vi).

Now, what is the causal effect of treatment on the particular unit i,the Individual Treatment Effect? This effect is defined as the differencebetween the states of an individual unit when it is subjected and notsubjected to the operation of the cause, in our case between the rate ofgrowth of labor force of a particular country at a particular time underdemocracy and autocracy. Formally,8

ITEi = y1i − y0i ≡ βi (9.8)

But even if the assignment of regimes to countries were random,this question could not be answered without making some assumptionsabout hypothetical situations that would have occurred had a countrythat did not get treatment (had not been exposed to the potential causes)received it or had a country that did receive treatment not received it.Since these states did not occur, they are contrary to fact, counterfactual.9

And since counterfactuals cannot be observed, assumptions about coun-terfactuals cannot be directly tested.10 Hence, the effect of a cause onan individual unit cannot be determined without making assumptionsabout counterfactuals. These assumptions cannot be tested.

What assumption would identify the individual treatment effect underrandom assignment?

Assumption 1: Unit homogeneity (Holland 1986).For any i, j ∈ N,

if {xi, vi} = {xj, vj}, then y0i = y0j and y1i = y1j.

This assumption says that if any two units have the same values of covari-ates, they would have the same states under control and the same statesunder treatment. When this assumption is true, the process of selectioncan be ignored: it does not matter which of two identical units is subjectto treatment and which serves as control.

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 296


This assumption identifies the causal effect of treatment. Assume thatwe observed i in state 1 and j in state 0. Applying the homogeneityassumption yields

ITEi = y1,i − y0,i = y1,i − y0,j,

where now both y1,i and y0,j are observed.What does “identify” mean? Intuitively “to identify” is to be able to

infer relations among variables (or the parameters of multivariate distri-bution) on the basis of all the possible observations (Koopmans, 1949;in Manski, 1995, p. 6). But very often this is possible only by assumingsomething that may or may not be testable. As Manski (1995, p. 18)observed, “Theories are testable where they are least needed, and are nottestable where they are most needed. Theories are least needed to deter-mine conditional distributions P(y|x) on the support of P(x). They aremost needed to determine these distributions off the support.” We haveseen that since each unit can be observed only in one state at one timeit is not possible to identify the individual causal effect without makingsome assumptions. Hence, we need identifying assumptions, such as unithomogeneity. This assumption is not testable. But it seems reasonable.

Now we can ask about the Average Treatment Effect (ATE). Specifically,under what assumptions

βATE = E(Y1 − Y0|X) = E(β|X) = y1 − y0 = β,

so that the observed mean difference identifies the average treatmenteffect? The answer is “conditional mean independence”:

Assumption 2: Conditional Mean Independence.

E(Y1|X, T = 1) = E(Y1|X, T = 0) = E(Y1|X)

E(Y0|X, T = 0) = E(Y0|X, T = 1) = E(Y0|X)

This assumption says that conditional on observed covariates we canexpect the units not exposed to treatment to react to it identically tothose observed under treatment and the units exposed to treatment notto differ in their control state from those observed under control.11 Underrandom assignment this assumption is trivially true. And it implies thatthe observed difference identifies the average causal effect:12

β = E(Y1|X, T = 1) − E(Y0|X, T = 0) = E(Y1 − Y0|X).

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 297

Adam Przeworski 297

Hence, if the assignment to treatment is random, then the difference ofthe observed means identifies the average causal effect of treatment.

Now, let U stand for the effect of V on Y and assume linear separability.Then

E(Y |X, V) = E(Y |X) + U . (9.9)

Substituting into (9.7) (and dropping the i subscript) yields

Y = E(Y0|X) + T [E(Y1 − Y0|X)] + {T(U1 − U0) + U0}(9.10)

= β0(X) + β(X)T + U ,

where β(X) = E(Y1 − Y0|X) is the average causal effect, discussed furtherbelow.

To identify the causal effect, we need to ensure that

U = T(U1 − U0) + U0 = 0,

where U is the impact of unobserved factors in Y = β0(X) + β(X)T + Uand β(X) is the average causal effect conditional on X. The basic concernin identifying causal effects is thus whether E(U) = 0.

9.4 Potential biases

Baseline bias

Note first that the causal effect of interest need not be the effect on theaverage unit but on those units that are actually observed as treated.13

This estimand is typically referred to as the Average effect of Treatmenton the Treated (ATT), defined as

βATT = E(Y1 − Y0|X, T = 1). (9.11)

The value of this parameter tells us how the treatment changes the out-come for those unit that were observed as treated. Note that E(Y1|T = 1) isobserved, while E(Y0|T = 1) is the missing counterfactual. Now considerthe bias of the observed difference, β, as an estimator of βATT :

β − βATT = E(Y1|X, T = 1) − E(Y0|X, T = 0) − E(Y1 − Y0|X, T = 1)

= E(Y0|X, T = 1) − E(Y0|X, T = 0) (9.12)

= E(U0|T = 1) − E(U0|T = 0),

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 298


where the last expression is the difference in the control state betweenthose units that were treated and those that were not, typicallyreferred to as the “baseline bias.” Suppose, for example, that anomitted variable, say human capital, H , is correlated with the treat-ment and it affects the employment prospects of a country, so thatE(U0|H = high, T = 1) > E(U0|H = low, T = 0) . Since countries in countriesobserved under T = 1 employment would have grown faster under T = 0than those actually observed under T = 0, the observed difference overes-timates the causal effect of T . This bias is sometimes referred to as “the”selection bias, but we will see that there are other potential selectionbiases than the baseline bias.

Self-selection bias

Now, return to ATE. The bias of β as the estimator of βATE is

β − βATE = E(Y1|X, T = 1) − E(Y0|X, T = 0) − E(Y1 − Y0|X). (9.13)

Adding and subtracting E(Y0|T = 1) yields

β − βATE = {E(Y0|X, T = 1) − E(Y0|X, T = 0)}+ {E(Y1 − Y0|X, T = 1) − E(Y1 − Y0|X)}

(9.14)= {E(U0|T = 1) − E(U0|T = 0)}

+ {E(U1 − U0|T = 1) − E(U1 − U0)}.

The term in the first curly brackets is the by now familiar baseline bias.The term in the second brackets, in turn, is best thought of as “self-selection” bias. This term is the difference between the effect of treatmenton those who were actually treated and on the average unit. But whywould the effect of the treatment on the treated differ from its effect onthose who are not? One reason is that recruitment to treatment dependson something not observed by the researcher but anticipated by the unit.This will occur if units seek treatment for some reasons other than theX’s observed by the researcher or if they comply differently with thetreatment depending on the X’s. Suppose – I am not asking you to believeit – that political elites which opt for democracy also know how to makeemployment grow faster. Then the effect of democracy on the growth ofemployment for the countries observed as democracies will differ fromthe effect on the average country: a self-selection bias.

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 299

Adam Przeworski 299

Post-treatment bias: manipulability and attributes

Thus far we have assumed that the X′s and the V ′s, called here “covari-ates,” do not change with treatment. The assumption was that causes canbe manipulated one-at-a-time. But suppose that some of the covariates –call this subset A for “attributes” – change as the effect of treatment: thisis called “post-treatment effect” by King and Zeng (2002). Now the treat-ment may have two effects: a direct one and an indirect via A. We needsome identification assumptions to tell these two effects apart.

Can we always make such assumptions? Here we enter into a com-plex and subtle issue. According to Holland (1986), to qualify as apotential cause, the particular variable must be vulnerable to (potential)manipulation. The critical feature of the notion of cause is that differentvalues of the cause can be realized under the same background con-ditions. This is why attributes, such race or gender, cannot be causes.“Causes,” Holland says, “are only those things that could, in principle, betreatments in experiments” (1986, p. 954). What distinguishes statisticalassociation from causation is manipulability: “the schooling a studentreceives can be a cause, in our sense, of the student’s performance on atest, whereas the student’s race or gender cannot.” It makes no sense tosay “Joe earns $500 less than Jim because Joe is black,” since skin color(called “race” in the United States) cannot be manipulated. Causal infer-ence is concerned with the effect of causes under specific backgroundconditions (“on specific units”) and attributes cannot be manipulatedwithout changing these conditions.

This arguments confounds two propositions: (1) T cannot be manip-ulated and (2) T cannot be manipulated without changing A. The firstone says that we cannot change the skin color of an individual. The sec-ond says that we can change it but if we change it, we will also changeother characteristics of this individual (or the treatment of this individ-ual by others). The confusion becomes apparent when we read that “Anattribute cannot be a cause in an experiment, because the notion ofpotential exposability does not apply to it. The only way for an attributeto change its value [so it can be changed!] is for the unit to change insome way and no longer be the same unit” (Holland 1986, p. 954). Now,if (1) holds, it may still be true that there are other units that have thesame background conditions but a different value of T and we can usethe conditional mean independence assumption to identify the causaleffect. Only if (9.2) is true, does identification become impossible.

Consider an example closer to our practice: the location of a country inAfrica, which in many analyses appears to affect civil strife and economic

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 300


growth. Does it make sense to say that “the effect of Africa on growthis β”? “Africa” is clearly an attribute by Holland’s definition, a set ofrelated unobserved characteristics. If history had placed Zimbabwe inLatin America, it would have no longer been Zimbabwe: it would differin various ways that make Africa distinct from Latin America. Hence,relying on the Africa dummy to generate counterfactuals would generatea “post-treatment bias.”

King and Zeng (2002, p. 21) emphasize that controlling (matching)for variables that are endogenous with regard to treatment generatesbias. This can be seen as follows. For simplicity, suppose that assign-ment is random, so that there is no baseline or self-selection bias, butX1 = X0 + δT . Then conditioning on X,

E(Y1 − Y0|X) = E(Y1|X0 + δT) − E(Y0|X0)(9.15)

= E(Y1 − Y0|X0) + {E(Y1|X0 + δT) − E(Y1|X0)},

where the last term is the “post-treatment bias.” For example, sup-pose that capital stock grows slower under dictatorships. Conditioningon the growth of growth of capital stock would then generate post-treatment bias.

Non-independence bias: SUTVA

One final implicit assumption concerns independence of the Y variablesacross units. This assumption is called SUTVA, for “stable unit treatmentvalue.” Suppose that the units are individuals and that they learn fromone another, so that yi = f (yj). This means that the performance of thetreated may affect the performance of the untreated, or vice versa. InLucas’s (1988) growth model, a young plumber learns from the experi-enced one. Hence, if we take the difference in their productivity as theeffect of experience, it will be underestimated because of the externality.Or take T to be ‘export-oriented’ strategy. South Korea adopted this strat-egy early and had high growth rates. Brazil adopted it late. But supposethat Brazil had adopted it early: would the growth rate of Korea havebeen the same? If it would not have been the same, the values observedfor Korea under treatment depend on the realization of the treatmentvariable for Brazil: hence the Korean values are not “stable.”

In our context, this assumption is particularly dubious. In an openeconomy, the rate of growth of employment in one country depends onits growth in other countries. Hence, if country i that is an autocracyin which wages grow slowly and employment quickly (say China) were

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 301

Adam Przeworski 301

to become a democracy in which employment would grow slower, therate of growth of employment in other countries would accelerate. Oneneeds some kind of a world equilibrium model to identify the causaleffect when this assumption is violated.

9.5 Types of estimators

How can we identify causal effects when the assignment is not random?14

Basically, we can adopt two approaches: drop the observations that arenot “comparable,” restricting identification of causal effects to those thatare, or keep all the observations and generate hypothetical matches foreach of them. Matching procedures would eliminate (or give almostzero weights to) all the observations that do not have close matches,while procedures generating hypothetical counterfactuals would fill allthe growth cells for which history did not generate the information.

Matching

One way to proceed is to match on observables.15 Say we want to examinethe effect of guaranteed income programs on labor supply. We observesome wealthy countries with such programs (Revenue minimum d’insertionin France) and many countries, rich and poor, without them. We wouldnot want to match the wealthy treatment cases with controls from poorcountries. Hence, we use as controls countries with comparable per capitaincome, and restrict our causal inference to such countries.

Matching takes the assignment of causes as given and calculates causaleffects conditional on the assignment of causes realized by history,relying on the conditional mean independence assumption

E(Yj|X, T = j) = E(Yj|X) ∀j, (9.16)

which says that the value of Y in any state j does not depend on the stateT in which a unit is observed once it is conditioned on the observedcovariates. This is the same assumption as conditional mean indepen-dence introduced above, but written more generally to emphasize thatthe cause may assume any set of values.

Matching estimators are vulnerable to two problems:

(1) Dropping observations reduces the scope of generality. Sometimes, asin the example of minimum income programs, this is not a loss. It isnot a loss because the probability that a poor country would institutethese programs is zero: poor countries cannot afford such programs,

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 302


Democracies

Autocracies

0.5

11.

52

2.5

Labo

r fo

rce

grow

th

0 5000 10000 15000 20000

Per capita income

Shaded area is the 95% confidence interval

Figure 9.4 Growth of employment as a function of per capita income, by regime

so that the question how these programs would affect labor supplyin poor countries is moot. But how should we proceed when thisprobability is positive under all conditions, yet very differently dis-tributed with regard to these conditions, as in the case of politicalregimes? What to do with observations without a close match? Fig-ure 9.4 shows a semi-parametric (fractional polynomial) regressionof labor force growth on per capita income, by regime. Since somedemocracies are located in the range where there are no autocracies,different matching algorithms will either drop these observations ofdemocracies or assign to them very low weights.16 In either case,we have to worry whether the causal effect is the same for thoseobservations with close matches and those without them. Moreover,as King and Zeng (2002) emphasize, extrapolations out of range ofcommon support are highly sensitive to the form of the function.

(2) We can match on observables. But should we not worry about unob-servables? Suppose that leaders of some countries go to study inCambridge, where they absorb the ideals of democracy and learnhow to promote employment. Leaders of other countries, how-ever, go to the School for the Americas, where they learn how torepress and nothing about economics. Autocracies will then gener-ate lower growth because of the quality of the leadership, which is notobserved. Since this is a variable we could not observe systematically,

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 303

Adam Przeworski 303

we cannot match on it. And it may matter. Conditional mean inde-pendence – the assumption that unobserved factors do not matter –is very strong, and likely to be often false in cross-national research.

All that was said about matching applies to regression models thatcontrol for the observables. Matching is just a non-parametric regression:both generate means of Yconditional on X and T . Moreover, as observedrespectively by Manski (1995) and Achen (1986), both matching andparametric regressions that control for observables may in fact exacerbatethe biases due to selection on unobservables.

Both matching and parametric regression estimates can be subjected tosensitivity analysis. Given assumptions about the unobservables, one cancalculate the range of estimates that are compatible with the observeddata (Manski, 1995). Rosenbaum (2002, chapter 4) presents methodsfor quantifying the sensitivity of the estimates of causal effects underdifferent assumptions. Obviously, the more plausible the assumptionand the narrower the bounds, the more credible is the estimate.

Instrumental variables

Instrumental variables estimator is based on the assumption of condi-tional mean independence in the form:

E(Yj|X, Z, T = j) = E(Yj|X, Z) ∀j. (9.17)

The idea is the following. Suppose that after conditioning on X, Yj stilldepends on j, in other terms that cov(T , U) �= 0. Now, suppose that thereis a variable Z, called an “instrument,” such that

cov(Z, T) �= 0 (9.18)

and

cov(Z, U) = 0. (9.19)

Then conditioning on X and Z satisfies (9.11). Thinking in regressionterms, let Y = f (Z) and T = g(Z). Then, by assumption (9.17), β in Y = βTis that part of the causal effect of T on Y which is independent of U .

To qualify as an instrument, a variable must be related to the cause andonly to the cause, so that its entire effect is transmitted by the cause. Notethat while the assumption that the instrument is related to the cause(conditional on all exogenous variables) can be and should be tested,

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 304


the assumption that it is independent of the conditions that also shapethe effect is not testable.

Instruments must be correlated with the cause. Weak instruments(those weakly correlated with the treatment) can generate biased esti-mates even with very large samples. But instruments cannot be toostrongly correlated with the cause. In the limit, if the instrument andthe cause are the same, the instrument is as endogenous as the cause:this is “the curse of strong instruments.” The causal effect cannot beidentified, because it is impossible to separate the impact of the causefrom that of the conditions that give rise to it.

In turn, the “exclusion restriction” (9.19) requires that the instrumenthave no effect that is not mediated by the cause. Moreover, given thatU = T(U1 − U0) + U0,

cov(Z, U) = cov(Z, U0) + cov(Z, T(U1 − U0)). (9.20)

Hence, the exclusion restriction has two parts, and Heckman (1996,2004) repeatedly makes the point that, even if cov(Z, U0) = 0 , in thepresence of unobserved self-selection the second covariance will notbe zero.

Selection on unobservables

Both matching and instrumental variables estimators condition onobserved covariates and both are vulnerable to the influence of unob-served variables that are correlated with the treatment. Another approachconditions on unobserved as well as on observed covariates. One wayto think of these estimators is that they emulate experiments, but dif-ferently than matching: not by eliminating observations that do nothave an observed match but by creating observations to match all theobserved values. The assumption is that if the conditioning is correct,then the resulting data have the same structure as if history had per-formed a random experiment assigning different values of treatment toeach unit. Since the conditional mean independence of the form

E(Yj|X, Z, V , T = j) = E(Yj|X, Z, V) ∀j (9.21)

holds whenever assignment is random, the only issue with regard tothese estimators is whether they correctly emulated random assignment.

The basic idea is the following. We first describe the process by whichthe observed assignment of causes was generated by history:

T∗ = Zα + V , T = 1(T∗ > 0), V ∼ (0, 1). (9.22)

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 305

Adam Przeworski 305

This equation says that the propensity toward being observed undertreatment depends on observable variables Z and unobserved factorsVand that we observe T = 1 if T∗ > 0.

Secondly, we exploit the possibility that cov(V , U) �= 0 , by expressingE(Uj|T = j) in

E(Yj|X, T = j) = E(Yj|X) + E(Uj|T = j), (9.23)

as

E(Uj|T = j) = θjE(V |T = j), (9.24)

where the latter expectation can be estimated from (9.22). Finally, wesubstitute, to obtain

E(Yj|X, T = j) = E(Yj|X) + θjE(V |T = j), (9.25)

which can be now estimated by least squares. The OLS coefficients ofE(Yj|X) = Xβj can be then used to generate counterfactual values of Yj forthe cases in which it is not observed, thus filling all the missing matches.Finally, for j = {0, 1},

βATE = E(Y1|X) − E(Y0|X) = (β1 − β0)X,

is the estimator of the average causal effect.Note that we still have to be concerned about strong endogeneity of

treatment. In principle, it has to be true that 0 < Pr (T = 1|Z) < 1 ∀Z.

Otherwise, the counterfactuals cannot be realized given the mecha-nism by which history assigns treatments, so that the entire exerciseis moot. The main vulnerability of this class of estimators stems fromthe untestable assumption about the joint distribution (V , U1, U0).

9.6 Political regimes and the growth of employment

With this background, we return to the effect of political regimes onthe growth of employment between 1950 and 1990. We observe 1,595democratic years and 2,396 autocratic ones. The mean rate of growth oflabor force under democracy was 1.59 (s.d. = 1.20) and under autocracy2.28 (s.d. = 1.94), for a difference of −0.69. Obviously, comparing meansis the same as regression, so it generates the same result, with a standarderror of 0.05. Yet we already know that the difference of observed means

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 306


0.0

001

.000

2.0

003

.000

4.0

005

Den

sity

0.0

001

.000

2.0

003

.000

4.0

005

Den

sity

0 5000 10000 15000 20000

Level

Democracies

0 5000 10000 15000 20000

Level

Autocracies

Figure 9.5 Density of per capita income, by regime

is a biased estimator of the average causal effect if the assignment ofregimes is not random, and it is easy to see that it is not.

Figure 9.5 shows the density of per capita incomes by regime. As onewould expect, autocracies tend to be poor, while democracies can befound at all income levels. Indeed the wealthiest autocracy in the dataset, Singapore in 1990 with per capita income of $11,698, was poorerthan 200 years of democracies, with the US in 1989 leading the list at$18,095 in 1989.

Moreover, Figure 9.6 shows that democracies were somewhat morefrequent during the years when the average rate of growth of labor forcein the world was lower.

To identify the causal effects of regimes, we must, therefore, distin-guish it from the effect of the conditions under which these regimeswere found. To do so, we will augment the theoretically derived specifi-cation given in (9.6) by adding some controls. These include the laggedproportion of the population that is employed (since labor force cannotgrow when everyone is employed), the average rate of growth of laborforce in the world during a particular year (as a crude attempt to take intoaccount the world equilibrium effects), and per capita income (as a crudeattempt to control supply effects, on the assumption that preference for

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 307

Adam Przeworski 307

0.5

11.

52

Den

sity

0.5

11.

52

Den

sity

1 1.5 2 2.5 3

lfg_world

Democracies

1 1.5 2 2.5 3

lfg_world

Autocracies

Figure 9.6 Density of average world employment growth, by regime

leisure increases in income). Hence, we will be estimating models of theform:

lfg = β0 + β1ksg_lag + β2lfprop_lag + β3lfg_world + β4level + βATEREG + e,(9.26)

where lfg stands for labor force growth, ksg_lag is the lagged value ofcapital stock growth,17 lfprop_lag is the lagged value of labor force pro-portion in the population, lfg_world is the mean rate of growth of laborforce in the world in a particular year, and level is income per capita. Theparameter of interest is βATE.

In addition, in the instrumental variables and the selection on unob-servables models, we will augment (9.26) by a selection equation of theform

Pr(REG = 1) = Pr(Zα + V > 0) = F(Zα). (9.27)

The probit model uses the specification based on Przeworski et al.(2000). It includes per capita income (levlag), proportion of countries inthe world that are democracies in a particular year (odwp), the number ofcompleted spells of democracy in the history of the country (stra), whereall these values are lagged one year, and interactions of these variableswith the lagged regime (autocracy = 1).

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 308


Table 9.1 Different estimates of the average treatment effect

Estimator βATE s.e. t p

OLSa −0.27 0.110 −2.56 0.010Fixed effects −0.15 0.093 −1.63 0.104Match-Kernelb −0.29Match-Neighborb −0.30Match-Stratab −0.29Match-Imbensc −0.18 0.074 −2.44 0.015Heckmand −0.26 0.015 −17.33 0.000Heckmane −0.29 0.070 −4.14 0.0002SLSf −0.29 0.070 −4.13 0.000

Notes: a Panel corrected standard errors. Other OLS results are similar. b Standard errors are notgiven for these matching estimators since the average treatment effect was calculated as theweighted average of ATT and ATC (see below). c Imbens nnmatch with 5 matches. d Heckmantwo-steps estimator, with separate regressions for each regime. e Heckman estimator with allthe observations considered together. f With propensity score as the instrument. Using allthe instruments separately generates an almost identical result.

Table 9.2 Estimates of the effect of the treatment on the treated and on thecontrol group

Estimator Democracies as Autocracies Autocracies as DemocraciesβATT s.e. z βATC s.e. z

Kernel 0.43 0.11 3.93 −0.20 0.05 −3.62Neighbor 0.48 0.49 0.98 −0.19 0.12 −1.59Strata 0.39 0.08 4.74 −0.23 0.06 −3.92Heckman 0.44 0.02 27.7 −0.02 0.03 −0.81

The results are surprisingly robust.18 Here is a table that summarizesthe estimates of βATE.

Similar conclusions apply to the estimates of the effect of the treatmenton the treated (ATT) and the effect of the treatment on the control (ATC).Since we took the treatment to be democracy, the first estimates tell uswhat would have been the difference in the growth of employment forthe units observed as democracies had they been autocracies under theidentical conditions, while the second inform us what would have beenthe difference for the countries observed as autocracies had they beendemocracies (The signs are inverted to facilitate the interpretation.)

These estimates indicate that countries observed as democracies(which tend to be more developed) would have had a much fasteremployment growth had they been autocracies. In turn, the countries

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 309

Adam Przeworski 309

observed as autocracies would have had a somewhat slower employmentgrowth as democracies, but the difference is about a half of that for theobserved democracies.

Thus, from the methodological point of view, this exercise turnedout to be disappointing. There appear to be no selection biases withregard to the growth of labor force, so that all the estimators generatesimilar results.19 It seems safe to conclude that the growth of employ-ment is somewhat slower in democracies by some amount between 0.15and 0.30.

9.7 Globalization

Yet all the estimators we used are based on the assumption of stable unittreatment value, which is unlikely to be satisfied in a globalized economy.Note that all the parametric analyses indicate that the rate of growth oflabor force in each country is positively affected by its rate of growth inthe world as a whole. Suppose that a country experienced a transition todemocracy. Since employment grows slower in democracies, the worldaverage would become lower, thus affecting the rate of growth in eachcountry. This is clearly a violation of SUTVA, even if the bias it generatesmay be small.

To analyze the consequences of the mobility of capital and of com-modities, one would have to study a world economic equilibrium, whichI will not do. Instead, we can analyze the effect of the average growthof employment in autocracies in a particular year on the growth ofemployment in each of the democracies during this year.

The estimates are again robust for all parametric models (differentversions of OLS, IV, and Heckman), so I do not enter into details.Re-estimating all the parametric models with the effect of the meangrowth of employment in autocracies increases the estimate of the effectof democracy on the growth of employment from a ball park number of−0.27 to about −0.43. With only minor variations, these analyses showthat (1) labor force in democracies grows when employment increasesin autocracies, indicating that autocracies and democracies respondsimilarly to fluctuations in world demand, but (2) this effect interactsnegatively with per capita income. Since (using estimates from Heckman-two) the effect of growth in autocracies on the growth in democraciesis +0.52 and the effect of interaction with per capita income of eachof the democracies is −0.0393 per thousand, in a poor democracy likeIndia, with per capita income of about $1,000, the rate of growth oflabor force increases by 0.48, while in a wealthy democracy such as the

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 310


United States, with an income of $18,000, employment growth slowsdown by about 0.19. when employment in autocracies increases by 1%.Democracies with per capita income of 0.52/0.0393 = 13,343, the incomeof Switzerland in 1971, neither benefit nor lose when employment inautocracies increases. Note that in 1990 the unweighted average of percapita incomes in the OECD countries was $13,650. Thirteen countrieswith incomes equal or higher to that of Iceland were net employmentlosers because of competition with autocracies.

Here, then, is the story: At each income level, capital stock grows atabout the same rate in the two regimes. Yet autocracies repress wages;hence, they employ workers with lower marginal product; hence, theiremployment grows faster. In turn, this implies that democracies employonly workers with higher marginal product; hence, their employmentgrows slower. Since, as Figure 9.1 shows, the gap in wages opens up withper capita income, the effect of competition with autocracies is greaterin the more developed democracies.

Notes

1. All the dollar numbers are in 1985 purchasing power parity dollars from PennWorld Tables, release 5.6.

2. For a thoughtful discussion of this issue, see Kaniyathu (2006).3. For a more extensive discussion of these issues, see Przeworski (2006).4. The labor force series is obtained by dividing product per worker by product

per capita from PWT 5.6. According to the ratings of data quality providedby PWT, the reliability of the product series varies greatly across countries.Przeworski et al. (2000: Appendix 3.1) calculated that this quality is muchhigher in democracies but did not find data quality to be a source of bias.

5. Indeed, the World Bank stopped publishing them.6. This particular data set is used because this is the only period during which the

information about capital stock is available. Economic data are from PWT5.6and political data from Przeworski et. al. (2000).

7. Although for simplicity I assume that the cause is a binary variable, every-thing said here holds for any discrete or continuous values of T .

8. For simplicity, I will ignore time in the theoretical dicusssion.9. The idea of counterfactuals goes back to Pascal (1669, sec. 162): “Le nez de

Cléopâtre: s’il eût été plus court, toute la face de la terre aurait changé.” On thedistinctions among different types of conditional propositions, see Edgington(2001). On the logical problems with counterfactuals, see Quine (1953), Lewis(1973), Mackie (2002 [1973]), Goodman (1979), and Stalnaker (1987).

10. For a statistical view of causality without counterfactuals, see Dawid (2000),who rejects them as metaphysical.

11. To help with the notation, E(Y1|T = 1) is to be read as “the expected valueof the outcome under treatment, given that the units have been observed

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 311

Adam Przeworski 311

as treated,” while E(Y1|T = 0) as “the expected value of the outcome undertreatment, given that the units have been observed as not treated.”

12. According to a theorem by Rosenbaum and Rubin (1983), if the conditionalmean independence holds in the form specified in the text, then it also holdsin the form in which p(X) = Pr(T = 1|X) is substituted for X, where p(X) is the“propensity score.”

13. This effect is of particular interest in remedial policy programs. As Heckmanrepeatedly points out, it makes no sense to ask what would be the effect ofmanpower training program on millionaires. In turn, we want to know theeffectiveness of such programs for the people who need them and get them.

14. For overviews of estimators see Angrist and Krueger (1999), Berk (2004, Chap-ter 5), Dufflo (2002), Persson and Tabelini (2003, chapter 5), or Winshipand Morgan (1999). For reasons of space, I do not discuss difference-in-difference estimators, for which see Woolridge (2002) and Bertrand, Duflo,Mullainathan (2004).

15. On matching estimators, see Rosenbaum (2002), Imbens (2002), Becker andIchino (2002), and, more critically, Heckman (2004).

16. Depending on the algorithm, matching estimators treat differently obser-vations that cannot be matched exactly. When matching is restricted tocommon support or when it is confined to balanced strata, observations with-out a match are ignored. When some kind of distance measure is employed,distant matches obtain weights approaching zero.

17. Przeworski et al. (2000) performed various tests and found that the growthof capital stock is exogenous with regard to regimes.

18. The parametric estimators are not sensitive to the specification of the selec-tion equation but matching estimates become lower when the variable strais dropped from this equation.

19. In some other contexts, different estimators generate highly disparate con-clusions. See Przeworski (2006).

References

Acemoglu, Daron. 2003. “Root Causes: A Historical Approach to Assessing the Roleof Institutions in Economic Development.” Finance and Development 27–30.

Achen, Christopher. 1986. The Statistical Analysis of Quasi-Experiments. Berkeley,CA: University of California Press.

Amemyia, Takeshi. 1994. Introduction to Statistics and Econometrics. Cambridge,MA: Harvard University Press.

Angrist, Joshua D. and Alan B. Krueger. 1999. “Empirical Strategies in Labor Eco-nomics.” Chapter 23 in The Handbook of Labor Economics, volume III, edited byO. Ashenfelter and D. Card. Amsterdam: North-Holland.

Angrist, Joshua D. and Alan B. Krueger. 2001. “Instrumental Variables and theSearch for Identification: From Supply and Demand to Natural Experiments.”Journal of Economic Perspectives 15: 69–85.

Becker, Sascha O. and Andrea Ichino. 2002. “Estimation of Average TreatmentEffects Based on Propensity Scores.” The Stata Journal 7: 1–19.

Berk, Richard A. 2004. Regression Analysis: A Constructive Critique. Thousand Oaks,CA: Sage.

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 312


Bertrand, Marianne, Esther Duflo, and Sendhill Mullainathan. 2004. “HowMuch Should we Trust Differences-in-differences Estimates?” Quarterly Journalof Economics 119: 249–75.

Dawid, A.P. 2000. “Causal Inference without Counterfactuals.” Journal of theAmerican Statistical Association 95: 407–24.

Duflo, Esther. 2002. “Empirical Methods.” Class notes. Department of Eco-nomics, MIT.

Edgington, Dorothy. 2001. “Conditionals.” Pp. 385–414 in The Blackwell Guide toPhilosophical Logic, edited by Lou Goble. Oxford: Blackwell.

Fearon, James. 1991. “Counterfactuals and Hypothesis Testing in PoliticalScience.” World Politics 43: 169–95.

Goodman, Nelson. 1979. Fact, Fiction, and Forecast, 4th edition. Cambridge, MA:Harvard University Press.

Hawthorn, Geoffrey. 1991. Plausible Worlds: Possibility and Understanding in Historyand the Social Sciences. Cambridge: Cambridge University Press.

Heckman, James J. 1996. “Instrumental Variables: A Cautionary Tale.” Techni-cal Working Paper No. 185. Cambridge, MA: National Bureau of EconomicResearch.

Heckman, James J. 1997. “Instrumental Variables: A Study in Implicit BehavioralAssumptions Used in Making Program Evaluations.” Journal of Human Resources32: 441–62.

Heckman, James J. 1992. “Randomization and Social Policy Evaluation” in Eval-uating Welfare and Training Programs, edited by C. Manski and I. Garfinkel.Cambridge, MA: Harvard University Press.

Heckman, James J. 2004. “The Scientific Model of Causality.” Working Paper.Department of Economics, University of Chicago.

Holland, Paul W. 1986. “Statistics and Causal Inference.” Journal of the AmericanStatistical Association 81: 945–60.

Imbens, Guido W. 2002. “Semiparametric Estimation of Average Treatment Effectunder Exogeneity: A Review.” Working Paper. Department of Economics,University of California at Berkeley.

Kaniyathu, Sunny. In progress. The Balance Sheet of Colonialism: Economic Develop-ment in the Colonial Period. PhD Dissertation. Department of Politics, New YorkUniversity.

King, Gary, and Langche Zeng. 2002. “When Can History be Our Guide? ThePitfalls of Counterfactual Inference.” Available at gking.harvard.edu.

Lewis, David. 1973. Counterfactuals. Cambridge, MA: Harvard UniversityPress.

Mackie, J.L. 2002 [1973]. “The Logic of Conditionals.” Pp. 106–14 in Philosophyof Science: Contemporary Readings, edited by Yuri Balashov and Alex Rosenberg.

London: Routledge.Manski, Charles F. 1995. Identification Problems in the Social Sciences. Cambridge,

MA: Harvard University Press.North, Douglass C. 1997. “Some Fundamental Puzzles in Economic History/

Development.” In The Economy as an Evolving Complex System II, editedby W. Brian Arthur, Steven N. Durlauf, and David A. Lane. Addison-Wesley.

North, Douglass C. and Robert Paul Thomas. 1973. The Rise of the Western World:A New Economic History. Cambridge: Cambridge University Press.

9780230_202573_10_cha09.tex 5/6/2008 15: 41 Page 313

Adam Przeworski 313

Przeworski, Adam. 2006. “Is the Science of Comparative Politics Possible?”In Oxford Handbook of Comparative Politics, edited by Carles Boix andSusan C. Stokes. New York: Oxford University Press.

Przeworski, Adam, José Antonio Cheibub, Fernando Limongi, and Michael E.Alvarez. 2000. Democracy and Development. New York: Cambridge UniversityPress.

Quine, W.V. 1953. From the Logical Point of View. Cambridge, MA: HarvardUniversity Press.

Rodrik, Dani. 1998. “Democracies Pay Higher Wages.” Working Paper 6364.Cambridge, MA: National Bureau of Economic Research.

Rodrik, Dani, Arvind Subramanian, and Francesco Trebbi. 2002. “InstitutionsRule: The Primacy of Institutions Over Geography and Integration in EconomicDevelopment.” Unpublished.

Rosenbaum, Paul R. 2002. Observational Studies, 2nd edition. New York: Springer-Verlag.

Rosenbaum, Paul R. and D.B. Rubin. 1983. “The Central Role of the PropensityScore in Observational Studies.” Biometrika 70: 41–55.

Stalnaker, Robert C. 1987. Inquiry. Cambridge, MA: MIT Press.Tocqueville, Alexis de. 1964 [1856]. L’ancien Régime et la Révolution. Paris:

Gallimard.Winship, Christopher and Stephen L. Morgan. 1999. “The Estimation of Causal

Effects from Observational Data.” Annual Review of Sociology 25: 659–707.Woolridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel Data.

Cambridge, MA: MIT.

This page intentionally left blank

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 315

Index

affirmative action 6analytical variables 31annual data 35–6anti-discrimination laws 6Australia

employment change 3, 71, 83employment level 3employment protection 73high wage increases 73low earnings inequality 73men’s employment 5payroll/consumption taxes 73public employment 73unemployment benefit 73, 84women’s employment 5

Austriachildcare provision 151, 152,

210, 211children and employment rate

222civilian government employment

149employment change 3employment levels 3extended leave 153family policies 214left cabinet incumbency 148maternity leave 150, 212men’s employment 5, 198public sector employment 213state benefits 154women’s employment 5, 154,

198, 199women’s working hours 247

autocorrelation 44Durbin’s M test 44

average effect of treatment on thetreated 297

average treatment effect 296

bad employment performance 69baseline bias 297–8

Belgiumchildcare provision 151, 152,



149employment change 3, 71, 83employment levels 3employment protection 73extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154,


between-country variation 41see also individual countries and

parametersBoolean analysis 1breadwinner state 229–35

part-time employment 235–41British Household Panel Survey 227

Canadachildcare provision 151, 152, 210,

211children and employment rate


149employment change 3, 71, 83employment levels 3

315

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 316

316 Index

Canada – Continuedemployment protection 73extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198payroll/consumption taxes 73public sector employment 213state benefits 154women’s employment 5, 154,

198, 199case studies 14causal complexity 95, 120causal conditions 72–4

fuzzy-set scores vs raw values 73causal configurations 67causal effects 169

political regimes 290–313causal necessity 150, 160causal pathways 14, 82causal relations 161–3, 169causal sufficiency 9, 10, 150, 160

family policies 198–202childcare provision 5, 92, 142, 151,

152, 204, 210, 211and women’s educational

attainment 203and women’s employment 142–4,

200see also individual countries

children, and employment rate 222civilian government employment

149clearly conforming cases 82clearly not conforming cases 82Cobb-Douglas economy 291coefficient of determination 13comparative analysis 35comparative employment

performance 2–6fuzzy-set analysis 67–90

compliers average causal effects(CACE) 20, 149, 156

estimation of 177–9women’s employment 164–6

conceptual map 267

conditional mean independence296–7

consistency scores 13, 76, 79consumption taxes 72, 86, 88coordinated market economies 267,

283–4counterfactual cases 77country dummies 208, 209covariates 299coverage score 13, 79–80, 82

raw coverage 79unique coverage 79–80

cross-sectional analysis 48–57cultural factors in women’s

employment 114–19cumulative indices 41cumulative left cabinet incumbency

138, 148, 151, 152, 155, 158, 160,163, 174

day care see childcare provisionDenmark

childcare provision 151, 152,210, 211

children and employment rate222

civilian government employment149

early-exit rate 274early-exit regimes 268employment change 3, 71, 83employment levels 3employment protection 73exit pathways 279extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198, 269payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154, 198,

199, 271

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 317

Index 317

deterministic relationships 8detrending 36Durbin’s M test 44

early-exit rate 272–4men 273, 274women 274, 275

early-exit regimes 22, 260–89comparative analysis 262–8partnership (mediation) 266–7,

268production (push) 261, 262, 281–5protection (pull) 261, 264, 280–1

earnings inequality 86economic growth 32economic outcomes 31economic structure, and women’s

employment 94educational attainment of women

201–2, 203employment change 3, 71

by causal configuration 83, 84cross-country variation 71–4fuzzy-set scores vs raw values 71low-end private sector services 70poor performance 86

employment population ratio 230employment protection regulations

42, 72, 88private sector consumer services

50–1and replacement rate 51, 56

employment rate 2–6, 32decline in 268–72effect of political regime 305–9low-wage sector 38–41men 5, 198, 269and presence of children 222private sector consumer services

42–61women see women’s employment

encompassing comparison 267estimand 297estimators 301–5European Community Household

Panel 227European Employment Strategy 224,

246

European UnionDirective 97/81/EC 233Part-Time Directive 246

exit pathways 279extended leave 153

family policies 4–5, 102–7, 139–46,214

causal sufficiency 198–202and women’s employment 20,

91–134, 196–220see also individual countries

female friendliness 135Finland

childcare provision 151, 152, 210,211



employment change 3, 71, 83employment levels 3employment protection 73extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154,


first differencing 36Fisher test for nonstationarity 39, 41fixed unit effects 46, 209France

childcare provision 102–3, 151,152, 210, 211



9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 318

318 Index

France – ContinuedComité du Travail Féminin 108, 122cultural factors 114–15dénatalité 114dépopulation 114early-exit rate 274early-exit regimes 268economic factors 107–8employment change 3, 71, 83employment levels 3exit pathways 279extended leave 153family policies 214family policy 102–4high wage increases 73labor market policies 108–10left cabinet incumbency 148loi Roudy 109maternity leave 102, 150, 212men’s employment 5, 198, 269Ministere des Droits de la Femme

122Mouvement Démocratique Féminin

108parental care leave 103, 121part-time employment 99public employment 73public sector employment 213single salary allowance 102state benefits 154unemployment benefit 73, 84women’s employment 5, 97, 98,

99, 107–8, 154, 198, 199, 271women’s working hours 247work and family values 101

free choice 115full-time work for women 98, 99fuzzy-set qualitative comparative

analysis 9, 12, 18, 20, 67–90advantages of 67–8goodness-of-fit tests 160, 161,

176–7membership scores 173–4outcome 69–71women’s employment 156–63

gender employment gap 221–2, 223gender equality 136, 141

gendered job creation 136, 138gendered roles 117, 140German Socio-Economic Panel 227Germany




early-exit rate 274early-exit regimes 268employment change 3, 71, 83employment levels 3employment population ratio 230employment protection 73exit pathways 279extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198, 269payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154, 198,

199, 221–59, 271women’s labor market status 236women’s working hours 247

globalization 309–10global variables 31golden handshakes 282goodness-of-fit tests 160, 161,

176–7, 179–87Greece


women’s working hours 247

heteroskedasticity 43, 49

ideals of care 94identifying assumptions 296institutional complementarity 265

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 319

Index 319

institutionalist approach 264institutional variables 33, 34institutions 31instrumental variables 303–4intention-to-treat analysis 153,

155–6Ireland




early-exit rate 274early-exit regimes 268employment change 3, 83employment levels 3exit pathways 279extended leave 153family policies 214left cabinet incumbency 148maternity leave 150, 212men’s employment 5, 198,

269public sector employment 213state benefits 154women’s employment 5, 154, 198,


Italychildcare provision 151, 152,



149early-exit rate 274early-exit regimes 268employment change 3, 71, 83employment levels 3employment protection 73exit pathways 279extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198, 269

payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154, 198,


Japanearly-exit rate 274early-exit regimes 260–89employment change 3, 71, 83employment levels 3employment protection 73exit pathways 279low earnings inequality 73men’s employment 5, 269partial exit model 284payroll/consumption taxes 73public employment 73women’s employment 5, 271

labor force growth 291–4labor market institutions 4,

29–66labor market policies 108–13labor market rigidities 4labor market status of women 236,

238–40labor share 294last in-first out rules 282left cabinet incumbency 148left governance 148, 163, 164

high levels 167impact on policy 159–61low levels 165

liberal market economies 266, 267,282

lock-in effects 34low earnings inequality 72low-end wages 4low-wage sector employment 38–41

autoregression 40cross-sectional and time variance

components 40main variables 39nonstationarity 41

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 320

320 Index

macrocomparative analysis 6macro-comparison 1–2macro-level variables 30–8manipulability 299–300market outcomes 46matching 301–3maternity leave 139, 140, 144–5,

150, 204, 212and women’s educational

attainment 203and women’s employment 200see also individual countries

membership scores for fuzzy-setanalysis 173–4

menearly-exit rate 273, 274employment 5, 198, 269

method of agreement 16method of difference 15methodology 6–18Mill, John Stuart

method of agreement 16method of difference 15

motherhoodand employment 22lone mothers 112–13reinforcement of role 117see also women’s employment

multicollinearity 54multiple pathways 68

nearly always sufficient condition10, 11

necessary condition 16necessity 9, 10Netherlands

Adjustment of Hours Act 112Breed Platform 116childcare provision 104–6, 113,

151, 152, 210, 211children and employment rate


149cultural factors 115–19early-exit rate 274early-exit regimes 268Emancipatiekommissie 116, 122

employment change 3, 71, 83employment levels 3employment population ratio

230employment protection 73Equal Treatment Act 111exit pathways 279extended leave 153family policies 104–7, 214high wage increases 73labor market policies 110–13left cabinet incumbency 148low earnings inequality 73maternity leave 102, 106, 150,

212men’s employment 5, 198, 269parental sharing 107part-time clause 232part-time employment 99payroll/consumption taxes 73public employment 73public sector employment 213single mothers 112–13state benefits 154Stimulation Measure on Childcare

105unemployment benefit 73, 84Wassenaar agreement 232welfare restructuring 111welfare system 104women’s employment 5, 97, 98,

99, 154, 198, 199, 221–59, 271women’s labor market status 236women’s working hours 247Work and Care Act (2001) 121work and family values 101

new institutionalism 290New Zealand

employment change 3employment levels 3men’s employment 5women’s employment 5

non-independence bias 300–1nonstationarity 38–9, 41, 44, 46Nordic countries see Denmark;

Finland; Norway; SwedenNorway


9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 321

Index 321



employment change 3, 71, 83employment levels 3employment protection 73extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154, 198,

199

omitted variable bias 86ordinal comparison 14, 263, 285OSA Labor Supply Panel 227

panel data 29, 63–4parental care leave 5, 103, 121, 144,

145–6see also maternity leave

parental sharing 107partial exit model 284partnership (mediation) regimes

266–7, 268part-time employment 6

breadwinner determinants 235–41as coping strategy 241–6women 98, 99, 100, 223–4, 226

pay equality 42payroll taxes 72, 86, 88pension schemes 264–5

public-private mix 265period-demeaning 36policy variables 31, 32, 34political economic variables 38–42

employment regulation 42low-wage sector employment

38–41

pay equality 42reservation wage 41–2

political regimeseffect on employment 305–9and per capital income 306

pooled regression 7–9, 18limitations of 8–9see also regression analysis

Portugalchildren and employment rate

222men’s employment 5women’s employment 5women’s working hours 247

post-treatment bias 299–300potential exposability 299potentially conforming cases 82Prais-Winston transformation 46pre-school childcare see childcare

provisionprincipal components analysis 204principal factor analysis 204private sector consumer services

42–61annual variation 47, 62cross-sectional analysis 48–57employment protection regulations

50–1, 56long-time changes 48, 55low-end, employment rate 70pooled analysis 42–8, 57–9replacement rate 43, 45, 56, 59,

60, 61process tracing 14, 95production (push) regimes 261, 262,

265–6, 268, 281–5productivity, and wage growth 293protection (pull) regimes 261, 264,

265, 266, 268, 280–1public childcare see childcare

provisionpublic sector employment 6, 72, 73,

88, 204expansion of 137–46women 94, 137–9, 162, 200, 213and women’s educational

attainment 203see also individual countries

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 322

322 Index

qualitative comparative analysis 1,7, 9–13, 196

crisp-set 9fuzzy-set 9, 12, 18, 20, 67–90limitations 12

ratchet effects 34raw coverage 79regression analysis 196–220

bivariate 205–8multivariate 208–17see also pooled regression

relative risk ratio 229remainders 77replacement rate 43, 59, 60, 61

and employment regulations51, 56

panel models 45scores 59

reservation wage 41–2

school scheduling 92selection bias 23self-selection bias 298single mothers 100–1, 112–13single salary allowance 102small-N analysis 1, 7, 13–18social care 135social democratic governance

139–41social policy 93solution sets 77–8Spain


employment change 3employment levels 3men’s employment 5women’s employment 5women’s working hours 247

stable unit treatment value 300–1state benefits 154structural variables 31subsidiarity 96sufficient condition 16Sweden

childcare provision 140, 151, 152,210, 211



early-exit rate 274early-exit regimes 268employment change 3, 71, 83employment levels 3employment protection 73exit pathways 279extended leave 153family policies 214high wage increases 73left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198, 269payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154,


Switzerlandemployment change 3employment levels 3men’s employment 5women’s employment 5

taxation policies 6and women’s employment 93–4

tendential relationships 8The OECD Jobs Study 4threshold effects 34time series 36time-series regression 1transitional labor market 224truth tables 74–6

UKchildcare provision 151, 152,



149

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 323

Index 323

early-exit rate 274early-exit regimes 268employment change 3, 71, 83employment levels 3employment population ratio

230employment protection 73exit pathways 279extended leave 153family policies 214high wage increases 73Job Release Scheme 280left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198, 269payroll/consumption taxes 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154,

198, 199, 221–59, 271women’s labor market status

236women’s working hours 247Working Hours Adjustment Act

235unemployment benefits 73, 84, 88unemployment rate 32unique coverage 79–80unit homogeneity 295–6unobservables, selection on

304–5USA




early-exit rate 274early-exit regimes 260–89employment change 3, 71, 83employment levels 3employment protection 73exit pathways 279extended leave 153family policies 214high wage increases 73

left cabinet incumbency 148low earnings inequality 73maternity leave 150, 212men’s employment 5, 198, 269payroll/consumption taxes 73public employment 73public sector employment 213state benefits 154unemployment benefit 73, 84women’s employment 5, 154, 198,

199, 271

vector space 74voluntarist bargaining systems 267

wage growth 72, 86and productivity 293

wage levels 4women 93–4

welfare systems 93, 104see also family policies

Why We Need a Welfare State 135women

childcare role 22, 221early-exit rate 274, 275educational attainment 201–2,

203, 216employment preferences 201,

246–52labor market status 236, 238–40marginalization of 224–5public sector jobs 6wage levels 93–4working hours 247

women’s employment 5, 20, 136,141–6, 271

CACE analysis 164–6causal relations 161–3, 200childcare provision 142–4, 200cross-country comparisons 154cultural factors 114–19and economic structure 94and educational attainment 216explanatory model of 120factors shaping 92–5family policies affecting 20,

91–134, 196–220full-time 98, 99

9780230_202573_11_ind.tex 5/6/2008 15: 42 Page 324

324 Index

women’s employment – Continuedfuzzy-set analysis 156–63maternity leave 200mothers see motherspart-time 98, 99, 100, 223–4, 226patterns of 94, 197–8, 221–59with pre-school children 142public sector 94, 137–9, 162, 200

single mothers 100–1, 112–13taxation policies 93–4see also individual countries

workand family values 101importance of 117–18

work–family conflicts 135,226

Method and Substance in Macrocomparative Analysis (Research Methods Series)

Documents

Transcript of Method and Substance in Macrocomparative Analysis (Research Methods Series)