jbm-vol-16-01

Published by Chapman University’s Argyros School of Business and EconomicsSponsored by the Western Decision Sciences Institute

Vol. 16 No. 1

Journal of Business and Management

J.B.M.

Editors

Amy E. Hurley-Hanson, Ph.D.Cristina M. Giannantonio, Ph.D.

WDSI

WESTERN DECISION SCIENCES INSTITUTE

The Western Decision Sciences Institute is a regional division of the Decision SciencesInstitute. WDSI serves its interdisciplinary academic and business members primarilythrough the organization of an annual conference and the publication of the Journal ofBusiness and Management. The conference and journal allow academicians and busi-ness professionals from all over the world to share information and research withrespect to all aspects of education, business, and organizational decisions.

PRESIDENTMahyar Amouzegar

California State University, Long Beach

PRESIDENT-ELECTNafisseh Heiat

Montana State University-Billings

PROGRAM CHAIR/VICE PRESIDENT FOR PROGRAMS/PROCEEDINGS EDITORJohn Davies

Victoria University of Wellington

VICE PRESIDENT FOR PROGRAMS-ELECTSheldon R. Smith

Utah Valley State College

VICE PRESIDENT FOR MEMBER SERVICESDavid Yen

Miami University of Ohio

SECRETARY/TREASURERRichard L. Jenson

Utah State University

DIRECTOR OF INFORMATION SYSTEMSAbbas Heiat

Montana State University - Billings

IMMEDIATE PAST-PRESIDENTG. Keong Leong

University of Nevada, Las Vegas

REGIONAL VICE PRESIDENTVijay Kannan

Utah State University

WDSI

Journal of Business and Management – Vol. 16, No. 1, 2010

Journal of Business and ManagementVolume 16, Number 1 2010

EDITORS

Amy E. Hurley-Hanson, Chapman UniversityCristina M. Giannantonio, Chapman University

Journal of Businessand Management

EDITORS

Amy E. Hurley-Hanson, Chapman UniversityCristina M. Giannantonio, Chapman University

EDITORIAL BOARD

Nancy BorkowskiFlorida International University

Krishna S. DhirBerry College

Sonia M. GoltzMichigan Tech University

Miles G. NichollsRMIT University

Richard L. JensonUtah State University

Terri A. ScanduraUniversity of Miami

Jeffrey A. SonnenfeldYale University

Victor H. VroomYale University

PAST EDITORS

Burhan Yavas, California State University Dominguez Hills 1993-1999Raymond Hogler, Colorado State University 2000-2004

EDITORIAL STAFF

Rosalinda Monroy, Chapman University Publications Jaclyn Witt, Editorial Assistant

J.B.M.


We would like to thank the many ad hoc reviewers who shared their expertise toreview the manuscripts submitted to JBM over the past few years. Their time andeffort greatly contributed to the Journal of Business and Management.

iii

Hank AdlerChapman University

David AhlstromThe Chinese University of Hong Kong

Ayca AltintigChapman University

Susan Wills AmatUniversity of Miami

Queen E. BookerMinnesota State University – Mankato

William J. CarnesMetropolitan State College of Denver

Natasa ChristodoulidouCalifornia State University Dominguez Hills

Richard ClodfelterUniversity of South Carolina

John DaviesVictoria University of Wellington

Don W. Davis Sr.Penn State University

Jeanette A. DavyWright State University

Kathy Lund DeanIdaho State University

Richard F. DeckroAir Force Institute of Technology

Lidija DediUniversity of Zagreb

Stephanie DellandeUniversity of New Orleans

Peter T. DiPaoloNova Southeastern University

Rafik Z. EliasCalifornia State University Los Angeles

Grigori ErenburgChapman University

Lori Baker-EvelethUniversity of Idaho

Debora J. GilliardMetropolitan State College of Denver

Thomas E. GriffinNova Southeastern University

Mary L. GrimesSouth Carolina State University

Gary HackbarthIowa State University

Owen P. Hall Jr.Pepperdine University

John HathornMetropolitan State College – Denver

J. Kline HarrisonCalloway School of Business and Accountancy

Mary F. HazeldineGeorgia Southern University

Abbas HeiatMontana State University – Billings

Roger W. HuttArizona State University –

Polytechnic Campus

Waithaka N. IrakiUniversity of Nairobi

Ronald D. JohnsonNorth Dakota State University

Stefanie K. JohnsonUniversity of Colorado – Denver

Gary M. KernIndiana University – South Bend

Donald KerrGriffith Business School

Steven KoHong Kong Polytechnic University

Austin KwagUtah State University


Jennifer LeonardMontana State University – Billings

June LuUniversity of Houston – Victoria

Christine B. MahoneyCleveland State University

Ronald G. McGarveyRAND

Paul F. MessinaUniversity of Texas, San Antonio

John G. MichelNotre Dame University

James MooreIndiana University-Purdue University,

Fort Wayne

Lewis A. Myers Jr.St. Edward’s University

Niklas MyhrChapman University

Carmela R. NantonPalm Beach Atlantic University

Vivek S. NatarajanLamar University

Paul L. NesbitMacquarie Graduate School of Management

Prashanth NyerChapman University

Wilhelmina N. OkunborUniversity of Maryland – Eastern Shore

Asbjorn OslandSan Jose State University

Melanie C. PageOklahoma State University

Mahour Mellat ParastUniversity of Nebraska – Lincoln

Ekin K. PellegriniUniversity of Missouri – St Louis

Richard PetersLouisiana State University – Shreveport

Karen L. ProudfordMorgan State University

Sharon L. PurkissCalifornia State University Fullerton

Kendra ReedLoyola University – New Orleans

Mooweon RheeUniversity of Hawaii

Barbara A. RibbensWestern Illinois University

Filippina RisopoulosSustainable University Graz

Melissa St. JamesCalifornia State University Dominguez Hills

Carol H. SawyerUniversity of La Verne

Paul L. SchumanMinnesota State University – Mankato

Sharon SegrestUniversity of Florida – St. Petersburg

Lois M. SheltonCalifornia State University Northridge

Sung J. ShimSeton Hall University

Ashraf I. ShiraniSan Jose State University

Kenneth SmallCoastal Carolina University

Faye L. SmithMissouri Western State University

Richard L. SmithUniversity of California – Riverside

James C. SpeeUniversity of Redlands

Gerald SteinerUniversity of Graz

Ahmad SyamilArkansas State University

Chong Leng TanUniversity of Idaho

iv


Liz ThachSonoma State University

Donna TillmanCalifornia State Polytechnic University

Pomona

Teresa C. TompkinsPepperdine University

Romica TrandafirTechnical University of Civil Engineering

Bucharest

Bruce O. TreadwayIPG Converting

William TsaoChung-Yuan Christian University

Francis D. TuggleChapman University

Nicholas W. TwiggMcNeese State University

Donna WileyCalifornia State University East Bay

John L. WilsonNova Southeastern University

Timothy L. WilsonUmeå University

Melien WuChung-Yuan Christian University

Hui-Hua Ou-YangChing Yun University

Limao YangHubei University

Chen Yen YaoShih Hsin University

Jeffrey D. YoungMount Saint Vincent University

Gail M. ZankTexas State University – San Marcos

v

Journal of Business and Management – Vol. 16, No. 1, 2010vi

Contents

Microcredit and Rural Women Entrepreneurship Development in Bangladesh:

A Multivariate Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

Sharmina Afrin, Nazrul Islam and Shahid Uddin Ahmed

Heterogeneity in Consumer Sensory Evaluation as a Base for Identifying

Drivers of Product Choice. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

Oded Lowengart

Group Attributional Style: A Predictor of Individual Turnover Behavior

in a Manufacturing Setting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

Laura Riolli and Steven M. Sommer

Business Failure Prediction for Publicly Listed Companies in China . . . . . . . . . 75

Ying Wang and Michael Campbell

Executive Compensation as a Moderator of the

Innovation – Performance Relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

Kathleen K. Wheatley and D. Harold Doty

7Journal of Business and Management – Vol. 16, No. 1, 2010 vii

9Afrin, Islam and Ahmed

Microcredit and Rural Women Entrepreneurship Development

in Bangladesh: A Multivariate Model

Sharmina Afrin

Khulna University, Bangladesh

Nazrul Islam

East West University

Shahid Uddin Ahmed

University of Dhaka

Microcredit programs have a positive socioeconomic impact on the ruralfemale borrowers of Bangladesh. This study suggests that the microcreditprograms do not help the borrowers to develop any entrepreneurialcapabilities other than survival. Thus, this paper aims at identifying thefactors related to the development of entrepreneurship among rural womenthrough the microcredit programs of providers. A multivariate analysistechnique (Factor Analysis) was conducted to identify the factors related toentrepreneurship development. Structural Equation Modeling (SEM) wasused to identify the relationship between microcredit programs and thedevelopment of rural female entrepreneurship in Bangladesh. Results showthat financial management skills are the most important factor and have asignificant relationship with the development of rural women andentrepreneurship. Results also show that the group identities of the femaleborrowers have a significant relationship with the rural entrepreneurshipdevelopment in Bangladesh. A borrower’s experience from the parents’families and the limitation of options also lead to the development ofentrepreneurship among the rural female borrowers of Bangladesh.

About 84% of the 140 million people living in Bangladesh reside in rural areas. Halfof this population is women. Men who live in the rural areas are primarily engaged inagricultural and related activities. Females however, remain idle in their houses due toa number of social and cultural barriers. They are discouraged from working outsideof their homes. This situation can be attributed to the dominant patriarchal society andstrong religious influence (Purdah) in Bangladesh (Ahmed et al., 1997; Cain &Khanam & Nahar, 1979). Barriers can also be attributed to the lack of access to funds,the knowledge of agro-based production technology and the market, as well as thesupport from other family members. Research shows that a large number of ruralwomen in Bangladesh are compelled by macroeconomic factors to enter into the labormarket. Hence, the overwhelming majority of women in Bangladesh are poor and alsocaught between two vastly different worlds: the world determined by culture andtradition that confines their activities inside family homesteads, and the world shapedby increasing landlessness and poverty that drive them outside into wage employment(Chowdhury, 1998).

In the last two decades, microcredit programs have been operated by government(GOs) and nongovernmental organizations (NGOs) in Bangladesh. The primeobjective of these programs is to enhance the income-earning potential of femaleborrowers of rural families, and empower them socially and economically. Thisprogram helped rural women working in paddy husking, poultry farming, pettytrading (e.g., grocery), pond aquaculture, animal husbandry, weaving, mini-garments,handicrafts, dairy farming, and plant nursery activities (which all tend to be home-based in nature). Microcredit programs substantially contribute to the socioeconomicdevelopment of the rural women in this country. Studies show that the microcreditprograms have created significant positive differences in the socioeconomic lives of therural women in Bangladesh (Hashemi, 1998; Schuler, Hashemi & Riley, 1997).Microcredit programs have also helped the rural women to be involved in home-basedeconomic activities, which in turn, have created enormous opportunities for them tobe independent and self-sufficient. Studies also show that the involvement of ruralwomen in home-based economic activities through microcredit programs has apositive socioeconomic impact on their lives, as well as their families. However, it isnot apparent whether these programs are actually making the rural female borrowersentrepreneurial or not (Hashemi, 1998).

The positive impact of microcredit programs can be discussed in two ways. Firstly,microcredit programs create employment opportunity, increased productivity, provideeconomic security, give nutritional and health status, and improve the housingconditions of the rural women. The positive impact on income has increased theirasset position and has created wealth for their families (Hulme & Mosely, 1998).Secondly, microcredit programs create a significant influence on rural women in thearea of social empowerment, awareness and education, self-esteem, sense of dignity,organizational and management skills, mobilization of collective strengths, etc. (Pitt &Khandaker, 1996). This positive socioeconomic change subsequently helps them to bemore independent and more financially solvent in their families and localities.

Microcredit providers assert that the important impact of their programs is thesustainable development of the socioeconomic lives of rural women. But the reality is

10 Journal of Business and Management – Vol. 16, No. 1, 2010

that the developments are hardly prolonged. Observation shows that rural women areunable to be completely self-reliant even if they are involved in microcredit programsfor a long period of time (i.e., 10 to 15 years). This indicates that the credit programsare making the women more dependent on the credit provider instead of making themindependent. Thus, concerns have been raised by the researchers about thesustainability of the socioeconomic developments of the rural women. These concernsare very much relevant to the development of rural women and their entrepreneurshipin Bangladesh.

The development of rural entrepreneurship in Bangladesh primarily depends onthe socioeconomic development of the people. It is necessary to develop ruralentrepreneurship in order to foster the development of the capabilities of theborrowers. Once the rural women are self-sufficient, they will be able to initiate theirown projects that result in self-independence. In order to encourage rural women’sentrepreneurship in a developing country like Bangladesh, three types of activitiesmight be performed. These activities include stimulatory, supporting, and sustaining(Rahman, 1979, 1999; Katz, 1991a). All three types of activities are partially performedby the microcredit providers that are helping the borrowers to survive. In addition, thedegree of the differences in sustainability is significant in both governmental andnongovernmental programs (Amin, 1994b).

For the development of rural female entrepreneurship, stimulatory supports areessential, as the women tend to be unaware of their capabilities. Interaction with theborrowers and the microcredit providers, as well as direct observation, education, andtraining in selecting products, projects, and other technoeconomic informationmotivate rural women to be more enthusiastic and entrepreneurial. The next step is tosupport the entrepreneurs and their different qualifications. Once the women areencouraged to engage in homestead economic activities, they require a different kindof support to start and run their own business. This support can be related to thesupply of scarce raw materials, access to different facilities, such as fund, technology,production methods and procedures, the marketing of products, reinvestments, etc.The question of sustainability comes at the third stage of the entrepreneurialdevelopment process. Once the business is run, rural female entrepreneurs requiresupport for sustaining their projects in order to foster growth in the future. Thesesustaining activities are related to the help in modernization, diversification,additional financing for full capacity utilization, deferring repayment/interest,diagnostic industrial extension, product reservation, new adventures for marketing,quality testing, and improving services. Rural women can benefit from the creditproviders by obtaining support facilities, which are helpful in order for them toincrease the level of sustainability of their economic activities. Therefore, the researchquestions of this study are as follows:

(i) Are the rural female borrowers becoming independent by their involvementin microcredit programs?

(ii) Are they gaining any knowledge from the income-generating projectsinitiated by the credit?


(iii) If not, how can the women borrowers be made entrepreneurial in operatinghome-based economic activities?

(iv) Is there any difference in the rural female entrepreneurship developmentbetween governmental and nongovernmental programs?

This study primarily focuses on how to identify the factors related to the developmentof entrepreneurship among the rural women borrowers of Bangladesh. The presentresearch also analyzes the sustainability of the socioeconomic impact on rural women,which is termed in this study as rural entrepreneurship development. The specificobjectives of the study are as follows:

1. To identify and explain the factors related to entrepreneurship developmentthrough microcredit programs.

2. To test the appropriateness of the factors.3. To develop a model related to the development of entrepreneurship among

the rural women through microcredit programs.4. To recommend a policy framework for the credit providers to develop rural

women entrepreneurship in Bangladesh.

Microcredit program and the entrepreneurship development

Over the last two decades, microcredit became an important tool for alleviatingpoverty in Bangladesh (Khandkar & Chowdhury, 1996). The overall success ofmicrocredit programs depends not only on immediate alleviation of poverty, but alsoon long-term sustainability. Long-term sustainability then depends on accumulationassets (Chowdhury, 2004). In Bangladesh, the Grameen Bank started microcreditprograms in 1976 as a pilot project. Now, more than 3000 nongovernmentalorganizations (NGOs), national commercial banks, and specialized financialinstitutions operate microcredit programs in Bangladesh. Such programs have provento be a strong means to alleviate poverty through the social and economicempowerment of rural and disadvantaged women (Puhazhendhi & Badatya, 2002).Such a group savings program can help the rural women to bring economic securityinto their lives (Secretary General, UN, 1998). The changing role of women shows asteady upward growth in the economic activities in Bangladesh (Arefin & Chowdhury,2008). Studies show that female entrepreneurs are doing better in the service sectorthan in the manufacturing sector in Bangladesh (Begum, 2002).

Microcredit is a structured program under which microlevel loans are given todisadvantaged residents, especially to poor rural women, without collateral security.It is a group-based and intensively supervised loan program. The uniqueness of thisloan program is that there is no requirement of collateral security from theborrowers. Anyone can apply for this credit and is also eligible to receive credit. It isa small loan that varies from Tk. 1,000.00 to Tk. 10,000.00 for each borrower. Thepurpose of this microcredit program is to give loans for self-employment thatgenerates income and allows them to care for themselves and their family members(Sankaran, 2005).


There are three C’s to the microcredit program: character, capacity and capital(Yunus, 2003). Character is defined as the historical record of the borrowers such ashow a borrower has handled his/her past debt obligations, his/her background, and aborrower’s honesty and ability to repay the loan. Capacity is termed as how much debta borrower can actually handle, according to their income, and still be able to pay thatdebt off. Capital is all the current available assets that the borrower has that will alsohelp him/her to repay the loan on time.

Microcredit programs have a significant impact on the income and economicsecurity of the lives of rural women. These programs increase income and help thefemale borrowers to spend more in order to foster the development of their families.Such programs also help to increase household income which in turn, improves theconsumption patterns and lifestyles of the families (Hossain & Sen, 1992; Navajas etal., 2000). The access to the microcredit program for rural women improves theirlifestyles through economic solvency and self-sufficiency; the single most importantneed of destitute women in Bangladesh (Apte, 1988). Microcredit encourages femaleborrowers to save for the future, which is an important source of capital accumulationfor the rural families and for the economy. Increased income indirectly improves thelevel of education of the borrowers and the awareness about consumption andsanitation needs as well. The improvement of education among the rural borrowershelps to increase consciousness about their health and the future of the nextgeneration. Credit programs increase productive resources for rural families and theirhousing conditions and also result in economic security for the borrowers.

The needs of low-income microcredit clients would be best served by highlyflexible financial services that enable them to conduct frequent transactions both forsmall savings and for borrowing at irregular intervals (Sinha, 2003). The mainobjective of microcredit providers is to create self-employment opportunities for therural unemployed women. These opportunities are largely in nonfarm relatedindustries. Before joining microcredit programs, many borrowers were employed asday laborers. Now they are more self-sufficient and can work on their own projects,whereas previously they had very little chance to participate in economic activitiesunder the socioeconomic conditions in Bangladesh. Microcredit programs havecreated the opportunity to reduce their dependency on others in their families. Theimmediate macroeconomic effect of microcredit is the reduction of labor supply andthe raising of the wage rate, given the local demand for labor. Wages remain at thehigh level if the credit program induces a large demand for food and other localproducts. Hence, the result of microcredit programs is the increase of placement inrural areas (Ghai, 1984).

Rural wage is a reflection of rural economic conditions. The growth of self-employment has been achieved at the expense of wage employment (Shahidur, 1998).The self-employment of borrowers was much higher than the reduction in wageemployment in rural areas. The immediate impact of microcredit is on the labor forceparticipation rate and the total hours worked. A survey on Grameen Bank shows thatmicrocredit programs generated new employment for about one third of its members(Hossain, 1986). Most of the new employment was created for the female borrowers.It has also reduced the dependency ratio in the village families. Rural development is


based on the investments that promote economic growth in rural areas. Increased farmproductivity is the main emphasis for this microcredit programs (Jha, 1991). Abilityand efficiency are considered here in order to denote the productivity of rural womenborrowers. Through this variable, an inquiry was made to discover whether theproduction of goods was increased by the borrowers after the involvement in credit-financed projects. In addition, women’s group memberships seriously shifted overalldecision-making patterns from norm-guided behavior and male decision-making to amore joint and female decision-making approach (Holvoet, 2005). In Vietnam, themicrocredit program has also reduced the poverty rate of the participants (Cuong,2008). Microcredit programs have increased the agricultural productivity of small andmarginal farm households. The use of high-yielding variety is higher among theborrowers, which helps them produce more products for the locality (Alam, 1988).

The nonfarming activities of Bangladesh include harvesting livestock, poultry,fisheries, trading, and shop keeping. The increase in shop keeping activities hasincreased the volume of trade in the rural areas. It is reported by the Grameen Bankthat 46% of its total trade loans given to the trade sector went to crop trading in 1985,while 22% went to livestock and fisheries. Trading and shop keeping activities have apositive impact on the development of local markets by boosting local production andcreating new market opportunities for selling those products locally (Shahidur et al.,1998). A housewife or part-time farmer can link this business to the local productionand consumption, as well as outside economic activity. The less fortunate are actuallyable to work and increase their working days after joining the rural credit programs(Hossain, 1988).

The empowerment of women is another main purpose of microcredit programs.Empowerment is about a change in favor for those who previously exercised littlecontrol over their lives. This change is two-sided. The first side is control overresources (financial, physical, and human), and the second is control over ideology(beliefs, values, and attitudes) (Sen, 1997). The next question is for whom are theempowerment benefits for? Such benefits are undoubtedly for the rural women inBangladesh who are governed by the two powerful forces of patriarchy and classstructures (Amin et al., 1994a). The literature on microcredit and femaleempowerment provides examples of a number of empowerment measures, includinga borrower’s control over loan (Goetz & Gupta, 1996; Montgomery, 1996), knowledgeof the enterprises accounts (Ackerly, 1995), mobility, intra-household decision makingpower, and general attitudes about children’s lives (Amin & Pebley, 1994b; Hashemi etal., 1996). A woman’s control over resources and incidence of domestic violence is alsoa factor (Naved, 1994).

Social empowerment is essential for the development of poor rural women inBangladesh. The positive argument is that microcredit programs help rural women tobe more socially empowered (Zaman, 1999; Acharya, 1994). Empowerment ischaracterized as the mobility of women, economic security, ability to make purchases,involvement in major household decisions, political and legal awareness, andinvolvement in public protest and political campaigns. Women’s participation in suchprograms increases their ability to visit market places for buying products, medicalcenters for medication, cinemas for watching movies, other homes in the village, and


outside villages for more social relations. Participation also enhances the ability of thewomen to make both small and large purchases. Small purchases include small itemsused for daily preparation for the family (e.g., kerosene oil, cooking oil, spices), foroneself (e.g., hair oil, soap, glass, etc), or items like ice cream or sweets for thechildren. The large purchases are usually things like pots and pans, children’s clothing,personal clothing (e.g., Saries), and a family’s daily food.

Microcredit increases the ownership of productive assets for the women. Themicrocredit programs also influence legal and political awareness and participation inpublic campaigns. Such campaigns are often for the members themselves, thechairman, the locale, and political leaders. The longer the involvement of a woman ina credit program, the greater the likelihood will be of that woman being empowered.She is likely to contribute more to not only her family, but to society as a whole in thelong run. Credit programs enable women to negotiate gender barriers that increase thecontrol of women over their own lives, improve their freedom in the family, andincrease their persuasive power. As a result, credit programs improve the relativepositions of women in their families, and in society as well.

Another positive result of microcredit programs is the improvement of nutritionand the health conditions of the rural women and their family members (Srinivasan &Bardhan, 1990; Hossain, 1986). Microcredit increases awareness about the access tomodern medication facilities. Tube well water is not normally used by the rural peoplein Bangladesh. Things such as sanitary latrines and urinals, which to some areeveryday conveniences, are a dream for the villagers. One of the major indicators ofpoverty is the nonavailability of such facilities. The rural credit providers usually tryto address this problem in order to improve the quality of life of the rural population.Studies show that the credit programs have even increased the daily intake of proteinand calories for the rural people (Shahidur, 1996). The children of microcreditborrowers tend to have better nutritional health compared to the children ofnonborrowers. Rural credit projects help increase the income of the rural women,which leads to higher food security and a better life overall. The ability to spend moreon sanitation and the health care activities also is increased by the use of creditprograms. Female borrowers can also improve their housing conditions from themoney they earn from the credit-supported projects. This is often considered to be aninsurance against rural poverty in Bangladesh.

Rural credit also increases education and awareness among the rural women. Theinvolvement of women in income-generation activities changes their attitudes also(Ahmed et al., 1997). With the help of fellow borrowers and loan providers, womenoften feel the need to further their education (an education that will likely benefit theirchildren, their husbands, and themselves). Credit programs actually increase thelikelihood for female education more than for male education (Pitt & Khandaker, 1996;Kabeer, 2001). Due to the increase in income, they then are able to send their childrento school also. Microcredit programs create awareness among the rural women throughinteractions with the group members and health workers. Because women are likely tobecome more educated after enrolling in a microcredit program, the use of contraceptionand birth control increases greatly. The exchange of ideas with others, social support forthe legitimization of innovative reproductive behavior, and group interactions encourage


rural women to use more contraception in their day-to-day lives (Amin et al., 1994a).Microcredit in turn, decreases the level of desire for additional children in rural families.Once a woman obtains economic security and is able to contribute to her family, she willhave the freedom of mobility, freedom from domination by only the family, better controlof her body, and birth control options. Mobility in the village, and being able to traveloutside of the village, helps women to seek family planning information, and other typesof educational assistance (Schular et al., 1997). Women earning independently andcontributing to their families are less insecure and less vulnerable to the threat thatabandonment by their husbands can pose. Acquiring their own money and other assetsmakes these women less fearful of the repercussions of having more children, shouldthey choose to do so. Studies show that in almost all cases, the impacts of microcreditare positive in terms of returns on investments, household income, employment in thenonagricultural sector, the labor force participation rate, socioeconomic empowerment,household expenditure and consumption patterns, human capital, and fixedinvestments (Hossain, 1988; Rahman, 1996).

Rural entrepreneurship is a key to economic development in many countries acrossthe globe (OECD, 1998, 2003; UN, 2004). About half of the population of Bangladeshis women who usually remain idle and unproductive within their homes. They haveno method of participation in the economy and no resources for income-generatingactivities except taking care of their family. Thus, these women can become moreproductive by getting involved in economic activities. By providing stimulatory andsustaining supports, these women can be made able to initiate businesses and otherincome-generating projects. Hence, both the developed and developing countries arefocusing more on groups such as rural women in order to engage them in income-generating activities (Chowdhury, 2002). Countries focus on female entrepreneurshipdevelopment by demonstrating that financial assistance can lead to reduced fertilityand an increase in the economic growth of the country.

Rural entrepreneurship has been defined by different scholars and has alsochanged over time in Bangladesh (Islam & Mamun, 2000). Studies show theshifting focus of entrepreneurial success factors. Before 1990, the focus was onpersonal and psychological factors, while after 1990, the focus was shifted tomanagerial and environmental factors. The common aspects found in thedefinitions are the entrepreneur, innovation, organization, value creation,opportunity taking, profit or nonprofit, growth, uniqueness, flexibility, dynamism,and risk taking propensity. These aspects can be put into overlapping typologies.There are five different perspectives of entrepreneurship, which include: (1) aneconomic function, (2) a form of behavior, (3) a set of characteristics, (4) a smallbusiness, and (5) creation of wealth (Ahmed & McQuaid, 2005; Deshpande &Joshi, 2002). In almost all definitions of entrepreneurship, there is agreement thatentrepreneurs behaviors include (1) initiative-taking, (2) organizing andreorganizing of social and economic mechanisms, and (3) the acceptance of risk orfailure.

Entrepreneurship has a wide range of meaning and has been debated amongscholars, educators, researchers and policy makers since the early 1700s when theterm was first coined. The idea of entrepreneurship is an elusive concept (McQuaid,


2002). Since the expectations and perspectives of various stakeholders are different,their views regarding enterprise, entrepreneurship and small business are alsodifferent. Rosa (1992) argued that the vagueness of enterprise definition has been tothe advantage of both government and academics in the 1990s in their attempts in theUK to change the national culture. Katz (1991b) commented on this debate, sayingthat small business is a subset of entrepreneurship, while others argue that smallbusiness commencement is an integral part of entrepreneurship. Kearney (1996)asserted that enterprise is the capacity and willingness to initiate and manage creativeaction in response to opportunities, wherever they appear, in an attempt to achieveoutcomes of added value. These outcomes can be personal, social, and cultural.Typically, enterprise involves facing degrees of uncertainty as well. The associated risksare not necessarily financial, but may be physical, intellectual, or emotional.

Innovation Innovation is an important characteristic for an entrepreneur. Austrian economist

Schumpeter (1949) defined entrepreneurship as focusing on innovation in fourdifferent areas such as new products, new production methods, new markets, and newforms of organization. Anyone who combines inputs in an innovative manner togenerate value to the society, results in a creation of some kind of wealth. Accordingto Schumpeter (1949), the use of new combinations defines enterprise and theindividuals whose function it is to carry them out. The Industrial Revolution alsoadded to this dimension in the entrepreneurial concept. Audretsch (1995) andCunningham and Lischeron (1991) emphasized the innovation issue of anentrepreneur. They identified three levels of the term of entrepreneurship: (1) smallfirms and enterprise level, (2) new firm formation, and (3) innovation and a system-wide coordination of complex production. Innovation and system-wide coordinationis also emphasized in other studies (Malechi, 1997; Casson 1990; Casson, 1999).Behavioral and social scientists also focused on risk-taking, innovation, and initiative-taking capabilities in their definitions of entrepreneurship (Weber, 1930; Hoselitz,1952; Chell, Haworth & Brearley, 1991, Gartner, 1988). These characteristics arerelated to the cognitive aspects of entrepreneurship.

Risk-taking Risk-taking is the prime factor for the success of an entrepreneur. When an

entrepreneur initiates a business venture, that person has to take risk and faceuncertainty. In the 18th century, the French term entrepreneur was first used byCantillon to describe a ‘go-between’ or a ‘between-taker’ whereby they boughtgoods at certain prices but sold at uncertain prices and when they purchased suchgoods at a given price, they could not be sure what price they would be able to sellthem for. So, he/she bore the risk and uncertainty of a venture, but kept the surplusafter the contractual payments had been made (Ahmed & McQuaid, 2005). In 1971,Peter F. Drucker also supported the view point of Cantillon and said that risk-takingis an important characteristic of an entrepreneur. Ahmed (1981) found anentrepreneur to be a risk-taker since he/she invests money and is involved inmaking decisions, the success of which brings rewards; and the failure of which


could lead to the loss of those rewards. An entrepreneur could also face the loss oftheir principal (i.e., invested money). Therefore, it is very logical to place risk-taking as the focal point of entrepreneurship. Hence, the person who takes risks inorder to establish new ventures, or who has the capability of taking moderate riskscan be defined as an entrepreneur (Ahmed, 1982; 1987). A person can also bedefined as entrepreneurial when they have a very strong eagerness to achieve, anidea which was emphasized by McClelland (1961). McClelland (1961) also foundthat achievement motivation is an important characteristic of a successfulentrepreneur. The person who strives to reach the top of the success ladder bytaking moderate risks is achievement and motivation-oriented. An entrepreneurshould not only initiate new business ventures, but also be able to run the businessefficiently. In this regard, Jean-Baptiste Say identified a few dimensions ofentrepreneurship, with the ideas proposed by Cantillon: planning, supervising,organizing, and even owning the factors of production. These activities are primarilyrelated to business management.

Opportunity-seekingAnother characteristic of an entrepreneur is opportunity-seeking. Stevenson

(2000) explained that entrepreneurship is an approach to management that can bedefined as the pursuit of opportunity without regard to the currently controlledresources. He examined five critical dimensions of business practices: strategicorientation, commitment to opportunity, control of resources, management structure,and reward philosophy, all of which are related to entrepreneurial development.

Entrepreneurship is the pursuit of a discontinuous opportunity involving thecreation of an organization with the expectation of value-creation for the participants.The entrepreneur is the individual or team that identifies the opportunity, gathers thenecessary resources, and is ultimately responsible for the performance of theorganization. As a catalyst agent, an entrepreneur creates the forces of change andutilizes it in accelerating the socioeconomic value-addition of a country throughresource utilization, employment generation, capital accumulation, andindustrialization (Rahman, 1979; 1996). Hence, self-employment is the result of thedevelopment of entrepreneurship. Entrepreneurs create employment for themselvesand for others in order to work with innovative and economic-centered projects.People who are self-employed and have ownership of the business are calledentrepreneurs (Chowdhury, 2002). They are the owners of the business enterprises aswell. In this regard, women entrepreneurs are defined as conventional entrepreneurs,radical proprietors, and domestic traders (Begum, 2003).

Therefore, it is evident that some definitions of entrepreneurship are concernedwith business development aspects, while some are concerned more with thebehavioral aspects of the entrepreneur (Ahmed & McQuaid, 2005). Businessdevelopment aspects can be defined by opportunity seeking, initiative taking forestablishing new business venture, and creating wealth. While, in contrast, behavioralaspects are related to achievement motivation, risk-taking propensity, inner urge to dosomething valuable for oneself and for the society as a whole. Essentially,entrepreneurship is the dynamic process of creating incremental wealth, which is


created by the individual. This can be achieved by adopting risks in terms of equity,time, and career commitment. It is the process of creating something new by devotingthe time and effort, assuming the accompanying financial, psychic, and social risks,and receiving the rewards of monetary, personal satisfaction, and independence.Hence, entrepreneurship can emerge through the actions of four factors. These are asupport system, socio-sphere system, resource system, and a self-sphere system. First,a support system includes structure, organizational goals /policies, activities, technicalcompetence, organizational climate, and style of functioning. A sociosphere systemincludes value orientation (which is defined by work) independence, initiative,innovations, and risk-taking norms. Third, a resource system, includes manpower,market, raw material, transport communication, other industries and enterprises,technology, and technical manpower. A self-sphere system includes motivation and skillwhere motivation is explained by personal efficiency, coping capability and skill isdefined by a selection of product/process, project development, and by establishingand managing enterprises.

The emergence of women entrepreneurs in a society depends mainly upon variouseconomic, social, religious, cultural, and psychological factors (Habib, Roni & Haque,2005). The motivations for starting a business by rural women are significant andinclude earning an attractive source of income, enjoying a better life, the availabilityof loans, and general security.

One of the key factors for the development of female entrepreneurship inBangladesh is recognition (Saleh, 1995). When activities are performed by familymembers or by neighbors, rural women feel encouraged to participate. Therefore,whatever rural women do, it must first be recognized by their husbands, then by thefamily members, then by others. The type of family in the rural areas has an impact onthe development of rural women entrepreneurship. Studies show that rural womenthat come from a nuclear family (a family consisting of a father, mother, and theirchildren living under the same roof) tend to become more entrepreneurial than if theycame from a joint family (Surti & Sarupia, 1983). The level of family liability can alsoattribute to this.

The age of the rural women is another factor that affects the development of ruralfemale entrepreneurship. Studies show that the majority of rural female entrepreneursstart a business at the age of 20-29 years (Punitha, Sangeeta & Padmavathi, 1999). Atthis age, they no longer have many family bindings, and they can work freely in theirbusiness projects. There are many places in Bangladesh where there is no realeconomic development, but because of the presence of the rural microcredit programsin those areas, rural women are becoming more enthusiastic about initiating neweconomic projects. Therefore, properly supervised microcredit can help to improvesocioeconomic conditions of these women in Bangladesh (Begum et al., 2005).However, a lack of family and community support, an ignorance of availableopportunities, the lack of motivation in initiating new projects, shyness andapprehensiveness to get involved with economic activities, and a preference fortraditional occupations are all factors that inhibit the promotion of grassrootsentrepreneurship development among rural women (Rao, 1991).


Methodology

The Bangladesh Rural Development Board (BRDB) is the largest service-orientedgovernment institution and is directly engaged in rural development and povertyalleviation activities in Bangladesh. The ASA was developed in an atempt to graduallyeradicate poverty from society in Bangladesh. BRDB started its credit activities in thestudy area in 1993, while the inception of the ASA was in 1996. The target people ofBRDB for credit programs are poor farmers and rural women who have at least someproductive assets. On the other hand, the focus of the ASA is to give credit to the poorwomen who have no productive assets. ASA provided microcredit to 1,200 womenand 295 for the BRDB study area. BRDB gave loans for the purpose of povertyalleviation primarily in the projects of agriculture, fish culture, poultry raising, andpetty trading. ASA gave credits for poverty alleviation in the areas of paddy husking,rice frying, running small hotels, petty trading (i.e., vegetables trading, molassestrading, etc.), transportation, purchasing cows, fish culture, and raising poultry. Theminimum amount of credit given by BRDB is Tk. 2,500 and the maximum is Tk.7,000. The ASA ranged from Tk. 3,000 to Tk. 12,000. Along with microcredit, theASA also has microinsurance services. BRDB does not offer an insurance policy.However, BRDB does provide advice in family planning along with microcredit, butthe ASA does not. The ASA is significantly more strict about installments that aresupposed to be given every week. BRDB’s loanees repay monthly installments, whichis less strict in comparison to the ASA.

Characteristics of the RespondentsThe respondents of this study are rural female borrowers of two leading NGOs,

the ASA in the private sector and the BRDB in the public sector. All the borrowersof BRDB are Hindu, while the borrowers of the ASA are comprised of 77.60%Muslims and 22.40% Hindus. The age distribution of the borrowers of the ASA andBRDB is different. About 29% of ASA’s borrowers are between the ages of 20 and 25,followed by 30 to 35 years (24.10%), 35 to 40 years (22.40%), 25 to 30 years(18.40%), and 15 to 20 years old (6.10%). On the contrary, 49% of the borrowers ofBRDB are between 35 and 40 years old. About 21% of this group is between the ageof 25 and 30 years followed by 20 to 25 years (15.00%), and 30 to 35 years (15.00%)(Table 1). The average age for the borrowers of the ASA is 29 years and for the BRDBis 32 years.

Table 1: Age Distribution of the Microcredit Borrowers


About 88% of the borrowers of the BRDB and 98% of ASA are married. The differencebetween the educational qualifications of the borrowers of the ASA and BRDB has beenobserved. About 33% of the ASA’s borrowers are self-literate. They become literate afterjoining microcredit programs to manage financial matters. About 29% of them areprimary educated, followed by illiterate (22.00%), and secondarily educated (16.00%).About 36% of the borrowers of BRDB are secondarily educated. Those who areilliterate are also similar (36.40%). The self-literate borrowers in BRDB are 15.20%,and primary educated borrowers are 12.10% (Table 2). This educational statusindicates that the female borrowers were self-literate after their involvement withcredit programs.

Table 2: Educational Qualifications of the Microcredit Borrowers

The training status of the rural female borrowers shows that the majority of therespondents have no training in technology or marketing. More than 75% of theborrowers in both the groups did not receive any formal training from the creditproviders. Only 18% of the borrowers of ASA and 12% of BRDB have receivedtechnical training from anything other than loan providers. Only 8.20% of ASA’sborrowers and 12.10% of BRDB’s borrowers obtained nontechnical training from thecredit providers. The nature of this training is only to give ideas about technology andother aspects of the business (Table 3). This study noted that ASA and BRDB have noarrangement for organized training in the study area.

Table 3: Training Status of the Microcredit Borrowers

Sample Design and DeterminationBangladesh is divided into six divisions. To select the sample respondents, the

second level administrative unit of Bangladesh, the Khulna division, was selected.Under this division, Khulna is an important district (a district refers to the thirdadministrative unit of Bangladesh). A group of Thanas constitutes a district. Under thisdistrict, there are 10 Thanas: Khulna Sadar, Batiaghata, Dacope, Daulatpur, Dumuria,Koyra, Paikgacha, Phultala, Rupsa, and Terokhada. A Thana is also called Upa-Zila. Itis the fourth level administrative unit of Bangladesh. It consists of a group of Unions,


and every Union is formed with a group of villages. The reason for selecting theKhulna district is that the most densely populated district is the Khulna Division.There are about 2.38 million people living in this district with approximately 375,000households (BBS, 2005). About 50% of population in this district is female.

Batiaghata Thana was selected as the sampling area which is located adjacent toKhulna City. This Thana consists of 7 Unions, with 159 villages. The population of thisThana is 128,184, with 516 persons per sq. km. The land is about 1,468.38 acres. Only37.70% of the population is literate. There are 23,698 families in this Thana. The totalnumber of dairy and poultry farms is 12 and 57 respectively. There are 12,088 sanitarylatrines and 1,024 tube wells in the Thana. The numbers of deep tube wells are 896.Most of the families are involved in agricultural farming followed by petty trading,fishing, pottering, paddy husking, gold-making business, kamar, and spinning. Thereare 26 village hat/bazaars in the Thana.

Borrowers who are already engaged in 3-10 years or more with the credit programsare used as respondents. Sample respondents were selected by using two samplingmethods: the purposive sampling method and the random sampling method.

Purposive Sampling Method This method was used to select the types of activities of rural female borrowers

including fish culture, paddy husking, poultry farming, petty trading, grocery, animalhusbandry, weaving, handicrafts, dairy farming, and plant nursery. All the femaleborrowers of BRDB were selected from the Rajbadh village, and 25% of the borrowersfrom the ASA were selected purposively from Hatbati, Wazed Akundi Nagar, Sachibuniavillages who have been involved in microcredit programs. The individual selection wason a random basis to reduce the biases of the sample selection in this study.

Three criteria were used to select two Unions of Batiaghata Thana for this survey:(1) the intensity of credit programs, (2) the density of population, and (3) theintensity of poverty. Under each Union there are about 14 to 17 villages. One villagenamed Rajbadh was selected for interviewing the borrowers of BRDB.

Sachibunia have been selected for interviewing the borrowers of ASA. ASA andBRDB are intense microcredit programs in these selected villages because of largepopulation size and high poverty.

The sample size was determined by using a formula suggested by Yamane (1967).The following formula was used to determine the sample size of the study:

n = N/1+N(e)2

where,

n = sample size, N = population, e = precision Levels, and where Confidence Level is93%, and P = .50 (degree of Variability).

The degree of variability in the attributes being measured refers to the distributionof attributes in the population. The more heterogeneous a population, the larger thesample size required to obtain a given level of precision. The less varied (more


homogeneous) a population is, the smaller the sample size. Note that a proportion of50% indicates a greater level of variability than either 20% or 80%. This is because 20%and 80% indicate that a large majority do not or do, respectively, have the attribute ofinterest. Because a proportion of .5 indicates the maximum variability in a population,it is often used in determining a more conservative sample size. The sample size maybe larger than if the true variability of the population attribute were used. The totalnumber of female borrowers interviewed was 246, 198 of which were from the ASAand 48 from BRDB.

Designing Measurement Instruments This study was based on primary data collected from the survey of rural women. A

survey was conducted among the rural female borrowers of BRDB and ASA to collectinformation about the development of rural women entrepreneurship throughmicrocredit programs, with the help of a structured questionnaire. A structuredquestionnaire in a 5-point scale was developed for the variables relating to thedevelopment of rural women entrepreneurship. A five-point scale ranging from 1 to 5,with 1 indicating strongly disagree and 5 indicating strongly agree, was used in thisregard. This study used 40 entrepreneurship-related variables to explain the chance ofrural women for being entrepreneurial-identified from the literature. The dependentvariable is explained by four variables: independence, ability to make complexdecisions, ability to seek and grasp opportunity, and ability to take risk and initiative.The survey has been conducted with the assistance of MBA students from KhulnaUniversity, who explained the questions to the borrowers in detail. The interviewerswere trained on the variables representing the questionnaire for data collection beforestarting the interview. Borrowers were surveyed from January 2006 to March 2007.

Data Analysis Along with descriptive statistics, multivariate analysis techniques including factor

analysis and Structural Equation Modeling (SEM) were used to analyze therelationships of the variables relating to the development of rural femaleentrepreneurship. A principal factor analysis with an orthogonal Varimax rotation,using the SPSS statistical package, was performed on the survey data and was used toseparate the factors for developing entrepreneurship. The relationship ofentrepreneurial factors with the overall entrepreneurship development is assessedthrough the Analysis of Structural Equation Modeling by using Amos version 4.

It was the ultimate intention of this study to test the conceptual model developedfrom the theoretical analysis and to estimate the parameters for the structural equationmodel. Hence, data were analyzed through the SEM using Analysis of MomentStructures (AMOS) to perform path analysis. Amos’s method of computing parameterestimates is called maximum likelihood. Hypothesis testing procedures, confidenceintervals, and claims for efficiency in maximum likelihood or generalized least squaresestimation by Amos depend on certain statistical distribution assumptions. First,observations must be independent. Second, the observed variables must meet certaindistributional requirements. For instance, it will suffice if the observed variables havea multivariate normal distribution. Amos implements this general approach to the


SEM data analysis, also known as analysis of covariance structures, or causal modeling.SEM is a computer program for estimating the unknown coefficients within a systemof structural equations, and is one of several computer-based covariance structuremodels for conducting such analysis. LISERAL or Lineral Structural Relations, is aspecial purpose statistical software package that estimates structural equation modelsfor manifest and latent variables. AMOS, like LISREL, is useful when the researcherdesires to explore the causal relationships among a set of variables. The method iscalled covariance structure analysis because the implications of the simultaneousregressions are studied primarily at the level of correlations or covariances. Typically,a covariance structure model is specified through a simultaneous set of structurallinear regressions of particular variables on other variables. The field of covariancestructure analysis actually covers a wide range of topics, including confirmatory factoranalysis, path analysis, and simultaneous equation and structural equation modeling.Much research in the social sciences including business involves the measurement oflatent constructs. The method is useful for analysis of structural equations involvingexperimental data. In business applications, theoretical constructs are typicallydifficult to operationalize in terms of a single measure, and the measurement error isoften unavoidable. As a result, given an appropriate statistical testing method, thestructural equation models are likely to become indispensable for theory evaluation inbusiness research. The approach provides a means for examining causal relationshipsamong multiple variables, the magnitude of hypothesized relationships, and the extentof measurement error of constructs in application of experimental designs (Bagozzi,1977). When researchers attempt to measure constructs such as perceptions tosomething, they are attempting to gauge unobservable cognitive processes withmeasurement devices that can only approximate the latent constructs of interest. Thisprocess is typically fraught with measurement error. Because of their ability to controlor allow for such measurement error when estimating the relationships betweenvariables, covariance structure models have been gaining in popularity in businessstudies (Bagozzi, 1980, 1981). Howard (1977) suggests in this regard that structuralmodeling sharply highlights the intimate, powerful, mutually reinforcing relationshipbetween theory and measurement. In this study, it was perceived that structuralequation modeling would be the best approach to understand the relationshipsbetween the constructs.

In this study, covariance and structural modeling was performed in two distinctstages. First, observed variables are linked to unobserved variables through aConfirmatory Factor Analytic (CFA) model. CFA is a means of discovering anunderlying structure in one’s data, given some prior theoretical or empiricalinformation. The set of connections between the observed and unobserved variables isoften called the measurement model. The measurement model specifies how the latentvariables are measured in terms of observed indicators and explicitly introducesmeasurement error. Second, the causal relationships between the resulting latentvariables are examined in a structural equation model. The model componentconnecting the unobserved variables to each other is often called the structural model.The structural equation model specifies the causal relationships among the latent andunobserved variables.


Results of Factor AnalysisA Multivariate Analysis technique, factor analysis, was used to identify the factors

responsible to development women entrepreneurship in the rural areas of Bangladeshwith the support of microcredit. A principal factor analysis with an orthogonal rotationusing the SPSS statistical package was performed on the survey data and was used toseparate the factors. Factor analysis of 40 variables in the rural womenentrepreneurship survey identified 13 main factors that account for 75.74% of thevariance in the data (Table 4). The initial factor structure derived from varimaxrotation extracted thirteen factors. Scrutiny shows that some of the factors wereunclear, particularly when several items loaded simultaneously on more than onefactor. All of these factors are reflected in Table 4.

Table 4: Women Entrepreneurship Development Factors

The first factor, financial management skill and group identity, accounts for 18.16%of the variance in the data. The development of financial skill and the creation of groupidentity by the microcredit is the most important factor for the development of ruralwomen’s entrepreneurship in Bangladesh. The eigenvalue of this factor is 7.26.Financial management skill and group identity are related to six variables, includingincreased family relationships and cohesiveness (0.536), involved rural women-folk(0.822), development of financial management skills (0.866), realized self andcollective identity (0.880), getting adult education (0.621), and developing awarenessof health and women’s rights (0.696). A relatively higher level of factor loading ofalmost all the variables indicates that these variables are very important to constitutethe rural women entrepreneurship development factor. The communality values forthese variables are 0.705, 0.818, 0.835, 0.901, 0.742, and 0.630 respectively. Thehigher level of communality of the variables associated with financial management


skill and group identity indicates that each variable is very much related to the factor.The next important factor is creative urge and self interest with an eigenvalue of

3.57. The variance of this factor is 9.73%. It indicates that creative urge and selfinterest is an important factor for the development of rural female entrepreneurship.Seven variables constituted this factor. The variables are creative urge (0.843), self-interest and self dependent (0.815), inadequacy of family supplement income (0.538),family support is required (0.534), attractive source of income (-0.441), competent totake and use loan (-0.426), and getting educated (0.416). These variables are highlyimportant for determining the entrepreneurial status of the rural women borrowers.The communality of the variables is also higher.

Family funds and female involvement is the third important factor for the ruralfemale entrepreneurship development with an eigenvalue of 2.76. This factor explains6.10% of the variance. The women borrowers are concerned with self-independence(0.852), family peace (0.787), gaining social prestige (0.664), ability to accumulatefamily fund (0.525), and alleviation of gender discrepancies (0.488). Anotherentrepreneurship factor is employment of family members and the creation of new jobswith eigenvalue of 2.75 and variance of 6.87%. This factor is constituted by fourvariables: can employ others (0.827), new work and work environment (0.761),training (0.758), and scope to utilize own skills and talents (0.549). Independence andkeeping oneself busy is the fifth factor for the development of rural womenentrepreneurship in Bangladesh. The eigenvalue and the variance of this factor are2.205 and 6.51% respectively. The variables forming this factor includes doingsomething independently (0.920), can keep myself busy (0.825) and career and familysecurity (-0.447). Family experience and option limitation is the next important factorfor the development of rural women entrepreneurship in Bangladesh. Two variablesconstituted this factor such as, experience and competencies (0.835) and no otheroption available (0.764).

Other factors like knowledge of business, economic necessity of the family, selfconfidence, technical knowledge of business, money earning, unable to find suitablework or job, and contribute to the economic growth were found not significant tobuild the model.

Results of Structural Equation Modeling (SEM) Analysis

The data of this study were analyzed in two stages. First, the measurement modelwas assessed to confirm that the scales were reliable. Second, when the reliability ofthe measures had been established, the structural model was tested. This testingdetermined the strength of individual relationships, goodness of fit of the model, andthe various hypothesized paths.

The first step of the analysis was a test of the measurement model. Objectives of thistest were: (1) to contain the validity and reliability of measures, and (2) to select the bestsubset of observed measures for use in testing the structural model. The data depicted anormal distribution with acceptable skewness and kurtosis values. Coefficient alpha wascomputed for each set of observed measures associated with a given latent variable, anda Confirmatory Factor Analysis (CFA). Alpha values of each item in each dimension


were performed separately and were found acceptable. Estimation of Measurementmodel for the six constructs (factors) of interest was performed using AMOS 4.01.

The results of overall structural model fit as indicated by the chi-square statistic,was significant chi-square = 707.80; df = 168; p = 0.000 (Table 5). The overall fit ofthe confirmatory factor analysis model to the sample variance/covariance matrix, asmeasured by chi-square, provides a test of the overall reliability of observed measure(Bagozzi, 1980). The statistic is computed under the null hypothesis that theobserved covariances among the answers came from a population that fits themodel. A statistically significant value in the goodness of fit test would suggest thatthe data do not fit the proposed model, i.e., that the observed covariance matrix isstatistically different than the hypothesized matrix. The assumptions required toemploy chi-square as a significance test (in support of the hypothesis that thepredicted covariance matrix does not differ from the sample covariance matrix) aretypically violated in most covariance structure analysis. Accordingly, when theresults of chi-square analysis are favorable, it is best to say that the fit betweenpredicted and observed covariance matrices is “acceptable” rather than “significant”(Joreskog & Sorbom, 1986). In this study, however, both terms are usedinterchangeably to mean “acceptable”.

Table 5: Fit Indices of the Model

The fit of the structural model was estimated by various indices and the resultsdemonstrated good fit. For models with good fit, most empirical analyses suggest that theratio of chi-square normalized to degree of freedom (chi-square/df) should not exceed3.0 (Carmines & Mclver, 1981). In addition, the obtained goodness-of-fit (GFI) measurewas 0.809 and the adjusted goodness-of-fit (AGFI) measure was 0.737 respectively,which are both higher than the suggested values. The other two indices of goodness-of-fit (GFI), the normalized fit index (NFI), and the comparative fit index (CFI) arerecommended to exceed 0.90. The results also meet these requirements. Finally, thediscrepancies between the proposed model and population covariance matrix, asmeasured by the root mean square error of approximation (RMSEA), are in line with thesuggested cutoff of 0.08 for good fit (Byrne, 1998). The complete model of microcreditprogram and the development of rural women entrepreneurship is shown in Figure 1.


Figure 1: A Model for the Development of Rural Women Enterpreneurshipthrough Micro Credit Program

Table 6 shows that the relationships of the factors that built the model for thewomen entrepreneurship development in Bangladesh through microcredit programs.After identifying the female entrepreneurial development factors, a hypothesis wasdeveloped for each construct and the important factors that were significantlyassociated with the rural female entrepreneurship development.

Table 6: Standardized Regression Weights


Financial Management Skill and Group IdentityIn Hypothesis 1 (H1), it was predicted that the financial management skill and the

group identity have a direct and positive relationship with the female entrepreneurialdevelopment (WED) in rural areas of Bangladesh. It was presumed that higherfinancial management skill and group identity will lead to higher level ofencouragement among the rural borrowers for taking new initiative of business. Theresults show that the direct effect of financial management skill and group identity onthe development of women entrepreneurship is positive and significant (β = 0.24, p <0.008). This result indicates that the higher the financial management skill and betterthe group involvement, the higher the chance of being entrepreneurial. In Bangladesh,many people who live in rural areas are illiterate, including the female borrowers.Therefore, they face the problems of financial planning, financial record keeping,financial calculations, and the identification of profits, etc. In addition, there is also agroup effect on the development of women entrepreneurship in Bangladesh.

Family Experience and Option LimitationHypothesis 6 (H6) states that family experience and option limitation has a direct

positive effect on the development of rural female entrepreneurship in Bangladesh.This means that if the rural woman has a business orientation from her parent’s familyand if she has some fund from the microcredit providers, she will take initiative to dobusiness or she will initiate economic projects which will help her to earn money andobtain social status. This hypothesis was supported by the analysis that providespositive and significant values (β = 0.13, p < 0.11). Although this factor is significantat 11%, it’s an important factor to be entrepreneurial for the rural women throughmicrocredit programs. Since this study is the first of its kind, this result is acceptable.

Independence of the Women and the Urge to Keep BusyIn Hypothesis 5 (H5), we hypothesized that the independence of the rural women and

the urge to be kept busy can make them entrepreneurial which has a positive andsignificant effect on female entrepreneurial development in the rural areas ofBangladesh. This indicates that more independence and more enterprising by a ruralwomen will lead to a higher level of entrepreneurship. The results support thishypothesis and positive and significant (β = 0.08, p < 0.13). We also accept this resulton the grounds that the significant level is 13%.

Other factorsIn Hypothesis 2 (H2), we predicted that the relationship between creative urge and

self-interest and the rural female entrepreneurship is positive and significant. But theresults show that the relationship between these constructs are negative and notsignificant (β = -0.063, p > 0.38). This indicates that if there is a change in the creativeurge and self-interest factor, it will not lead to the development of rural womenentrepreneurship through microcredit programs in Bangladesh. That means throughmicrocredit programs, the creative urge and self-interest is not developed among therural female borrowers, as it depends on environmental factors which are unfavorablefor the rural women in Bangladesh.


In Hypothesis 3 (H3), it was predicted that the relationship between family fundsand involvement in business and rural female entrepreneurship is positive andsignificant. However, the results show the opposite situation in this regard (β = -0.120,p > 0.21). This indicates that the change in financial status and female involvementwith money matters will not change in the entrepreneurship developmentcharacteristics among the rural women in Bangladesh. If the rural families arefinancially solvent, they will not lean towards doing business in Bangladesh where itis culturally discouraged.

In Hypothesis 4 (H4), it was perceived that there is a positive and significantrelationship between a new job and the employment of family members with rural femaleentrepreneurship development. But the results show that there is no significantrelationship between the two constructs (β = 0.035, p > 0.67). This indicates thatemployment of family members and the new job will not develop any entrepreneurialcharacteristics among the rural female borrowers through microcredit programs.

Conclusions and Recommendations

It is generally perceived that the microcredit program helps to developsocioeconomic status of the rural women in Bangladesh. In addition, it is perceivedthat microcredit is helping not only to bring socioeconomic changes, but also to makethe borrowers entrepreneurial. This study tried to resolve these questions byconstructing a model which was supported by the results of multivariate analysis.

This study identified that factors like the financial management skill of theborrowers and group identity, experience from the fathers’ family and option limitation,independence of the rural women, and the urge to make them entrepreneurs have asignificant relationship with the rural female entrepreneurship development inBangladesh. On the other hand, factors such as creative urge and self interest, familyfund and previous involvement in business, and job and employment of the familymembers are not significantly related to the rural women entrepreneurshipdevelopment. SEM analysis shows that among seven hypotheses, only three hypothesesare supported by the analysis. This indicates that other factors are not appropriate forthe development of rural women entrepreneurship in Bangladesh.

The most important finding of this study is that the financial management skill andthe group identity of the borrowers have a direct and significant relationship with thedevelopment of rural women entrepreneurship (WED) through microcredit programs.When rural women receive financial support from the microcredit providers, they feelencouraged to involve themselves in the financial projects that subsequently increasethe financial management skills of the borrowers. Microcredit also provides groupidentity to the rural women. When women acquire knowledge of financialmanagement and get group identity, they become more enthusiastic to initiate newbusiness projects. These significant relationships indicate that if the microcreditborrowers can enhance this skill among the rural female borrowers, it would lead themtowards the development of entrepreneurship. As a result, the borrowers will be ableto stand on their own feet.

The second important finding of this study is that the experience from the parent’s


family of the borrower and option limitation have a direct positive impact on thedevelopment of rural women entrepreneurship in the rural areas of Bangladesh. Thismeans that if a rural female has a business orientation from her parent’s family and atthe same time, has some funds at her disposal, she will initiate new business oreconomic projects which will help her to earn a profit and obtain social status as well.

The third important finding is that the rural women who are independent by natureand would like to keep busy with economic activities could be identified by theborrowers. This section of rural female has the highest potential to be entrepreneurial.This study supports this observation for the rural women borrowers in Bangladesh.

The main problem of any small business in Bangladesh is the management skillsrelated to financial affairs of the business. The businessmen or entrepreneurs areunable to make financial plans and maintain financial accounts of the businessbecause of their illiteracy. Most of the people in rural areas are illiterate in Bangladeshand women are in a more disadvantageous position in this regard. Hence, microcreditproviders should give importance to the development of the financial managementskills of the borrowers and create group identity of the borrowers. They also shouldidentify the rural women who have their family experience and no other options butto do business or get involved with loan providers. Loan providers should also bemindful of the fact that the rural women of Bangladesh have an independent mentalityand they would like to take on the challenge of being entrepreneurs. Therefore, todesign and implement a loan program, microcredit providers should keep thisindependence in mind. If these aspects are properly addressed by the loan providers,rural female borrowers will be more entrepreneurial and as a result, the borrowers willbe able to stand on their own feet and rural women entrepreneurship will be developedin Bangladesh.

References

Acharya, J. (1994). Rural Credit and Women Empowerment: Case Study of theJanashakthi Program in Hambantota District, Srikant. Unpublished master’s thesis,Asian Institute of Technology, Bangkok, Thailand.

Ackerly, B. (1995). Testing Tools of Development: Credit Programs, Loan Involvement and Women’s Empowerment. IDS Bulletin, 26(3): 56-68.

Ahmed, S. M., Adams, A., Chowdhury, A. M. R. & Bhuiya, A. (1997). Income-Earning Women from Rural Bangladesh: Changes in Attitudes and Knowledge.Empowerment – A Journal of Women for Women, 4: 1-12.

Ahmed, A. & McQuaid, R. W. (2005). Entrepreneurship, Management, andSustainable Development. World Review of Entrepreneurship, Management andSustainable Development, 1(1): 6-30.

Ahmed, S. U. (1981). Entrepreneurship and Management Practices among Immigrants from Bangladesh in the United Kingdom. Unpublished doctoral dissertation, BrunelUniversity, London.

Ahmed, S. U. (1982). Entrepreneurship and Economic Development. DhakaUniversity Studies, Part-C, 3(1): 41-53.

Ahmed, S. U. (1987). Entrepreneurship Development with Some Reference to


Bangladesh, Entrepreneurship and Management in Bangladesh. In Mannan, M. A.(Ed.), Research Monograph. Bureau of Business Research (BBR). Dhaka:Bangladesh.

Alam, J. (1988). Rural Poor Program in Bangladesh. Dhaka: UNDP.Amin, R., Ahmed, A.U., Chowdhury, J. & Ahmed, M. (1994a). Poor Women’s

Participation in Income-Generating Projects and Their Fertility Regulation in RuralBangladesh: Evidence from a Recent Survey. World Development, 22(4): 555-565.

Amin, S. & Pebley, A. (1994b). Gender Inequality within Households: The Impact of Women’s Development Program in 36 Bangladeshio Villages. Special Issue onWomen, Development and Change. The Bangladesh Development Studies, 22(2 &3): 121-155.

Apte, J.K. (1988). Coping Strategies of Destitute Women in Bangladesh. Food and Nutrition Bulletin, 10(3): 121-155.

Arefin, K. & Chowdhury, M. R. (2008). Small Women Entrepreneurship in DhakaCity: An Appraisal. Maneggiare, 4(Dec): 45-55.

Audretsch, D. B. (1995). Innovation and Industry Evolution. Cambridge, MA: MITPress.

Bagozzi, Richard P. (1977). Structural Equation Models in Experimental Research. Journal of Marketing Research, 14(May): 209-226.

Bagozzi, Richard P. (1980). Causal Models in Marketing. New York: John Wiley & Sons, Inc.

Bagozzi, Richard P. (1981). Attitudes, Intentions, and Behavior: A Test of Some Key Hypotheses. Journal of Personality and Social Psychology, 41: 607-627.

Begum, R. (2002). Entrepreneurial Performance of Women Entrepreneurs inBangladesh. Journal of Business Studies, 23(2): 343-355.

Begum, R. (2003). Determining Entrepreneurial Success Status of WomenEntrepreneurs. Journal of Business Studies, 24(1): 127-136.

Begum, R., Uddin, Islam Md., & Ahmed, Masoom Md. (2005). Role of the GrameenBank in Enhancing the Socio-Economic Condition of Women: A Study on BazarPara at Gomastapur. Journal of Business Studies, 26(1): 15-35.

Byrne, B.M. (1998). Structural Equation Modeling with Lisrel, Prelis, and Simplis:Basic Concepts, Applications, and Programming. Mahwah, NJ: Lawrence ErlbaumAssociates

Cain, M., Khanam, S. R. & Nahar, S. (1979).Class, Patriarchy, and Women’s Work in Bangladesh. Population and Development Review, 5: 405-438.

Cantillon, R. (1755). Essai Sur La Nature Du Commerce in General, London:MacMillan.

Carmines, E.G. & McIver, J.P. (1981). Analyzing Models with Unobserved Variables:Analysis of Covariance Structures. In Bohrnstedt, George W. & Borgatta, Edgar F.(Eds.), Social Measurement: Current Issues. Beverly Hills, CA: Sage Publications.

Casson, M. (1990). Enterprise and Competitiveness, Oxford: Clarendon Press.Casson, M. (1999). Entrepreneurship and the Theory of the Firm, In Acs, Z. J.,

Carlsson, B. & Larisson, C. (Eds.), Entrepreneurship, Small and Medium SizedEnterprises and the Macro Economy. Cambridge: Cambridge University Press.

Chell, E., Haworth, J.M. & Brearley, S.A. (1991). The Entrepreneurial Personality:


Concepts, Cases and Categories, London: Routledge.Chowdhury, M. M. R. (1998). Women Entrepreneurs: Emerging as Leaders of Rural

Bangladesh. Journal of the Faculty of Arts, The Dhaka University Studies, 55(1): 113-143.

Chowdhury, M. M. R. (2002). The Emerging Role of Women as Entrepreneurs in Bangladesh. The Dhaka University Studies, Part-D, 19(1): 157-168.

Chowdhury, M. J. A. (2004). Micro-Credit and Sustainability of Poverty Alleviation: ACase Study of Grameen Bank in Bangladesh, Social Science Review, The DhakaUniversity Studies, Part-D, 21(1): 105-120.

Cunningham, J. B., & Lischeron, J. (1991). Defining Entrepreneurship, Journal ofSmall Business Management, 29: 45-61.

Cuong, N. (2008). Is A Governmental Micro-Credit Program for the Poor Really Pro-Poor? Evidence from Vietnam. The Developing Economies, 46(2): 158-187.

Deshpande, M., & Joshi, J. V. (2002). Study of Women Entrepreneurship inMarathwada, Indian Journal of Marketing, 32(5 & 6): 14-18, 32.

Gartner, W.B. (1988). Who is an Entrepreneur? Is the Wrong Question.Entrepreneurship Theory and Practice, 12: 47-67.

Ghai, D. (1984). An Introduction of the Grameen bank Project, Dhaka: Grameen Bank.Goetz, A. M. & Gupta, R. S. (1996). Who Takes the Credit? Gender, Power and

Control over Loan Use in Rural Credit Programs in Bangladesh. World Development,24(1): 45-63.

Habib, W.M., Roni, N.N. & Haque, T. (2005). Factors Affecting WomenEntrepreneurship in India: A Multivariate Analysis. Journal of Business Studies,26(1): 249-258.

Hashemi, S. M., Schuler, S. R. & Riley, A. P. (1996). Rural Credit Programs and Women’s Empowerment in Bangladesh. World Development, 24(4): 635-653.

Hashemi, S. M. (1998). Those left Behind: A Note of Targeting the Hard-Core-Poor. InWood, G. & Sharif, I. A., (Eds.), Who needs Credit – Poverty and Finance inBangladesh. London: Zed Books.

Hoselitz, R. F. (1952). Entrepreneurship and Economic Growth. American Journal of Economics and Society, 12(1).

Holvoet, N. (2005). The Impact of Microfinance on Decision-Making Agency:Evidence from South India. Development and Change, 36(1): 75-102.

Hossain, M. & Sen, B. (1992). Rural Poverty in Bangladesh: Trends and Determinants. Asian Development Review, 10: 2-34.

Hossain, M. (1986). Credit for Alleviation of Rural Poverty: The Experience of theGrameen Bank in Bangladesh, Washington, DC: International Food Policy ResearchInstitute and Dhaka: Bangladesh Institute of Development Studies (BIDS).

Hossain, M. (1988). Credit for Alleviation of Rural Poverty: The Grameen Bank in Bangladesh, Research Report 65, Washington, DC: International Food Policy Research

Institute and Bangladesh Institute of Development Studies (BIDS).Howard, J. (1977). Consumer Behavior: Application of Theory, New York: McGraw-Hill

Book Company.Hulme, D. & Mosely, P. (1998). Micro Enterprise Finance: Is There a Conflict between

Growth and Poverty Alleviation? World Development, 26(5): 783-790.


Islam, N. & Mamun, MZ. (2000). Entrepreneurship Development: An OperationalApproach (1st Ed). Dhaka: University Press Limited.

Jha, S.C. (1991). Rural Development in Asia: Issues and Perspectives, AsianDevelopment Review, 15(3): 83-99.

Joreskog, Karl G. & Sorbom, Dag. (1986). LISREL VI, 4th Ed. Mooresville, Indiana: Scientific Software Inc.

Kabeer, N. (2001). Conflicts over Credit: Reevaluating the Empowerment Potential of Loans to Woman in Rural Bangladesh, World Development, 29(1): 63-84.

Katz, J.A. (1991a). The Institution The Institution and Infrastructure ofEntrepreneurship, Entrepreneurship Theory and Practice, 15(3): 85-102.

Katz, J.A. (1991b). Endowed Positions: Entrepreneurship and Related Fields, Entrepreneurship Theory and Practice, 15(3), 53067.

Khandkar, S. & Chowdhury, O. H. (1996). Targeted Credit Programs and RuralPoverty in Bangladesh. World Bank Discussion Paper: 336.

Kearney, P. (1996). The Relationship Between Developing of the Key Competencies inStudents and Developing of the Enterprising Student, Canberra, Australia: Education,Training and Youth Affairs.

McClelland, D. C. (1961). The Achieving Society. New York: The Free Press.Malechi, E. J. (1997). Technology and Economic Development: The Dynamics of Local,

Regional and National Competitiveness, (2nd ed). London: Addison WesleyLongman Limited.

McQuaid, R.W. (2002). Entrepreneurship and ICT Industries: Support from Regionaland Local Policies. Regional Studies, 36(8): 909-919.

Montgomery, R. (1996). Credit for the Poor in Bangladesh- The BRAC RuralDevelopment Program and the Government Thana Resource Development andEmployment Program.

Navajas, S., Schreiner, M., Meyer, R., Gonzalez-Vega, C. & Rodriguez-Meza, J. (2000).Micro credit and the Poorest of the Poor: Theory and Evidence from Bolivia. WorldDevelopment, 28(2): 333-346.

Naved, R. (1994). Empowerment of Women: Listing to the Voices of Women. In Amin(Ed.), The Bangladesh Development Studies, Special Issue on Women, Developmentand Change, 22(2 & 3): 121-155.

OECD. (1998). Fostering Entrepreneurship: A Thematic Review, Paris: OECD.OECD. (2003). Entrepreneurship and Local Economic Development: Program and

Policy Recommendations, Paris: OECD.Pitt, M. & Khandaker, S.R. (1996). Household and Intra-household Impacts of the

Grameen Bank and Similar Targeted Credit Programs in Bangladesh. World BankDiscussion Papers: 320.

Puhazhendhi, V. & Badatya, K. C. (2002). SHG Bank Linkage Program for Rural Poor-An Impact Assessment. Mumbai, India: National Bank for Agriculture and RuralDevelopment.

Punitha, M., Sangeeta, S. & Padmavathi, K. (1999). Women Entrepreneurs: TheirProblems and Constraints. The Indian Journal of Labor Economics, 42(4): 707-716.

Rahman, A.H.M.H. (1979). Entrepreneurship and Small Enterprise Development in Bangladesh, Bureau of Business Research, Dhaka: Bangladesh.


Rahman, A.H.M.H. (1996). Promoting Entrepreneurship Through Micro EnterprisesDevelopment in Bangladesh. Paper presented at the AMDIB Seminar, Dhaka,Bangladesh.

Rahman, A.H.M.H. (1999). Micro-credit Initiatives for Equitable and Sustainable Development: Who Pays? World Development, 27(1): 67-82.

Rao, H.C. (1991). Promotion of Women Entrepreneurship. SEDME, 18(2): 21-28.Rosa, P. (1992). Entrepreneurial Training in the UK: Past Confusion and Future Promise.

Paper presented at the Scottish Enterprise Foundation Conference, SterlingUniversity, Scotland.

Saleh, A. (1995). A Profile of the Women Entrepreneurship in Bangladesh. Faculty of Business Studies, University of Dhaka. Dhaka University Journal of Business Studies,

16(1).Sankaran, M. (2005). Microcredit in India: An Overview, World Review of

Entrepreneurship. Management and Sustainable Development, 1(1): 91-100.Schumpeter, J. A. (1949). The Theory of Economic Development. Cambridge, MA:

Harvard University Press.Schuler, S. R., Hashemi, S.M. & Riley, R.A. (1997). The Influence of Women’s

Changing Roles and Status in Bangladesh’s Fertility Transition: Evidences from aStudy of Credit Programs and Contraceptive Use, World Development, 25(4): 563-575.

Sen, G. (1997). Empowerment as an Approach to Poverty. Human Development Report.Manuscript submitted for publication. New York: UNDP.

Shahidur, R. K. (1996). Grameen Bank; Impact, Cost, and Program Sustainability.Asian Development Review, 14(1): 97-130.

Shahidur, R. K. (1998). Fighting Poverty with Micro Credit: Experience in Bangladesh. Dhaka: University Press Limited.

Shahidur, R., Khandkar, S., Hussain, A., Samad, A. & Zahid, K.H. (1997). Income andEmployment Effects of Micro-credit Programs: Village-level Evidence fromBangladesh. The Journal of Development Studies, 35(2): 96-124.

Sinha, S. & Patole, M. (2003). Microfinance and the Poverty of Financial Services: APerspective from Indian Experience. South Asia Economic Journal, 4(2): 301-318.

Srinivasan, T. N. & Bardhan, P.K. (1990). Rural Poverty in South Asia. Impact ofGrameen Bank Programs on Rural Poverty Alleviation in Bangladesh. Unpublishedmaster’s thesis, Asian Institute of Technology, Bangkok.

Stevenson, H. (2000). The Six Dimensions of Entrepreneurship. In Birley, S. &Muzyka, D. (Eds.), Financial Times Mastering. London: Pearson Education Limited.

Surti, S.P. & Saruptia, A. (1983). Psychological Factors Affecting WomenEntrepreneurs: Some Findings. Indian Journal of Social Work, 44(3): 287-295.

United Nations. (1998). Secretary General’s Report on Role of Micro Credit in the Eradication of Poverty, New York: United Nations.

United Nations. (2004). Unleashing Entrepreneurship: Making Business Work for the Poor, Report of the Commission of the Private Sector and Development to the UN. Weber, M. (1930). The Protestant Ethic and the Spirit of Capitalism. New York: Charles

and Sons.Yunus, M. (2003). Banker to the Poor: Micro-Lending and the Battle Against World


Proverty. New York: Pubic Affairs. Homepage of Grameen Credit.Yamane, T. (1967). An Introductory Analysis of Statistics. New York: Harper and Row.Zaman, H. (1999). Assessing the Poverty and Vulnerability Impact of Micro-Credit in

Bangladesh: A Case Study of BRAC, World Bank Policy Research Working Paper: No.2145.


37Lowengart

Heterogeneity in ConsumerSensory Evaluation as a Base

for Identifying Drivers ofProduct Choice

Oded Lowengart

Ben Gurion University

In this paper we propose a multiattribute choice modeling approach to explorethe heterogeneity in the saliency of product attributes in the process of aproduct choice that is based on sensory evaluations. We demonstrate this ideaby using data about consumers’ red wine evaluation. Such an approachenables managers to add knowledge about consumers' needs and wantsbeyond traditional art and the experience of wine makers into the process ofdesigning a product. We utilized a choice model that enables us to identifysuch attributes and, simultaneously, to estimate the choice probabilities foreach different wine. Our results, based on four different red wines, indicatethat based on their sensory evaluation, consumers tend to utilize several wineattributes in their choice process. The saliency of these attributes varies indifferent consumer segments such as gender and frequency of wine drinking.

Choosing among products characterized by many different types of attributes isdifficult for consumers, as it requires a considerable cognitive effort. This isparticularly true when the product category offers many different alternatives withvarious tastes. In such cases, consumers can rely on extrinsic (i.e., signals of qualitysuch as brand name or package) or intrinsic (i.e., taste of the product) productcharacteristics to choose among alternative products. The latter might be more reliablethan the former, as consumers can develop their own direct evaluation criteria (theirown taste) and test that product. Wines, for example, provide consumers with a widevariety of products with different tastes, qualities, prices, and other related attributes.Choosing a specific wine, therefore, is a complex task for consumers. Furthermore,

verifying the qualities of such products is usually possible only after actually using theproduct. Moreover, due to the wide selection of possible alternative products,consumers cannot be sure they made the right decision even after consuming theproduct. This makes wine a typical credence product – products that are difficult toevaluate before as well as after consumption (Darby & Karni, 1973), as opposed tosearch products (that can be evaluated prior to consumption) and experience products(that can be evaluated after consumption) (Nelson, 1974). It is logical to expect thatconsumers cannot solely rely on their own taste test for wine choice, since, in manypurchasing situations, this option is not easily available. As a result, other methods ofreducing uncertainty can be used by consumers. For example, Lynch and Ariely(2000) found that electronic shopping can reduce search costs and price sensitivity,while maximizing the transparency of quality information specifically for adifferentiated product such as wine. Nevertheless, a taste test is still the more reliableselection criteria for choosing such products when possible.

Consumers can use their own sensory evaluation to verify product qualities, whenpossible. Shepherd and Towler (1992), for example, argue that experience (andvaluation) of consumers with food products is shaped by sensory attributes andparticularly, by taste. Koivisto and Sjóden (1996) argue that taste is a good explanatoryvariable for food choices. As in many aspects of consumer products, there isheterogeneity among consumers when it comes to the exact combination of marketingmix variables that fit their needs. Heterogeneity stemming from personal differences(e.g., gender) geographical, behavioral (e.g., experience with the product) and othersources can have an effect on the desired product characteristics and preferences for it.For example, Scarpa, Philippidis and Spalatro (2005) found a variation in choice thatis associated with socioeconomic variables in several food products. Hu et al. (2004)found gender differences in a latent class model analysis of choice of geneticallymodified ingredients of food products. The same type of difference was also found inwine (Goodman, Lockshin & Cohen, 2008).

The current study explores how the effect of consumer sensory evaluations on thechoice among different products can provide diagnostic information about productmodification, or new product development. In order to demonstrate this approach, weanalyze red wines, where sensory evaluation plays a significant role, as this product ischaracterized by a variety of attributes that are evaluated by different sensors (e.g.,taste, smell). To expand our understanding about the potential difference amongconsumer segments with respect to such product modifications, we explore twodifferent sources of potential heterogeneity in consumers’ evaluation: genderdifferences (personal source) and frequency of drinking wine (behavioral source).

To address the objective of this paper, a probabilistic choice model formulation isused to identify the salient product attributes in choice formation. The results of theanalysis reveal that such attributes can be identified and consumers' heterogeneity insensory evaluation that is reflected in the saliency of the wine attributes exist acrossdifferent consumer segments. That is, a segment-to-segment difference is revealed. Betterunderstanding of such a pattern of results can provide a better understanding of sensory-based evaluation methods and scenarios and, at the same time, provide insight into thetype of product (wine) managers should develop to better cater to their target markets.


The rest of the paper is organized as follows. In Section 2, we present thebackground for this study and present the conceptualization of the research at hand.In Section 3, the methodology used in this study is presented. This is followed bySection 4, where the results of the analysis are laid out. Section 5 provides thediscussion, conclusion, and summarizes the study.

Problem Conceptualization

Traditionally, winemakers make wines that preserve the qualities of the differentwine varieties and at the same time, attempt to create a wine that will appeal to thepalettes of wine consumers. In a sense, it is an art of blending two aspects of productcreation into the resulting outcome: wine taste. Research aimed at improving grapequality in the agricultural area is grounded in extensive accumulated knowledge thatcan provide wine growers with better agro-technical methods to improve thecultivation of their vineyards (Weaver, 1976; Seguin, 1986) or improve the technologyof wine making (Pretorius & Bauer, 2002) and bottling (Prescott et al., 2002).

Substantial research has also been conducted on the other domain of importanceto winemakers; consumer preferences. Such research is mainly concerned with tastetests and the development of information cues that try to assist consumers inidentifying and selecting wines (Johnson et al., 2001). The latter includes the effect ofcountries and regions within a country on the evaluation of wine (Orth, Wolf & Dodd,2005; Skuras, 2002) and branding (Thode & Maskulka, 1998; Walker, 2003). Anothertype of research has focused on consumer heterogeneity with respect to winepreference. This has taken the form of appropriate methodology for heterogeneitydetection (Mueller, Francis & Lockshin, 2009; Cordelle, Lange & Schlich, 2004) orconsumer demographic effects (Scarpa et al., 2005; Hu et al., 2004), among others. Asnoted earlier, tasting the wine is probably the best method consumers can use inselecting a wine, as it is probably more reliable in examining wine qualities. Indeed,winemakers frequently use taste tests to persuade consumers to test different wineblends for qualities such as aroma, bouquet, after taste, and other characteristics.

Sensory Evaluation and PreferenceSince wine can be considered as a credence quality type of good, consumers use a

variety of direct and indirect product attributes to evaluate the product sinceconsumers cannot be sure they made the right decision. To address these difficulties,wine producers, for example, try to influence potential consumers by reducing someof the uncertainty concerning their wines. To this end, producers create several winebrands for the same varieties based on the quality of the grape juice, which could be asignal or self-declaration of quality. Other indicators are vintage, winery andreputation, geographical location, and other external characteristics that may classifythe wine. All these indicators serve as a proxy to the product quality. Since winequality is marked by relatively high heterogeneity, even when dealing with the samevariety and the same production year, the best tool for consumers to evaluate thequality of the wine is still their own tasting experiment. It is very difficult forconsumers to taste all wines they might like to buy before an actual purchasing has

39Lowengart


taken place. Wine marketers usually provide sampling procedures to foster suchtesting. This procedure provides marketers with primarily two types of informationfrom consumers: 1) the opportunity to gain insight into the overall preference for acertain wine, and 2) evaluation of the different wine qualities based on consumers'sensory evaluation (Lesschaeve, 2007). Relating the information from the secondsource to the first (attributes to preference) can reveal more insight about theformation of consumer preferences. This is particularly important to winemakers, as itwill allow them to lower the number of blends they create to better target the desiredpreferred wine. In other words, identifying attributes that drive consumer preferencescan indicate to winemakers what aspects of the wine need to be changed to increaseits preference among consumers. Wine testing and a short follow-up questionnairecompleted by consumers after tasting the wine regarding product attributes that areevaluated by their taste and smell sensors can indicate what kind of product attributescreate a preferred product.

We frame the consumer decision of whether to buy a certain wine in this study tothe sensory evaluation case. The decision about such a purchase, therefore, dependson consumer perceptions of these sensory-based product attributes. On regularpurchasing occasions, consumers are faced with more than a single alternative of winefrom which they can choose. The purchasing decision in such real-world scenariosbecomes even more complex to analyze as there are common attributes acrossproducts and one choice decision that, in a sense, captures a competitive scenariobetween alternative products. Since this case is probably more important towinemakers than the single (i.e., monopoly type) case where only one wine isconsidered, rather limited work aimed at modeling this purchasing decision process inwines has been done. More specifically, no complete understanding exists of thecompetitive intensity between various wines available to the consumer that is based onsensory-type attribute evaluations. Furthermore, the effect of the wine attributes onthe purchase decision has not been adequately addressed in the literature. To fill thisvoid, we propose a probabilistic modeling approach that will address these issues. Inparticular, we employ a multinomial Logit choice model to examine the choiceprobabilities of different red wines as a function of the wine sensory-based attributes.

Researchers have tried to define wine quality according to objective characteristicsbased on chemical and instrumental analyses of wine attributes. Such characteristicsinclude acidity, color, volatile components, and other aroma-related and measurableattributes. Wine’s compositional and sensory profiles are widely documented, andseveral models have been proposed to identify and classify wine quality and origin,based on these profiles (Cliff & Dever, 1996; Vanier, Brun & Feinberg, 1999). Thesemeasures, however, are not fully appreciated by consumers, who generally rely ontheir own perception of product qualities.

Some characteristics are not easily measurable either. For example, the aroma andsensory attributes of wine are complex and difficult to measure and describe. Hence,a sensory evaluation of wine is generally performed by wine experts, who evaluate thewine and describe its attributes to potential consumers. However, consumers willfrequently rely on their own judgments about these qualities. Since consumers makethe purchasing decision, it would be prudent for winemakers to use a consumer

41Lowengart

sample to evaluate such wine qualities and preferences to better identify the preferredwine taste.

Consumer HeterogeneityAs noted above, wine tasting is a common method for selecting a wine in wineries

or wine stores, as it reduces uncertainty about the product qualities. What are theattributes that most affect consumers in such a choice process? Do these attributesdiffer across different consumer segments? In other words, does heterogeneity amongconsumers have an effect on the saliency of the wine attributes in a choice context?From the marketers' perspective, the answer to these questions might indicate apotential for constructing a marketing strategy based on those important attributes.Such a strategy might be more effective and efficient than others, because it wouldfocus on the potential drivers of consumer preferences and choice. That said, a lack ofunderstanding continues to exist with respect to the salient attributes of red wineswhich differ from white wines in their complex characteristics and the variation indifferent consumer segments.

The aforementioned discussion about winemaking that is primarily based on thewinemakers’ experience and consumer evaluations primarily based on their sensoryevaluation, yields some inconsistencies regarding the issue of how to develop a winewith the highest consumer preference. The art of winemaking, as exhibited by theknowledge of the winemakers, was eventually tested by consumer sensors. Such winetaste tests evaluate the overall quality of the product and give winemakers anindication as to whether they are on the right track. This type of test has oneshortcoming since it involves a sequential evaluation of each wine, one at a time, withan evaluation of that wine on its own. That is, there is no provision for the relativeeffect of the one wine characteristic on the relative preference of this wine comparedto other wines in the choice set. This issue becomes even more complex as poolingconsumers evaluation might lean to an “average” wine taste that will not necessarilyfit the desired preference of a certain segment. It is therefore essential to identifyheterogeneity among consumers in terms of preference formation to reveal the driversof this preference formation.

Heterogeneity in consumer sensory evaluation is well documented in the literature(Tomlins et al., 2007). In a study conducted by Weaver (2001), heterogeneity in foodpreference based on sensory evaluation was observed, to a certain degree, betweenmen and women. In addition, preference and frequency of consumption were alsocorrelated. Differences between consumers based on gender behavior of alcoholconsumption have been widely documented (Ricciardelli et al., 2001). Since heavyalcohol drinkers may be more experienced in wine styles, segmenting the marketbased on the frequency of drinking wine might be valuable in gaining more insightinto different consumer needs.

Figure 1 summarizes the proposed framework of analysis of this study.In short, this study is aimed at filling the void in the literature on gaining

additional insight into sensory-based attributes and their effect on consumer choicein a heterogeneous consumer group. Winemaking is considered by many as acombination of art and science, so we worked to increase knowledge of the exact


wine attributes that drive consumer choice and, therefore, provide managers withmore “knowledge to improve art,” while capturing the competitive intensity thatprevails in such product category.

Figure 1: Sensory-based Evaluation Analysis

Methodology

In terms of methodology, we used a descriptive research approach that was basedon two stages. In the first, we identified the relevant red wine attributes thatconsumers consider when purchasing red wines. In the second stage, we conducted ablind taste test experimental design to capture the effect of the wine qualities only (i.e.,not the brand effect or other external cues). We used the following list ofcharacteristics as representative of the wine attributes: color intensity, aroma, bouquet,taste, tannic, harmony, and after-taste sensation. This set of wine attributes conformsto the generally accepted rules of wine tasting (Kolpan, Smith & Weiss, 1996).

Procedure and Data collectionThe subjects used for this study were students, visitors and staff members at a large

university. The taste tests were conducted during a time period of two days that lastedfrom late morning to late afternoon. One hundred and thirty-five respondentsparticipated in the study. The tasting experiment was performed in the lobby of a largebuilding complex to attract potential participants. The researchers suggested winetasting to the visitors who walked through the building. They presented four bottlesof wine wrapped in brown paper. All of the wines tested were presented to the subjectssimultaneously, without any information about the wine. Furthermore, randommixing of the alternatives across participants was carried out to avoid potentialprimary or recency effects. Overall, four red generic wines of different brands weretested (i.e., an unknown producer with a private label, a well-known brand, a winefrom a boutique winery, and a very well-known brand).

43Lowengart

Overall, 135 participants took part in the wine tasting procedure and answered thequestions pertaining to this test. The sample was formed by 88 males and 47 females.The participants were mostly young adults, 41 of which were between the ages of 18and 24 (since the legal drinking age is 18 in the area where the study was performed),89 between 25 and 40, and 5 over 40. It is acknowledged that this sample might beskewed toward younger male customers. Further exploration of other demographicvariables can be carried out in future research. With respect to income level, 81 of theparticipants earned less than the average salary, 46 at about the average, and 14 abovethe average income. The level of employment ranged from full-time, 64, to part-time,8, and full-time students (unemployed), 62. Subjects were asked to taste the wine andto rate each of the following wine attributes described earlier: color intensity, aroma,bouquet, taste, tannic, harmony, and aftertaste. Respondents were asked to rate theirresponses on an interval scale of 1 (very low level) to 10 (very high level). Forinstance, a respondent would be asked: “On a scale of 1 to 10, where 1 is very lightand 10 is very dark, how would you rate the color intensity of this brand?”Descriptions for the scales used for the other attributes are also given in Table 1.

Table 1: Attributes Involved in Product Evaluation

Respondents were informed about the characteristics of the different productattributes. For example, aromas are the smell stemming from the grape, bouquet is thesmell coming from the production process (e.g., aging in oak barrels) of the wine andnot the grape itself. Harmony is the balance between the wine components, whiletannic is the dry feeling in the mouth after drinking the wine, and so on. Similarmeasures were used in other studies (Nerlove, 1995; Hughson & Boakes, 2001).

In addition, respondents were asked to rate their overall evaluation of each wineand to rate their overall preference for each of the four wines they tasted (Cohen &Lowengart, 2003).

Choice Model The main objectives of this study, as noted above, are twofold: 1) estimating the

probability that a consumer would choose a specific wine from a set of alternativewines, and 2) identifying the red wine attributes that most affect customers in theirpurchasing decision. The latter will assist managers and winemakers in deciding whichwine attribute they need to modify to improve the choice probability of their wine.

We employed a probabilistic multinomial Logit choice model (McFadden, 1974) to


analyze the data. The MNL model is a simultaneous compensatory attribute choicemodel that incorporates the concepts of thresholds, diminishing returns to scale andsaturation levels (McFadden, 1974). Furthermore, the MNL is based on theassumption that the overall preference of a consumer for a choice alternative (i.e., thepreferred wine) is a function of the perceived relative utility that the alternative (wine)holds for the consumer.

Let Uij be the utility of alternative product j for customer i, and m the number ofalternative products. The utility function can be separated into a deterministiccomponent Vij (measured in terms of perceived value associated with thecharacteristics of the products), and an unobserved random component, eij, which isassumed to be drawn from independent and identically distributed such that:

Uij = Vij + eij (1)

The distribution of eij is assumed to be exponential (Gumbel type II extreme value)and thus the probability that alternative product j will be chosen by customer i isrepresented by:

Pij = (2)

Utility SpecificationThe deterministic component of the utility function is a product of the weighted

sum of the product attributes identified earlier and has the following form:

Vij = a1COLORij + a2AROMAij + a3BOUQUETij + a4TASTEij + (3)a5TANNICij + a6HARMONYij + a7AFTERTASTEij

where, COLORij – consumer i' perceptions of the color intensity of wine alternative j AROMAij – consumer i' perceptions of the aroma of wine alternative jBOUQUETij – consumer i' perceptions of the bouquet of wine alternative jTASTEij – consumer i' perceptions of the bouquet of wine alternative jTANNICij – consumer i' perceptions of the tannic of wine alternative jHARMONYij – consumer i' perceptions of the harmony of wine alternative jAFTERTASTEij – consumer i' perceptions of the aftertaste of wine alternative jfor j=1,2,3,4.a1a2a3a4a5a6a7 – parameters to estimate.

Results and Discussion

The estimated parameters �a1,…,�a7 for all subjects tasting red wine are presentedin Table 2. The data indicate that four wine attributes are salient in the choice process– namely, taste and harmony and to a lesser degree, bouquet and aftertaste. Thus, wine

exp(Vij)

S exp(Vij)j = m

j = 1

45Lowengart

producers and marketers should focus on these wine attributes, while targeting wineconsumers similar to those in our study.

Table 2: Multinomial Logit Coefficients – Aggregate Level

Understanding consumer preferences and what drives their choice is essential isdeveloping marketing strategies. Based on the results in Table 1, it can be concludedthat changing the wine taste and harmony will have a significant effect on the choiceprobability of red wines, and a marginal effect when improving the bouquet andaftertaste of the wine at the aggregate level. The exact attribute level can be determinedin a different study when several categories, or values, of each variable are consideredto find the optimal level of the specific attribute. The choice-based model was able toidentify those attributes that drive wine choice among four alternative red wines.

As a next step in identifying drivers of wine choice in a heterogeneous consumermarket, we employed the same multinomial logit analysis for different segments basedon gender, frequency of wine drinking (less than once a week and twice a week ormore, for low and high frequency wine drinking), and wine involvement.

With respect to male/female segmentation scheme, our results, presented in Table3, show that taste is a salient attribute for both males and females. These two segments,however, are different with respect to other wine attributes. Harmony plays animportant role in the male segment (harmony is recognized as the balance among allwine attributes) and, to a lesser degree, aftertaste. Bouquet is also significant in thefemale segment. A possible justification for this finding might be that bouquet isconsidered as the feeling in the mouth while drinking the wine, and not the actualmeaning of bouquet, which is the combination of aromas and odors developed in thewine during fermentation and aging.

In sum, the gender segmentation variables revealed interesting dissimilaritiesbetween segments such that the male segment was concerned with intrinsic productcharacteristics that are taste sense-based evaluated. The preference for red wine in thefemale consumer segment, in contrast, was also driven by external productcharacteristics that are other sensor-based evaluated (smell). Based on these results, itcan be seen that personal differences in consumers, such as gender, have an effect ofthe formation of preferences and choice.


Table 3: Multinomial Logit Coefficients - Male and Female Segments

The next step of the analysis is exploring heterogeneity in consumers’ frequency ofdrinking alcohol beverages that is a proxy to their experience with the product.Analyzing the results of this analysis (Table 4), it can be seen that bouquet is a salientattribute in the low frequency wine drinkers’ segment (Table 4). Both segmentsappreciate taste and harmony. The high frequency segment is also affected, to a certaindegree, by the aftertaste and color of the wine. It comes as no surprise that lessexperienced and knowledgeable consumers tend to evaluate products with a smallerset of attributes (Sujan, 1985).

Table 4: Multinomial Logit Coefficients - Low and High Drinking Frequency Segment

To verify whether our segmentation schemes are meaningful (i.e., whetherseparating the sample into two segments should result in better data fitting than in anaggregate sample), we conducted log-likelihood tests, –2 log l, where l = (LLsegments

– LLaggregate), (Gensch, 1985) on the different segmentation schemes. The results ofthis analysis are presented in Table 5.

47Lowengart

Table 5: Segmentation Scheme Log Likelihood Tests

All of these tests are significant at least at the 0.05 level, thus indicating that oursegmentation schemes are meaningful and such consumer groups do behavedifferently in their choice decisions.

Discussion and Conclusions

The purpose of this study was aimed at exploring the effect of sensory-basedproduct attributes on consumer choice, and in a heterogeneous consumer market inparticular. It therefore, presented a general approach for obtaining diagnosticinformation about the saliency of product attributes in a choice context. In order todemonstrate this framework, the paper focused on seven sensory based wine attributesthat were identified as part of consumer considerations. We employed a probabilisticchoice model to address this issue and were able to identify those wine attributes. Inaddition, we estimated the effect of a change in these attributes on the probability ofchoosing a wine. This methodological approach enabled us to gain insight intoconsumer preferences that are driven by attributes that can be managed scientifically,as well as practically, by winemakers. That is, the proposed method added science intoart in the sense that part of the winemaking decision can be based on consumerresearch preferences and perceptions, and not just on expert opinion or trendguessing. However, this is not to say that the other methods supporting product designdecisions are not important in consumers decision of wine purchase. This is also notto say that other product attributes (i.e., price, image, etc.) are of less importance. Forexample, mapping techniques that combine consumer perceptions and preferencescan provide insight into the desired (ideal) product and the proximity of alternativeproducts in the category to this ideal point (Ghose & Lowengart, 2001). There areother quantitative methods that utilize consumers’ sensory evaluation to examineproduct preference that can be found in the literature (Saguy & Moskowitz, 1999;Lesschaeve, 2007). The approach proposed in this study provides a different tool to getbetter accuracy in understanding consumer needs and wants through choice processformation and relevant diagnostic information.

When constructing a marketing strategy for a red wine and utilizing the results ofthis study, marketers can increase the choice probability of their wines by improvingthe taste and emphasizing the wine’s harmony. This can be done either bytechnological improvements or by blends with other varieties of grapes. Naturally, it is


not easy to delineate what is the exact taste and harmony for a preferred wine; rather,this study can indicate which sensory wine attributes are those that influence thechoice process. Wine marketers, therefore, need to construct further sensoryevaluations tests to identify the most preferred tastes and flavors for their wines.Namely, we can indicate “what” should be improved and the question "how much" canbe answered in another study.

Our results also indicated variation in the saliency of the wine attributes acrossdifferent consumer segments that can be incorporated into a better understanding ofcustomer preferences. This market-to-market variation in the male segment, for example,can be translated into offering a wine that is a bit more complex in that it will includeindications of its harmony and aftertaste. A different approach, one that offers a wine thatindicates the bouquet of the wine, can be targeted for the female segment. Such diagnosticinformation can aid wine marketers in constructing more effective marketing strategiesto increase their market share. This can be done by introducing two different wines withdifferent marketing communication strategies that will fit each segment. Such marketingresponses will be more effective than marketing the same wine to both male and femalesegments. Similarly, the consumer segment that purchases wines at low frequency can beeducated about the bouquet of the wine with appropriate communication schemes toincrease the choice probability of purchasing specific red wine.

It should be noted that the proposed framework provides diagnostic informationabout which attribute is salient in the choice process that allows managers to designmarketing strategies for product modifications, or new product development. It doesnot, however, provide insight about the exact level of such (salient) an attribute andthe exact tactic to obtain it. This can be obtained in different research that can examinethe different levels of this attribute.

Overall, this study presented a choice model-based approach for gainingknowledge about current product modifications, as well as developing new productsin categories that are characterized by the high importance of consumer sensoryevaluation in forming preferences toward brands. This is particularly valid incategories where product design decisions are based on experience and art. Identifyingthe salient product attributes for the aggregate and disaggregate markets providemanagers and winemakers with information about the exact product attributes thatneed to be modified. Improving the relevant product attributes will increase consumerchoice probabilities for the specific product (wine) alternative.

The current study introduced a framework for future studies that can focus on theeffect of other consumer characteristics, demographics and others, on wine selection,as well as the manufacturer’s (i.e., winery) effect on the choice of such a product. Thatis, exploring whether consumer heterogeneity in responsiveness to various wineattributes might aid marketers in tailoring marketing strategies that are more targetedand therefore more efficient.

Many different factors can affect the choice decision of a product in a productcategory. These include tangible (e.g., price, quality, packaging, taste, etc.) andnontangible aspects (e.g., reputation, image). For complex products, those that havemany different types of product characteristics, or that have experience or credence innature, the choice task of consumers is even more difficult.

49Lowengart

References

Cliff, M.A. & Dever, M.C. (1996). Sensory and compositional profiles of BritishColumbia Chardonnay and Pinot Noir wines. Food Research International, 29 (3-4):317-323.

Cohen, E. & Lowengart, O. (2003). Exploring consumers wine choice preference.ANZMAC 2003 Conference Proceedings, Adelaide: 259-263.

Cordelle, S., Lange, C. & Schlich, P. (2004). On the consistency of liking scores:insights from a study including 917 consumers from 10 to 80 years old. FoodQuality and Preference, 15: 831–841.

Darby, M.R. & Karni, E. (1973). Free competition and the optimal amount of fraud,Journal of Law & Economics, 16: 67-86.

Gensch, D. H. (1985). Empirically testing a disaggregate choice model for segments,Journal of Marketing Research, 22: 462-467.

Ghose, S. & Lowengart, O. (2001). Taste tests: impacts of consumer perceptions andpreferences on brand positioning strategies, Journal of Targeting, Measurement andAnalysis for Marketing, 10(1): 26-41.

Goodman, S., Lockshin, L. & Cohen, E. (2008). Examining market segments andinfluencers of choice for wine using the Best-Worst choice method. Marketing &Communication, Market Management (Revue Internationale des Sciences Sociales),8(1): 94-112.

Hu, W., Hünnemeyer, A., Veeman, M., Adamowicz, W. & Srivastava, L. (2004).Trading off health, environmental and genetic modification attributes in food.European Review of Agricultural Economics. 31(3): 389-408.

Hughson, A. L. & Boakes, R. A. (2001). Perceptual and cognitive aspects of wineexpertise. Australian Journal of Psychology, 53(2): 103-108.

Johnson, L.F., Bosch, D.F., Williams, D.C. & Lobitz, B.M. (2001). Remote sensing ofvineyard management zones: implications for wine quality. Applied Engineering inAgriculture, 17: 557–560

Koivisto, U-K & Sjödén, P-O. (1996). Food and general neophobia in swedishfamilies: parent–child comparisons and relationships with serving specific foods.Appetite, 26(2): 107-118.

Kolpan, S., Smith, B.H. & Weiss, M.A. (1996). Exploring Wine. New York: VanNostrand Reinhold.

Lesschaeve, Isabelle (2007). Sensory evaluation of wine and commercial realities:review of current practices and perspectives. American Journal of Enology andViticulture, 58(2): 252-258.

Lynch, J. G. & Ariely, D. (2000). Wine online: search costs affect competition on price,quality, and distribution. Marketing Science, 19(1): 83–103.

McFadden, D. (1974). Conditional Logit Analysis of Qualitative Choice Behavior. InZarembkam, P. (Ed.), Frontiers in Econometrics. New York: Academic Press.

Mueller, S., Francis, I.L. & Lockshin, L. (2009). Comparison of best–worst andhedonic scaling for the measurement of consumer wine preferences. AustralianJournal of Grape and Wine Research.


Nelson, P. (1974). Advertising as information, The Journal of Political Economy, 82(4):729-754.

Nerlove, M. (1995). Hedonic price functions and the measurement of preferences: Thecase of Swedish wine consumers. European Economic Review, 39:1697-1716.

Orth, U.R., Wolf, M.M. & Dodd, T.H. (2005). Dimensions of wine region equity andtheir impact on consumer preferences. Journal of Product & Brand Management,14(2): 88-97.

Prescott, J., Norris, L., Kunst, M. & Kim, S. (2002). Estimating a “consumer rejectionthreshold” for cork taint in white wine. Food Quality and Preference, 16(4): 345-349.

Pretorius, I. S. & Bauer, F. F. (2002). Meeting the consumer challenge throughgenetically customized wine-yeast strains. Trends in Biotechnology, 20(10): 426-432.

Ricciardelli, L.A., Connor, J.P., Williams, R.J. & Young, R. (2001). Gender stereotypesand drinking cognitions as indicators of moderate and high risk drinking amongyoung women and men. Drug and Alcohol Dependence, 61:129-136.

Saguy, I.S. & Moskowitz, H.R. (1999). Integrating the consumer into new productdevelopment. Food Technology, 53: 63-73.

Scarpa, R., Philippidis, G. & Spalatro, F. (2005). Product-country images andpreference heterogeneity for Mediterranean food products: a discrete choiceframework. Agribusiness, 21(3): 329–349.

Seguin, M. (1986). Terroirs’ and pedology of wine growing. Cellular and Molecular LifeSciences, 42(8): 861-873.

Shepherd, R. & Towler, G. (1992). Application of Fishbein and Ajzen's expectancy-value model to understanding fat intake. Appetite, 18(1): 15-27.

Skuras, D. (2002). Consumers’ willingness to pay for origin labelled wine: A Greekcase study. British Food Journal, 104(11): 898-912.

Sujan, M. (1985). Consumer knowledge: effects on evaluation strategies mediating.Journal of Consumer Research, 12(6): 31-46.

Thode, S. F. & Maskulka, J. M. (1998). Place-based marketing strategies, brand equityand vineyard valuation. Journal of Product & Brand Management, 7(5): 379-99.

Tomlins, K., Sanni, I .L., Oyewole, O., Dipeolu, A., Ayinde, I., Adebayo, K. & Westby,A. (2007). Consumer acceptability and sensory evaluation of a fermented cassavaproduct (Nigerian fufu). Journal of the Science of Food and Agriculture, 87(10):1949-1956.

Vanier, A., Brun, O.X. & Feinberg, M.H. (1999). Application of sensory analysis ofchampagne wine characterization and discrimination. Food Quality and Preference,10: 101-107.

Walker, L. (2003). The surge from brand Australia. Wines & Vines, 84(7): 28-30.Weaver, Michelle R. (2001). Food preferences of men and women by sensory

evaluation versus questionnaire. Family and Consumer Sciences Research Journal,29(3): 288-301.

Weaver. R.J. (1976). Grape growing. New York: John Wiley.

Group Attributional Style: A Predictor of Individual Turnover Behavior in aManufacturing Setting

Laura Riolli

California State University – Sacramento

Steven M. Sommer

Pepperdine University, Irvine Graduate Campus

Separate research streams have examined (1) teamwork and (2) turnover.We examined the interaction of group beliefs on team member turnoverbehavior. We hypothesized that groups with more pessimistic attributionalstyles would experience greater turnover than optimistic attributional stylegroups. This effect would be independent of influences of group potency andsocial identity. A study of fifty intact work teams in a manufacturing facilitywas conducted, with special attention devoted to recommendations forenhancing the validity of multilevel research. The results supported thehypotheses. Implications for attributional processes, shared team mentalmodels, and social capital are discussed.

Work teams and groups continue to receive increasing attention in managementtheory, research and practice (Goodman, Ravlin & Schminke, 1990; Guzzo &Dickson, 1996; Hackman, 1990; Jackson, Stone & Alvarez, 1993; Labianca, Brass &Gray, 1998; Liden, Wayne & Bradway, 1997). To date, this research has identifiedantecedents of group effectiveness (Earley & Mosakowski, 2000; Gibson, Randell &Earley, 2000; Lau & Murninghan, 1998; Stevens & Campion, 1999) and more oftenhas examined resultant performance (Hollenbeck et al., 1998; Jung & Avolio, 1999;Sparrowe et al., 2001). However, groups are “a collection of individuals with a definite

51Riolli and Sommer

sense of membership and shared beliefs” (Bar-Tal, 1990, p. 41). Those beliefs, in turn,guide group behaviors that relate to collective issues. Only recently has attentionspecifically focused on the formation process of group beliefs (Eby et al., 1999;Gibson, 1999; Guzzo et al., 1993) and their implications for organizational outcomes(Kirkman & Shapiro, 2001).

Group beliefs are neither absolutely an aggregation of individual characteristics nora set of wholly group-level characteristics (Crocker & Luhtanen, 1990; Guzzo et al.,1993; Sayles, 1958). Rather, as members interact, they develop a collective sense oftheir role requirements, behavior patterns, and the connectedness of their actions(Weick & Roberts, 1993). Thus, individual attributes and the group’s contextualcharacteristics meld together to create a team mental model (Eby et al., 1999; Klimoski& Mohammed, 1994) of shared expectations and rules (Hackman, 1990) that guidefuture action. It is important to note these “shared cognitions” need not be identicalamong group members. However, there must be a significant overlap of understandingamong members (Earley & Mosakowski, 2000; Waller et al., 2001) so that they “holdcompatible models that lead to common expectations” (Klimoski & Mohammed,1994, p. 421). In particular, while there is much debate about the nature of sharedmental models (Klimoski & Mohammed, 1994), there is consensus that they are“shared understandings of task demands, environmental contingencies, andappropriate behavior” (Eby et al., p. 367). Thus, they are cognitive frameworks withmotivational potential.

While studies of group processes have examined a variety of topics (e.g., teamworkexpectations, group efficacy, social identity, communication), we propose that one areaof shared cognition thus far overlooked that impacts subsequent behavior isattributional style—the manner in which groups collectively interpret good and badevents relevant to the group. Numerous studies in the behavioral sciences haveexamined the possibility that certain individuals favor some explanations over othersfor different events (Peterson, Buchanan & Seligman, 1995). That is, rather thanindependently evaluating the cause of each experience, over time they develop aconsistent cognitive orientation and interpretive framework. Furthermore, anindividual’s future expectations (e.g., self-efficacy) and behavior (e.g., expended effort)are significantly influenced by their perceptions and explanation of past events(Luthans, 2002a; Stajkovic & Sommer, 2000; Weiner, 1986). Although the majorinterest in attributional style has been at the individual level of analysis (Martinko,1995), some work has been done to extend attribution theory to group settings,especially in sports (Zaccaro et al., 1987; Rettew & Reivich, 1995). This streamsuggests that when individuals work in groups, they also generate sharedunderstandings of the relationship between group attributes and group outcomes(Hewstone, Jaspair & Lalljee, 1982).

It should be noted that attributional style is specifically a cognitive process(Seligman & Csikszentmihalya, 2000) rather than an affective disposition like positiveand negative affectivity (George, 1990; Judge, Locke & Durham, 1997). Dispositionsmay become more state-like upon accumulation of experience (Chen, Gully & Eden,2001), but do not always influence actual behavior (Fishbein & Ajzen, 1975). Forexample, team member negative affectivity does not influence teamwork expectations


(Eby et al., 1999). Attributional style, however, is a cognitive process shown to bemotivational (thus impacting behavior) even when manifesting different emotions(Weiner, 1986). Similar to the distinction Chen et al. (2001) makes between self-efficacy and self-esteem, attributional processes involve motivational evaluations ofself in the context of internal and external criteria, whereas self-esteem is an evaluationresulting in an affective orientation towards self based on external characteristics.These unique implications of attributional style for group behavior relative to othergroup dynamics in the extant research further suggest the need for study.

We investigated the implications of attributional style in the context of turnoverbehavior. Employee turnover is frequently cited as the most prominently studied andamong the most practically relevant topics in organizational behavior research(Luthans, 2002b; Robbins, 2003). Recent discussions have gone beyond indirect costconsiderations and have demonstrated the significance of turnover on firmperformance (Guthrie, 2001). These efforts seem equally divided betweenexaminations of the process of an individual voluntarily leaving an organization (Leeet al., 1999) and the impact of involuntary turnover events like downsizing (Brockneret al., 1987; Mishra & Spreitzer, 1998). While some suggest teams should strengthenan individual’s attachment to an organization (Kirkman & Shapiro, 2001), recent workon social networks and social capital indicate that an individual’s turnover behaviormay be greatly (and negatively) influenced by the turnover behavior of relevant others(Dees & Shaw, 2001; Mollica & DeWitt, 2000; Shah, 2000). Thus, we suggest that anindividual’s turnover behavior is influenced by the collective explanation of the group’sexperiences, and that a shared mental model of pessimistic explanations will create asnowball effect (Krackhardt & Porter, 1985) on turnover.

Attributional Style

Attribution theory is concerned with how individuals perceive causes of events andthe consequences of those perceptions. There is no single theory of “attribution”(Kelley & Michella, 1980; Martinko, 1995). However, research performed by Heider(1958) on how people explain their own actions and those of others’ is widelyconsidered the birth of attribution theory. Subsequent research shows beliefs aboutcausation affects mood, expectations, and subsequent behavior (Stajkovic & Sommer,2000; Weiner, 1986). Weiner’s (1986) theory of achievement motivation deals withhow individuals explain their successes and failures and how this impacts subsequentmood and behavior (self perspective). Kelley (1967) and Green and Mitchell’s (1979)models are concerned primarily with how observers assign responsibility for theoutcomes of others. The application of attribution theory to group settings wouldsuggest members of the group also generate a naive theory of the relationship betweengroup characteristics and group outcomes. The key to understanding the groupexplanation of good and bad events is to be found in the ongoing interaction processamong the group members. While these group level effects have been postulated(Brown, 1984), they have rarely been empirically examined.

One specific stream within the attribution literature—“explanatory style”—examines “one’s tendency to offer similar sorts of explanations for different events”

53Riolli and Sommer

(Peterson, Buchanan & Seligman 1995, p. 4) or, simply put, the habitual way in whichpeople explain the favorable and unfavorable events that happen to them (Peterson &Seligman, 1984). For example, an individual who habitually explains bad events as “Icaused it,” “It’s an ongoing thing and everything else will go wrong” (i.e., internal,stable, and global), is labeled as having a “pessimistic” explanatory or attributionalstyle. In contrast, one who attributes failure to external, unstable, and specific causesis labeled as having an “optimistic” explanatory style. Research on learnedhelplessness shows an individual with a “pessimistic” style is more likely to exhibithelplessness deficits when confronted with bad events than individuals with anoptimistic style (Seligman et al., 1979; Peterson, 2000; Seligman & Schulman, 1986),which will likely lead to dysfunctional consequences in terms of future behavior andperformance (Luthans, 2002a). At this point, we should mention that what has beencalled “explanatory style” in the psychology research has been referred to as“attributional style” in the sparse management literature (Furnham, Sadka & Brewin,1992; Martinko, 1995) on the topic. From here on, we will use the latter term.

There is a prolific body of literature showing that an individual’s attributional stylehas significant effects on their mental and physical well-being, task persistence, andperformance success (Peterson, 2000), and helped to launch the growth of theresearch stream called ‘positive psychology’ (Seligman & Csikszentmihalya, 2000). Todate, attributional style has been related to such diverse outcomes as physical illnesses(Peterson, Seligman & Vaillant, 1988), anxiety (Seligman et al., 1979), academicperformance (Peterson & Barret, 1987), burnout (Wade, Cooley & Savicki, 1986),work exhaustion (Moore, 2000), low self-esteem (Kao, Nagata & Peterson, 1997),hardiness (Hull, Van Treuren & Propson, 1988), and workplace aggression (Douglas& Martinko, 2001). Indeed, a meta-analysis of over 100 studies supported theproposition that depression is positively related to internal, stable, and globalattributions for failure and external, unstable, and specific attributions for success(Sweeney, Anderson & Bailey, 1986). Findings from two decades of research haveshown that an individual’s conclusions that outcomes were uncontrollable wereassociated with cognitive, motivational, and emotional deficits (Abramson, Seligman& Teasdale, 1978; Seligman & Csikszentmihalya, 2000). The motivational deficits area result of the expectation that responses are in vain (Peterson, 2000). The cognitivedeficit is comprised of difficulties in learning, given that one’s responses are not seenas producing outcomes (Hjelle, Busch & Warren, 1996). Finally, the depressed affect(e.g., frustration or sadness) is a consequence of believing outcomes are independentof responses (Garber, Miller & Abramson, 1980).

Attribution theory has long received significant attention in both the clinical andorganizational research (Knowlton & Ilgen, 1980; Liden & Mitchell, 1985; Heneman,Greeberger & Anonyou, 1989; Ployhart & Ryan, 1997). Attributional style, however,has only recently garnered the level of attention in organizational behavior thatapproaches the interest shown in the clinical and social arenas (Furnham et al., 1992;Judge & Martocchio, 1996; Moss & Martinko, 1998; Wunderley, Reddy & Dember,1998). A few relevant studies have examined productivity and turnover amonginsurance sales staff. For example, Seligman and Schulman (1986), using a sample of94 experienced life insurance sales agents, found that individuals who interpreted


failure as internal, stable, and global were less persistent, produced less, initiated fewersales attempts, and quit more frequently. Corr and Gray (1996) replicated thesefindings in a study of an insurance sales staff in the U.K.

The focus of attributional style research has thus far examined individual processesand implications. We reiterate the importance of the work team and repeated calls forresearch to examine group-level influences on theories traditionally examined at theindividual level of analysis (Eby et al., 1999; Pelled & Xin, 1999; Yammarino &Dubinsky, 1990). Prior research has demonstrated the ease with which an individualidentifies with a group. For example, a nominal cue like wearing similar clothing cancue a significant in-group categorization effect (Dovidlio et al., 1995). Once perceivinghimself or herself as a member, the individual is prone to adopt similar attitudes(George, 1990; Salancik & Pfeffer, 1978) and personalize the group’s success andfailures (Ashforth & Mael, 1989). Given the importance of the individual being insync with the group on key coordination and perception issues, if the group is to beeffective (Waller et al., 2001), it seems reasonable to determine if psychologicalwithdrawal (e.g., being in a bad mood) effects found for individuals (Judge &Martocchio, 1996; Pelled & Xin, 1999) would also occur at the group level.

While some attribution work has looked at athletic performance in sports teams(Rettew & Reivich, 1995), attributional style research at the group level is scant. Thisstudy seeks to determine if a construct of group attributional style exists. We definegroup attributional style (GAS) as the group’s habitual and collective manner ofexplaining the causes of bad and good events happening to them. As discussed, groupscreate shared cognitions and collective mental models through their interactions(Earley & Mosakowski, 2000). We again propose that one set of beliefs involves thecollective sense, making governing the explanation of good and bad events happeningto the group. Extrapolating from the existing research, members engaged inpessimistic attributional style discussions may share feelings of helplessness. Thisgroup will experience and collectively amplify/reinforce debilitating deficits that willhinder efforts to correct or improve their activities. Consequently, members ofpessimistic groups will be more inclined to withdrawal, express thoughts of quitting,and intentions to search for alternative employment. When one member quits to takea job elsewhere, others may reevaluate their job status (Dess & Shaw, 2001; Mowday,1981) and likely quit as well. Studies (Lee et al., 1999; Mollica & DeWitt, 2000;Sheehan, 1995) empirically show this shock may lead to potential turnover even if theindividual is satisfied with their current position. We propose this effect will be morepronounced in teams given the more active discussions and higher sense of collectiveexpectations. Therefore:

Hypothesis 1: A pessimistic group attributional style will lead to higher turnover thanturnover in groups characterized by an optimistic attributional style.

Accounting for the Effects of Other Group Beliefs

Group attributional style (GAS) will not occur in a vacuum (Klein & Kozlowski,2000). In particular, we expect the role of GAS to operate in conjunction with other

55Riolli and Sommer

group beliefs that have been demonstrated to influence shared mental models (Eby etal., 1999). In particular, research has demonstrated that group potency and socialidentity produce significant impacts on group dynamics. For example, whileattributional style will result in the collective assessment of an event’s causality, futurebehavior will also be influenced by the confidence the group has that they cansuccessfully mobilize the necessary subsequent resources and tactics. So, whereas,atributional style concerns “Why did it happen?” and “Do we want to do somethingabout it?” we present group potency as the “Can we do something about it?” dynamic.Furthermore, the extent to which the individual is emotionally attached to theparticular group will also add to the desire to remain or turnover. This we consider the“do I care?” or social identity dynamic.

Group Potency The efficacy literature has extensively focused on individual self-regulating

behaviors (Bandura, 1997). However, since the early 1980’s, attention has also beengiven to team-and-group performance beliefs. This trend started with the concept ofteam-potency, which was defined as “a shared conception of group ability acrosssituations” by Guzzo et al. (1993, p. 87). This definition shares elements of and in fact,is often cited as a precursor to the term shared mental models (Eby et al., 1999). Thereis empirical evidence of the relation of potency to performance related criteria(Gibson, 1999). This research stream demonstrates that in different settings, groupbeliefs have a significant effect on different group outcomes. Subsequent work ongroup potency more often uses the term “group efficacy,” defined as the collectivebelief of a group that it can successfully perform a specific task (Gibson et al., 2000;Lindsley, Brass & Thomas, 1995).

Similar to discussions of shared mental models described above, this belief is notthe simple sum of group members’ efficacy beliefs but an “emergent” expectationgenerated through collective sense making (Bandura, 1997). By emergent, we meanthe process of member interactions and accumulated experiences that lead to aframework of shared cognition. The empirical evidence shows this collective (oftencalled the group discussion) approach best predicts group outcomes (Gibson et al.,2000; Little & Madigan, 1997). By suggesting that collective efficacy is deeplygrounded in self-efficacy, Bandura (1997) was among the first researchers to see theconnection between performance beliefs across the two levels of analysis. In anattempt to discriminate between efficacy at the individual and collective levels,Bandura stated:

Linking efficacy assessed at the individual level to performance at the grouplevel does not necessarily represent a cross-level relation. An assessment focusat the individual level is steeped in processes operating within the group. Nordoes a focus at the group level remove all thought about the individuals whocontribute to the collective effort (1997, p. 478).

Consistent with past research (Silver, Mitchell & Gist, 1995), we expect turnoverbehavior to be further influenced by the group’s collective feeling of potency. Teams


that have a strong sense of potency are less likely to experience the sense ofhelplessness that would lead to turnover. Teams with low potency are likely to createan environment that amplifies feelings of helplessness, especially related to assemblingthe skills necessary to succeed. As a result, the effect of potency can potentially maskattributional dynamics such as members relieving their dissonance by leaving theorganization (Abraham, 1999).

Hypothesis 2: High potency groups will experience a lower rate of turnover than low potency groups.

Social Identity Another important group characteristic that may detract from the effect of group

attributional style is the level of member identification with the group. Individualsdefine themselves and others not simply in interpersonal terms, but also in terms oftheir various category memberships (Hewstone et al., 1982) and group ororganizational affiliations (Tajfel & Turner, 1979). As defined by Tajfel (1982), socialidentity is “that part of an individual’s self-concept that derives from their knowledgeof their membership in a social group (or groups) together with the value andemotional significance attached to that membership” (1982, p. 2). Turner (1982)stressed that cohesiveness is the concept of belonging based on affection, whereassocial identity is related to the member’s cognition regarding criteria describing thegroup’s characteristics. Social groups possess specific behavioral expectations, and thisshared understanding of group characteristics and expectations is again relevant to thedefinition of shared mental models.

Individuals with strong identification tend to exert more effort towards groupobjectives and engage in more prosocial behavior (Mael & Ashforth, 1992; Kirkman& Shapiro, 2001). Indeed, one can extrapolate from the social capital literature thatindividuals with strong referent identity tend to build more supportive networks,cooperative work relationships, and higher levels of trust, all of which would reduceone’s motivation to leave the organization (Nahapiet & Ghoshal, 1998).Contemporary work on diversity has emphasized the need to focus on cognitions ofsalient work-related characteristics rather than or in addition to emotional reactions todemographic attributes. Other work on social identity theory (Ashforth & Mael, 1989;Tajfel & Turner, 1986) describes the process by which members will seek todistinguish the group, and by which members’ strength of identification with and ego-enhancement from the group influence intentions to remain.

Thus, in spite of a pessimistic interpretation of group experiences, a highidentification with the group itself may counter turnover. That is, the group itselfprovides a source of support that outweighs experience generated image concerns.

Hypothesis 3: Groups with strong social identity will experience lower turnover than low identity groups.

At this point it might be prudent to restate Hypothesis 1 as follows:

57Riolli and Sommer

Hypothesis 1: A pessimistic group attributional style will lead to higher turnover thanturnover in groups characterized by an optimistic attributional style, independent of theeffects of group potency and social identity.

Methods

SampleThe sample consisted of teams drawn from a major division of a large

manufacturing operation located in the Midwest. This division was responsible formail inserting for large financial institutions. The teams worked around mail insertingmachines and had to collectively coordinate several activities in order to minimizeperformance defects; for example, loading and unloading of different materials (e.g.,envelopes, ink, paper statements), synchronization with machine speed, maintenanceand repair, general work area cleanliness while maintaining or repairing the machines.Each team had 3 or 4 members, a size determined by industrial engineering to be mosteffective for the technology. No difference due to group size was observed for any ofthe study variables. These were permanent groups that spent a considerable amount oftime working in close proximity and socializing with each other during breaks andlunch. Pooling of effort from the group members was an important factor for the teamperformance. Thus, the sample meets the proximity, similarity, interdependence, andinteraction criteria to be considered a team.

Data were retained only for employees where a team’s complete membershipcompleted surveys. This resulted in 180 employees comprising 50 wholly intact teamsused in the actual analyses. Age was measured categorically. The median and mostfrequent response was “3” representing the individual as being in the 26-30 year oldage range. Average organizational tenure was 22.7 months, and 50% of therespondents were female. A comparison of the demographic composition of the wholecompany and the demographics of the investigated division showed the sample thatcompleted the questionnaires was representative of the plant population. Therefore,we are confident there was no sampling bias.

This organization was selected for several expected contributions to validity. Forone, performance was dependent on both human and technological inputs. Themeasure of performance was the number of envelopes zip-sorted in a month. Countersattached to the employee machines tabulated completed envelopes. Managersrecorded group membership and attendance. While a Within-and-Between Analysis(WABA) analysis determined that performance was more variable across groups thanwithin (E = 2.06, significant at 150 [a < .01]), the variance in performance was quiterestricted (3%). Indeed, an ANOVA showed no difference across teams, and aregression of the study variables on performance was also not significant. This addsto the power of the design as group beliefs and subsequent behavior will be more afunction of differences in collective interaction and perceptions than a product ofactual productivity differences.

Finally, the group size provided reasonable opportunity to satisfy Bar-Tal’s (1990)four requirements for effectively measuring group beliefs. The ability to survey the


entire group membership addressed the first two requirements that the constructsreflect the group as a whole, and that members agree with regard to the construct(typically addressed by scale construction). The use of WABA techniques would helpaddress the other two requirements concerning group differentiation and within-groupprocesses. Finally, group beliefs are more predictive of group outcomes when based ongroup interaction processes (Gibson, 1999; Gibson et al., 2000); in this case groupdiscussions of many topics were promoted by the fact that the groups worked togetheraround the machines.

ProcedureRespondents were invited to participate in the study as part of an ongoing project

within the context of reducing turnover at the organization level. The surveyinstrument was directly distributed and collected by the senior author. Completion ofthe survey required approximately 30 minutes and was done on company time.Respondents were given the choice of completing the survey at the start or end of theirshift. All participants were provided with explanations of the general purpose andnature of the study prior to responding. Confidentiality of individual responses wasemphasized in the instructions, and it was stated that only the summaries of theresearch would be provided to management.

Measures

Independent VariablesGroup Attributional Style. Several instruments have been developed to assess

attributional style at the individual level: the Organizational Attribution ScaleQuestionnaire (OASQ) developed by Kent and Martinko (1995), the Attribution StyleQuestionnaire (ASQ) adapted by Peterson et al. (1988), and the OccupationalAttributional Style Questionnaire developed by Furnham et al. (1992). Each isworded specific to a certain population or application. The Group Attributional Stylemeasure (GASQ) used here was based on the Kent and Martinko measure, as it hasshown strong psychometric properties and is specifically worded to tap work-relatedevents. The items were modified to reference group-level opinions in order to followrecommendations for considering issues related to research crossing multiple levelsof analysis (Eby et al., 1999; House, Rousseau & Hunt, 1995; Klein, Dansereau &Hall, 1994). In particular, shifting the item referent from the individual to the grouptends to enhance the level of within group agreement and between group variancerequired to alleviate construct validity threats due to level of analysis (Klein &Kozlowski, 2000).

The measure was presented to the respondents with the following directions:Read each of the situations and imagine a time it happened to you and to yourgroup. Even if it is unlikely that the situation will actually occur, still imagineit is happening and respond to the questions. Based on what you know aboutyourself, your group, and the organization in which you are employed, writedown what you think is the one major cause of the event in the space provided

59Riolli and Sommer

(e.g., bad luck). Respond to each of the items that follow the event by circlingthe number on the scale which best describes the cause you identified.

Following the instructions were the 12 modified events from the Kent andMartinko instrument: 6 good and 6 bad outcomes. A sample reworded negative eventwas “Members of your group have great difficulty in getting along with each other.” Asample reworded positive event was “All the feedback your group has received fromyour supervisor lately concerning the group’s performance has been positive.”Consistent with recommendations for usage, following each event was a space toprovide their narrative cause then parallel questions along the three dimensions ofinternality, stability, and globality. Response anchors used a 7-point Likert format (e.g.,1 = completely external to the group; 7 = completely internal to the group).

Given the high intercorrelations among the three attributional dimensions in pastresearch, Reivich (1995) recommends the use of composite scores for determiningattributional style (Corr & Gray, 1996; Seligman & Schulman, 1986). Again followingKent and Martinko’s recommendations, the composite negative and the compositepositive attributional style scores were calculated first. These scores represent thecombined mean responses across the three dimensions for the 6 events in eachcategory. Next, the total score (CPCN) is obtained by calculating the compositepositive minus the composite negative scores. Higher scores reflect a more optimisticattributional style. Past research (Peterson & Seligman, 1984; Seligman & Schulman,1986; Reivich, 1995; Corr & Gray, 1996) indicates the CPCN is the most validempirical predictor of attributional style at the individual level of analysis. Thus, weexpect the Group-CPCN to be a valid extrapolation to the group level of analysis.TheCronbach alpha for the measure was .76.

Group Potency. We assessed collective self-efficacy with the eight-item scaledeveloped by Guzzo et al. (1993). Other researchers have used different methods tomeasure group-efficacy. For example, Gibson (1999) employed a method called“group discussion procedure,” where a group is presented with a rating scale to use informing a single consensus response to a question about its sense of efficacy withregard to a given task. Limitations of this method include the inability to calculatestatistical indicators of agreement (Gibson, 1999), and that group interaction duringthe process of arriving at an efficacy estimate may change a group’s efficacy to the pointthat it is unrealistic (Bandura, 1997). Even so, Gibson et al. (2000) has shown potencymeasures like Guzzo’s to be unidimensional and sound and furthermore, that the twomeasures are equally sound. While the potency measure was shown to have lowerpredictive validity (Gibson et al., 2000), we propose the nature of the workplacementioned above might exploit both approaches. These individuals work side-by-sideand discuss several topics (including work), so it is highly probable that eachindividual’s response reflected the group’s attitude, thus providing the “referent shiftconsensus” necessary for crossing levels of analysis (Chan, 1998).

Scale items included “My group has confidence in itself,” “My group expects tohave power around here,” and “My group believes it can become unusually good atproducing high-quality work.” Group members completed the eight items using a ten-point format (1 = To no extent, 3 = To a limited extent, 5 = To some extent, 7 = To a


considerable extent, and 10 = To a great extent). The Cronbach alpha was .91.Social identity was measured using a modified version of the Mael and Ashforth

(1992) six-item organizational identification scale. Items included “If someone wereto criticize this group, it would feel like a personal insult,” “When I talk about thisgroup, I say ‘we’ rather than ‘they,’” and “If someone were to praise this group, it wouldfeel like a personal compliment.” Participants were asked to indicate their agreementwith each statement on a five-point scale (5 = To a very great extent; 1 = To no extent).Mael and Ashforth (1992) reported a reliability of .79 and the reliability for this studywas an acceptable .70.

Dependent VariableTurnover Behavior. Organizational records were examined for the 6-month period

following the study to identify study participants who terminated their employment.Unfortunately, we did not obtain sufficient information on date of turnover and thus,could not perform a stronger test using survival analysis.

Significant discussion has recently focused on methods for measuring turnover, aswell as potential flaws in past data collection efforts. For example, turnover is oneform of role transition (Ashforth, 2001) and some transitions can mask withdrawalbehaviors. Transferring jobs or geographic relocation within an organization is a moreacceptable way to leave an unsatisfactory position than quitting, especially ifopportunities in other organizations are limited. Similarly, turnover is defined as “theentire cycle in organizations of entries and leaving” (Bluedorn, 1982, p. 78-79) andagain includes transfers, promotions, and relocations. Given this, and that theconstructs examined here classify as “shared team properties” that are influenced bythe members (Klein & Kozlowski, 2000, p. 215), we utilized a more elaborateoperationalization of turnover.

Exit interview data listed the reason for turnover. For example, some left to returnto school, some were fired, some had immigration problems, and some weretransferred or requested transfers to other departments. Seven categories, includingone for still employed, were created. Consistent with the literature on voluntaryturnover (Lee et al., 1999), these causes were coded into three categories: (1) stillemployed, (2) voluntary turnover or school, requested transfer, and (3) involuntaryturnover—fired, immigration. As described later, since the dependent variable iscategorical, logit regression techniques were used to analyze the data. One analysis wasconducted to examine those who turned over (coded as 0) versus those who stayed(coded as 1), while a second analysis was performed to examine those who leftvoluntarily (coded as 0) versus those who left involuntarily (coded as 1).

Results

Test of Group-Level Effects (WABA)Discussions of multilevel research provide cautions for examining the impact of group

level constructs on individual level variables (House et al., 1995; Klein & Kozlowski, 2000).In this study, we examined how group effects may influence actual turnover behavior at theindividual level. Beyond the climate and intent created by discussion, the mere act of a friend

61Riolli and Sommer

leaving is a significant predictor that the individual may also leave (Krackhardt &Porter, 1985) since closeness (e.g., social identity) is a form of friendship.

In order to justify the group level of analysis, it is important to demonstratehomogeneity within the group and further, that two people within the same group aremore similar than two people who are members of different groups (Bar-Tal, 1990;Florin et al., 1990). WABA was conducted in order to verify the existence of grouplevel effects. This technique uses the within-group correlation and the between-groupcorrelation (called “eta”) to determine if there is more variance among memberswithin a group than variance accounted for by differences across groups. The squaredeta’s (η2’s, similar to R2) are tested relative to one another with F-tests of statisticalsignificance and an E-test of practical significance, making it a more robust measureof group level effects (Klein & Kozlowski, 2000). The cutoff value to concludeconstruct validity at the group level is E larger than 1.3 for the 15° angle test,comparable to a = .01 (Dansereau, Alutto & Yammarino, 1984). Values less than .77indicate individual differences are greater than group effects, and that individual-leveldata cannot be aggregated. Cutoff scores for the F-tests are obtained from critical valuetables and determined by the degrees of freedom (n – J and J – 1). WABA is a usefuland one of the more rigorous tools for determining appropriate levels of analysis inmultilevel research (Klein & Kozlowski, 2000). However, George (1990) points outone should not expect to find extremely large differences across groups when allmembers of a group belong to the same organization and are performing the same task.

Even so, the results of the WABA indicated group level effects for these data. TheE scores for the three independent variables (1.42 to 1.60) all surpassed the 15° valuefor practical significance. Furthermore, the F (df = 130, 49) scores (2.05 to 2.09)surpassed the bracketed critical values of 1.84 (df = 100, 48) and 1.76 (df = 200,50) listed inAppendix A for a = .01 (Dansereau et al., 1984). Thus, variation between groups wassignificantly greater than the variations within groups for the variables of interest, soone can conclude that there is an effect of group membership on the measures.Therefore, we can conclude the existence of collectively shared mental models withinthese groups. Given this, the remaining analyses and testing of hypotheses wereperformed using data aggregated to the group-level. Finally, examination of the datasuggested no violations of normality.

Descriptive StatisticsInspection of the correlation matrix reveals this organization experiences a

moderately high rate of turnover (annualized 60%) that varies across the groups. Olderemployees had lower education levels, and higher levels of turnover which is likely aresult of the physical demands of the job. While there was great variance in theattributional style across the manufacturing teams, the mean was modestly optimistic.Additionally, the groups reported relatively strong levels of potency and identity. Givennone of the demographic measures were related to the dependent variable, they wereexcluded from further analyses. Thus, Hypothesis 1 could be initially tested by examiningthe correlation between attributional style and turnover behavior. Given the codingscheme (lower CPCN = more pessimistic; turned over = 0), the positive correlation showsthat pessimistic groups were likely to produce higher individual turnover.


Hypothesis TestingModerated regression is typically used to study the unique and relative effects of

independent variables on a dependent variable when controlling for each other(Pedhazur, 1982). However, when analyzing a categorical dependent variable withcategorical and continuous independent variables, logistic regression analysis isrecommended (Goodman & Blum, 1996; Tansey et al., 1996).

Hypothesis 1 predicting group explanatory style would relate to group memberturnover was supported. Groups with a more pessimistic explanatory style led tohigher turnover among members than groups with an optimistic explanatory style (b= .23, t < .01), even after accounting for cross-level effects and holding constant theeffects of group potency and social identity (see Table 2). In addition, group memberswho left for voluntary reasons were likely to have left more pessimistic groups (b = .21,t < .05) than individuals who turned over for involuntary reasons (see Table 3).

Table 1: Descriptive Statisticsa

63Riolli and Sommer

Table 2: Logistic Regression (Employees That Left Versus These That Stayed With The Company)

Table 3: Logistic Regression (Employees That Left For Voluntary Versus Involuntary Reasons)

Hypothesis 2 proposed an inverse relationship between group potency andturnover. Again, Tables 2 and 3 present the results of the analysis. Groups with higherpotency experienced lower turnover (b = .67, t < .05). However, there was nodifference for high versus low potency groups in terms of voluntary versus involuntaryreasons for leaving.

Hypothesis 3 was supported in that higher social identity groups experienced lowerturnover in general (b = 2.41, t < .01), and a lower turnover rate due to voluntaryreasons (b = 2.41, t < .01).

Discussion

This study proposed and tested the idea that the attributional style constructexisted at the group level of analysis. WABA analysis of responses from 50 work teamsillustrated the attributional style measures (in fact all the measures of interest) diddisplay group level characteristics in accordance with existing recommendations for


65Riolli and Sommer

conducting multilevel research (House et al., 1995; Klein et al., 1994; Klein &Kozlowski, 2000). It was further proposed that this group level phenomenon wouldinfluence turnover behavior. This proposition was also supported. This impact wasestablished while simultaneously accounting for the influences of group potency andsocial identity firmly established in shared group cognition research. The followingdiscussion will outline theoretical and practical implications.

Consistent with expectations, group attributional style had a significant impact onindividual turnover. This finding adds to the rich body of previous research on factorsthat contribute to employee withdrawal behavior. Previous related research examinedthe increasing effect on individual turnover of communication networks (Krackhard& Porter, 1985), the centrality of network position (Eisenberg, Monge & Miller,1983), and social influence (Kincaid, 1993). More recent studies have shown thatothers can indirectly influence an individual’s turnover behavior—a friend’s turnoverwill increase the chance of an individual’s departure (Mollica & DeWitt, 2000; Shah,2000). Findings from this study showed that a group member’s decision to stay at orto leave a particular job is a function of the quality and pattern of interaction withother group members (Feeley & Barnet, 1997). For example, Krackhardt and Porter(1985) argued that the ‘closer’ the friends who leave the organization are to the personwho stays, the stronger the effect will be on the latter’s turnover considerations. Again,recent research on voluntary turnover shows ‘such a shock’ may induce an individualto leave a job, even if they are not personally dissatisfied (Dess & Shaw, 2001; Lee etal., 1999).

As shown here, and consistent with prior work of shared team mental models, teammember interactions likely developed a group level schema regarding the nature andcauses of their experiences. In particular, the negative attributional style likelyreflected interactions among group members that created collectively agreed upon,ego-protecting explanations for perceived failure. Furthermore, a normative belief inthe immutability of the situation collectively indicated such experiences were notnecessarily indicative of personal failure on any one member’s part. Consequently, thisprocess created mutual social support for leaving. This effect was compounded by thedegree to which the group collectively felt incapable of mobilizing an effectiveimprovement response (potency).

The significant finding for social identity speaks to the issue of who groupmembers may see as “friends” in making cognitions. Similarity across group membersmay enhance social integration (i.e., the degree to which an individual ispsychologically linked to others in the group) and in turn lead to a lower likelihoodof leaving (O’Reilly, Caldwell & Barnett, 1989). Historically, research has shown thatdemographically similar people create a supportive identity group that can reducepressures to leave. Prior research has found higher turnover rates in demographicallydiverse work groups (Jackson et al., 1993), and O’Reilly et al. (1989) discovered thatdisagreement within heterogeneous groups accelerates the departure of members. Thetraditional perspective holds that the presence of demographically or socially“different” members of an otherwise homogeneous group may make the othermembers of the group uncomfortable.

Our findings were to the contrary. The teams in this study were ethnically diverse

(Caucasian, Hispanic, Vietnamese, Native American) and displayed no significantdifferences related to the measures. One explanation may be that communication iseasier and support stronger between individuals with shared social experiences(Zenger & Lawrence, 1989) and the physical arrangement of the workplace createdclose proximity and prolific social interaction as well. The groups in our sample wereheterogeneous, yet the results show the existence of high within-group consensuswhich is consistent with recent research (Earley & Mosakowski, 2000; Lau &Murninghan, 1998; Pfeffer, 1998) showing the “right balance” of diversity in groups isnecessary for greater organizational effectiveness. This study thus shows people in theworkplace can be attracted to each other and create a shared identity because of thework even when they are not similar demographically. This would be consistent withthe growing body of literature claiming social identity can be created around task-related as well as demographic criteria.

As with all research, this study had its strengths and limitations. First, this was thefirst known attempt to specifically examine attributional style as a group levelconstruct, and one of the few to examine how attributional style affects turnoverbehavior. This issue is especially important given the increase in the use of groups andteams in today’s organizations and how little we understand group versus individualmotivation (Sundrom, DeMeuse & Futrell, 1990). Second, as a field study in amanufacturing organization it adds to the generalizability of attributional style beyondthe traditional insurance and sales domains. Third, this research studied groups intheir natural settings with hard outcome measures, thus responding to the need forresearch studies on groups in real organizations (Langfred, 1998), as well as avoidingcommon threats like common method variance. Fourth, this study sought to reduceinternal validity threats by examining teams of a fixed size doing the same task. Finally,our analysis sought to control for multilevel issues that would potentially contaminateour variable of interest.

One important limitation of this study was the sample size. Survey responses wereonly obtained from 50 intact teams. While this number minimally met the thresholdfor sufficient power (.8 at α = .05), small sample sizes tend to be problematic wheninvestigating complex phenomena (Cohen & Cohen, 1983). Thus, interpretations ofthese results must be made cautiously. Even so, our sample of identical and fully intactteams meets or exceeds the sample size typically obtained in research in this stream,and we hope future studies with larger samples will provide support for themoderating effects we proposed.

While this study was longitudinal in terms of the independent variables precedingthe dependent variable, further research using repeated measures designs might resultin a more robust and comprehensive understanding of how the group beliefs identifiedhere influence group outcomes. As mentioned before, we were not able to obtain datesof turnover though such an analysis would have provided a robust insight intopossible relationships between degrees of attributional style and time to turnover. Ithas been noted that many findings concerning groups may not equally apply to “newlyborn” groups (Jackson et al., 1993). The distinguishing characteristic of “newly born”groups is that prior to the formation of the group none of the group’s members haveany formal experience working with one another. In this organization there is a


67Riolli and Sommer

practice of occasionally collapsing remnants of former groups into a new group. Thus,it would be interesting to look at the antecedents and the formation process ofoptimistic and pessimistic group norms in newly-created groups versus groups wheremembers are carried over from prior (optimistic or pessimistic) groups. Indeed, suchan approach would be extremely valuable given the rotation practice commonly usedin professional services firms.

The results of this research suggest that organizations should pay close attention tothe habitual explanations of work groups. The findings of this study indicated groupattributional styles (optimistic or pessimistic) did impact turnover. Companies thatrequire persistence and initiative due to frequent frustration, rejection, and even defeatshould focus on more training that might instill optimism in their employees(Luthans, 2002b). More importantly, these effects were found even thoughperformance was not highly variable. Thus, turnover behavior here was almostexclusively due to differences in collective sense making than to differences in actualoutcomes. Therefore, efforts to help frame experiences optimistically might benefitthe organization even when no actual change in objective circumstances may beneeded (Luthans, 2002a).

Furthermore, our findings relate to work on social networks and social capital. Aperson’s position in a social network can greatly influence their impact on anorganization (Sparrowe et al., 2001), including being a key player that instigates acohort to turnover (Dess & Shaw, 2001). Professionals commonly demonstrate greaterloyalty to their network than the organization (Cappelli, 2000) and thus, a snowballof turnover is common (Dess & Shaw, 2001). However, we, like Krackhardt and Porter(1985), found similar behavior among low-level production employees and this, inand of itself, deserves recognition beyond the implication it suggests for lessorganizationally committed cohorts. It takes considerable time for an organization todevelop the unique knowledge, memory, and interpersonal connections that result inincreased efficiency (Nahapiet & Ghoshal, 1998). Yet, collective turnover of a socialcohort can impede the development of or even eliminate valuable human and socialcapital. While recent recommendations emphasize investing in developing people(Pfeffer, 1998), organizations would be foolhardy to do so, knowing they will leavebefore any return is realized (Guthrie, 2001).

Seligman (1991) explains “learned optimism gets people over the wall—and notjust as individuals but the whole team” (p.256). Workshops for optimism trainingcould teach members of the group how to cope with adversity. How to utilize work-related team attitudes instead of individual differences related to demographics couldbe the source of social support that makes the difference between renewedcommitments to the organization versus pessimistic abandonment through leaving.

References

Abraham, R. (1999). The impact of emotional dissonance on organizationalcommitment and turnover. The Journal of Psychology, 133: 441-455.

Abramson, L.Y., Seligman, M.E.P. & Teasdale, J. (1978). Learned helplessness inhumans: Critique and reformulation. Journal of Abnormal Psychology, 87: 32-48.

Ashforth, B. E. (2001). Role transitions in organizational life: An identity-basedperspective. Mahwah, NJ: Lawrence Erlbaum Associates.

Ashforth, B.E. & Mael, F. (1989). Social identity theory and the organization.Academy of Management Review, 14: 20-39.

Bandura, A. (1997). Self-efficacy: The exercise of control. New York, NY: W.H. Freeman& Co.

Bar-Tal, D. (1990). Group Beliefs. New York: Springer-Verlag.Bluedorn, A. C. (1982). A taxonomy of turnover: Causes, effects, and meanings.

Research in the Sociology of Organizations, 1: 75-128.Brockner, J., Grover, S., Reed, T., DeWitt, R. & O’Malley, M. (1987). Survivor reactions

to layoffs: We get by with a little help from our friends. Administrative SciencesQuarterly, 32: 526-541.

Brown, K. A. (1984). Explaining group poor performance: An attributional analysis.Academy of Management Review, 9: 54-63.

Cappelli, P. (2000). A market-driven approach to retaining talent. Harvard BusinessReview, 78: 103-113.

Chan, D. (1998). Functional relationships among constructs in the same contentdomain at different levels of analysis: A typology of composition models. Journal ofApplied Psychology, 83: 234-246.

Chen, G., Gully, S. M. & Eden, D. (2001). Validation of a new General Self-Efficacyscale, Organizational Research Methods, 4: 62-83.

Cohen, J. & Cohen, P. (1983). Applied multiple regression/correlation analysis for thebehavioral sciences (2nd ed). Hillsdale, NJ: Erlbaum.

Corr, P. J. & Gray, J. A. (1996). Attributional style as a personality factor in insurancesales performance in the UK. Journal of Occupational and Organizational Psychology,69: 83-87.

Crocker, J. & Luhtanen, R. (1990). Collective self-esteem and in-group bias. Journalof Personality and Social Psychology, 58: 47-60.

Dansereau, F., Alutto, J. & Yammarino, F.J. (1984). Theory testing in organizationalbehavior: The varient approach. Englewood Cliffs, NJ: Prentice-Hall.

Dess, G. G. & Shaw, J.D. (2001). Voluntary turnover, social capital, and organizationalperformance. Academy of Management Review, 26: 431-445.

Douglas, S. C. & Martinko, M. J. (2001). Exploring the role of individual differencesin the prediction of workplace aggression. Journal of Applied Psychology, 86(4):547-559.

Earley, P. C. & Mosakowski, E. (2000). Creating hybrid team cultures: An empiricaltest of transnational team functioning. Academy of Management Journal, 43: 26-49.

Eby, L. T., Meade, A. W., Parisi, A. G. & Douthitt, S. S. (1999). The development ofan individual-level teamwork expectations measure and the application of a within-group agreement statistics to assess shared expectations for teamwork.Organizational Research Methods, 2: 366-394.

Eisenberg, E.M., Monge, P.R. & Miller, K.I. (1983). Involvement in communicationnetworks as a predictor of organizational commitment. Human CommunicationResearch, 10: 179-201.

Feeley, T. H. & Barnet, G.A. (1997). Predicting employee turnover from


69Riolli and Sommer

communication networks. Human Communication Research, 23: 370-387.Fishbein, M. & Ajzen, I. (1975). Belief, attitude, intention and behavior: An introduction

to theory and research. Reading, MA: Addison-Wesley.Florin, P, Giamartino, G.A., Kenny, D.A. & Wandersman, A. (1990). Levels of analysis

and effects: Clarifying group influence and climate by separating individual andgroup effects. Journal of Applied Social Psychology, 20: 881-900.

Furnham, A., Sadka, V. & Brewin, C.R. (1992). The development of and occupationalstyle questionnaire. Journal of Organizational Behavior, 13: 27-39.

Garber, J., Miller S.M. & Abramson, L.Y. (1980). On the distinction between anxietyand depression: Perceived control, certainty, and probability of goal attainment. InGarber, J. & Seligman, M.E.P. (Eds.), Human helplessness. Theory and Applications.New York: Academic Press.

George, J.M. (1990). Personality, affect, and behavior in groups. Journal of AppliedPsychology, 75: 107-116.

Gibson, C. B. (1999). Do they do what they believe they can? Group efficacy andgroup effectiveness across tasks and cultures. Academy of Management Journal, 42:138-152.

Gibson, C. B., Randall, A. E. & Earley, P. C. (2000). Understanding group efficacy: Anempirical test of multiple assessment methods. Group and OrganizationManagement, 25: 67-97.

Goodman, J. S. & Blum, T. C. (1996). Assessing the non-random sampling effects ofsubject attrition in longitudinal research. Journal of Management, 22: 627-652.

Goodman, P.S., Ravlin, E.C. & Schminke, M. (1990). Understanding Groups inorganizations. In Cummings, L.L. & Staw, B.M. (Eds.), Leadership Participation,and Group Behavior. Greenwich, CT: JAI.

Green, S.G. & Mitchell, T.R. (1979). Attributional processes of leaders in leader-member interactions. Organizational Behavior and Human Performance, 23: 429-458.

Guthrie, J. P. (2001). High involvement work practices, turnover, and productivity:Evidence from New Zealand. Academy of Management Journal, 44: 180-190.

Guzzo, R.A., Yost, P.R., Campbell, R.J. & Shea, G. P. (1993). Potency in groups:Articulating a construct. British Journal of Social Psychology, 83: 87-106.

Guzzo, R.A. & Dickson, M.W. (1996). Teams in organizations: Recent research onperformance and effectiveness. Annual Review of Psychology, 47: 307-106.

Hackman, J.R. (1990). Groups that work (and those that don’t). San Francisco:Jossey Bass.

Heider, F. (1958). The Psychology of Interpersonal Relations. New York: Wiley.Heneman, R.L., Greeberger, D.B. & Anonyuo, C. (1989). Attributions and exchanges:

The effects of interpersonal factors on the diagnosis of employee performance.Academy of Management Journal, 32: 486-476.

Hewstone, M., Jaspars, J. & Lalljee, M. (1982). Social representations, socialattribution and social identity: The intergroup images of ‘public’ and the‘comprehensive’ schoolboys. European Journal of Social Psychology, 12: 241-269.

Hjelle, L. A., Busch, E. A. & Warran, J. E. (1996). Explanatory style, dispositionaloptimism, and reported parental behavior. Journal of Genetic Psychology, 157:489-500.

Hollenbeck, D. R., Ilgen, D. R., LePine, J. A. & Colquitt, J. H. (1998). Extending themultilevel theory of team decision making: effects of feedback and experience inhierarchical teams. Academy of Management Journal, 41: 269-282.

House, R., Rousseau, D. M. & Hunt, M. (1995). The meso paradigm: A framework forthe integration of micro and macro organizational behavior. In. Cummings, L. L.& Staw, B. M. (Eds.), Research in Organizational Behavior, 17: 71-114.

Hull, J., Van Treuren, R. & Propsom, P. (1988). Attributional style and the componentsof hardiness. Personality and Social Psychology Bulletin, 14: 505-513.

Jackson, S.E, Stone, V.K. & Alvarez, E.B. (1993). Socialization amidst diversity: Theimpact of demographics on work team oldtimers and newcomers. Research inOrganizational Behavior, 15: 45-109.

Judge, T. A, Locke, E. A. & Durham, C. C. (1997). The dispositional causes of jobsatisfaction: A core evaluations approach. Research in Organizational Behavior, 19:151-158.

Judge, T. A. & Martocchio, J. J. (1996). Dispositional influences on attributionsconcerning absenteeism. Journal of Management, 22: 837-861.

Jung, D. & Avolio, B. J. (1999). Effects of leadership style and followers’ culturalorientation on performance in group and individual task conditions. Academy ofManagement Journal, 42: 208-218.

Kao, E. M., Nagata, D. K. & Peterson, C. (1997). Explanatory style, familyexpressiveness, and self-esteem among Asian-American and European-Americancollege students. The Journal of Social Psychology, 137: 435-444.

Kelley, H.H. (1967). Attribution theory in social psychology. In Levine, D. (Ed),Nebraska Symposium on Motivation. Lincoln: University of Nebraska Press.

Kelley, H.H. & Michella, J.L. (1980). Attribution theory and research. Annual Reviewof Psychology, 31: 457-501.

Kent, R.L. & Martinko, M.J. (1995). The development and evaluation of a scale tomeasure Organizational Attribution Style. In Martinko, M.J. (Ed.), Attributiontheory: An organizational perspective. Delaray Beach Florida: St. Lucie Press.

Kincaid, D. L. (1993). Communication network dynamics: Cohesion, centrality, andcultural evolution. In Richards, W.D. & Barnett, G.A. (Eds.), Progress inCommunication Sciences, 12: 111-134. Norwood, NJ: Ablex.

Kirkman, B. L. & Shapiro, D. L. (2001). The impact of cultural values on job satisfactionand organizational commitment in self-managing work teams: The mediating role ofemployee resistance. Academy of Management Journal, 44: 557-569.

Klein, K. J., Dansereau, F. & Hall, R. J. (1994). Level issues in theory development,data collection, and analysis. Academy of Management Review, 37: 191-229.

Klein, K. J. & Kozlowski, S.W. J. (2000). From micro to meso: Critical steps inconceptualizing and conducting multi-level research. Organizational ResearchMethods, 3: 211-236.

Klimoski, R. & Mohammed, S. (1994). Team mental model: Construct or metaphor?Journal of Management, 20: 403-437.

Knowlton Jr., W.A. & Ilgen, D.R. (1980). Performance attributional effect on feedbackfrom superiors. Organizational Behavior and Human Performance, 25: 441-456.

Krackhardt, D.M. & Porter, L.W. (1985). When friends leave: a structural analysis of


71Riolli and Sommer

the relationship between turnover and stayers’ attitudes. Administrative ScienceQuarterly, 30: 242-261.

Labianca, G., Brass, D.J. & Gray, B. (1998). Social network and perceptions ofintergroup conflict: The role of negative relationships and third parties. Academy ofManagement Journal, 41: 55-67.

Langfred, C.W. (1998). Is group cohesiveness a double-edged sword? Aninvestigation of the effects of cohesiveness on performance. Small Group Research,29: 124-143.

Lau, D. C. & Murninghan, J. K. (1998). Demographic diversity and faultlines: Thecompositional dynamics of organizational groups. Academy of Management Review,23: 325-340.

Lee, T. W., Mitchell, T. R., Holton, B., McDaniel, L. & Hill, J. W. (1999). The unfoldingmodel of voluntary turnover: A replication and extension. Academy of ManagementJournal, 42: 450-463.

Liden, R.C. & Mitchell, T.R. (1995). Reactions to feedback: The role of attributions.Academy of Management Review, 20.

Liden, R. C., Wayne, S. & Bradway, L. K. (1997). Task interdependence as a moderatorof the relationship between group control and performance. Human Relations, 50:169-181.

Lindsley, D.H., Brass, D.J. & Thomas, J.B. (1995). Efficacy-performance spirals: amultilevel perspective. Academy of Management Review, 20: 645-678.

Little, B. L. & Madigan, R. M. (1997). The relationship between collective efficacy andperformance in manufacturing work teams. Small Group Research, 28: 517-534.

Luthans, F. (2002a). Positive Organizational Behavior (POB): Implications for theworkplace. Academy of Management Executive, 16: 57-75.

Luthans, F. (2002b). Organizational behavior (9th ed.). Burr Ridge, IL: McGraw-Hill.Mael, F. & Ashforth, B.E. (1992). Alumni and their alma mater: a partial test of the

reformulated model of organizational identification. Journal of OrganizationalBehavior, 13: 103-123.

Martinko, M. J. (1995). The nature and function of attribution theory within theorganizational sciences. In Martinko, M.J. (Ed.), Attribution theory: Anorganizational perspective. Delaray Beach Florida: St. Lucie Press.

Mishra, A. K. & Spreitzer, G. M. (1998). Explaining how survivors respond todownsizing: The roles of trust, empowerment, justice, and work redesign. Academyof Management Review, 23: 567-588.

Mollica, K. A. & DeWitt, R. (2000). When others retire early: What about me?Academy of Management Journal, 43: 1068-1075.

Moore, J. E. (2000). Why is this happening? A causal attribution approach to workexhaustion consequences. Academy of Management Review, 25: 335-349.

Moss, S. E. & Martinko, M. J. (1998). The effects of performance attributions andoutcome dependence on leader feedback behavior following poor subordinateperformance. Journal of Organizational Behavior, 19: 259-274.

Mowday, R.T. (1981). Viewing turnover from the perspective of those who remain:The influence of attitudes on attributions of the causes of turnover. Journal ofApplied Psychology, 66: 120-123.

Nahapiet, J. & Ghoshal, S. (1998). Social capital, intellectual capital, and theorganizational advantage. Academy of Management Review, 23: 242-266.

O’Reilly, C.A., Caldwell, D.F. & Barnett, W.P. (1989). Work group demography, socialintegration, and turnover. Administrative Science Quarterly, 34: 21-37.

Pedhazur, E.J. (1982). Multiple regression in behavior research. New York: Holt,Rinehart & Winston.

Pelled, L. H. & Xin, K. R. (1999). Down and out: An investigation of the relationshipbetween mood and employee withdrawal behavior. Journal of Management, 25:875-895.

Peterson, C. (2000). The future of optimism. American Psychologist, 55: 44-55.Peterson, C. & Barret, L.C. (1987). Explanatory style and academic performance among

university freshmen. Journal of Personality and Social Psychology, 53: 603-607.Peterson, C., Buchanan, G.M. & Seligman, M.E.P. (1995). Explanatory style: History

and evolution of the field. In Buchanan, G. M. & Seligman, M.E.P. (Eds.),Explanatory Style. Hillsdale New Jersey: Lowrence Erlbaum Associates Publishers.

Peterson, C. & Seligman, M.E.P. (1984). Causal explanations as a risk factor fordepression: Theory and evidence. Psychological Review, 91: 347-374.

Peterson, C., Seligman, M.E.P. & Vaillant, G.E. (1988). Pessimistic explanatory styleis a risk factor for physical illness: A thirty-five year longitudinal study. Journal ofPersonality and Social Psychology, 55: 23-27.

Pfeffer, J. (1998). The human equation. Boston: Harvard Business School Press.Ployhart, R.E. & Ryan, A.M. (1997). Toward explanation of applicant reactions: An

examination of organizational justice and attribution frameworks. OrganizationalBehavior and Human Decision Processes, 72: 308-335.

Reivich, K. (1995). The measurement of explanatory style. In Buchanan, G. M. &Seligman, M. E. P. (Eds.), Explanatory Style. Hillsdale New Jersey: LowrenceErlbaum Associates Publishers.

Rettew, D. & Reivich, K. (1995). Sport explanatory style. In Buchnan, G. M. &Seligman, M.E.P. (Eds.), Explanatory Style. Hillsdale, New Yersey: LawrenceErlbaum Associates, Publisher.

Robbins, S. P. (2003). Essentials of organizational behavior (7th ed.). Upper SaddleRiver, NJ: Prentice Hall.

Salancik, G. R. & Pfeffer, J. (1978). A social information processing approach to jobattitudes and task design. Administrative Science Quarterly, 23: 224- 253.

Sayles, L.R. (1958). The behavior of industrial work groups. New York: Wiley.Seligman, M.E.P., Abramson, L.Y., Semmel, A. & Von Baeyer, C. (1979). Depressive

attributional style. Journal of Abnormal Psychology, 88: 242-247.Seligman, M. E. P. & Csikszentmihalya, M. (2000). Positive psychology. American

Psychologist, 55: 5-14.Seligman, M.E.P. & Schulman, P. (1986). Explanatory style as a predictor of

productivity and quitting among life insurance sales agents. Journal of Personalityand Social Psychology, 50(4): 832-838.

Seligman, M.E.P. (1991). Learned Optimism. New York: Knopf.Shah, P. P. (2000). Network destruction: the structural implications of downsizing.

Academy of Management Journal, 43: 101-112.


73Riolli and Sommer

Sheehan, E. P. (1995). Affective responses to employee turnover. The Journal of SocialPsychology, 135: 63-70.

Silver, W.S., Mitchell, T.R. & Gist, M.E. (1995). Responses to successful andunsuccessful performance: The moderating effect of self-efficacy on therelationship between performance and attributions. Organizational Behavior andHuman Decision Processes, 62: 286-299.

Sparrowe, R. T., Liden, R. C., Wayne, S. J. & Kraimer, M. L. (2001). Social networksand the performance of individuals and groups. Academy of Management Journal,44: 326-338.

Stajkovic, A. & Sommer, S. M. (2000). Self-efficacy and causal attributions: Direct andreciprical links. Journal of Applied Social Psychology, 30: 707-737.

Stevens, M. J. & Campion, M. A. (1999). Staffing work teams: development andvalidation of selection tests for teamwork settings. Journal of Management, 25:207-228.

Sweeney, P., Anderson, K. & Bailey, S. (1986). Attributional Style in depression: a metaanalytic review. Journal of Personality and Social Psychology, 50: 974-991.

Tajfel, H. (1982). Social identity and intergroup relations. Cambridge: CambridgeUniversity Press and Paris: Editions de la Maison des Sciences del’Homme.

Tajfel, H. & Turner, J.C. (1979). An integrative theory of intergroup conflict. In Austin,W. G. & Worchel, S. (Eds.), The social psychology of intergroup relations.Monterey, Calif.: Brooks/Cole.

Tajfel, H. & Turner, J.C. (1986). The social identity theory of intergroup behavior. InWorchel, S. & Austin, W.G. (Eds.), Psychology of intergroup relations. Chicago:Nelson-Hall.

Tansey, R., White, M., Long, R. G. & Smith, M. (1996). A comparison of log-linearmodeling and logistic regression in management research. Journal of Management,22: 339-358.

Turner, J.C. (1982). Towards a cognitive redefinition of the social group. In Tajfel, H.(Ed.), Social identity and intergroup relations. Cambridge: Cambridge UniversityPress and Paris: Editions de la Maison des Sciences de l”Homme.

Wade, D., Cooley, E. & Savicki, V. (1986). A longtitudinal study of burnout. Childrenand Youth Services Review, 8(2): 161-173.

Waller, M. J., Conte, J. M., Gibson, C. B. & Carpenter, M. A. (2001). The effect ofindividual perceptions of deadlines on team performance. Academy of ManagementReview, 26: 586-600.

Weick, K.E. & Roberts, K.H. (1993). Collective mind in organizations: Heedfulinterrelating on flight decks. Administrative Science Quarterly, 38: 357-381.

Weiner, B. (1986). An attributional theory of motivation and emotion. New York:Springer-Verlag.

Yammarino, F. J. & Dubinsky, A. J. (1990). Salesperson performance and manageriallycontrollable factors: An investigation of individual and work group effects. Journalof Management, 16: 87-106.

Zenger, T.R. & Lawrence, B.S. (1989). Organizational demography: The differentialeffects of age and tenure distribution on technical communications. Academy ofManagement Journal, 2: 353-376.

75Wang and Campbell

Business Failure Prediction forPublicly Listed Companies

in China

Ying Wang


Michael Campbell


This study uses data from Chinese publicly listed companies for the periodof September 2000-September 2008 to test the accuracy of Altman’s Z-scoremodel in predicting failure of Chinese companies. Prediction accuracy wastested for three Z-score variations: Altman’s original model, a reestimatedmodel for which the coefficients in Altman’s model were recalculated, and arevised model which used different variables. All three models were found tohave significant predictive ability. The reestimated model has higherprediction accuracy for predicting nonfailed firms, but Altman’s model hashigher prediction accuracy for predicting failed firms. The revised Z-scoremodel has a higher prediction accuracy compared with both the reestimatedmodel and Altman’s original model. This study indicates that the Z-scoremodel is a helpful tool in predicting failure of a publicly listed firm in China.

Developing countries are attracting more foreign investment than ever before. Since2000, foreign direct investment inflows have rocketed from $165.5 billion to anestimated $470.8 billion in 2007. According to the World Bank, China draws the mostforeign investments, attracting $84 billion of investment in 2007 and representing 18%of the total. Although China is an attractive place for investment, publicly listedChinese companies suffer credibility issues. All three stock exchange markets –Shanghai, Shenzhen, and Hong Kong – are to varying degrees, known for governmentintervention and a club type atmosphere. Investors need guidelines to distinguish low-risk investments from higher-risk ones. The objective of this study is to determine if the

information available in the annual reports of Chinese publicly listed companies isuseful to predict which companies are likely to fail.

The following research questions are considered in this paper: Is Altman’s Z-scoremodel effective for predicting company failure in China during the period of 2000-2008? Is the model effective for predicting company failure for many different typesof firms, not solely for manufacturing companies? Will recalculation of the coefficientsof Altman’s variables result in more accurate failure prediction? Can other variables besubstituted in the basic Z-score model to create a more accurate model?

Previous Studies

The prediction of company failure has been well-researched using developedcountry data (Beaver, 1966; Altman, 1968; Wilcox, 1973; Deakin, 1972; Ohlson, 1980;Taffler, 1983; Boritz, Kennedy & Sun, 2007). A variety of models have been developedin the academic literature using techniques such as Multiple Discriminant Analysis(MDA), logit, probit, recursive partitioning, hazard models, and neural networks.Summaries of the literature are provided in Zavgren (1983), Jones (1987), O’Leary(1998), Boritz et al. (2007) and Agarwal and Taffler (2007). Despite the variety ofmodels available, both the business community and researchers often rely on themodels developed by Altman (1968) and Ohlson (1980) (Boritz et al., 2007). A surveyof the literature shows that the majority of international failure prediction studiesemploy MDA (Altman, 1984; Charitou, Neophytou & Charalambous, 2004).

Beaver (1966) presented empirical evidence that certain financial ratios, mostnotably cash flow/total debt, gave statistically significant signals well before actualbusiness failure. Altman (1968) extended Beaver’s (1966) analysis by developing adiscriminant function which combines ratios in a multivariate analysis. Altman (1968)found that his five ratios outperformed Beaver’s (1966) cash flow to total debt ratio andcreated the final discriminant function:

Z=1.2X1+1.4X2+3.3X3+0.6X4+0.999X5

where,

X1 = working capital/total assetsX2 = retained earnings/total assetsX3 = earnings before interest and taxes/total assetsX4 = market value of equity/book value of total liabilitiesX5 = sales/total assets

Firms with Z-scores less than 2.675 are predicted to be bankrupt, and firms withZ-scores greater than 2.675 are predicted to not be bankrupt.

Boritz et al. (2007) reestimated the model using Canadian company data andobtained the following:

Z=2.149X1-0.624X2+1.354X3-0.018X4+0.463X5


The cutoff point is 0.27.

Taffler (1983) developed a UK-based Z-score model as follows:

Z=3.20+12.18X1+2.50X2-10.68X3+0.029X4

where,

X1 = profit before tax/current liabilitiesX2 = current assets/total liabilitiesX3 = current liabilities/total assetsX4 = (quick assets-current liabilities)/daily operating expenses with the denominatorproxied by (sales-PBT-depreciation)/365

Sandin and Porporato (2007) use data from a developing country, Argentina, andretain 2 out of 13 ratios after stepwise selection and come up with the final model:

As=15.06R5+16.11S3-4.14

where,

R5 = operative income/net salesS3 = shareholder’s equity/total assets

Despite the popularity of the MDA technique in constructing failure classificationmodels, questions were raised regarding the restrictive statistical requirementsimposed by the models (Ohlson, 1980). To overcome the limitations, Ohlson (1980)employed logistic regression to predict company failure, but the model was suggestedto be insensitive to financial distress situations (Grice & Dugan, 2001).

Boritz et al. (2007) question the suitability of using the Altman (1968) and Ohlson(1980) models for Canadian companies since the Altman-Ohlson models weredeveloped using data from U.S. firms. They contend that new models must bedeveloped and validated for use with Canadian firms because of various differences inthe environments in which firms of the two countries operate. This argument appliesequally well to the need to develop and validate new models for evaluating Chinesefirms. Along these same lines, Grice and Ingram (2001) argue that original Z-scorecoefficients should be reestimated when examining firms of different time periods orin different industries.

Methodology

As mentioned earlier, the majority of international failure prediction studies employMDA (Altman, 1984; Charitou et al., 2004). This study employs MDA to allow bettercomparison with other international studies. This research plan avoids one previouscriticism of MDA analysis. Ohlson (1980) is concerned about using predictors of failure

77Wang and Campbell

that are derived from information published after bankruptcy has occurred. In this study,all information is from reports published at least three months before a company wasdelisted. Agarwal and Taffler (2007) emphasize the importance of testing the predictiveability of models against an entire population instead of using only a relatively smallsample. The authors plan to address this issue in a subsequent study. The currentresearch plan is to test the predictive ability of three Z-score based models using thematched pair technique. Two of the models are actually developed in this study.

Selection of Failed FirmsIn order to select failed firms, we must define “failure” first. “Failure” is defined

as the inability of a firm to pay its financial obligations as they mature (Beaver, 1966).In another words, insolvency. In the analysis in this paper, we work exclusively withfirm insolvencies on the basis that these are clean measures. Because firm insolvencyis such a stringent criterion, this approach potentially weakens the predictive abilityof the Z-score model, in particular in terms of increasing the type II error rate –misclassification of nondelisted firms as delisted.

The failed firms in this sample are firms that were publicly listed in Shanghai StockExchange Market (SHSE) or Shenzhen Stock Exchange Market (SZSE) for at least twoconsecutive years and then were delisted during 2000-2008 due to financial problems.According to the “Public Listing Regulation” published in 2000 by the China SecuritiesRegulatory Committee, four situations will lead to the delisting of a publicly listedcompany. The first situation is privatization or other changes of shareholderscomposition. The second situation is failing to disclose financial information orfinancial fraud. The third situation is illegal activities by the listing firm. The fourth isbeing unprofitable for three consecutive years. This study selected only those firms thatwere delisted for either situation two or four. For firms delisted because of situation one,the company is not considered failed, only that the shareholders have decided toprivatize the company or the company is merged into another company. For situationthree, this study believes that firms delisted because of illegal activities are different fromfirms delisted because of financial problems. Firms delisted because of illegal activitiesmight still be financially sound and thus cannot be predicted with financial ratios. Wetreat the event of being delisted as a clear signal of firm failure. We look at firm failurefrom the investors’ standpoint. Once the firm is delisted, its stocks become worthlesssince there is no platform for exchange of the stocks any more. The delisted firm ingeneral will continue operating for a period of time, but shareholders have essentiallylost their investment. Although there have been continuous demands for establishingplatforms for exchanges of stocks of delisted firms, no such platform has been created.

Selection of matching firmsThe selection process was based upon a paired-sample design. For each delisted

firm in the sample, a nondelisted firm of the same industry and asset size was selected.If the exact match of asset size could not be found, the firm which had the closest assetsize was chosen. The asset size was based upon the asset size reported on the lastfinancial statement of the delisted firm and the asset size of the matching firm reportedfor that same year.


Data CollectionFor every delisted and matching nondelisted firm, the financial data were

manually collected for up to two years prior to delisting from www.sina.com.cn.According to Altman (1968), the bankruptcy prediction model is an accurateforecaster of failure for up to two years prior to bankruptcy. Accuracy diminishessubstantially as the lead time is increased. A total of 42 delisted firms (16manufacturing companies) were collected along with 42 (16 manufacturingcompanies) matching nondelisted firms. We then randomly selected 12 out of the 42delisted firms along with their matching nondelisted firms as the prediction or holdout sample to test the validity of our Z-score model. The final sample was divided intotwo subsamples: the estimation sample which includes 30 delisted firms and 30matching nondelisted firms, and the prediction sample which includes 12 delistedfirms and 12 matching nondelisted firms.

Results

Descriptive statisticsThe average time between the actual delisting date and submission of the last

financial report prior to the delisting for the 30 failed firms was eight months, rangingfrom three months to 23 months. The average asset size for the delisted firms was466,629,673 Chinese dollars versus 747,952,379 for the nondelisted firms for the firstyear prior to failure. The respective numbers are 882,387,177 and 693,322,301 for thesecond year prior to delisting. There was a sharp decrease of mean total asset size ofthe delisted firms between the two financial reporting periods prior to delisting, whilethe total assets of the nondelisted firms increased.

The sizes of the firms vary. The total assets of the delisted companies range fromRMB 21,514,900 to RMB 1,211,942,318 the first year before delisting. The sales of thedelisted firms range from 0 to RMB 433,961,140 in the first year before delisting. Thetotal assets of the nondelisted firms range from RMB 208,295,652 to RMB2,394,944,689 for the corresponding year. The sales of the nondelisted firms rangefrom RMB 5,918,570 to RMB 986,715,195 for the corresponding year.

The means of the financial ratios using the financial reports one and two yearsprior to delisting are summarized in Tables 1 and 2, respectively. The results areconsistent between the estimation and prediction groups for both years. A comparisonof the delisted and nondelisted variable means indicates that working capital/totalassets (X1), retained earnings/total assets (X2), earnings before interest and taxes/totalassets (X3), market value of equity/book value of total debt (X4) and sales/total assets(X5) are lower in the delisted than in the nondelisted group. The p-values for the testof mean differences between delisted and nondelisted companies are significant foreach of these variables. The results are similar to those reported by Altman (1968) forhis estimation sample except for the sales/total assets variable (X5), which is notsignificantly different between his bankrupt and non-bankrupt groups. The resultsreported by Grice and Ingram (2001) do not find significant differences between thedistressed and nondistressed groups for variables X4 and X5.

79Wang and Campbell

Table 1: Descriptive statistics for estimation subsample and prediction subsamples using the annual financial report one year prior to delisting


Table 2: Descriptive statistics for estimation subsample and prediction subsamples using the annual financial report two years prior to delisting

81Wang and Campbell

Classification accuracy of Altman’s (1968) Z-score modelWe evaluated the classification accuracy of Altman’s (1968) Z-score model using

the estimation sample and prediction sample respectively. The Z-scores are derived forboth samples using two years of financial data. The accuracy of the Z-score model iscalculated by dividing the number of firms correctly predicted by the total number offirms in the sample.

Table 3 reports results of tests of Altman’s (1968) model. The model does fairly wellfor predicting the delisting of a firm, with accuracy ranging from 91.67% to 100%. Themodel tends to misclassify a nondelisted firm into the delisted group with Type II errorranging from 16.67% to 43.33%. The model does well using financial data 2 years priorto delisting, with an overall accuracy of 85% for the estimation sample and 87.5% forthe prediction sample. The tendency to misclassify a nondelisted firm into the delistedgroup persists.

Table 3: Comparisons of classification accuracies using coefficients from Altman’s (1968) model

Classification accuracy of the reestimated model (one year prior to delisting)Additional evidence of the stationary nature of the Z-score model is obtained by

reestimating the model’s coefficients using our estimation sample, then testing theprediction accuracy of our model using the prediction sample. Table 4 reports resultsfor the reestimated model.

The sample of 30 delisted firms and the 30 corresponding nondelisted firms isexamined using MDA. Since the discriminant coefficients and the group distributionsare both derived from this sample, a high degree of successful classification is expected.

The Z-score model derived is:Z=0.8059X1-0.2898X2+0.0440X3+0.1971X4+6.3327X5

Firms with Z-scores less than 2.2373 are predicted to be delisted and Z-scoresgreater than 2.2373 are predicted to be nondelisted.


Table 4: Comparisons of classification accuracies using newly derived coefficients

The model correctly predicted 90% of firms (54 out of 60), with both type I andtype II error at 10%. This is higher than 76.67% overall accuracy for the estimationsample using Altman’s (1968) model. However, the Type I error is lower using Altman’s(1968) model (3.33%) compared with our model. Altman’s (1968) model misclassified13 out of 30 nondelisted firms into delisted group while it only misclassified 1 out of30 delisted firms into nondelisted group. The model this study derives misclassified 3out of 30 nondelisted firms into the delisted group and 3 out of 30 delisted firms intothe nondelisted group.

Classification accuracy of the reestimated model (two years prior to delisting)The second test is made to observe the discriminating ability of the model for firms,

using data from two years prior to delisting. Fifty two out of 60 firms are properlyclassified (86.67%), with a Type I error of 10% and a Type II error of 16.67%. Theprediction power of the model is quite constant across the two years. The predictionaccuracy is 85% using Altman’s (1968) model with a Type I error of 0 percent and TypeII error of 30%. Our model correctly classified 27 out of 30 delisted companies and 25out of 30 nondelisted firms, while Altman’s (1968) model correctly classified all the 30delisted firms and misclassified 9 out of the 30 nondelisted firms.

Cross-validationIt is important to cross-validate the result using hold out data. Using data one year

prior to delisting, 21 out of 24 of the prediction group firms (87.5%) are correctlyclassified using the derived Z-score model, with a Type I error of 16.67% and a Type IIerror of 8.33%. The model misclassified 2 out of the 12 delisted firms and 1 out of the12 nondelisted firms. Altman’s (1968) model has an overall accuracy of 83.33%. Itcorrectly classified all the delisted firms and misclassified 4 out of the 8 nondelistedfirms.

Using two years prior to delisting data, our model arrives at exactly the sameresults as using one year prior to delisting data. Altman’s (1968) model has an overall

83Wang and Campbell

accuracy of 87.5% and it misclassified 1 out of the 12 delisted firms and 2 out of the12 nondelisted firms.

Analysis

During the process of data collection, we noticed that the delisted firms’ total assetsdecreased over the two year period, while the nondelisted firms’ total assets increased.Although no previous research has taken this into consideration, we believe it worthfurther exploration. We thus added another variable into the discriminant function.The sixth variable is defined as follows: (Total assets one year prior to delisting – Totalassets two years prior to delisting)/Total assets two years prior to delisting. We thenapplied a backward elimination procedure. Three variables remained after theprocedure with a significance level of p<0.05. The specific p values are shown in Table5. The three variables are: X4, X5 and X6. Using these three variables, we createdanother Z-score model, the revised Z-score model.

Z=0.2086X4+4.3465X5+4.9601X6

Table 5: Variables retained after backward elimination procedure

Firms with a Z-score smaller than 1.5408 are predicted to be delisted, while firmswith a Z-score larger than 1.5408 are predicted to be nondelisted. The prediction resultsusing the revised Z-score model are reported in Table 6. The revised model correctlyclassified 95% of the firms in the estimation sample. It misclassified 3 out of the 30delisted firms and correctly classified all the nondelisted firms. The estimation sampleoverall accuracy rates of the revised model are 95% and 91.67% respectively for oneyear and two years prior to delisting. These rates were comparatively more accuratethan those of Altman’s model at 76.67% and 85% and the re-estimated model at 90%and 86.67%. The cross validation results are also reported in Table 6. Using theprediction sample, the revised model yields superior results one year prior to delisting,and all 3 models yield the same overall accuracy for two years prior to delisting.

Relatively few studies of this type have been done in emerging countries. Sandinand Porporato (2007) did a study for Argentine companies. New Z-score variablesspecific to Argentine companies were developed. Then, both the new model andAltman’s original Z-score model were tested. Both models were found to havepredictive ability, with the new model enjoying enhanced predictive power. Thus,both the current study on Chinese companies and Sandin and Porporato (2007)support the contention that the Z-score model is an effective predictor of company


failure in emerging countries, especially when the model is revised based on datafrom the specific country being studied.

Table 6: Comparisons of classification accuracies using revised Z-score model

Because of the relatively small number of failed firms during the period understudy, all failed firms were included regardless of their industry. As mentioned earlier,of the 84 firms used in this study, only 32 (16 failed and 16 healthy) weremanufacturing firms. It is interesting to note that even though Altman’s original Z-score model was developed based only on manufacturing firms, it performed well onthis cross section of Chinese firms.

Our study shows that the revised model with three variables has a comparativelymore accurate prediction than both the Altman’s model and the reestimated modelusing one year prior to delisting data for both the estimation and the predictionsample. The revised model also has comparatively more accurate prediction thanboth the Altman’s model and the reestimated model using two years prior to delistingdata for the estimation sample. The three models perform the same using two yearsprior to delisting data for the prediction sample. Table 7 summarizes the results.Table 8 shows the results in separate charts to facilitate a comparison.

Table 7: Comparison of classification accuracies of different models

85Wang and Campbell

Table 8: Comparison of classification accuracies of different models (Chart presentation)

Conclusion

Our study supports the effectiveness of the Z-score methodology for predictingcompany failure in China. Overall, the re-estimated model with recalculatedcoefficients but the same five financial ratios as Altman’s model has a higher predictionaccuracy for the nondelisted group, while Altman’s (1968) model has higherprediction accuracy for the delisted group. Our revised model with three financialratios has higher overall prediction accuracy than both the re-estimated model andAltman’s model. The revised model includes a financial ratio that is not considered inthe other two models. It is defined as follows: X6 = (Total assets one year prior to


delisting – Total assets two years prior to delisting)/Total assets two years prior todelisting. This variable indicates the extent of asset decrease from two years to one yearprior to delisting.

Our models use companies from various industries. The models developedshould apply to a wide variety of firms. Due to the limitations of data access and thematched sample method when estimating Z-score model, this study uses a relativelysmall sample. One of the criticisms of failure prediction models in general, is thatthey have not been tested on an entire underlying population (Agarwal & Taffler,2007). Future research is planned to test the 3 models in this paper against the entirepopulation of Chinese listed companies for a longer period. Future research also isplanned to employ Ohlson’s (1980) logit model with a large sample or wholepopulation. It then may be possible to compare the efficacy of MDA versus logit forChinese listed companies.

References

Agarwal, V. & Taffler, R.J. (2007). Twenty-five Years of the Taffler Z-score Model: DoesIt Really Have Predictive Ability? Accounting and Business Research, 37(4): 285-300.

Altman, E.I. (1968). Financial Ratios, Discriminant Analysis and the Prediction of Corporation Bankruptcy. The Journal of Finance, 23: 589-609.

Altman, E.I. (1984). The Success of Business Failure Prediction Models: An International Survey. Journal of Banking & Finance, 8: 171-98.

Beaver, W.H. (1966). Financial Ratios as Predictors of Failure. Journal of Accounting Research, 4: 71-111.

Boritz, J., Kennedy, B. & Sun, J.Y. (2007). Predicting Business Failure in Canada. Accounting Perspectives, 6(2): 141-65.

Charitou, A., Neophytou, E. & Charalambous, C. (2004). Predicting corporate failure: Empirical Evidence for the UK. European Accounting Review, 13(3): 465-97.

Deakin, E.B. (1972). A Discriminant analysis of Predictors of Business Failure. Journal of Accounting Research, 10: 167-79.

Grice, J.S. & Dugan, M.T. (2001). The Limitations of Bankruptcy Models: SomeCautions for the Researcher. Review of Quantitative Finance & Accounting, 17(2):151-166.

Grice, J.S. & Ingram, R.W. (2001). Tests of the Generalizability of Altman’s Bankruptcy Prediction Model. Journal of Business Research, 54(1): 53-61.

Jones, F. (1987). Current Techniques in Bankruptcy Prediction. Journal of Accounting Literature, 6: 131-64.

Ohlson, J.A. (1980). Financial Ratios and the Probabilistic Prediction of Bankruptcy. Journal of Accounting Research, 18(1): 109-31.

O’Leary, D. (1998). Using Neural Networks to Predict Corporate Failure. InternationalJournal of Intelligent Systems in Accounting, Finance and Management, 7(3): 187-97.

Sandin, A. & Porporato, M. (2007). Corporate Bankruptcy Prediction ModelsApplied to emerging Economies: Evidence from Argentina in the Years 1991-1998.International Journal of Commerce and Management, 17(4): 295-311.

Taffler, R.J. (1983). The Assessment of Company Solvency and Performance Using a

87Wang and Campbell

Statistical Model. Accounting and Business Research, 15(52): 295-308.Wilcox, J.W. (1973). A Prediction of Business Failure Using Accounting Data. Journal

of Accounting Research, 11: 163-79.Zavgren, C. (1983). The Prediction of Corporate Failure: The State of the Art. Journal

of Accounting Literature, 2: 1-38.


89Wheatley and Doty

Executive Compensation as aModerator of the Innovation –Performance Relationship1

Kathleen K. Wheatley

University of Tennessee at Chattanooga

D. Harold Doty

University of Texas at Tyler

Little research has been done to try and connect type of compensation withthe use of a specific competitive strategy. We propose that compensation(percentage of base, bonus, options-granted, and stock for the topmanagement team) will moderate the innovation strategy to performancerelationship based on risk and time horizon. Analyses of panel data from1994 to 1998 for 380 firms show that the innovation strategy to performancerelationship is moderated by bonus and options-granted compensation.These findings suggest that implementing an innovation strategy and usinga high percentage of bonus compensation will lead to greater performance.Alternately, implementing an innovation strategy and using a low percentageof options granted will create the best outcome. Our findings help shed lighton the firm-specific mechanisms that enable strategy implementation.

Recent global and economic conditions have reduced the slack available toorganizations and have also heightened the need for effective strategy implementation.Given global economic realities, it is critical that firms focus on all aspects of theorganization necessary to implement their chosen strategy. Previous research hasdemonstrated that a variety of organizational attributes are critical to implementationefforts. These include supply chain coordination, organizational design, workforceconfiguration, and human resource management policies (Shaw, Gupta & Delery,

1 This study has been partially funded by the Snyder Innovation Research Center at Whitman School ofManagement, Syracuse University.


2001; Slater & Olson, 2001). Firms that establish a better fit between organizationalattributes and their strategy are better able to implement the strategy and haveperformance advantages as well (Allen & Helms, 2002; Gomez-Mejia, 1992; Lerner &Wulf, 2007; Slater & Olson, 2001; Yanadori & Marler; 2006; Xue, 2007).

A second area of popular concern, particularly after highly visible corporatecollapses, bankruptcies, and accounting scandals, is the role of executivecompensation in firm performance. Much of the current compensation research hasbeen framed using agency theory (Fama & Jensen, 1983; Jensen & Meckling, 1976)and has provided inconsistent findings (Barkema & Gomez-Mejia, 1998). As a resultof these divergent findings, some researchers have suggested looking outside of theagency framework (Garen, 1994; Jensen & Murphy, 1990). We agree that limiting ourviewpoint to agency theory and considering only the direct relationship betweencompensation and performance is too restrictive. This restriction is not onlyresponsible for some of the divergent compensation results, but has also delayed theintegration of compensation research in the area of strategy implementation.

An issue at the intersection of the implementation and compensation research isthe role of executive compensation in strategy implementation. As noted by Barkemaand Gomez-Mejia:

An unresolved issue that remains to be explored is the extent to which the designof a CEO compensation package supports the implementation of a given strategyor instead, helps determine a firm’s strategic choices (1998, p. 139).

While Barkema and Gomez-Mejia do not focus explicitly on strategy implementation,they do provide a general framework for understanding executive compensation basedon criteria, governance, and contingencies.

Our research contributes to the literature by examining the importance ofexecutive compensation for firms implementing an innovation strategy. We chose toinvestigate innovation strategies since such strategies incorporate two constructsrelevant to compensation research: time horizon and risk. Time horizon, as used incompensation research, typically is defined as either short-term or long-term. Timehorizon is especially important to innovation strategy since innovation itself isgenerally considered a long-term commitment. There is a great deal of up-frontresearch and development (R&D) expenditure that must be undertaken beforereceiving any future benefit. In addition, innovation strategies incorporate greaterstrategic risk. As strategy risk increases, executives will attempt to reduce theirexposure to this risk (Harrison & March, 1984; Miller & Friesen, 1982) even thoughrisk-taking has been shown to have a positive effect on firm performance (Aaker &Jacobsen, 1987; Gilley, Walters & Olson, 2002).

In summary, this study is intended to extend the compensation and innovationliteratures in three ways. First, we attempt to understand the role compensation plays inenabling the implementation of an innovation strategy. Second, we base our moderatingarguments on the role of risk and time-horizon in combination with compensation andstrategy. Finally, we employ a panel data methodology (380 firms over a 5 year timeperiod) in order to benefit from both cross-sectional and time series data.

91Wheatley and Doty

Innovation Strategy and Executive Compensation

The major thrust of our argument is that the appropriate executive compensationpolicy will facilitate the implementation of an innovation strategy. Thus, we expectexecutive compensation to moderate the relationship between innovation strategy andfirm performance. To develop our argument, we begin by briefly discussing innovationand then exploring four elements of executive compensation as a function of timehorizon and risk.

Innovation StrategyOne way in which firms try to compete within (and buffer against) the competitive

landscape and environmental uncertainty is through the increased use of innovation,either for preemptive reasons or in response to internal or external environmentalchange (Damanpour, 1991; Hage, 1980; Thompson, 1967). A defining component ofan innovation strategy is the firm’s spending on R&D. The operationalization ofinnovation as R&D spending is well-suited for the purposes of this study for threereasons. First, R&D decisions are directly related to the implementation of aninnovation strategy. Second, R&D spending is under the direct control of the CEO andtop management team (TMT). Thus, executive compensation policies are likely tohave a greater effect on the firm’s R&D spending. Third, decisions about R&Dspending incorporate (either explicitly or implicitly) statements about risk preferencesand organizational time horizons. Each of these two constructs is used below tocharacterize important elements of executive compensation.

Executive CompensationMany of the important differences between the various forms of compensation can

be represented by two interdependent constructs: risk and time-horizon (Table 1). Wesuggest that risk is a crucial factor in the compensation-performance relationship.Risk reduction is dependent on the type of compensation provided. If executives arenot in fear of losing compensation based on performance, they may be more likely totake on the additional strategy risk. If their compensation is tied directly to firmperformance and a loss of compensation is possible, the need to reduce their riskwould be more likely, resulting in the desire to implement a less risky strategy.

Table 1: Compensation Time Horizon and Risk Relationship

Base compensation. Quadrant 1 (Table 1) shows slow risk and short-term and isdefined as basic cash compensation that an employer provides in exchange for workperformed. Because of this low compensation risk, executives would feel more atliberty to attempt implementation of a higher-risk strategy (i.e. base would be


considered over bonus because of the lower risk). Innovation strategy is defined ashigh risk/high return (Hansen & Hill, 1991; Hitt, Hoskisson & Ireland, 1990).Therefore, if executive compensation is not contingent upon implementation success,as in base compensation, the strategic leadership would enjoy more freedom toattempt to implement a higher risk strategy. Firm executives would be motivated toimplement an innovation strategy because of the transparent potential for payout.

As the proportion of base pay increases, the strategic leaderships’ comfort with risktaking would also increase (especially when compared to bonus). An alternateperspective on base compensation is that if no compensation risk were involved,executives would be less likely to implement a more risky strategy due to their desireto follow the status quo. However, as defined, innovation strategy is high risk/highreturn. Anticipation of high return may be one factor that drives risk-taking andenables the implementation of an innovation strategy.

Therefore we hypothesize,

Hypothesis 1: The percentage of base compensation moderates the relationship between innovation strategy and firm performance.

Bonus compensation. Quadrant 2 (Table 1) reflects high-risk and short-termbonus compensation and ties compensation to short-term success or performancemeasures. Bonus is considered high-risk because of the short-term nature and thecontingency on performance (especially when compared to base). Bonus pay is oftenpredicated on specific performance standards, thus the TMT is aware of what needs tobe accomplished in order to capitalize on the bonus pay component. The risk of notbeing granted a bonus is an important factor to consider. However, bonus has a short-term time frame which provides the TMT with less ambiguity and better forecastingtechniques. It is easier to forecast the result of a decision in the short-term versusconsidering the long-range implications of decisions as in the case of optionscompensation. Compared with strictly base compensation, bonus compensation hasgreater risk in implementing a high-risk innovation strategy.

A short-term, results-based bonus, especially if it constitutes a large portion of thecompensation package, will discourage executives from taking the long-term riskinvolved with innovation strategy because of the lack of predictable compensation.Implementing an innovation strategy is a long-term endeavor. A firm needs to make aconscious decision to pursue innovation and needs to provide ample resources. If afirm were to provide short-term compensation in the form of bonus, this would notsupport the long-term orientation of the innovation strategy. Thus, no relationshipwould be present to tie bonus and firm performance together. If executives arepresented with specific performance criteria for bonus compensation, they will mostlikely do whatever is necessary to gain that bonus, instead of focusing on the long-term implications. Another aspect of bonus compensation is the difference between abonus being available (which motivates future performance) and the actual awardingof a bonus which rewards prior performance. Stock and options are similar in that theyreward future performance with anticipation as the motivator and realization (ornonrealization) as the reward.

93Wheatley and Doty

Hypothesis 2: The percentage of bonus compensation moderates the relationship between innovation strategy and firm performance.

Options compensation. Options compensation which is low-risk and long-term(see Quadrant 3, Table 1), provides the most flexibility for executives. The individualexecutive has the most control over options, as individuals choose whether or not toexercise them. This compensation method provides the strategic leadership with theability to hedge against a negative outcome using their incentive compensation. In theevent that their projects/innovations are unsuccessful, the strategic leadership couldchoose not to exercise their options and instead wait until the firm moves into a morefavorable position. This flexibility promotes risk-taking by the TMT and mitigates theinherent risk of an innovation strategy.

To better understand options compensation, we contrast it with stock compensationbased on three key differences: 1) amount of control and flexibility, 2) downside risk,and 3) ability to buffer. Ultimately, stocks and options are the same piece of companyownership. However, the options alternative gives individuals the choice of whether ornot they want that piece of ownership at a specific point in time, with a specific priceand value. Options must be exercised to become shares of stock, with the decision oftiming being made somewhat by the individual. The second major difference isdownside risk. With stock compensation, downside risk is always present. If the firm'sstock begins to fall, the strategic leadership has no way to change their compensation.However, with options, if the stock begins to fall, the strategic leadership could choosenot to exercise their options and thus, endure no downside risk. Although the risk ofoptions is much lower, and the downside risk is minimal, there are some who wouldargue that options do carry with them an opportunity cost, which should be figured intodownside risk. Finally, because the environment is constantly changing, the use ofoptions provides executives with the opportunity to buffer against poor performanceand fluctuations in internal and external environments.

Options carry with them no downside risk essentially, whereas stock compensationdoes carry some of that risk. It is this lack of risk that promotes more risk-taking instrategy implementation. The risk literature provides support for the distinctionbetween stock and options compensation by suggesting that as contingentcompensation increases, managers’ risk-taking propensity decreases (Finkelstein &Hambrick, 1988; Zajac, 1992). Presumably, options are given in lieu of a higher levelof base compensation, with the thought that executives will be positively motivated tolook for long-run increases in the stock’s value. The lack of downside risk alignsoptions compensation with innovation strategy and should improve firm performance.From the dynamic perspective (as opposed to a static one), options do carry risk. Thisis especially apparent in today's economic environment where executives and directorshave lost substantial amounts of money because of the increased use of optionscompensation. As the firm's stock price falls below the options purchase price, thevalue of the compensation becomes worthless.

So with innovation strategy (high-risk), options will provide less compensationrisk than that of stock compensation. Therefore,


Hypothesis 3: The percentage of options compensation moderates the relationship between innovation strategy and firm performance.

Stock. The final quadrant, Quadrant 4 (Table 1), is stock compensation (high-riskand long-term). As compensation risk increases, so does the strategic leaderships’ riskaversion, making it less likely that they will attempt to implement a risky endeavorsuch as an innovation strategy (Beatty & Zajac, 1994; Gomez-Mejia, 1994; Gray &Cannella, 1997; Hill & Phan, 1991; Wiseman & Gomez-Mejia, 1988). Stockcompensation is considered pay for performance and the strategic leadership does nothave discretionary control over this type of compensation. Restricted and commonstock is awarded to executives without their making the decision to exercise (unlikeoptions compensation). Similar to bonus type compensation, a specified level ofperformance is defined, and if the strategic leadership meets or exceeds this target,they are rewarded (i.e. bonus is also high-risk on the short-term continuum). Becauseof this lack of exercise choice and long-term characteristic, stock carries the most riskfor executives. Stock compensation is used to align the interests of the TMT with theshareholders by providing rewards for increasing shareholder value (Jensen &Murphy, 1990). The TMT's fear of adversely affecting present shareholder value woulddeter the TMT from taking what they perceive to be high-risk actions. In the case ofhigh innovation strategy (high-risk), a low-risk compensation type would be preferred(i.e. base or options). Thus,

Hypothesis 4: The percentage of stock compensation moderates the relationship between innovation strategy and firm performance.

Methods

Sampling and Data CollectionPublicly traded firms were selected from ten industries that varied based on R&D

intensity. Only publicly traded firms were used because of the sensitive nature ofcompensation data. A two-stage process was employed during sample identification.First, compensation data were collected by industry from the Execucomp databasewhich contains data on companies in the Standard & Poor (S&P) 1500. Next, thesedata were matched to data from Compustat, removing companies with missing R&Ddata. We selected the final sample based on industries with the greatest number ofmatches and varying levels of R&D intensity (measured by R&D expenditure/numberof employees) (Hill & Snell, 1988; Scherer, 1984).

Compensation data covered a 5-year time span (1994-1998). Performance datawere lagged to cover 1995-1999 in order to better estimate the effect of compensationon future performance (Finkelstein & Boyd, 1998). Our final sample consisted of1900 observations and included data on 380 firms. All dollar values were adjusted forinflation and all data were archival. In addition, outliers were removed from thesample and normality was checked for each variable. Variables that were not normalwere transformed when possible by using the natural log.

95Wheatley and Doty

Independent and Moderator VariablesInnovation strategy. Innovation strategy was measured using R&D expenditure per

sales as an indicator of what is being accomplished from R&D money spent, controllingfor firm size. This strategy also provides a richer variable than using R&D expenditurealone (Hansen & Hill, 1991; Hay & Morris, 1979; Meyer-Krahmer & Reger, 1999;Scherer, 1984). This is an important indicator of an innovation strategy since the focusis on how companies transform R&D money into a successful outcome.

Compensation. Compensation data came from the S&P’s Execucomp database,which is compiled from SEC Filings requiring compensation information for the CEOand the 4 highest paid executives. Compensation was divided into base, bonus,options granted, and stock representing both short- and long-term compensation. Allcompensation was reported in dollars. The value of options granted was estimatedusing a Black-Scholes based (1973) option valuation model, which incorporates theexercise price of the option, the option term until exercise, an interest rate factor, avolatility factor, and dividend rate.

To calculate percent compensation, we summed each compensation type over allexecutives listed. A grand total of all compensation (base, bonus, stock, optionsgranted) for each TMT was then calculated for use in generating the percentagecompensation figure. These percentages were used for hypothesis testing, trying totease out the role each compensation type plays in enabling the implementation of aninnovation strategy.

Dependent and Control VariablesFinancial performance. Return On Assets, Return On Equity, and Earnings Per

Share data were collected from the Compustat database maintained by the S&P. Afterpreliminary analysis provided similar results for all three financial measures, weperformed a factor analysis to assess the number of factors present (Gomez-Mejia, Tosi& Hinkin, 1987; Tosi & Gomez-Mejia, 1994). This analysis suggested the presence ofonly one factor with all component loadings greater than 0.5. The loadings were asfollows: EPS (.781); ROA (.882); ROE (.862); Eigenvalues (2.132); Percent ofVariance=71.057. In order to create one aggregate measure, we multiplied thevariable’s z-score by the factor loading, then summed the three weighted scores tocreate the final variable called financial performance (Gomez-Mejia et al., 1987; Tosi& Gomez-Mejia, 1994).

Control variables. We controlled for industry using dummy variables based on a2-digit SIC code. Company and year were also controlled through dummy variablesfrom our use of the least squares dummy variable (LSDV) analysis, which categorizesdata into groups.

AnalysisWe employed a panel data methodology using LSDV because of our use of cross-

sectional (380 firms) as well as time series data (5-years). Two of the key problemswith panel data methodology are heteroscedasticity and auto-correlation (Hannan &Young, 1977). In this case, ordinary least squares (OLS) are ineffective in determiningthe regression estimates.


To interpret the direction of the moderating term, a graphing procedure was usedwhereby the independent variable (innovation strategy) was categorized as high orlow, as was the moderator variable (i.e. high percent base compensation and lowpercent base compensation) (Cohen & Cohen, 1983; Dwyer & Fox, 2000; Hitt et al.,2001; McFarlin & Sweeney, 1992; Welsh & Dehler, 1988). This information was thengraphed, resulting in two representative lines plotted against the independent variable(x-axis) and the dependent variable (y-axis). For example if the moderator of interestwas percent base compensation the resulting lines would be high percent basecompensation and low percent base compensation. The lines were then interpreted forthe direction of slope, as well as interception of the two lines.

Results

The correlations, means, and standard deviations of all the study variables arepresented in Table 2. Innovation strategy (measured by R&D/Sales) is positively andsignificantly correlated with percent base compensation and percent optionscompensation. Alternatively, innovation strategy is negatively and significantlycorrelated with percent bonus compensation and percent stock compensation.Financial performance is significantly correlated with all independent andmoderator variables.

Table 2: Descriptive Statistics and Zero-Order Correlation Coefficients

Table 3 presents the results of hypothesis testing. The results are presented inhierarchical fashion to better represent the effect of the interaction betweeninnovation strategy and compensation. Model 1 includes dummy variables forcompany, year, and industry (coefficients not shown), innovation strategy, andcompensation. Model 2 expands on Model 1 by adding the interaction betweeninnovation strategy and compensation.

97Wheatley and Doty

Hypothesis 1, which predicted a significant moderating effect of base compensationon the innovation strategy-performance relationship, was not supported. Thecoefficient for base compensation was not significant with financial performance as thedependent variable.

Table 3: Results of Generalized Least Squares Regression Analysis of Innovation Strategy and Base Compensation Effects on Firm Financial Performance

Hypothesis 2, which predicted a significant moderating effect of bonuscompensation on the innovation strategy-performance relationship, was supported.All of the eight models with the interaction term entered were significant. Thecoefficients for percent bonus compensation were both positive and significant withfinancial performance as the dependent variable (β=.37, p<.001; F=32.92, p<.001).

Hypothesis 3, which predicted a significant moderating effect of optionscompensation on the innovation strategy-performance relationship, was alsosupported. The coefficient for percent options granted compensation was bothnegative and significant with financial performance as the dependent variable (β=-.20,p<.001; F=20.45, p<.001).

Hypothesis 4, which predicted a significant moderating effect of stockcompensation on the innovation strategy-performance relationship, was notsupported for financial performance.

The models with significant interaction effects were further analyzed to correctlyinterpret the interaction effects. We followed Dwyer and Fox (2000) and graphicallyrepresented the moderating effect of compensation on innovation strategy andperformance. Figure 1 illustrates the bonus compensation interaction for financialperformance. The interaction graph for bonus compensation suggests that for bothlow and high innovation strategy (measured as R&D/Sales), the use of high bonuscompensation is most beneficial. We interpret the results in this manner because thelow and high base compensation lines do not intersect (nor are they parallel).


Figure 1: Interaction Between R&D/Sales and TMT Percent Bonus Compensation

The options compensation graph (Figure 2) has the most interesting interpretationbecause the high and low compensation lines intersect. This suggests that for highinnovation strategy (measured as R&D/Sales), the use of low-percent options grantedcompensation is most beneficial. Alternatively, for low innovation strategy, the use ofhigh-percent options granted appears to provide improved financial performance.

Figure 2: Interaction Between R&D/Sales and TMT Percent Options Granted Compensation

Discussion and Conclusion

In this paper, we investigated the relationship between innovation strategy andfirm performance, especially under various conditions of short- and long-termcompensation. Our findings provided a road map for companies that are pursuing aninnovation strategy and need to design the most beneficial compensation package for

99Wheatley and Doty

their top management team. Companies pursing an innovation strategy should designtheir compensation packages in such a way as to be heavy on bonus and light onoption type pay. For companies not focused on innovation, compensation packagesshould still be heavy on bonus type pay, but also heavy on option pay.

We drew on agency theory as well as the risk and time horizon relationship inorder to frame our ideas and explain this relationship. Analyses of data from 380 firmsover 5 years support some of our assertions. Results indicated that compensation doesmoderate the innovation strategy to the firm performance relationship whenconsidering bonus and options compensation. More specifically, we found that short-and long-term compensation have different driving mechanisms in organizationdecision-making when regarding strategy implementation.

We used a two-by-two matrix to model our arguments and show the distinctionbetween types of compensation. These arguments were also framed using risk to tryand understand what is driving managers’ decision-making. Our results suggest thatall strategies, whether they be low- or high-risk require short-term compensation.This provides additional support for the focus of compensation being placed on thetime component of compensation, as opposed to the risk component. Our findingsdefined this difference by showing that high-percent bonus compensation is relatedto greater performance levels, no matter the strategy risk involved. We believe thesefindings emphasize the pay-for-performance relationship (one that is especiallyprevalent in today’s organizations) and highlight the positive benefits of bonuscompensation. Bonus compensation has the added benefit of being a clearer, morepredictable form of compensation since bonus pay occurs in the short-term. It iseasier for managers to forecast and predict short-term effects of strategyimplementation than long-term effects.

Alternatively, long-term compensation and level of risk provide different findings.Our findings suggest that if low-risk strategies are being implemented, compensationcan be tied directly to performance in the form of long-term compensation withoutany reduction in firm performance. In contrast, when high-risk strategies are beingimplemented, long-term compensation must not be tied directly to performance inorder to foster better firm results. This result is an important finding and should beconsidered when determining compensation packages.

Contributions, Limitations, and Future Directions for ResearchWe made three significant contributions to the strategic management literature.

First, we tried to address and translate Barkema and Gomez-Mejia’s (1998) call forresearch into how compensation is related to strategy implementation. This paper isone of the first to treat compensation as a moderating factor and suggests thatcompensation enables the implementation of a specific strategy. Secondly, we extendedthe compensation literature by basing this moderating relationship not only oncompensation time-horizon, but the risk relationship as well. Finally, we utilized paneldata methodology, which maintains the richness of cross-sectional and time-series data.

In spite of the above contributions, there are some important limitations to thisresearch. One such limitation was the use of completely archival data. Although somewould argue that archival data are more accurate than informant data, archival also has

limited richness. The main limitation for this study arises when measuring innovationstrategy. We were looking to capture the broadest possible conceptualization ofinnovation strategy. However, using archival sources limited our measuring capability.Although we set out to cast a broad net, the R&D measure used is skewed towardproduct innovation.

Sample selection was also a problem. In the original design of the study, weattempted to sample from 6 industries (2 low R&D intensity, 2 medium R&Dintensity, 2 high R&D intensity) providing a "balanced" sample. In addition, wehoped to stratify the sample by size to focus on business level decisions, as opposedto corporate level ones. The available data did not allow for this split. Of the 380companies in the final sample, 348 fell in the greater-than $100,000,000 salescategory. Our final sample selection consisted of 10 industries. This change in thedesign was necessary due to limited compensation data. The final sample was alsosomewhat unbalanced. A single industry, Chemical and Allied Products (SIC 28),considered high R&D intensity had 86 companies. At the next level of R&D intensity,4 industries were represented with 202 companies. At the low R&D intensity end, 5industries were represented with 92 companies.

This study moved research a step closer to understanding the intricacies of strategyimplementation. Although this study did not open the “black box” of implementation,it did shed some light on mechanisms that enable implementation. Future studiesmight look to broaden the sample with additional industries and a more balanceddesign to enhance the generalizability. Investigating other strategies and the role of theenabling mechanism holds many possibilities as well.

There are also additional opportunities in considering other enabling mechanisms.For instance, options research is becoming much more popular and useful inexamining the incentive relationship. We merely scratched the surface looking atoptions granted as representative of long-term compensation in this study. A muchmore in-depth investigation of options may help to shed more light on this “special”compensation type, especially as ethical and legal issues surround this form ofcompensation. Options have many more components to consider such as type ofoptions granted, time period for vesting, and awards schedule, all of which may proveto be a driving factor for the interaction between strategy and compensation.

References

Aaker, D. & Jacobson, R. (1987). The role of risk in explaining differences inprofitability. Academy of Management Journal, 30: 277-296.

Allen, R.S. & Helms, M.M. (2002). Employee perceptions of the relationship betweenstrategy, rewards and organizational performance. Journal of Business Strategies, 19:115-139.

Barkema, H.G. & Gomez-Mejia, L.R. (1998). Managerial compensation and firmperformance: A general research framework. Academy of Management Journal,41(2): 135-145.

Beatty, R.P. & Zajac, E.J. (1994). Managerial incentives, monitoring, and risk bearing:A study of executive compensation ownership, and board structure in initial public


offerings. Administrative Science Quarterly, 39: 313-335.Black, F. & Scholes, M. (1973). The pricing of options and corporate liabilities. Journal

of Political Economy, (May/June): 637-659.Cohen, J. & Cohen, P. (1983). Applied multiple regression/correlation analysis for the

behavioral sciences (2nd ed.). Hillsdale, NJ: Erlbaum.Damanpour, F. (1991). Organizational innovation: A meta-analysis of effects of

determinants and moderators. Academy of Management Journal, 34: 555-590. Dwyer, D.J. & Fox, M.L. (2000). The moderating role of hostility in the relationship

between enriched jobs and health. Academy of Management Journal, 43: 1086-1096.Fama, E. & Jensen, M.C. (1983). Separation of ownership and control. Journal of Law

and Economics, 26: 375-393.Finkelstein, S. & Boyd, B.K. (1998). How much does the CEO matter? The role of

managerial discretion in the setting of CEO compensation. Academy of ManagementJournal, 41(2): 179-199.

Finkelstein, S. & Hambrick, D.C. (1988). Chief executive compensation: A synthesisand reconciliation. Strategic Management Journal, 9: 543-558.

Garen, J.E. (1994). Executive compensation and principal-agent theory. Journal ofPolitical Economy, 102: 1175-1199.

Gilley, K.M., Walters, B.A. & Olson, B.J. (2002). Top management team risk takingpropensities and firm performance: Direct and moderating effects. Journal ofBusiness Strategies, 19(2): 95-114.

Gomez-Mejia, L. (1992). Structure and process of diversification, compensationstrategy, and firm performance. Strategic Management Journal, 13: 381-398.

Gomez-Mejia, L. (1994). Executive compensation: A reassessment and a futureresearch agenda. In Ferris, G.R. (Ed.), Research in personnel and human resourcesmanagement, 12: 161-222. Greenwich, CT: JAI Press.

Gomez-Mejia, L.R., Tosi, H. & Hinkin, T. (1987). Managerial control, performance,and executive compensation. Academy of Management Journal, 30: 51-70.

Gray, S.R. & Cannella, A.A. (1997). The role of risk in executive compensation.Journal of Management, 23: 517-540.

Hage, J. (1980). Theories of organization. New York: Wiley Interscience.Hannan, M. & Young, A. (1977). Estimation in multi-wave panel models: Results on

pooling cross-sections and time series. In Heise, D. (Ed.), Sociological Methodology:52-83. San Francisco: Jossey-Bass.

Hansen, G.S. & Hill, C.W.L. (1991). Are institutional investors myopic? A time-series study of four technology-driven industries. Strategic Management Journal,12(1): 1-17.

Harrison, J.R. & March, J.G. (1984). Decision making and postdecision surprises.Administrative Science Quarterly, 29(1): 26-43.

Hay, D.A. & Morris, D.J. (1979). Industrial economics: Theory and evidence. Oxford:Oxford University Press.

Hill, C.W.L. & Phan, P. (1991). CEO tenure as a determinant of CEO pay. Academy ofManagement Journal, 34: 707-717.

Hill, C.W.L. & Snell, S.A. (1988). External control, corporate strategy, and firmperformance in research-intensive industries. Strategic Management Journal, 9: 577-590.

101Wheatley and Doty

Hitt, M.A., Bierman, L., Shimizu, K. & Kochhar, R. (2001). Direct and moderatingeffects of human capital on strategy and performance in professional service firms:A resources-based perspective. Academy of Management Journal, 44: 13-28.

Hitt, M.A., Hoskisson, R.E. & Ireland, R.D. (1990). Mergers and acquisitions andmanagerial commitment to innovation in M-form firms. Strategic ManagementJournal, 11(SI): 29-48.

Jensen, M.C. & Meckling, W. (1976). Theory of the firm: Managerial behavior, agencycosts, and ownership structure. Journal of Financial Economics, 3: 305-360.

Jensen, M.C. & Murphy, K.J. (1990). Performance pay and top managementincentives. Journal of Political Economy, 98: 225-264.

Lerner, J. & Wulf, J. (2007). Innovation and incentives: Evidence from corporateR&D. The Review of Economics and Statistics, 89: 634-644.

McFarlin, D.B. & Sweeney, P.D. (1992). Distributive and procedural justice aspredictors of satisfaction with personal and organizational outcomes. Academy ofManagement Journal, 35: 626-638.

Meyer-Krahmer, F. & Reger, G. (1999). New perspectives on the innovation strategiesof multinational enterprises: Lessons for technology policy in Europe. ResearchPolicy, 28(7): 751-776.

Miller, D. & Friesen, P.H. (1982). Innovation in conservative and entrepreneurialfirms: Two models of strategic momentum. Strategic Management Journal, 3: 1-25.

Scherer, F.M. (1984). Innovation and growth: Schumpeterian perspectives. Cambridge,MA: MIT Press.

Shaw, J.D., Gupta, N. & Delery, J.E. (2001). Congruence between technology andcompensation systems: Implications for strategy implementation. StrategicManagement Journal, 22: 379-386.

Slater, S.F. & Olson, E.M. (2001). Marketing’s contribution to the implementation ofbusiness strategy: An empirical analysis. Strategic Management Journal, 22: 1055-1067.

Thompson, J.D. (1967). Organizations in action. New York, NY: McGraw Hill.Tosi, H.L. & Gomez-Mejia, L.R. (1994). CEO compensation monitoring and firm

performance. Academy of Management Journal, 37: 1001-1017Welsh, M.A. & Dehler, G.E. (1988). Political legacy of administrative succession.

Academy of Management Review, 31: 948-961.Wiseman, R.M. & Gomez-Mejia, L.R. (1998). A behavioral agency model of

managerial risk taking. Academy of Management Review, 25: 133-152.Xue, Y. (2007). Make or buy new technology: The role of CEO compensation contract

in a firm’s route to innovation. Review of Accounting Studies, 12: 659-690.Yanadori, Y. & Marler, J.H. (2006). Compensation strategy: Does business strategy

influence compensation in high-technology firms? Strategic Management Journal,27: 559-570.

Zajac, E.J. (1992). CEO preferences for incentive compensation: An empirical analysis.Academy of Management Proceedings: 47-51.


103Wheatley and Doty

Invitation To Review Manuscripts

The review process is a critical step in publishing a quality journal. The editors ofthe Journal of Business and Management invite you to participate in the ongoingactivities necessary to make JBM a reputable scholarly outlet. If you would like us tosend you manuscripts to review, please complete the form below or email us with theinformation at [email protected].

Name ____________________________________________________________________

Address __________________________________________________________________

__________________________________________________________________________

__________________________________________________________________________

Email ____________________________________________________________________

Please list your major areas of interest:

Please indicate how many manuscripts you would be willing to review in anacademic year:

■ 1 ■ 2 ■ 3 ■ 4 ■ 5

Please return the form to:Amy E. Hurley-Hanson, Ph.D.

Cristina M. Giannantonio, Ph.D.Editors, Journal of Business and ManagementArgyros School of Business and Economics

Chapman UniversityOne University DriveOrange, CA 92866

FAX (714) 532-6081

J.B.M.

Subscription Form

The Journal of Business and Management is published by the Argryros School of

Business and Economics, Chapman University. It is sponsored by the Western

Decision Sciences Institute.

The annual subscription fee is $50.00 for individuals and $100.00 for institutions.

Your check or money order should be made out to CHAPMAN UNIVERSITY/JBM.

Please complete your mailing information below.

Name ________________________________________________________________________

Department ____________________________________________________________________

Address Line 1__________________________________________________________________

Address Line 2__________________________________________________________________

Address Line 3__________________________________________________________________

Tel ________________________________________________________________________

Fax________________________________________________________________________

Mail this form with your check or money order to:

Journal of Business and Management

Argyros School of Business and Economics

Chapman University

One University Drive

Orange, CA 92866

email: [email protected]

J.B.M.

ISSN: 1535-668XEditorial Offices: Journal of Business and Management

Argyros School of Business and EconomicsChapman UniversityOne University DriveOrange, CA 92866

Fax: (714) 532-6081

E-mail: [email protected]

Submission information is available on our website: http://jbm.chapman.edu

The Journal of Business and Management is published by Chapman University’s Argyros School of Business and Economicswith the sponsorship of the Western Decision Sciences Institute. The primary editorial objective of the Journal of Businessand Management is to provide a forum for the dissemination of theory and research in all areas of business, management,and organizational decisions which would be of interest to academics and practitioners.

The views expressed in published articles are those of the authors and not necessarily those of the editors, editorial board,WDSI or Chapman University.

Copyright @ 2010 Argyros School of Business and Economics, Chapman University.

jbm-vol-16-01

Documents

Transcript of jbm-vol-16-01