Human Capital, Technology and Inequality › 122964 › 1 › Zainab_Asif_Thesis.pdf · Zainab Asif...

Human Capital, Technology and Inequality

Zainab Asif

M.Sc. (Economics); MPhil (Economics)

Submitted in fulfilment of the requirements for the degree of

Doctor of Philosophy

School of Economics and Finance

QUT Business School

Queensland University of Technology

2018

i

Keywords

Cognitive skills

Economic growth

Educational achievements

Educational attainments

Decomposition

Human Capital

Inequality

Technology

TIMSS

ii

Abstract

This thesis comprises of two essays that explore issues related to human capital,

technology and inequality. In both essays human capital is at the heart of the analyses performed.

The common motivation for these essays stems from the literature on economic growth that

highlights the direct contribution of human capital to growth, as well as its indirect contribution

through facilitation of the adoption and diffusion of technologies, and provides evidence of

inequality in human capital impacting upon economic performance.

The first study argues that the empirical literature on the link between human capital and

technological diffusion is inconclusive, with controversies pertaining to both the measurement of

human capital as well as that of technological adoption and diffusion. In this study we revisit this

issue, by examining this link using newly created measures for both of these concepts.

Specifically, we examine the impact of qualitative measures of human capital (based on data on

tests of cognitive skills), and direct measures of technology adoption using country level panel

data for the period 1964-2003. Our measure of cognitive skills is drawn from Trends in

Mathematics and Science Study (TIMSS). Based on the nature of these test scores, skills

manifest in mathematics scores are labeled as “generic” while science are labeled as “specific”

skills. For measures of technology we use the Cross Country Historical Adoption of Technology

(CHAT) data set due to Comin and Hobijn (2009), which presents measures of intensity and

timing of adoption for a large number of technologies from various sectors of the economy. Our

analysis suggests that the link between human capital and technological adoption and diffusion is

a conditional one, which rests on various aspects of human capital and the nature of the

technology in question. We find, for example, that technologies in transport, tourism and health

exhibit a stronger evidence of correlation between our measures of technology adoption and

human capital, than technologies from “traditional” sectors such as agriculture. Our

interpretation for the lack of correlation in the latter sector is not that human capital does not

matter in agriculture; rather, other unmeasured aspects of human capital such as “learning by

doing” could matter more. Our analysis, which also controls for institutional variables and other

factors that determine technological adoption, therefore suggests that future explorations of the

link between human capital and technological adoption need to be more comprehensive, in that

iii

they take into account the appropriate dimensions of human capital associated with the nature of

the technology in question.

The second study examines qualitative measures of human capital from a microeconomic

perspective by analyzing the composition and determinants of inequality in human capital. We

use advanced mathematics raw test scores from 2008 Trends in Mathematics and Science Study

(TIMSS) to construct generalized entropy measures of human capital inequality for 10 countries.

In common with previous literature, we find that, at the aggregate level, within-country

inequality is higher than between-country inequality. Hence, we further decompose within-

country inequality at the school level to extract insights about the micro-composition and

determinants of inequality. This decomposition reveals that within-school inequality is greater

than between-school inequality. We further examine, for each country, the school and teacher

characteristics that underpin this within-school inequality. Our analysis reveals that each country

has a unique set of determinants of within-school inequality. Compared to aggregated

approaches used in extant literature, our findings suggest that a disaggregated, stepwise

exploration of this type is more fruitful in identifying the root causes of inequality in human

capital and as such more informative in determining appropriate educational policies.

iv

Table of Contents

Keywords .................................................................................................................................. i

Abstract .................................................................................................................................... ii

Table of Contents .................................................................................................................... iv

List of Figures ......................................................................................................................... vi

List of Tables ......................................................................................................................... vii

List of Abbreviations ........................................................................................................... viii

Statement of Original Authorship ........................................................................................... ix

Acknowledgements ..................................................................................................................x

Chapter 1: Introduction………………………………………………………...1

Chapter 2: Related Literature and Motivation.……………………………….9

2.1 Introduction ...................................................................................................................9

2.2 Human Capital, Growth and Technology .....................................................................10

2.3 Direct Measures of Technology ....................................................................................14

2.4 Qualitative Measures of Human Capital .......................................................................19

2.5 Prespectives on Income and Human Capital Inequality ..............................................23

2.6 Human Capital and Inequality .....................................................................................26

2.7 Conclusion ...................................................................................................................29

Chapter 3: Human Capital and the Adoption and Diffusion of Technology……31

3.1 Introduction ..................................................................................................................31

3.2 Empirical Metholodgy ..................................................................................................39

3.2.1 Measures of Technology Adoption and Diffusion .......................................................39

3.2.2 Measures of Cognitive Skills ........................................................................................43

3.2.3 Econometric Methodology ...........................................................................................45

3.3 Emprical Evidence on Measures of Human Capital and Usage Intensity of

Technology………………………………………………………………………….....47

3.4 Empirical Evidence on Measures of Human Capital and Technology Usage lags .. ....55

3.5 Additional Robustness Checks .....................................................................................61

3.6 Concluding Remarks ....................................................................................................64

Chapter 4: Deconstructing Human Capital Inequalities: A new approach based on

measures of educational achievement……………………………………………..67

4.1 Introduction ..................................................................................................................67

v

4.2 Methodology .................................................................................................................74

4.2.1 Generalized Entropy Measures of Inequality ..............................................................74

4.2.2 Data and Data sources ..................................................................................................76

4.2.3 Empricial Framework for Country-wise Analysis ........................................................77

4.3 Empricial Evidence on Inequalities in Educational Achivements ................................80

4.3.1 Skill-Inequalities:A Cross-Country Analysis ...............................................................81

4.3.2 Skill-Inequality Indices at Country and School level………………………………...83

4.3.2.2 Results of Country-wise Analysis with a Common Set of Variables ……………...85

4.3.2.3 Country-specific Analysis ..........................................................................................86

i Lebanon ........................................................................................................................86

ii Netherlands ...................................................................................................................90

ii Russia ............................................................................................................................93

iv Iran ................................................................................................................................95

v Slovenia ........................................................................................................................97

vi Philippines ..................................................................................................................100

vii Norway .......................................................................................................................102

viii Armenia ......................................................................................................................103

ix Itlay .............................................................................................................................106

x Sweden ........................................................................................................................108

4.4 Cross-Country Analysis ..............................................................................................112

4.5 Concluding Remarks…………………………………………………………………..115

Chapter 5: Summary and Conclusions………….……………………………….117

Bibliography……………………………………………………………………….128

Appendices…………………………………………………………………………141

Appendix A Definitions and Descriptive Statistics ..............................................................141

Appendix B Results for Human Capital and Technology Adoption ....................................150

Appendix C Descriptive Statistics for Mathematics and Science Panel for Technology Usage

Lags as Measures of Technology Diffusion. .....................................................................156

Appendix D Results for Human Capital and Technology usage lags ..................................160

Appendix E Additional Robustness Checks ………………………………………………165

Appendix F Definitions and Descriptive Summary. .............................................................177

Appendix G Combined/Cross Country Human Capital Inequality ......................................180

Appendix H Description of Selected Variables Country Wise Analysis ..............................183

Appendix I Cross-Country Analysis: Individual Variable Regression Models……………195

vi

List of Figures

Figure 2.1 Technology Adoption Lags………………………………………………18

Figure 2.2 Educational Attainments and Economic Growth…………………………22

Figure 3.3 Graphical Representation of Technology Usage Lags...............................41

vii

List of Tables

Table 2.1 Technologies and Measures in Historical Cross-Country Technology Adoption Dataset

(1788-2001)…………………………………………………………………………………..15

Table 3.1 Usage Intensity of technologies, Mathematics Skill Panel Estimations.................49

Table 3.2 Usage Intensity of technologies, Science Skill Panel Estimations.........................50

Table 3.3 Usage Lags of technologies, Mathematics Skill Panel Estimations.......................56

Table 3.4 Usage Lags of technologies, Science Panel Estimations........................................57

Table 4.1 Regression Results for Lebanon……………………………….............................88

Table 4.2 Regression Results for Netherlands........................................................................91

Table 4.3 Regression Results for Russia................................................................................94

Table 4.4 Regression Results for Iran.....................................................................................96

Table 4.5 Regression Results for Slovenia..............................................................................98

Table 4.6 Regression Results for Philippines.........................................................................100

Table 4.7 Regression Results for Norway..............................................................................102

Table 4.8 Regression Results for Armenia.............................................................................105

Table 4.9 Regression Results for Italy ..................................................................................107

Table 4.10 Regression Results for Sweden……………………………….............................109

viii

List of Abbreviations

Cross Country Historical Adoption of Technology CHAT

Generalized Entropy Measures GE

Generalized methods of moments GMM

Gross Domestic Product GDP

Historical Cross-Country Technology Adoption Dataset HCCTAD

International Student Achievement Tests ISAT

Organization for Economic Cooperation and Development OECD

OECD standardized group OSG

Programme for International Student Assessment PISA

Research and Development R&D

Total factor productivity TFP

Trends in Mathematics and Science Study TIMSS

United Nations Development Program UNDP

United Nations Fund for Population Activities UNFPA

World Development Indicators WDI

ix

Statement of Original Authorship

The work contained in this thesis has not been previously submitted to meet requirements

for an award at this or any other higher education institution. To the best of my knowledge and

belief, the thesis contains no material previously published or written by another person except

where due reference is made.

Signature: _________________________

Date: _________________________

n9341862

Stamp

n9341862

Stamp

x

Acknowledgements

First and the foremost, I am thankful to Almighty Allah who has helped me in His

mysterious ways in undertaking and completing this study.

No amount of words can express and do justice when it comes to thanking my principal

supervisor Dr Radhika Lahiri. This thesis would not have been possible without her immense

guidance, help, support and encouragement. She is a true mentor who has tremendously refined

my research and writing skills. I consider myself extremely blessed to have worked and learnt

from an extraordinary supervisor like her. I would also like to thank my associate supervisor Dr

Vincent Hoang for all his guidance, feedback and support during this time.

I am grateful to Professor Pascalis Raimondos (Head of School, QUT Business School,

Economics and Finance) for being understanding and supportive in the time when my PhD

journey was virtually over. I would also like to thank him and Dr. Stephen Cox (Director, Higher

Degree Research Studies, QUT Business School) for reviewing my confirmation document and

their valuable feedback. I am thankful to QUT for providing me the scholarship to undertake my

PhD studies. I would also like to thank my friends and colleagues who supported me and made

my stay at QUT a memorable one. I am especially thankful to Sharmila, Eucabeth, Thames,

Hong hong, Minh, Wangsit, Uttam and Javeir for all their help.

I am thankful to my Mother and sister Rabia, for all their prayers, support and

constructive criticism that kept me going throughout this time. I would also like to thank my

elder sister Fatima for her support. To my husband Asif, who was also enrolled in PhD but was

extremely understanding of the stress and pressures faced by me. I could not have done this

without his encouragement, support, patience and faith in me at times when I had lost all hope. I

am grateful to my Father for raising me as a strong and resilient woman in a world dominated by

xi

men. I can still feel the warmth of his hand on my right shoulder and his voice saying to me “you

know what it takes, all you have to do is to put in a bit of hard work”. What makes me sad is that

he is no longer with me to see that his little girl did put in a bit of hard work this time.

Lastly, I am thankful to the women in my life, my girls Amal and Hafsa. They have

missed out a lot on their mum for the past four years. During this time they have stood by me,

tolerated all the terrible behavior and shouting when I was exhausted and all alone. Still they

love me like hell and to them I am the best mum. I can never forget the shine in their eyes while

they wait at the train station for me to come home. The moment they use to get a glimpse of me

coming out from the platform they just use to run towards me and gave a warm hug. That hug

gave me the courage, stamina and strength to move on and will continue to do so in the future as

well. I dedicate this thesis to my Father and my daughters Amal and Hafsa.

1

Chapter 1

Introduction

This thesis constitutes essays that explore issues related to human capital, technology and

inequality. The first essay focuses on the link between qualitative constructs of human capital

and direct measures of technology diffusion and adoption. The second essay looks at the

inequality of human capital by developing an inequality index which reveals the structure and

composition of within and between-country human capital inequality and further traces its roots

by decomposing the within country inequality at a disaggregated microeconomic level. The

common thread that links the two essays is the concept of human capital.

The first study presented in Chapter 3 of this thesis empirically explores the association

between human capital and technology, employing measures of educational quality and direct

measures of technology adoption and diffusion. This study broadly falls within the literature on

human capital, technology and growth initiated by Nelson and Phelps (1966) who suggest that

human capital accumulation, through its impact on technology adoption and diffusion, influences

an economy’s growth prospects. Since then, subsequent models of economic growth recognize

the contribution of human capital in the form of better education, as it impacts on productivity

growth directly as an input in the production process as well as indirectly by facilitating

technological adoption and diffusion (Lucas, 1988; Romer, 1990; Mankiw et al, 1992; Aghion

and Howitt, 1998; Barro, 1998; Vandenbussche, 2006; Madsen, 2014).

However, there are studies within this literature which bring the growth accounting and

productivity measurement methods under scrutiny (Jorgenson & Griliches, 1967; Hulten, 2001).

According to this line of thought, the Solow’s (1957) residual suffers from measurement bias as

2

it not only captures changes in technology but other unmeasured inputs in the process of

production (Basu, 1996; Burnside et al, 1995; Weil, 2005). These unmeasured inputs include

variations in capacity utilization and labour hoarding. Hence, technological change measured as

a residual may be inappropriate as it does not constitute changes attributed specifically to

technology.

This essay, however, is motivated by Comin et al (2008) and Comin and Mestieri (2013)

who introduce direct constructs of technology. They consider the intensive margin of technology

adoption, which captures the intensity of use of technology in an economy -i.e. how many units

of a particular technology are in use relative to the size of a country as measured by per capita

GDP or population. In addition, they consider the extensive margin of adoption which refers to

the timing of adoption of technology. This refers to the first time a new technology is adopted

within a country. They argue that indirect and traditional measures of technology such as total

factor productivity or residual are unable to differentiate between the extensive and intensive

margins of technology, which should be central to any analysis that explores the channels

through which technology impacts on growth.

This study is also inspired by one of the earlier studies of Comin and Hobijn (2004)

which employs enrollment levels as quantitative measures of human capital and direct measures

of technology, and shows that human capital is among the important determinants influencing

the rate of adoption of a technology. However, their panel data analysis pools a large set of

technologies which makes it difficult to understand the association that a specific technology

3

may have with a particular type of human capital, thus, making it complicated to address the

issue of skill-technology specificity.1

Further motivation for this study emerges from Hanushek and Kimko (2000) and

Hanushek and Woessmann (2012) who employ cognitive skills as qualitative measures of human

capital and examine their link with growth. Their analysis reveals that test scores representative

of skills have a robust and positive relationship with economic growth.2 This literature does not

examine the link with human capital and technology; however, one possible interpretation of the

evidence is that presence of skills among labour force improves the ability of a country to imitate

technology, and this contributes to its growth prospects. The aim of chapter 3 of this thesis is to

examine this particular mechanism of growth, namely technology adoption and diffusion and its

link with human capital.

We argue that this link should be examined by employing direct measures of technology

and qualitative measures of human capital. Given this, the current study contributes to the

literature as a first attempt to examine the link between human capital and technology adoption

and diffusion by incorporating disaggregated, qualitative measures of human capital and direct

measures of technology. We also argue that a specific type of human capital may be more or less

relevant in facilitating the adoption and diffusion processes depending on the type of technology

under discussion. Therefore, we distinguish between different types of human capital and

examine their relative impact on the adoption of technologies. For this purpose we employ a

1 By skill-technology specificity, we mean that a specific skill facilitates a specific technology in a particular sector.

2 In the human capital literature, qualitative measures of human capital are cognitive skills measured as test scores

also termed as educational achievements. On the other hand, quantitative measures of human capital are average

years of schooling also referred to as educational attainments (Hanushek and Woessmann, 2012). To provide the

reader with clarity, in what follows, when we generate a comparison between the two measures of human capital, we

may use the terms interchangeably, for instance qualitative versus quantitative, cognitive skills (test scores)versus

average years of schooling, and educational achievements versus attainments.

4

comprehensive set of measures of human capital. These measures are reflective of various

dimensions of human capital such as learning-by-doing, cognitive skills, stock of knowledge and

health. We discuss these measures in detail in Chapter 3.

As we explore the association between different types of human capital and direct

measures of technology, our results support our premise regarding the technology-specific nature

of the link between human capital and technology. We also find support for our premise that the

nature of human capital matters and some dimensions of human capital are more important than

others. In particular, the learning-by-doing dimension, reflected in the extent of past usage of a

technology is of primary importance, followed by cognitive skills and other measures such as the

extent of education and health capital. However, this is a broad conclusion. As we have

emphasized previously, the link is technology specific and the ranking/importance of any

measure of human capital may change across different types of technologies.

While the first study provides a macroeconomic perspective of human capital, the second

study examines qualitative measures of human capital inequality from a microeconomic

perspective. If human capital accumulation influences growth through technology adoption and

diffusion, then the distribution of human capital may also impact the economy in several ways.

The literature on economic growth acknowledges inequality in human capital as one of the

factors having implications for economic performance (Checchi, 2004; Castello and Climent

2010; Hanushek and Woessmann, 2015). However, this literature is less well developed

compared to the literature on income inequalities and growth, and requires further exploration.

To that end, the objective of Chapter 4 of this thesis is to construct a human capital inequality

index and study the structure, composition and determinants of within and between sub-group

inequalities at a microeconomic level, an aspect macro cross-country studies do not examine.

5

This study is inspired by ideas initiated by Sen (1979, 1985, 1987) who suggests that

income is not a sufficient measure of well-being and other aspects of human life such as

education, health, political freedom and civil liberties also play a role in improving the quality of

life. Within this strand of literature, theoretical studies show that educational inequality as a

measure of human capital inequality is a key determinant of income inequality (Glomm and

Ravikumar, 1992; Saint-Paul and Verdier, 1993; Galor and Tsiddon, 1997). Furthermore, there

are empirical studies which provide evidence that educational inequalities are also among the

factors that influence growth, income distribution and lead to differences in productivity

(Gregorio and Lee, 2002; Checchi, 2004; Acemoglu and Dell, 2010; Blanden and Mcnally,

2015). Hence, the main argument of this literature is that human capital inequality influences

economic growth, which underscores the need to investigate composition and determinants of

educational inequality.

Our second study is also inspired by Hanushek and Kimko (2000), Hanushek and

Woessmann (2012) and Woessmann (2014) in that it constructs inequality indices based on

qualitative measures that directly determine educational achievements rather than attainments.

Further motivation stems from the strand of literature that employs aggregate level standardized

or average test scores as educational quality measures to explore variations in human capital and

develop international comparisons. In particular, Sahn and Younger (2007) use standardized

mathematics and science test scores to construct generalized entropy index and decompose

inequality into within and between-country components.3 Their work indicates that within-

country inequality dominates between-country inequality.

3 Shorrocks and Wan (2005) explain the construct of generalized entropy measure based on the concept of income as

a measure of inequality.

6

Against this background, the second study employs educational quality measures based

on TIMSS (2008) and uses Generalized Entropy Measures to decompose within and between

sub-group inequality at three sub levels, i.e., cross-country, country and school levels. This

study, to the best of our knowledge, is the first attempt to decompose human capital inequality

using micro-level educational achievement data based on raw pupils’ test scores to construct

within and between measures of dispersion for human capital. Based on this decomposition

exercise, our findings reinforce the dominance of within-country over between-country

inequality. Hence, we argue that human capital inequality has a country-specific dimension. This

implies that disparities in human capital originate from differences in educational quality within

a country rather than between countries, and suggests that the pattern and factors associated with

inequalities are specific to a country.

As we aim at uncovering micro-level country-specific nature of human capital inequality,

we further decompose inequality by considering two sub-levels: country and schools within each

country. For this we consider student sub-groups from each of the schools that participated in the

test and develop country-specific analyses of human capital inequality. Our results at country

level indicate that within-school inequality dominates the between-school component for all

countries.

Given these findings, we aim to identify the factors associated with inequalities at the

school level. We therefore develop country-wise regressions employing decomposed school-

level inequalities and a standard regression framework with school and teacher related attributes

among the possible determinants of skill inequality. Our results show that school and teacher

attributes are the important factors influencing human capital inequality; however, the specific

attributes differ across countries. Based on this evidence, we suggest that if the determinants of

7

educational quality are different across countries then the factors contributing to within-school

inequalities must be different.

The studies described above have several insights of relevance to policy. In the context of

human capital and its link with technological diffusion, the evidence suggests that many

dimensions of human capital are of relevance, although their relative ranking and importance

varies depending on the nature of technology in question. Based on these findings, a coherent

policy developed on a multi-dimensional approach to improving human capital is likely to be

more useful in contrast to a broad-brush approach aimed at increasing the educational attainment

of the population.

A similar analogy applies in the context of the second essay, in which we find that the

pattern and set of factors associated with human capital inequalities are country-specific. The

plausible strategies to reduce inequalities may include subsidizing the students from

economically disadvantaged backgrounds, provision of financial support to schools to reduce

class size or improving the quality of teachers through better hiring, pay and retention practices.

However, the factors of importance change across countries, suggesting that single country

microeconomic analyses of the type we conduct may be of greater relevance to the design of

policy.

The remainder of the thesis is organized as follows. Chapter 2 constitutes a review of

literature related to the studies comprising this thesis. Chapter 3 presents the first essay,

examining the association between human capital and technology employing measures of

educational quality and direct measures of technology adoption and diffusion. Chapter 4 presents

the second essay, which deconstructs a human capital inequality index and reveals the structure,

8

composition and determinants of within and between sub-group inequalities at a micro economic

level. Chapter 5 presents the concluding remarks.

9

Chapter 2

Related Literature and Motivation

2.1 Introduction

This chapter discusses some of the themes which are relevant to the studies that comprise

this thesis. Given that this thesis constitutes two independent, albeit related studies, we review

the topics that are relevant to each study separately. A common thread runs through both strands,

however, given that both essays relate to the topic of human capital. The first study empirically

explores the role of human capital in the context of technology adoption and diffusion. We

therefore motivate this study from the literature on economic growth that highlights the direct

contribution of human capital to growth, as well as its indirect contribution through the

facilitation of the adoption and diffusion of technologies.

We begin by reviewing studies which use indirect and traditional measures of technology

and quantitative constructs of human capital to explain the cross-country differences in

technological progress, and suggest that lack of educated and skilled human capital is one of the

most important barriers to technology adoption and diffusion. This brief account comprises of

Section 2.2. We then discuss the literature which provides an alternate construct of technology

by employing direct measures of technology and explain the dynamics of technology and its

association with economic growth in Section 2.3. This literature is particularly of relevance to

Chapter 3, which uses similar measures to proxy our concept of technology adoption and

diffusion. Section 2.4 presents studies which use qualitative measures of human capital, as

measured by cognitive skills, assessed using test scores as described in the introduction. These

10

studies reveal a strong and consistent cognitive skill-growth relationship thereby motivating the

use of this variable as a proxy for our human capital measure in Chapter 3.

The second study explores inequality in human capital in the context of qualitative

measures of education by developing microeconomic case-by-case country analyses of these

measures. This study entails a deconstruction of inequality by exploring its occurrence at a

disaggregated level, and then attempts to explain the causes of inequality at these levels. In

Section 2.5 we provide an account of studies that examine the impact of human capital inequality

on economic growth. These studies provide evidence of how variations in within and between-

country inequalities contribute towards overall inequality at a macroeconomic level, thus,

providing a motivation to study inequality at a disaggregated level. Developing on this literature

we direct our attention towards inequality in human capital in the context of education in Section

2.6. This literature employs both quantitative and qualitative measures of education to examine

variations in human capital, and provide an evidence of an association between inequality in

human capital and growth. Section 2.7 concludes the chapter.

2.2 Human Capital, Growth, and Technology

The debate about the role of human capital in facilitating the adoption and diffusion of

technology originates in the mid-1960s. Nelson and Phelps (1966) suggest that prospects of

introduction and adoption of a new technology improve due to higher level of education. A

skilled labour force facilitates adoption as it is better able to learn new technologies and is more

suited to adoption and improves an economy’s growth prospects. Models of economic growth

suggest that human capital impacts the economy in two ways. Firstly, human capital contributes

to productivity growth as an input. Secondly, it increases productivity as one of the channels

11

facilitating the adoption of technology (Lucas, 1988; Romer, 1990; Makiw, et al, 1992; Aghion

and Howitt, 1992, 1998).

Focusing on the notion of human capital and technology, Atkinson and Stiglitz (1969)

suggest that a country selects a particular technology which corresponds to its factor

endowments or which is similar to technologies already in use. Based on this they argue that

technological progress is localized and leads to higher productivity of only those technologies

that correspond to the capital-output ratio of a particular economy. Given that each country uses

a specific technology, a technology employed by a country may not be appropriate for another

country. The larger interdisciplinary literature also suggests that a developing country cannot set

up industries employing technologies of the developed countries as it lacks the required

organization, finance, infrastructure and appropriate capital-labour ratio for this type of a

production process (Schumacher, 1973). Hence, developing economies remain unable to adapt to

advanced technologies as they have barriers to adoption in the form of lack of physical capital,

skill shortages and insufficient depth of human capital (Basu and Weil, 1998). This argument is

reinforced by Acemoglu and Zilibotti (2001) who suggest that skill-technology mismatch

contribute to output gaps between developed and developing regions. Therefore, improvement of

human capital leads to reduction of barriers facilitating technological adoption.

Caselli and Coleman (2006) develop a model of endogenous technological choice and

analyze cross-country differences in aggregate production in a framework where skilled and

unskilled workers are imperfect substitutes. The results of their calibrated model emphasize that

higher income economies are better suited to adopt technology because they face lower skill

barriers due to presence of educated and skilled human capital. Moreover, the technological view

of human capital also receives empirical support in several studies which show that both the

12

initial level of schooling and its interaction with a measure of technology gap are positively

associated with growth (Benhabib and Spiegel, 1994; Barro and Sala-i-Martin, 1995; Barro,

1998; Galor and Weil, 1999 and 2000).4 The literature on human capital and technology adoption

also acknowledges the need of education in the process of industrialization where innovations

lead to introduction of new machines which requires skilled human capital for the purpose of

technological adoption (Champernowne, 1963).

Vandenbussche et al. (2006) contribute to the argument regarding growth, human capital

and technology by suggesting that different degrees of skills impact on technology in a different

manner. They do so by decomposing educational attainment at different levels and show that

primary and secondary education are appropriate for technology adoption while higher education

leads to innovation of technologies. Furthermore, Madsen (2014) examines the role of education

in labour productivity using age specific school enrollment levels as measures of human capital.

An important feature of his work is that he distinguishes between the role of education in raising

labour productivity and impact of education through human capital on technology. The

simulations results show that a mean country experiences a growth of 146% in labor productivity

due to education for the period spanning 140 years. On the other hand, productivity grows by

around 19% due to higher educational attainment. He also uses the interaction between

educational attainment and distance to technology frontier in simulations and finds that

interaction improves productivity by 127% in an average country.

Another way of looking at the role of human capital in adoption and diffusion of

technology is from the perspective of a particular technology in a specific sector of an economy.

Studies based on agricultural sector regarding adoption of new technologies such as pesticides,

4 A technology frontier can be defined as the stock of knowledge of technology available to innovators from all the

sectors in all the countries Aghion and Howitt, (2009).

13

fertilizers, weedicides highlight that education measured as years of schooling improves human

skills as it develops mental attitudes that facilitate acceptance and adoption of new technologies

(Feder and Slade, 1984). Similarly evidence suggests that the level of education of farmers is a

significant factor determining the adoption of new technologies and their cropping practices

(Waller et al, 1998; Caswell et al, 2001). Studies in the health sector also provide evidence that

more education leads to assisting in adoption of technologies in the field of medicine as health

gaps across education groups start to widen in the past few years (Fogel, 1994; Mackuc et al,

1989; Elo and Preston, 1996; Muney and Lichtenberg, 2002).

Furthermore, Caselli and Coleman (2001) analyze the impact of human capital in the

diffusion process for computers. They use data on the import value of computers for 90 countries

dated 1970-1990 to examine the hypothesis whether imports of computers as measures of

technology are influenced by different measures of human capital. The country-level random

effects specifications include controls for per capita income, continent and year dummies. The

results show that a one percentage point increase in the proportion of population with more than

primary schooling leads to a one percent increase in the import value of computers. Riddell and

Song (2012) examine a similar hypothesis but in a different setting. Their analysis constitutes

micro level data from the Canadian workplace and employee survey. They employ time and state

variation in compulsory education laws and instrument it for educational attainments of workers

as a measure of human capital. Their estimations show that graduating from high school is

associated with the probability of using a computer at work by 37 percentage points. Likewise,

an additional year of school leads to an increase of 7 percentage points in the probability of

computer usage.

14

2.3 Direct Measures of Technology

In this section we elaborate on direct measures of technology mentioned previously in the

introduction as they are of particular relevance to the first study. Comin et al (2008) and Comin

and Hobijn (2009) argue that the literature on growth accounting employs indirect measures of

technology based on the Solow residual. The residual not only measures the changes in output

due to technology, but also captures the changes attributed to other unmeasured inputs in the

process of production. Hence, the earlier studies fail provide answers to the two main questions.

Firstly, how big are cross-country differences in technology? Secondly, can these differences be

compared to differences originating in the income per capita of various economies?

In order to answer these, Comin et al (2008) suggest the use of direct measures of

technology. They develop the concepts of technology and income usage lags, and provide

explanations regarding cross-country differences in adoption and diffusion of technology. Their

analysis reveals that technology usage lags across countries are sizeable. In addition, these lags

show high correlations across technologies and are also highly correlated with the economic

development level of an economy, measured by the per-capita income.

In one of their earlier studies, Comin and Hobjin (2004) examine diffusion of more than

20 technologies across 23 leading industrialized economies of the world. In this study they use

direct measures of technology rather than total factor productivity (TFP)/residual to measure

technology. The direct measures are disaggregated constructs of technology contained in the

Historical Cross-Country Technology Adoption Dataset (HCCTAD). They categorize

technologies into eight groups and explain the adoption of specific technologies by using six

different measures for the level of adoption. Table 2.1 includes the technologies and their

measures contained in the HCCTAD data set.

15

Table 2.1 Technologies and Measures in Historical Cross-Country Technology Adoption

Dataset (1788-2001)

Group Technology Measures

1 Steel Fraction of tonnage of steel produced using Bessemer

method

Fraction of tonnage of steel produced using Open

Hearth furnaces

Fraction of tonnage of steel produced using Blast

Oxygen furnaces

Fraction of tonnage of steel produced using Electric

Arc furnaces 2 Textile Fraction of spindles that are mule spindles

Fraction of spindles that are ring spindles

3 Transportation

(rail, road and airways) Freight traffic on railways (TKMs) per unit of real

GDP

Passenger traffic on railways (PKMs) per capita

Trucks per unit of GDP

Passenger cars per capita

Aviation cargo (TKMs) per unit of real GDP

Aviation passengers (PKMs) per capita

4 Telecommunications Mail per capita

Telegrams per capita

Telephones per capita

Mobile phones per capita

5 Mass communications Newspapers per capita

Radios per capita

Televisions per capita

6 Information

Technology Personal computers per capita

Industrial robots per unit of real GDP

7 Transportation

(shipping) Fraction of merchant fleet (tonnage) made up of sail

ships

Fraction of merchant fleet (tonnage) made up of

steamships

Fraction of merchant fleet (tonnage) made up of motor

ships

8 Electricity MWh of electricity produced per unit of real GDP Source: Comin and Hobijn (2004)

16

In Table 2.1 we can see that for steel technologies the measures include shares of output

produced. For textile technologies the constructs are based on capital shares which measure the

fraction of capital stock that is made up of equipment that embodies a specific technology. The

measure for transportation based technologies is capital output ratios. For technologies in this

sector that do not have data on capital output ratios the data employed is based on production to

GDP ratio. For the rest of the technologies in the sample the measures are based on capital stock

per capita such as mobile phones per capita and consumption per capita for mail and telegram

technologies.

Comin and Hobjin (2004) employ these direct measures of technology contained in

HCCTAD and suggest that leading economies of the world are the main innovators of the

majority of technologies. In the first stage the adoption takes place in these innovating

economies followed by technological adoption in economies that lag behind. Moreover, better

human capital and higher income per capita are associated with increase in the rate of adoption

of technologies. Based on this evidence they argue that use of direct measures of technology has

three apparent advantages over measures such as TFP/ Solow residual when explaining the

differences in cross-country technology and adoption levels of specific technologies. Firstly,

direct measures of technology are relatively more disaggregated measures and their use prevents

the problem of heterogeneity which exists in aggregate measures across countries and overtime.

Second, the availability of data on a particular set of technologies allows the possibility of

analyzing interactions across their adoption levels. Lastly, these measures are “micro” constructs

of technology; therefore the correlations with aggregate explanatory variables are interpreted as

“causal relations”. Moreover, their panel data analysis shows that human capital endowments are

17

among the important determinants influencing the timing of adoption of technology in an

economy.

Comin and Mestieri (2013) extend their HCCTAD dataset by including more

technologies and countries and develop a more comprehensive data known as Cross Country

Historical Adoption of Technology (CHAT) data set. This data set covers 104 technologies for

more than 150 countries over the last 200 years. The direct measures of technologies in this data

set are based on the notion of technology that captures technology “as a manner of

accomplishing a task employing technical processes, methods or knowledge”. Developing on

this concept the measures for technology in the CHAT data set are (i) the amount of capital

goods required to completing specific tasks (ii) number of particular tasks that have been

completed (iii) the number of users of specific manner to complete a task.

These measures of technology can be divided into two categories: intensive and extensive

measures. The intensive measures of technology adoption are based on units of a specific

technology in use relative to the size of the economy whereas extensive measures refer to timing

of adoption of a technology. This data set includes both the intensive and extensive measures to

explain adoption of technologies from eight sectors of economy which include agriculture,

finance, health, steel, telecommunications, textiles, tourism and transportation.

Comin and Mestieri (2013) present a more formal argument for direct measures of

technology and suggest that these measures of adoption of technology should be central to any

examination of mechanisms through which technology adoption and diffusion influences

economic growth. In addition, based on these measures of technology various studies explain the

dynamics and role of technology in growth (Comin and Hobijn, 2004 and 2009; Comin et al,

2008; Comin and Hobijn, 2010; Comin and Mestieri, 2013). In particular, Comin and Hobjin

18

(2010) explain the dynamics of adoption lags by employing usage lags which refer to the

difference in the usage level of a specific technology at a particular point in time relative to

technology leader. They suggest that over a period of time these lags have become smaller. They

plot the average adoption lags of a set of 15 technologies relative to their date of invention given

in Figure 2.1 below.

Figure 2.1 Technology Adoption Lags Adapted from Comin and Hobijn (2010, page. 2049)

Figure 2.1 shows that newer technologies invented in the recent years have a faster rate of

diffusion compared to the older technologies. Comin and Hobijn (2010) also argue that the pace

of technological adoption gained momentum before the digital revolution. This implies that

technological diffusion has been going on at this rate for the past 200 years or so. Given this if

technological diffusion continues at this pace it can have a major impact on the cross-country

differences in TFP that exists between the rich and poor economies of the world. They predict

that the TFP gap in the future between these economies will reduce due to faster technological

diffusion.

Steam and steam motor ships

Railways-passengers

Railways-freight

Cars

Trucks

Aviation-passenger

Aviation-freight Telegrams

Telephone

Cell phones

PCs

Internet users

MRIs

Blast oxygen

Electricity

0

20

40

60

80

100

120

140

1775 1800 1825 1850 1875 1900 1925 1950 1975 2000

TE

CN

OL

OF

Y

AD

OP

TIO

N

YEARS

19

The literature using direct measures of technology also explores improvement in human

capital and its impact on adoption and diffusion of specific technologies. Comin and Hobijn

(2004) use secondary school enrollment as a measure of human capital and examine the diffusion

and adoption of 25 technologies in 15 advanced countries for the past 200 years. The empirical

specifications examining the impact of human capital in diffusion of technology are similar to

Caselli and Coleman (2001) and allow for capturing the different effects of enrollment rates

before and after 1970. Their findings show that until 1970, secondary school enrollment is

positively associated with technology adoption. More specifically, adoption of mass

communication technologies such as newspapers, radio and televisions are positively associated

with secondary schooling. In addition, adoption of transportation technologies has a positive

association with primary enrollment levels and adoption of computers increases due to

improvement in tertiary education.

2.4 Qualitative Dimension of Human Capital

The literature on technology reviewed above suggests that human capital is one of the

important determinants of technology adoption and diffusion. These studies employ average

years of schooling or enrollments as quantitative measures of human capital and reveal that it has

a positive and significant association with economic growth and is one of the possible

mechanisms assisting adoption and diffusion of technology. However, as suggested by Hanushek

and Kimko (2000) quantitative measures of human capital are not correct measures of human

capital as these implicitly assume that a year of schooling delivers the same amount of increase

in knowledge and skills to a student regardless of the quality of educational system of a country.

According to them it is unrealistic to consider one year of secondary schooling in Egypt being

20

equal to a year of schooling at the same grade in United States. In addition, another drawback of

the earlier literature is that it assumes that educational outcomes are weakly influenced by the

changes in the quality of non-school factors.

Hanushek and Kimko (2000) introduce a new construct of human capital measured as

educational achievements, also termed as qualitative measures of human capital or cognitive

skills. We elaborate on this strand of literature which uses cognitive skills as a measure of human

capital because the hypothesis of the first study employs this definition of human capital for

empirical analysis. Hanushek and Kimko (2000) examine the association between quality of

labour force and economic growth by employing data on international student achievement test

scores from 1960 to 1990 for a set of 31 countries as measures of human capital. The results of

their analysis suggest that labour force quality measured using test scores has strong and

significant impact on economic growth. In addition, they do not find any evidence suggesting

that this association between the two is a result of growth being associated with improving the

quality through an investment in the school resources. Later on, Hanushek and Woessmann

(2011) document a series of analysis based on different tests and specifications to reinforce the

robust relationship that exists between cognitive skills and growth.

Furthermore, Hanushek and Woessmann (2012) present the advantages of using test

scores as measures of cognitive skills and human capital. Their measures of human capital are an

extension to Hanushek and Kimko (2000). The data set in their study includes additional

international tests for countries along with several time and country dimensions. More

specifically, it includes international test scores of mathematics, science and reading from

TIMSS and PISA for different age groups administered between 1964 and 2003. They advocate

the use of test scores can capture the variations that exist in the knowledge and ability produced

21

by schools and can be considered as school outputs which have impact on economic growth. In

addition, these test scores can be treated as the total outcomes of education which are due to

different sources such as family, schools or one’s own ability. Lastly, variations in test scores are

indicators of differences in student performance which are attributed to the differences in school

quality. Based on these arguments supporting use of test scores as measures of human capital

they also examine the association between human capital and gross domestic product (GDP)

growth. Their cross country examination reveals a strong cognitive skills-growth relationship,

which is robust to sensitivity analyses allowing for changes in specifications, country samples

and time period.

Hanushek and Woessmann (2012) in the context of technology suggest that countries

need highly skilled human capital for the purpose of imitation of technology. Based on the

country evidence for Taiwan, Singapore and Korea they show that skilled human capital

accelerates the process of economic convergence. More specifically, according to them the

exceptional growth rate of these countries in the past is attributed to the large share of high

performers in their population. Hence, if a country wants to be better able to imitate and

innovate, the technology and strategies developed by highly skilled human capital such as

scientists it must also have a labour force with at least basic knowledge and skills.

Given that differences in cognitive skills across countries contribute to significant

differences in economic growth, Hanushek and Woessmann (2015) plot variations in test scores

and growth rates of countries referred to as conditional test scores and growth rates. Figure 2.2

below explains this association between test scores and growth rates for European Union (EU)

economies constituting their sample.

22

Figure 2.2 Educational Attainments and Economic Growth

Adapted from Hanushek and Woessmann (2015, p. 7)

As evident from Figure 2.2 economies that do well in terms of better test scores tend to

have higher long run growth rates compared to countries with poor educational attainments. In

addition, there is a very strong effect between variation in test scores and growth rates. Their

estimations indicate that half a standard deviation in test scores amounts to 1 percentage point

higher long run growth for a country. These results show a presence of a close link between

educational achievements of nation’s population and its long run growth rate.

In summary, the discussion for the first study presents insights about the nexus between

human capital, technology and economic growth. The literature reviewed suggests that human

capital perhaps is the most important factor influencing economic growth, as it is one of the

channels facilitating adoption and diffusion of technology. Differences in human capital lead to

differences in the ability of countries to adopt new technologies and impact upon their growth

prospects. Furthermore, countries with educated and skilled labour force are more suited to

technology adoption as they face lower skill barriers in comparison to economies with shortage

23

of such work force. This technological view of human capital has received support in studies

employing both quantitative and qualitative measures of human capital as well as indirect and

direct measures of technology.

Given that the above mentioned literature shows that human capital is linked to growth,

the relationship between inequality in human capital and economic growth is also an issue that

has been examined extensively in macroeconomics and development literature. As mentioned in

the introduction the second essay aims to unearth the composition and factors associated with

human capital inequality, the following section therefore, reviews inequality in the context of

income and most importantly human capital within and across countries and its impact on

economic growth.

2.5 Perspectives on Income and Human Capital Inequality

The first part of this section will look at the literature on growth and income inequality,

while the next part will review relevant studies on inequality in the context of human capital.

Finally, we discuss literature examining inequality and its determinants using quantitative and

qualitative dimensions of education as a measure of human capital.

The relationship between income inequality and growth was initially explored by

Kuzents (1955). He observes an initial decline in rate of growth and a simultaneous decline in

income inequality in the United States, United Kingdom, and Germany. Based on this the

Kuznets’ curve suggests that income inequality increases in the earlier stages of industrialization

process and shrinks in the later phases of growth. A possible interpretation of this relationship is

that higher inequality stirs up growth which leads to reduction in inequality through a “trickle

down” mechanism. Later on, Acemoglu and Robinson (2002) revisit this hypothesis and provide

support to Kuznets’ curve for several countries. However, countries such as Japan, Norway,

24

Netherlands, South Korea and Taiwan show that development does not necessarily result in a

concurrent decline in inequality. Hence, in general the literature is inconclusive about the

Kuznets type phenomenon; some studies reveal a relationship similar to the inverse U-shaped

curve, others find it to be either negative or inconclusive (Borissov & Lambrecht, 2009; Shin,

2012).

There are several studies that attempt at establishing the direction in which inequality in

income impacts on growth. For instance, Alesina and Rodrik (1994) using cross sectional data

and Gini coefficient as a measure of inequality show a negative effect of inequality on growth. In

a similar empirical setting using the third quintile share of income Persson and Tabellini (1994),

also suggest that reduction in inequality seems to encourage growth. Supporting studies include

Clarke, (1995); Deininger and Squire, (1998); Castello and Domenech, (2002); and Knowles,

(2005). Studies based on panel data also support the presence of an inverse association between

inequality and economic growth (Banerjee and Dufflo, 2003; Castello, 2010; and Ostryet al,

2014). In brief, this line of literature shows that inequality does not favour long-run growth.

On the contrary, empirical evidence also indicates that an increase in inequality promotes

economic growth. For instance, Li and Zou (1998) using panel data and Gini coefficient as

measure of inequality show that inequality has a positive effect of growth. Forbes (2000) finds

that in short and medium term income inequality positively and significantly impacts on growth.

Recent studies employing dynamic panel estimations also reveal a positive impact of inequality

on growth (Deininger and Olinto, 2000; Halter et al, 2014).

Another line of studies in the literature on inequality investigates the income distributions

within and between countries. Li et al (1998), use a comprehensive data of Gini coefficients for

112 countries for the years 1947-94. According to them inequality in income is comparatively

25

stable within rather than between countries. In order to explain international and inter-temporal

variations in income they employ the arguments based on capital market imperfections and

political economy. Their empirical analysis shows that financial depth and initial distribution of

land along with civil liberties and secondary schooling are among important factors influencing

inequality. Milanovic and Yitzhaki (2002), also decompose world income inequality into within

and between country components. They perform an international comparison for five continents:

Africa, Asia, North America, Oceania, Western and Eastern Europe. In case of Asia they find

that between-country component of inequality accounts for a larger proportion of total inequality

than within-country component. On the contrary, this is not the case for Africa and Latin

America where the between-country inequality is smaller in proportion to within-country

inequality. In case of North America and Western Europe they found that both within and

between inequalities are low. For the transition economies situated in the Eastern Europe both

components of inequality seem to show a similar percentage contribution to total inequality.

The above-mentioned literature argues that income inequality is one of the factors that

influence economic growth. However, another strand of literature emphasizes that inequality in

human capital also impacts on growth. It highlights that human capital inequality is another

important measure of quality of life along with income. The following section therefore reviews

the literature which examines these hypotheses because they are linked to the empirical analysis

performed in this thesis.

26

2.6 Human Capital and Inequality

This strand of literature focuses on studies examining inequality in human capital which

characterizes differences in standards of living, its impact on growth and distribution of human

capital across countries. In particular, Sen (1979, 1985, 1987) suggests that income is not a

sufficient measure of human well-being. According to him other dimensions of human capital

such as education, health, civil liberties and political freedom are equally important factors

influencing quality of life. Hence, inequality in society measured exclusively on the basis of

income does not account for inequality in other facets of an individual’s life such as education

and health (Oppedisano and Turati, 2011). Moreover, literature on growth underscores the role of

better education in accumulation of human capital as a key to economic growth (Hanushek and

Kimko, 2000; Krueger and Lindhal, 2001 and De La Fuente and Domenech, 2006).

There exists evidence which suggests that educational inequality is among the main

determinants of income inequality (Glomm and Ravikumar, 1992; Saint- Paul and Verdier, 1993

and Galor and Tsiddon, 1997). Supporting studies further confirm this evidence and argue that

educational inequalities not only influence income distribution but also lead to differences in

productivity (Park, 1996; Gregorio and Lee, 2002; Checchi, 2004; Acemoglu and Dell, 2010). In

addition, studies also analyze inequality in human capital bearing an influence on variables other

than economic growth. For example, empirical evidence points out that educated human capital

contributes to better health and labour market outcomes along with a lesser possibility to engage

in crime (Harmon et al, 2003; Lochner and Moretti, 2004; and Grossman, 2006).

Castello and Domenech (2002) employ one the most comprehensive data sets on human

capital by Barro and Lee (2001) and examine the association between Gini coefficient in terms

of years of schooling and economic growth. Their results indicate that the variations in

27

educational attainments are associated with lower investment rates. They argue that countries

exhibiting higher inequality in education tend to have lower investment in education which

translates into lower income and economic growth rate. Moreover, variations in educational

achievements between countries are greater across countries rather than within countries. Some

studies highlight the role of demographics as one of the mechanisms through which inequalities

in education may impede growth (Castello and Climent, 2010a and 2010b). This finding rests on

the argument that uneducated groups have higher fertility rates and lower life expectancy which

inhibit investment in education. Furthermore, an increase in number of literates causes a decline

in human capital inequality without reducing income inequalities in the world (Castello and

Domenech, 2014). Overall, using macro-economic data on average years of schooling as

quantitative measures of education and human capital, these studies provide evidence of human

capital inequality influencing growth.

On the other hand, Woessman (2014) uses qualitative measures of education such as test

scores and shows that skills acquired are more important measures of educational attainment

than years of schooling. An increase in educational achievements (test scores) contribute to

higher economic growth in the long run. Hanushek and Woessman (2008, 2012, 2015) suggest

that the performance of a country’s population on achievement tests particularly in mathematics

and science is highly linked to an economy’s long run growth rate. Given that educational

achievements affect growth, the analysis should not be restricted only to the use of years of

schooling as a measure of human capital.

The literature on human capital inequality does not restrict itself to establishing an

association between educational inequality and growth. It also includes studies that investigate

educational achievement distributions within and between countries and reveal the causes of

28

educational inequality. In particular, Sahn and Younger (2007) employ standardized mathematics

and science country test scores from Trends in Mathematics and Science Study (TIMSS) for the

years 1999 and 2003. They decompose inequality in educational achievements using generalized

entropy index. Their cross-country macroeconomic analysis reveals that more than half of the

inequality stems from within-country differences in educational achievements. A comparison of

inequality across the two years indicates that decline in inequality took place for more countries

in case of science in contrast to mathematics test scores. The analysis also shows presence of an

obvious correlation between dispersion and average test scores. Moreover, for a broad set of

inequality values there exists a narrow band of science and mathematics test scores. This implies

that countries with comparable levels of test scores can have a different degree of educational

dispersion level from each other.

Freeman et al (2010) use fourth and eighth grade mathematics test scores from the 2000

and 2009 data sets for Programme for International Student Assessment (PISA). Their measure

of inequality is calculated individually for each country in the sample. They first calculate a

median score of students, then the measure of dispersion of these scores within the country is

calculated as the ratio of the difference between the 95th

and 5th

percentile score divided by the

median. Based on this set of calculations they reveal wide cross-country variation in level and

dispersion of test scores with highest scores associated to countries having least inequality in

scores.

Oppedisano and Turati (2011) examine the evolution of human capital inequality

between 2000 and 2006 in nine European countries by focusing on PISA reading test scores.

Their examination reveals a decline in inequality in only Germany and Spain, while an increase

was observed for France, Italy, Greece, Norway, Portugal, Sweden and United Kingdom. They

29

also decompose inequalities into their causes and analyze the trends over time for a few selected

countries such as; France, Germany and Italy. Their results show that parental and school

characteristics are important determinants of inequalities in educational achievements among

students. Other studies using PISA mathematics test scores show that students who are more

socio-economically advantaged score 39 points higher compared to less advantaged students.

This difference between test scores implies that students from less advantaged background are

one year behind in schooling compared economically advantaged students (OECD 2013a).

Furthermore, studies for OECD countries have even suggested a bigger difference of 95 PISA

test points; this is similar to being two and a half years behind in school (OECD, 2013b). In

brief, the above-mentioned literature employs both educational quantity and quality measures of

human capital and reveals that it is one of the important measures of human well-being which is

also associated with the growth of an economy.

2.7 Conclusion

To summarize, the above discussion presents a review of the main themes that are of

relevance to the two studies that constitute the thesis. In particular, we discuss the role of human

capital determining the pace of economic growth through adoption and diffusion of technology.

Within this literature we review studies which examine this role employing alternate measures of

both human capital and technology. We also provide an account of the literature examining

inequality in income and its impact on growth. Following the inequality debate we focus on

inequality in human capital and its intricate links with economic growth. In addition, we review

studies that explore human capital inequality by examining measures such as cognitive skills

rather than years of schooling, and further study their distribution and determinants across

countries. These studies provide the background and context to subsequent chapters. The next

30

chapter focuses on the theme of human capital while Chapter 4 looks at the decomposition of

human capital inequality and its determinants.

31

Chapter 3

Human Capital and the Adoption and Diffusion of Technology

3.1 Introduction

A substantial strand of literature on the relationship between education and technological

diffusion stems from the work of Nelson and Phelps (1966), who show that human capital

accumulation, through its impact on technology adoption and diffusion, influences an economy’s

ability to catch up with more developed economies. Benhabib and Speigel (1994) extend this

approach by emphasizing that human capital not only helps in the adoption of more sophisticated

technologies but also facilitates development of new technologies at the frontier through better

innovation. They show that the positive link between human capital and economic growth rests

critically on both of these mechanisms. Subsequent empirical developments present evidence

that is either supportive of this view (as in Barro and Sala-i-Martin, 1995 and Barro, 1998), or

supportive with caveats pertaining to the level of development (as in Krueger and Lindahl, 2001)

or the measure of human capital used (as in Vandenbussche, 2006; Messinis and Abdullahi, 2010

and Madsen, 2014).

One of the drawbacks of the previously mentioned studies is that they consider changes in

total factor productivity as a measure of technological change. However, changes in productivity

growth do not properly account for changes in technology (Hulten 2000, Lipsey and Carlaw

2004), given that total factor productivity is a “residual” from growth accounting exercises which

can be related not only to technological change, but other unmeasured inputs in the process of

production. Moreover, as suggested by Comin and Mestieri (2013), indirect and traditional

measures do not distinguish between the extensive and intensive margins of technology adoption,

which should be central to any examination of mechanisms through which technology adoption

32

impacts on growth. The intensive margin refers to the intensity of use of a new technology in a

given economy while the extensive margin refers to the timing of adoption – i.e lag in adoption

of a technology for the first time relative to the leading adopter of a technology. This concept is

termed as usage lag was first defined in Comin et al (2008). If, as the human capital and

technology diffusion literature mentioned above suggests, human capital influences growth

through its impact on technology adoption and diffusion, the appropriate empirical exercise to

address this issue should focus on direct measures of both human capital and technology

diffusion.

A key objective of this study, therefore, is to empirically investigate and analyze the link

between technology and educational quality in the light of direct measures of technology

adoption and diffusion, as well as of educational quality. To that end, we examine the impact of

educational quality, as measured using the data set on cognitive skills created by Hanushek and

Woessman (2012) on direct measures of technology adoption and diffusion based on the recently

created Cross Country Historical Adoption of Technology (CHAT) data set due to Comin and

Hobijn (2009).5 To our knowledge, this is the first attempt to examine the link between human

capital and technology adoption and diffusion by incorporating disaggregated qualitative aspects

of education (in the form of cognitive skills measured as Trends in International Mathematics

and Science Study (TIMSS) test scores and direct measures of technology.

The literature on cognitive skills and growth suggests that the quality of human capital has a

close, consistent and stable relationship with economic growth (Hanushek and Kimko, 2000;

Hanushek and Woessman, 2012).6 However, in this paper we suggest that the mechanisms which

5 The Cross Country Historical Adoption of Technology (CHAT) data set captures both the extensive and intensive

margins of 104 technologies from 8 sectors for a sample of more than 150 countries, over a period of 1800-2000. 6 The literature that uses quantitative measures of human capital, such as years of schooling and enrolment rates in

contrast exhibits mixed evidence on the link between human capital and economic growth.

33

transform human capital into output are intrinsically related to the nature of technology in

question, an issue that is relatively neglected in this literature. For example, certain technologies

require a higher embodiment of skills and educational quality than others, and this is one of the

premises of our exploration. This premise is in part inspired by the findings presented in Comin

and Hobijn (2004) who explore the link between quantitative measures of human capital and

technology adoption, and suggest that human capital is an important determinant of the intensity

of adoption. However, their regressions pool a large set of technologies into one panel, making it

difficult to address this specificity.

Following this idea, we suggest that in an analogous sense, specific types of qualitative

measures of human capital may be more or less appropriate or relevant in facilitating adoption

depending on the type of technology in question. For example, cognitive skills as represented by

science scores may be more relevant to the adoption and diffusion of medical technologies, while

mathematics scores, which arguably embody analytical skills of a more generic nature, could be

relevant for a larger set of technologies including medical technologies, computers or digital

technologies and technologies relating to transportation. In the analysis to follow, therefore, we

prefer to refer to the human capital measure associated with mathematics scores as “generic

human capital”. The human capital measure associated with science scores is referred to as

“specific human capital”.7

Apart from the two dimensions of human capital mentioned above –i.e. ‘generic’ and

‘specific’ human capital, a third dimension pertains to what has often been referred to as

“learning by doing” in several theoretical and empirical studies of technology adoption (Parente,

7 This may be justifiable in the sense that the mathematics test consists of basic mathematical knowledge applied to

set of analytical problems. The science test, in contrast, is more knowledge specific rather than analytical. Of course,

this may be contentious and the reader may not agree with our interpretation. Our choice of the labels ‘specific” and

“generic”, however proves convenient as well as intuitive in the context of discussing and interpreting the results to

follow.

34

1994; Jovanovic and Nyarko,1996; Conley and Udry, 2010). This aspect of technology adoption

stresses the notion that the productivity of technologies depends on the experience of using and

adapting the technology to local conditions, and the insufficiency of this type of human capital

can present barriers to the adoption of such technologies (Basu and Weil, 1998; Acemoglu and

Zilibotti, 2001; Lahiri et al 2018). However, while direct measures of such human capital are not

available in disaggregated technology-specific form, a simple way of capturing this aspect is to

examine the impact of past levels of usage intensity and usage lags of the technology in question.

Therefore, another objective of our study is to capture this aspect and examine its implications

for technology adoption. In terms of our methodology, we do so by incorporating lags of the

dependent variable in our regressions, along with the human capital measures based on the

TIMSS data set.8

As our study analyzes two dimensions of technology, usage intensity and usage lags of

technologies, we may also argue that a change in the measure or dimension of technology may

bring a change in the association between a particular technology and skill under discussion. For

instance, human capital embodying knowledge of numeracy skills may not be as relevant in

reducing adoption lags of a digital technology, since the invention of that technology took place

elsewhere and other factors, such as trading relationships and property rights have a greater

bearing on when the transfer of that technology takes place. However the usage intensity after

adoption may depend more critically on such human capital.9

8 In addition to our reasoning above Comin et al (2008) suggest that past level of technology adoption is a strong

predictor of current levels; as such a dynamic specification is appropriate. In Comin and Hobijn (2004), which to our

knowledge is the only other study analyzing the impact of human capital on technology measures based on the

CHAT data set, the lagged variable is not considered and the focus is on quantitative measures of human capital

such as secondary school enrollment. 9 While such technologies do not require mathematics skills per se, their prevalence requires human capital in the

form of qualified technicians and engineers to provide maintenance and technical support services. It is in this sense

that we suggest that the generic nature of mathematics skills is relevant. Following Hanushek and Kimko (2000) we

interpret these measures as an indirect proxy of the quality of the labour-force of an economy.

35

In order to explore these issues we create two panels based on science and mathematics

scores from TIMSS and technology adoption measures from CHAT for the years 1964-2003 and

1973-2003 respectively. Given that we add a lagged measure of technology in our empirical

specifications in addition to other human capital measures, dynamic-panel methodologies are

required. For this purpose, we employ the dynamic generalized method of Moments (GMM)

methodology due to Arellano and Bond (1991). In our specifications we also include certain

control variables that may be of relevance to technology adoption and diffusion, such as health

and foreign direct investment (FDI), but have received less attention in previous literature

pertaining to these issues.10

Further, in order to compare the impact of qualitative and

quantitative measures of human capital, we also include the average years of schooling measure

from Barro and Lee (2010).

The results support our premise regarding the technology-specific nature of the link between

human capital and technology adoption. For example, our analysis of cognitive skills based on

mathematics test scores suggests that the generic type human capital associated with these scores

is more likely to have a positive impact on the usage intensity of same technologies we consider,

particularly in the transportation, tourism and health sectors. We note, however, that not all

regressions yield positive and significant coefficients for the human capital variable in these

sectors. Furthermore, this type of human capital does not seem to exhibit any clear-cut link with

technology adoption in agriculture as regressions based on a variety of technologies in this sector

have coefficients of human capital that are either negatively significant or positive but not

significant. In our interpretation this does not necessarily suggest that human capital does not

10

Barro (2013) uses life expectancy to measure the dimension intrinsic to human capital by introducing it in the

literature on economic growth. Sinani and Myer (2004) and Branstetter (2006) highlight the role of foreign direct

investment on technological spillovers which contribute to physical capital accumulation, increasing domestic

employment and generating positive effects on domestic industries and firms. We introduce these measures in this

study to control for possible determinants of usage intensity and usage lags of technology.

36

matter for the adoption of agricultural technologies. Adoption of technologies in agriculture, for

example, may require a different dimension of human capital in the form of “learning by doing”

of the type suggested by Foster and Rosenzweig (1995) in the context of technologies such as

high-yield varieties of seeds.

Indeed, the lag of the technology measure, which we interpret as representative of the

experiential, learning-by-doing aspect of adoption, is positive and significant not only in the case

of agricultural technologies, but highly significant across all regressions. This measure remains

positive and significant in the regressions based on usage lags of the same technologies,

suggesting that shorter time lags in adoption in the past lead to even shorter lags in the present,

quickening the pace of adoption as more time has been spent on learning a particular technology.

Regarding the usage intensity regressions, it is interesting to note that the “generic” human

capital measure associated with mathematics scores yields a positive and significant impact on

usage intensity in only 10 out of the 21 technologies we consider. In the case of usage lags

evidence regarding the hypothesis that human capital facilitates adoption by reducing adoption

lags is substantially weaker; only 3 regressions yield a negatively significant coefficient for the

variable representing human capital. The evidence based on the “specific” measure of science

further reinforces this point. In this case, the human capital measure has the hypothesized impact

on usage intensity in only 5 out of the 21 regressions. Likewise, we find that the coefficient of

the human capital variable in the usage lag regressions is negative and significant only for 4 of

these technologies. The lagged technology measure, however, remains positive and significant

across all regressions.

Furthermore, the regressions also suggest that qualitative measures of generic human capital

matter more relative to quantitative measures such as average years of schooling and other

37

measures of human capital such as life expectancy; there are very few regressions for which the

coefficients of these variable is positive and significant.

However, the above summary of results is indicative the broad themes that emerge based on

the number of times a particular variable is significant in all of the regressions. If we look at the

regressions for various technologies individually there are a few exceptions where we find the

impact of quantitative measures of education and life expectancy are larger compared to

qualitative dimensions of human capital. When the former measures are significant, their

coefficients in the corresponding regressions can be much larger in comparison to those

associated with qualitative measures.

Referring back to the literature suggesting a strong and stable positive impact of human

capital as measured by cognitive skills on economic growth, as in Hanushek and Woessmann

(2012) it is perhaps surprising that the impact of this measure is not persuasively positive in the

context of technology adoption, which is regarded as a mechanism through which growth takes

place. Even so, we believe analyses of this type, focusing on mechanisms of growth rather than

growth per se are more informative from the point of view of policy. Here, the insight that

emerges is that the notion of human capital relevant for different types of technologies is diverse,

and not easily captured by either the qualitative measures (such as test scores) or quantitative

measures (such as years of schooling). Further, there is robust and clear-cut evidence to suggest

that the learning-by-doing aspect associated with technology adoption matters, given the

significance of the lagged measure of technology in all specifications considered in our analysis.

The appropriate design of policy, then, is better informed by examining the nature of

technologies and the types of human capital more relevant for their adoption.

38

Furthermore, we consider a few robustness exercises. Based on the finding in Comin et al

(2008) that the coefficient of income or GDP lag, a proxy for the extent to which a country has

caught up with the most developed economy, is important for the diffusion of technology. We

therefore include this variable in our usage lag regressions and find that while the coefficient of

this variable is positive in majority of them, it is insignificant in most cases. Hence, the dynamics

of technology and qualitative measures of human capital appear to be more important

determinants of technology diffusion rather than the dynamics of income. Lastly, we also

consider the inclusion of institutional quality, in the form of political rights and civil liberties

along with GDP per capita and R&D expenditures to carry out some additional robustness

checks. Even when we control for these variables the signs of the coefficients of human capital

as measured by cognitive skills and the lagged dependent variable remain similar to baseline

regressions in both the mathematics and science panel estimations.

To summarize, our study considers the impact of qualitative (generic and specific) and

quantitative measures of human capital on technology adoption. We also consider the

experiential, learning associated aspect through the presence of past levels of technology in our

specifications. We find that the most important determinant of technology adoption is the past

level of technology, reflecting the importance of the learning-by-doing aspect of technology

adoption. Qualitative measures also matter, but are conditional on the nature of technology, with

generic skills being more relevant compared with specific skills. Finally, quantitative measures

such as average years of schooling matter even less in comparison with qualitative measures.

Based on our analysis, we suggest a multi-dimensional approach to studying barriers to

technology adoption may be more informative from a policy making point of view. However,

39

given the importance of learning-by-doing a common prescription that emerges pertains to

technology specific and vocational training approaches to deepen human capital.

The remainder of this chapter is organized as follows: Section 3.2 outlines the main features

of theoretical and empirical framework relevant to our study. In section 3.3 we summarize results

analyzing the role of cognitive skills in the process of technology adoption. Section 3.4 examines

this role from the perspective of diffusion of technologies within and across selected sectors.

Section 3.5 provides the results for robustness checks. Lastly section 3.6 presents our

conclusions.

3.2 Empirical Methodology

In what follows we provide a brief review of our measures of adoption and diffusion of

technology and cognitive skills. We also present the econometric specifications examining the

role of cognitive skills in the process of adoption and diffusion of technologies.

3.2.1 Measures of Technology Adoption and Diffusion

In this section we briefly explain our measures of technology adoption and diffusion,

which we borrow from Comin, et al (2008).11

They consider two measures: usage intensity and

usage lags. The former is relatively simple and captures the intensity with which each adopter

uses the technology-i.e., intensive margin.12

In our study usage intensity is measured as the

number of technology employed at a particular point in time scaled by the population in a

11

Comin and Hobijn (2009) develop their notion of technology drawing from the definition in Merriam-Webster’s

Collegiate Dictionary. It defines technology as “a manner of accomplishing a task especially using technical

processes, methods, or knowledge”. Given this definition the basic idea behind technological measures in CHAT is

to cover these various aspects of technology. For example, it includes the quantity of capital goods required to

achieve a specific task, the number of times a specific task has been completed and the number of users of a the

specific manner in which the task was accomplished. A more elaborate discussion of these ideas is presented in

Chapter 2. 12

Comin and Mestieri (2013), use a different theoretical construct for intensive margin of technology in their

theoretical framework.

40

country.13

Therefore, the usage intensity of technology conceptually measures the per capita

usage of technology instead of measuring technology adoption simply as the number of units of a

particular technology available in an economy for each year in our analysis. Using this technique

we estimate usage intensity of technology for 14 technologies in six sectors given in the CHAT

data set.14

However, the latter measure, i.e. usage lags is more complex, we choose to explain it in

this section for the benefit of the reader. Our discussion is similar to Comin et al (2008).

However, we believe that the discussion is worthy of reiteration for the sake of reader’s

convenience. To provide an intuitive explanation for the concept of technology usage lag, we

plot the usage levels for internet for Australia, US, France, Japan and Netherlands in Figure 3.1,

and perform an exercise similar to Comin et al (2008).15

Specifically, we ask the question: how

many years before the year 2002 did the United States last have the usage level that Japan had in

2002? As is visible from the figure, US last passed Japan’s 2002 usage level in 2000, 2 years

before 2002. Similarly we can perform this exercise for other countries in our sample to find that

in 2002 US led Australia by a few months, France by 4 years and Netherlands by a year. This

illustration makes it somewhat easier to understand the theoretical definition provided in Comin

et al (2008). Again, since our analysis heavily draws on this measure we reiterate its method of

calculation here, rather than inconvenience the reader by omitting the explanation presented

below.

13

Comin and Mestieri (2013), suggest use of population or Gross Domestic Product as scaling factors. 14

We have 21 technologies in our sample, but the tables in the main text include results for 21 technologies. For the

complete sector-wise picture including all technologies refer to Appendix B. 15

This graphical representation is based on author’s own calculations for usage levels.

41

Figure 3.1 Graphical Representation of Technology Usage Lags.

0

100

200

300

400

500

600

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002

Us

ag

e le

vel

Years

Australia USA France Japan Netherlands

Australia

Netherlands

Japan

France

42

From a theoretical perspective technology usage lag x for a country c at time t explains

the time in terms of number years before a leader last had the same level ofusage of technology.

This shows difference in time period in the usage and adoption of a technology between a

country c and that of leader. Following Comin et al (2008), we denote Xj,t as the technology

usage intensity of a specific technology for country 𝑗 at a time period 𝑡. We evaluate this usage

level in country j with the past time series of the leader. Then the time series of U.S is given as

{𝑋𝑈.𝑆,𝑠} where the observations over time are indexed as S. As the time series data for U.S has

missing observations, let 𝑆 denote the set of observations available in the past data. In this time

series 𝑆 for U.S they further select two observations each indicating a technology usage intensity

level. In the first case they select�̅� :

�̅� = arg 𝑚𝑖𝑛𝑠∈𝑆

{𝑠|𝑋𝑈.𝑆,𝑠′ ≥ 𝑋𝑗,𝑡𝑓𝑜𝑟 𝑎𝑙𝑙 �́� ∈ 𝑆𝑎𝑛𝑑�́� ≥ 𝑠} (3.1)

In equation (3.1) 𝑠̅ is the set of observation that denotes the first time U.S passed the level of

technology usage 𝑋𝑗,𝑡 for country𝑗. On the other hand, the second observation𝑠 denotes the last

time U.S recorded a level of technology usage which was either equal or lower than 𝑋𝑗,𝑡 which is

given as:

𝑠 = arg 𝑚𝑖𝑛𝑠∈𝑆

{𝑠|𝑋𝑈.𝑆,𝑠 ≥ 𝑋𝑗,𝑡} (3.2)

Given these two observations, we denote 𝜏 as the last time U.S had technology usage level 𝑋𝑗,𝑡

which can be computed as follows:

𝜏 = (𝑋𝑗,𝑡−𝑋𝑈.𝑆𝑠

𝑋𝑈𝑆,𝑠−𝑋𝑈𝑆,𝑠) (𝑠 − 𝑠) (3.3)

43

Equation (3.3) shows, it is known that 𝑠 comes after observation 𝑠 in the historical time

series data for U.S, then technology usage lag between country 𝑗 and U.S at time 𝑡 can be given

as 𝑡 − 𝜏.

3.2.2 Measures of Cognitive Skills

Furthermore, following Hanushek and Woessmann (2012) we develop the measure of

educational quality to incorporate the dimension of human capital in our model. Educational

quality reflects the educational achievement measured as cognitive skills which are averages of

all observed mathematics and science scores for international tests conducted during the time

period (1964-2003) for a set of more than 50 countries.16

The common metric of educational

quality assists in tracking the distribution of cognitive skills and developing comparisons across

countries, time and tests. Hanushek and Woessmann (2012) develop this metric first by

standardizing the performances of students to make it comparable across time. This metric takes

US as the benchmark country, as it is the only country that has participated in all the

international tests. Given the time series evidence on test score performance for students from

US, the metric scales the current level of each International Student Achievement Tests (ISAT)

relative to the known previous comparable performance of students from students which is

expressed as:

𝑈𝑎,𝑠,𝑡𝑈𝑆 = (𝑁𝐴𝐸𝑃𝑎,𝑠,𝑡

𝑈𝑆 − 𝑁𝐴𝐸𝑃𝑎,𝑠,1999𝑈𝑆 )

𝑆𝐷𝑠𝑈𝑆,𝑃𝐼𝑆𝐴

𝑆𝐷𝑎,𝑠𝑈𝑆,𝑁𝐴𝐸𝑃 (3.4)

In equation (3.4), 𝑈 is the standardized performance difference of students from the

benchmark country US, 𝑎 is the age of student and 𝑠 denotes subject at relative time 𝑡, which is

16

The measure developed here is an extension of Hanushek and Kimko (2000). Details for countries and tests are

present in Hanushek and Woessmann (2012).

44

in this case year 1999. 𝑆𝐷𝑠𝑈𝑆,𝑃𝐼𝑆𝐴

is the subject specificstandard deviation of U.S students on

Programme for International Student Assessment (PISA) test, while 𝑆𝐷𝑎,𝑠𝑈𝑆,𝑁𝐴𝐸𝑃

is the age and

subject specific standard deviation of U.S students onNational Assessment of Educational

Progress (NAEP) test.

Moreover in order to bring in variation in test scores over time comparable across

countries they select a group of OECD countries as a benchmark to develop a comparable scale

for the variation on different ISATs.17

The framework transforms original test scores denoted as

𝑂 of country 𝑖, for each age 𝑎 and subject 𝑠 at time 𝑡 into a transformed test score 𝑋 which is

expressed as:

𝑋𝑎,𝑠,𝑡𝑖 = (𝑂𝑎,𝑠,𝑡

𝑖 − 𝑂𝑎,𝑠,𝑡𝑂𝑆𝐺̅̅ ̅̅ ̅̅ ̅)

𝑆𝐷𝑠,𝑃𝐼𝑆𝐴𝑂𝑆𝐺

𝑆𝐷𝑎,𝑠,𝑡𝑂𝑆𝐺 (3.5)

Given equation (3.5), the transformed test score 𝑋 has mean zero among the OECD

standardized group countries. Furthermore it shows that between country standard deviation

among the OSG and group of countries on the PISA test is the same in a particular subject. The

variation in the metric of rescaled test score termed as 𝑋 in the above equation is comparable

across tests. In order to generate the common metric for educational quality that is comparable

across time, country and subject, they combine equation (3.4) and (3.5), where the standardized

test score can be formally expressed as:

𝐼𝑎,𝑠,𝑡𝑖 = 𝑋𝑎,𝑠,𝑡

𝑖 − 𝑋𝑎,𝑠,𝑡𝑈𝑆 + 𝑂𝑠,𝑃𝐼𝑆𝐴

𝑈𝑆 + 𝑈𝑎,𝑠,𝑡𝑈𝑆 (3.6)

17

This group of countries is called OECD standardized group (OSG) which include countries: Austria, Belgium,

Canada, Denmark, France, Germany, Iceland, Japan, Norway, Sweden, Switzerland, United Kingdom and United

States.

45

Equation (3.6) gives the standardized test score 𝐼𝑎,𝑠,𝑡𝑖 . It determines the performance in ISAT for

all participating countries on a common scale that can be compared across ISATs. After

performing the standardization procedures this exercise provides cognitive skills measured as a

simple average of all standardized science and mathematics test scores of the ISAT’s for a

participating country.

3.2.3 Econometric Methodology

This section explains the empirical methodology used to examine the link between

technology diffusion and human capital. The specifications are shown in equations (3.7) and

(3.8) below.

𝑇𝑐,𝑡𝑖 = 𝛼𝑐 + 𝛾𝑇𝑐,𝑡−1

𝑖 + 𝛽1𝐶𝑆𝑐,𝑡 + 𝛽2𝐴𝑆𝑐,𝑡 + 𝛽3𝑋𝑐,𝑡 + 𝜇𝑐,𝑡 (3.7)

𝐿𝑎𝑔𝑐,𝑡𝑖 = 𝜃𝑐 + 𝛾𝐿𝑎𝑔𝑐,𝑡−1

𝑖 + 𝛺𝑌𝑐𝑡−𝑠 + 𝛽4𝐶𝑆𝑐,𝑡 + 𝛽5𝐴𝑆𝑐,𝑡 + 𝛽6𝑋𝑐,𝑡 + 𝜀𝑐,𝑡 (3.8)

In equation (3.7), 𝑇 is the usage intensity of technology, 𝐶𝑆 are the cognitive skills,𝐴𝑆 is

average of schooling, 𝑋 is a set of control variables and 𝜇𝑐,𝑡 is the error term. The subscripts

𝑖, 𝑐, 𝑡 denote a specific technology 𝑖, country 𝑐 and year𝑡 respectively. In equation (3.8) 𝐿𝑎𝑔 is

the usage lag of technology diffusion and the rest of the variables are the same as equation (3.7).

The dynamics of technology and the dimension of “learning by doing” are introduced as 𝑇𝑐,𝑡−1𝑖

and 𝐿𝑎𝑔𝑐,𝑡−1𝑖 to denote the lag of the dependent variables in period 1 in equation (3.7) and (3.8)

respectively. Here, we expect the sign of 𝛾 > 0. This implies a positive association between

previous period’s usage intensity and usage lag of technology with the current period’s usage

intensity and usage lag.

To further capture the dynamics of technology, in equation (3.8) we follow Comin et al

(2008) and introduce 𝑌𝑐𝑡−𝑠 which is the per capita income or GDP lag of a country. It is

46

measured in analogous fashion to technology usage lags, i.e. how far behind a country 𝑐 is in

GDP at time t compared to the GDP leader 𝑠 in the world. In this case the world GDP leader is

United States, which is also the world technology leader in our analysis. We expect the sign of

Ω >0 which implies that reduced income lags are associated with shorter technology usage lags.

While estimating these equations there is a possibility of the error term being correlated with any

of the explanatory variables in the model or with the lagged dependent variable. To address this

we employ the dynamic GMM estimators of Arellano and Bond (1991). These GMM estimators

take into account the dynamic nature of the model and correlation generated due to introducing

the lag of the dependent variable.18

In our analysis cognitive skills are a measure of human capital and educational quality. In

equation (3.7) we expect the sign of 𝛽1 > 0. This implies that human capital embodying skills

increases usage intensity of a given technology. In equation (3.8) we expect the sign of 𝛽4 <

0 which implies that better skills result in reducing timing of adoption of a given technology. In

addition, we include a quantitative measure of human capital as average years of schooling based

Barro and Lee (2010). In equation (3.7) the expected sign of the coefficient of this variable 𝛽2 >

0. This indicates that human capital with higher educational attainments enhances usage intensity

of technology. In equation (3.8) we expect 𝛽5 < 0 indicating that higher educational attainments

reduce the timing of adoption of technologies.

The control variables in our analysis include health and foreign direct investment (FDI) as

facilitators to technology adoption and diffusion. We include health as a second dimension of

human capital as it has gained importance in economic growth literature since the early 1990s.

Many studies suggest that health is one of the main components of human capital formation

18

In this estimation procedure we instrument current variables at time t by their past lags, which eliminate

correlation between explanatory variables and error term. For further details see Arellano and Bond (1991).

47

which contributes to economic growth as it facilitates the acquisition of skills and adds to

productivity (Ainsworth and Over, 1994; Jamison et al, 1998; Barro, 2013). However, there is a

dearth of studies that examine the role of health in technology diffusion and adoption from the

human capital perspective. We therefore add life expectancy in order to incorporate the health

dimension of human capital borrowing from Barro (2013). We obtain data for life expectancy for

the years 1964-2003 from World Development Indicators (WDI) of the World Bank (2015). It is

measured as life expectancy at birth in total years.19

Moreover, the literature on technology suggests that FDI inflows may contribute to spillovers

and affect domestic industries and firms (Sun, 2011). However, the empirical evidence in

relation to FDI affecting technology diffusion and adoption remains mixed (Aitken and Harrison,

1999; Li et al, 2001; Sun, 2011). Nevertheless, based on empirical support for the positive

impact of FDI as a determinant of technology adoption (Meyer and Sinani, 2009), we use it as

control variable in our analysis. The measure for FDI is drawn from the WDI of the World Bank

(2015) data for the years 1964-2003 and measured as net inflows of FDI as percentage of Gross

Domestic Product.20

3.3 Empirical Evidence on Measures of Human Capital and Usage Intensity of

Technology

We begin by estimating equation (3.7) to examine the association between human capital of

different types and technology diffusion as measured by usage intensity. We consider a larger set

of 21 technologies however, 14 are presented here. We include technologies considered in

Comin et al (2008), and in the interest of a more detailed analysis, some other technologies that

19

See www.worldbank.org 20

See www.worldbank.org

48

were not included in that paper.21

Specifically, we consider technologies in transportation,

tourism, telecommunications and information, health, electricity production and the agricultural

sector.22

In the interest of a succinct and salient presentation we provide results of selected

technologies in tables 3.1 and 3.2.23

Table 3.1 includes results of cognitive skills based on

mathematics test scores and usage intensity of technologies. Table 3.2 presents results of

cognitive skills based on science test scores and usage intensity of technologies. We reiterate

here that the former measure of cognitive skills is interpreted as being more generic in nature,

while the latter reflects skills that are of a specific nature.

A key finding of our empirical analysis is that the lagged dependent variable has a coefficient

that is positive and significant across almost all of the regressions we consider. The results are

presented in tables 3.1 and 3.2. As mentioned earlier we consider the lagged dependent variable

as reflective of the “learning-by-doing” dimension of human capital. We stress that this is only

our interpretation of the result; the caveat applies that such dimensions of human capital are hard

to measure directly.

21

Comin et al (2008) include technologies such as; electricity production, internet, personal computers, telephones,

cell phones, cars, trucks, passenger and cargo planes and tractors. 22

Appendix A contains definitions and descriptive evidence regarding the data used for analysis. Sector-wise results

of the 21 technologies in our sample are reported in the Appendix B. Appendix B consists of several tables

organized as follows: each table presents a sector of economy. The left-hand side panel reports results for generic

skills measured as mathematics test scores while the right-hand side report results for specific based cognitive skills.

In turn each of these panels consists of various sub-panels, which represent a particular technology in that sector. 23

Appendix B contains tables which provide complete sector-wise overview of these results and includes a larger set

of technologies. The more succinct presentation of these results in the form of tables in the main text does not affect

the overall findings and interpretation of the analysis.

49

Table 3.1 Usage Intensity of Technologies, Mathematics Skill Panel Estimations.

Variables

Aviation pkm/

Air

Shipton

Steam

motor/

Sea

Transplant

Liver

Transplant

Lung

Transplant

Bone

marrow

Cable TV Mail

Lagged

dependent

variable

0.889***

(0.032)

0.955***

(0.027)

0.7933***

(0.077)

0.2136

(0.1171)

0.817***

(0.06)

0.8615***

(0.03)

0.9075***

(0.028)

Cognitive

Skills

0.00087***

(0.0003)

-0.00005

(0.0004)

0.000012**

(0.000006)

0.000026***

(0.000006)

0.0000052

(0.00001)

0.1072***

(0.03)

0.000122**

(0.00004)

Years of

Schooling

-0.138***

(0.02)

0.0056

(0.003)

0.00069

(0.0005)

-0.00089**

(0.0003)

-0.00056

(0.001)

1.529

(2.61)

0.00048

(0.003)

Life

Expectancy

0.0413***

(0.013)

0.00011

(0.001)

-0.000303

(0.0002)

0.0004***

(0.0001)

0.0014

(0.0009)

0.0402

(1.5)

-0.0022

(0.001)

FDI

0.0136

(0.011)

0.00111

(0.001)

-0.000012

(0.00004)

0.000002

(0.00002)

-0.00003

(0.0001)

-1.113***

(0.31)

0.0033**

(0.001)

Observations 170 111 83 68 106 212 163 Standard errors in parenthesis; *,**,*** imply 10%, 5%, and 1% significance levels respectively. Lagged dependent variable

indicates AR1.

Table 3.1 (continued): Usage Intensity of Technologies, Mathematics Skill Panel

Estimations.

Standard errors in parenthesis; *,**,*** imply 10%, 5%, and 1% significance levels respectively. Lagged dependent variable

indicates AR1.

Variables Computers

Internet

User Telephone

Cell

phones

Visitor

beds Harvester Fertilizer

Lagged

dependent

variable

1.0137***

(0.015)

0.945***

(0.03)

1.0002***

(0.025)

1.001***

(0.018)

0.7342***

(0.054)

0.8828***

(0.02)

0.834***

(0.03)

Cognitive

Skills

0.1891***

(0.057)

0.449*

(0.25)

-0.0666

(0.04)

-0.158*

(0.08)

0.011***

(0.003)

-0.0009**

(0.0004)

-0.040***

(0.01)

Years of

Schooling

7.331**

(3.57)

7.018

(10.009)

0.514

(2.63)

11.365***

(6.35)

-0.584

(0.211)

0.032

(0.037)

-3.24***

(1.12)

Life

Expectancy

3.0355

(2.20)

24.95***

(6.44)

1.926

(1.23)

21.265***

(4.27)

-0.794

(0.13)

-0.0084

(0.019)

3.099***

(0.64)

FDI 0.2325

(0.4)

0.591

(1.02)

2.753

(0.70)

2.473***

(0.93)

-0.687

(0.03)

-0.011

(0.007)

-0.133

(0.21)

Observations 178 150 190 258 157 287 293

50

Table 3.2 Usage Intensity of Technologies, Science Skill Panel Estimations.

Variables

Aviation

pkm air

Shipton

Steam

motor/ sea

Transplant

Liver

Transplant

Lung

Transplant

Bone

marrow

Cable TV

Mail

Lagged

dependent

variable

1.0220***

(0.03)

0.81053***

(0.06)

0.6794***

(0.08)

0.40465***

(0.104)

0.7417***

(0.062)

0.89391***

(0.02)

0.9496***

(0.31)

Cognitive

Skills

-0.000028

(0.0001)

0.00003***

(0.000005)

-0.000006***

(0.000002)

-0.000005***

(0.000001)

0.000017**

(0.000007)

0.0184*

(0.01)

-0.000046**

(0.00001)

Years of

Schooling

-0.0681**

(0.03)

0.0037***

(0.001)

0.00031

(0.0004)

-0.00124***

(0.0003)

-0.0021

(0.001)

-0.6624

(2.23)

-0.00052

(0.004)

Life

Expectancy

0.28313*

(0.016)

-0.00154***

(0.0005)

0.000401

(0.0002)

0.00023

(0.0001)

0.00115

(0.0007)

0.39457

(1.38)

0.00581***

(0.001)

FDI

0.00299

(0.109)

-0.00074***

(0.0002)

0.000032

(0.00003)

-0.0000024

(0.00002)

-0.00007

(0.0001)

-0.9187***

(0.26)

0.00237

(0.001)

Observations 162 88 90 72 109 253 153


indicates AR1.

Table 3.2 (continued): Usage Intensity of Technologies, Science Skill Panel Estimations.

Variables Computers

Internet user

Telephone

Cell phones

Visitor

rooms

Harvester

Fertilizers

Lagged

dependent

variable

1.0159***

(0.13)

0.92361***

(0.028)

0.9379***

(0.02)

1.0266***

(0.015)

0.85242***

(0.033)

0.83778***

(0.02)

0.08068***

(0.027)

Cognitive

Skills

0.01487

(0.01)

0.07052

(0.061)

-0.00452

(0.01)

0.00006

(0.027)

0.00185**

(0.007)

-0.00049**

(0.0001)

-0.00746

(0.005)

Years of

Schooling

6.4531**

(3.14)

2.3008

(8.45)

-1.026

(2.63)

19.439***

(5.77)

0.11308

(0.13)

0.0342

(0.366)

-2.5538***

(0.98)

Life

Expectancy

5.0570**

(1.97)

33.118***

(6.36)

2.414

(1.634)

17.655***

(3.67)

-0.082805

(0.077)

-0.00816

(0.016)

2.2730

(0.46)

FDI 0.35111

(0.37)

0.3104

(0.94)

2.3892***

(0.58)

1.9098**

(0.81)

-0.01549

(0.02)

-0.000841

(0.007)

-0.219

(0.18)

Observations 215 177 162 304 269 288 305


indicates AR1.

51

In contrast to learning-by-doing, the qualitative dimensions of human capital exhibit a

relatively weak association with usage intensity of technologies. For instance, a review of the

results in Table 3.1 reveals that the cognitive skills based on mathematics test scores – which we

interpret as “generic” in nature have a positive and significant association with 8 out of 14

technologies such as aviation pkm/air, transplant liver, transplant lung, cable TV, mail,

computers, internet users and visitor beds. Furthermore, in Table 3.2 the evidence based on

science scores – which we interpret as reflective of specific skills (i.e. knowledge of science) – is

weaker. The coefficient of the qualitative measure of human capital is now positive and

significant in only 4 out of 14 regressions for technologies such as shipton steam motor/sea,

transplant bone marrow, cable TV and visitor rooms.24

Overall, our interpretation is that a

workforce equipped with generic in contrast to specific skills may serve as a more appropriate

channel to enhance technological diffusion. However, the results also suggest that the evidence is

weak relative to previous literature examining the contribution of human capital in facilitating

technology adoption. Given the significance of the lagged dependent variables in all regressions,

we again suggest that more important drivers of technology adoption and diffusion are to be

found in other, less measurable dimensions of human capital, such as those developed via

learning-by-doing.

Furthermore, our results reveal that embodiment of a certain skill is not positively associated

with adoption of all technologies within a sector. For example, in Table 3.1, the first four

24

In the sector-wise analysis presented in Appendix B, we have 21 technologies each in the mathematics and

science panels. The coefficient of mathematics and science skills is significant in 10 and 5 intensity of usage of

technologies respectively. Hence, we may suggest that overall mathematics skills are more suitable for improving

the adoption of technology. A sector-wise or technology-specific study can also be undertaken based on specific set

of determinants of a technology or sector under discussion. For example, for medical technologies data on number

of medical graduates per capita or in case of aviation number of pilots per capita can be used as variables

representing human capital. However, it is difficult to obtain a comprehensive cross-country data set of this sort

which restricts in developing a sector or technology-specific analysis.

52

columns include results for technologies from the telecommunications and information sector.

As can be seen that the association of mathematics (i.e. generic) skills is positive and significant

for computers and internet, it is certainly not the case for telephone and cell phone of

technologies. This implies that a certain skill relevant for particular technology within a sector

may not be appropriate to assist in the adoption of another technology from the same sector. A

possible interpretation is that the link between a particular type of human capital and technology

is a conditional one which rests on various aspects of human capital as well as the nature of the

technology in question.

Some remarks are in order in relation to the counter-intuitive results we find in the context of

a few technologies. Interestingly results for mathematics (i.e. generic) and science (i.e. specific)

cognitive skills are significant in majority of technologies in agriculture but the association is

negative. This is visible in the last two columns of tables 3.1 and 3.2, where the coefficient for

these skills is negative and significant for usage intensities of harvester and fertilizer

technologies. While these results are hard to interpret, they still connect with earlier empirical

evidence by Foster and Rosenzweig (1995) who suggest that agricultural technologies are

associated to a greater degree with “learning by doing” and time spent acquiring formal

knowledge of certain subjects or disciplines represents an opportunity cost. However, the results

certainly do not rule out the significance of specific knowledge; rather the qualitative measures

in our regressions do not adequately address the specificity of knowledge required in agriculture.

Our empirical analysis also includes a quantitative dimension of human capital measured as

average years of schooling, and the evidence for the human capital and technology diffusion link

is the weakest in this case. For instance, in Table 3.1 for the mathematics scores panel the

variable is positively and significantly associated with only 2 technologies. Table 3.2 which

53

presents science panel estimations, we find the association between average years of schooling

positive and significant for only 3 technologies. In addition we also include another measure of

human capital as life expectancy in our analysis. Our results show that life expectancy is

positively and significantly associated with aviation pkm/ air, lung transplant, internet, cell

phone and fertilizer usage intensity of technologies in generic panel results. In the specific panel

estimations the coefficient for life expectancy is positive and significant for aviation pkm/ air,

computer, internet user and cell phone usage intensity of technologies.

Based on the above one may conclude that there is a hierarchy in the effectiveness of

different types of human capital. In the regressions above human capital associated with

learning-by-doing seems to be the most important contributor to technology adoption followed

by other types of human capital reflected in qualitative and quantitative measures. Of the latter

measures there is some evidence to suggest that human capital of the generic type as reflected in

mathematics test scores is important in the context of technology adoption.

A caveat applies to this discussion; some of the cases presented in tables 3.1 and 3.2 are

worthy of discussion especially in relation to alternate measures of human capital such as

average years of schooling and life expectancy. In technologies such as cell phones it seems that

average years of schooling and life expectancy are more relevant as they have a relatively larger

impact than qualitative measures of human capital. We find that an increase in one year of

schooling is associated with 11.36% increase in the usage intensity of cell phone technologies. In

addition, a similar increase in life expectancy leads to 21.26% increase in adoption of cell

phones. On the other hand, we find a negative and significant association between cognitive

skills and these technologies.

54

We also control for other possible determinants of technology adoption and introduce foreign

direct investment in both sets of analyses presented in tables 3.1 and 3.2. We find foreign direct

investment shows a positive and significant association with cell phones, mail and telephone

usage intensity of technologies. Overall, this control variable shows inconclusive evidence as

indicated by the signs of the coefficients. This probably suggests that macroeconomic, aggregate

level variables may have relatively lower explanatory power in the context of specific

technologies and we would need sectoral, microeconomic counterparts of these variables to get a

more accurate idea of their relevance.

Furthermore, some caveats apply to variables selected for the analysis. For example, generic

skills may not be important for adoption of medical technologies such as bone marrow transplant

in comparison to measures such as per capita number of medical graduates or surgeons. In

addition, number of pilots per capita may be a more relevant determinant of adoption of aviation

technologies in comparison to specific human capital reflected in science scores. This implies

that for a particular technology a specific knowledge variable and set of determinants is required

for a technology-specific discussion. However, apart from issues relating to availability of data

the aim here is to find common determinants of technology in addition to specific ones. With

regard to the latter, it is difficult to find comparable and consistently measured variables for all

of the countries in the sample. For example, finding per capita measures of the number of

medical graduates or surgeons is difficult because such kind of data is mostly available from

country specific sources rather than international databases.25

In regard to the former – i.e.

common determinants - the contribution of the current analysis is that learning-by-doing and

generic skills matter relatively more compared other dimensions of human capital.

25

For example, World Development Indicators (WDI) do not have such indicators for health. However, some

country specific studies do provide this information from their respective national databases (Ceppa et al, 2012).

55

3.4 Empirical Evidence on Measures of Human Capital and Technology Usage Lags

We estimate equation (3.8) to examine the contribution of human capital to technology usage

lags.26

We present results in tables 3.3-3.4.27

Table 3.3 includes mathematics (generic) panel and

Table 3.4 presents science (specific) panel results with selected technologies.

Similar to our evidence for usage intensity of technologies we also find a strong association

between the past levels of usage lags with current usage lags of technologies. As can be seen

from tables 3.3 and 3.4 lagged dependent variable is positively and significantly associated with

usage lags of technologies in almost all regressions. A similar interpretation, pertaining to the

learning-by-doing dimension of human capital is applicable here. Specifically the pace of

technology adoption is small (representing quicker adoption) if past levels of the usage lag is

small. If the gap in usage is small “learning by doing” has occurred to a greater degree.

26

In total the sample consists of 18 technologies and results of all technologies are presented in Appendix D. In

some cases U.S was not the technology leader and in other cases there were not enough observations for lags to

perform regressions. Therefore, we were not able to calculate the lags for all the technologies included in the usage

intensity of technology sample. Specifically, we consider usage lags for technologies in tourism,

telecommunications and information, health, electricity production and agriculture. 27

Appendix C contains descriptive evidence regarding data used for analysis. Appendix D contains tables which

provide a complete sector-wise overview of these results including a larger set of technologies. The more succinct

presentation of these results in the form of tables in the main text does not affect the overall findings and

interpretation of the analysis.

56

Table 3.3 Usage Lags of Technologies, Mathematics Skill Panel Estimations.

Variables Computers Internet

User Telephone Mail Cable TV

Cellphones

Transplant

Lung

Lagged

dependent

variable

0.88***

(0.470)

0.551***

(0.106)

0.736***

(0.037)

0.804***

(0.051)

0.581***

(0/075)

0.7633***

(0.057)

0.7715***

(0.075)

Cognitive

Skills

-0.015***

(0.004)

-0.184*

(0.0101)

-0.003

(0.005)

-0.0601

(0.013)

-0.0106

(0.006)

0.0075**

(0.057)

-0.099***

(0.056)

Years of

Schooling

-0.290

(0.226)

-0.042

(0.27)

0.0563

(0.367)

0.109

(1.08)

0.713

(0.439)

-0.709***

(0.202)

12.72***

(3.54)

Life

Expectancy

0.183

(0.120)

-0.0012

(0.163)

0.070

(0.201)

0.098

(0.523)

0.347

(0.260)

-0.238*

(0.134)

-0.0286

(1.11)

FDI 0.018

(0.028)

0.0016

(0.032)

-0.400***

(0.106)

-1.068**

(0.043)

0.047

(0.051)

-0.044

(0.046)

0.009

(0.151)

GDP/income

lag

-0.0073

(0.027)

0.003

(0.047)

-0.117*

(-0.117)

0.135

(0.124)

0.0095

(0.051)

-0.0138

(0.028)

0.253

(0.033)


indicates AR1.

Table 3.3 (continued) Usage Lags of Technologies, Mathematics Skill Panel Estimations.

Variables

Transplant

Heart

Transplant

Kidney

Transplant

Liver

Visitor

rooms

Visitor

beds

Tractor

Fertilizers

Lagged

dependent

variable

0.856***

(0.057)

0.6304***

(0.06)

0.121

(0.12)

0.835***

(0.046)

0.535***

(0.118)

0.961***

(0.018)

0.609***

(0.062)

Cognitive

Skills 0.007

(0.004)

0.0011

(0.007)

-0.0103

(0.01)

-0.0063

(0.005)

0.0086

(0.008)

0.0017

(0.001)

0.020***

(0.006)

Years of

Schooling 0.310

(0.32)

0.261

(0.565)

1.278

(0.997)

-0.655

(0.404)

1.073**

(0.484)

0.024

(0.068)

1.407***

(0.505)

Life

Expectancy 0.418**

(0.203)

1.249***

(0.399)

2.613***

(0.555)

0.594

(0.209)

0.549*

(0.319)

0.0415

(0.054)

0.387

(0.257)

FDI

0.016

(0.02)

0.183**

(0.087)

-0.0101

(0.07)

0.051

(0.055)

0.004

(0.06)

0.015

(0.012)

0.012

(0.073)

GDP/income

lag

-0.0072

(0.024)

-0.032

(0.071)

0.001

(0.067)

0.124**

(0.053)

-0.0033

(0.094)

-0.0105

(0.013)

0.012

(0.06)

Observations 58 150 60 182 100 214 183


indicates AR1.

57

Table 3.4 Usage Lags of Technologies, Science Skill Panel Estimations.

Variables Computers Internet User Mail Cable TV

Cell

phones

Electricity

Production

Transplant

Heart

Lagged

dependent

variable

0.866***

(0.044)

0.385***

(0.11)

0.568***

(0.067)

0.603***

(0.058)

0.849***

(0.042)

0.792***

(0.043)

0.856***

(0.057)

Cognitive

Skills

-0.002**

(0.001)

0.0016

(0.002)

-0.011*

(0.006)

-0.002

(0.001)

0.0013

(0.001)

-0.0015

(0.001)

0.007

(0.004)

Years of

Schooling 0.047

(0.204)

-0.434

(0.002)

-0.144

(1.42)

0.792**

(0.365)

-0.830

(0.180)

0.175

(0.278)

0.310

(0.32)

Life

Expectancy

0.029

(0.109)

-0.063

(0.165)

-0.128

(0.653)

0.347

(0.212)

-0.154

(0.109)

0.275*

(0.163)

0.418**

(0.203)

FDI -0.006

(0.027)

0.020

(0.034)

0.083

(0.421)

0.051

(0.04)

-0.192

(0.033)

0.039

(0.072)

0.016

(0.02)

GDP/income

lag -0.0108

(0.021)

-0.003

(0.33)

0.022

(0.87)

0.005

(0.03)

0.0132

(0.018)

0.105*

(0.045)

-0.0072

(0.024)


indicates AR1.

Table 3.4 (continued) Usage Lags of Technologies, Science Skill Panel Estimations.

Variables

Transplant

Bone

marrow

Transplant

Kidney

Transplant

Lung

Visitor

rooms

Visitor

beds

Tractor

Fertilizers

Lagged

dependent

variable

0.693***

(0.087)

0.296***

(0.079)

0.208

(0.163)

0.790***

(0.047)

0.624***

(0.093)

0.979***

(0.018)

0.790***

(0.039)

Cognitive

Skills

-0.004*

(0.002)

0.012***

(0.004)

-0.0019

(0.008)

-0.003*

(0.002)

0.002

(0.003)

0.00058

(0.0005)

0.0013

(0.001)

Years of

Schooling

0.217

(0.509)

-1.408*

(0.835)

5.668

(2.106)

-0.505

(0.395)

0.588

(0.430)

0.124

(0.080)

0.747***

(0.266)

Life

Expectancy

0.847***

(0.26)

1.951***

(0.423)

0.117

(0.991)

0.800***

(0.243)

0.635**

(0.279)

-0.0388

(0.064)

0.426***

(0.134)

FDI 0.0417

(0.041)

0.270**

(0.138)

0.071

(0.130)

0.071

(0.061)

0.015

(0.058)

-0.008

(0.014)

0.028

(0.036)

GDP/income

lag

-0.0286

(0.034)

0.020

(0.104)

-0.032

(0.109)

0.050

(0.046)

-0.031

(0.085)

-0.001

(0.012)

0.026

(0.02)


indicates AR1.

58

Our empirical evidence indicates a weaker link for both generic and specific skills with

usage lags of technologies when compared to learning-by-doing aspect of human capital. In

terms of generic skills based on mathematics test scores the results presented in Table 3.3 show

the hypothesized negative and significant association in only 3 technologies such as computers,

internet, and transplant lung. In Table 3.4 the estimations for specific skills measured as science

test scores also provide a similar picture a negative and significant association with only 4 usage

lags of technologies such as computers, mail, transplant bone marrow and visitor rooms. This

inverse association, when present, indicates that the presence of a workforce with generic and

specific skills tends to reduce usage lags, thereby improving diffusion of technologies.

In contrast to previous results on usage intensity of technologies the hierarchy in the

degree of importance of various types of human capital is not the same here. Both generic and

specific skills seem to be of equal and limited importance compared to learning-by-doing which

is negatively and significantly associated with almost all usage lags of technologies. However, in

terms of average years of schooling and life expectancy as other dimensions of human capital

our results are similar to the evidence presented for usage intensity of technologies. Overall, we

find learning-by-doing to be the most appropriate determinant of technology adoption followed

by both qualitative measures of human capital, average years of schooling and life expectancy

respectively.

The association between skills and usage lags is weaker when compared to results for

usage intensity of technologies in the previous section. Referring to the introduction, one way to

interpret this is perhaps that the timing of adoption and intensity of use are determined

differently. For example, we find that generic skills are positively and significantly associated

with improving the per capita usage of cable TV technology. However, generic skills are

59

insignificant in reducing the usage lag of cable TV technology. As mentioned before one

plausible argument here is that in case of usage lags of technologies it is not just the presence of

skills among the potential adopters that matters for an economy. There may be other factors

which inhibit diffusion of a technology such as governmental and political motives, industrial

policy dynamics and demographic or cultural factors. These factors are better explored in single

country and single technology studies aimed at unearthing country-specific issues pertaining to

technology adoption.

One of the most important points to be made here is that even though our empirical evidence

for human capital is weaker for usage lags of technologies, it still lends support to our hypothesis

that the human capital and technology is conditional one which rests on various aspects of

human capital and the nature of technology under question. In common with results for usage

intensity of technology our evidence for usage lags of technology also reinforces that the

association between human capital and technology diffusion varies within and across sectors. A

glance at the first 5 technologies in Table 3.3 highlights this variation of association between

generic skills and technologies in telecommunications and information sector. In the first and

second columns we find a negative and significant association between generic skills and

computer and internet usage lags. However, in the third and fourth column the impact is

insignificant for telephone and mail usage lag of technologies, while in the fifth column for cell

phone usage lags of technology the impact is positive and significant. This shows that

technologies within a sector have a different link with the same measure of human capital. This

suggests that the association between generic skills and technologies within a particular sector

need not necessarily imply a similar association with other technologies in that particular sector.

60

In the case of the agricultural sector our results indicate an absence of skill-technology link

for generic and specific skills. We find this evidence similar to our results for usage intensity of

technologies. This weak association could be perhaps due to the reason that these technologies

have greater degree of association with informal channels of diffusion such as learning from

social networks and may not require a proper understanding of the subjects under discussion

(Conley and Udry, 2001).

In contrast to our usage intensity analysis here we also include income or GDP lags in the

usage lag of technology estimations. We follow Comin et al (2008) who employ these lags to

capture the dynamics of technology. They argue that if a country is progressing well in terms of

reducing its income lags then it should have shorter technology lags as well. Hence, higher

economic growth should improve the diffusion prospects of technology in an economy. Our

results indicate that GDP lags are positive and significant in only 3 estimations across these two

panels. This weak significance presented in our results reinforces the findings of Comin et al

(2008) and indicates that technology lags measuring the past level of technology are relatively

more important that the income lag of a country. Therefore, the dynamics of technology itself

play a more vital role in the process of technological diffusion rather than the dynamics of an

economy measured as income usage lags.

To summarize, we find indirect evidence that learning-by-doing is perhaps the most

appropriate dimension of human capital which is highly relevant for both adoption and diffusion

of technologies. Furthermore, based on our evidence for usage intensity and usage lags of

technologies we find qualitative measures of human capital relatively better facilitators of

technology compared to quantitative constructs such as average years of schooling. In general

61

presence of human capital embodying generic compared to specific skills perhaps is more

conducive to improving the technological prospects of an economy.

Our results also highlight that conclusions about the human capital and technology adoption

link based on a single technology (as is the case with microeconomic studies) or aggregate

measures of technology (such as total factor productivity) can be misleading. Studies looking at

aggregate measures, for example, may find a positive impact of human capital leading to a “one-

size-fits-all” recommendations for investment in a certain type of human capital. Likewise

evidence based on a single technology yields information of relevance to only that particular

technology. By following a comprehensive approach that looks at different measures of human

capital and a large set of technologies, we have taken a more cautious approach, leading to the

insights that learning-by-doing or technology-specific education may be a better facilitator of

technology adoption and diffusion, and that qualitative measures reflective of generic skills are

of greater relevance in an overall sense in comparison with specific skills and quantitative

measures of human capital.

In the light of these results, studies using qualitative measures of human capital in growth

regressions may also be interpreted differently. As mentioned in Chapter 2, the evidence in

favour of such measures positively impacting on growth is relatively robust. Given the relatively

weak results here, it may be the case that mechanisms other than technology adoption are more

relevant when considering the impact of human capital on economic growth.

3.5 Additional Robustness Checks

In what follows we carry out three additional robustness checks for separate panels based

on generic and specific cognitive skills for adoption and diffusion of technology respectively.

62

Firstly, we control for the quality of institutions using measures of political rights and civil

liberties. Secondly, we use GDP per capita to examine the influence of economic growth on both

adoption and diffusion of technology. Lastly, we use expenditure on research and development

(R&D) as percentage of GDP and evaluate its impact on technology adoption and diffusion. The

robustness results are reported in Appendix E. Tables 1-2 of this appendix summarize results for

adoption of technologies for generic and specific skills respectively. In addition, tables 3-4

contain results for diffusion of technologies for both generic and specific skills.28

Each panel in

these tables represents a specific technology. The first column in each panel reports results

regarding the impact of institutional quality on adoption and diffusion of selected technologies,

while second and third columns describe how GDP per capita and R&D expenditures affect

adoption and diffusion of technology respectively.

There has been recent emphasis on the role of institutions in the process of economic

growth as poor quality institutions adversely affect economic performance of a country

(Acemoglu et al 2005). On the other hand, good quality institutions ensure efficient allocation of

resources, protect and safeguard political rights and civil liberties, reduce uncertainties, enable

investment in high return projects and facilitate coordination among economic agents (North,

1990; Aghion et al, 2008; Meyer and Sinani, 2009; Rodrik et al, 2004; Glaeser et al, 2004;

Flachaire et al, 2014; Jude and Levieuge, 2015). Based on these findings we are interested in

exploring the role of institutions from the perspective of technology adoption and diffusion.

Adoption and diffusion responses of majority of technologies are similar as in baseline

regressions with the inclusion of institutional quality in both generic and specific panel

estimations. In relation to the role of institutions, we obtain variable coefficients for the proxies

28

We perform robustness checks for all technologies in our sample and report results of a few selected set of

technologies in Appendix E

63

of institutional quality.29

More specifically, civil liberties used as a proxy for institutions

positively and significantly influence adoption of cable TV, vehicle car and agriculture

technologies in both generic and specific estimations. These results show that our measure of

civil liberty capturing several dimensions of equality, freedom, legality and fairness in society

facilitates adoption of these technologies in the sample.

Furthermore, improvements in technology through investment in human capital lead to

economic growth. Developed economies experience higher growth because they are

technologically more advanced than developing economies (Romer, 1990; Aghion and Howitt,

1992). Given these findings we examine whether economic performance of an economy

influences the process of technology adoption. We introduce GDP per capita as a measure of

economic performance in our empirical analysis as another determinant of adoption and

diffusion of technology. The results for robustness checks for both generic and specific human

capital reinforce earlier findings of baseline regressions. Our results for adoption of technologies

indicate that liver and lung transplant procedures, vehicle and cable TV usage intensity of

technologies respond positively to GDP per capita in both mathematics and science panels. In

addition, we see that increase in GDP leads to a significant reduction in usage lags of fertilizer

and visitor bed technologies.

Lastly, literature on technology suggests that expenditure on research and development is

linked with technological innovations (Acemoglu and Zillibotti, 2001). We therefore, examine

whether expenditure on R&D impacts upon technology adoption and introduce expenditure on

29

A possible reason might be that the measures of institutional quality used in this study are perhaps unable to

capture the soundness of institutions appropriately as there is lack availability of authentic data on institutions

beginning from early 60’s, as the data on institutional quality from World Bank starts from mid-1990s. The current

study employs Freedom House data set on political rights and civil liberties as a proxy for institutional quality which

begins in the early 1970s. See for details; Freedom House official website for access to data and Freedom in the

World Report 2016.

64

R&D as the third determinant for robustness checks. Our results show that coefficient for skills

and lagged dependent variables remain similar to baseline regressions. Moreover, expenditures

on R&D are significant in case of usage intensity of liver transplant and vehicle usage in

mathematics scores panel. Based on the above empirical evidence, we see that the basic result

suggesting that adoption and diffusion of technologies respond positively to disaggregate

measures of skills remains robust even after controlling for other determinants of technological

adoption and diffusion.

3.6 Concluding Remarks

This study analyzes the link between human capital and technology in the light of direct

measures of technology adoption and diffusion and educational quality. Earlier literature in the

field of human capital and economic growth uses average measures of educational quality and

quantity (Barro, 1997; Hanushek and Woessmann, 2012). However, it focuses more on the link

between human capital and economic outcomes and ignores the channels through which human

capital affects growth of an economy. We hypothesize that one of the channels through which

human capital may impact economic growth is its role in improving adoption and diffusion of

technologies. This study bridges this gap by investigating the missing link between human

capital and technology adoption and diffusion using direct measures of technology and

educational quality. We contribute in the literature by examining this relationship of how

disaggregated measures of educational quality facilitate technology adoption and diffusion

through improvement in human capital. Moreover, we also differentiate between different forms

of human capital and examine their relative impact on the adoption and diffusion of

technologies.

65

In testing the hypothesis whether educational quality enhances technology adoption and

diffusion, we use cognitive skills data for international mathematics and science test scores along

with data on direct measures of technology adoption. We use Hanushek and Woessman (2012),

measure of educational quality and further decompose average cognitive skills into mathematics

and science skills and construct separate panels for both the set of skills from 1964-2003 and

1973-2003 respectively. Moreover, we use CHAT data set developed by Comin and Hobijn

(2009) to obtain direct measures of technology. In order to empirically analyze our hypothesis of

learning-by-doing dimension of technology, we follow the econometric approach by Comin et al

(2008) based on dynamic panel specification and incorporate the lagged effect of technology.

Based on empirical analysis our main finding reveals that the link between human capital

and technological adoption and diffusion is a conditional one, which rests on various aspects of

human capital and technology under consideration. Moreover, this skill-technology association

indicates that appropriateness of skills required for adoption and diffusion of technologies

changes within and across sectors. In summary for technology adoption, technologies from

transportation, tourism and health sectors positively respond to both disaggregated measures of

cognitive skills. However, telecommunication and information based technologies are more

influenced by generic in contrast to specific skills. On the other end, for usage lags as a measure

of technology diffusion, mathematics based generic skills assist diffusion of certain technologies

in telecommunications and information, electricity production and health sectors. Empirical

evidence for science indicates that specific skills reduce lags associated with technologies in

telecommunications and information, electricity production and health sectors. However, to our

surprise skill implications for both adoption and diffusion of technologies in agriculture are weak

as compared to other technologies in our sample.

66

Another noteworthy finding of this analysis is that the most important determinant for

technology adoption and diffusion is the past level of technology. This highlights the presence of

learning-by-doing aspect of technology across all sectors in our analysis. Our evidence shows

that qualitative measures of education are one of the channels facilitating adoption and diffusion

of technology. More specifically, generic human capital measured as mathematics test scores are

more relevant in comparison to specific science based skills. We also find that quantitative

measures of human capital such as average years of schooling to be of lesser relevance in

comparison with qualitative measures. Finally, the impact of cognitive skills remains robust

even after controlling for other determinants of technology adoption and diffusion which include

institutional quality, GDP per capita and R&D expenditures. Against this background, we

suggest that to develop more relevant policy insights, an appropriate approach inclusive of

different types of human capital can provide better understanding about barriers to technology

adoption and diffusion.

67

Chapter 4

Deconstructing Human Capital Inequalities: A new approach based on measures of

educational achievement

4.1 Introduction

One of the debates surrounding the relationship between growth and income inequality stems

from the famous Kuznets’ curve, which suggests that inequality rises at the beginning of the

industrialization process and shrinks in subsequent phases of growth in an economy (See

Kuznets, 1955 for the original articulation of this idea). Several studies extend this literature and

highlight inequality as an important factor affecting economic growth; this impact can be

negative, positive or ambiguous, depending on the framework in question (Aghion et al, 1999;

Galor, 2011; Ostry et al, 2014; Halter et al, 2014). Regardless of the unresolved nature of this

link in theoretical papers, as well as the mixed evidence found empirically (Cingano, 2014), the

literature on the measurement of inequality motivates it as an indicator of economic

performance. Taking this view, however, implicitly recognizes a negative link between

inequality and development. In recent years, in fact, the emphasis seems to have shifted in favor

of recognizing the adverse effects of inequality, in both academic and policy circles (See for

example’s Piketty, 2013, 2015). In addition, the inclusion of dimensions of inequality in the

United Nations Sustainable Development Goals (SDGs) such as reduction of inequality within

and across countries underscores the need to develop policies that focus on the needs of

marginalized and disadvantaged groups among the population of countries.

Given these developments, measurement of inequality and its many dimensions has become

even more important. The aim of this chapter is to focus on one of these dimensions, namely

human capital inequality. Furthermore, the approach taken here is microeconomic in flavor,

68

recognizing that the roots of human capital inequality can be traced to educational inequality at

the disaggregated level of educational institutions, such as primary or secondary schools. To that

end, this chapter constructs a measure of human capital inequality based on the qualitative

measure of human capital – test scores - considered in earlier chapters of this thesis, and then

decomposes this measure by using data at the disaggregated level of schools.

Of motivational relevance to this study is the literature that suggests the inadequacy of

income as a measure of human well-being. This literature argues that other aspects of human life

such as education, health, civil liberties and political freedom also play a role in improving the

quality of life (Sen 1979, 1985, 1987). Therefore, measuring inequality in a society solely on the

basis of income does not reflect inequality in other important dimensions of human life such as

education and health (Oppedisano and Turati, 2011).

In comparison to literature on income inequality and its link to growth, the literature on

inequalities in education and its economic implications is relatively less well-developed.

However, there are studies which provide a rationale for, or evidence that educational

inequalities may be one of the factors influencing growth, income distribution and productivity

differences (Saint- Paul and Verdier, 1993; Park, 1996; Galor and Tsiddon, 1997; Gregorio and

Lee, 2002; Checchi, 2004; Acemoglu and Dell, 2010; Castello and Domenech, 2002, 2014). The

underlying argument of this literature is that inequality influences the economy in several ways.

For instance, empirical evidence suggests that less educated segments have higher fertility and

lower life expectancy levels relative to more educated segments of the society. Both these

characteristics inhibit investment in education which leads to lower economic growth due to

reduced productivity of human capital (Castello and Climent, 2010). Furthermore, educational

inequality among individuals imposes constraints on their borrowing capacity relative to their

69

future incomes which influences the distribution of resources and investment patterns of an

economy (Perotti 1996). Hence, inequalities in human capital may be more relevant as education

of individuals is associated with their health status, investment behaviours, labour market

outcomes and political participation in the democratic process. It is therefore important to study

educational inequalities, which may have externalities leading to undesirable gaps in economic,

social and political dimensions.

One of the drawbacks of these studies is that they are restricted to comparisons of inequality

across time and employ macroeconomic data sets that use standardized or average measures of

educational attainment. Woessmann (2014) argues that qualitative measures, such as skills

acquired as a result of education are more important measures of human capital rather than years

of education; it is the increase in educational achievements that contribute to higher economic

growth in the long run. Following this idea we use Trends in Mathematics and Science Study

(TIMSS) 2008 test scores for advanced mathematics for final year secondary students in order to

measure human capital inequality.

Against this background, this study decomposes within and between sub-group inequalities

using Generalized Entropy Measures based on the TIMSS (2008) data at three levels, i.e.,

combined or cross-country, country and school level.30

It therefore focuses on decomposing

inequalities in educational achievement at a disaggregated level by employing a comparable

cross-country microeconomic data set for which such disaggregation is available.31

To the best of

our knowledge this is the first attempt to use the TIMSS 2008 raw pupils’ test scores to construct

within and between measures of dispersion for human capital. Such an exercise is supported by

30

We discuss these measures in section 4.2. For a more elaborate discussion see Shorrocks (1980); Cowell and

Kuga, (1981). 31

This is the only data set that provides information on raw tests scores disaggregated at this level

70

analogous explorations in the context of income inequality; Blundell and Etheridge (2010), for

example, find a strong correlation between aggregate measures of income inequality and

microeconomic dimensions of inequality such as those associated with earnings, gender and

consumption.32

Motivated by such strands of literature, we argue that our disaggregated

empirical analysis could yield additional and more relevant insights about inequalities in human

capital.

Other studies using standardized and average test scores have made an attempt to provide

international comparisons of educational inequality on the basis of achievement rather than

attainment. However, they do so at an aggregated level, using the generalized entropy index to

decompose within and between-country inequality in the TIMSS mathematics and science test

scores, as in the study by Sahn and Younger (2007). Their work indicates that within-country

inequality dominates between-country inequality. Freeman et al (2010) use fourth and eighth

grade TIMSS scores to show wide cross-country variation in the level and dispersion of test

scores with the highest scores associated with countries having the least inequality in scores.

Moreover, Oppedisano and Turati (2011) examine the evolution of inequality for 2000 and 2006

in nine European countries by focusing on reading test scores in the two waves of Programme for

International Student Assessment (PISA) study. They propose that parental and school

characteristics are important determinants of inequalities in educational achievements among

students.

This study further contributes to the literature on human capital in several ways. First, we

employ a unique microeconomic data set constituting raw test scores at a higher level of

32

They find this correlation in the context of UK labour market. Using micro-data for the year 1978 on income,

consumption and earnings, they develop a consistent analysis of these three variables to examine inequality and

build a logical and comprehensive link between microeconomic and macroeconomic evolution of inequality. They

indicate that the inequality boom of the 1980s was led by the increase associated with inequality in earnings.

71

education for a different set of countries to construct a cross country human capital index. Earlier

evidence on human capital inequalities indicates that the within-country component of inequality

overshadows between-country inequality.33

If our evidence supports the previous findings we

may argue that human capital inequality has a country-specific dimension. Hence, the

appropriate approach to address the issue of inequalities is to focus on the factors associated with

the educational system of individual countries rather than drawing inferences based on cross-

country analyses of such inequalities. To that end, the analysis that follows focuses on the former

approach by undertaking an inequality decomposition exercise, and exploring the determinants

of inequality at a more disaggregated level.

To elaborate, in the context of examining the underlying reasons for within-country

inequalities, we decompose human capital inequality and consider student sub-groups from each

of the schools that participated in that test and develop country-specific analyses. Such country-

wise analysis is aimed at revealing the composition and structure of human capital at a

microeconomic level by exploring its determinants at the school level. For this purpose we

employ a standard regression analysis with school and teacher related attributes as potential

factors influencing inequality in skills. We anticipate that the results of a country-wise empirical

analysis will reinforce our hypothesis that factors influencing inequality in human capital are

specific to a country and can be traced down to an institutional level. Using this approach we

anticipate identifying country and perhaps school-specific factors that lie at the core of human

capital inequalities.

Further, studies suggest that constructing inequality indices using standardized tests scores is

an approach subject to flaws that have been overlooked in the earlier literature on educational

33

Sahn and Younger (2007), use eighth grade test scores on mathematics and science TIMSS 1999 and 2003. We

employ TIMSS, 2008 advanced mathematics raw test scores for final year secondary school students.

72

inequality (Ferreira and Gignoux, 2013).34

Therefore, we consider decomposing inequalities

using pupils’ raw tests scores in mathematics. Finally, it is important to note that we focus on a

higher grade of education, by examining the achievement of a cohort of final year secondary

school students who are on the verge of completion of studies and planning to continue for

university education or enter the job market. We believe information on human capital

inequalities at this level of education is better reflective of disparities that have a more immediate

impact on overall inequalities, as well as an economy’s potential for growth, given that skills

acquired at this stage are more salient for the human capital of the workforce.

Our focus on within and between measures of dispersion for human capital employing

disaggregated micro data on mathematics test scores yields the following findings. Firstly, our

results reinforce earlier evidence on educational achievement inequalities by Sahn and Younger

(2007) and reveal that within-country disparities overshadow between-country educational

quality dispersion. In our analysis of 10 countries, the value of the human capital inequality

index indicates on average a within-country component of 70% and between-country component

of 29.35%. This evidence confirms the result that inequality in human capital is country-specific.

In addition, we obtain a slightly greater value for within-country component as compared to

findings obtained by Sahn and Younger (2007). This could be due to the fact that we employ a

higher grade of test scores (12th

grade as opposed to 8th

grade) to construct our index. This may

imply that the composition of within and between components of dispersion in mathematics test

34

The study indicates use of standardized test scores based on raw test scores affect results of statistical analysis

based on measures of central tendency and dispersion in particularly for less developed countries. Many other

studies also provide similar evidence (Micklewright and Schnepf, 2006; Brown et al, 2007; Ferreira and Gignoux,

2013). Furthermore, standardization is suitable in cases where a cross-country analysis is undertaken, such as in the

panel estimations of Chapter 3. In this study we develop a country by country analysis by examining one country at

a time. Therefore, raw test scores which reflect upon the actual scores achieved by the students for a country under

discussion are more appropriate.

73

scores changes, and perhaps deepens at higher levels of educational achievement.35

However, in

either case, the intrinsic attribute of human capital inequality associated with educational

achievement indicates the dominance of the within over the between component of inequality.

Our results are in contrast to earlier work, based on income and educational attainment

inequalities, which show that between-component is greater than the within-inequality

component (Li et al, 1998; Castello and Domench, 2002).36

More importantly, we further decompose within-county inequalities to within and between

school inequalities. This empirical exercise provides information to better understand and

explain the grassroots composition and factors associated with inequality for each country in our

sample. Interestingly, we find that within-school inequality is greater than between-school

inequality in all countries. We believe this to be a very striking result which supports our

intuition that microeconomic factors are the fundamental causes of human capital inequality. In

what follows we therefore focus on the exploration of within-school inequalities at a country-

specific level. Our case-by-case country-wise regressions employing decomposed school level

inequalities reinforce the use of school and teacher attributes in examining human capital

inequalities as suggested in Oppedisano and Turati (2011).37

Our empirical evidence reveals that these attributes are among important determinants of

inequality in human capital at a disaggregated level; however, the specific attributes differ across

countries. We argue that as the factors that determine the level of education in each country are

different, which also implies that factors contributing to the within-school inequalities may be

different. Hence, based on a specification analysis the model that best fits a given country may

35

We cannot assert this as both the samples for grade 12 and 8 consist of test scores of a different set of students. 36

Castello and Domench (2000) employ educational attainment data which consists of average years of schooling as

constructs of human capital to inequality. 37

They perform country-level analysis for selected European Union countries by using the standardized test scores

of students on reading, maths and science based on Program for International Student Assessment 2000 and 2006.

74

include different set of attributes, confirming the country-specific nature of inequality posited

earlier. For example, we find that the set of factors impacting upon school inequality in Iran are

entirely different from the determinants of school inequality in the model for Sweden. In other

cases the country may have a few similar variables but the association may vary. For instance,

student-teacher ratio is positively and significantly associated with school inequalities in

Lebanon. On the contrary a higher student-teacher ratio leads to a significant decline in school

inequalities for Slovenia.

Furthermore, we estimate a cross-country version of a common set of variables for the

purpose of a robustness analysis. Our cross-country analysis at an aggregated level lends support

to our findings presented in country-specific analysis. However, such an aggregated level

analysis conceals the differences in the determinants of inequality among different countries.

Therefore, we suggest that a microeconomic approach to examine inequalities better reveals the

differences present in the composition and determinants of dispersion in human capital in

comparison to cross-country macro-data analysis.

The structure of the essay is as follows, Section 4.2 outlines the data and empirical

framework relevant to the current study. Section 4.3 analyzes the indices decomposing within

and between human capital inequalities, and uses case-by-case country regression analyses to

determine the factors influencing these inequalities. Section 4.4 contains the cross-country

analysis. Finally, Section 4.5 presents the main conclusions.

4.2 Methodology

4.2.1 General Entropy Measures of Inequality

This section describes the framework employed to obtain the measures of skill

inequalities in human capital for our sample of countries using advanced mathematics raw test

75

scores. We choose Generalized Entropy Measures (GE) of inequality as these indices are

attributed to have solid axiomatic foundations (Cowell and Kuga, 1981 and Foster 1983). In

particular they are decomposable into within-group and between-group inequalities (Shorrocks,

1980; Cowell and Kuga, 1981).We follow Shorrocks and Wan (2005), who explain the construct

of generalized entropy measures based on the concept of income as a measure of inequality.38

Our analysis uses test scores as educational achievements in place of income.

Let N= {1, 2,…, n} be a population of students, with test scores given by the vector v=

{v1, v2, …,vn}. The mean of these test scores is denoted by µ. Inequality in test scores can be

captured by an inequality index which follows the standard properties of measures of relative

inequality.39

We can write Generalized Entropy Index to calculate skill inequalities as:

𝐺𝐸(𝛼) = (1

𝛼(𝛼−1)) (

1

𝑛∑ (

𝑣𝑖

𝜇)

𝛼𝑛𝑖=0 − 1) (4.1)

In expression (4.1), 𝑣𝑖 is student i’s test score, µ is the mean test score and α is a

parameter that can take any real value. This expression defines a particular class of generalized

entropy index as 𝐺𝐸(𝛼) which is the index can assume different forms depending upon the value

assigned to 𝛼. Morespecifically, if assigned a positive and large αthe index becomes more

sensitive to what happens in the upper tail of the distribution. On the other hand a positive and

small α makes the index more sensitive to what happens at the bottom tail of the distribution.

The most commonly employed and indices are α= 0 referred to as mean log deviation, α=1 as

Theil index and α = 2 coefficient of variation.

Any GE index can be easily decomposed into within and between group components

given j exhaustive and mutually exclusive groups. Then the expression (4.1) can be written as:

38

We only present main equations of the theoretical framework. For more details, See Shorrocks and Wan (2005), 39

For more details, See Deutsch and Silber (1999).

76

𝐺𝐸(𝛼) = (1

𝛼(𝛼−1)) (∑

𝑛𝑗

𝑛

𝑗𝑗=1 (

𝜇𝑗

𝜇)

𝛼

(1

𝑛𝑗∑ (

𝑣𝑖

𝜇)

𝛼

− 1𝑛𝑗

𝑖=1) + (∑

𝑛𝑗

𝑛

𝑗𝑗=1 (

𝜇𝑗

𝜇)

𝛼

− 1)) (4.2)

In this expression the generalized entropy index of the entire population of students is the

weighted average of each group’s GE index, termed as within-group index and the between

component is based on each group’s mean test score value.

4.2.2 Data and Data sources

To construct skill-inequality indices using GE measures we use final year secondary

school pupils’ raw test scores in advanced mathematics for the year 2008. These scores are one

in a series of Trends in Mathematics and Science Study (TIMSS) assessments conducted by

International Association for the Evaluation of Educational Achievement. To the best of our

knowledge this is the only series in TIMSS that provides micro level information on raw pupils’

test scores. The first cycle of TIMSS was conducted in the year 1995 for mathematics and

science at several grade levels including senior secondary school pupil in the final year of

secondary school. The achievement scores that we employ are the outcome of the second

advanced TIMSS 2008 assessments for students only studying advanced mathematics in their

last year of secondary school completion. These achievement tests are designed to measure

cognitive skills in advanced mathematics with a focus towards improving learning and teaching

practices. The tests in mathematics cover three main domains, namely algebra, calculus and

geometry. Moreover, the questions are designed to test thinking behaviors or three cognitive

domains namely knowledge, application and reasoning.

A total of 10 countries with a divergent socio-economic background, different set of

cultures and geographic parts of the world participated in the 2008 advanced mathematics

assessment by TIMSS. These include: Armenia, Iran, Italy, Lebanon, Netherlands, Norway,

77

Philippines, Russian Federation, Slovenia and Sweden. All these countries participating in the

TIMSS advanced 2008 have a different overall size of their age cohorts and number of students

enrolled in the advanced mathematics programme. Therefore, the sample includes students

enrolled in the final year of their secondary school or adjacent grades with the same age cohort.

Moreover, TIMSS advanced 2008 employs a two-stage stratified cluster design, with schools

sampled first followed by the selection of one or more classes from a list of eligible classes in the

school.40

The procedures involved in assessments and data collection were quality controlled at

each step and monitored by control observers arranged by the International Association for the

Evaluation of Educational Achievement (IEA) secretariat. Appendix F contains the definitions

and descriptive evidence regarding the data used for the analysis.

4.2.3 Empirical Framework for Country-wise Analysis

In order to develop our country-wise analysis we adopt a standard regression framework.

In the equation school inequalities are considered a function of a vector of school-level variables

X1, and a vector of teacher-level variables X2. Note that, while we have used a uniform notation

to represent each country’s regression, the vectors X1 and X2 constitute a different set of

variables for each country. The equation takes the following form:

Si = α0 + α1X1 + α2X2 + εi (4.4)

In equation (4.4) Si is the inequality for schools within a country. As outlined in the

previous sub-section our data set allows us to estimate within school skill-inequalities specific to

each country in our sample. In addition to raw test scores, the TIMSS data set includes

information on an array of individual, family, school and teacher characteristics. For reasons

40

Details regarding sampling, assessment procedures, curriculum and other material relevant to TIMSS advanced

2008 can be found in the TIMSS advanced User Guide 2008, which is available at the International Association for

the Evaluation of Educational Achievement website.

78

mentioned in the sections to follow, we only focus on school and teacher characteristics which

can be aggregated at the level of schools. Furthermore, as mentioned earlier and explained in

further detail below we abstract from features pertaining to geographical location of the country,

its income level or institutional characteristics.

First we elaborate on the variables that, depending on their importance – where the

importance is determined based on specification tests for each country - may or may not

constitute the vector X1. To construct these variables, we extract information about school and

class characteristics from the TIMSS (2008) school level questionnaires for each participating

country. The key information of interest relates to the percentage of students from an

economically disadvantaged background, the percentage of students with language of test as

their native language, total number of students enrolled, student-teacher ratio and location of the

school. In what follows, we describe the details of the construction of these variables and the

hypotheses in the related literature regarding their impact on school performance.

Earlier evidence from PISA and OECD countries indicates that students from advantaged

backgrounds and language of test as their native language score twice as high as compared to

students belonging to disadvantaged socio-economic background or lacking proficiency of native

language (OECD, 2013; Oppedisano and Turati, 2015). In order to incorporate these dimensions

we use information present in the school questionnaire and include two variables in our analysis.

To represent economic background, the vector X1 includes four categorical dummy variables,

which represent the extent to which the student population constitutes economically

disadvantaged students. These variables take the value of 0 or 1 depending on whether the

school’s percentage of students coming from low socio-economic background is respectively 0-

10%, 11-25%, 26-50% and more than 50%. Similarly, to represent language proficiency we

79

include four categorical dummy variables depending on whether the percentage of students with

low language proficiency is respectively more than 90%, 76-90%, 50-75% and less than 50%.

Moreover, studies examining the impact of size of class or school on educational

achievement of pupils provide inconclusive evidence (Angrist and Lavy, 1999; Lee and Leob,

2000; Akabayashki and Nakamura, 2014). This strand of literature shows that small class and

school size has a significant and positive effect on the educational attainment levels of students.

On the other hand, studies focusing on TIMSS data set indicate no systematic and significant

effect of class size on academic achievement and inequality levels of students (Pong and Pallas,

2001; Li and Konstantopoulos, 2017). To capture these features we include total number of

students enrolled as a numerical variable. In addition, we construct a variable to measure student-

teacher ratio using the information on number of students and teachers provided in the data set.

Furthermore, earlier evidence on the location of a school also provides contradictory

conclusions. Axtell and Bowers (1972) find that students from rural schools perform

significantly better than their urban counterpart in verbal aptitude tests. On the contrary, Owoeye

and Yara (2011) and show that schools located in areas with low population density also have a

negative impact on student performance levels. To assess the impact of location, we again

construct a set of 6 categorical dummy variables depending on whether the school is located in

the following locations: more than 500,000 people, 100,001 to 500,000, 50,001 to 100,000,

15,001 to 50,000, 3,001 to 15,000 and 3,000 people or fewer.

80

The vector X2 includes teacher related attributes.41

These attributes pertain to the subject-

specific experience (in this case years teaching mathematics), and their job-satisfaction levels.

We extract this information on teacher related attributes from TIMSS (2008) school and teacher

level questionnaires. While none of the extant literature studies the implications of these factors

for inequality, we expect these to be relevant given that other studies consider their impact on

student performance, which has an indirect bearing on inequality. For example, Hanushek and

Woessmann (2014) find that cognitive skills or quality of teachers are among the important

factors which determine international differences in student performance levels. Moreover, better

skills of a teacher tend to improve the skills of students coming from disadvantaged socio-

economic backgrounds in contrast to student from economically well-off backgrounds. Barro and

Lee (2001), also indicate a positive impact of teachers’ salary levels on student achievements.

Similarly, Woessman (2003) using information on TIMSS shows a positive association between

teacher experience and student achievement levels. In the light of this evidence we include six

variables: one variable pertaining to total years of experience teaching mathematics and five

categorical dummy variables relating to whether job-satisfaction is very high, high, medium, low

and very low.

4. 3 Empirical Evidence on Inequalities in Educational Achievements

This analysis starts by estimating generalized entropy indices for a sample of 10 countries

which decompose human capital inequality into two main components; inequality between and

within countries. Second, we construct inequality indices at a country-specific level by grouping

students into schools and estimating within-school inequalities. Subsequently we perform a

41

Again, this is the total set of variables we consider. As we focus on a country-wise analysis, specification tests

determine which of these variables are important or pertinent in determining a given countries within-school

inequality.

81

country-wise analysis of the determinants of these inequalities. This analysis assists in revealing

the underlying patterns and causes associated with variation in human capital specific to the

educational system of a country.

4.3.1 Skill-Inequalities: A Cross-Country Analysis

The results for all three generalized entropy indices are presented in Appendix G, table 1,

which show within and between human capital inequalities for all countries included in our

sample. The first panel contains results for inequality measured as mean log deviation (α= 0).

The second panel provides results for Theil index (α=1) and the last panel explains results for

coefficient of variation (α = 2).

Overall, the results using raw test scores indicate that within-country human capital

inequality contributes to around 70% in total educational achievement inequality.42

This implies

that within-country inequality dominates between-country inequalities in case of mathematics

skills as measures of educational achievement. Interestingly, this evidence on skill inequality is

in contrast to literature on inequalities in human capital measured using average years of

schooling, which are suggestive of greater differences between countries than within each

country (Castelló and Doménech, 2000). Moreover, our finding is also contrary to earlier

empirical evidence for income inequality, which highlights that within-country inequality is

lesser than between-country inequality (Li et al, 1998).

Furthermore, our results lend support to earlier evidence on human capital inequalities in

educational achievements (Sahn and Younger 2007). However, our shares of within-country

inequalities across all three measures of dispersion indicate a value of 70% as compared to 52%

42

The figures are average value of shares of within-country inequality for all three general entropy measures.

82

reported in earlier study associated with mathematics skills. One possible explanation for this

difference in results is that we employ a higher level of educational achievement based on

secondary school levels in contrast to grade eight scores used in Sahn and Younger (2007).

Therefore, there is a possibility that dispersion in mathematics test scores increase over time,

thereby widening human capital inequalities.43

Also, skill inequalities in mathematics may vary

with a change in the content or nature of the subject under consideration, as greater specialization

and difficulty of the curriculum exacerbating the inequalities, given that “falling behind” has a

cumulative effect on students at the lower end of the performance spectrum.

In the background of this evidence measuring skill-inequalities using data on educational

achievements, we suggest that skill inequalities stem from disparities that exist in educational

quality within a particular a country rather than across countries. To understand the sources of

this inequality of we estimate inequality indices by grouping students into schools for each

country in our sample that participated in the 2008 advanced mathematics test. One of the

reasons for grouping students into schools is based on earlier literature which envisages schools

as a key institution that play a vital role in shaping national strategies, goals and targets

(Prucha,1979; Smith, 2006; Lenzi et al, 2014). Hence schools are not only confined as individual

entities that impart knowledge through formal learning but contribute in the development and

behavior of individuals as citizens. In addition, we also find literature in macroeconomics

suggestive of differences in institutions across countries as one of the fundamental causes which

lead to differences in economic growth and development (Acemoglu et al, 2005). If, as the

literature above suggests, schools may be considered as institutions, and political and economic

43

Note that we are not comparing the same group of students, so we cannot say that inequalities within a cohort of

students have increased. We can only conclude that human capital inequalities in general are higher when evaluated

at a higher level of education.

83

institutions influence growth, then the appropriate empirical exercise is to examine school related

characteristics that result in skill-inequalities specific to each country’s educational system.

Secondly, we find that school characteristics have not received due attention in the

literature on skill-dispersions. This strand focuses more on student, parental and family

attributes, socio-cultural and economic status as determinants of skill- inequality (Shavit and

Blossfeld, 1993; Breen et al, 2009; Freeman et al, 2010). Few studies which include school

related variables in empirical analysis treat these as a part of school fixed effects (Oppedisano

and Turati, 2015). They indicate a large and significant impact of these effects on skill dispersion

levels and emphasize the need to open this “black box” in order to reveal the school

characteristics which cause skill-inequalities. In doing so it is possible to unearth specific targets

of relevance to policy. For example, whether the inequality occurs due to an unfavourable

student-teacher ratio or teachers’ experience has different implications for policy, and is

therefore important to identify.

4.3.2 Skill-Inequality Indices at Country and School level

Appendix G, Table 2, reports within and between-school inequalities estimated for all

countries. A review of measures of dispersion indicates that, analogous to the country-level,

within-school inequalities exceed between-school inequalities. Appendix G, Table 3, ranks

countries in an ascending order based on within-school skill inequality levels measured as mean

log deviation. Our estimations show that the top five countries with highest within-school skill

inequalities also have the lowest average test scores. Likewise, countries with lowest within-

school skill inequality are those with highest average test scores in TIMSS 2008 advanced

mathematics tests. This finding is somewhat similar to Sahn and Younger (2007), who employ a

cross-country data set for TIMSS 1999 and 2003 and find a negative correlation between value

84

of inequality and respective average country test scores. Moreover, Freeman et al (2010) using

eight grade mathematics test scores from 1997 and 2007 waves of TIMSS, suggest that countries

with highest test scores also exhibit least inequality in test scores. We find this relationship rather

obvious as the calculation of skill-dispersion employs average test score in a manner that the

relationship should present a negative correlation between the two; we therefore do not include

average test scores in our regressions.44

Apart from country-specificity of inequality in human capital, there is the possibility that

certain broad determinants characterizing groups of countries may have an impact on school-

level inequalities. Since we wish to perform a parsimonious regression analysis focusing on

school-specific features only, it may be of interest to examine these features from the point of

view of eliminating control variables that are unnecessary and including only those aggregate

level variables that are more salient. To that end, we categorize countries on the basis of

educational system, income and geographical location, factors that have been hypothesized to be

of importance in student performance levels.45

For example, earlier literature provides evidence

indicating that decentralized system of schooling has a positive association with equity in

educational achievements of pupils (Causa and Chapuis, 2009; Rodriguez-Pose and Ezcurra,

2009).

44

We refer to equation (4.1) in the methodology section 4.2 for this study. It presents the formula for generalized

entropy index. Here we find μ is the mean or average test score stated in the denominator and points to an obvious

inverse association between inequality (α) and average test scores. 45

We categorize countries on the basis of system of education in table 3 column 5. Educational systems of countries

fall into two main categories; centralized and decentralized. The former has decision making power vested in the

central government or ministries of education. In the later the powers are deconcentrated, with the control, shared or

transferred to sub-governments at different levels. We obtain the information about the type of educational system in

a country form respective official sources such as ministry of education websites.

For the purpose of income categorization in column 6, we employ the World Bank, World Development Indicators

2017, categories classifying countries on the basis of their income levels. Lastly, column 7 uses the geographical

categorization of countries also based on the World Bank, World Development Indicators 2017, data-set. For details,

see, www.worldbank.org

85

A closer look at Table 3, Appendix G, columns 1 and5, reveals no clear pattern of

association between the type of education system and level of skill-inequality. We also perform

this exercise for income (column 6) and geographical location (column 7) and find no link

between these categories and level of skill inequality. In subsequent analysis, we therefore focus

on school-specific characteristics only.

4.3.2.2 Results of Country-wise Analysis with a Common Set of Variables

We begin our analysis by developing a model of human capital inequality based on a

common set of explanatory variables discussed above in section 4.2.46

We do not discuss this

analysis in much detail here; the point we wish to make is that these estimations do not indicate

any specification that fits the data well for all countries. This confirms our intuition that causes

of educational inequality are specific to each country. As a consequence, we are of the view that

a more meaningful approach entails an exploration that aims to find a specification suitable for

each country by performing case-by-case analysis of each country rather than attempting to find

a common specification that fits the empirical experience of all countries.

Therefore, we move to the second part of our analysis, which is based on an individual

country analysis of human capital inequality. To find the empirical specification that is

representative of a particular country we perform a specification analysis for each country. Since

this analysis is based on a large number of regressions, we do not present it here.47

In what

follows we present a discussion of the end result of this analysis by examining only the model

selected for each country following the specification analysis.

46

Appendix H, table 1 contains the definitions for all the variables discussed in section 2. Table 2 contains the

results for the regression analysis. 47

The detailed specification analysis is available upon request. For the benefit of the reader we present the analysis

for Netherlands in Appendix G Table 3.

86

4.3.2.3 Country-Specific Analysis

In this section we present country-specific regression analysis and individually analyze

composition of inequality in 10 countries that participated in the advanced 2008 (TIMSS)

secondary school mathematics test.48

We begin our analysis with the country that has least

within-school inequalities followed by the rest in an ascending order.

i) Lebanon

We begin our analysis by focusing on Lebanon. The results of human capital index indicate

that Lebanon exhibits the lowest within-school inequality across all countries in our sample. A

review of the country’s educational profile indicates that the adult literacy rate is around 93.9%

and 54.2% of the population has at least some secondary education.49

These macroeconomic

figures are perhaps also reflected in our micro level analysis which places Lebanon with lowest

inequality in human capital. Lebanon has the third lowest number of students (1612) and highest

number of schools (212).

Our decomposed skill dispersion data on educational quality for Lebanon at the level of

schools reveal that 50% of schools attain mean test scores higher than the overall mean

calculated for all schools.50

In the context of skill-disparities, 58.1% of schools have skill-

48

Appendix F, table 1, includes summary descriptive statistics for the sample of students and schools for all

participating countries. Appendix G, table 3, ranks countries in ascending order based on within-school human

capital inequality index. Appendix H, table 4, contains information on composition of human capital inequality at

the level of schools. In what follows we present the results of country estimations in a pursuit to identify the

plausible causes of inequalities specific to their respective educational systems. Appendix H, table 5 (i-x) presents

the summary statistics for variables included in country-wise analysis. 49

United Nations Development Programme Country Profiles. 50

The overall school mean test score, is the average test scores of all schools, divided by total number of schools.

This gives the overall school mean test score or average mean skill level across all schools. We develop this as a

benchmark measure for skills to indicate the performance of schools within a country. In our analysis the schools

with higher than overall school mean test scores from now onwards will be referred as high skill achieving schools.

87

inequality levels less than average skill inequality across schools.51

Our results also show that

35.5% of schools have both the above mentioned skill related attributes. This implies that not all

high skill achieving schools are also low skill disparity schools. We employ our decomposed

skill-inequality indices as an attempt to uncover the possible factors linked to dispersion in

mathematics skills at final year secondary school level. We obtain the following factors

associated with inequality in schools for Lebanon:

51

Our estimations calculate a skill-inequality level for each school in the sample. We calculate average skill-

inequality as a summation of these skill-inequality levels for all schools divided by total number of schools. We

develop this value as a benchmark measure for skill inequality. All schools with skill inequality below this

benchmark inequality in our analysis from now onwards will be referred as low-skill disparity schools.

88

Table 4.1 Regression Results for Lebanon

Dependent variable : Human capital inequality (Si)

Variables

Student-teacher ratio (𝑠𝑡) 0.0013***

(0.00047)

Location of School

category 1: More than 500,000 people (𝐿1)

category 2: 100,001 to 500,000 people (𝐿2) 0.052

(0.0077)

category 3: 50,001 to 100,000 people (𝐿3) 0.012*

(0.007)

category 4: 15,001 to 50,000 people (𝐿4) 0.013**

(0.006)

category 5: 3,001 to 15,000 people (𝐿5) 0.0145**

(0.0067)

category 6: 3,000 people or fewer (𝐿6) 0.0076

(0.008)

Teacher's job satisfaction

category 1: very high (𝐽1) -0.0468*

(0.027)

category 2: high (𝐽2) -0.0464*

(0.0268)

category 3: medium (𝐽3) -0.0351

(0.0268)

category 4: low (𝐽4)

Teacher's Experience teaching mathematics

(𝑒𝑥𝑝𝑚)

0.00051

(0.0001)

Constant 0.057**

𝑅2 0.118

Observations 186

Note: The sample size is representative of 1612 students grouped into 186 schools. Standard Errors in parenthesis;

*, **, *** imply 10%, 5%, 1% significance levels. The reference category for location is; location option 1 which

includes population density of more than 500,000 people. Other categories included (L2, L3, L4, L5 and L6),

contain population densities lesser that location option 1. The reference category for job satisfaction is: option 4

which is low satisfaction level. Other categories included (J1, J2 and J3) contain job satisfaction levels higher than

option 4.

In Table 4.1 we find a positive and significant relationship between student-teacher ratio

(st) and school level inequalities (𝑆𝑖). This implies that, in the case of Lebanon, higher student

teacher ratio may lead to a reduction in individual interactions between students and teacher,

more potential management and discipline issues and lesser time spent on learning. This is

consistent with the broader evidence on educational achievement which shows a positive

89

association between small class size and performance levels of students (Krueger, 1999; Nye et

al, 2000b). However, there is mixed evidence on class size reduction and impact on reducing

achievement gaps between students (Finn and Achilles, 1990; Nye et al, 2000a;

Konstantopoulos, 2008; Li and Konstantoploulos, 2017).

The second variable is location of school. It is a categorical variable consisting of six

categories (L2, L3, L4, L5 and L6). Our results show that all locations with smaller relative to

highest population density school locations are positively and significantly associated with

dispersions in mathematics test scores. This suggests that schools situated in small towns and

rural locations relative to schools in big studies and metropolitan areas have higher human

capital inequality. Again, literature on educational achievements shows that students from

schools located in rural or sub-urban settings have relatively low performance levels in

comparison to students belonging to schools located in urban settings. These schools are fraught

with several issues such as lack of qualified teachers, unavailability of means of travel and

communication and disparity in the distribution of resources which may lead to gaps in

achievement between the urban and rural schools (Hallak, 1977; Balogun, 1982). We also relate

this argument to Lebanon where public spending is biased in favour of urban schools. This bias

is causing a perpetuating rural-urban divide between schools leading to inefficient spending and

poor quality of public education (UNDP 2006, UNFPA 2011-12).52

The presence of this divide

perhaps reinforces our empirical evidence, indicating that schools located in rural locations are

associated with higher dispersion in skills as they lack financial resources to meet their

educational requirements.

52

United Nations Development Programme 2006. Towards the Rise of Arab Women: The Arab Human

Development Report. The United Nations Population Fund, formerly the United Nations Fund for Population

Activities (2011-12), Lebanon: An Overview Context, Evolving Demographics for Women, Sexual and

Reproductive Health, Poverty and Women, Gender and Rights

90

Furthermore, the model includes job satisfaction as a series of 4 one-zero dummy

variables (J1, J2 and J3). We find that higher job satisfaction levels of teachers are associated

with lower skill-inequality in mathematics test scores. In addition, high and medium job

satisfaction levels significantly reduce disparities in skills. Intuitively, teachers satisfied with

their work environment should put more effort in imparting knowledge and education to

students. It may lead to improved educational achievement among students thereby reducing

dispersion in performance levels. In the context of Lebanon, we find that conventional wisdom

prevails as all the variables present in the estimated model have the expected signs as

hypothesized in the literature on school performance.

Hence, we may suggest that policy makers aiming to reduce disparities in skills should

target schools with a higher number of students per teacher and schools located in less densely

populated areas, and attempt to uncover the reasons attributed to their poor performance levels in

terms of school characteristics. However, these policies naturally exert financial pressures and

may have redistributive and institutional implications for a society. In that case, the choice of

expenditure geared towards reducing human capital inequalities originating from a particular set

of schools may be at the cost of the sections of society who do not directly benefit from such

schools.

ii) Netherlands

Netherlands is ranked second in terms of within-school inequality as shown in Appendix H,

Table 3. To further elaborate on the specificity of skill composition for Netherlands, our skill-

dispersion estimations show that 50.89% of schools are high skill achieving schools and 59.82%

fall in the category of low-skill disparity schools. In addition, 38.39% of the schools feature both

these skill based attributes. This indicates that Netherlands has a higher percentage of high skill

91

achieving schools as compared to Lebanon as well as a slightly higher percentage of low-skill

disparity schools. In order to understand the skill-dispersion composition, we attempt to identify

some of the possible factors associated with dispersion in mathematics test scores for schools in

Netherlands. In the context of Netherlands our specification analysis unearths the following

regression results:

Table 4.2 Regression Results for Netherlands


Variables

Student-teacher ratio (𝑠𝑡)

0.0021**

(0.0048)

Enrollment in the twelfth grade (E)

-

0.00021***

(0.00008)

Percentage of Students from economically

disadvantaged background

category 1: 0-10% (D1)

-0.028***

(0.009)

category 2: 11-25% (D2)

-0.036***

(0.01)

category 3: 26-50% (D3)

-0.026**

(0.012)

category 4: More than 50% (D4)

Constant

0.054***

(0.009)

R2 0.243

Observations 97 Note: This sample size is representative of 1537 students grouped into schools, hence the N= 97. Standard Errors in

parenthesis; *, **, *** imply 10%, 5%, 1% significance level. In this particular case we have four categories of

percentage of students from economically disadvantaged background represented as D1, D2, D3 and D4. The

reference category for this is option D4 referring to more than 50% of students belonging to economically

disadvantaged background.

Table 4.2 includes; student-teacher ratio (st), total number of students enrolled (E) and

the percentage of students from economically disadvantaged background (D) as a categorical

92

variable. All these variables belong to the set of school related attributes, and have coefficients

that are significant and have the expected sign.

In common with Lebanon we also find student-teacher ratio positively and significantly

associated with human capital inequality for Netherlands. More specifically, we comment here

on our results that size of school measured as number of students enrolled (E) is negatively and

significantly associated with skill-inequalities in schools for Netherlands. This perhaps implies

the presence of the economies of scale in large schools. Previous literature on this aspect presents

evidence that is mixed; there are studies indicating that school performance can be positively or

negatively associated with number of enrollments. In this context, we focus on the former strand,

which highlights potential reason underlying the positive association we find here.53

For

example, some literature suggests that as the school gets bigger resource wastage is minimized

leading to more savings and increased efficiency (Buzacott, 1982). In addition, large schools

seem to have enough students with similar needs to warrant spending on specialized programmes

catering to them. On the contrary, small schools tend to focus their resources towards basic

programmes and students with marginalized needs remain excluded from such activities (Monk

and Haller, 1993). Given this literature, we may argue that large schools in Netherlands perhaps

have better management, organizational and resource infrastructure which positively impacts the

performance of students and leads to reduced disparities in skills.

To provide some context, Netherlands has the second highest Gross Domestic Product

per capita (GDP per capita, US$ 56,928.82) among countries included in our sample.54

Provision

of education is free and compulsory at primary and partially supported at the secondary school

53

We will comment on the other strand in the context of Iran, where the association between enrolments and skill

inequality is negative. 54

Gross Domestic Product per capita current US$, World Bank, World Development Indicators 2016. As our

analysis is based on TIMSS 2008 data set, we refer to the figures for the year 2008.

93

levels. In addition, the Dutch system of schooling aims at improving both quality as well as

equity in education through various policies.55

These statistics reflect higher economic prosperity

and the likely presence of educational support mechanisms that facilitate weaker students.

iii) Russia

Russia is ranked third in the list of countries for within-school inequalities. Overall, it has

the second highest number of students (3185) as well as schools (143) across all countries. Our

human capital inequality index for schools in Russia indicates that half of the schools (50.34%)

are high skill achieving and 58.74% low-skill disparity schools. Russia has the second highest

percentage of schools (45.45%) that have both the skill attributes in comparison to other

countries in the sample. We obtain the following set of factors specific to human capital

inequality for Russia:

55

For details see Educational Policy Outlook, Netherlands 2014. Organization for Economic Co-operation and

Development (OECD).

94

Table 4.3 Regression Results Russia


Variables

Location of School


-0.0047***

(0.016)

category 2: 100,001 to 500,000 people (𝐿2)

-0.044***

(0.015)


-0.038**

(0.016)


-0.037**

(0.017)


category 6: 3,000 people or fewer (𝐿6)

-0.012

(0.028)

Percentage of Students with language of test as

their native language

category 1: More than 90% (𝑁1)

-0.034**

(0.011)

category 2: 75 to 90% (𝑁2)

-0.022

(0.015)


-0.024

(0.017)

Less than 50% (𝑁4)


-0.0016

(0.0002)

Constant

0.129***

(0.017)

R2 0.184

No of Observations 139 Note: This sample size is representative of 3185 students grouped into schools, hence N= 139. Standard Errors in

parenthesis; *, **, *** imply 10%, 5%, 1% significance level. In this case we have six categories for location (L1,

L2, L3, L4, L5 and L6) and the reference category is option L5, referring to a location having 3,001 to 15,000

people. We only interpret the results for the significant categories which are; L1, L2, L3, L4.In this case we have

four categories for percentage of students with language of test as their native language (N1, N2, N3 and N4) and

the reference category is N4, referring to less than 50% students.

The first variable in Table 4.3 is the location of school, which is a series of 6 one-zero

dummy variables (L1, L2, L3, L4 and L6). The results obtained here are quite similar to the case

of Lebanon and suggest that schools located in high relative to less densely populated locations

95

are negatively and significantly associated with dispersion in mathematics skills. As in the case

of Lebanon, it may be that schools in bigger cities face relatively fewer constraints in relation to

teacher quality, physical infrastructure and financial resources in contrast to schools situated in

less populated or rural areas.

Furthermore, the model contains percentage of students with language of test as their

native languages series of 4 one-zero dummy variables (N1, N2, N3 and N4). We find that the

category N1, i.e., schools with more than 90% of students with language of test as their native

language, has a negative and significant coefficient, suggesting that beyond this threshold level

human capital inequality is lower relative to other categories.56

Russia has the world’s second

largest immigrant population after United States (UN Population Division estimates, 2013). Such

large numbers of migrants are reflective of a multi-lingual society with diverse set of languages

spoken. A review of the educational policy of Russia shows that majority of schools located in

bigger cities such as Moscow offer proper Russian language classes in schools to foreign

students. Such policies perhaps improve the proficiency of migrants over native language

resulting in better performance and reduced human capital inequality.

iv) Iran

Iran is ranked fourth in terms of within-school inequality. It has the third highest number

of students (2425) and schools (119) across all countries. Iran’s educational quality disparity

structure shows that less than fifty percent of schools (39.94%) are high skill achieving schools.

In the context of skill inequality, more than fifty percent of schools (58.82%) are classified as

low-skill disparity schools. Both these characteristics are present in 36.13% of schools in the

56

Note that, in the case of Russia, the language of the test is Russian. In our sample of schools for Russia, more than

80% of the schools fall in the N1 category with 90% or more students with language of test as their native language.

96

sample. In order to further reveal and examine the underlying causes of disparities in educational

quality in Iran, our specification analysis reveals the following set of factors specific to

dispersion in educational quality for Iran:

Table 4.4 Regression Results Iran


Variables

Enrollment in the twelfth grade (𝐸)

0.00006***

(0.0002)

Location of School


-0.142***

(0.047)


-0.13***

(0.048)

category 3: 50,001 to 100,000 people(𝐿3)

-0.12**

(0.05)


-0.142***

(0.051)


-0.128**

(0.05)

category 6: 3,000 people or fewer (𝐿6)


(𝑒𝑥𝑝𝑚)

-0.00019

(0.0001)

Constant

0.211**

(0.047)

R2 0.139

Observations 111 Note: This sample size is representative of 2425 students grouped into schools, hence N= 111. Standard Errors in

parenthesis; *, **, *** imply 10%, 5%, 1% significance level. In this case we have six categories for location (L1,

L2, L3, L4, L5 and L6) and the reference category is option L6, referring to a location having 3,000 people or fewer.

In Table 4.4 number of students enrolled (E) is positively and significantly associated

with human capital inequality. This finding is in contrast to Netherlands where schools with

higher enrolment rate tend to reduce disparities in human capital. As mentioned earlier, evidence

regarding enrollments and performance is mixed. In the context of Iran we draw on literature

suggestive of a negative link between enrollments and performance in order to provide a

potential explanation for this result. For example, Lee and Smith (1997) suggest that students

97

enrolled at relatively smaller schools in contrast to large schools have higher achievement gains

in the subject of mathematics. Moreover, sociological evidence considers size of school as an

ecological feature and treats it as a social structure which influences our physical and social

interactions. This strand of literature shows that social relations are more positive in small rather

than big schools (Bryk and Driscoll, 1988; Lee et al, 1993). Lastly we have the categorical

dummy variable location of school (L1, L2, L3, L4 and L5) in our analysis. In common with

Russia we also find here that schools located in densely populated areas relative to schools

situated in less densely populated area negatively and significantly associated with human capital

inequalities.

v) Slovenia

Slovenia has sixth highest number of students (2156) and second lowest number of

schools (79) across all countries in our sample. Its skill-dispersion structure shows that 50.63%

of schools are high skill achievers and 62.02% are low-skill disparity schools. Moreover, 37.97%

of schools seem to have both the above mentioned characteristics. We employ our decomposed

skill-inequality indices to uncover the possible factors influencing dispersion in mathematics

skills at final year secondary school level. The following set of variables is obtained after

specification tests conducted for human capital inequality in Slovenia:

98

Table 4.5 Regression Results for Slovenia


Variables

Percentage of Students from economically disadvantaged

background

category 1: 0-10% (𝐷1)

-0.038**

(0.015)

category 2: 11-25% (𝐷2)

-0.026*

(0.013)

category 3: 26-50% (𝐷3)

category 4: More than 50% (𝐷4)

-0.007

(0.018)


-0.0019*

(0.051)

Teacher's Experience teaching mathematics(𝑒𝑥𝑝𝑚)

-0.00047

(0.0005)


category 1: very high (𝐽1)

category 2: high(𝐽2)

0.039

(0.024)

category 3: medium(𝐽3)

0.049**

(0.0023)

category 4: low(𝐽4)

0.017

(0.037)

Constant

0.131***

(0.031)

R2 0.274

No of Observations 68 Note: The sample size of is representative of students is 2156 grouped into 68 schools. Standard Errors in

parenthesis; *, **, *** imply 10%, 5%, 1% significance level. In this case there are four categories for percentage of

students belonging to economically disadvantaged backgrounds (D1, D2, D3, D4) and the reference category is D3,

referring to 26 to 50% of students. In this case there are four job satisfaction categories (J1, J2, J3 and J4) and the

reference category is J1 referring to very high job satisfaction level.

The results in Table 4.5 show, that percentage of students from economically

disadvantaged background (D1, D2 and D4) has a negative association with human capital

inequality. More specifically, the first two categories with a combined range of 0-25% are

inversely and significantly related to skill-inequalities in mathematics skills. This implies that

relatively lower percentage of students in schools from these backgrounds does not lead to an

increase in skill disparities. To provide some context the PISA (The Programme for International

99

Student Assessment 2012) test scores also indicate a low impact of socio-economic background

on student performance level for Slovenia. A review of the educational policy for Slovenia

indicates presence of system-level policies and mechanisms promoting equity in education,

which provide support to students from socio-economically disadvantaged background across

regions.57

Counter intuitively, the coefficient of student-teacher ratio (st) is negative and significant

in our estimations obtained for Slovenia. On the contrary, we found a positive and significant

association of student-teacher ratio in the case of both Lebanon and Netherlands. Note that in our

sample of countries Slovenia has the second highest average student-teacher ratio (22.28) after

Philippines (34.75) across countries in the sample. Given this high student teacher ratio, the

expected sign of coefficient should have indicated a positive association with skill-disparities.

The results for teachers’ experience teaching mathematics perhaps provide further intuition to the

negative association of the student teacher ratio with skill inequalities. The coefficient for

teachers’ experience indicates higher experience associated with lower human capital dispersion.

Hence, we suggest that presence of experienced teachers in schools with greater number of

students per teacher offsets the negative impact of bigger class size on skill inequality.

Furthermore, lower job satisfaction levels (J2, J3 and J4) of teachers are positively associated

with higher level of skill disparities. More specifically, any level of job satisfaction relative to

very high, even the categories of high and medium significantly lead to an increase in the level of

skill-inequalities.

57

For details see, Education Policy Outlook Slovenia, OECD (2016).

100

vi) Philippines

A review of our human capital index shows that Philippines is ranked sixth in the context

of within-school inequality. It is third in terms of number of schools (118) and has the highest

number of students (4091) across all countries. These figures indicate that the quantity of eligible

students in Philippines who participated in the test is the highest compared to the other

participating countries. Our human capital inequality index for Philippines shows that 41.52%

are high skill achieving and 55.93% are low-skill disparity schools. In addition, 28.81% of

schools exhibit both these skill based attributes. In comparison to other countries Philippines is

placed in the bottom three countries in terms of percentage of high skill achieving schools. Table

4.6 includes the possible determinants of within-school inequalities obtained for Philippines:

Table 4.6 Regression Results for Philippines


Variables


0.000061

(0.00003)


background

category 1: 0-10%(D1)

category 2: 11-25% (D2)

0.0052

(0.011)

category 3: 26-50% (D3)

0.0162*

(0.009)


0.0057

(0.008)

Teacher's Experience teaching mathematics (expm)

-0.000014

(0.0006)

Constant

0.089***

(0.015)

R2 0.03

No of Observations 4091 Note: The sample is representative of 4091 students into 111 schools. Standard Errors in parenthesis; *, **, ***

imply 10%, 5%, 1% significance level. In this case there are four categories of percentage of students belonging to

economically disadvantaged homes (D1, D2, D3 and D4). The reference category in this case is D1 which refers to

0-10% of students coming from economically disadvantaged homes.

101

The results in Table 4.6 show that the coefficient for student-teacher ratio (st) is positive,

which implies that greater number of students per teacher lead to greater human capital

inequality. On the other hand for Lebanon and Netherlands the association was not only positive

but also significant. Philippines has the highest average student-teacher ratio (34.57) across all

countries in our sample. Exploring the background in relation to the country we find that the

educational system of Philippines faces issues such as: big class size poorly paid teachers,

schools lacking proper teaching materials and wide regional disparities in school completion

rates. For this reason the Filipino children belonging to affluent class or expats do not attend

public schools.58

Furthermore, we find that higher percentage of students belonging to economically

disadvantaged backgrounds (D2, D3 and D4) implies higher level of skill disparity. More

specifically, if the percentage of students falls in the category of 26-50% (D3), which is almost

half of the student body, the impact on inequality in skills is both positive and significant. The

final variable in the estimated school inequality equation for Philippines is teachers’ experience

teaching mathematics (expm). The coefficient of this variable indicates a negative impact on

skill-disparities. One plausible argument for this is that teachers lack proper teaching aids,

materials and school facilities in Philippines as mentioned above. Hence, without proper teaching

infrastructure and facilities higher teaching experience may not be the sufficient enough to

reduce inequality in human capital.

58

For details see, World Education News and Review, Education in Philippines, Edition 2015 July.

102

vii) Norway

Norway is ranked seventh in terms of within-school human capital inequalities. It is also

seventh on the basis of number of schools (107) and students (1932) across all countries in our

sample. The results for skill disparity index indicate that 52.33% of schools are high skill

achieving and 56.07% are low-skill disparity schools. However, only 30.84% of schools have

both the above stated attributes. A comparison of our inequality composition reveals that Norway

has the highest percentage of high skill achieving schools across other countries. In order to

identify country specific agents associated with dispersion in educational quality for Norway, our

specification unearths the following set of variables:

Table 4.7 Regression Results for Norway


Variables


0.0012

(0.0007)

Percentage of Students with language of test as their native

language

category 1: More than 90% (𝑁1)

-0.126***

(0.041)


-0.118***

(0.042)


Teacher's Experience teaching mathematics (𝑒𝑥𝑚𝑝)

-0.0003

(0.0002)

Constant

0.211***

(0.042)

R2 0.11

No of Observations 104 Note: The sample is representative of 1932 students grouped into 104 schools. Standard Errors in parenthesis; *, **,

*** imply 10%, 5%, 1% significance level.In this case there are three categories of percentage of students with

language of test as their native language included in the model (N1, N2 and N3). The reference category is N3

referring to 50-75% of students with language of test as their native language.

In Table 4.7 the coefficient for student-teacher ratio (st) is positive which is similar to the

results obtained for Philippines. The second variable is percentage of students with language of

103

test as their native language (N1 and N2). The coefficients of both categories have a negative and

significant impact on skill-inequalities at the level of schools. This implies that greater

percentages of students relative to the category of 50-75% will reduce skill-disparities in

mathematics test scores.59

We find this evidence present in case or Russia and refer to the earlier

literature suggesting that speaking a non-national language in the country of residence has a

negative and significant impact on achievement levels measured as reading test scores. Hence,

the children of immigrants achieve lower educational scores in the country of residence

(Oppedisano and Turati, 2015). However, in case of Norway the impact of this variable is

significant if the student body is composed of 75% of such students, unlike Russia where the

impact becomes significant at later stage of 90% and above. In the case of Norway the medium

of instruction is based on native languages and English is rarely spoken and understood in

educational institutions. A review of the educational policy reveals that the Norwegian

government encourages the migrants to send their children to Kindergartens at an early age to

familiarize them with native languages. For this purpose, the government also provides subsidies

to the migrant families.60

viii) Armenia

Our estimated human capital inequality index places Armenia eighth in terms of within-

school inequalities. Overall, it has the lowest number of schools (38) and students (858)

participating in mathematics test. Our results for school indices reveal that it has the lowest

percentage of high skill achieving schools (39.47%) across all countries. In addition, we find

59

Norway and Lebanon are the only two countries that administered the test in two languages. However, in Lebanon

the test was administered in native and English language whereas in Norway both were native languages (Bokmal

and Nynorsk). 60

For details see, Education from Kindergarten to Adult Education, Norwegian Ministry for Education and

Research.

104

more than fifty percent of schools (55.26%) as low-skill disparity schools and 31.57% schools

with both these attributes. Table 4.8 explains set of factors associated with human capital

inequalities for Armenia.

Table 4.8 includes percentage of students from economically disadvantaged background

as one of the factors attributed with inequality in skills (D1, D2 and D3). The results show that

the coefficient for all three categories is positive but significant for category two only. Hence,

student percentages of 26-50% relative to the category of 50% or more, lead to a significant

increase in skill disparities in mathematics tests scores. This implies that even smaller percentage

of students from weak economic backgrounds positively and significantly impact skill

inequalities. This evidence in contrast to Netherlands where the results suggest that lower

percentage of such students is not associated with higher skill-inequalities. Armenia is a lower

middle income country and lags far behind in terms of GDP per capita (US$4,010.027) than

Netherlands (US$ 56,928.82).61

Therefore, for countries such as Armenia with relatively lower

economic prosperity a smaller percentage of students from weak backgrounds could be a source

of greater skill-dispersion. On the other end, for high income countries such as Netherlands with

relatively higher economic prosperity and advanced social and educational systems, smaller

percentages of students may not be a cause of concern.

61

Gross Domestic Product per capita current US$, World Bank, World Development Indicators 2016. The values

mentioned here are for the year 2008 as our analysis for skill-inequality is based on TIMSS 2008 data set.

105

Table 4.8 Regression Results for Armenia


Variables


background

category 1: 0-10% (D1)

0.046

(0.002)

category 2: 11-25% (D2)

0.119***

(0.03)

category 3: 26-50% (D3)

0.064**

(0.027)


Location of School

category 1: More than 500,000 people (L1)

0.0034

(0.021)

category 2: 100,001 to 500,000 people (L2)

0.063

(0.039)




0.095**

(0.055)

category 6: 3,000 people or fewer (L6)

Teacher's Experience teaching mathematics (𝑒𝑥𝑝𝑚)

-0.0016***

(0.0004)

Constant

0.0559***

(0.025)

R2 0.70

No of Observations 26 Note: This sample is representative of 858 students grouped into 26 schools. Standard Errors in parenthesis; *, **,

*** imply 10%, 5%, 1% significance level. There are four categories of percentage of students belonging to

economically disadvantaged homes (D1, D2, D3 and D4). The reference category in this case is D4 which refers to

more than 50% of students coming from economically disadvantaged homes. In this case there are six categories for

location (L1, L2, L3, L4, L5 and L6) and the reference categories are options L3, L4 and L6 are omitted or referred

categories.

In addition, Table 4.8 includes location of school (L1, L2 and L5) as one of the factors

influencing disparities in skills. In common with Lebanon, we find schools located in relatively

less populated locations positively and significantly influence skill-inequality. We refer to

secondary education institutions in Armenia that are funded based on their location and number

of students. There are huge gaps between what the schools receive financially and what they

106

require to meet their expenditures.62

Therefore, in such a case schools located in rural

backgrounds possibly face greater constraints compared to schools located in urban regions.

Finally, the equation includes teacher related attribute in the form of teachers’ experience

teaching mathematics (expm). The coefficient of experience has a negative and significant

association with skill-inequality.

ix) Italy

We move to Italy which has the sixth highest number of students (2143) grouped into 91

schools. Our results for human capital indices show that 43.95% of schools are high skill

achieving and half of the schools (50%) are low-skill disparity schools. However, only 29.67%

of schools fall in the combined skill category of schools. A comparison of decomposed skill-

inequality indices also shows that Italy has the least percentage of low-skill disparity schools.

Table 4.9, includes the possible factors associated with inequalities in human capital obtained

after our specification tests for Italy:

62

Access to School Education in Armenia (2012), Turpanjian Center for Policy Analysis, American University of

Armenia.

107

Table 4.9 Regression Results for Italy


Variables


-0.004**

(0.001)

Teacher's Experience teaching mathematics (𝑒𝑥𝑝𝑚)

-0.009*

(0.00005)


background

category 1: 0-10% (𝐷1)

-0.044**

(0.018)

category 2: 11-25% (𝐷2)

-0.03

(0.019)

category 3: 26-50% (𝐷3)

-0.029

(0.021)

category 4: More than 50% (𝐷4)

Constant

0.226***

(0.034)

R2 0.14

No of Observations 91 Note: This sample is representative of 2143 students grouped into 91 schools. Standard Errors in parenthesis; *, **,

*** imply 10%, 5%, 1% significance level. In this case there are four categories of percentage of students belonging

to economically disadvantaged homes (D1, D2, D3 and D4). The reference category in this case is D4 which refers

to more than 50% of students coming from economically disadvantaged homes.

Table 4.9 shows that the coefficient for student- teacher ratio (st) is negative and

significant. This indicates that higher the number of students per teacher, the lower the dispersion

in mathematics skills. In case of Lebanon and Netherlands student-teacher ratio has a positive

and significant association with inequality in skills, even though the average student teacher ratio

is lesser than Italy. In other words, we find that increase in student-teacher ratio is inversely

associated with skill-inequality in Italy even with a higher average student teacher ratio

compared to other countries. One possible argument could be the presence of a significant and

inverse association of teachers’ experience teaching mathematics at final year secondary school

level in the above estimated model.

108

In addition, our dummy variable percentage of students coming from economically weak

households (D1, D2 and D3) is negatively associated with inequalities in mathematics skills.

This implies that schools with relatively lesser proportion of students coming from this segment

of society are not associated with higher-dispersion in test scores. In Italy parents of children

from weak economic backgrounds have access to financial support. This support includes no

payment of fee and compulsory education form age 6 to 16 years. Moreover, after the age of 16,

student and his/her family is provided financial support as they do not pay fee except for a minor

enrollment tax paid at the beginning of the academic year.

x) Sweden

Our final country-wise study constitutes an examination of skill-disparities in educational

quality of Sweden. It has the highest within-school inequality and, ranked fourth and fifth in

number of schools and students respectively. Our estimates for skill-inequality indices show that

out of 116 schools, 47% of schools are high skill achievers and 28% are low-skill disparity

schools. In contrast to other countries Sweden has the lowest percentage of schools (28.44%)

with both the aforementioned attributes. We obtain the following factors associated with

inequality in schools for Sweden:

109

Table 4.10 Regression Results for Sweden

Note: This sample size is representative of 2303 students grouped into 89 schools. Standard Errors in parenthesis; *,

**, *** imply 10%, 5%, 1% significance level.In this case there are four categories of percentage of students

belonging to economically disadvantaged homes (D1, D2, D3 and D4). The reference category in this case is D3

which refers to 26 to 50% of students coming from economically disadvantaged homes. There are four categories of

percentage of students with language of test as their native language included in the model (N1, N2, N3 and N4).

The reference category is N3 referring to 50-70% of students with language of test as their native language.

Surprisingly, in Table 4.10 we find that determinants of inequality in human capital for

Sweden constitutes primarily of school factors without any teacher related attributes. This

perhaps implies that the factors associated with the institutional set up of the educational are

more relevant compared to teacher attributes in contributing to variation in human capital. The

coefficient of student-teacher ratio (st) is positive and significant, similar to the case for Lebanon


Variables

Student-teacher ratio ( 𝑠𝑡)

0.001***

(0.007)


background

category 1: 0-10% ( 𝐷1)

0.032

(0.02)

category 2: 11-25% ( 𝐷2)

0.035*

(0.02)

category 3: 26-50% ( 𝐷3)

0.076**

(0.035)

category 4: More than 50% ( 𝐷4)


language

category 1: More than 90% ( 𝑁1)

-0.046**

(0.018)

category 2: 75 to 90% ( 𝑁2)

-0.0047**

(0.019)

category 3: 50 to 75% ( 𝑁3)

category 4: Less than 50% ( 𝑁4)

-0.004

(0.03)

Constant

0.1073***

(0.027)

R2 0.09

No of Observations 89

110

and Netherlands. Note that in our sample Sweden is ranked 5th

with the average student-teacher

ratio of 15.99 across all countries.

Secondly, the coefficient of all the categories of percentage of students coming from

economically weak status (D1, D2 and D4) is positive. This indicates that higher disparities in

human capital are associated with children belonging to weak economic backgrounds.

Furthermore, the impact on skill-inequality is significant if quarter or more than half of students

in schools belong to this category. These results imply that moderate to higher percentages of

these students lead to significantly higher inequality in mathematics test scores. Sweden has a

relatively higher living standard and lower income inequality compared to the other European

Union countries.63

Lastly, the coefficient for percentage of students with language of test as their native

language (N1, N2 and N4) has a negative and significant impact on skill-inequalities which is

similar to Russia. However in case of Russia the impact became significant with 90% category of

students with language of test as their native language, unlike Sweden where the impact is

significant with 75% of such students in schools. These findings reinforce earlier evidence

presented in studies on OECD and TIMSS countries where the performance gap in reading

scores is greater among native and non-native language speaking students. Therefore, the focus

of educational policy reforms should be to identify and target schools with a higher student-

teacher ratio along with higher percentage of non-native language speaking students.

In the light of our country-wise evidence, we find that the pattern and set of factors

associated with human capital inequalities is different for every country. Hence, it would be

63

Source Eurostat data set, European Commission (2015).

111

unrealistic to think that one particular policy approach can cater to inequality attributed to skills

across countries. However, we do attempt and suggest a few measures to tackle inequality in

human capital based on our country-wise empirical estimations. For instance, student-teacher

ratio is positively and significantly associated with human capital inequality in Lebanon,

Netherlands and Philippines. One possible approach could be to recruit more teachers to cater to

a larger number of students and dividing bigger classes into sub-sections. Aforementioned

literature on school performance indicates that smaller class size positively impacts on

educational attainment levels of students. If smaller class size leads to better student

performances, then reducing the class size to design an appropriate student-teacher ratio perhaps

could lead to more equitable educational outcomes in these countries.

A look at evidence from Russia and Lebanon indicates that schools situated in rural

locations may be associated with inequality human capital. We find from aforementioned

literature that schools in rural locations face financial, infrastructure and teacher related

constraints. These constraints negatively impact on the performance levels of students and may

lead to inequitable educational outcomes. Therefore, one of the plausible approaches is to

redirect resources from other financially stable schools or sectors of the economy towards

schools in rural locations. In addition, results for Armenia and Philippines indicate students

belonging to weak economic backgrounds are associated with inequitable outcomes. Earlier

evidence indicates that resource-based interventions targeting pupils from such background leads

to reduction in achievement disparities (Gibbons and McNally, 2013). These measures to reduce

disparities in educational quality exert financial pressures and may have redistributive and

institutional implications for a society. In which case, the choice of expenditure geared towards

reducing human capital inequalities originating from a particular set of schools may be at the

112

cost of the sections of society who do not directly benefit from such schools. The economic

significance of our results stems from the empirical evidence presented in Chapter 2 which

suggests that skills learned by population of a country are closely associated with its long run

economic growth. This implies that greater variations in skills will also lead to wider cross-

country income and growth rate differences. Moreover, differences in individuals’ skill levels

determine their prospects of employability which overall impacts upon the unemployment rate of

an economy. Individuals equipped with better skills not only have better job prospects but also

have substantial returns in the form of higher earnings. At a microeconomic level, due to

variation in skills this may result in income differences among individuals. Its macroeconomic

impact becomes apparent when these variations in skills lead to significant variations in labour

market income levels which amount to greater income inequalities and result in widening

differences in the standards of living of the people.

In summary, an appropriate educational policy approach should be to address inequalities

in human capital with a microeconomic focus developing macro-level equity enhancing policies

that may lead to improvement of overall quality of education for a particular country. This may

require further research focusing more at evaluating educational quality outcomes rather than

attainments of education and training based programmes of different schools. This, in turn, will

help in finding out the best possible alternatives for fostering the skills of individuals leading to

reduction in human capital inequality.

4.4 Cross-Country Analysis

In this section, we consider running cross-country regressions as robustness checks as

some of the country specific empirical models may have an issue of limited degree of freedom.

While this type of a cross-country analysis may be helpful in increasing the degree of freedom, it

113

has the tendency to ignore the individual country heterogeneity by treating the data as one big

sample.64

In what follows we carry out two cross country analyses as robustness checks. Firstly,

we estimate regression models with human capital inequality as our dependent variable and a

single independent variable chosen from school and teacher attributes mentioned earlier in the

methodology section. Second, we increase the number of independent variables one at a time in

our estimations leading to a specification that includes all variables from the country-specific

regressions. Appendix I includes the results for our cross-country estimation exercise. Tables 1-7

include the results for individual variables analyses. Table 8 contains results for the second

cross-country analysis.

The first individual variable regression model controls for student-teacher ratio as a

determinant of human capital inequality. This model is followed by individual variable models

(2-7) that control for teacher’s experience teaching mathematics, percentage of students from

economically disadvantaged backgrounds, location of school, enrollment in the twelfth grade,

teacher’s job satisfaction and percentage of students with language of test as their native

language respectively. Overall, we find that the coefficients of independent variables in our

individual cross-country regressions remain statistically significant as in country-wise

regressions. Hence, these results support our country-specific analysis. However, this common

set of regression analysis at a cross county level hides the richness of analysis performed at the

64

We do acknowledge the limitations associated with cross-country estimations. A cross-country analysis is unable

to capture the temporal nature of variables. It is limited to analyzing the behaviour of variables at a specific point

rather than over a period of time. In addition, it may be fraught by an omitted variable bias due to ignoring the

country-specific variables such as cultural factors associated with a particular country. Our data set does not include

information on the cultural or social aspects of a country we acknowledge this as a limitation of our data set rather

than the technique itself.

114

disaggregated level as it remains unable to reveal the differences in the determinants of

inequality among countries.

In the interest of succinct argument, we provide explanations of a few selected models.

Appendix I, Table 1 shows that an increase in student-teacher ratio leads to a significant increase

in human capital inequalities at the level of schools. This evidence is similar to our country-

specific regression results for Lebanon, Netherlands and Sweden. Furthermore, in Table 2 we

examine the association between human capital inequality and teacher’s experience teaching

mathematics. These results show similarity to our country-specific cases for Armenia and Italy.

The third model presented in Table 3, exhibits an inverse and significant association between

percentage of students from economically disadvantaged backgrounds and human capital

inequality, while in Table 4 we find schools located in urban areas are associated with lesser

variations in mathematics skills. In Table 5, we find that a larger school size leads to a higher

variation in skills which is similar to the country-specific estimations for Iran. However, it is in

contrast to the evidence for Netherlands where the association is negative and significant. The

evidence in cross-country analysis mostly supports the arguments provided earlier regarding the

importance of these variables employed as possible determinants of inequality in the country-

wise estimations.

In the second stage we estimate six cross country regression models. Appendix I, Table 8

presents the results for these models. Each column 1-7 of Table 8 includes the results for a

regression model. In the first regression we examine the impact of student-teacher ratio and

teacher’s experience teaching mathematics on human capital inequality. Consistent with our

country-specific and cross-country analysis we find a positive and significant association

between student-teacher ratio and variation in skills. Moreover, the results indicate a negative

115

and significant impact of teacher’s experience teaching mathematics on human capital inequality.

In the second regression, we control for percentage of students from economically disadvantaged

backgrounds and find a negative and statistically significant association with human capital

inequality. The third regression controls for location of school and shows that rural based schools

located in less populated regions relative to schools in high population density locations are

positively and significantly associated with dispersion in mathematics skills across countries.

The evidence obtained here is similar to Lebanon and Russia implying that schools in bigger

cities face relatively fewer constraints regarding quality of teachers, financial resources and

physical infrastructure issues in comparison to schools that are located in rural or less populated

areas.

In the fourth model, greater number of students enrolled at secondary school level is

associated with significant increase in the human capital inequality. In the light of this evidence

the cross-country exercise performed serves its purpose as it helps in developing a broad idea

regarding the association of these variables with human capital inequality at an aggregated level.

However, compared to our country-wise disaggregated analyses which identifies country-

specificity of inequality a cross-country analysis of inequality lacks information about the

country-specific dimensions of inequality.

4. 5. Concluding Remarks

This study analyzes the composition and determinants of within and between human

capital inequalities. For this purpose, we use a unique micro data set on TIMSS 2008 that has

information on raw advanced mathematics test scores for students in final year of secondary

school. We first construct human capital index for 10 participating countries in our sample by

employing generalized entropy measures, as they have the advantage of being considered

116

additively decomposable into within-group and between-group inequalities. Our results indicate

that intra-country disparities overshadow inter-country educational quality dispersion levels as

more than half of inequality in skills is due to within-country differences. These results are in

contrast to earlier work on income and educational attainment inequalities which shows that

between-inequality is greater than within-inequality component (Li et al, 1998; Castelló and

Doménech, 2000).

We therefore extend our analysis and take the approach of examining the structure and

factors associated with human capital inequalities within each of the individual countries in our

sample. We consider that one fruitful way of addressing the issue of inequality in educational

quality is to develop strategies explicit to a specific country. Our country-wise results highlight

that school and teacher attributes are among important determinants of human capital inequality.

More specifically, the composition and causes of inequality in human capital are different for

each of the individual countries in our sample.

Consequently, our study provides a comprehensive and deeper understanding of

decomposed skill-inequality patterns at various levels using a comparable cross-country micro

data. This decomposition exercise identifies the possible factors associated with inequality in

skills at the level of schools for each country. From a macroeconomic policy perspective,

strategies targeting higher economic growth should not only be based on increasing educational

attainments, but also incorporate policies which enhance equity in educational achievements.

These policies should be formed on the basis of the evidence unearthed at the “grassroots” level

– in this case schools and educational institutions. In particular, policy makers interested in

reducing human capital inequalities should focus on school and teacher related attributes rather

than student, family or societal attributes.

117

Chapter 5

Summary and Conclusions

This chapter provides a summary of objectives and the main outcomes of the two essays

that investigate issues associated with human capital, technology and inequality. The first essay

analyzes the association between qualitative measures of human capital and direct constructs of

technology adoption and diffusion. The second essay examines human capital inequality by

constructing an inequality index which unearths the composition and determinants of within and

between- country inequalities at a microeconomic level. Furthermore, we discuss some of the

policy recommendations and conclude by outlining the limitations and directions for future

research.

The first essay draws inspiration from the literature on human capital, technology and

growth which throws light on the contribution of human capital in productivity growth as an

input as well as a facilitator of technology adoption in the process of production (Nelson and

Phelps, 1966; Lucas, 1988; Romer, 1990; Mankiw et al, 1992; Aghion and Howitt, 1998; Barro,

1998; Madsen, 2014). In this essay we employ usage intensity and usage lags of technology

based on direct measures of technology as constructs of technology. These measures are

introduced by Comin et al. (2008) and Comin and Mestieri (2013) who suggest that direct

measures are better able to account for technology dynamics and income differences across

countries. To include a measure of human capital in this study, we seek further inspiration from

Hanushek and Kimko (2000) and Hanushek and Woessmann (2012). They employ international

test scores as a proxy of human capital and show that the association between cognitive skills

and economic growth is robust to different specifications.

118

Following the above-mentioned literature, we focus on direct measures of technology and

qualitative measures of human capital and contribute in the literature by examining the

association between human capital and technology adoption and diffusion. Our analysis

incorporates different dimensions of human capital such as learning by doing, cognitive skills,

average years of schooling and life expectancy. We employ these measures of human capital and

suggest that specific types of human capital may be more or less relevant in facilitating adoption

and diffusion processes depending on the type of technology. To examine this hypothesis, we

form two panels of mathematics and science tests scores as qualitative measures of human

capital based on TIMSS for the period 1964-2003. In addition, we include quantitative measures

of human capital as average years of schooling from Barro and Lee (2010). We obtain

information on direct measures of technology adoption and diffusion from CHAT data set due to

Comin and Hobijn (2009).

As we examine the association between human capital and technology, the results

support our hypothesis regarding the technology-specific, nature of link between human capital

and technology. We find that the learning-by-doing dimension of human capital is the most

important determinant of adoption and diffusion of technology. On the other hand, evidence for

cognitive skills is weaker for usage intensity and usage lags of technology followed by average

of schooling and life expectancy as other dimensions of human capital. Based on these findings

we argue that the concept of human capital to facilitate the adoption and diffusion of different

types of technologies is diverse and neither qualitative nor quantitative measures of human

capital can completely account for it.

In the second study we examine qualitative dimension of human capital from a

microeconomic perspective. For this purpose we construct human capital inequality index that

119

reveals the structure, composition and factors associated with within and between sub-group

inequalities at a microeconomic level, which macroeconomic cross-country studies hitherto

remain unable to unearth. This study draws motivation from the literature which employs

education as one of the measures of human wellbeing and provides theoretical and empirical

evidence that educational inequality influences income inequality, economic growth and leads to

differences in productivity (Sen, 1979; 1985; 1987; Gloom and Ravikumar, 1992; Galor and

Tsiddon, 1997; Park, 1996; Acemoglu and Dell, 2010). Similar to the first essay as mentioned

before, we motivate the second study and use cognitive skills as measures of human capital from

Hanushek and Kimko (2000), Hanushek and Woessmann (2012) and Woessmann (2014).

Further motivation stems from the literature that employs international test scores to reveal and

examine the variations in human capital and develops international comparisons among different

countries (Sahn and Younger, 2007; Freeman et al, 2010; Oppedisano and Turati, 2011).

Developing on these strands of literature, the second study uses qualitative measures of

education based on TIMSS (2008) and employs Generalized Entropy Measures to construct

inequality index, which decompose within and between sub-group inequalities at three sub-

levels, i.e., cross-country, country and school level. Our decomposed inequality index reveals

that within-country inequality overshadows between-country inequality. This implies that

disparities in human capital stem from the differences in educational quality within a country

rather than between countries. Therefore, we suggest that the pattern and factors associated with

inequality in educational quality are specific to a country and human capital inequality has a

country-specific dimension.

Based on this evidence we further decompose inequality at the level of each country by

considering sub-groups of students from each of the schools and develop country-specific case

120

studies. At the country level our index reveals that within-school inequality dominates the

between-school inequality component for all countries. Given these findings we aim to identify

the determinants of inequality at the level of schools and develop case-by-case country-wise

regressions. Our regressions employ decomposed school level inequalities with school and

teacher attributes as possible determinants of inequality in human capital. Our analysis reveals

that these attributes are one of the important factors influencing human capital inequality;

however, the specific attributes differ across countries. Thus, the decomposition approach that

we employ in this study reveals that the composition and determinants of educational quality are

different across countries; therefore, human capital inequality is country-specific.

The empirical analysis undertaken in the first study exploring the skill-technology nexus

suggests that qualitative rather than quantitative measures of human capital are more relevant for

adoption and diffusion of wider range of technologies. In terms of policy implications, this

implies that educational policies should recognize the importance of learning outcomes in the

form improvement of cognitive skills instead of overemphasizing access to schools. Earlier

evidence shows that countries considered relatively more educated with greater proportion of

school going population or higher educational attainment levels lag behind in terms of economic

growth (Hanushek and Woessmann, 2009). The poor economic performance of these countries is

attributed to the differences between the quality and quantity of education. Estimations for lower

middle income countries for the year 2015 show that achieving universal enrollment leads to a

gain of 206%, however, a simultaneous access to basic skills results in a 1302% gain in GDP for

these economies (Hanushek and Woessmann 2015). This indicates that gains in economic

growth attributed to mere achievement of universal education are lesser than the gains which

accompany improvement in basic skills.

121

Hence, educational policies targeting improvement in learning outcomes impact upon the

skill development of each country’s workforce. Over a period of time the learning capital of the

nation will develop and gradually more skilled and educated workers enter the labour force.

Lower human capital barriers in the form of better skilled and more educated labour force will

improve diffusion prospects of technologies for an economy. Moreover, when the workers

employ these technologies, it raises the overall productivity of the labour force. As technology

embodies productivity, it thereby increases the country’s productivity growth and improves the

growth prospects of an economy.

In the first essay, we also provide evidence suggesting that a specific skill contributes in

the adoption of a particular technology. For example, our empirical evidence shows that generic

skills based on mathematics test scores contribute significantly in the adoption of technologies

from telecommunications and information sector such as cable TV, computer and internet.

Evidence from previous work documents that higher adoption and diffusion of these network

based broadband technologies is associated with rise in GDP per capita of 2.7-3.9% (Czernich,

2011). In the light of this background the pragmatic policy approach requires investments in

mathematical and analytical skills of a country’s population. This will result in improved

diffusion prospects of such technologies later contributing to higher economic growth.

Another example stems from our finding that mathematics based skills facilitate the

usage intensity of electricity production. Our data set categorizes this technology as a general

purpose technology which is used in almost all sectors of an economy. There is plethora of

research which suggests that electricity production and consumption are strongly associated with

the growth prospects of a country (Masih and Masih, 1996; Shiu and Lam, 2004; Yoo, 2005;

Czernich et al, 2011; Iyke, 2015). The skill-technology implication that emerges from this

122

evidence is that a nation should ensure presence of generic skills among its population to

promote economic growth. Intuitively, a skilled population contributes in increasing the

consumption of electricity as they substitute tasks performed manually by employing

technologies that require electricity. On the production side of the economy, more skilled

workers will enter the workforce in the form of technicians and engineers with the passage of

time. This skilled human capital encourages automation and results in increased supply of

electricity through better services such as maintenance of the grid network and use of newer

technologies in electricity production. This process involving skilled human capital raises

productivity levels and contributes in higher economic growth. Given this evidence, a nation

aiming to overcome constraints on energy consumption and production should invest in

improving generic skills of its population as it enhances the usage intensity of electricity which

later electricity stimulates economic growth.

This first essay also has implications which develop from the weak skill-technology

associations between qualitative measures of human capital and agricultural technologies. This

suggests that presence of skilled human capital is not an important prerequisite for countries

aiming to increase the diffusion of agricultural technologies. Earlier evidence for these

technologies suggests that they are mostly associated with developing economies and their

direction of diffusion is more from East to West rather than North to South. This feature makes

these technologies geography-specific (Diamond, 1998).65

In this case where spatial factors are

involved technological knowledge is more easily transmitted between countries or adopters of

technologies which are located close to each other compared to countries in far off locations.

Hence, human capital in the form of skills or years of schooling is no longer the main driver of

65

For details, See, Jared Diamond (1998), Guns and Germs and Steel.

123

diffusion and adoption of such technologies rather acquisition of knowledge which constitutes

interactions with other agents becomes important (Comin and Hobjin, 2013). Studies that have

examined this idea suggest that location matters as in the agricultural sector a neighbors’

adoption decision affects the agents’ own decision to adopt a particular technology (Foster and

Rosenweig 1995; Conley and Udry, 2010).

Hence, our finding of a weak skill technology association in agricultural sector has

implications for countries that are keen to increase the adoption of agricultural technologies.

These countries should formulate policies that reduce the technological knowledge constraints by

developing proper channels to facilitate the interaction among the current and potential adopters

of a technology to increase the flow of knowledge between the two agents. This entails providing

platforms or forums to improve their regular interactions where agents can meet and share their

knowledge and experiences regarding different technologies or use of agricultural production

methods. Greater access to agricultural support services can provide farmers or agents with latest

information regarding a particular technology such as a new pesticide or water based technology

helping them in protecting their crops against a pest or a natural hazard such as draught.

Enhanced access to network based technologies such as internet, cell phones and TV/cable TV

can also increase rapid flow of global information on current technologies that are cost effective

and raise the productivity of the area under cultivation. All these forms of interactions will add to

the knowledge pool of the workforce involved in agricultural sector thereby raising the overall

productivity of the farms leading to higher economic growth.

Further implications develop from the weaker evidence of association between skills and

usage lags of technologies. This implies that the technological diffusion process is not only

inhibited by lack of skilled human capital but other factors also determine the distance of a

124

nation in terms of technological adoption relative to the technology leader. Extant evidence in

the literature suggests that diffusion of technologies is also restricted due to presence of

inadequate institutions. The main premise of this argument is that poor institutions are unable to

protect the rights of the adopters or the income generated by employing a particular technology

which makes the adopters reluctant to invest in adopting new technologies leading to lower

diffusion of technology (Comin and Mestieri, 2013). In addition, evidence shows that diffusion

of technology is hindered by the political parties, lobbies or other political agents as adoption of

certain technologies reduces the political power of these groups (Acemoglu and Robinson,

2000). Hence, our evidence for diffusion of technology perhaps entails reducing barriers in the

form of presence of sound institutions which are able to protect and safeguard the rights and

income of potential adopters of a technology. This can be done by designing appropriate policies

that ensure development of sound institutions which will result in higher diffusion of technology

leading to increase in individual incomes accompanied with a rise in overall productivity of an

economy.

The second essay reveals the composition and determinants of human capital inequality

at grass root level and shows that variations in human capital are associated with differences in

educational quality within rather than across countries. As we explore the within-country

distribution of human capital at the level of schools our analysis suggests that within-school

inequality overshadows between-school inequality and school and teacher characteristics are

important determinants of variations in human capital. The implication of this finding leads to a

policy debate whether the educational decision-making authority should be centralized or

delegated to schools as the institutions imparting skills and the source of micro-level inequality

in human capital. One plausible policy approach could be that schools should be empowered and

125

local autonomy in educational decision making should be upheld as local decision makers are

well aware of the demands on school and their service capacity. In this situation, the role of

central authorities will be limited to keeping a watchful eye on the overall educational outcomes

of schools. Another policy option could be that decisions regarding administration and

management related issues could be the sole discretion of the schools. However, in the context of

academic curriculum standardization at a macro-level can be must be considered.

Moreover, studies show that the level of economic development and presence of sound

institutions also impacts upon the skill outcomes of schools. Autonomous decision making on

behalf of schools leads to a positive impact on the educational outcomes of schools in developed

economies with higher economic prosperity and strong institutional framework levels. This type

of educational decision making may impact the skill outcomes of schools negatively for

developing economies with lower economic development and weak institutions (Hanushek et al,

2011). Overall, the broader implication that emerges from this debate is that educational policies

cannot be generalized across countries as each country has specific economic and institutional

framework. This implication is in line with our findings which suggest that in order to address

inequality in human capital country-specific policies are required. Hence, policies should be

formulated keeping in mind a country’s own economic and institutional composition which can

minimize the negative impact on the skill outcomes of schools.66

Finally, our analysis also has implications for labour market outcomes. We suggest that

variation in human capital will be associated with the level of economic prosperity of a nation

due to the link that exists between the educational outcomes and future labour marker income

66

We control of intuitional quality in our robustness checks conducted in Chapter 3, Section 3.5. Our results show

variable coefficients for the proxies of institutions. We do acknowledge in the particular section and mention it here

as well that due to data limitations for measures of institutional quality beginning from early 60’s the measures

included may lack in capturing the soundness of institutions appropriately.

126

levels. This underscores the importance of our analysis suggesting deeper understanding of these

variations at a grass root level. The implication that emerges from this is that nations that are

aiming to reduce future disparities in labour market income levels should design educational

policies that can reduce current variations in skills resulting in a higher level of economic

prosperity.

The thesis has certain limitations and also provides some directions for further research.

Our first study examines the contribution of human capital in technology adoption and diffusion,

and develops an analysis employing standard cross-country macroeconomic variables across

sectors. Due to data constraints we are unable to develop sector or technology-specific case

studies which constitute specific set of determinants based on the technology or sector under

discussion. Hence, our empirical evidence may lack in providing sector or technology-specific

interpretations regarding the players involved in the adoption and diffusion of technology. Based

on this, we suggest that future research may develop specific case studies which constitute

technology or sector-specific set of determinants across sectors of an economy. This perhaps

requires data collection based on primary survey methods explicitly focusing on identify the

factors influencing the pace and adoption of a specific technology across countries. Such an

exercise will result in developing coherent cross-country macroeconomic data sets allowing for

panel data analysis constituting a large sample of countries.

The second study constitutes an exploration of composition and causes of inequality for a

set of 10 countries that participated in the TIMSS 2008 advanced mathematics tests. As we aim

to uncover the country-specific determinants of variations in human capital perhaps this large

number of countries has limited us in providing a detailed extensive review of each country.

However, our analysis does reveal that the composition and determinants of human capital

127

inequality are country-specific and the practical approach is a case-by-case grass roots level

exploration rather than a cross-country analysis employed in earlier literature. Thus, a future

direction for research perhaps is to select one country at a time from this sample and develop a

more comprehensive country-wise analysis which may be even more useful to identify in detail

the country-specific dimensions of human capital inequality.

128

Bibliography

Acemoglu, D. (1998). Why do new technologies complement skills? Directed technical change

and wage inequality. The Quarterly Journal of Economics, 113(4), 1055-1089.

Acemoglu, D. (2002). Directed technical change. The Review of Economic Studies, 69(4), 781-

809.

Acemoglu, D., & Robinson, J. A. (2002). The political economy of the Kuznets Curve. Review of

Development Economics, 6(2), 183-203.

Acemoglu, D., Johnson, S., & Robinson, J. A. (2005). Institutions as a fundamental cause of

long-run growth. Handbook of economic growth, 1, 385-472.

Acemoglu, D., & Dell, M. (2010). Productivity differences between and within

countries. American Economic Journal: Macroeconomics, 2(1), 169-88.

Acemoglu, D., Akcigit, U., Alp, H., Bloom, N., & Kerr, W. R. (2013). Innovation, reallocation

and growth (No. w18993). National Bureau of Economic Research.

Acemoglu, D., & Zilibotti, F. (2001). Productivity differences. The Quarterly Journal of

Economics, 116(2), 563-606.

Aghion, P., & Howitt, P. (1992). A Model of Growth through Creative

Destruction. Econometrica, 60(2).

Aghion, P., Howitt, P., Brant-Collett, M., & García-Peñalosa, C. (1998). Endogenous growth

theory. MIT press.

Aghion, P., Caroli, E., & Garcia-Penalosa, C. (1999). Inequality and economic growth: the

perspective of the new growth theories. Journal of Economic literature, 37(4), 1615-

1660.

Aghion, P., Boustan, L., Hoxby, C., & Vandenbussche, J. (2005). Exploiting States’ Mistakes to

Identify the Causal Impact of Higher Education on Growth.

Aghion, P., Meghir, C., & Vandenbussche, J. (2005). Growth, Distance to Frontier and

Composition of Human Capital (No. 4860). CEPR Discussion Papers.

Aghion, P., Alesina, A., & Trebbi, F. (2007). Democracy, Technology, and Growth (No. 13180).

National Bureau of Economic Research, Inc.

Aghion, P., Alesina, A. & Trebbi, F. (2008), Democracy, technology, and growth, in E.

Helpman, ed., Institutions and Economic Performance, Harvard University Press.

Aghion, P., & Howitt, P. W. (2008). The economics of growth. MIT press.

Ainsworth, M., & Over, M. (1994). AIDS and African development. The World Bank Research

Observer, 9(2), 203-240.

129

Aitken, B. J., & Harrison, A. E. (1999). Do domestic firms benefit from direct foreign

investment? Evidence from Venezuela. American Economic Review, 89(3), 605-618.

Akabayashi, H., & Nakamura, R. (2014). Can small class policy close the gap? An empirical

analysis of class size effects in Japan. The Japanese Economic Review, 65(3), 253-281.

Alesina, A., & Perotti, R. (1996). Income distribution, political instability, and

investment. European Economic Review, 40(6), 1203-1228.

Alesina, A., & Rodrik, D. (1994). Distributive politics and economic growth. The Quarterly

Journal of Economics, 109(2), 465-490.

Angrist, J. D., & Lavy, V. (1999). Using Maimonides' rule to estimate the effect of class size on

scholastic achievement. The Quarterly Journal of Economics, 114(2), 533-575.

Archer, S. (2016). Late Afternoon Concurrent Sessions: Training and Education: Presentation:

Pilot Study: Secondary Aviation/Aerospace Education Organization Design Survey.

Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo

evidence and an application to employment equations. The Review of Economic

Studies, 58(2), 277-297.

Arrow, K. J. (1962). The economic implications of learning by doing. The Review of Economic

Studies, 29(3), 155-173.

Atkinson, A. B., & Stiglitz, J. E. (1969). A new view of technological change. The Economic

Journal, 79(315), 573-578.

Axtell, B., & Bowers, J. (1972). Rural urban effects on the common entrance

examination. TEDRO RP, 104.

Balogun, T. A. (1982). Improvisation of science teaching equipment. Journal of the Science

Teachers Association, 20(2), 72-76.

Banerjee, A. V., & Duflo, E. (2003). Inequality and growth: What can the data say?. Journal of

Economic Growth, 8(3), 267-299.

Barro, R. J. (2001). Human capital and growth. American economic review, 91(2), 12-17.

Barro, R. J., & Sala-I-Martin, X. (1995). Economic growth theory. New York: Mac Graw-Hill.

Barro, R. J. (1998). Human capital and growth in cross-country regressions. Harvard University.

Barro, R. J. (2000). Inequality and Growth in a Panel of Countries. Journal of Economic

Growth, 5(1), 5-32.

Barro, R. J., & Lee, J. W. (2001). International data on educational attainment: updates and

implications. Oxford Economic papers, 53(3), 541-563.

Barro, R. J., & Lee, J. W. (2010). A new data set of educational attainment in the world, 1950–

2010. Journal of development economics, 104, 184-198.

Barro, R. J., & Lee, J. W. (2013). A new data set of educational attainment in the world, 1950–

2010. Journal of development economics, 104, 184-198.

130

Barro, R. J. (2013). Health and economic growth. Annals of Economics and Finance, 14(2), 329-

366.

Basu, S. (1996). Procyclical productivity: increasing returns or cyclical utilization?. The

Quarterly Journal of Economics, 111(3), 719-751.

Basu, S., & Weil, D. N. (1998). Appropriate technology and growth. The Quarterly Journal of

Economics, 113(4), 1025-1054.

Benhabib, J., & Spiegel, M. M. (1994). The role of human capital in economic development

evidence from aggregate cross-country data. Journal of Monetary economics, 34(2), 143-

173.

Blanden, J., & McNally, S. (2015). Reducing inequality in education and skills: Implications for

economic growth. EENEE Analytical report.

Blatchford, P., Russell, A., Bassett, P., Brown, P., & Martin, C. (2007). The effect of class size

on the teaching of pupils aged 7–11 years. School Effectiveness and School

Improvement, 18(2), 147-172.

Blundell, R., & Etheridge, B. (2010). Consumption, income and earnings inequality in

Britain. Review of Economic Dynamics, 13(1), 76-102.

Borissov, K., & Lambrecht, S. (2009). Growth and distribution in an AK-model with

endogenous impatience. Economic Theory, 39(1), 93-112.

Branstetter, L. (2006). Is foreign direct investment a channel of knowledge spillovers? Evidence

from Japan's FDI in the United States. Journal of International economics, 68(2), 325-

344.

Breen, R., Luijkx, R., Müller, W., & Pollak, R. (2009). Non-persistent inequality in educational

attainment: Evidence from eight European countries. American Journal of

Sociology, 114(5), 1475-1521.

Brown, G., Micklewright, J., Schnepf, S. V., & Waldmann, R. (2007). International surveys of

educational achievement: how robust are the findings?. Journal of the Royal statistical

society: series A (statistics in society), 170(3), 623-646.

Bryk, A. S., & Driscoll, M. E. (1988). The High School as Community: Contextual Influences

and Consequences for Students and Teachers.

Burnside, C., Eichenbaum, M., & Rebelo, S. (1995). Capital utilization and returns to

scale. NBER macroeconomics annual, 10, 67-110.

Buzacott, J.A. 1982. Scale in production system. New York: Pergamon

Caselli, F. (2005). Accounting for cross-country income differences. Handbook of economic

growth, 1, 679-741.

Caselli, F., & Coleman, W. J. (2001). Cross-country technology diffusion: The case of

computers. American Economic Review, 91(2), 328-335.

131

Caselli, F., Coleman, I. I., & John, W. (2006). The world technology frontier. American

Economic Review, 96(3), 499-522.

Castelló-Climent, A. (2010). Inequality and growth in advanced economies: an empirical

investigation. The Journal of Economic Inequality, 8(3), 293-321.

Castelló-Climent, A. (2010). Channels through which human capital inequality influences

economic growth. Journal of Human Capital, 4(4), 394-450.

Castelló-Climent, A. (2010). Inequality and growth in advanced economies: an empirical

investigation. The Journal of Economic Inequality, 8(3), 293-321.

Castelló, A., & Doménech, R. (2002). Human capital inequality and economic growth: some

new evidence. The economic journal, 112(478).

Castelló-Climent, A., & Doménech, R. (2014). Human capital and income inequality: some facts

and some puzzles. Retrieved from BBVA Research https://www. bbvaresearch. com/wp-

content/uploads/migrados/WP_1228_tcm348-430101. pdf.

Caswell, M., Fuglie, K., Ingram, C., Jans, S., & Kascak, C. (2001). Adoption of agricultural

production practices: lessons learned from the US Department of Agriculture Area

Studies Project. (Agricultural Economic Report, No. 792). Washington, DC: US

Department of Agriculture, Economic Research Service.

Causa, O., & Chapuis, C. (2009). Equity in Student Achievement Across OECD Countries.

Economic Department Working Papers, (708).

Ceppa, D. P., Kosinski, A. S., Berry, M. F., Tong, B. C., Harpole, D. H., Mitchell, J. D., ... &

Onaitis, M. W. (2012). Thoracoscopic lobectomy has increasing benefit in patients with

poor pulmonary function: a Society of Thoracic Surgeons Database analysis. Annals of

surgery, 256(3), 487.

Chakraborty, S., & Das, M. (2005). Mortality, human capital and persistent inequality. Journal

of Economic growth, 10(2), 159-192.

Champernowne, D. (1961). A dynamic growth model involving a production function. In The

Theory of Capital (pp. 223-244). Palgrave Macmillan, London.

Champernowne; A (1963). Dynamic growth model involving a production function. In: F. A.

Lutz and D. C. Hague (eds), The theory of capital, New York: Macmillan

Checchi, D. (2004). Does educational achievement help to explain income

inequality?. Inequality, growth and poverty in an era of liberalization and globalization.

Cingano, F. (2014). Trends in income inequality and its impact on economic growth.

Clarke, G. R. (1995). More evidence on income distribution and growth. Journal of Development

Economics, 47(2), 403-427.

Comin, D., & Hobijn, B. (2004). Cross-country technology adoption: making the theories face

the facts. Journal of Monetary Economics, 51(1), 39-83.

132

Comin, D., & Hobijn, B. (2007). Implementing technology (No. w12886). National Bureau of

Economic Research.

Comin, D., Hobijn, B., & Rovito, E. (2008). Technology usage lags. Journal of Economic

Growth, 13(4), 237-256.

Comin, D. A., & Hobijn, B. (2009). The CHAT dataset (No. w15319). National Bureau of

Economic Research.

Comin, D., & Hobijn, B. (2009). Lobbies and technology diffusion. The Review of Economics

and Statistics, 91(2), 229-244.

Comin, D., & Mestieri, M. (2014). Technology diffusion: measurement, causes, and

consequences. In Handbook of economic growth (Vol. 2, pp. 565-622). Elsevier.

Comin, D. A., Dmitriev, M., & Rossi-Hansberg, E. (2012). The spatial diffusion of

technology (No. w18534). National Bureau of Economic Research.

Comin, D., & Mestieri, M. (2013). Technology diffusion: Measurement, causes and

consequences. NBER Working Paper. 19052.

Conley, T., & Udry, C. (2001). Social learning through networks: The adoption of new

agricultural technologies in Ghana. American Journal of Agricultural Economics, 83(3),

668-673.

Conley, T. G., & Udry, C. R. (2010). Learning about a new technology: Pineapple in

Ghana. American economic review, 100(1), 35-69.

Cowell, F. A., & Kuga, K. (1981). Additivity and the entropy concept: an axiomatic approach to

inequality measurement. Journal of Economic Theory, 25(1), 131-143.

Czernich, N., Falck, O., Kretschmer, T., & Woessmann, L. (2011). Broadband infrastructure and

economic growth. The Economic Journal, 121(552), 505-532.

Day, J. D., Metes, D. M., & Vodovotz, Y. (2015). Mathematical modeling of early cellular innate

and adaptive immune responses to ischemia/reperfusion injury and solid organ all

transplantation. Frontiers in immunology, 6, 484.

Deininger, K., & Squire, L. (1998). New ways of looking at old issues: inequality and

growth. Journal of development economics, 57(2), 259-287.

Deininger, K. & Olinto, P. (2000). Asset Distribution, Inequality, and Growth, World Bank

Development Research Group Working Paper, No.2375.

Diamond, J. M. (1998). Guns, germs and steel: a short history of everybody for the last 13,000

years. Random House.

Doménech, R. (2006). Human capital in growth regressions: how much difference does data

quality make?. Journal of the European Economic Association, 4(1), 1-36.

De la Fuente, A. and Doménech, R., 2006. Human capital in growth regressions: How much

difference does data quality make? Journal of the European Economic Association, 4(1),

pp 1-36.

133

Deutsch, J., & Silber, J. (1999). Inequality decomposition by population subgroups and the

analysis of inter-distributional inequality. In Handbook of income inequality

measurement(pp. 363-403). Springer, Dordrecht.

Dobbie, W., & Fryer Jr, R. G. (2011). Are high-quality schools enough to increase achievement

among the poor? Evidence from the Harlem Children's Zone. American Economic

Journal: Applied Economics, 3(3), 158-87.

Elo, I. T., & Preston, S. H. (1996). Educational differentials in mortality: United States, 1979–

1985. Social science & medicine, 42(1), 47-57.

Else-Quest, N. M., Hyde, J. S., & Linn, M. C. (2010). Cross-national patterns of gender

differences in mathematics: a meta-analysis. Psychological bulletin, 136(1), 103.

Feder, G., & Slade, R. (1984). The acquisition of information and the adoption of new

technology. American Journal of Agricultural Economics, 66(3), 312-320.

Makuc, F. J., Kleinman, J., & Cornoni-Huntley, J. (1989). National Trends in Educational

Differences in Mortality,“. American Journal of Epidemiology.

Feng, C. G., Lau, T. Y., Atkin, D. J., & Lin, C. A. (2009). Exploring the evolution of digital

television in China: An interplay between economic and political interests. Telematics

and Informatics, 26(4), 333-342.

Ferreira, F. H., & Gignoux, J. (2013). The measurement of educational inequality: Achievement

and opportunity. The World Bank Economic Review, 28(2), 210-246.

Finn, J. D., & Achilles, C. M. (1990). Answers and questions about class size: A statewide

experiment. American Educational Research Journal, 27(3), 557-577.

Flachaire, E., García-Peñalosa, C., & Konte, M. (2014). Political versus economic institutions in

the growth process. Journal of Comparative Economics, 42(1), 212-229.

Forbes, K. J. (2000). A Reassessment of the Relationship between Inequality and

Growth. American economic review, 90(4), 869-887.

Fogel, R. W. (1994). Economic growth population theory and physiology: the bearing of long-

term processes on the making of economic policy. American Economic, 84(3), 369-95.

Foster, J. E. (1983). An axiomatic characterization of the Theil measure of income

inequality. Journal of Economic Theory, 31(1), 105-121.

Foster, A. D., & Rosenzweig, M. R. (1995). Learning by doing and learning from others: Human

capital and technical change in agriculture. Journal of political Economy, 103(6), 1176-

1209.

Freeman, R. B., Machin, S., & Viarengo, M. (2010). Variation in educational outcomes and

policies across countries and of schools within countries (No. w16293). National Bureau

of Economic Research.

Galor, O., & Tsiddon, D. (1997). The distribution of human capital and economic

growth. Journal of Economic Growth, 2(1), 93-124.

134

Galor, O., & Weil, D. N. (1999). From Malthusian stagnation to modern growth. American

Economic Review, 89(2), 150-154.

Galor, O., & Weil, D. N. (2000). Population, technology, and growth: From Malthusian

stagnation to the demographic transition and beyond. American economic review, 90(4),

806-828.

Galor, O., 2011. Inequality, human capital formation and the process of development (No.

w17058). National Bureau of Economic Research.

Gerschenkron, A. (1962). Economic Backwards in Historical Perspective: A Book Essays.

Belknap Press of Harvard University Press.

Gibbons, S., & McNally, S. (2013). The effects of resources across school phases: A summary of

recent evidence. Centre for Economic Performance, LSE.

Glaeser, E. L., La Porta, R., Lopez-de-Silanes, F., & Shleifer, A. (2004). Do institutions cause

growth?. Journal of economic Growth, 9(3), 271-303.

Glomm, G., & Ravikumar, B. (1992). Public versus private investment in human capital:

endogenous growth and income inequality. Journal of political economy, 100(4), 818-

834.

Gregorio, J. D., & Lee, J. W. (2002). Education and income inequality: new evidence from

cross-country data. Review of income and wealth, 48(3), 395-416.

Grossman, M. (2006). Education and nonmarket outcomes. Handbook of the Economics of

Education, 1, 577-633.

Pappas, G., Queen, S., Hadden, W., & Fisher, G. (1993). The increasing disparity in mortality

between socioeconomic groups in the United States, 1960 and 1986. New England

journal of medicine, 329(2), 103-109.

Perotti, R. (1996). Growth, income distribution, and democracy: What the data say. Journal of

Economic growth, 1(2), 149-187.

Hallak, J. (1977). Planning the Location of Schools: An Instrument of Educational Policy.

Halter, D., Oechslin, M., & Zweimüller, J. (2014). Inequality and growth: the neglected time

dimension. Journal of economic Growth, 19(1), 81-104.

Hansen, L. P. (1982). Large sample properties of generalized method of moments

estimators. Econometrica: Journal of the Econometric Society, 1029-1054.

Hanushek, E. A. (1986). The economics of schooling: Production and efficiency in public

schools. Journal of economic literature, 24(3), 1141-1177.

Hanushek, E. A. (1997). Assessing the effects of school resources on student performance: An

update. Educational evaluation and policy analysis, 19(2), 141-164.

Hanushek, E. A., & Kimko, D. D. (2000). Schooling, labor-force quality, and the growth of

nations. American economic review, 90(5), 1184-1208.

135

Hanushek, E. A., & Woessmann, L. (2008). The role of cognitive skills in economic

development. Journal of economic literature, 46(3), 607-68.

Hanushek, E. A., & Woessmann, L. (2012). Do better schools lead to more growth? Cognitive

skills, economic outcomes, and causation. Journal of economic growth, 17(4), 267-321.

Hanushek, E. A., & Woessmann, L. (2015). The economic impact of educational

quality. Handbook of International Development and Education, 6-19.

Hanushek, E. A., Piopiunik, M., & Wiederhold, S. (2014). The value of smarter teachers:

International evidence on teacher cognitive skills and student performance (No. w20727).

National Bureau of Economic Research.

Hanushek, E. A., & Woessmann, L. (2017). School resources and student achievement: A review

of cross-country economic research. In Cognitive Abilities and Educational Outcomes

(pp. 149-171). Springer, Cham.

Harmon, C., Oosterbeek, H., & Walker, I. (2003). The returns to education:

Microeconomics. Journal of economic surveys, 17(2), 115-156.

Halter, D., Oechslin, M., & Zweimüller, J. (2014). Inequality and growth: the neglected time

dimension. Journal of economic Growth, 19(1), 81-104.

Hulten, C. R. (2000). Measuring innovation in the New Economy. Unpublished paper,

University of Maryland.

Husain, M., & Millimet, D. L. (2009). The mythical ‘boy crisis’?. Economics of Education

Review, 28(1), 38-48.

Iyke, B. N. (2015). Electricity consumption and economic growth in Nigeria: A revisit of the

energy-growth debate. Energy Economics, 51, 166-176.

Jamison, D. T., Lau, L. J., & Wang, J. (1998). Health’s contribution to economic growth, 1965-

90. Health, health policy and economic outcomes, 61-80.

Jorgenson, D. W., & Griliches, Z. (1967). The explanation of productivity change. The review of

economic studies, 34(3), 249-283.

JovANovIc, B. (1996). Learning by doing and the choice of technology. Econometrica, 64(6),

1299-1310.

Jude, C., & Levieuge, G. (2015). Growth effect of FDI in developing economies: the role of

institutional quality.Working papers 559, Banque de France.

Kanbur, R. (2000). Income distribution and development. Handbook of income distribution, 1,

791-841.

Knowles, S. (2005). Inequality and economic growth: The empirical relationship reconsidered in

the light of comparable data. The Journal of Development Studies, 41(1), 135-159.

Konstantopoulos, S. (2008). Do small classes reduce the achievement gap between low and high

achievers? Evidence from Project STAR. The Elementary School Journal, 108(4), 275-

291.

136

Konstantopoulos, S., & Sun, M. (2014). Are teacher effects larger in small classes?. School

Effectiveness and School Improvement, 25(3), 312-328.

Krueger, A. B. (1999). Experimental estimates of education production functions. The quarterly

journal of economics, 114(2), 497-532.

Krueger, A. B., & Lindahl, M. (2001). Education for growth: Why and for whom?. Journal of

economic literature, 39(4), 1101-1136.

Kuznets, S. (1955). Economic growth and income inequality. The American economic review, 1-

28.

Lahiri, R., Ding, J., & Chinzara, Z. (2017). Technology adoption, adaptation and

growth. Economic Modelling.

Lee, J. W., & Barro, R. J. (2001). Schooling quality in a cross–section of

countries. Economica, 68(272), 465-488.

Lee, V. E., Bryk, A. S., & Smith, J. B. (1993). Chapter 5: The organization of effective

secondary schools. Review of research in education, 19(1), 171-267.

Lee, V. E., & Smith, J. B. (1997). High school size: Which works best and for

whom?. Educational Evaluation and Policy Analysis, 19(3), 205-227.

Lee, V. E., & Loeb, S. (2000). School size in Chicago elementary schools: Effects on teachers'

attitudes and students' achievement. American Educational Research Journal, 37(1), 3-

31.

Lenzi, M., Vieno, A., Sharkey, J., Mayworm, A., Scacchi, L., Pastore, M., & Santinello, M.

(2014). How school can teach civic engagement besides civic education: The role of

democratic school climate. American journal of community psychology, 54(3-4), 251-

261.

Li, H., Squire, L., & Zou, H. F. (1998). Explaining international and intertemporal variations in

income inequality. The economic journal, 108(446), 26-43.

Lipsey, R. G., & Carlaw, K. I. (2004). Total factor productivity and the measurement of

technological change. Canadian Journal of Economics/Revue

canadienned'économique, 37(4), 1118-1150.

Li, W., & Konstantopoulos, S. (2017). Does class-size reduction close the achievement gap?

Evidence from TIMSS 2011. School Effectiveness and School Improvement, 28(2), 292-

313.

Li, X., Liu, X., & Parker, D. (2001). Foreign direct investment and productivity spillovers in the

Chinese manufacturing sector. Economic systems, 25(4), 305-321.

Li, H., Squire, L., & Zou, H. F. (1998). Explaining international and intertemporal variations in

income inequality. The economic journal, 108(446), 26-43.

Lleras-Muney, A., & Lichtenberg, F. R. (2002). The effect of education on medical technology

adoption: are the more educated more likely to use new drugs (No. w9185). National

Bureau of Economic Research.

137

Lochner, L., & Moretti, E. (2004). The effect of education on crime: Evidence from prison

inmates, arrests, and self-reports. American economic review, 94(1), 155-189.

Lucas Jr, R. E. (1988). On the mechanics of economic development. Journal of monetary

economics, 22(1), 3-42.

Madsen, J. B. (2014). Human capital and the world technology frontier. Review of Economics

and Statistics, 96(4), 676-692.

Mankiw, N. G., Romer, D., & Weil, D. N. (1992). A contribution to the empirics of economic

growth. The quarterly journal of economics, 107(2), 407-437.

Masih, A. M., & Masih, R. (1996). Energy consumption, real income and temporal causality:

results from a multi-country study based on cointegration and error-correction modelling

techniques. Energy economics, 18(3), 165-183.

Messinis, G., & Ahmed, A. D. (2013). Cognitive skills, innovation and technology

diffusion. Economic modelling, 30, 565-578.

Meyer, K. E., & Sinani, E. (2009). When and where does foreign direct investment generate

positive spillovers? A meta-analysis. Journal of International Business Studies, 40(7),

1075-1094.

Micklewright, J., & Schnepf, S. V. (2006). Inequality of learning in industrialised countries.(No.

2517). Institute for the Study of Labor (IZA).

Milanovic, B., & Yitzhaki, S. (2002). Decomposing world income distribution: Does the world

have a middle class?. Review of income and wealth, 48(2), 155-178.

Monk, D. H., & Haller, E. J. (1993). Predictors of high school academic course offerings: The

role of school size. American Educational Research Journal, 30(1), 3-21.

Nelson, R. R., & Phelps, E. S. (1966). Investment in humans, technological diffusion, and

economic growth. The American economic review, 56(1/2), 69-75.

North, D. C. (1990). Institutions, institutional change, and economic performance. Cambridge;

New York: Cambridge University Press.

Nye, B. A., Hedges, L. V., & Konstantopoulos, S. (2000a). Do the disadvantaged benefit more

from small classes? Evidence from the Tennessee class size experiment. American

journal of education, 109(1), 1-26.

Nye, B., Hedges, L. V., & Konstantopoulos, S. (2000b). The effects of small classes on academic

achievement: The results of the Tennessee class size experiment. American Educational

Research Journal, 37(1), 123-151.

Oppedisano, V., & Turati, G. (2011). What are the causes of educational inequalities and of their

evolution over time in Europe? Evidence from PISA.

Oppedisano, V., & Turati, G. (2015). What are the causes of educational inequality and of its

evolution over time in Europe? Evidence from PISA. Education Economics, 23(1), 3-24.

138

Organization for Economic Cooperation and Development (OECD). 2013. OECD skills outlook

2013: First results from the survey of adult skills. Paris: Organisation for Economic Co-

operation and Development.

Organization for Economic Cooperation and Development (OECD). 2013. PISA 2012 results:

What students know and can do – Student performance in mathematics, reading and s

Ostry, M. J. D., Berg, M. A., & Tsangarides, M. C. G. (2014). Redistribution, inequality, and

growth. International Monetary Fund.

Owoeye, J. S., & Yara, P. O. (2011). School location and academic achievement of secondary

school in Ekiti State, Nigeria. Asian social science, 7(5), 170.

Parente, S. L. (1994). Technology adoption, learning-by-doing, and economic growth. Journal of

economic theory, 63(2), 346-369.

Parente, S. L., & Prescott, E. C. (1994). Barriers to technology adoption and

development. Journal of political Economy, 102(2), 298-321.

Parente, S., & Prescott, E. (2004). A unified theory of the evolution of international income

levels. Federal Reserve Bank of Minneapolis. Research Department Staff Report 333.

Park, K. H. (1996). Educational expansion and educational inequality on income

distribution. Economics of education review, 15(1), 51-58.

Persson, T., & Tabellini, G. (1994). Is inequality harmful for growth?. The American Economic

Review, 600-621.

Piketty, T. (2015). About capital in the twenty-first century. American Economic Review, 105(5),

48-53.

Piketty, T., & Saez, E. (2013). Top incomes and the great recession: Recent evolutions and

policy implications. IMF economic review, 61(3), 456-478.

Pong, S. L., & Pallas, A. (2001). Class size and eighth-grade math achievement in the United

States and abroad. Educational evaluation and policy analysis, 23(3), 251-273.

Pritchett, L. (1997). Divergence, big time. Journal of Economic perspectives, 11(3), 3-17.

Prucha, F. P. (1979). The Churches and the Indian Schools, 1888-1912. University of Nebraska

Press, 901 N. 17th St., Lincoln, NE 68588-0520.

Rahman, A., Kamarulzaman, N. H., & Sambasivan, M. (2013). A study on organizational

culture, performance, and technological adoption behaviours of Malaysian food-

processing SMEs. Pertanika Journal of Social Sciences & Humanities, 21(spec. June),

231-256.

Rashidi, A. (2016). A mathematical model provides new insights into solid organ transplant

associated acute graft versus host disease. Clinical transplantation, 30(9), 1173-1177.

Riddell, W. C., & Song, X. (2012). The Role of Education in Technology Use and Adoption:

Evidence from the Canadian Workplace and Employee Survey.

139

Rodrik, D., Subramanian, A., & Trebbi, F. (2004). Institutions rule: the primacy of institutions

over geography and integration in economic development. Journal of economic

growth, 9(2), 131-165.

Rodrik, D. (2011). Unconditional convergence (No. w17546). National Bureau of Economic

Research.

Rodrik, D. (2013). Unconditional convergence in manufacturing. The Quarterly Journal of

Economics, 128(1), 165-204.

Rodríguez-Pose, A., & Ezcurra, R. (2009). Does decentralization matter for regional disparities?

A cross-country analysis. Journal of Economic Geography, 10(5), 619-644.

Rogers Everett, M. (1995). Diffusion of innovations. New York, 12.

Romer, P. M. (1990). Endogenous technological change. Journal of political Economy, 98(5,

Part 2), S71-S102.

Sahn, D. E., & Younger, S. D. (2007). Decomposing world education inequality. Cornell Food

and Nutrition Policy Program Working Paper No. 187.

Saint-Paul, G., & Verdier, T. (1993). Education, democracy and growth. Journal of development

Economics, 42(2), 399-407.

Sala-i-Martin, X. (2002). The disturbing" rise" of global income inequality (No. w8904).

National Bureau of Economic Research.

Sen, A. (1979). Personal utilities and public judgements: or what's wrong with welfare

economics. The economic journal, 537-558.

Sen, A. (1985). Commodities and Capabilities. Amsterdam: North Holland.

Sen, A. (1987). The standard of living. Cambridge University Press.

Schumacher, E. F.(1973): Small is Beautiful: Economics as if People Mattered. Blond & Briggs,

London.

Shavit, Y., & Blossfeld, H. P. (1993). Persistent Inequality: Changing Educational Attainment in

Thirteen Countries. Social Inequality Series. Westview Press, 5500 Central Avenue,

Boulder, CO 80301-2847.

Shin, I. (2012). Income inequality and economic growth. Economic Modelling, 29(5), 2049-

2057.

Shorrocks, A. F. (1980). The class of additively decomposable inequality

measures. Econometrica: Journal of the Econometric Society, 613-625.

Shorrocks, A., & Wan, G. (2005). Spatial decomposition of inequality. Journal of Economic

Geography, 5(1), 59-81.

Shiu, A., & Lam, P. L. (2004). Electricity consumption and economic growth in China. Energy

policy, 32(1), 47-54.

140

Sinani, E., & Meyer, K. E. (2004). Spillovers of technology transfer from FDI: the case of

Estonia. Journal of comparative economics, 32(3), 445-466.

Smith, R. (2006). Schools as institutions for peace in Northern Ireland: pupils’, parents’ and

teachers’ perspectives on the community relations dimension. Educate, 1(1), 123-153.

Solow, R. M. (1957). Technical change and the aggregate production function. The review of

Economics and Statistics, 312-320.

Sun, S. (2011). Foreign direct investment and technology spillovers in China's manufacturing

sector. Chinese Economy, 44(2), 25-42.

Iranzo, S., & Peri, G. (2009). Schooling externalities, technology, and productivity: Theory and

evidence from US states. The Review of Economics and Statistics, 91(2), 420-431.

Ten Raa, T., & Shestalova, V. (2011). The Solow residual, Domar aggregation, and inefficiency:

a synthesis of TFP measures. Journal of Productivity Analysis, 36(1), 71-77.

Vandenbussche, J., Aghion, P., & Meghir, C. (2006). Growth, distance to frontier and

composition of human capital. Journal of economic growth, 11(2), 97-127.

Waller, B. E., Hoy, C. W., Henderson, J. L., Stinner, B., & Welty, C. (1998). Matching

innovations with potential users, a case study of potato IPM practices. Agriculture,

ecosystems & environment, 70(2-3), 203-215.

Weil, D. (2005). Economic growth. Boston, Addison-Wesley.

Woessmann, L. (2003). Schooling resources, educational institutions and student performance:

the international evidence. Oxford bulletin of economics and statistics, 65(2), 117-170.

Woessmann, L. (2014). The Economic Case for Education.

Wozniak, G. D. (1993). Joint information acquisition and new technology adoption: Late versus

early adoption. The Review of Economics and Statistics, 438-445.

Yeaple, S. R. (2005). A simple model of firm heterogeneity, international trade, and

wages. Journal of international Economics, 65(1), 1-20.

Yoo, S. H. (2005). Electricity consumption and economic growth: evidence from Korea. Energy

Policy, 33(12), 1627-1632.

Yusuff, R. M., Chek, L. W., & Hashmi, M. S. J. (2005). Advanced manufacturing technologies

in SMEs. CACCI J. Commer. Ind, 1(1-11).

Zeira, J. (2009). Why and how education affects economic growth. Review of International

Economics, 17(3),1602-61.

141

Appendices

Appendix A: Definitions and Descriptive Statistics:

Variable Name Definition Source

Mathematics Cognitive

skills

Mathematics test scores for grade 8 National Center for Education Statistics (1992).

Report on TIMSS and PIRLS byInternational Study

Center, Lynch School of Education, Boston College

& International Association for the Evaluation of the

Educational Achievement. 2011.

Science Cognitive

Skills

Science test scores for grade 8 National Center for Education Statistics (1992).

Report on TIMSS and PIRLS byInternational Study

Center, Lynch School of Education, Boston College

& International Association for the Evaluation of the

Educational Achievement. 2011.

Years of Schooling Average years of total schooling Barro and Lee 2010

Life Expectancy Life expectancy at birth, total (years) World Bank, World Development Indicators.(2015)

Foreign Direct

Investment

Foreign direct investment, net inflows (% of

GDP)

World Bank, World Development Indicators.(2015)

Real GDP per capita Gross Domestic Product (GDP) measured in 1990

International Geary-Khamis dollar.

Maddison Data Set (2018)

Unemployment Rate Unemployment, total (% of total labor force)

(national estimate)

World Bank, World Development Indicators.(2015)

Harvester Number of self‐propelled machines that reap and

thresh in one

Operation

Comin and Hobijn (2009)

142

Milking machine Number of installations consisting of several

complete milking units


Tractor Number of wheel and crawler tractors (excluding

garden tractors)

used in agriculture


Fertilizer Metric tons of fertilizer consumed. Aggregate of

25 individual types listed in source


Bone marrow

Transplant

Number of bone marrow transplants performed Comin and Hobijn (2009)

Heart Transplant Number of heart transplants performed Comin and Hobijn (2009)

Kidney Transplant Number of kidney transplants performed Comin and Hobijn (2009)

Liver Transplant Number of liver transplants performed Comin and Hobijn (2009)

Lung Transplant Number of lung transplants performed. Comin and Hobijn (2009)

Cable TV Number of households that subscribe to a multi‐channel television

service delivered by a fixed line connection


Cellphone Number of users of portable cell phones Comin and Hobijn (2009)

Mail Number of items mailed/received, with internal

items counted one and cross‐border items counted

once for each country. May or may not include

newspapers sent by mail, registered mail, or


143

parcel post

Newspaper Number of newspaper copies circulated daily.

Note that there is a tendency for news circulation

to be under‐reported, since data for weekly and

biweekly publications are not included


Radio Number of radios Comin and Hobijn (2009)

Telephones Number of mainline telephone lines connecting a

customer's

equipment to the public switched telephone

network as of year end


Internet Number of people with access to the worldwide

network


Computer Number of self‐contained computers designed for

use by one

person


Visitor beds Number of visitor beds available in hotels and

elsewhere visitor rooms


Visitor rooms Number of visitor rooms available in hotels and

elsewhere.

years)


Aviation pkmp/air Civil aviation passenger‐KM traveled on Comin and Hobijn (2009)

144

scheduled services by companies registered in the

country concerned. Not a measure of

travel through a country’s airports

Shipton Steam

motor/sea

Tonnage of steam and motor ships (above a

minimum weight) in use at midyear


Vehicle car/land Number of passenger cars (excluding tractors and

similar vehicles) in use. Numbers typically

derived from registration and licensing records,

meaning that vehicles out of use may occasionally

be included.


Electricity production Gross output of electric energy (inclusive of

electricity consumed in power stations) in KwHr


Population Population


Political Rights Countries are ranked on the scale of 1-7 with

countries and territories with a rating of 1 enjoy a

wide range of political rights. These include free

and fair elections. Candidates who are elected

actually rule, political parties are competitive, the

opposition plays an important role and enjoys real

power, and the interests of minority groups are

well represented in politics and government.

Freedom in the World Report (2016)

Civil Liberties Countries are ranked on the scale of 1-7 with Freedom in the World Report (2016)

145

countries and territories with a rating of 1 enjoy a

wide range of civil liberties. These include

freedoms of expression, assembly, association,

education, and religion. They have an established

and generally fair legal system that ensures the

rule of law (including an independent judiciary),

allow free economic activity, and tend to strive

for equality of opportunity for everyone,

including women and minority groups.

GDP per capita GDP per capita is gross domestic product divided

by midyear population. GDP is the sum of gross

value added by all resident producers in the

economy plus any product taxes and minus any

subsidies not included in the value of the

products. It is calculated without making

deductions for depreciation of fabricated assets or

for depletion and degradation of natural resources.

Data are in current U.S. dollars.

World Bank, World Development Indicators (2015)

Expenditure on

R& D

Expenditures for research and development are

current and capital expenditures (both public and

private) on creative work undertaken

systematically to increase knowledge, including

knowledge of humanity, culture, and society, and

the use of knowledge for new applications. R&D

covers basic research, applied research, and

experimental development.

World Bank, World Development Indicators (2015)

146

Appendix A (continued): Descriptive Statistics for Mathematics Cognitive Skills Panel for

Usage Intensity as measure of Technology Adoption (1964-2003).

Variable Observations Mean Std Dev Min Max

Mathematics

Cognitive skills 480 424.064 111.452 122.4 609

Years of

Schooling 1000 8.001 2.582 0.92 12.64

Life Expectancy 1038 71.68183 6.006965 44.92385 81.76

Foreign Direct

Investment 672 2.015817 2.899249 -0.6519227 22.38404

Unemployment

Rate 495 7.360606 4.875179 0.9 36.7

Harvester 856 2.217481 2.440021 0.0000996 10.21824

Milking machine 494 4.761195 5.033899 0.0067787 21.15954

Tractor 894 13.80816 12.18866 0.001072 58.20502

Fertilizer 893 46.68103 40.69915 0.599535 229.3602

Bone marrow

Transplant 176 0.0214691 0.0183561 0.0001206 0.0746298

Heart Transplant 165 0.0049186 0.0033667 0 0.0147882

Kidney

Transplant 398 0. 0210578 0.0128271 0.0000962 0.0507885

Liver Transplant 176 0.0064384 0.0046055 0 0.0184819

Lung Transplant

120 0.0017186 0.0012383 0 0.0046925

Cable TV 457 94.04078 109.0733 0 401.346

Cellphone

671 97.31201 205.5867 0 1026.304

147

Mail 554 0.1775561 0.1306173 0.0023467 0.6652465

Radio 845 0.5986825 0.4161835 0.0517568 2.147192

Telephone

678 284.8042 234.0749 2.089562 1013.462

Internet 310 105.475 150.6394 0 573.1446

Computer 368 159.2686 153.634 .8779043 696.3917

Visitor beds 491 13.39978 9.011875 0.2283789 40.57656

Visitor rooms 577 6.565598 4.312404 0.2860303 17.09582

Aviation

pkmp/air 600 0.9307605 1.56559 0.001707 13.57749

Shipton

Steam motor/sea 389 0.2673012 0.5718443 0.0018271 3.300755

Vehicle car/land 798 216.9602 177.0675 0. 5360206 791.4692

Electricity

production 909 5465321 5501502 35040.34 3.12e+07

Population 962 39562.11 63868.29 1017 291200

Political Rights 742 2.448787 1.926992 1 7

Civil Liberties 742 2.568733 1.797432 1 7

GDP per capita 903 9823.426 9927.19 105.1262 50111.66

Expenditure

R&D 168 1.611943 0.9913625 0.10166 4.22244

148

Appendix A (continued): Descriptive Statistics for Science Cognitive Skills Panel for Usage

Intensity as measure of Technology Adoption (1973-2003).


Science Cognitive

skills

418 131.2 367.3213 151.1557 580

Years of

Schooling 713 8.317363 2.450953 1.79

12.64

Life Expectancy 744 72.27863 5.067044 53.47881 81.76

Foreign Direct

Investment 578 2.219367 2.994509 -0.6519227 22.38404

Unemployment

Rate 432 7.180787 5.252941 0.9 36.7

Harvester

552 2.557819 2.6157 0.0007337 10.21824

Milking machine 271 4.700926 5.172172 0.0067787 21.15954

Tractor 610 14.06253 13.14737 0.0085992

58.20502

Fertilizer 609 47.60572 42.00031 0.599535 229.3602

Bone marrow

Transplant

136 0.0179578 0.0167072 0.0001206 0.067665

Heart Transplant 117 0.0041423 0.0025224 0 0.0089563

Kidney

Transplant

272 0.0227518 0.0129084 0.0006714 0.0507885

Liver Transplant 121 0.0063818 0.0046032 0 0.0184819

Lung Transplant 95 0.0018226 0.0013436 0 0.0046925

Cable TV 396 65.82861 82.7646 0 278.7279

Cellphone

615 89.85775 192.0515 0 939.4391

149

Mail 329 0.1697851 0.1492096 0.0036425 0.6652465

Radio

567 0.6420931 0.4513853 0.0857526 2.147192

Telephones

431 292.6813 251.6995 6.327229 1013.462

Internet

276 102.5083 149.4308 0 573.1446

Computer

332 147.8118 158.4431 0.8779043 696.3917

Visitor beds

410 12.78704 9.883768 0.2283789 40.57656

Visitor rooms

536 6.199816 4.514737 0.2860303 17.09582

Aviation kmp/air

357 1.196812 1.939648 0.0357245 13.57749

Shipton Steam

motor/sea

253 0.3441496 0.6950479 0.0042609 3.300755

Vehicle car/land

519 225.8919 189.7235 2.190707 791.4692

Electricity

production

612 6120185 6228404 181876.8 3.12e+07

Population

669 42246.37 67901.2 1674 291200

Political Rights

640 2.720312 1.965642 1 7

Civil Liberties

640 2.829687 1.853152 1 7

GDP per capita

633 10574.13 10243.84 269.8519 50111.66

Expenditure R&D 149 1.372285 0.9553502 0.10166 3.91382

150

Appendix B: Results for Human Capital and Technology Adoption.

Table 1: Human capital and Usage Intensity of Technology in Transportation

Mathematics Skills Panel Science Skills Panel

Variables

(1)

Aviation pkm

air

(2)

Vehicle

car/land

(3)

Shipton

Steam motor/sea

(1)

Aviation

pkm

air

(2)

Vehicle

car/land

(3)

Shipton

Steam motor/sea

Cognitive Skills

0.00087***

(0.0003)

0.016

(0.03)

-0.00005

(0.0004)

-0.000028

(0.0001)

0.01673

(0.01)

0.00003***

(0.000005)

Years of

Schooling

-0.138***

(0.02)

-1.0545

(2.55)

0.0056

(0.003)

-0.0681**

(0.03)

-0.75244

(2.45)

0.0037***

(0.001)

Life Expectancy

0.0413***

(0.013)

1.016

(1.66)

0.00011

(0.001)

0.28313*

(0.016)

1.0993

(1.45)

-0.00154***

(0.0005)

FDI

0.0136

(0.011)

0.088

(0.48)

0.00111

(0.001)

0.00299

(0.109)

0.13722

(0.41)

-0.00074***

(0.0002)

Lagged dependent

variable

0.889***

(0.032)

0.9483***

(0.026)

0.955***

(0.027)

1.0220***

(0.03)

0.93327***

(0.025)

0.81053***

(0.06)

Observations 170

241 111 162 250 88

Standard errors in parenthesis; *,**,*** imply 10%, 5%, and 1% significance levels respectively.

151

Table 2: Human Capital and Usage Intensity of Technology in Health


Variables (1)

Transplant

Liver

(2)

Transplant

Lung

(3)

Transplant

Heart

(4)

Transplant

Bone

Marrow

(5)

Transplant

Kidney

(1)

Transplant

Liver

(2)

Transplant

Lung

(3)

Transplant

Heart

(4)

Transplant

Bone

Marrow

(5)

Transplant

Kidney

Cognitive Skills 0.000012**

(0.000006)

0.000026***

(0.000006)

0.0000011

(0.000006)

0.0000052

(0.00001)

0.000012

(0.000008)

-0.000006***

(0.000002)

-0.000005***

(0.000001)

-0.0000011

(0.000001)

0.000017**

(0.000007)

-0.0000015

(0.000003)

Years of

Schooling

0.00069

(0.0005)

-0.00089**

(0.0003)

-0.00064

(0.0004)

-0.00056

(0.001)

0.000031

(0.0006)

0.00031

(0.0004)

-0.00124***

(0.0003)

-0.00064

(0.0004)

-0.0021

(0.001)

0.00059

(0.0007)

Life

Expectancy

-0.000303

(0.0002)

0.0004***

(0.0001)

0.000064

(0.0001)

0.0014

(0.0009)

-0.000072

(0.0003)

0.000401

(0.0002)

0.00023

(0.0001)

0.00016

(0.0001)

0.00115

(0.0007)

0.000521

(0.0003)

FDI -0.000012

(0.00004)

0.000002

(0.00002)

-0.000021

(0.00004)

-0.00003

(0.0001)

-0.00019*

(0.0001)

0.000032

(0.00003)

-0.0000024

(0.00002)

-0.000015

(0.00004)

-0.00007

(0.0001)

-0.00014

(0.0001)

Lagged

dependent

variable

0.7933***

(0.077)

0.2136

(0.1171)

0.761***

(0.075)

0.817***

(0.06)

0.757***

(0.046)

0.6794***

(0.08)

0.40465***

(0.104)

0.70625***

(0.083)

0.7417***

(0.062)

0.730***

(0.05)

Observations 83 68 93 106 196 90 72 92 109 209


152

Table 3a: Human Capital and Usage of Intensity of Technology in Telecommunications & Information (mathematics panel)

Variables

(1)

Cable TV

(2)

Mail

(3)

Computers

(4)

Internet

user

(5)

Radio

(6)

Telephone

(7)

Cell phones

Cognitive

Skills

0.1072***

(0.03)

0.000122**

(0.00004)

0.1891***

(0.057)

0.449*

(0.25)

-0.000004

(0.00006)

-0.0666

(0.04)

-0.158*

(0.08)

Years of

Schooling

1.529

(2.61)

0.00048

(0.003)

7.331**

(3.57)

7.018

(10.009)

0.0258***

(0.04)

0.514

(2.63)

11.365***

(6.35)

Life

Expectancy

0.0402

(1.5)

-0.0022

(0.001)

3.0355

(2.20)

24.95***

(6.44)

0.0021

(0.002)

1.926

(1.23)

21.265***

(4.27)

FDI -1.113***

(0.31)

0.0033**

(0.001)

0.2325

(0.4)

0.591

(1.02)

-0.0013

(0.0008)

2.753

(0.70)

2.473***

(0.93)

Lagged

dependent

variable

0.8615***

(0.03)

0.9075***

(0.028)

1.0137***

(0.015)

0.945***

(0.03)

0.844***

(0.021)

1.0002***

(0.025)

1.001***

(0.018)

Observations 212 163 178 150 257 190 258


153

Table 3b: Science based Cognitive Skills and Usage of Intensity of technology in Telecommunications & Information (science

panel)

Variables

(1)

Cable TV

(2)

Cell phones

(3)

Radio

(4)

Computers

(5)

Internet

user

(6)

Mail

(7)

Telephone

Cognitive

Skills

0.0184*

(0.01)

0.00006

(0.027)

0.00003

(0.00002)

0.01487

(0.01)

0.07052

(0.061)

-0.000046**

(0.00001)

-0.00452

(0.01)

Years of

Schooling

-0.6624

(2.23)

19.439***

(5.77)

0.01096**

(0.005)

6.4531**

(3.14)

2.3008

(8.45)

-0.00052

(0.004)

-1.026

(2.63)

Life

Expectancy

0.39457

(1.38)

17.655***

(3.67)

0.00466*

(0.002)

5.0570**

(1.97)

33.118***

(6.36)

0.00581***

(0.001)

2.414

(1.634)

FDI -0.9187***

(0.26)

1.9098**

(0.81)

0.00099

(0.0008)

0.35111

(0.37)

0.3104

(0.94)

0.00237

(0.001)

2.3892***

(0.58)

Lagged

dependent

variable

0.89391***

(0.02)

1.0266***

(0.015)

0.9672***

(0.02)

1.0159***

(0.13)

0.92361***

(0.028)

0.9496***

(0.31)

0.9379***

(0.02)

Observations 253 304 265 215 177 153 162


154

Table 4: Human Capital and Usage Intensity of Technology in Electricity Production and Tourism.


Variables

(1)

Electricity

production

(2)

Visitor beds

(3)

Visitor rooms

(1)

Electricity

production

(2)

Visitor beds

(3)

Visitor rooms

Cognitive Skills 4451.838***

(1461.9)

0.011***

(0.003)

0.004**

(0.002)

256.248

(560.69)

0.00115

(0.001)

0.00185**

(0.007)

Years of

Schooling

12885

(92761.2)

-0.584

(0.211)

0.305**

(0.15)

50849.27

(100069.4)

-0.10322

(0.18)

0.11308

(0.13)

Life Expectancy 39326.89

(53344.6)

-0.794

(0.13)

-0.0761

(0.08)

169130.4***

(55758.08)

0.19507*

(0.11)

-0.082805

(0.077)

FDI 4499.4

(18550.1)

-0.687

(0.03)

-0.0387

(0.02)

6740.894

(19316.95)

-0.04587

(0.03)

-0.01549

(0.02)

Lagged

dependent

variable

0.7402***

(0.04)

0.7342***

(0.054)

0.8539***

(0.036)

0.73240***

(0.041)

0.81504***

(0.042)

0.85242***

(0.033)

Observations 279 157 244 289 190

269


155

Table 5: Human Capital and Usage Intensity of Technology in Agriculture


Variables

(1)

Harvester

(2)

Fertilizers

(3)

Milking

machine

(4)

Tractor

(1)

Harvester

(2)

Fertilizers

(3)

Milking

machine

(4)

Tractor

Cognitive

Skills

-0.0009**

(0.0004)

-0.040***

(0.01)

0.000029

(0.001)

-0.0027*

(0.001)

-0.00049**

(0.0001)

-0.00746

(0.005)

0.00095***

(0.0003)

-0.0013**

(0.0006)

Years of

Schooling

0.032

(0.037)

-3.24***

(1.12)

-0.1537**

(0.06)

0.0672

(0.12)

0.0342

(0.366)

-2.5538***

(0.98)

-0.06548

(0.05)

0.03154

(0.13)

Life

Expectancy

-0.0084

(0.019)

3.099***

(0.64)

0.0126

(0.044)

-0.0049

(0.64)

-0.00816

(0.016)

2.2730

(0.46)

-0.0581**

(0.02)

-0.00188

(0.57)

FDI -0.011

(0.007)

-0.133

(0.21)

0.0207

(0.02)

-0.0053

(0.23)

-0.000841

(0.007)

-0.219

(0.18)

0.0395**

(0.18)

-0.00566

(0.024)

Lagged

dependent

variable

0.8828***

(0.02)

0.834***

(0.03)

0.998***

(0.015)

0.8823***

(0.016)

0.83778***

(0.02)

0.08068***

(0.027)

0.98338***

(0.012)

0.9099***

(0.018)

Observations 287 293 170 293 288 305 174 305


156

Appendix C. Descriptive Statistics for Mathematics Cognitive Skills Panel for Technology

Usage Lags as measures of Technology Diffusion (1973-2003)


Mathematics

Cognitive skills

440 428.1243 110.2615 122.4 609

Years of

Schooling

960 7.866635 2.518101 0.92 11.76

Life Expectancy 998 71.5944 6.093759 44.92385 81.76

Foreign Direct

Investment

638 2.082505 2.956014 -0.6519227 22.38404

Real GDP lag per

capita

975 29.08615 18.07572 4 73

Unemployment

Rate

471 7.415711 4.980973 0.9 36.7

Harvester 706 22.52295 11.17289 1.803589 40.99407

Tractor 744 21.43499 11.06449 -2.269452 40.99862

Fertilizer 701 20.55985 11.52588 -2.603976 43.00351

Bone marrow

Transplant

116 1.486409 3.937542 -6.049613 12.04657

Heart Transplant 102 7.153791 4.400163 2.974864 14.98844

Kidney

Transplant

329 5.504315 7.862458 16.97348 27.2596

Liver Transplant 148 5.309744 4.722574 2.164151 21.02741

Lung Transplant 76 3.244937 4.922544 7.649384 14.53378

Cable TV 235 13.71066 8.345276 -18.62067 27.51526

Cellphone 333 2.391144 4.276329 -7.869115 12.14841

Mail 526 60.61741 25.1219 7.968489 107.4928

157

Radio 787

32.74951 10.11304 13.13484 60.64052

Telephone 613 42.08814 26.05534 -0.1616383 87.49915

Internet 277 2.755459 2.542218 -1.744918 9.251847

Computer 344 7.733123 4.754848 -0.7463593 19.51483

Visitor beds 365 9.568224 7.048247 -3.591774 23.91281

Visitor rooms 498 10.2466 8.419268 -24.9582 25.92233

Aviation pkmp/air 535

16.37194 10.78567 -17.4344 45.87047

Electricity

production

753 31.76108 17.77717 -18 67.01201

Political Rights 710

2.514085 1.944708 1 7

Civil Liberties 710 2.639437 1.805686 1 7

GDP per capita 863

9446.723 9693.34 105.1262 50111.66

Expenditure on

R& D

160 1.565586 0.9932934 0.10166 4.22244

158

Appendix C continued: Descriptive Statistics for Science Cognitive Skills Panel for

Technology Usage Lags as measures of Technology Diffusion (1964-2003)


Mathematics

Cognitive skills

387 372.5238 149.0293 131.2 580

Years of Schooling 682

8.147155 2.366785 1.79 11.76

Life Expectancy 713 72.16753 5.138295 53.47881 81.76

Foreign Direct

Investment

547 2.298172 3.054491 -0.6519227 22.38404

Real GDP lag per

capita

681 33.68674 19.32983 4 73

Unemployment Rate 408 7.233824 5.389327 0.9 36.7

Harvester 438 27.26247 8.415413 11.3886 40.99407

Tractor 498 26.09841 8.535559 6.340208 40.99862

Fertilizer 494 24.48208 9.807492 -1.955834 40.96936

Bone marrow

Transplant

89

1.729173 4.164594 -7.323859 12.04657

Heart Transplant 93 23.71823 47.77922 -6.02362 216.5907

Kidney Transplant 224

6.097456 9.264138 -27.43828 34.35971

Liver Transplant 102

5.430726 4.998367 -2.164151 21.02741

Lung Transplant 69 2.885949 5.33661 -7.649384 14.53378

Cable TV 201 15.65594 7.227921 -12.21838 43.27872

Cellphone 317 2.550117 4.362253 -7.869115 12.14841

Mail 307

67.58051 27.22576 7.968489 107.4928

Radio 518 36.40362 9.782321 13.13484 60.64052

159

Telephone 386

49.21271 27.22868 -0.1616383 87.66544

Internet 244 3.099174 2.666004 -1.744918 9.251847

Computer 309 8.416949 5.145492 -0.7463593 19.51483

Visitor beds 282 9.689905 7.206189 -2.562439 23.91281

Visitor rooms 445 25.69847 66.30322 -24.9582 327.4338

Aviation pkmp/air 309 19.55163 11.03387 -17.4344 45.87047

Electricity

production

483 35.49276 17.83551 -18 67.01201

Political Rights 609 2.807882 1.975392 1 7

Civil Liberties 609 2.922824 1.851996 1 7

GDP per capita 602 9996.65 9918.367 269.8519 50111.66

Expenditure on

R& D

141 1.306084 0.939365 0.10166 3.91382

160

Appendix D: Results for Human Capital and Technology Usage Lags.

Table 1a: Human Capital and Technology Usage lags in Telecommunications & Information (mathematics panel).

Variables

(1)

Internet user

(2)

Telephone

(3)

Computers

(4)

Mail

(5)

Cable TV

(6)

Cell phones

(7)

Radio

Cognitive Skills -0.184*

(0.0101)

-0.003

(0.005)

-0.015***

(0.004)

-0.0601

(0.013)

-0.0106

(0.006)

0.0075**

(0.057)

0.0036*

(0.031)

Years of

Schooling

-0.042

(0.27)

0.0563

(0.367)

-0.290

(0.226)

0.109

(1.08)

0.713

(0.439)

-0.709***

(0.202)

-0.332

(0.151)

Life Expectancy -0.0012

(0.163)

0.070

(0.201)

0.183

(0.120)

0.098

(0.523)

0.347

(0.260)

-0.238*

(0.134)

0.188**

(0.088)

FDI

0.0016

(0.032)

-0.400***

(0.106)

0.018

(0.028)

-1.068**

(0.043)

0.047

(0.051)

-0.044

(0.046)

0.017

(0.02)

GDP/income

lag

0.003

(0.047)

-0.117*

(-0.117)

-0.0073

(0.027)

0.135

(0.124)

0.0095

(0.051)

-0.0138

(0.028)

-0.0146

(0.0207)

Lagged

dependent

variable

0.551***

(0.106)

0.736***

(0.037)

0.88***

(0.470)

0.804***

(0.051)

0.581***

(0/075)

0.7633***

(0.057)

0.0892***

(0.031)

Observations 125 154 157 140 123 142 222


161

Table 1b: Human Capital and Technology Usage lags in Telecommunications & Information (science panel).

Variables

(1)

Mail

(2)

Cable TV

(3)

Computers

(4)

Internet

user

(5)

Telephone

(6)

Cell phones

(7)

Radio

Cognitive

Skills -0.011*

(0.006)

-0.002

(0.001)

-0.002**

(0.001)

0.0016

(0.002)

0.012

(0.008)

0.0013

(0.001)

0.004***

(0.0009)

Years of

Schooling

-0.144

(1.42)

0.792**

(0.365)

0.047

(0.204)

-0.434

(0.002)

1.334

(0.008)

-0.830

(0.180)

-0.427***

(0.161)

Life

Expectancy

-0.128

(0.653)

0.347

(0.212)

0.029

(0.109)

-0.063

(0.165)

-1.014

(0.716)

-0.154

(0.109)

0.518***

(0.097)

FDI

0.083

(0.421)

0.051

(0.04)

-0.006

(0.027)

0.020

(0.034)

-0.314

(0.351)

-0.192

(0.033)

-0.026

(0.028)

GDP/income

lag

0.022

(0.87)

0.005

(0.03)

-0.0108

(0.021)

-0.003

(0.33)

-0.179

(0.247)

0.0132

(0.018)

0.059***

(0.017)

Lagged

dependent

variable

0.568***

(0.067)

0.603***

(0.058)

0.866***

(0.044)

0.385***

(0.11)

-0.011

(0.09)

0.849***

(0.042)

0.698***

(0.038)

Observations 133 134 194 157 125 200 234


162

Table 2: Human Capital and Technology Usage lags in Health.


Variables (1)

Transplant

Bone

marrow

(2)

Transplant

Lung

(3)

Transplant

Liver

(4)

Transplant

Kidney

(5)

Transplant

Heart

(1)

Transplant

Bone

marrow

(2)

Transplant

Lung

(3)

Transplant

Liver

(4)

Transplant

Kidney

(5)

Transplant

Heart

Cognitive

Skills

-0.001

(0.006)

-0.099***

(0.056)

-0.0103

(0.01)

0.0011

(0.007)

0.007

(0.004)

-0.004*

(0.002)

-0.0019

(0.008)

0.0001

(0.002)

0.012***

(0.004)

0.065**

(0.028)

Years of

Schooling

-0.152

(0.371)

12.72***

(3.54)

1.278

(0.997)

0.261

(0.565)

0.310

(0.32)

0.217

(0.509)

5.668

(2.106)

-0.094

(0.438)

-1.408*

(0.835)

-12.813**

(6.471)

Life

Expectancy

0.665***

(0.194)

-0.0286

(1.11)

2.613***

(0.555)

1.249***

(0.399)

0.418**

(0.203)

0.847***

(0.26)

0.117

(0.991)

1.501***

(0.388)

1.951***

(0.423)

0.682

(2.683)

FDI 0.026

(0.293)

0.009

(0.151)

-0.0101

(0.07)

0.183**

(0.087)

0.016

(0.02)

0.0417

(0.041)

0.071

(0.130)

0.0143

(0.038)

0.270**

(0.138)

0.230

(0.541)

GDP/income

lag

-0.009

(0.034)

0.253

(0.033)

0.001

(0.067)

-0.032

(0.071)

-0.0072

(0.024)

-0.0286

(0.034)

-0.032

(0.109)

-0.011

(0.031)

0.020

(0.104)

-0.138

(0.454)

Lagged

dependent

variable

0.7715***

(0.075)

-0054

(0.207)

0.121

(0.12)

0.6304***

(0.06)

0.856***

(0.057)

0.693***

(0.087)

0.208

(0.163)

0.523***

(0.103)

0.296***

(0.079)

0.897***

(0.045)

Observations 59 33 60 150 58 67 48 76 166 71


163

Table 3: Human Capital and Technology Usage lags in Tourism and Electricity Production.


Variables

(1)

Visitor beds

(2)

Visitor

rooms

(3)

Electricity

production

(1)

Visitor beds

(2)

Visitor

rooms

(3)

Electricity

production

Cognitive Skills 0.0086

(0.008)

-0.0063

(0.005)

-0.0029

(0.006)

0.002

(0.003)

-0.003*

(0.002)

-0.0015

(0.001)

Years of

Schooling

1.073**

(0.484)

-0.655

(0.404)

-0.0293

(0.375)

0.588

(0.430)

-0.505

(0.395)

0.175

(0.278)

Life Expectancy 0.549*

(0.319)

0.594

(0.209)

0.1748

(0.268)

0.635**

(0.279)

0.800***

(0.243)

0.275*

(0.163)

FDI

0.004

(0.06)

0.051

(0.055)

0.0603

(0.110)

0.015

(0.058)

0.071

(0.061)

0.039

(0.072)

GDP/income lag -0.0033

(0.094)

0.124**

(0.053)

0.059

(0.061)

-0.031

(0.085)

0.050

(0.046)

0.105*

(0.045)

Lagged

dependent

variable

0.535***

(0.118)

0.835***

(0.046)

0.818***

(0.049)

0.624***

(0.093)

0.790***

(0.047)

0.792***

(0.043)

Observations 100 182 197 101 198 203


164

Table 4: Human Capital and Usage lags of Technology in Agriculture


Variables (1)

Fertilizers

(2)

Harvester

(3)

Tractor

(1)

Fertilizers

(2)

Harvester

(3)

Tractor

Cognitive Skills 0.020***

(0.006)

0.00019

(0.0001)

0.0017

(0.001)

0.0013

(0.001)

-0.003

(0.002)

0.00058

(0.0005)

Years of

Schooling

1.407***

(0.505)

-0.002

(0.006)

0.024

(0.068)

0.747***

(0.266)

-0.008

(0.036)

0.124

(0.080)

Life

Expectancy

0.387

(0.257)

0.0035

(0.005)

0.0415

(0.054)

0.426***

(0.134)

-0.049*

(0.026)

-0.0388

(0.064)

FDI 0.012

(0.073)

0.0015

(0.001)

0.015

(0.012)

0.028

(0.036)

-0.008

(0.006)

-0.008

(0.014)

GDP/income

lag

0.012

(0.06)

0.0004

(0.001)

-0.0105

(0.013)

0.026

(0.02)

-0.0002

(0.005)

-0.001

(0.012)

Lagged

dependent

variable

0.609***

(0.062)

0.995***

(0.002)

0.961***

(0.018)

0.790***

(0.039)

1.019***

(0.008)

0.979***

(0.018)

Observations 183 179 214 215 192 210


165

Appendix E: Additional Robustness Checks

Table 1: Robustness Checks for Mathematics Skills Usage Intensity of Technology.

(1)

Transplant Liver

(2)

Transplant Lung

(3)

Visitor beds

Variables (1) (2) (3) (1) (2) (3)

(1) (2) (3)

Cognitive Skills

0.000013

(0.000006)

0.0000065

(0.000006)

0.000073

(0.000068)

0.00002***

(0.000006)

0.000020***

(0.000007)

-.0000179

(0.0000633)

0.01187***

(0.0034)

0.007998*

(0.00438)

-0.023136

(0.02688)

Years of Schooling

0.0007603

(0.00052)

0.00007416

(0.000522)

0.0015205

(0.00132)

-0.000898 **

(0.00036)

-0.001069***

(0.000369)

-0.000356

(0.00118)

-0.019472

(0.21529)

-0.079915

(0.21935)

0.10662

(0.56180)

Life Expectancy

-0.000268

(0.00027)

-0.000217

(0.000275)

-0.0014993

(0.00077)*

0.000416**

(0.000163)

0.0004319***

(0.00016)

-0.0000532

(0.00075)

0.05572

(0.14597)

0.075843

(0.14708)

-0.060272

(0.21628)

FDI

-0.000012

(0.00004)

-0.000008

(0.00004)

-0.0000985

(0.00011)

-0.000001

(0.00002)

-0.0000015

(0.000025)

(0.00010)

(0.00010)

-0.07433**

(0.03480)

-0.07094**

(0.03503)

-0.10202

(0.10031)

Political Rights

-0.00067

(0.00077)

-0.00022

(0.00081)

-0.29977

(0.26496)

-0.248431

(0.26874)

-0.06508

(0.47271)

Civil Liberties

0.000173

(0.00074)

-0.000006

(0.00074)

-0.000222

(0.00058)

-0.00043

(0.00058)

-0.03345

(0.33508)

-0.027362

(0.336527)

-0.327765

(0.56605)

GDP Per capita

0.00000006

(0.0000003)*

0.00000007

(0.00000007)

0.00000005**

(0.00000002)

0.00000004

0.00000007

0 .000043

(0.00003)

-0.0000686

(0.000142)

Research &

Development

0.0094451

(0.00438)**

0.0028803

(0.00413)

0.0026051

(1.2998)

Lagged dependent

variable

0.7909***

(0.07891)

0.751695***

(0.08180)

0.2716344

(0.22203)

0.2529***

(0.11872)

0.268187**

(0.117072)

-0.10466

(0.81608)

0.7162***

(0.05667)

0.7228***

(0.0565)

0.6902***

(0.1841)

Observations 83 83 16 68 68 16 157 157 32


166

Table 1 continued: Mathematics Panel

(4)

Vehicle car/land

(5)

Cable TV

(6)

Computer

Variables (1) (2) (3) (1) (2) (3) (1) (2) (3)


(0.03530)

-0.040395

(0.03768)

0.543810

(0.95384)

0.1142***

(0.0353)

0.06643*

(0.03926)

-0.239419

(0.82716)

0.20109***

(0.05913)

0.21552***

(0.07104)

0.873177

(0.67931)

Years of Schooling -1.4853

(2.7145)

-1.436219

(2.6172)

-3.651587

(10.026)

1.550353

(2.6539)

1.315192

(2.6218)

1.799311

(9.2926)

7.683769**

(3.59070)

7.852866**

(3.62043)

15.3934

(10.954)


(1.8851)

1.710958

(1.8221)

4.333874

(5.7518)

-.608835

(1.50703)

-.9733144

(1.49295)

4.953014

(4.70411)

2.635373

(2.2624)

2.581063

(2.2829)

4.982002

(5.9151)

FDI

0.09599

(0.50518)

0.072433

(0.48711)

-2.778173

(2.4257)

-1.045***

(0.31227)

-0.92542***

(0.31161)

-0.168778

(0.88461)

0.20785

(0.40405)

0.18115

(0.40856)

-1.86743*

(1.1231)

Political Rights -0.72536

(2.6632)

0.358174

(2.5846)

6.11402

(7.3948)

-8.769***

(2.7068)

-8.513206***

(2.6737)

-6.928899

(8.1358)

-0.0467878

(2.5885)

-0.09809

(2.6046)

13.3344

(10.406)

Civil Liberties 4.943798*

(2.9863)

1.892107

(2.9968)

1.896459

(7.81754)

5.316003**

(2.4019)

3.424606

(2.4759)

-0.9417895

(7.71469)

-2.68915

(3.3899)

-2.64764

(3.4598)

-9.09771

(9.5067)

GDP Per capita 0.0011***

(0.00030)

-0.0000721

(0.00124)

0.000583***

(0.00021)

-0.0011603

(0.00110)

-0.000115

(0.000353)

.0011837

(0.001381)

Research &

Development

67.29925*

(35.716)

9.355411

(16.2555)

-41.75186

(26.099)

Lagged dependent

variable

0.9425***

(0.02786)

0.8667***

(0.03387)

0.5731***

(0.1569)

0.8569***

(0.03136)

0.8442***

(0.03133)

0.5337***

(0.1466)

1.0147***

(0.0154)

1.015***

(0.0157)

1.011***

(0.0548)

Observations 227 227 26 212 212 54 178 178 63


167

Table 1 continued: Mathematics Panel (7)

Tractor

(8)

Fertilizer

Variables (1) (2) (3) (1) (2) (3)


(0.00162)

-0.000582

(0.00198)

0.029655

(0.06046)

-0.0442351***

(0.01492)

-0.00104908

(0.01703)

-0.1352981

(0.12974)

Years of Schooling 0.029415

(0.13385)

0.034499

(0.13383)

0.464117

(0.72843)

-3.601329***

(1.1844)

-3.52864**

(1.171003)

0.4274924

(1.58566)


(0.06975)

0.018735

(0.07005)

0.033375

(0.35982)

3.437704***

(0.70235)

3.9067***

(0.70433)

-1.824247**

(0.80098)

FDI 0.0048781

(0.02503)

0.0027866

(0.02487)

-0.0097704

(0.0986123)

-0.1256564

(0.221213)

-0.1922653

(0.2193273)

-0.1750385

(0.2115905)

Political Rights 0.1729604

(0.14262)

0.1287412

(0.14399)

-0.0546575

(0.67600)

0.2797649

(1.24716)

-0.61966

(1.2536)

-1.11979

(1.4850)

Civil Liberties 0.2545829*

(0.15025)

0.364131**

(0.15872)

0.073563

(0.844675)

2.09251

(1.36454)

4.102527***

(1.44108)

0.591595

(1.8202)

GDP Per capita -0.0000308**

(0.000014)

-0.0000616

(0.000101)

-0.0005***

(0.00012)

-0.0005043**

(0.00023)

Research &

Development

-0.5875405

(1.47783)

-4.359462

(3.1697)

Lagged dependent

variable

0.8800***

(0.0198)

0.88801***

(0.0201)

0.7385***

(0.0895)

0.8244***

(0.0346)

0.7930***

(0.03514)

0.134858

(0.14849)

Observations 279 279 50 279 279 50


168

Table 2: Robustness Checks for Science Skills and Usage Intensity of Technology.

(1)

Transplant Liver

(2)

Transplant Lung

(3)

Visitor beds

Variables (1) (2) (3) (1) (2) (3) (1) (2) (3)

Cognitive Skills 0.000006**

(0.000002)

0.000005**

(0.000002)

(0.00002)**

(0.000009)

0.000004***

(0.000001)

0.000004**

(0.000001)

-0.00001

0.0000007

0.00157

(0.00129)

-0.00013

(0.0014)

0.00712

(0.006)

Years of Schooling

0.000312

(0.00043)

0.00028

(0.00043)

-0.000350

(0.00127)

-0.0012427***

(0.00037)

-0.0013***

(0.0003)

-0.00026

(0.0009)

-0.07229

(0.1886)

-0.16391

(0.1895)

0.371475

(0.4379)

Life Expectancy

-0.00041

(0.00026)

-0.000381

(0.00026)

0.000532

(0.0006)

0.00023

(0.00017)

0.00028*

(0.0001)

0.00008

(0.0004)

0.16711

(0.1131)

0.15817

(0.1123)

-0.067138

(0.1762)

FDI

0.000035

(0.00004)

0.000036

(0.00004)

0.00025***

(0.00008)

-0.0000016

(0.00002)

-0.000003

(0.00002)

0.000013

(0.00006)

-0.05184*

(0.0312)

-0.04324

(0.0312)

-0.002219

(0.0506)

Political Rights

-0.00073

(0.00058)

-0.000373

(0.00062)

-0.23799

(0.2235)

-0.195155

(0.2222)

-0.19829

(0.4002)

Civil Liberties

.0001867

(0.00038)

0.000162

(0.00038)

-0.000446

(0.0006)

-0.000594

(0.0005)

-0.09759

(0.216)

-0.053984

(0.2155)

0.03184

(0.4145)

GDP Per capita 0.000000005

(0.00000003)

0.0000001*

(0.00000008)

0.00000005**

(0.00000002)

0.000000056

(0.00000005) 0.0000651***

(0.00002) -0.000005

(0.0001)

Research &

Development

-0.0053***

(0.00173)

0.00036

(0.00167) -0.52165

(0.9345)

Lagged dependent

variable

0.6807***

(0.0867)

0.65282***

(0.08798)

0.498817

(0.20351)

0.4123***

(0.10644)

0.4101***

(0.1038)

0.14207

(0.5139)

0.79578***

(0.0445)

0.7813***

(0.0445)

0.7563***

(0.1548)

Observations 90 90 22 72 72 21 190 190 44


169

Table 2 continued: Science Panel

(4)

Vehicle car/land

(5)

Cable TV

(6)

Computer

Variables (1) (2) (3) (1) (2) (3) (1) (2) (3)

Cognitive

Skills

0.01653

(0.01064)

0.002474

(0.01166)

-0.06959

(0.11351)

0.0189*

(0.01065)

-0.0086

(0.0124)

0.03256

(0.158)

0.023054*

(0.0133)

0.01626

(0.0168)

0.041007

(0.1515)

Years of

Schooling

-1.035602

(2.4878)

-2.363515

(2.4916)

-1.92958

(9.6117)

-0.8155

(2.242)

-0.477803

(2.202)

2.9533

(8.248)

6.446825**

(3.125)

6.466533**

(3.1075)

12.5284

(9.481)

Life

Expectancy

1.319532

(1.5129)

0.88258

(1.4846)

0.17474

(4.794)

0.344787

(1.387)

-0.03924

(1.367)

4.67587

(3.999)

4.7476**

(1.966)

5.060857**

(2.020)

3.783758

(5.208)

FDI 0.1464266

(0.41812)

0.23612

(0.41285)

0.5974

(0.97135)

-0.8591***

(0.2698)

-0.68453**

(0.2689)

-0.1067703

(0.7349)

0.3075012

(0.370)

0.32611

(0.3704)

-1.244442

(0.9531)


(2.2263)

-0.7617381

(2.1790)

0.16289

(6.5259)

-7.934***

(2.439)

-8.12762***

(2.398)

-6.833525

(7.417)

-0.354014

(2.625)

-0.3511

(2.6207)

11.24717

(9.7400)

Civil Liberties 1.579393

(2.2949)

0.4503156

(2.2914)

(-0.42590)

(7.7340)

4.426275**

(2.029)

2.908508

(2.029)

-0.84805

(7.086)

-5.355732**

(2.723)

-5.523123**

(2.731)

-10.03891

(8.862)

GDP Per capita 0.0007***

(0.0002)

0.000908

(0.00093)

0.0008***

(0.0002)

-0.00131

(0.0009)

0.0002313

(0.0003)

0.00134

(0.0012)

Research &

Development

24.47498

(20.646)

8.946207

(13.985)

-50.38995**

(21.225)

Lagged

dependent

variable

0.930445***

(0.0257)

0.8994***

(0.02788)

0.8647***

(0.11925)

0.8854***

(0.0262)

0.8680***

(0.0261)

0.5317***

(0.136)

1.013***

(0.0133)

1.011***

(0.013)

1.033***

(0.047)

Observations 250 250 36 253 253 62 215 215 75


170

Table 2 continued: Science Panel

(7)

Tractor

(8)

Fertilizer

Variables (1) (2) (3) (1) (2) (3)

Cognitive Skills -0.0016292**

(0.00068)

-0.000848

(0.00077)

0.003432

(0.00950)

-0.0096503*

(0.00525)

0.00383

(0.00579)

0.03638

(0.02373)


(0.14000)

0.062953

(0.140382)

0.2297

(0.57673)

-3.077685***

(0.98504)

-2.402649**

(0.97232)

-0.224506

(1.4558)


(0.05957)

0.06487

(0.06216)

0.10401)

(0.29298)

2.628***

(0.47795)

3.3597***

(0.48842)

-0.9713

(0.7097)

FDI 0.000843

(0.02464)

-0.00264

(0.02465)

-0.003406

(0.07303)

-.1793301

(0.1886)

-0.30106

(0.18593)

-0.16421

(0.1819)


(0.13841)

0.145708

(0.13867)

-0.00545

(0.60253)

0.89101

(1.058)

0.49752

(1.0384)

-0.2900002

(1.520)


(0.138376)

0.248134*

(0.14002)

2.365319**

(1.1027)

3.359013

(1.0956)

1.61003

(1.8532)

GDP Per capita -0.000029**

(0.00001)

-0.0005***

(0.0001)

-0.000307

(0.00021)

Research &

Development

-0.51270

(1.1698)

-3.010473

(2.9443)

Lagged dependent

variable

0.9100***

(0.01899)

0.91481***

(0.01912)

0.7356***

(0.07871)

0.7905***

(0.02794)

0.7590***

(0.0280)

0.4419***

(0.11920)

Observations 305 305 61 305 305 61


171

Appendix E Continued: Robustness Checks for Mathematics and Science Skills and Technology Usage Lags.

Table 3: Mathematics Panel and Technology Diffusion. (1)

Cable TV

(2)

Computers

(3)

Transplant Bone marrow

Variables (1) (2) (3) (1) (2) (3) (1) (2) (3)

Cognitive Skills -0.100

(0.006)

-0.0107

(0.009)

-0.129

(0.139)

-0.016***

(0.005)

-0.014**

(0.005)

-0.030

(0.031)

-0.009

(0.006)

-0.006

(0.008)

0.074

(0.270)

Years of

Schooling

0.761

(0.446)*

0.791

(0.452)*

0.388

(1.223)

-0.305

(0.231)

-0.293

(0.23)

-0.605

(0.514)

-0.3003

(0.411)

-0.313

(0.424)

2.274

(1.719)


(0.263)

0.342

(0.263)

0.367

(0.584)

0.189

(0.122)

0.202

(0.124)

-0.043

(0.234)

0.675***

(0.197)

0.628***

(0.207)

0.106

(1.279)

FDI 0.043

(0.052)

0.045

(0.053)

-0.068

(0.127)

0.023

(0.029)

0.019

(0.029)

-0.046

(0.064)

0.029

(0.030)

0.029

(0.0306)

-0.181

(0.315)


(0.441)

0.609

(0.443)

2.057*

(1.156)

0.0508

(0.165)

0.051

(0.165)

0.205

(0.478)

-0.667

(0.738)

-0.661

(0.762)

Civil Liberties -0.227

(0.389)

-0.248

(0.416)

-0.289

(1.011)

0.1073

(0.223)

0.139

(0.228)

-0.589

(0.543)


(0.00005)

0.00002

(0.0002)

-0.003

(0.027)

-0.00002

(0.00002)

-0.000005

(0.00008)

-0.0001

(0.00004)

-0.00002

(0.0001)

Research &

Development

-1.884

(2.621)

3.129***

(1.078)

9.106

(14.99)

GDP/income lag

0.0009

(0.053)

0.003

(0.060

-0.071

(0.231)

-0.012

(0.030)

0.054

(0.0905)

-0.012

(0.035)

-0.0364

(0.035)

-0.143

(0.305)

Lagged

dependent

variable

0.568***

(0.077)

0.569***

(0.079)

0.436***

(0.168)

0.880***

(0.046)

0.892***

(0.047)

0.857***

(0.073)

0.763***

(0.077)

0.779***

(0.077)

0.4206

(0.657)

Observations 123 123 43 157 157 56 59 59 12


172

Table 3 continued: Mathematics Panel and Technology Diffusion.

(4)

Lung Transplant

(5)

Fertilizers

(6)

Tractor

Variables (1) (2) (3) (1) (2) (3) (1) (2) (3)


(0.058)

-0.108*

(0.059)

-0.0511 0.019***

(0.006)

0.016**

(0.007)

-0.006

(0.074)

0.0014*

(0.0008)

0.001

(0.0009)

-0.004*

(0.002)

Years of Schooling 12.726***

(3.686)

11.993***

(3.793)

1.337**

(0.519)

1.303**

(0.521)

0.842

(0.939)

-0.015

(0.052)

-0.0241

(0.053)

-0.012

(0.03)

Life Expectancy -0.028

(1.153)

0.5903

(1.36)

-3.691 0.443

(0.271)

0.415

(0.273)

0.515

(0.495)

-0.015

(0.052)

0.077*

(0.045)

0.037*

(0.019)

FDI 0.009

(0.157)

-0.003

(0.159)

2.611 0.0204

(0.076)

0.028

(0.076)

-0.247

(0.120)

0.002

(0.009)

0.003

(0.009)

-0.0005

(0.003)


(0.412)

0.025

(0.415)

0.819

(0.851)

-0.021

(0.053)

-0.015

(0.054)

0.025

(0.03)


(0.482)

0.208

(0.513)

-0.452

(0.993)

0.0403

(0.066)

0.024

(0.069)

0.008

(0.033)


(0.0003)

-0.0005 0.00004

(0.00005)

-0.0002*

(0.0001)

0.000002

(0.000005)

Research &

Development

-0.861

(1.779)

GDP/income lag 0.253

(0.35)

0.311

(0.357)

-0.319 0.026

(0.066)

0.044

(0.069)

0.005

(0.149)

0.00002

(0.0106)

0.003

(0.011)

0.0046

(0.006)

Lagged dependent

variable

-0.0054

(0.215)

-0.075

(0.231)

1.003 0.6009***

(0.064)

0.595***

(0.064)

0.621***

(0.169)

0.968***

(0.015)

0.968***

(0.01)

0.98***

(0.009)

Observations 33 33 6 177 177 40 204 204 38


173

Table 3 continued: Mathematics Panel and Technology Diffusion.

(7)

Visitor Beds

Variables (1) (2) (3)


(0.08)

-0.003

(0.011)

0.185***

(0.185)

Years of

Schooling

1.119**

(0.508)

0.892*

(0.519)

1.651

(1.147)


(0.368)

0.729*

(0.372)

0.106

(0.704)

FDI 0.008

(0.063)

0.017

(0.062)

0.055

(0.193)


(0.411)

0.468

(0.412)

0.646

(1.178)


(0.581)

-0.123

(0.578)

0.120

(1.22)


(0.00007)

0.00005

(0.0005)

Research &

Development

-4.998

(4.26)


(0.095)

0.036

(0.098)

0.00005

(0.005)

Lagged

dependent

variable

0.516***

(0.121)

0.497***

(0.12)

0.968***

(0.23)

Observations 100 100 28


174

Table 4: Science Panel and Technology Diffusion.

(1)

Cable TV

(2)

Computer

(3)

Transplant Bone marrow

Variables (1) (2) (3) (1) (2 (3) (1) (2) (3) Cognitive Skills -0.0016

(0.001)

-0.001

(0.002)

0.001

(0.024)

-0.003**

(0.0012)

-0.002

(0.0014)

-0.008

(0.006)

-0.003*

(0.002)

-0.002

(0.002)

0.001

(0.007)

Years of

Schooling

0.823*

(0.37)

0.887**

(0.371)

0.505

(1.063)

-0.009

(0.205)

0.008

(0.2054)

-0.543

(0.424)

0.12

(0.542)

0.316

(0.543)

2.421**

(1.105)


(0.215)

0.286

(0.215)

0.176

(0.459)

0.073

(0.112)

0.058

(0.113)

-0.021

(0.188)

0.849***

(0.264)

0.805***

(0.26)

0.949*

(0.533)

FDI 0.046

(0.041)

0.044

(0.042)

-0.044

(0.099)**

0.003

(0.028)

0.001

(0.028)

-0.044

(0.049)

0.045

(0.042)

0.044

(0.041)

-0.037

(0.067)


(0.369)

0.500

(0.37)

2.221

(0.988)

0.088

(0.177)

0.089

(0.176)

0.258

(0.423)

-0.617

(1.049)

-0.274

(1.047)


(0.317)

-0.222

(0.328)

-0.062

(0.866)

0.375

(0.195)

0.405**

(0.197)

-0.571

(0.468)

GDP Per capita -0.0001

(0.00004)

0.0001

(0.001)

-0.00003

(0.00002)

-0.00001

(0.00006)

-0.00009*

(0.00005)

-0.00005

(0.00006)

Research &

Development

-0.066

(2.011)

3.152***

(0.851)

3.874***

(1.441)

GDP/income lag 0.001

(0.036)

0.036

(0.088)

0.014

(0.023)

0.008

(0.238)

0.047

(0.039)

-0.029

(0.035)

-0.046

(0.035)

0.012

(0.03)

Lagged

dependent

variable

0.600***

(0.060)

0.608***

(0.06)

0.400***

(0.117)

0.847***

(0.044)

0.851***

(0.044)

0.886***

(0.059)

0.687***

(0.089)

0.656***

(0.089)

0.201

(0.142)

Observations 134 134 50 194 194 68 67 67 17


175

Table 4 continued: Science Panel and Technology Diffusion.

(4)

Transplant Lung

(5)

Fertilizer

(6)

Tractor

Variables (1) (2) (3) (1) (2) (3) (1) (2) (3)

Cognitive Skills -0.001

(0.0084)

-0.002

(0.0088)

0.028

(0.041)

0.0012

(0.001)

0.00008

(0.001)

-

0.0012***

(0.003)

0.00052

(0.0005)

0.0006

(0.00056)

-0.161

(0.017)

Years of

Schooling

5.772***

(2.162)

5.588**

(2.371)

2.273

(5.674)

0.726***

(0.268)

0.674**

(0.263)

-0.139

(0.22)

0.117

(0.081)

0.123

(0.082)

0.210

(0.404)


(1.019)

0.313

(1.191)

0.689

(2.025)

0.458***

(0.141)

0.424***

(0.138)

0.105

(0.165)

-0.028

(0.065)

-0.025

(0.065)

0.714

(0.244)

FDI 0.067

(0.133)

0.066

(0.134)

0.118

(0.268)

0.033

(0.037)

0.056

(0.0372)

-0.013

(0.026)

-0.007

(0.014)

-0.008

(0.014)

-0.098

(0.052)


(0.205)

0.005

(0.201)

-0.072

(0.255)

-0.028

(0.079)

-0.035

(0.079)

0.210

(0.434)


(2.973)

2.787

(3.008)

0.196

(0.213)

0.226

(0.214)

0.556

(0.262)

0.060

(0.084)

0.076

(0.087)

-0.297

(0.486)


(0.0002)

-0.0001

(0.0009)

0.00008***

(0.0002)

0.00006

(0.0003)

-0.000006

(0.00001)

-0.000005

(0.0007)

Research &

Development

0.300

(0.465)

0.353

(0.889)


(0.112)

-0.021

(0.114)

0.03

(0.124)

0.025

(0.026)

0.047*

(0.026)

0.08***

(0.019)

-0.002

(0.079)

-0.004

(0.012)

0.026

(0.088)

Lagged

dependent

variable

0.211

(0.167)

0.201

(0.176)

-0.310

(0.731)

0.785***

(0.039)

0.74***

(0.041)

1.024***

(0.064)

0.977***

(0.018)

0.980***

(0.019)

0.912***

(0.113)

Observations 48 48 11 215 215 51 210 210 42


176

Table 4 continued: Science Panel and Technology Diffusion.

(7)

Visitor Beds

Variables (1) (2) (3)


(0.004)

-0.001

(0.005)

0.035

(0.031)


(0.446)

0.735

(0.452)

0.710

(1.053)

Life Expectancy 0.735**

(0.315)

0.819**

(0.309)

0.650

(0.572)

FDI 0.233

(0.600)

0.035

(0.058)

-0.076

(0.134)


(0.385)

0.377

(0.375)

1.447

(1.060)


(0.525)

0.066

(0.508)

-0.165

(0.981)


(0.0006)

-0.00007

(0.0004)

Research &

Development

-8.153**

(3.696)

GDP/income lag

-0.028

(0.087)

-0.012

(0.085)

-0.813**

(0.315)

Lagged dependent

variable

0.603***

(0.097)

0.531***

(0.103)

0.898***

(0.216)

Observations 101 101 34


177

Appendix F. Definitions and Descriptive Summary.

Table 1: Sample for Advanced Mathematics.

Country

Number of

Students

Number of

Classes

Number of

Schools

Average Student per

Class

Armenia 858 77 38 11.14

Iran 2425 197 119 12.31

Italy 2143 130 91 16.48

Lebanon 1612 242 212 6.66

Netherland 1537 118 112 13.02

Norway 1932 120 107 16.1

Philippines 4091 118 118 34.66

Russia 3185 143 143 22.27

Slovenia 2156 95 79 22.69

Sweden 2303 150 116 15.35

178

Appendix F (continued), Table 2: Definition and Data Sources:

Variable Name Definition Source

Raw Mathematics Test

scores

Number of Score points obtained by a student on the

advanced mathematics items in his or her assigned

booklet

TIMSS and PIRLS, International

Study Center, Lynch School of

Education , Boston College (2009)

179

Appendix F (continued), Table 3: Descriptive Summary for Advanced Mathematics Raw

Test Scores

Country Observations Mean

Standard

Deviation Min Max

Armenia 858 10.7634 6.088512 0 31

Iran 2425 14.85732 7.387345 0 37

Italy 2143 11.38964 5.992668 0 33

Lebanon 1612 17.59988 5.538585 3 36

Netherland 1537 18.53155 5.285145 3 36

Norway 1932 11.25414 5.322544 0 34

Philippines 4091 7.846248 4.483501 0 37

Russia 3185 19.26217 7.461713 1 39

Slovenia 2156 12.26438 6.085062 0 34

Sweden 2303 10.39687 5.666493 0 33

180

Appendix G.

Table 1: Combined/Cross Country Human Capital Inequality Indices

GE(0) GE(1) GE(2)

Within Between Within Between Within Between

Raw Test Scores 0.12793 0.04695 0.10467 0.04591 0.10397 0.04614

% Share 73.15302 26.84698 69.51122 30.48878 69.26254 30.73746

181

Appendix G (continued)

Table 2: Within and Between School Human Capital Indices.

Country Name

GE(0) GE(1) GE(2)

Within Between Within Between Within Between

Armenia 0.10552 0.0786 0.07999 0.07709 0.07748 0.08002

% Share 57.31045 42.68955 50.9231 49.0769 49.19365 50.80635

Iran 0.09096 0.05318 0.07135 0.05386 0.06607 0.05672

% Share 63.10531 36.89469 56.98427 34.28826 53.80731 46.19269

Italy 0.10603 0.05574 0.08584 0.05257 0.08512 0.05264

% Share 65.54367 34.45633 62.01864 37.98136 61.78862 38.21138

Lebanon 0.03781 0.01909 0.03332 0.01819 0.03182 0.01767

% Share 66.44991 33.55009 64.68647 35.31353 64.29582 35.70418

Netherland 0.03891 0.00484 0.03638 0.00483 0.03579 0.00485

% Share 88.93714 11.06286 88.27954 11.72046 88.06594 11.93406

Norway 0.1014 0.02032 0.09034 0.01899 0.09301 0.01814

% Share 83.30595 16.69405 82.63057 17.36943 83.67971 16.32029

Philippines 0.09789 0.05828 0.0847 0.06236 0.09186 0.07087

% Share 62.68169 37.31831 57.59554 42.40446 56.44933 43.55067

Russia 0.05483 0.03848 0.04364 0.03615 0.03982 0.03519

% Share 58.76112 41.23888 54.69357 45.30643 53.08626 46.91374

Slovenia 0.09547 0.04184 0.08237 0.03858 0.08495 0.03692

% Share 69.5288 30.4712 68.10252 31.89748 69.70542 30.29458

Sweden 0.13544 0.02306 0.11979 0.0225 0.12486 0.02247

% Share 85.4511 14.5489 84.18722 15.81278 84.74852 15.25148

182

Appendix G (continued)

Table 3Skill-Inequality, Average scale scores, Educational, Income, Spatial and Categorization.

Rank County

Within-School

Inequality Average Scale Score

Type of Educational

System

Income

Categories Spatial Categories

1 Lebanon 0.03781 545 Centralized Upper Middle Middle East

2

Netherland

s 0.03891 552 Decentralized High Europe

3 Russia 0.05483 561 Centralized Upper Middle Central Asia/

Europe

4 Iran 0.09096 497 Centralized Upper Middle Middle East

5 Slovenia 0.09547 457 Centralized High Europe/ Central

Asia

6 Philippines 0.09789 355 Decentralized Lower Middle East Asia

7 Norway 0.1014 439 Decentralized High Europe

8 Armenia 0.10552 433 Centralized Lower Middle Europe/ Central

Asia

9 Italy 0.10603 449 Centralized High Europe

10 Sweden 0.13544 412 Decentralized High Europe

Source: TIMSS Advanced 2008 User Guide for International Database, Lynch School of Education, Boston College, World

Development Indicators (World Bank 2017)& Author's own calculations

183

Appendix H: Table 1: Description of Selected Variables Country Wise Analysis

Variables Description (TIMSS)

School variables


background Set of four categories: 0-10%, 11-25%, 26-50% and More than 50%


language Set of four categories: More than 50%, 76-90%, 50-75% and less than 50%

Location of School Set of Six categories: More than 500,000 people, 100,0001 to 500,000, 50,0001 to

100,000, 15,001 to 50,000, 3,001 to 15,000 and 3,000 people or fewer

Enrollment in the twelfth grade Total enrollment of twelfth graders in the school

Student-teacher ratio Ratio of total number of students to total number of teachers

Teacher variables

Teacher's Experience teaching mathematics Total number of years teaching mathematics at secondary school level

Teacher's job satisfaction Set of five categories: Very high, High, Medium, low and very low

Source: TIMSS 2008, School and Teacher’s questionnaires.

184

Appendix H, Table 2: Specifications Analysis Standard Variable Country Models

Armenia Iran Italy Lebanon Netherlands Norway Philippines Russia Slovenia Sweden

Teacher's

Experience

teaching

mathematics

-

0.001633** 0.000002 -0.001316* -0.000008 0.000027 -0.000367 -0.000204 -0.000181 -0.000061 -0.000082

(0.000553) (0.000705) (0.000528) (0.000146) (0.000214) (0.000300) (0.000661) (0.000245) (0.000702) (0.000811)

Enrollment 0.000014 0.000007 -0.000024 0.0000026 -0.000001 0.000003 -0.000002 0.000014 -0.000023 0.000004

(0.000042) (0.000009 (0.000021) (0.000003) (0.000005) (0.000014) (0.000001) (0.000009) (0.000026) (0.000012)

Student-teacher

ratio -0.000766 0.000006 -0.003708* 0.001071* 0.001059** 0.000762 0.000718 -0.000793 -0.001811 0.00094

(0.002890) (0.000928) (0.001785) (0.000606) (0.000484) (0.000889) (0.000493) (0.000569) (0.001231) (0.000852)

Teacher's job

satisfaction

category 1: very

high 0.057958 -0.00439 0.069399 0 0 0 0.024087 -0.012114 -0.012798 0

(0.054355) (0.02566) (0.076210) (0.033953) (0.012568) (0.041670)

category 2: high R -0.014196 0.073 0.004001 -0.010676 0.004477 0.006606 -0.002529 0.026368 -0.019908

(0.023658) (0.064969) (0.006356) (0.021959) (0.011836) (0.033003) (0.006168) (0.034328) (0.015967)

category 3:

medium 0.028498 0.004142 0.09887 0.014665** -0.007859 0.000951 0.012228 0 0.03612 0.005451

(0.024395) (0.024521) (0.064647) (0.007386) (0.022220) (0.015530) (0.033098) (0.032799) (0.023210)

category 4: low - 0 0.109631 0.052619* -0.02802 - 0.026842 - 0 -

(0.067251) (0.028341) (0.039630) (0.040642)

category 5: very

low - 0.007401 0 - - - 0 - - -

(0.041701)

Percentage of

Students with

language of test as

their native

language

category 1: More

than 90% 0 -0.033214 0.091462 0.007429 0 -0.126664*** -0.006528 -0.009607 0.012672 0.01402

(0.022766) (0.062210) (0.009369) (0.045210) (0.043482) (0.014062) (0.031354) (0.036231)

category 2: 75 to 0.013182 0 0.072651 -0.002784 0.008464 -0.111954** -0.025967 -0.002706 0 0.013826

185

90%

(0.056268) (0.067038) (0.012721) (0.006884) (0.045744)

(0.047052) (0.017216)

(0.036796)

category 3: 50 to

75% - -0.031311 0.104625 0 0.014171 0 0 0 -0.002581 0.053353

(0.036981) (0.090101) (0.030157) (0.043123) (0.040394)

category 4: Less

than 50% - -0.035318 0 0.009259 0.083112** - -0.011224 0.015529 0 0

(0.025001) (0.007782) (0.037281) (0.034453) (0.017892)

Percentage of

Students from

economically

disadvantaged

backgrounds

category 1: 0-10% 0.056264 0.02283 -0.03181 -0.009605 0.016366 -0.007601 0.015296 -0.03082 -0.044377

(0.033766) (0.013887) (0.020713) (0.010383) (0.033678) (0.012466) (0.017840) (0.020431) (0.033820)

category 2: 11-

25% 0.12927*** 0 -0.020549 0 0.010144 0 0.018825 -0.019155 -0.039745

(0.034944) (0.021397) (0.033351) (0.017930) (0.017885) (0.033744)

category 3: 26-

50% 0.065102* 0.019162 -0.020637 -0.000917 0.015977 0.01472 0.020702 0.009081 -0.071532*

(0.033766) (0.014139) (0.024161) (0.008051) (0.025440) (0.011880) (0.020680) (0.020501) (0.039742)

category 4: More

than 50% 0 -0.00419 0 -0.002743 0 0.009638 0 0 0

(0.014443) (0.006930)

(0.011099)

Location of School

category 1: More

than 500,000

people 0 -0.129809** 0.068527* 0 -0.004365 -0.017103 -0.008738 -0.044893** 0.022002 0.095756*

(0.050723) (0.031750) (0.016307) (0.023372) (0.011179) (0.018393) (0.024526) (0.057024)

category 2:

100,001 to 500,000

people 0.069101 -0.117807** 0.015872 0.004513 -0.003176 -0.027585 -0.006772 -0.044455** 0.028431 0.085479

(0.027640) (0.051227) (0.027471) (0.008286) (0.012781) (0.021389) (0.011662) (0.018152) (0.034113) (0.055233)

category 3: 50,001

to 100,000 people 0 -0.113181** 0.018807 0.014599* 0.000075 0 0.00701 -0.038807** 0.017676 0.088487

(0.053621) (0.025735) (0.007690) (0.012797) (0.012198) (0.018840) (0.023190) (0.056703)

186

category 4: 15,001

to 50,000 people -0.000699 -0.121972** 0.007481 0.014112* -0.001248 -0.007402 -0.002474 -0.03962** 0.023954 0.087853

(0.027640) (0.055779) (0.024845) (0.007419) (0.012111) (0.019624) (0.011677) (0.019149) (0.022174) (0.054299)

category 5: 3,001

to 15,000 people 0.082716** -0.122557** 0 0.014552** 0 -0.020589 0 0 0 0.066411

(0.031469) (0.055310) (0.007335) (0.019618)

(0.055066)

category 6: 3,000

people or fewer 0 0 - 0.009783 0 -0.005398 -0.022363 -0.072465* - 0

(0.009367) (0.025632) (0.026207) (0.038653)

constant 0.035071 0.256234*** 0.040009 0.003475 0.01752 0.230927*** 0.076853 0.095515*** 0.105733* 0.060602

(0.054753) (0.063138) (0.095796) (0.012951) (0.043773) (0.052471) (0.052842) (0.030811) (0.058028) (0.074995)

N 26 108 91 169 84 103 103 135 65 83

R2 0.74 0.2 0.27 0.13 0.3 0.15 0.13 0.19 0.31 0.27

Mean log deviation (GE0) as a measure of inequality in educational quality for school is the dependent

variable. Standard errors in parenthesis, *,**,*** imply 10%, 5%, and 1% significance levels respectively. The

reference category for categorical variable states 0 and the categories that do not exist in the data for the

country are marked blank (-).

187

Appendix H (continued), Table 3: Specification Tests Netherlands

Models I II III IV V VI VII VIII IX X XI XII

Student-teacher ratio 0.0007157** 0.00067* 0.0019*** 0.00084** 0.0019*** 0.0018*** 0.00181*** 0.002*** 0.0021*** 0.00214***

(0.0003247) (0.0003) (0.0004) (0.0004) (0.005) (0.0005) (0.0004) (0.0004) (0.0005) (0.0004)

Enrollment in the twelfth

grade

-

0.0000015

-

0.00023****

-

0.00024***

-

0.00023** -0.0016**

-

0.0002**

-

0.00022** -0.00021**

(0.00006)

(0.00008)

(0.00086) (0.00008) (0.00008) (0.00008) (0.00008) (0.00008)

Percentage of Students from

economically disadvantaged

backgrounds

category 1: 0-10%

-0.0013

-0.0082

-

0.0282*** -0.0286***

(0.01)

(0.01) (0.0102) (0.0098_

category 2: 11-25%

-0.0074

-0.0162

-

0.0356*** -0.0362***

(0.01)

(0.011) (0.0107) (0.0101)

category 3: 26-50%

-0.0298** -0.0265**

(0.0142) (0.0128)

category 4: More than 50%

0.0199

0.0199

(0.014)

(0.014)

Location of School

category 1: More than 500,000

people

0.0359*

(0.0197)

category 2: 100,001 to

500,000 people

0.0218

(0.019)

category 3: 50,001 to 100,000

people

0.0251

(0.0192)

category 4: 15,001 to 50,000

people

0.0204

(0.0189)

category 5: 3,001 to 15,000

people

0.0154

(0.0204)

category 6: 3,000 people or

fewer

188

Teacher's Experience teaching

mathematics

0.0003 0.00059

(0.0001) (0.00017)

Percentage of Students with

language of test as their native

language


0.0012

(0.009)

category 2: 75 to 90%

0.006

(0.0101)


category 4: Less than 50%

0.0627***

(0.0151)


category 1: very high

0.0263 0.231

(0.028) (0.026)

category 2: high

-0.00349

0.0194 0.0186 -0.0105

(0.867)

(0.019) (0.01) (0.019)

category 3: medium

-0.0024

0.0182 0.0185 -0.0111

(0.021)

(0.019) (0.018) (0.0197)

category 4: low

-0.0195

-0.0358

(0.029)

(0.029)

constant 0.0276**** 0.0369*** 0.0279*** 0.0392*** 0.0294*** 0.0295 0.0066 0.0119 0.0041 0.046* 0.0536*** 0.0548***

(0.004) (0.0049) (0.0054) (0.009) (0.005) (0.022) (0.018) (0.019) (0.198) (0.023) (0.0101) (0.0092)

N 110 101 103 99 101 101 100 99 99 95 91 97

R2 0.043 0.038 0.066 0.129 0.061 0.187 0.139 0.32 0.25 0.235 0.243


189

Appendix H (continued), Table 4: Composition of Human Capital Inequality at Micro School Level.

Country High Skill Achieving Schools

(%)

Low Skill Disparity Schools (%) High Achieving & Low Skill Disparity Schools (%)

Lebanon 50 58.1 35.5

Netherlands 50.89 59.82 38.39

Russia 50.34 58.74 45.45

Iran 39.94 58.82 36.13

Slovenia 50.63 62.02 37.97

Philippines 41.52 55.93 28.81

Norway 52.33 56.07 30.84

Armenia 39.47 55.26 31.57

Italy 43.95 50 29.67

Sweden 47 28 28.44

190

Appendix H (continued): Summary Statistics for Country-wise analysis:

Table 1 Lebanon

Variables Observations Mean

Standard

Deviation Min Max

Mean log Deviation 211 0.034146 0.0264132 0 0.18168

Location of School

category 1: More than 500,000 people 203 0.142857 0.3507922 0 1

category 2: 100,001 to 500,000 people 203 0.128079 0.3350037 0 1




category 6: 3,000 people or fewer 203 0.083744 0.2776881 0 1

Student-teacher ratio 212 6.270047 4.034258 0.5 31

Teacher's Experience teaching mathematics 202 30.46535 16.54487 4 129


category 1: very high 205 0.160976 0.3684081 0 1

category 2: high 205 0.609756 0.488999 0 1

category 3: medium 205 0.22439 0.4182014 0 1

category 4: low 205 0.004878 0.069843 0 1

Table 2 Netherlands


Standard

Deviation Min Max

Mean log Deviation 112 0.036873 0.0195631 0.00492 0.12248



category 1: 0-10% 99 0.666667 0.4738035 0 1

category 2: 11-25% 99 0.252525 0.4366719 0 1

category 3: 26-50% 99 0.040404 0.197907 0 1

category 4: More than 50% 99 0.040404 0.197907 0 1

Enrollment in the twelfth grade 101 69.30693 30.55642 20 186

Student-teacher ratio 110 12.74091 5.709614 2 27

191

Table 3 Russia


Standard

Deviation Min Max

Mean log Deviation




category 2: 75 to 90% 143 0.083916 0.2782365 0 1

category 3: 50 to 75% 143 0.041958 0.201198 0 1

Less than 50% 143 0.06993 0.2559255 0 1

Location of School








Table 4 Iran


Standard

Deviation Min Max


Location of School







Enrollment in the twelfth grade 113 205.6637 186.6912 8 700


192

Table 5 Slovenia


Standard

Deviation Min Max




category 1: 0-10% 70 0.242857 0.4319056 0 1

category 2: 11-25% 70 0.442857 0.5003105 0 1

category 3: 26-50% 70 0.2 0.4028881 0 1



category 1: very high 75 0.053333 0.2262105 0 1

category 2: high 75 0.426667 0.4979236 0 1

category 3: medium 75 0.493333 0.5033223 0 1

category 4: low 75 0.026667 0.1621922 0 1



Table 6 Philippines


Standard

Deviation Min Max




category 1: 0-10% 112 0.160714 0.3689179 0 1

category 2: 11-25% 112 0.142857 0.3514998 0 1

category 3: 26-50% 112 0.232143 0.4240972 0 1




193

Table 7 Norway


Standard

Deviation Min Max





category 2: 75 to 90% 106 0.113208 0.3183515 0 1

category 3: 50 to 75% 106 0.009434 0.0971286 0 1



Table 8 Armenia


Standard

Deviation Min Max




category 1: 0-10% 37 0.189189 0.3970613 0 1

category 2: 11-25% 37 0.243243 0.4349588 0 1

category 3: 26-50% 37 0.297297 0.4633732 0 1


Location of School








194

Table 9 Italy


Standard

Deviation Min Max




category 1: 0-10% 91 0.395604 0.4916892 0 1

category 2: 11-25% 91 0.296703 0.4593354 0 1

category 3: 26-50% 91 0.153846 0.3628001 0 1




Table 10 Sweden


Standard

Deviation Min Max




category 1: 0-10% 91 0.494506 0.5027397 0 1

category 2: 11-25% 91 0.384615 0.4891996 0 1

category 3: 26-50% 91 0.087912 0.2847358 0 1


Percentage of Students with language of test

as their native language


category 2: 75 to 90% 114 0.289474 0.4555204 0 1

category 3: 50 to 75% 114 0.078947 0.2708471 0 1

category 4: Less than 50% 114 0.061404 0.2411289 0 1


195

Appendix I Cross-Country Analysis: Individual Variable Regression Models

Table 1: Human Capital Inequality and Student-teacher ratio

Human Capital Inequality

Student-teacher ratio 0.00139***

(0.0001)

Constant 0.0547***

(0.0024)

N 1093

R2 0.0457


Table 2: Human Capital Inequality and Teacher's Experience teaching mathematics


Teacher's Experience teaching mathematics -0.00076***

(0.0001)

constant 0.09163***

(0.0024)

N 1093

R2 0.0457


Table 3 Human Inequality and Percentage of Students from economically disadvantaged

backgrounds



disadvantaged backgrounds

category 1: 0-10% -0.0143***

(0.0044)

category 2: 11-25% -

category 3: 26-50% -0.0018

(0.0053)

category 4: More than 50% -0.01405***

(0.0046)

constant 0.0833***

(0.0033)

N 962

R2 0.0164


196

Table 4 Human Capital Inequality and Location of School


Location of School

category 1: More than 500,000 people -0.00528

(0.005)

category 2: 100,001 to 500,000 people -0.00507

(0.006)

category 3: 50,001 to 100,000 people -

category 4: 15,001 to 50,000 people -0.00567

(0.005)

category 5: 3,001 to 15,000 people 0.0037

(0.006)

category 6: 3,000 people or fewer -0.00843

(0.010)

Constant 0.0769***

(0.004)

N 932 R2 0.0051


Table 5 Human Capital Inequality and Enrollment in Twelfth grade


Enrollment in the twelfth grade 0.00003***

(0.000006)

Constant 0.0715***

(0.001)

N 1101

R2 0.0328


197

Table 6 Human Capital Inequality and Teacher’s job satisfaction



category 1: very high -0.0277

(0.014)

category 2: high -0.0379***

(0.0127)

category 3: medium -0.0212*

(0.012)

category 4: low

-

category 5: very low -0.0337

(0.0243)

constant 0.1249***

(0.012)

N 324

R2 0.0419


198

Table 7 Human Capital Inequality and Percentage of Students with language of test as





category 1: More than 90% -0.00828

(0.006)

category 2: 75 to 90% -

category 3: 50 to 75% -0.0113

(0.009)

category 4: Less than 50% -0.0182***

(0.0064)

Constant 0.086***

(0.0056)

N 950

R2 0.0111


199

Table 8: Results Cross-Country Analysis

Human Capital Inequality I II III IV V VI VII

Student-teacher ratio 0.00139*** 0.00109*** 0.00106*** 0.0011*** 0.00076*** 0.00036 0.00042

(0.00015) (0.00016) (0.00017) (0.00017) (0.0002) (0.00031) (0.00036)

Teacher's Experience teaching

mathematics -0.0005*** -0.0006*** -0.00063*** -0.00062*** 0.00013 0.0001

(0.00011) (0.00012) (0.00017) (0.00012) (0.00043) (0.00045)

Percentage of Students from

economically disadvantaged

backgrounds

category 1: 0-10%

-0.0173*** -0.01265** -0.01536 - -0.0199**

(0.00425) (0.00517) (0.00463)

(0.00919)

category 2: 11-25%

- 0.00383 - 0.0201** -

(0.00557)

(0.0091)

category 3: 26-50%

0.00037** - -0.00619 0.0178** -0.0036

(0.00516)

(0.0056) (0.00905) (0.00913)


-0.01066 -0.00625 -0.01439*** 0.01794** -0.0025

(0.00453) (0.00533) (0.00498) (0.00847) (0.00864)

Location of School

category 1: More than 500,000

people

0.01357 -0.00486 -0.03909 -0.03791

(0.01073) (0.00556) (0.0249) (0.02509)

category 2: 100,001 to 500,000

people

0.01336 -0.00446 -0.03432 -0.0325

(0.01082) (0.00569) (0.02513) (0.0254)

category 3: 50,001 to 100,000

people

0.0173 - -0.01998 -0.0198

(0.0111)

(0.0256) (0.02592)

category 4: 15,001 to 50,000

people

0.01308 -0.00371 -0.03048 -0.0297

(0.01077) (0.00567) (0.02503) (0.0253)

category 5: 3,001 to 15,000

people

0.01944* 0.00311 -0.02978 -0.02773

(0.0111) (0.00648) (0.02587) (0.02637)

200

category 6: 3,000 people or

fewer

-0.01448 - -

(0.01106)

Enrollment in the twelfth grade

0.000022*** 0.0000002 0.000003

(0.000006) (0.000007) (0.0000079)


category 1: very high

-0.0061 -0.0030113

(0.0171) (0.0256)

category 2: high

-0.02000 -0.01795

(0.0163) (0.0248)

category 3: medium

-0.0074 -0.00408

(0.01691) (0.025)

category 4: low

- -0.00337

(0.0287)

category 5: very low

0.00423 -

(0.02868)

Percentage of Students with

language of test as their native

language


-0.00327

(0.01962)


0.02089

(0.02535)


category 4: Less than 50%

-0.00200

(0.01943)

Constant 0.0547*** 0.0689*** 0.0766*** 0.0521*** 0.0752*** 0.115*** 0.133***

(0.00245) (0.0041) (0.0053) (0.0117) (0.0068) (0.0314) (0.0433)

N 1093 1089 924 748 737 214 210

R2 0.0457 0.0832 0.1269 0.1547 0.1688 0.1054 0.1194

Human Capital, Technology and Inequality › 122964 › 1 › Zainab_Asif_Thesis.pdf · Zainab Asif...

Documents

Transcript of Human Capital, Technology and Inequality › 122964 › 1 › Zainab_Asif_Thesis.pdf · Zainab Asif...