Essays on Entrepreneurship and Economic Development, Exercises of Entrepreneurship

Download Essays on Entrepreneurship and Economic Development and more Exercises Entrepreneurship in PDF only on Docsity! Essays on Entrepreneurship and Economic Development Dissertation zur Erlangung des wirtschaftswissenschaftlichen Doktorgrades der Wirtschaftswissenschaftlichen Fakultät der Universität Göttingen vorgelegt am 7. 9. 2007 von Jagannadha Pawan Tamvada aus New Delhi Eidesstattliche Erklärung Hiermit versichere ich an Eides statt, dass ich die eingereichte Dissertation Essays on Entrepreneurship and Economic Development selbständig verfasst habe. Anderer als der von mir angegebenen Hilfsmittel und Schriften habe ich mich nicht bedient. Alle wörtlich oder sinngemäß den Schriften anderer Autoren entnommenen Stellen habe ich kenntlich gemacht. Göttingen, den 7. September 2007, Jagannadha Pawan Tamvada i Publications The paper Religion and Entrepreneurship, co-authored with David Audretsch and Werner Boente, is published as a Center for Economic Policy Research (CEPR) Discussion Paper. The other papers in the dissertation are authored by me and have been presented at international conferences, doctoral colloquiums and fac- ulty seminars. The research work in this dissertation has been accepted for presentation at the First World Congress of Spatial Econometrics (Cambridge, 2007), the 44th European Regional Science Association’s Annual Congress (Paris, 2007), the International Council for Small Business Research (Finland, 2007), the IZA- World Bank Conference on Employment and Development (Bonn, 2007) and the Second Annual Max Planck Indian Institute of Science (IISc) International Conference on Entrepreneurship, Innovation and Economic Growth (Bangalore, 2007). The research in this dissertation has been presented at the Schumpeter Con- ference (Nice, 2006), the 20th Research in Entrepreneurship Conference (Brues- sels, 2006), the First Annual Max Planck India Workshop on Entrepreneurship, Innovation, and Economic Growth (Bangalore, 2006), Hellenic Workshop on En- trepreneurship and Productivity (Patras, 2006), the European Summer School in Industrial Dynamics (Corsica, 2006), the Babson Doctoral Consortium (Bloom- ington, 2006), Augustin Cournot Doctoral Days (Strasbourg, 2006), the Technol- ogy Transfer Society’s Annual Conference (Kansas City, 2005) and the G-Forum’s Annual Conference (Jena, 2006). The work has also been presented at internal seminars at the Max Planck Institute of Economics, Jena and at the Faculty of Economics, University of Göttingen. iv Acknowledgements I am greatly indebted to Prof. Stephan Klasen for giving me an opportunity to pursue doctoral studies in economics, a dream I had cherished since my high school days. Without his intellectual guidance, this dissertation would not have seen its completion. I am also grateful to Prof. David Audretsch, whose inspi- ration, support, and kindness have helped me to complete the task at hand. I owe enormous gratitude to him not just for giving me the prestigious Max Plank PhD scholarship, but also for introducing me to entrepreneurship and constantly guiding me through the scholarship. I also show my gratitude to Prof. Walter Zucchini for his guidance and encouragement. My discussions with my supervi- sors form the foundations of this dissertation, and I would like to express my gratitude for giving me their precious time and intellectual support. I am grateful to Prof. Amartya Sen for his inspiration not just for me, but for many young Indians to study economics. I thank him for sparing some of his valuable time for me when I visited him at Harvard. I also thank Professors TVS Ram Mohan Rao and Vishwanath Pandit for their constant guidance and support. I also thank Swami Supernanada, a monk of the Rama Krishna Mission order, who introduced me to this fascinating subject. I thank all my colleagues at the Max Planck Institute, especially Werner Boente and Max Keilbach. I would also like to thank Taylor Aldridge, Melanie Aldridge, Iris Beckman, Saradindu Bhaduri, Andreas Chai, Andrea Conte, Sameeksha Desai, Stephan Heblich, Anja Klaukien, Stefan Krabel, Adam Led- erer, Prashanth Mahagaonkar, Erik Monsen, Pamela Mueller, Holger Patzelt, Stephan Schütze, Jörg Zimmerman, and all my colleagues for making my stay in Jena both intellectually and personally rewarding. My special thanks go to Kerstin Schük, Madeleine Schmidt, and Lydia Nobis for their continuous efforts at making my stay at the Institute memorable. I also show my gratitude towards v Thomas Bauman and his IT team and Katja Müller and her library team at the Max Plank Institute for attending to all my academic requirements. I also thank Marten Koppenhagen, Thilo Klein, and particularly Alex Audretsch for providing valuable research assistance. Thus, I thank the Max Planck Institute of Economics and all its wonderful staff for providing an environment that is so conducive to research. I also would like to thank my colleagues at the Chair for Development Eco- nomics at Göttingen, particularly Micheal Grimm, Melanie Grosse, Isabal Gün- ther, Andrey Launov, Felicitas Nowak-Lehman Danzinger, Ken Harttgen, Mark Misselhorn, Jan Priebe, Dana Schüler, Sebastian Vollmer and Julian Wiesbrod. My special thanks also go to Michaela Beckmann. Roswitha Brinkmann and the international office of the Goettingen University have my great appreciation for giving me a scholarship during my stay there in 2004, as does Prof. Manfred Denker for accepting me as a member of the Center for Statistics in Göttingen. I give my sincerest thanks to the Lindau Nobel Council for selecting me to participate at the Second Lindau meeting of Nobel laureates, giving me an un- paralleled opportunity to meet and listen to some of the greatest living legends in the field of economics. I also extend my thanks to the Max Planck Society for giving me a grant to organize the First Annual Max Planck India Workshop on Entrepreneurship, Innovation, and Economic Growth in partnership with the Indian Institute of Science, Bangalore. This conference reaffirmed to me that fur- ther entrepreneurship research focusing on developing countries such as India is an absolute necessity to promoting their development. Finally, I would like to thank the Kauffman Foundation for sponsoring my participation at the Babson Doctoral Consortium. The Ministry of Small Scale Industries provided much assistance to me by providing me firm-level data, as did the Reserve Bank of India by inviting me for a research stay at the Rural Credit and Policy Department. I am greatly indebted to my parents for their love, affection, prayers, and blessings. In storm and in calm, they have stood by me. I am equally grateful to Sai Baba, who showed me that man is born not for pursuing self-interest, but to serve humanity. He stayed with me through-and-through. Like my parents, he always showered unconditional love and affection on me and inspired me to follow my heart and pursue this path. As his student in Prashanti Nilayam, I vi 3 Religion and Entrepreneurship 42 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.2 Religion, Entrepreneurship and the Indian Context . . . . . . . . 44 3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.4 Empirical Analysis: Discrete Choice Models . . . . . . . . . . . . 49 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4 The Geography of Start-up Size 65 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.2 The Start-Up Size . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.3 Geoadditive Models . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.5 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 5 Entrepreneurship and Welfare 87 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 5.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . 88 5.2.1 Occupation, Welfare and Economic Development . . . . . 88 5.2.2 Occupational Selection and Determinants of Welfare . . . . 90 5.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 5.3.1 Quantile Regressions . . . . . . . . . . . . . . . . . . . . . 93 5.3.2 Selection Models for Multiple Outcomes . . . . . . . . . . 93 5.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 5.5 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 96 5.5.1 Entrepreneurship and Welfare . . . . . . . . . . . . . . . . 96 5.5.2 Endogenous Non-random Occupational Selection . . . . . . 102 5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 6 The Dynamics of Entrepreneurship 122 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 6.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 6.2.1 Repeated Cross Section Analysis . . . . . . . . . . . . . . 123 6.2.2 Pseudo Panel Approach . . . . . . . . . . . . . . . . . . . 124 6.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 ix 6.4 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 127 6.4.1 Repeated Cross Sections . . . . . . . . . . . . . . . . . . . 127 6.4.2 Pseudo Panel Analysis . . . . . . . . . . . . . . . . . . . . 129 6.4.3 Reconciling the Results . . . . . . . . . . . . . . . . . . . . 133 6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 7 Conclusion 151 7.1 Exogenous Constraints and Entrepreneurship . . . . . . . . . . . . 151 7.2 The Dual Theory of Entrepreneurship . . . . . . . . . . . . . . . . 152 7.2.1 Entrepreneurship, Start-Up Size, and the Spatial Location 154 7.2.2 A Simple Model . . . . . . . . . . . . . . . . . . . . . . . . 155 7.2.3 Entrepreneurship and Economic Development: The Dual Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 7.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 Bibliography 160 x List of Figures 2.1 Non-linear Effects of Age on Self-employment . . . . . . . . . . . 31 2.2 Spatial Effects on Self-employment Choice . . . . . . . . . . . . . 35 2.3 Spatial Effects in ‘Nonagriculture’ . . . . . . . . . . . . . . . . . . 36 2.4 Spatial Effects in ‘Agriculture’ . . . . . . . . . . . . . . . . . . . . 37 3.1 Entrepreneurship and Religion . . . . . . . . . . . . . . . . . . . . 54 3.2 Entrepreneurship and Caste System in Hinduism . . . . . . . . . 54 4.1 Spatial Effects in Model I . . . . . . . . . . . . . . . . . . . . . . 81 4.2 Spatial Effects in Model II . . . . . . . . . . . . . . . . . . . . . . 82 5.1 Consumption and Occupation(Un-normalised) . . . . . . . . . . . 108 5.2 Quantile Plots-Households . . . . . . . . . . . . . . . . . . . . . . 111 5.3 Quantile Plots-Households (continued) . . . . . . . . . . . . . . . 112 5.4 Occupation and Poverty Plots . . . . . . . . . . . . . . . . . . . . 114 5.5 Occupation and Inequality Plots at Median . . . . . . . . . . . . . 114 6.1 Non-linear Effect of Age on Self-employment (2000) . . . . . . . . 137 6.2 Non-linear Effect of Age on Self-employment (2004) . . . . . . . . 138 6.3 Spatial Effects on Self-employment Choice . . . . . . . . . . . . . 142 6.4 Spatial Effects in ‘Nonagriculture’ . . . . . . . . . . . . . . . . . . 143 6.5 Spatial Effects in ‘Agriculture’ . . . . . . . . . . . . . . . . . . . . 144 7.1 Entrepreneurship and Economic Development . . . . . . . . . . . 158 xi Chapter 1 Introduction Almost four decades ago, Baumol (1968, p. 71) proclaimed that “in a growth con- scious world I remain convinced that encouragement of the entrepreneur is the key to the stimulation of growth.” Entrepreneurship, however, remained hidden and elusive from the grasp of economists. Fortunately, in recent years, the economics of entrepreneurship emerged as a compelling subject, providing insights into the entrepreneurial processes. Bringing together this literature on entrepreneurship, Parker (2004, p. 1) notes that “entrepreneurship has only recently come to be regarded as a subject.” While the debate in scholarly community has still not conclusively accepted even the definition of entrepreneurship, a vast literature has emerged over the last two decades providing insights into the many facets of entrepreneurship. Though each such facet is incomplete by itself, together they offer a comprehensive understanding of the entrepreneurial choice, new firm for- mation and the role of entrepreneurship in economic growth. Reflecting a broad consensus that has emerged in recent times, Lazear (2002, p. 1) claims that “the entrepreneur is the single most important player in the economy.” This dynam- ically expanding subject, the economics of entrepreneurship, however, neglected entrepreneurship in less developed countries. This dissertation exploits recent ad- vances in Bayesian semiparametric methods and geoadditive models (Fahrmeir and Lang, 2001a) and large databases of individual and firm-level micro-data from India to provide fresh perspectives of the entrepreneurial processes and their relationship to economic development. This dissertation underlines the nexus be- tween the entrepreneur, the firm, and the region by emphasizing the role of the spatial location in simultaneously determining the entrepreneurship choice and 1 the size of new firms. The returns to occupational choice and the spatio-temporal dynamics of self-employment choice form another major part of this dissertation. The role of the caste system and religion in determining the entrepreneurship choice is studied, as such factors play a crucial and important role in determining the occupational choice in India. The theme of the second and third chapters is the determinants of self em- ployment and the role of exogenous constraints in occupational choice. While a vast literature has emerged examining the determinants of entrepreneurship, the role of spatial location and the neighborhood of an individual have rarely been considered as determinants of entrepreneurship choice. There are compelling rea- sons, however, to assume that such factors play an important role in shaping the occupational choice of people. Thus, in chapter 2, I analyze the role of geographic location as a micro-determinant of self-employment choice. I also study the im- pact of human capital accumulation on occupational choice in agricultural and nonagricultural sectors in India. In chapter 3, I analyze the role of religion as an exogenous constraint on the occupational choice of individuals. Recent studies (Iannaccone, 1998; McCleary and Barro, 2006a; Guisa et al., 2006) link religion with economy but the channels through which religion influences the economy are not examined by the existing literature. One such channel through which reli- gion might influence the economy is through entrepreneurship. Religions impose behavioral constraints and influence economic outcomes. For instance, the insti- tution of the caste system in Hindusim is likely to act as an exogenous constraint on the occupational choice of Hindus. In this paper, I examine the role of religion and class structures in promoting or inhibiting entrepreneurial behavior. The theme of the fourth chapter is the impact of ownership structure and geo- graphic location on the size of new entrants. In this chapter, I revisit the question of firm size at entry. A number of studies show that, for new entrants at least, the initial size influences growth and survival. The determinants of the size of firms at entry, however, remained under-researched and neglected in this discussion, for a long time. The few studies on start-up size show that the industry characteris- tics such as turbulence, minimum efficient scale, and industry growth (Mata and Machado, 1996; Mata, 1996) and human capital of entrepreneurs (Astebro and Bernhardt, 2005; Colombo et al., 2004; Colombo and Grilli, 2005), determine the start-up size of new firms. However, the role of spatial location on the start-up 2 size has never been studied. Chapter 4 incorporates ownership structure and ge- ographic location as micro-determinants of start-up size, using micro data from India. The theme of the fifth chapter is entrepreneurship and welfare. A growing body of literature identifies returns to self-employment in developed countries (Hamilton, 2000). Historically, the development economics literature has consid- ered self-employment in less developed countries, to be a part of the so-called informal sector (Harris and Todaro, 1970). More recently, a growing body of literature argues that the informal sector is a blend of a low-productive disad- vantaged sector and a voluntary competitive sector (Cunningham and Maloney, 2001; Fields, 2005; Günther and Launov, 2006). In chapter 5, I link occupational decisions of the household with a direct measure of welfare, per-capita consump- tion. Using quantile regressions, I estimate occupational choice as a determinant of welfare. Furthermore, using selection methods that allow for corrections af- ter multinomial logit estimation (Bourguignon et al., 2007), I test if a process of endogenous non-random selection determines the selection of individuals into different occupations. Thus, the underlying process of selection into occupations and subsequent returns in terms of welfare are examined to see whether peo- ple are compelled to opt for low-productivity self-employment or whether they voluntarily self-select based on their unobserved abilities, in a developing country. The theme of the sixth chapter is the evolution of the entrepreneurial choice over time and space. The post-liberalisation era of Indian economy has witnessed a surge in entrepreneurial activity. The dynamics of occupational choice in this context are not explored in the literature. Using two cross-sectional databases of the National Sample Survey Organization of India (NSSO) data, I examine the spatial dynamics of self-employment choice and in particular, the role of educa- tion as a determinant of entrepreneurship. In addition, using three surveys of the NSSO (1994-1995, 1999-2000 and 2004), I also construct a psuedo-panel (Deaton, 1997; Moffitt, 1993; Verbeek, 2006) to examine the dynamics of entrepreneurial activity in India. The final chapter constructs the dual theory of entrepreneur- ship, linking results of the chapters of this dissertation. This chapter showcases a coherent theory of self-employment, firm formation, and geographic location and concludes this dissertation. 3 minants of self-employment choice in one such growing economy, India, that has in recent years, experienced substantial leaps in both its entrepreneurial activity and growth rates. Household level data collected by the National Sample Survey Organization in 2004 are used for the empirical analysis. The effects of individual personal charac- teristics, educational background, household characteristics and non-linear effects of continuous covariates such as age and geographic location on the probability of being self-employed are jointly estimated using geoadditive models. The results suggest that outside of agriculture, educated individuals are more likely to be salaried employees while in the agricultural sector, educated individuals are more likely to be self-employed. Strong spatial patterns are observed and these are pri- marily attributable to spatial self-employment patterns in the agricultural sector. Consistent with earlier empirical studies on the determinants of entrepreneurship, the results suggest that Indian males, married and older citizens are more likely to be self-employed as well. The next section discusses the literature and states the hypotheses on the determinants of self-employment in a developing economy. The third and fourth sections describe the semiparametric geoadditive modeling techniques and the dataset. The fifth section presents the empirical analysis. The final section pro- vides conclusions and discusses possible avenues for future research. 2.2 Theoretical Background 2.2.1 Determinants of Self-employment Empirical research on occupational choice in developed economies suggests that individuals’ personal characteristics (Kihlstrom and Laffont, 1979; Evans and Leighton, 1989b) and regional factors (Georgellis and Wall, 2000) play an impor- tant role in influencing the entrepreneurial decisions. The decision of individuals to become entrepreneurs is generally modeled in terms of utility maximization, where the economic returns from entrepreneurship are compared to returns of wage employment (Lucas, 1978; Holmes and Schmitz Jr., 1990; Jovanovic, 1994). Individual-specific characteristics such as risk aversion (Kihlstrom and Laf- font, 1979), prior self-employment experience (Evans and Leighton, 1989b), edu- cation, human capital, and age (Zucker et al., 1998; Bates, 1990; Rees and Shaw, 6 1986; Blanchflower and Meyer, 1994) and personality characteristics (McCelland, 1964), are found to have an impact on an individual’s entrepreneurship choice. As Parker (2004, p. 106) succinctly summarizes the broadly agreed determinants of entrepreneurship, The clearest influences on measures of entrepreneurship (usually the likelihood or extent of self-employment) are age, labor market experi- ence, marital status, having a self-employed parent and average rates of income tax (all with positive effects). Greater levels of risk and higher interest rates generally have negative effects, although to date only a handful of studies have satisfactorily investigated the former. Region specific characteristics such as industry structure (Acs and Audretsch, 1989; White, 1982), unemployment rates (Blanchflower, 2000; Blanchflower and Oswald, 1998), local job layoffs (Storey and Jones, 1987), small business employ- ment (Reynolds et al., 1994) and public policy variables such as state retirement benefits (Blau, 1987), unemployment benefits (Carrasco, 1999), and adherence to welfare state principles (Fölster, 2002) are also found to influence occupational choice.4 2.2.2 Labor Markets in Developing Countries The disadvantage theory and the comparative advantage theory are two compet- ing theories of labor markets in developing countries. The disadvantage theory hypothesizes that people who are rationed out of the formal labor markets are compelled to take up self-employment or work as workers in household enterprises. Such people are considered to constitute the informal sector. Thus, beginning with the labor surplus model of Lewis (1954), the labor markets of developing countries are viewed as segmented dualistic markets along the formal-informal lines (also see Sen, 1966; Ranis and Fei, 1961; Harris and Todaro, 1970).5 4Other examples of studies analyzing the determinants of entrepreneurship include Evans and Jovanovic (1990) and Parker et al. (2005). 5 Lewis (1954) argued that if wage rate is determined competitively in the rural areas of a LDC then it will be below the subsistence levels. Harris and Todaro (1970) predicts that workers who migrate from rural to urban areas face unemployment and are forced to work in household enterprises at subsistence levels. Models of rural-urban migration following this line of thought hypothesize that the urban informal sector acts as a refuge for migrants and excess labor in urban areas are forced to take up low productivity self employment. 7 Many studies find evidence against these theories of low level subsisting self- employment in LDCs (Chiswick, 1976; Majumdar, 1981; Blau, 1986; Rosenzweig, 1980; Mohapatra et al., 2007).6 The comparative advantage theory, thus hypothe- sizes that individuals voluntarily choose employment in the so called informal sec- tor, when they perceive competitive opportunities there (Gindling, 1991; Magnac, 1991; Maloney, 2004).7 In this paper, we do not distinguish between the formal and the informal sectors for two reasons. First, Maloney (2004, p.1159) notes that, “as a first ap- proximation we should think of the informal sector as the unregulated, devel- oping country analogue of the voluntary entrepreneurial small firm sector found in advanced countries, rather than a residual comprised of disadvantaged work- ers rationed out of good jobs.” As most empirical research on the determinants of self-employment is based on data from the developed economies, the results will stand comparable to the results of earlier studies if we consider both the sectors together and treat the informal sector akin to the entrepreneurial small firm sector of the developed countries. Second, the other main purpose of the paper is to examine the determinants of self-employment choice in agriculture and nonagriculture in India through the lens of economic geography. Though the characteristics of the informal sector in a developing country are well debated in the literature, examining the determinants of self-employment in this light is an interesting avenue for future research. 2.2.3 Hypotheses: Determinants of Self-employment Though there are compelling reasons to posit that there are sectoral differences in self-employment choice, male, married and older individuals are more likely 6Blau (1985) positively tests for competitive labor markets in the nonagriculture sector in LDCs but finds negative selection into self-employment based on managerial ability in the farm sector. His results suggest that self-employed earn more than wage employees in urban areas whereas in rural areas the self-employed earn much less than the wage employees. 7More recently, a growing body of literature attempts to capture the heterogeneity within the informal sector. This strand of literature argues that the informal sector is a blend of both disadvantaged and competitive sectors (Cunningham and Maloney, 2001; Fields, 2005; Günther and Launov, 2006) and claims simultaneous presence of disadvantaged “lower” and voluntary “upper” tiers within the informal sector. Pratap and Quintin (2006) do not find any evidence for segmented labor markets in Argentina. Yamada (1996) finds evidence of voluntary self-selection and higher earnings in self-employment in informal sector in Peru. 8 This suggests that returns to salaried employment increase faster than returns to entrepreneurship as the per-capita income grows, and this makes individuals more risk averse and decreases their willingness to become entrepreneurs (also see Lucas, 1978). Thus, there are compelling reasons to posit that individuals who are more educated opt for salaried employment relative to self-employment in an LDC context (see Sluis et al., 2005, for a survey). Hence, we hypothesize that individuals with greater human capital might prefer salaried employment as opposed to self-employment. Another determinant of self-employment that is discussed in the literature is wealth. Wealth possessed by the individuals provides a degree of security for entering self-employment and helps them to ease their credit constraints.10 As Boháček (2006, p.2196) notes, In order not to default on loan contracts, entrepreneurs can borrow only limited amounts secured by collateral. This collateral (accumu- lated assets) guarantees not only the repayment of the loan but also positive consumption of the entrepreneur in the case of a project’s failure. As the financial constraint is endogenously related to a bor- rower’s wealth, entrepreneurship becomes positively correlated with wealth. Households with very high levels of wealth have a higher propensity to take risk (Carroll, 2000). Hurst and Lusardi (2004) argue that as households with higher levels of wealth have a higher tolerance for risk, they are most likely to be busi- ness owners.11 Blanchflower and Oswald (1998) find that inheritance increases the probability of self-employment. Banerjee and Neuman (1993) argue that wealth distribution determines the occupational structure. For these reasons, we hypoth- esize a positive relationship between household wealth and the entrepreneurship choice. Borjas and Bronars (1989) present differences in self-employment rates amongst racial minorities in US. They show that consumer discrimination af- 10Lindh and Ohlsson (1996) test if the presence of credit constraints inhibit people from becoming self-employed. Many other studies also find that credit constraints act as barriers to entry of individuals into self-employment (Evans and Jovanovic, 1989; Evans and Leighton, 1989b; Blanchflower and Oswald, 1998). 11However, Hurst and Lusardi (2004) find that the relationship between wealth and en- trepreneurship is flat over the majority of the wealth distribution. They discover a positive relationship only after the ninety-fifth percentile. They argue that the reason could be that capital needed for a start-up in the United States is relatively low (also see Bhidé, 2000). 11 fects the earnings of self-employed blacks and other minority communities, mak- ing them less likely to select into self-employment relative to whites. Some other studies find that self-employment is higher in minority communities (Clark and Drinkwater, 1998). In an Indian context, the presence of caste system leads us to hypothesize that individuals of the backward classes may have a lesser propensity to be self-employed. Based on insights from the theory of new economic geography (Krugman, 1991; Fujita and Krugman, 2003), we hypothesize that individuals in neighbor- ing regions exhibit similar occupational preferences and in some neighborhoods individuals are more likely to be self-employed than in others and that this effect is non-linear in shaping economic outcomes over space. The presence of many self-employed people in a wealthy neighborhood may induce others to choose self-employment. Thus, it may have an inducement effect on the local popula- tion. People in such regions are likely to be more entrepreneurial and risk loving. However, presence of many self-employed people in poor neighborhoods indicates that dearth of viable employment opportunities compells people to select into self-employment in such neighborhoods. 2.3 Bayesian Semiparametric Methodology Semiparametric regression technique based on Bayesian P-Splines and geoaddi- tive models is used for the empirical analysis. The methodology allows for the estimation of non-linear effects of the continuous variables and the neighborhood effects of spatial units on the probability of individuals selecting self-employment. A brief outline of the method is presented here.12 2.3.1 Geoadditive Models Let (yi, xi, vi) for i in {1,2,...N} describe a dataset of N observations. Let yi be the response variable and xi be a m-dimensional vector of continuous covariates and 12This section draws on Lang and Brezger (2004) and Brezger and Lang (2005). This method- ology has been applied earlier by Kandala et al. (2001) and Kandala et al. (2002) to examine the determinants of under-nutrition in African countries. 12 vi be a vector of categorical variables.13 Assume yi are independent and Gaussian with mean ηi = f1(xi1) + .... + fp(xip) + viγ, and a common variance σ2. If fi are unknown smooth functions of the continuous variables and viγ corresponds to the parametric part of the regression, the regression model is called the Additive Model or a Semiparametric regressor. Eilers and Marx (1996) use polynomial regression splines that are parameterized in terms of B-Spline basis functions, the P-Splines, in the context of an Additive Model, to estimate the smooth functions within the semiparametric framework. Fahrmeir and Lang (2001a,b) use simple random walk priors in a bayesian version of the Additive Model. Kammann and Wand (2003) introduce Geoadditive models within the Additive Mixed Model framework to deal with unobserved heterogeneity across different spatial units.14 Furthermore, Lang and Brezger (2004) and Brezger and Lang (2005) generalize the work of Fahrmeir and Lang (2001a,b) and develop the Bayesian version of the P-Spline approach of Eilers and Marx (1996), Bayesian P-Splines.15We use these methods in the empirical analysis. Assume that the unknown functions fj can be approximated by a l degree spline with equally positioned knots in the domain of xj (Eilers and Marx, 1996). By writing such a spline in the form of a linear combination of k B-Spline basis functions, Bjk, where k is equal to the number of knots plus the degree of the spline, fj(xj) = ΣβjkBjk and, in matrix notation, η = ΣXjβj + V γ. By defining a roughness penalty based on the differences of adjacent B-Spline coefficients, for ensuring smoothness of the estimated functions, the penalized likelihood assumes the form: L = l(y, β1, ....., βp, γ)− λ1Σ(4kβ1) 2 − .......λpΣ(4kβp) 2 (2.1) 13We first present the case of the gaussian response distribution and then show how the family of binomial probit models can be generalized to the family of gaussian response, using a link function. 14Generalized Additive Mixed Models (Lin and Zhang, 1999) for cases with unobserved het- erogeneity are extensions of Generalized Additive Models (Hastie and Tibshirani, 1990). For an overview of semiparametric regressions, see Fahrmeir and Tutz (2001). Additive Mixed Mod- els in the Bayesian framework have also been considered by Hastie and Tibshirani (2000) and Fahrmeir and Lang (2001a,b) but these approaches do not consider the unobserved heterogene- ity, the spatially correlated random effects. 15The difference penalties are replaced by Gaussian (intrinsic) random walk priors that serve as smoothness priors for the unknown regression coefficients. A related approach is the Bayesian smoothing splines methodology of Hastie and Tibshirani (2000). 13 tial patterns can be explained using one of the following econometric approaches. A simple strategy is to regress the mean residual spatial effects on the regional variables. Thus, after estimating the geoadditive model, the total spatial effect of each region is explained by regressing the posterior mean of the estimated spatial residual effect on the regional variables. However, this empirical strategy does not consider the estimated posterior variance of spatial effects. In order to overcome this problem, a discrete choice model of the 95% or 80% spatial effects can be estimated. In this case, a variable is constructed that takes a value of (-1) when the region has a significant negative effect, takes a value of (0) if the effect is insignificant and takes a value of (1) if the effect is significant and positive. This leads to a straightforward multinomial specification. This variable is then regressed on the regional variables. We employ both strategies to examine the determinants of the residual spatial patterns. 2.4 Data The data used for the analysis is the 60th round employment-unemployment sur- vey of the National Sample Survey Organization (NSSO) of India conducted in 2004. As the focus of the paper is on economically active individuals, we restrict the sample to those who are older than 15 years but younger than 70 years. This reduces the sample size from 303,811 to 204,298.16 While the principal economic activity of this sample ranges from domestic duties to full time employment (in the form of salaried employment, self-employment, casual labor or unemploy- ment), 17% of the individuals in this sample are engaged in subsidiary activities. For the rest of the analysis, we consider the principal economic activity alone for two reasons. First, all individuals are not engaged in subsidiary activities. Second, as less than one sixth of the entire sample are engaged in subsidiary activities, considering such activities would further complicate the analysis when individu- als report as both self-employed and paid employees. Furthermore, the principal economic activity is the activity to which the individuals devote most of their time. For these reasons, we consider only the primary occupation for classifying workers into self-employment and paid employment. Table 2.1 lists the number of 16We drop 17 individuals who adhere Zoharastrianism for reasons of consistency with the next chapter. 16 individuals in different occupational categories. We also drop individuals who are unpaid family workers, students, workers involved in domestic duties, pensioners, those who are unable to work due to disabilities and people who reported to belong to the occupational class ‘other’. This reduces the final sample to 88,623 economically active individuals.17 We thus only consider those who have reported their primary occupation as self-employed (includes own account workers and em- ployers), salaried employees, casual laborers, or unemployed.18 The descriptive statistics in Table 2.2 show that 65% percent of the individuals have attended at least primary school, 65% live in rural areas and 40% are in the agricultural sector. Table 2.3 presents the descriptive statistics of self-employed and others in agricultural as well as nonagricultural sectors. Self-employed are older in both sectors. 13% of the self-employed in nonagriculture have university education compared to 3.7% of those who are self-employed in agriculture. A higher proportion of educated individuals are self-employed in agriculture and a higher proportion of educated individuals are salaried employees in nonagricul- ture. In the absence of an appropriate measure for wealth, we proxy it using the land-possed by the household. We thus posit that individuals who own large areas of land are more likely to be self employed. While in agriculture, land enables self-employed farming, and this makes people to choose self-employment over other modes of occupation, in the nonagricultural sector, land serves as potential collateral to obtain credit for starting an enterprise.19 These descriptive tables also show that more than 50% of individuals in agri- 1721.91% of these individuals are engaged in some subsidiary economic activity but for reasons listed earlier, we only consider the primary occupation in classifying individuals as self-employed workers or paid employees. 18We merge the occupations into self-employment and paid-employment for the rest of the analysis in this chapter. In the next chapter, we consider the four occupational categories as distinct classes. 19On the one hand, self-employed individuals in agriculture may possess more land as they need it for agricultural purposes. On the other hand, only those who possess land may be able to choose self-employment. Thus, the land possessed is also likely to determine the self-employment status. Hence the problem of endogeneity with respect to land even in the agricultural sector may not be so severe. The dataset has some information on the purchases made on the some durable commodities for some households. However, the information is missing for a number of households and for a number of items in the representative consumption bundle. Hence, we are not in a position to use this data. Furthermore, as income data is not available for the majority of individuals in the sample, we are not able to instrument the land possessed using income data. 17 culture are self-employed in comparison to a relatively lower proportion in nona- griculture. The presence of agricultural sector in the data poses several problems in analyzing the determinants of self-employment. The farm sector is usually found in rural areas with mainly farmers as self employed individuals. There are compelling reasons to posit that they are different from self-employed individuals in nonagriculture. As some scholars have noted before, the process of economic development reduces participation in farm sector and this induces a bias when analyzing the changes in self-employment rates with time if the agricultural sector is included in the analysis (Parker, 2004).20 Researchers have usually analyzed the determinants of self-employment only in the non-farm sector in order to get around these problems. As the farm sector is very important in a developing country like India, we also study self-employment in this sector. 2.5 Empirical Analysis In order to use the entire data set on hand and to make robust inferences on the determinants of self-employment, three different models are estimated. 2.5.1 Aggregate Model In the first model, participation in the agricultural sector is controlled using a dummy variable. The following semiparametric geoadditive probit model is estimated: η = γconst + γfemale + γmarital_status + γeducation_general + γeducation_technical + γwealth+γurban+γagri+γhindu+γbackward+fage+fspatial(district)+frandom(district) The non-linear effect of age is modeled as third degree P-Spline with second order random walk penalty.21 Figure 2.1(a) shows that the probability of being 20However, as our study is cross-sectional and does not analyze self-employment rates over time, this limitation does not apply here. Furthermore, we analyze the determinants of self- employment in agriculture and nonagriculture separately. 21The number of equidistant knots is assumed to be 20. The structured spatial effects are estimated based on Markov random field priors and random spatial effects are estimated with gaussian priors. The variance component in all the cases are estimated based on inverse gamma priors with hyperparameters a=0.001 and b=0.001. The number of iterations is set to 110000 with burnin parameter set to 10000 and the thinning parameter set to 100. The autocorrelation files and the sampling paths show that the MCMC algorithm has converged. These plots are available from the author. 18 2.5.2 Sector Specific Models Agricultural and Nonagricultural Self-employment The first model assumes that the determinants of self-employment are same for all self-employed individuals in agricultural as well as nonagriculture. In order to examine the differences in the two sectors, the following semiparametric model is estimated for individuals in agricultural and nonagricultural sectors separately: η = γconst + γfemale + γmarital_status + γeducation_general + γeducation_technical + γwealth + γurban + γhindu + γbackward + fage + fspatial(district) + frandom(district) The parameters for a, b, the number of iterations, burnin, and the thinning parameter are set equal to the first model’s parameters.29 The relationship of age with self-employment is very close to being linear in the agricultural sector, as seen in Figure 2.1(e), while in the nonagricultural sector, as Figure 2.1(c) shows, the age function increases at a decreasing rate until the age of 55 years and then increases at an increasing rate. Table 2.5 and Table 2.6 show considerable differences in relative human capital endowments of self-employed individuals in the two sectors. While in the agricultural sector, those who are endowed with higher levels of human capital (proxied by age and education) are more likely to be self employed, in the nonagricultural sector such individuals are more likely to be salaried employees. Belonging to a backward class is significantly negatively related to being self-employed in both the sectors, and being a Hindu has a significant negative relationship only in nonagriculture. For people in nonagriculture, as maps in Figure 2.3 suggest, the north-south divide seen in the spatial effect on the self-employment choice for individuals in the aggregate model is less pronounced. People of Kerala and some districts of Tamil Nadu in the south, Maharastra and Madhya Pradesh in western and central parts of India, and the majority of districts in the north-eastern states are less likely to be self-employed. People living in Uttar Pradesh, Bihar, Rajasthan, some districts of Andhra Pradesh, and West Bengal are more likely to be self-employed. The maps of spatial effects in agriculture in Figure 2.4 show that the result of north-south spatial divide observed in the first model can be attributed mainly to such a phenomenon in the agricultural sector. In sharp contrast to some districts in the western and the northern parts of India, people are very less likely to be 29The autocorrelation files and plots of the sampling paths show that sufficient convergence is achieved in these models also. 21 self-employed in agriculture in southern and central states. As Figures 2.3(b) and 2.4(b) demonstrate, the unstructured random effects are negligible compared to the structured spatial effects. The confidence interval plots for the random spatial effects also show that the local effects are small and insignificant compared to the effects of structured spatial effects in all the three estimated models.30 2.5.3 Determinants of Residual Spatial Patterns The presence of spatial patterns, as shown by the empirical analysis, suggests that it is not just personal characteristics of individuals that totally explain their occupational choice. As discussed below, regional characteristics also play an important role in determining self-employment choice. In particular, financial constraints, level of economic development, unemployment and small business employment are found to influence the self-employment rates in a region by earlier studies. Hence, we hypothesize that these variables can explain the residual spatial patterns. We follow the empirical approach described in subsection 2.3.3. Holtz-Eakin et al. (1994) test the role of liquidity constraints in the formation of new enterprises. Their analysis suggests that the size of inheritance has an effect on entrepreneurial choice and also on investment in the capital of a new enterprise. Many studies find that credit constraints are barriers to entry for individuals into self-employment (Evans and Jovanovic, 1989; Evans and Leighton, 1989b; Blanchflower and Oswald, 1998). Lindh and Ohlsson (1996) test for the presence of credit constraints as inhibitors to self-employment, by seeing if those who win a lottery are more likely to enter self-employment. They also find that such individuals start firms with higher capital. Cabral and Mata (2003) find that the presence of binding financial constraints inhibit firms from growing to their optimal size. Hence, we hypothesize that the level of financial development in the region, measured by the per-capita credit or the credit-deposit ratio in a district can explain the residual spatial pattern. Lucas (1978) predicts that entrepreneurship decreases with economic devel- opment. Calvo and Wellisz (1980) show that the growth rate of total stock of knowledge requires greater ability of the marginal entrepreneur in a steady state equilibrium. This suggests that, given a fixed ability distribution in a population, the number of entrepreneurs decreases and average firm size increases with tech- 30These plots are available from the author. 22 nological progress. Empirical studies of Acs et al. (1994) and Fölster (2002) find that per-captia gross net product (GNP) is negatively related to self-employment. Acs et al. (1994) argue that self-employment decreases in the early stages of de- velopment as technological change shifts output from agriculture and small scale industry to large scale manufacturing. We thus hypothesize that level of economic development determines the propensity to be self-employed in a region. Cross-sectional evidence gives a mixed impression about the effect of unem- ployment on the propensity to be self-employed. The recession-push hypothesis claims that high unemployment decreases the probability of getting paid employ- ment and thus pushes individuals into self-employment. However, the prosperity- pull hypothesis suggests that high unemployment reduces demand for goods and services of the self-employed, leading to a reduction in self-employment. Many cross-sectional studies find a negative relationship between unemployment and the probability of self-employment (Taylor, 1996; Blanchflower and Oswald, 1998). However, many studies also indicate that the self-employed experience a spell of unemployment (Evans and Leighton, 1989b; Blanchflower and Meyer, 1994). As Storey (1991) notes, time series studies show a positive relationship but cross-sectional studies suggest a negative relationship. Hence we hypothesize that unemployment could explain the residual self-employment pattern. We also introduce a number of demographic controls. In particular, we control for size of the district and the population density. Armington and Acs (2002) sug- gest that these factors play an important role in explaining the spatial patterns of new firm formation. We also control for agglomeration, measured by the density of firms in the region, as presence of a large number of firms in the neighbor- hood is likely to result in spillovers that induce new firm formation. As Krugman (1991, p. 484) notes, “the concentration of several firms in a single location offers a pooled market for workers with industry-specific skills, ensuring both a lower probability of unemployment and a lower probability of labor shortage.” Further- more, as Armington and Acs (2002, p.38) argue, “informational spillovers give clustered firms a better production function than isolated producers have. The high level of human capital embodied in their general and specific skills is another mechanism by which new firm start-ups are supported.” Thus regions with high agglomeration are more likely to be associated with higher probability of people entering self-employment. 23 the determinants in urban and rural areas. We control for regional effects using a set of state level regional dummies. We estimate this for the sub-sample of individuals in the nonagricultural sector alone.33 We also check the robustness of the estimates, with respect to the presence of land variables, by running separate regressions with and without land variables. We estimate the regressions with the land variables excluded in the first specification and land variables included in the second specification (Table 2.10). However, the regression estimates for the two specifications are not very different. It can be argued that in the Indian context, wealth plays a definite role in self-employment choice. As argued earlier, this is possible if credit is rationed in favor of individuals possessing assets such as land. We interpret the results of the specification with the land variables, as Table 2.10 suggests that the estimates of models with and without them are similar. The results are broadly consistent with results of the semi-parametric estima- tion. The estimated signs of higher education variables are negative in rural as well as urban areas. The absolute value of the coefficients are, however, slightly higher in the rural areas suggesting that educated people in the rural areas have a still lower propensity for self-employment. The returns to self-employment in ru- ral areas may be lower in comparison to the returns to self-employment in urban areas and this could explain this result. This issue is analyzed more extensively in chapter 5. While technical education is insignificant in rural estimations, it is significant and negative in urban regressions. The land variables are positive and increase the propensity to be self-employed in rural and urban areas. However, the coefficients are larger in urban areas, indicating that people in urban areas with more land have a higher propensity to choose self-employment. This may be because land in urban areas is more expensive relative to land in rural areas. This has a direct implication for obtaining credit from financial institutions. The esti- mates of the religion and caste variables are consistent with the semi-parametric model for the nonagricultural sector estimated earlier and the coefficients are significant and negative. The absolute value of the coefficient of the ‘Hindu’ vari- able is larger in the urban regression than in the rural regression equation. This is counter intuitive to some degree, because cultural institutions responsible for lower likelihood of Hindus and individuals of backward classes to be self-employed are expected to be stronger in rural areas. A plausible explanation is that individ- 33As the agricultural sector is mostly found in the rural areas only, we restrict the urban-rural analysis to the nonagricultural sector. 26 uals of other religions face greater discrimination in urban areas when it comes to wage-employment. Thus the probability of Hindus entering wage-employment may be higher in urban areas. 2.6 Conclusion The field of entrepreneurship in economics provides insights into the individual determinants of the self-employment choice in developed countries. We contribute to one aspect of this literature that remained neglected for a long time. We use recent advances in Bayesian semiparametric methodologies to examine the spa- tial as well as individual determinants of self-employment choice in a developing country, India. Consistent with studies based on datasets from developed coun- tries, we find age to have a non-linear relationship with the probability to be self-employed, particularly in nonagriculture. A clear jump after the age of 55 is noticed, which could be a direct result of the retirement effect. The effect is linear and monotonically increasing in agriculture. Married individuals are more likely to be self-employed in both sectors. In nonagriculture, educated people are less likely to be self-employed while in agriculture, they are more likely. The results are consistent with empirical studies of developed economies and also shed light on the unexplored agricultural self-employment in a developing country context. The analysis further suggests that in the nonagriculture, self-employed people are more or less uniformly distributed across different spatial units but in agri- culture self-employed individuals are concentrated in certain geographic pockets. In both sectors, the regions with the highest propensity of self-employment are the states of Uttar Pradesh and Bihar. While it can be argued that these regions are more entrepreneurial, these regions are also the poorest regions in India, in terms of per-capita income and human development. This leads to an important conclusion that self-employment in Indian context may actually support the view that self-employment in a fast growing economy like India continues to be the main occupational option in the poorest neighborhoods and not for individuals with high human capital. Furthermore, an analysis of the determinants of nona- gricultural self-employment in rural and urban areas suggests that in rural areas educated individuals have still lower propensity to become self-employed. 27 Ta bl e 2. 1: D is tr ib ut io n of O cc up at io n To ta lN um be r P er ce nt ag e C um ul at iv e Se lf- em pl oy ed (O w n A cc ou nt W or ke rs ) 37 ,1 97 18 .2 1 18 .2 1 Se lf- em pl oy ed (E m pl oy er s) 92 2 0. 45 18 .6 6 H ou se ho ld H el pe rs (U np ai d Fa m ily W or ke r) 23 ,5 05 11 .5 1 30 .1 6 Sa la ri ed E m pl oy ee s 21 ,2 23 10 .3 9 40 .5 5 C as ua lL ab or (P ub lic ) 31 0 0. 15 40 .7 0 C as ua lL ab or (O th er ) 23 ,8 23 11 .6 6 52 .3 6 U ne m pl oy ed 5, 14 8 2. 52 54 .8 8 St ud en ts 25 ,8 53 12 .6 5 67 .5 4 O nl y D om es ti c D ut ie s 40 ,8 94 20 .0 2 87 .5 6 D om es ti c D ut ie s an d C ol le ct io n of W oo d et c. 18 ,0 45 8. 83 96 .3 9 P en si on er s 2, 64 5 1. 29 97 .6 8 N ot w or ki ng du e to di sa bi lit y 1, 38 1 0. 68 98 .3 6 B eg ga rs an d P ro st it ut es 33 52 1. 65 10 0 To ta l 20 4, 29 8 10 0 28 15 28.5 42 55.5 69 -1.04 -0.5 0.05 0.59 1.13 Effect of age age (a) Posterior mean of the non-linear ef- fect of ‘age’ together with 95% and 80% pointwise credible intervals in the Ag- gregate Model. 15 28.5 42 55.5 69 -0.023 0.01 0.043 0.075 0.108 Derivative of Effect of age age (b) Derivative of the posterior mean of the ‘age’ function with 95% and 80% pointwise credible intervals in the Ag- gregate Model. 15 28.5 42 55.5 69 -0.95 -0.41 0.13 0.67 1.21 Effect of age age (c) Posterior mean of the non-linear ef- fect of ‘age’ together with 95% and 80% pointwise credible intervals in Nonagri- culture. 15 28.5 42 55.5 69 -0.055 -0.0010 0.053 0.106 0.16 Derivative of Effect of age age (d) Derivative of the posterior mean of the ‘age’ function with 95% and 80% pointwise credible intervals in Nona- griculture. 15 28.5 42 55.5 69 -1.27 -0.64 -0.01 0.61 1.24 Effect of age age (e) Posterior mean of the non-linear ef- fect of ‘age’ together with 95% and 80% pointwise credible intervals in Agricul- ture. 15 28.5 42 55.5 69 -0.041 -0.0050 0.031 0.067 0.102 Derivative of Effect of age age (f) Derivative of the posterior mean of the ‘age’ function with 95% and 80% pointwise credible intervals in Agricul- ture. Figure 2.1: Non-linear Effects of Age on Self-employment 31 Table 2.4: Determinants of Self-employment Variable Mean Std. Dev. 2.5%-Qt. 97.5%-Qt. Personal Characteristics Female -0.398 0.014 -0.426 -0.372 Married 0.175 0.018 0.141 0.211 Divorced 0.317 0.029 0.259 0.376 General Education Informal 0.265 0.019 0.227 0.304 Primary School 0.332 0.014 0.304 0.360 High School 0.193 0.016 0.163 0.224 University -0.181 0.020 -0.218 -0.141 Technical Education Technical Degree -0.127 0.057 -0.232 0.016 Technical Diploma -0.117 0.026 -0.168 -0.068 Land Possessed 0.2< Land <0.4 Hectares 0.149 0.014 0.120 0.176 0.4< Land < 2 Hectares 0.791 0.017 0.758 0.824 Land > 2 Hectares 1.180 0.024 1.132 1.226 Location Urban 0.253 0.013 0.227 0.279 Agriculture 0.336 0.013 0.312 0.361 Religion & Social Group Hindu -0.205 0.014 -0.233 -0.179 Backward -0.183 0.012 -0.206 -0.160 Constant -0.545 0.027 -0.599 -0.492 N 86140 Deviance(Mean) 93422.587 Std. Dev. 36.196992 deviance(µ̄) 92973.92 pD 448.66642 DIC 93871.253 Notes: Dependent variable is binary self-employment status of the indi- vidual. Base categories for marital status, general education, technical education, land dummies are unmarried, no general education, no tech- nical education and less than 0.2 hectares of land respectively. 32 Table 2.5: Determinants of Self-employment in Nonagriculture Variable Mean Std. Dev. 2.5%-Qt. 97.5%-Qt. Personal Characteristics Female -0.256 0.018 -0.290 -0.221 Married 0.203 0.019 0.165 0.240 Divorced 0.218 0.042 0.137 0.298 General Education Informal 0.141 0.028 0.085 0.195 Primary School 0.130 0.021 0.086 0.169 High School -0.039 0.022 -0.078 0.004 University -0.349 0.024 -0.395 -0.301 Technical Education Technical Degree -0.109 0.057 -0.217 0.004 Technical Diploma -0.134 0.025 -0.183 -0.084 Land Possessed 0.2< Land <0.4 Hectares 0.151 0.015 0.122 0.181 0.4< Land < 2 Hectares 0.112 0.022 0.070 0.153 Land > 2 Hectares 0.160 0.033 0.097 0.222 Location Urban 0.029 0.015 0.001 0.059 Religion & Social Group Hindu -0.180 0.016 -0.213 -0.149 Backward -0.150 0.014 -0.179 -0.121 Constant -0.222 0.031 -0.282 -0.163 N 51674 Deviance(Mean) 60166.724 Std. Dev: 34.978124 deviance(µ̄) 59807.524 pD 359.20045 DIC 60525.925 Notes: Dependent variable is binary self-employment status of the indi- vidual. Base categories for marital status, general education, technical education, land dummies are unmarried, no general education, no tech- nical education and less than 0.2 hectares of land respectively. 33 -0.745864 0 0.681658 (a) Structured Non-linear Effect of ‘District’. Shown are the posterior means. -0.0960833 0 0.112093 (b) Unstructured Random Effect of ‘District’. Shown are the posterior means. (c) Non-linear Effect of ‘District’. Pos- terior probabilities for a nominal level of 95%. Black denotes regions with strictly negative credible intervals, white denotes regions with strictly pos- itive credible intervals. (d) Non-linear Effect of ‘District’. Posterior probabilities for a nominal level of 80%. Black denotes regions with strictly negative credible inter- vals, white denotes regions with strictly positive credible intervals. Figure 2.3: Spatial Effects in ‘Nonagriculture’ 36 -1.89128 0 2.67315 (a) Structured Non linear Effect of ‘District’. Shown are the posterior means. -0.240888 0 0.202827 (b) Unstructured Random Effect of ‘District’. Shown are the posterior means. (c) Non–linear Effect of ‘District’. Posterior probabilities for a nominal level of 95%. Black denotes regions with strictly negative credible inter- vals, white denotes regions with strictly positive credible intervals. (d) Non–linear Effect of ‘District’. Posterior probabilities for a nominal level of 80%. Black denotes regions with strictly negative credible inter- vals, white denotes regions with strictly positive credible intervals. Figure 2.4: Spatial Effects in ‘Agriculture’ 37 Ta bl e 2. 7: D et er m in an ts of Sp at ia lP at te rn s in F ig ur e 2. 2, F ig ur e 2. 3 an d F ig ur e 2. 4 A ll N on ag ri cu lt ur e A gr ic ul tu re F in an ci al D ev el op m en t P er -c ap it a C re di t 0. 00 62 2 -0 .0 18 3 0. 02 75 (0 .0 16 ) (0 .0 12 ) (0 .0 44 ) C re di t- D ep os it R at io -0 .1 02 ** * 0. 04 36 ** -0 .4 02 ** * (0 .0 23 ) (0 .0 18 ) (0 .0 61 ) E co no m ic D ev el op m en t P er -C ap it a N SD P -0 .3 10 ** * -0 .2 68 ** * -0 .2 91 ** * -0 .3 25 ** * -0 .4 18 ** * -0 .2 53 ** * (0 .0 34 ) (0 .0 30 ) (0 .0 26 ) (0 .0 23 ) (0 .0 91 ) (0 .0 78 ) U ne m pl oy m en t -0 .0 60 3* ** -0 .0 47 1* ** 0. 04 06 ** * 0. 03 69 ** * -0 .2 91 ** * -0 .2 39 ** * (0 .0 16 ) (0 .0 15 ) (0 .0 12 ) (0 .0 12 ) (0 .0 41 ) (0 .0 40 ) D em og ra ph ic s M id Si ze D is tr ic t 0. 00 32 5 0. 01 41 0. 08 69 ** * 0. 08 19 ** * -0 .1 91 ** -0 .1 47 * (0 .0 30 ) (0 .0 29 ) (0 .0 22 ) (0 .0 22 ) (0 .0 79 ) (0 .0 76 ) La rg e D is tr ic t 0. 02 80 0. 03 05 0. 07 50 0. 07 19 -0 .1 76 -0 .1 61 (0 .0 91 ) (0 .0 90 ) (0 .0 68 ) (0 .0 68 ) (0 .2 4) (0 .2 3) P op ul at io n D en si ty -0 .0 18 9 -0 .0 18 3 0. 05 94 ** * 0. 05 54 ** * -0 .1 29 ** * -0 .1 29 ** * (0 .0 15 ) (0 .0 14 ) (0 .0 11 ) (0 .0 11 ) (0 .0 40 ) (0 .0 37 ) A gg lo m er at io n In de x F ir m D en si ty -0 .0 07 67 -0 .0 02 13 -0 .0 02 90 -0 .0 08 74 -0 .0 17 9 0. 00 45 3 (0 .0 13 ) (0 .0 12 ) (0 .0 09 4) (0 .0 09 0) (0 .0 33 ) (0 .0 31 ) C on st an t 2. 92 6* ** 2. 53 4* ** 2. 61 8* ** 2. 74 9* ** 4. 28 0* ** 2. 77 8* ** (0 .3 5) (0 .3 6) (0 .2 7) (0 .2 7) (0 .9 5) (0 .9 3) O bs er va ti on s 53 4 53 4 53 1 53 1 53 2 53 2 R 2 0. 20 0. 23 0. 40 0. 40 0. 16 0. 22 F 19 .0 8 22 .4 6 49 .4 3 50 .3 0 14 .1 1 21 .4 2 R 2 A dj us te d 0. 19 2 0. 22 0 0. 39 0 0. 39 4 0. 14 7 0. 21 2 N ot es : *S ig ni fie s p< 0. 05 ; ** Si gn ifi es p< 0. 01 ; ** * Si gn ifi es p< 0. 00 1. St an da rd er ro rs ar e re po rt ed in pa re nt he se s. D ep en de nt va ri ab le is th e m ea n sp at ia le ffe ct pe r di st ri ct af te r es ti m at io n of th e ge oa dd it iv e m od el s. 38 Table 2.10: Self-employment in Nonagriculture Rural and Urban Regressions Model I Model II Independent Var. Rural Urban Rural Urban Personal Characteristics Age 0.0298*** 0.0332*** 0.0294*** 0.0335*** (0.0052) (0.0049) (0.0052) (0.0049) Age Square -0.0224*** -0.0229*** -0.0221*** -0.0239*** (0.0064) (0.0059) (0.0065) (0.0060) Female -0.232*** -0.275*** -0.231*** -0.276*** (0.027) (0.024) (0.027) (0.024) Married 0.252*** 0.298*** 0.255*** 0.302*** (0.028) (0.027) (0.028) (0.027) Divorce/Widow 0.376*** 0.250*** 0.380*** 0.268*** (0.061) (0.053) (0.061) (0.053) General Education Informal Education 0.175*** 0.0874** 0.170*** 0.0799** (0.038) (0.040) (0.038) (0.040) Primary School 0.159*** 0.0759*** 0.155*** 0.0614** (0.027) (0.028) (0.027) (0.029) High School -0.0540* -0.0248 -0.0567** -0.0510* (0.028) (0.029) (0.029) (0.030) Diploma/University Education -0.410*** -0.278*** -0.412*** -0.317*** (0.036) (0.032) (0.036) (0.032) Technical Education Technical Degree 0.168 -0.211*** 0.164 -0.220*** (0.12) (0.063) (0.12) (0.063) Technical Diploma 0.0251 -0.205*** 0.0262 -0.208*** (0.042) (0.033) (0.042) (0.033) Household Controls 0.2< Land <0.4 Hectares 0.117*** 0.166*** (0.027) (0.018) 0.4< Land <2 Hectares 0.0603** 0.226*** (0.030) (0.043) Land >2 Hectares 0.113*** 0.344*** (0.041) (0.066) Hindu -0.128*** -0.237*** -0.128*** -0.238*** (0.024) (0.020) (0.024) (0.020) Backward -0.117*** -0.157*** -0.119*** -0.157*** (0.021) (0.018) (0.021) (0.018) Total Observations 23916 28611 23895 28589 Log Likelihood -14191 -16930 -14169 -16865 LR (χ2) 2472 2685 2492 2789 Degrees of freedom 47 47 50 50 Pseudo R2 0.0801 0.0735 0.0808 0.0764 Notes: Probit estimation. *Signifies p< 0.05; ** Signifies p<0.01;*** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is ‘selfemployed’. State dummies are included in all the regressions and are not reported here. The coefficients of the constant are not reported. 41 Chapter 3 Religion and Entrepreneurship While considerable concern has emerged about the impact of religion on economic de- velopment, little is actually known about how religion impacts the decision making of individuals. This chapter examines the influence of religion on the decision for people to become an entrepreneur. Based on a large-scale data set of nearly ninety thousand workers in India, this chapter finds that religion shapes the entrepreneurial decision. In particular, some religions, such as Islam and Christianity, are found to be more con- ducive to entrepreneurship than Hinduism. In addition, the caste system is found to influence the propensity to become an entrepreneur. Individuals belonging to a back- ward caste exhibit a lower propensity to become an entrepreneur. Thus, the empirical evidence suggests that both religion and the tradition of the caste system influence entrepreneurship, suggesting a link between religion and economic behavior. 3.1 Introduction Religion and economics have had a tenuous relationship. On the one hand, schol- ars dating back at least to Adam Smith and Max Weber have argued that religion plays a fundamental role in shaping economics.1 On the other hand, only scant attention has recently been given as to how and why religion might influence eco- nomics. The omission of religion as a determinant of economic activity is startling, given the recent suggestion by Iannaccone (1998, pp. 1492) that “the economics of religion will eventually bury two myths - that of homo economicus as a cold 1Anderson (1988, p. 1068) notes, “In Wealth, Smith was not interested in theological issues or even in the nature of religious belief. Rather, he was concerned with two basic problems: (1) the economic incentives involved in the individual’s decision to practice religion and (2) the economic effects of different systems of religious belief as reflected in individual behavior. He did not attempt to develop an economic theory of the emergence of religious beliefs... Smith attempted the more limited task of defining the logical economic consequences of certain kinds of religious beliefs.” 42 creature with neither need nor capacity for piety, and that of homo religiosus as a benighted throwback to pre-rational times.” Moreover, as Edmund Phelps argues, “values and attitudes are as much a part of the economy as institutions and policies are. Some impede, others enable.”2 In India, for instance, Hinduism is strongly associated with the emergence of the caste system. Although some aspects of the caste system such as untoucha- bility, were abolished by the government, it remains formidable and imposing in practice. There remains a heated public debate in India on the impact of the caste system on the economic status of what is widely referred to “backward classes”. For example, in an article announcing, “Indian College Quota Law Suspended”, The New York Times reports that, “Caste discrimination is outlawed but contin- ues to persist in obvious and subtle ways, and the contest over the latest university admissions quotas revolve around how to best redress an entrenched and often ugly social bias.”3 Recent studies suggest the existence of a relationship between religion and economic performance (Barro and McCleary, 2003; McCleary and Barro, 2006b; Guisa et al., 2006). For example, Barro and McCleary (2003) estimate the im- pact of adherence to religious beliefs on economic performance using international survey data on religiosity. They find that increases in church attendance tend to reduce economic growth while increases in the belief in hell and an afterlife in- crease economic growth. These empirical findings raise several important but unanswered questions: (1) What are the channels by which religion influences economic activity? and (2) Is the impact of religion on economic activity homo- geneous across all religions? The purpose of this paper is to shed light on these questions by examining whether religion has any impact on one particular channel of economic decision- making influencing economic growth – the decision to become an entrepreneur. Recent studies suggest that entrepreneurship may be a key factor generating growth and development (Baumol, 2002). As Lazear (2002, p. 1) concludes, “The Entrepreneur is the single most important player in a modern economy.” Lazear’s conclusion is supported by considerable theoretical and empirical literature link- ing entrepreneurship to economic growth.4 2“It’s All About Attitude,” Newsweek International Edition, 30 April, 2007. 3“India College Law Suspended,” The New York Times, 29 March, 2007. 4See for example the studies by Holtz-Eakin and Kao (2003) and Audretsch et al. (2006). 43 in India. Compared to the other main religions of India, Hinduism provides little encouragement or value to change one’s situation in terms of material well being (Singer, 1966). According to Uppal (2001, p. 20), “The people of South Asia are deeply religious and all facets of their lives including their endeavors to achieve material advancement are affected greatly by religious beliefs and values.”7 According to Hinduism every human being is Amrutasya Putraha, a child of immortality and a spark of divinity. The purpose of life is to attain liberation which essentially is freedom from re-birth and the chain of cause and effect. One should live to understand reality and not for transitory material pursuits. Dharma Righteousness, Artha Earnings, Kama Desire, Moksha Liberation are supposed to guide the lives of Hindus. The scriptures ordain individuals to follow righteousness, perform duties and earn their livelihood, satisfy their desires and finally seek liberation. Dharma, Artha, Kama, Moksha can also be interpreted differently: one should righteously earn his livelihood and desire only for liberation (also referred to as self-realization). An individual has to do his duty as dictated by the scriptures and should not loose himself in material pursuits. Varna refers to classification of individuals into different classes, categories or castes. Historically Hindus were classified into four major castes. Initially their occupation determined their caste and caste affiliation akin to the religious iden- tity was passed on to their progeny. Brahmins were scholars, priests, advisors to kings, intelligentsia of the community. Kshatriyas were kings and noblemen. Their duties involved protection of the community from enemies and adminis- tration. Traders, businessmen and entrepreneurs were Vyshyas and people of all other occupations were classified as Shudras. Thus the Varna System that ini- tially categorized individuals into different classes persisted across generations and later determined the occupations of Hindus to a great extent. In his third major work on the sociology of religion, Weber (1958, pp. 103-104) states that “If the stability of the caste order could not hinder property differ- entiation it could at least block technological change and occupational mobility, which from the point of view of caste were objectionable and ritually danger- ous.” In summary, he claims that the impact of caste system on the economy is essentially negative (Medhora, 1965). In one of the few studies analyzing the effects of the caste system, Munshi and 7Uppal (2001) also provides an excellent overview of the philosophy of Hinduism. 46 Rosenzweig (2006) examine the influence of the caste within the context of an educational choice model in Bombay. They find that lower caste boys are more likely to study in schools where the medium of instruction is the local language and not English. This is very likely to lead them into traditional occupations as defined by the caste structure. Munshi and Rosenzweig (2006, p. 1230) note, “caste networks might place tacit restrictions on the occupational mobility of theirs members to preserve the integrity of the network” and “although these restrictions might have been welfare enhancing and indeed equalizing when they were first put in place, such restrictions could result in dynamic inefficiencies when the structure of the economy changes.” The clear demarcation of occupations based on castes, the persistence of oc- cupation decisions across generations and the other tenets that entail Hindus not to live a life of material pursuits, lead us to hypothesize that these factors might continue to influence the occupational choices of Hindus, and in particular inhibit the propensity to become an entrepreneur. We have no strong predictions how other religions in India, like Islam or Christianity, might influence an individual’s entrepreneurial decision. It is likely, however, that the impact of the caste system on economic behaviors is stronger for Hindus as compared to non-Hindus. In the following sections we will analyze whether Hinduism, as well as belong- ing to a lower caste, will influence the propensity to become an entrepreneur. 3.3 Data The main source of data to link religion and caste affiliation to entrepreneurship is the National Sample Survey Organization (NSSO) of India. We use the NSSO’s 60th round Employment-Unemployment Survey. This household level survey was conducted in 2004. Almost three hundred thousand individuals in sixty thousand households were questioned about their economic status, religious affiliation and personal background. The households were selected based on a stratified sampling methodology. Since the focus of this paper is on economically active individuals, we only consider those who have reported to be: self employed (includes own account workers and employers), salaried employees, casual laborers and unem- ployed. For similar reasons, we restrict our sample to those who are older than 15 years but younger than 70 years. We thus exclude from our analysis family 47 members who assist household enterprises, such as children and the elderly, as well as people classified into other miscellaneous occupational categories. These individuals can also be located according to their region. The final sample consists of 87,181 individuals. Table 3.1 provides the means and standard deviations of the independent vari- ables. 79% of the final sample are Hindus, 11.2% are Muslims, 5.6% are Christians, 1.4% are Sikhs, 0.3% are Jains, 1% are Buddhists and 1.1% are individuals of other religions or without religion. This roughly corresponds to the distribution of religion within the overall population of India.8 66.5% of Jains in the sample are self-employed, 50.4% of Christians and 48.6% of Muslims, 41% of Hindus and Sikhs and 38% of Buddhists. (Figure 3.1 and Table 3.2). Individuals included in the database are also classified according to class affil- iation. They belong to either one of the three backward classes (Schedule Castes, Schedule Tribes, Other Backward Classes) or to the forward castes. 12.5% of the sample belong to schedule castes, 18% to schedule tribes, 36.8% to other backward classes. These three classes combine to account for 67.5% of the entire sample. It should be emphasized that although the caste system is a distinct feature of Hin- duism and the Constitution of India (Schedule Castes) Order, 1950 notes that, “no person who professes a religion different from the Hindu, the Sikh or the Buddhist religion shall be deemed to be a member of a Scheduled Caste”, almost 66% of Christians are classified in the Schedule Caste. As Table 3.3 suggests, the other religions also have a share of their population that claims to be backward. While in Christianity this may be the result of conversion of individuals of the lower castes of Hinduism, in other religions this possibly reflects the economic backwardness rather than social backwardness. The presence of caste system, a characteristic of Hinduism, is also reflected in other religions in India. Within Islam certain sects are considered to be nobler than others. In Christianity, con- verts from lower castes of Hindu society are treated as lower caste members of Christianity. We cannot rule out conversions into Christianity giving rise to this phenomena. Also, we cannot rule out the possibility of the caste system diffusing into other religions in India. When we examine class based occupational behavior specifically in Hinduism, 8According to the 2001 Census, the religious composition of population in India is as follows: 80.9% are Hindus, 12.9% are Muslims, 2.4% are Christians, 1.9% are Sikhs, 0.4% are Jains, 0.8% are Buddhists, and 0.7% are others. See Premi (2004, p. 4294). 48 backward class but not members of other religions. One might therefore argue that the reservation system enables Hindu backward class to favor salaried em- ployment instead of self employment whereas members of other religions choose self employment. However, the values of estimated marginal effects suggest that the positive coefficients for salaried employment category are negligible compared to the negative coefficients in the self-employment category. This suggests that the effect of caste system in inhibiting Hindus from selecting self-employment is significant. In fact, the backward class Hindus have a higher propensity to be casual laborers. In order to focus on the impact of caste system we estimate the model based on the sample of Hindus only (Table 3.9) . The strong presence of class struc- tures within Hinduism leads us to posit that Hindu individuals belonging to the backward class might have a lower propensity to become an entrepreneur than Hindus belonging to the forward class. Thus, the impact of both religion and caste system, by being both a Hindu and a member of the backward class on the decision to become an entrepreneur was estimated and the results are presented in Table 3.9. The evidence suggests that a Hindu who is a member of the back- ward class scheduled caste is almost 14.6% and backward class scheduled tribe is 18% less likely to be self employed than a forward class Hindu. The four estimated models confirm our hypotheses that Hindus are less likely to be entrepreneurs than are individuals of other religions. This leads us to the last question. How does the propensity to enter into entrepreneurship compare between the non-Hindu and the Hindu religions? Thus, the results included in Table 3.10 take Hinduism as the base class and show the marginal effect on the probability to be self employed for individuals of other religions. The results sug- gest that Muslims are 7.9%, Christians 2.9%, and Jains 27% more likely to be self employed compared to Hindus. By contrast, individuals of other minor reli- gions and those without religion are almost 13.4% more likely to be entrepreneurs compared to Hindus. Buddhists and followers of Sikhism are pretty much in the same boat as Hindus. As a further check of the robustness of the results, we estimate a model by considering the self-employed separated as employers and only self-employed people (Table 3.11). It is startling to observe that the coefficients of the Hindu variable and the backward class variable are significant and negative even for 51 the employer group. This suggests that the Hindus have a lesser propensity to be entrepreneurs.13 An important qualification of the results is that the self- employed includes both agricultural and non-agricultural self-employed people. However, when the sample is restricted to non-agriculture (Table 3.12), the results confirm that there is virtually no difference. It is important to note that minority communities are associated with higher self-employment rates even in the developed countries (Clark and Drinkwater, 1998). However, the insight from our analysis is that even when we consider the Hindus alone, the caste system has an effect on the propensity to be self-employed. This supports our theory that the caste-system continues to exert an influence on the occupational choice of Hindus. 3.5 Conclusion Religion is rarely attributed to shaping economic phenomena. So it is with the de- cision to become an entrepreneur. While a rich and robust literature has emerged identifying a number of important characteristics and factors alternatively con- ducive to or impeding entrepreneurship, religion has been noticeably absent. The results of this paper suggest that religion matters. While India is rich with diverse religions, some of them, such as Islam and Christianity, are conducive to entrepreneurship. By contrast, others, and in particular Hinduism, inhibit entrepreneurship. We control for regional specific effects by introducing state level dummies and the results are robust to these controls as well. Similarly, the caste system is found to influence the propensity to become an entrepreneur. In particular, belonging to a backward caste inhibits entrepreneur- ship. The least entrepreneurial people tend to be Hindus in the lower class. One reason for this might be the long shadow of caste system that persists and limits the freedom of occupational choice to some extent not only to all individuals of backward classes but to Hindus in particular. Hence, the results of this paper suggest that elements of religion and the caste system need to be explicitly considered in understanding what influences important economic phenomena, such as entrepreneurship. Just as religion plays 13The marginal effects are very small but this could partly be attributed to the very small number of employers in the sample. 52 a major role in influencing entrepreneurial activity, so too does the caste system. At least in the case of India, Max Weber’s insight is found to hold - religion is an important influence on economic behavior. It may be fruitful for future research to consider not just the impact of religion on economic activity, such as entrepreneurship, but also the conditioning effect of the particular locational context. One clue about the importance of location is provided by the results of studies showing that Indian and other Asian im- migrants in the United Kingdom and North America actually exhibit a greater propensity for entrepreneurship (Clark and Drinkwater, 1998). While the specific religion of the immigrants is not explicitly identified, the inhibiting impact of a specific religion and particular caste may, in fact, disappear along with the change in location and institutional context. Without the painstaking future research, however, such a conjecture will remain simply that, a conjecture. 53 Table 3.2: Religion and Occupational Choice (Descriptives) Religion Self Salaried Casual Unemployed Total Employed Employee Labor Hinduism 41.30 23.90 28.99 5.81 100 Islam 48.62 20.92 24.28 6.17 100 Christianity 50.43 30.01 13.58 5.98 100 Sikhism 41.00 30.53 22.2 6.26 100 Jainism 66.54 28.08 4.23 1.15 100 Buddhism 37.97 26.00 32.15 3.88 100 Others 69.69 16.45 9.70 4.16 100 Total 43.01 23.95 27.23 5.81 100 Table 3.3: Religion and Caste System (Descriptives) Religion Backward Backward Backward Forward Total Caste(SC) Tribe(ST) Other(OB) Caste Hinduism 8.84 21.28 40.06 29.82 100 Islam 2.98 0.99 35.67 60.37 100 Christianity 66.24 3.69 11.60 18.47 100 Sikhism 0.56 31.56 19.90 47.98 100 Jainism 7.31 0.00 2.69 90.00 100 Buddhism 39.27 50.81 5.83 4.10 100 Others 85.36 1.30 11.68 1.67 100 Total 12.52 18.17 36.88 32.43 100 For explanation on SC, ST, OB see notes of Table 3.1. 56 Table 3.4: Caste System and Occupation (Descriptives) Social Self Salaried Casual Unemployed Total Group Employed Employee Labor Backward Caste(SC) 46.91 18.69 29.77 4.62 100 Backward Tribe(ST) 28.32 18.72 47.39 5.57 100 Backward Other(OB) 45.75 21.59 27.50 5.17 100 Forward Caste 46.62 31.59 14.66 7.13 100 Total 43.01 23.95 27.23 5.81 100 For explanation on SC, ST, OB see notes of Table 3.1. Table 3.5: Caste System and Occupation in Hinduism (Descriptives) Social Self Salaried Casual Unemployed Total Group Employed Employee Labor Backward Caste(SC) 36.10 13.72 45.70 4.48 100 Backward Tribe(ST) 28.78 18.29 47.45 5.47 100 Backward Other(OB) 45.67 21.44 27.84 5.05 100 Forward Caste 45.90 34.23 12.43 7.44 100 Total 41.3 23.9 29 5.8 100 For explanation on SC, ST, OB see notes of Table 3.1. 57 Table 3.6: Hinduism and Entrepreneurship (Marginal Effects after Multinomial Probit Estimation) Independent Self Salaried Casual Unemployed Employed Employee Labor Religion: Hinduism -0.0861*** 0.0293*** 0.0534*** 0.00346*** (0.0052) (0.0044) (0.0042) (0.00088) Personal Characteristics: Age 0.0123*** 0.00758*** -0.0160*** -0.00397*** (0.0011) (0.0010) (0.00093) (0.00031) Agesq/100 -0.00424*** -0.00834*** 0.00939*** 0.00318*** (0.0013) (0.0012) (0.0011) (0.00040) Female -0.133*** 0.0630*** 0.0425*** 0.0272*** (0.0055) (0.0052) (0.0048) (0.0019) Married 0.0883*** -0.0445*** 0.000897 -0.0447*** (0.0066) (0.0058) (0.0056) (0.0026) Divorced 0.106*** -0.0540*** -0.0375*** -0.0149*** (0.012) (0.0096) (0.0089) (0.0011) General Education: Informal Education 0.0308*** 0.0721*** -0.102*** -0.000700 (0.0084) (0.0087) (0.0045) (0.0026) Primary School 0.0148** 0.170*** -0.202*** 0.0171*** (0.0060) (0.0060) (0.0035) (0.0022) High School -0.0763*** 0.312*** -0.286*** 0.0499*** (0.0065) (0.0066) (0.0029) (0.0037) University -0.226*** 0.426*** -0.297*** 0.0958*** (0.0070) (0.0081) (0.0022) (0.0066) Technical Education: Technical Degree 0.0139 0.0930*** -0.107*** 0.000122 (0.025) (0.021) (0.027) (0.0033) Technical Diploma -0.00744 0.105*** -0.111*** 0.0134*** (0.010) (0.0090) (0.0084) (0.0021) Household Characteristics: Urban 0.0439*** 0.171*** -0.218*** 0.00384*** (0.0047) (0.0042) (0.0033) (0.00088) 0.2<Land<0.4 Hectares 0.0730*** -0.0762*** 0.00339 -0.000272 (0.0054) (0.0044) (0.0045) (0.0010) 0.4< Land < 2 Hectares 0.325*** -0.146*** -0.176*** -0.00309*** (0.0055) (0.0045) (0.0039) (0.0011) Land > 2 Hectares 0.397*** -0.154*** -0.237*** -0.00606*** (0.0053) (0.0047) (0.0026) (0.0012) Observations 87181 Notes: *Signifies p< 0.05; ** Signifies p<0.01;*** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is primary occupation of the individ- ual. Base categories for marital status, general education, technical education, land dummies are unmarried, no general education, no technical education and less than 0.2 hectares of land respectively. Full set of state level regional dummies are also included in the regression. 58 Table 3.9: Backward Classes and Entrepreneurship (Only Hindus) (Marginal Effects after Multinomial Probit Estimation) Independent Self Salaried Casual Unemployed Employed Employee Labor Religion and Class: Hindu SC -0.146*** -0.0331*** 0.183*** -0.00331** (0.0084) (0.0078) (0.0090) (0.0016) Hindu ST -0.181*** -0.0415*** 0.222*** 0.000495 (0.0063) (0.0054) (0.0067) (0.0012) Hindu OBC -0.0446*** -0.0425*** 0.0926*** -0.00547*** (0.0057) (0.0048) (0.0055) (0.0010) Controls: Personal Characteristics YES General Education YES Technical Education YES Household Characteristics YES Regional Dummies YES Observations 69705 Notes: *Signifies p< 0.05; ** Signifies p<0.01; *** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is primary occupation of the individual. Base category for the Hindu caste is Hindu Forward. Set of state level regional dummies that have nonzero observations in all the four categories are included in the regression. 61 Table 3.10: Religion and Entrepreneurship (Marginal Effects after Multinomial Probit Estimation) Independent Self Salaried Casual Unemployed Employed Employee Labor Religion and Class: Muslim 0.0792*** -0.0475*** -0.0271*** -0.00462*** (0.0063) (0.0052) (0.0052) (0.00098) Christian 0.0290** 0.0200** -0.0490*** -0.0000146 (0.012) (0.010) (0.0090) (0.0020) Sikh 0.00315 -0.0224 0.0145 0.00476 (0.021) (0.016) (0.020) (0.0048) Jain 0.271*** -0.132*** -0.124*** -0.0155*** (0.029) (0.018) (0.027) (0.00094) Buddhist -0.0194 0.0350* -0.0111 -0.00444 (0.021) (0.018) (0.016) (0.0031) Others 0.134*** -0.0493** -0.0827*** -0.00196 (0.022) (0.019) (0.017) (0.0044) Backward Class -0.0778*** -0.0150*** 0.0941*** -0.00126 (0.0047) (0.0041) (0.0039) (0.00087) Controls: Personal Characteristics YES General Education YES Technical Education YES Household Characteristics YES Regional Variables YES Observations 87175 Notes: *Signifies p< 0.05; ** Signifies p<0.01; *** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is primary occupation of the individual. Base category for religion is Hindu. 62 Ta bl e 3. 11 :S el f-e m pl oy ed an d E m pl oy er s (M ar gi na lE ffe ct s af te r M ul tin om ia lP ro bi t E st im at io n) In de pe nd en t Se lf E m pl oy er Sa la ri ed C as ua l U ne m pl oy ed E m pl oy ed E m pl oy ee La bo r R el ig io n an d C la ss : H in du is m -0 .0 72 0* ** -0 .0 01 61 ** 0. 01 09 ** * 0. 05 88 ** * 0. 00 38 6* ** (0 .0 04 8) (0 .0 00 72 ) (0 .0 04 2) (0 .0 03 9) (0 .0 00 87 ) B ac kw ar d C la ss -0 .0 72 7* ** -0 .0 07 31 ** * -0 .0 18 2* ** 0. 10 00 ** * -0 .0 01 68 * (0 .0 04 4) (0 .0 00 83 ) (0 .0 03 8) (0 .0 03 7) (0 .0 00 88 ) C on tr ol s: P er so na lC ha ra ct er is ti cs Y E S G en er al E du ca ti on Y E S Te ch ni ca lE du ca ti on Y E S H ou se ho ld C ha ra ct er is ti cs Y E S R eg io na lD um m ie s Y E S O bs er va ti on s 87 17 5 N ot es : E m pl oy er s ar e tr ea te d as a se pa ra te cl as s he re . *S ig ni fie s p< 0. 05 ; ** Si gn ifi es p< 0. 01 ; ** * Si gn ifi es p< 0. 00 1. St an da rd er ro rs ar e re po rt ed in pa re nt he se s. D ep en de nt va ri ab le is pr im ar y oc cu pa ti on of th e in di vi du al .B as e ca te go ry fo rr el ig io n is no n- H in du an d fo rc as te is no n- ba ck w ar d cl as s. Se t of st at e le ve lr eg io na ld um m ie s th at ha ve no nz er o ob se rv at io ns in al lt he fiv e ca te go ri es ar e in cl ud ed in th e re gr es si on . 63 The few studies on start-up size show that the industry characteristics (Mata and Machado, 1996; Mata, 1996) and human capital of entrepreneurs (Astebro and Bernhardt, 2005; Colombo et al., 2004; Colombo and Grilli, 2005), determine the start-up size of new firms. However, the role of spatial location on the start-up size has never been studied although the economic geography literature empha- sizes the geographic location as an important determinant in shaping economic activity (Krugman, 1991; Fujita and Krugman, 2003). This paper contributes to the growing literature on the start-up size by highlighting that the firm size distribution of start-ups (FSDS) is not independent of the spatial context. Using recent methodological advances in spatial econometrics and a dataset of 150,000 firms that registered as small firms in India from 1998-2000, we find that the FSDS is remarkably spatially skewed and displays distinct spatial patterns. The paper consists of five sections. In the next section, we discuss the theoret- ical framework and present the hypotheses on the FSDS in an Indian context. In the third section, we present the geoadditive modeling techniques with Bayesian inference based on Monte Carlo Markov Chain(MCMC) methods. In the fourth section, we give the empirical results linking the region with the FSDS. In the final section, we provide the conclusions and summary and present possible avenues for future research. 4.2 The Start-Up Size One of the stylized facts in the industrial dynamics literature is that the magni- tude of firm entry, across industries, time periods, and regions is quite startling. Firm size distribution is skewed and the majority of entrants are small (Cable and Schwalbach, 1991). The likelihood of survival for new entrants is low and those that do survive grow at a higher rate than the incumbents. Firms that have a higher start-up size have a higher likelihood of survival (Dunne et al., 1989; Guimaraes et al., 1995).3 Many empirical studies categorically reject the Gibrat’s Law which, in essence, claims that the firm growth is independent of size. Three 3However, there are some exceptions. Agarwal and Audretsch (2001) show that the entry size is more important in the early stages of the industry life cycle but not in the mature stages. Audretsch et al. (1999), however, find that there is no relationship between start-up size and firm survival in a sample of Italian firms. They also find that growth rates are even neagtively correlated with initial size. 66 important surveys (Geroski, 1995; Sutton, 1997; Caves, 1998) summarize these and other major findings of the literature on entry, growth, survival and exit of firms. While the effects of entry are extensively discussed, the determinants of the start-up size have received little attention. As Colombo et al. (2004, p. 1184) note, “if a larger start-up size positively affects the likelihood of survival of new firms and if surviving new firms that started operations at smaller scale struggle to catch up, the question arises why there are firms with small initial size.” The few empirical studies on the determinants of the start-up size of firms include Mata and Machado (1996), Mata (1996), Görg et al. (2000), Görg and Strobl (2002), Astebro and Bernhardt (2005), Colombo et al. (2004), Colombo and Grilli (2005). These studies examine the role of industry characteristics such as the minimum efficient scale (MES) of the industry, industry growth, effects of operation at suboptimal scale (defined as the proportion of those employed in firms that are operating at sub-optimal scale), impact of market size, role of human capital characteristics of founders, such as previous work experience and education, and credit constraints, on the initial size of firms. As Mata and Machado (1996, p. 1321)4 note, “entry on a relatively large scale in each industry is much more sensitive to the minimum efficient scale and to the extent of firm turnover in the industry than entry in small scale. Put differently, it seems that small new firms appear everywhere, while relatively large ones only appear where economies of scale make it crucial, or where sunk costs are low, therefore leading to low losses in case of failure.” A similar study on Irish firms shows comparable results, but finds a negative effect of industry size and positive effect of industry growth on start-up size (Görg et al., 2000). The start-up size increases with age and education of the founder, and is higher in industries with higher minimum efficient scale (MES), greater turbulence, and in industries where few suboptimal firms operate (Mata, 1996). Industry-specific professional knowledge and managerial and entrepreneurial experience have been found to have a greater positive impact than education and working experience on the start-up size (Colombo et al., 2004).5 4Mata and Machado (1996) analyze a sample of 1079 new firms from Portugal. In their sample, not more than 25% have greater than the average size of 17 employees, and 50% of the firms employ less than 10 people. 5Colombo et al. (2004) investigate start-up size of 391 technology based young Italian firms in both manufacturing and services. 67 Görg and Strobl (2002) find that the presence of multinationals negatively effects the size of domestic Irish entrants. Astebro and Bernhardt (2005) show that entrepreneurial human capital of founders co-determines their household wealth and the firms start-up capital. According to (Colombo and Grilli, 2005), firms receiving external private equity financing have greater start-up size. Advertising costs and R&D expenditures are important in determining the start-up size of large firms than small firms (Arauzo-Carod and Segarra-Blasco, 2005). Nurmi (2006) studies sectoral differences in start-up size in Finland and finds that results for manufacturing and service sectors are very similar. In addition, some studies show that start-up size is higher when entrepreneurs receive inheritances (Holtz- Eakin et al., 1994). Evans and Jovanovic (1989) discover the presence of binding liquidity constraints that limit start-up capital of entrepreneurs. They find that “entrepreneurs are limited to a capital stock that is no more than about one and one-half times of their wealth.” Thus, almost all entrepreneurs in their sample “devote less capital to their business than they would like to.” (p. 825) As mentioned earlier, we hypothesize that start-up size is not independent of the geographic region. The growing literature of economic geography (Krugman, 1991; Fujita and Krugman, 2003) gives us compelling reasons to hypothesize that the spatial location should play an important role in determining the size of new start-ups. In particular, there are compelling reasons to posit that some regions give birth to firms with a greater start-up size while others lead to creation of very small firms. We also hypothesize that initial knowledge endowments of the firm and the ownership structure influence the start-up size. Entrepreneurs who possess technical knowhow are more likely to start with larger firms. Firms that have single proprietary ownership are more likely to be small compared to those that have partnership or co-operative ownership structures. 4.3 Geoadditive Models We use semiparametric regression techniques based on Bayesian P-Splines and geoadditive models for the empirical analysis. The method allows estimating the non-linearities of continuous variables and the neighborhood effects on the start- up size of new firms.6 A brief outline of the methodology is presented here. 6This section draws from Lang and Brezger (2004); Brezger and Lang (2005). 68 Essays on Entrepreneurship and Economic Development Dissertation zur Erlangung des wirtschaftswissenschaftlichen Doktorgrades der Wirtschaftswissenschaftlichen Fakultät der Universität Göttingen vorgelegt am 7. 9. 2007 von Jagannadha Pawan Tamvada aus New Delhi Eidesstattliche Erklärung Hiermit versichere ich an Eides statt, dass ich die eingereichte Dissertation Essays on Entrepreneurship and Economic Development selbständig verfasst habe. Anderer als der von mir angegebenen Hilfsmittel und Schriften habe ich mich nicht bedient. Alle wörtlich oder sinngemäß den Schriften anderer Autoren entnommenen Stellen habe ich kenntlich gemacht. Göttingen, den 7. September 2007, Jagannadha Pawan Tamvada i Publications The paper Religion and Entrepreneurship, co-authored with David Audretsch and Werner Boente, is published as a Center for Economic Policy Research (CEPR) Discussion Paper. The other papers in the dissertation are authored by me and have been presented at international conferences, doctoral colloquiums and fac- ulty seminars. The research work in this dissertation has been accepted for presentation at the First World Congress of Spatial Econometrics (Cambridge, 2007), the 44th European Regional Science Association’s Annual Congress (Paris, 2007), the International Council for Small Business Research (Finland, 2007), the IZA- World Bank Conference on Employment and Development (Bonn, 2007) and the Second Annual Max Planck Indian Institute of Science (IISc) International Conference on Entrepreneurship, Innovation and Economic Growth (Bangalore, 2007). The research in this dissertation has been presented at the Schumpeter Con- ference (Nice, 2006), the 20th Research in Entrepreneurship Conference (Brues- sels, 2006), the First Annual Max Planck India Workshop on Entrepreneurship, Innovation, and Economic Growth (Bangalore, 2006), Hellenic Workshop on En- trepreneurship and Productivity (Patras, 2006), the European Summer School in Industrial Dynamics (Corsica, 2006), the Babson Doctoral Consortium (Bloom- ington, 2006), Augustin Cournot Doctoral Days (Strasbourg, 2006), the Technol- ogy Transfer Society’s Annual Conference (Kansas City, 2005) and the G-Forum’s Annual Conference (Jena, 2006). The work has also been presented at internal seminars at the Max Planck Institute of Economics, Jena and at the Faculty of Economics, University of Göttingen. iv Acknowledgements I am greatly indebted to Prof. Stephan Klasen for giving me an opportunity to pursue doctoral studies in economics, a dream I had cherished since my high school days. Without his intellectual guidance, this dissertation would not have seen its completion. I am also grateful to Prof. David Audretsch, whose inspi- ration, support, and kindness have helped me to complete the task at hand. I owe enormous gratitude to him not just for giving me the prestigious Max Plank PhD scholarship, but also for introducing me to entrepreneurship and constantly guiding me through the scholarship. I also show my gratitude to Prof. Walter Zucchini for his guidance and encouragement. My discussions with my supervi- sors form the foundations of this dissertation, and I would like to express my gratitude for giving me their precious time and intellectual support. I am grateful to Prof. Amartya Sen for his inspiration not just for me, but for many young Indians to study economics. I thank him for sparing some of his valuable time for me when I visited him at Harvard. I also thank Professors TVS Ram Mohan Rao and Vishwanath Pandit for their constant guidance and support. I also thank Swami Supernanada, a monk of the Rama Krishna Mission order, who introduced me to this fascinating subject. I thank all my colleagues at the Max Planck Institute, especially Werner Boente and Max Keilbach. I would also like to thank Taylor Aldridge, Melanie Aldridge, Iris Beckman, Saradindu Bhaduri, Andreas Chai, Andrea Conte, Sameeksha Desai, Stephan Heblich, Anja Klaukien, Stefan Krabel, Adam Led- erer, Prashanth Mahagaonkar, Erik Monsen, Pamela Mueller, Holger Patzelt, Stephan Schütze, Jörg Zimmerman, and all my colleagues for making my stay in Jena both intellectually and personally rewarding. My special thanks go to Kerstin Schük, Madeleine Schmidt, and Lydia Nobis for their continuous efforts at making my stay at the Institute memorable. I also show my gratitude towards v Thomas Bauman and his IT team and Katja Müller and her library team at the Max Plank Institute for attending to all my academic requirements. I also thank Marten Koppenhagen, Thilo Klein, and particularly Alex Audretsch for providing valuable research assistance. Thus, I thank the Max Planck Institute of Economics and all its wonderful staff for providing an environment that is so conducive to research. I also would like to thank my colleagues at the Chair for Development Eco- nomics at Göttingen, particularly Micheal Grimm, Melanie Grosse, Isabal Gün- ther, Andrey Launov, Felicitas Nowak-Lehman Danzinger, Ken Harttgen, Mark Misselhorn, Jan Priebe, Dana Schüler, Sebastian Vollmer and Julian Wiesbrod. My special thanks also go to Michaela Beckmann. Roswitha Brinkmann and the international office of the Goettingen University have my great appreciation for giving me a scholarship during my stay there in 2004, as does Prof. Manfred Denker for accepting me as a member of the Center for Statistics in Göttingen. I give my sincerest thanks to the Lindau Nobel Council for selecting me to participate at the Second Lindau meeting of Nobel laureates, giving me an un- paralleled opportunity to meet and listen to some of the greatest living legends in the field of economics. I also extend my thanks to the Max Planck Society for giving me a grant to organize the First Annual Max Planck India Workshop on Entrepreneurship, Innovation, and Economic Growth in partnership with the Indian Institute of Science, Bangalore. This conference reaffirmed to me that fur- ther entrepreneurship research focusing on developing countries such as India is an absolute necessity to promoting their development. Finally, I would like to thank the Kauffman Foundation for sponsoring my participation at the Babson Doctoral Consortium. The Ministry of Small Scale Industries provided much assistance to me by providing me firm-level data, as did the Reserve Bank of India by inviting me for a research stay at the Rural Credit and Policy Department. I am greatly indebted to my parents for their love, affection, prayers, and blessings. In storm and in calm, they have stood by me. I am equally grateful to Sai Baba, who showed me that man is born not for pursuing self-interest, but to serve humanity. He stayed with me through-and-through. Like my parents, he always showered unconditional love and affection on me and inspired me to follow my heart and pursue this path. As his student in Prashanti Nilayam, I vi 3 Religion and Entrepreneurship 42 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.2 Religion, Entrepreneurship and the Indian Context . . . . . . . . 44 3.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 3.4 Empirical Analysis: Discrete Choice Models . . . . . . . . . . . . 49 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4 The Geography of Start-up Size 65 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.2 The Start-Up Size . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4.3 Geoadditive Models . . . . . . . . . . . . . . . . . . . . . . . . . . 68 4.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 4.5 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 72 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 5 Entrepreneurship and Welfare 87 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 5.2 Theoretical Background . . . . . . . . . . . . . . . . . . . . . . . 88 5.2.1 Occupation, Welfare and Economic Development . . . . . 88 5.2.2 Occupational Selection and Determinants of Welfare . . . . 90 5.3 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92 5.3.1 Quantile Regressions . . . . . . . . . . . . . . . . . . . . . 93 5.3.2 Selection Models for Multiple Outcomes . . . . . . . . . . 93 5.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95 5.5 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 96 5.5.1 Entrepreneurship and Welfare . . . . . . . . . . . . . . . . 96 5.5.2 Endogenous Non-random Occupational Selection . . . . . . 102 5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 6 The Dynamics of Entrepreneurship 122 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 6.2 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123 6.2.1 Repeated Cross Section Analysis . . . . . . . . . . . . . . 123 6.2.2 Pseudo Panel Approach . . . . . . . . . . . . . . . . . . . 124 6.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126 ix 6.4 Empirical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 127 6.4.1 Repeated Cross Sections . . . . . . . . . . . . . . . . . . . 127 6.4.2 Pseudo Panel Analysis . . . . . . . . . . . . . . . . . . . . 129 6.4.3 Reconciling the Results . . . . . . . . . . . . . . . . . . . . 133 6.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 7 Conclusion 151 7.1 Exogenous Constraints and Entrepreneurship . . . . . . . . . . . . 151 7.2 The Dual Theory of Entrepreneurship . . . . . . . . . . . . . . . . 152 7.2.1 Entrepreneurship, Start-Up Size, and the Spatial Location 154 7.2.2 A Simple Model . . . . . . . . . . . . . . . . . . . . . . . . 155 7.2.3 Entrepreneurship and Economic Development: The Dual Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 7.3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158 Bibliography 160 x List of Figures 2.1 Non-linear Effects of Age on Self-employment . . . . . . . . . . . 31 2.2 Spatial Effects on Self-employment Choice . . . . . . . . . . . . . 35 2.3 Spatial Effects in ‘Nonagriculture’ . . . . . . . . . . . . . . . . . . 36 2.4 Spatial Effects in ‘Agriculture’ . . . . . . . . . . . . . . . . . . . . 37 3.1 Entrepreneurship and Religion . . . . . . . . . . . . . . . . . . . . 54 3.2 Entrepreneurship and Caste System in Hinduism . . . . . . . . . 54 4.1 Spatial Effects in Model I . . . . . . . . . . . . . . . . . . . . . . 81 4.2 Spatial Effects in Model II . . . . . . . . . . . . . . . . . . . . . . 82 5.1 Consumption and Occupation(Un-normalised) . . . . . . . . . . . 108 5.2 Quantile Plots-Households . . . . . . . . . . . . . . . . . . . . . . 111 5.3 Quantile Plots-Households (continued) . . . . . . . . . . . . . . . 112 5.4 Occupation and Poverty Plots . . . . . . . . . . . . . . . . . . . . 114 5.5 Occupation and Inequality Plots at Median . . . . . . . . . . . . . 114 6.1 Non-linear Effect of Age on Self-employment (2000) . . . . . . . . 137 6.2 Non-linear Effect of Age on Self-employment (2004) . . . . . . . . 138 6.3 Spatial Effects on Self-employment Choice . . . . . . . . . . . . . 142 6.4 Spatial Effects in ‘Nonagriculture’ . . . . . . . . . . . . . . . . . . 143 6.5 Spatial Effects in ‘Agriculture’ . . . . . . . . . . . . . . . . . . . . 144 7.1 Entrepreneurship and Economic Development . . . . . . . . . . . 158 xi Chapter 1 Introduction Almost four decades ago, Baumol (1968, p. 71) proclaimed that “in a growth con- scious world I remain convinced that encouragement of the entrepreneur is the key to the stimulation of growth.” Entrepreneurship, however, remained hidden and elusive from the grasp of economists. Fortunately, in recent years, the economics of entrepreneurship emerged as a compelling subject, providing insights into the entrepreneurial processes. Bringing together this literature on entrepreneurship, Parker (2004, p. 1) notes that “entrepreneurship has only recently come to be regarded as a subject.” While the debate in scholarly community has still not conclusively accepted even the definition of entrepreneurship, a vast literature has emerged over the last two decades providing insights into the many facets of entrepreneurship. Though each such facet is incomplete by itself, together they offer a comprehensive understanding of the entrepreneurial choice, new firm for- mation and the role of entrepreneurship in economic growth. Reflecting a broad consensus that has emerged in recent times, Lazear (2002, p. 1) claims that “the entrepreneur is the single most important player in the economy.” This dynam- ically expanding subject, the economics of entrepreneurship, however, neglected entrepreneurship in less developed countries. This dissertation exploits recent ad- vances in Bayesian semiparametric methods and geoadditive models (Fahrmeir and Lang, 2001a) and large databases of individual and firm-level micro-data from India to provide fresh perspectives of the entrepreneurial processes and their relationship to economic development. This dissertation underlines the nexus be- tween the entrepreneur, the firm, and the region by emphasizing the role of the spatial location in simultaneously determining the entrepreneurship choice and 1 the size of new firms. The returns to occupational choice and the spatio-temporal dynamics of self-employment choice form another major part of this dissertation. The role of the caste system and religion in determining the entrepreneurship choice is studied, as such factors play a crucial and important role in determining the occupational choice in India. The theme of the second and third chapters is the determinants of self em- ployment and the role of exogenous constraints in occupational choice. While a vast literature has emerged examining the determinants of entrepreneurship, the role of spatial location and the neighborhood of an individual have rarely been considered as determinants of entrepreneurship choice. There are compelling rea- sons, however, to assume that such factors play an important role in shaping the occupational choice of people. Thus, in chapter 2, I analyze the role of geographic location as a micro-determinant of self-employment choice. I also study the im- pact of human capital accumulation on occupational choice in agricultural and nonagricultural sectors in India. In chapter 3, I analyze the role of religion as an exogenous constraint on the occupational choice of individuals. Recent studies (Iannaccone, 1998; McCleary and Barro, 2006a; Guisa et al., 2006) link religion with economy but the channels through which religion influences the economy are not examined by the existing literature. One such channel through which reli- gion might influence the economy is through entrepreneurship. Religions impose behavioral constraints and influence economic outcomes. For instance, the insti- tution of the caste system in Hindusim is likely to act as an exogenous constraint on the occupational choice of Hindus. In this paper, I examine the role of religion and class structures in promoting or inhibiting entrepreneurial behavior. The theme of the fourth chapter is the impact of ownership structure and geo- graphic location on the size of new entrants. In this chapter, I revisit the question of firm size at entry. A number of studies show that, for new entrants at least, the initial size influences growth and survival. The determinants of the size of firms at entry, however, remained under-researched and neglected in this discussion, for a long time. The few studies on start-up size show that the industry characteris- tics such as turbulence, minimum efficient scale, and industry growth (Mata and Machado, 1996; Mata, 1996) and human capital of entrepreneurs (Astebro and Bernhardt, 2005; Colombo et al., 2004; Colombo and Grilli, 2005), determine the start-up size of new firms. However, the role of spatial location on the start-up 2 size has never been studied. Chapter 4 incorporates ownership structure and ge- ographic location as micro-determinants of start-up size, using micro data from India. The theme of the fifth chapter is entrepreneurship and welfare. A growing body of literature identifies returns to self-employment in developed countries (Hamilton, 2000). Historically, the development economics literature has consid- ered self-employment in less developed countries, to be a part of the so-called informal sector (Harris and Todaro, 1970). More recently, a growing body of literature argues that the informal sector is a blend of a low-productive disad- vantaged sector and a voluntary competitive sector (Cunningham and Maloney, 2001; Fields, 2005; Günther and Launov, 2006). In chapter 5, I link occupational decisions of the household with a direct measure of welfare, per-capita consump- tion. Using quantile regressions, I estimate occupational choice as a determinant of welfare. Furthermore, using selection methods that allow for corrections af- ter multinomial logit estimation (Bourguignon et al., 2007), I test if a process of endogenous non-random selection determines the selection of individuals into different occupations. Thus, the underlying process of selection into occupations and subsequent returns in terms of welfare are examined to see whether peo- ple are compelled to opt for low-productivity self-employment or whether they voluntarily self-select based on their unobserved abilities, in a developing country. The theme of the sixth chapter is the evolution of the entrepreneurial choice over time and space. The post-liberalisation era of Indian economy has witnessed a surge in entrepreneurial activity. The dynamics of occupational choice in this context are not explored in the literature. Using two cross-sectional databases of the National Sample Survey Organization of India (NSSO) data, I examine the spatial dynamics of self-employment choice and in particular, the role of educa- tion as a determinant of entrepreneurship. In addition, using three surveys of the NSSO (1994-1995, 1999-2000 and 2004), I also construct a psuedo-panel (Deaton, 1997; Moffitt, 1993; Verbeek, 2006) to examine the dynamics of entrepreneurial activity in India. The final chapter constructs the dual theory of entrepreneur- ship, linking results of the chapters of this dissertation. This chapter showcases a coherent theory of self-employment, firm formation, and geographic location and concludes this dissertation. 3 minants of self-employment choice in one such growing economy, India, that has in recent years, experienced substantial leaps in both its entrepreneurial activity and growth rates. Household level data collected by the National Sample Survey Organization in 2004 are used for the empirical analysis. The effects of individual personal charac- teristics, educational background, household characteristics and non-linear effects of continuous covariates such as age and geographic location on the probability of being self-employed are jointly estimated using geoadditive models. The results suggest that outside of agriculture, educated individuals are more likely to be salaried employees while in the agricultural sector, educated individuals are more likely to be self-employed. Strong spatial patterns are observed and these are pri- marily attributable to spatial self-employment patterns in the agricultural sector. Consistent with earlier empirical studies on the determinants of entrepreneurship, the results suggest that Indian males, married and older citizens are more likely to be self-employed as well. The next section discusses the literature and states the hypotheses on the determinants of self-employment in a developing economy. The third and fourth sections describe the semiparametric geoadditive modeling techniques and the dataset. The fifth section presents the empirical analysis. The final section pro- vides conclusions and discusses possible avenues for future research. 2.2 Theoretical Background 2.2.1 Determinants of Self-employment Empirical research on occupational choice in developed economies suggests that individuals’ personal characteristics (Kihlstrom and Laffont, 1979; Evans and Leighton, 1989b) and regional factors (Georgellis and Wall, 2000) play an impor- tant role in influencing the entrepreneurial decisions. The decision of individuals to become entrepreneurs is generally modeled in terms of utility maximization, where the economic returns from entrepreneurship are compared to returns of wage employment (Lucas, 1978; Holmes and Schmitz Jr., 1990; Jovanovic, 1994). Individual-specific characteristics such as risk aversion (Kihlstrom and Laf- font, 1979), prior self-employment experience (Evans and Leighton, 1989b), edu- cation, human capital, and age (Zucker et al., 1998; Bates, 1990; Rees and Shaw, 6 1986; Blanchflower and Meyer, 1994) and personality characteristics (McCelland, 1964), are found to have an impact on an individual’s entrepreneurship choice. As Parker (2004, p. 106) succinctly summarizes the broadly agreed determinants of entrepreneurship, The clearest influences on measures of entrepreneurship (usually the likelihood or extent of self-employment) are age, labor market experi- ence, marital status, having a self-employed parent and average rates of income tax (all with positive effects). Greater levels of risk and higher interest rates generally have negative effects, although to date only a handful of studies have satisfactorily investigated the former. Region specific characteristics such as industry structure (Acs and Audretsch, 1989; White, 1982), unemployment rates (Blanchflower, 2000; Blanchflower and Oswald, 1998), local job layoffs (Storey and Jones, 1987), small business employ- ment (Reynolds et al., 1994) and public policy variables such as state retirement benefits (Blau, 1987), unemployment benefits (Carrasco, 1999), and adherence to welfare state principles (Fölster, 2002) are also found to influence occupational choice.4 2.2.2 Labor Markets in Developing Countries The disadvantage theory and the comparative advantage theory are two compet- ing theories of labor markets in developing countries. The disadvantage theory hypothesizes that people who are rationed out of the formal labor markets are compelled to take up self-employment or work as workers in household enterprises. Such people are considered to constitute the informal sector. Thus, beginning with the labor surplus model of Lewis (1954), the labor markets of developing countries are viewed as segmented dualistic markets along the formal-informal lines (also see Sen, 1966; Ranis and Fei, 1961; Harris and Todaro, 1970).5 4Other examples of studies analyzing the determinants of entrepreneurship include Evans and Jovanovic (1990) and Parker et al. (2005). 5 Lewis (1954) argued that if wage rate is determined competitively in the rural areas of a LDC then it will be below the subsistence levels. Harris and Todaro (1970) predicts that workers who migrate from rural to urban areas face unemployment and are forced to work in household enterprises at subsistence levels. Models of rural-urban migration following this line of thought hypothesize that the urban informal sector acts as a refuge for migrants and excess labor in urban areas are forced to take up low productivity self employment. 7 Many studies find evidence against these theories of low level subsisting self- employment in LDCs (Chiswick, 1976; Majumdar, 1981; Blau, 1986; Rosenzweig, 1980; Mohapatra et al., 2007).6 The comparative advantage theory, thus hypothe- sizes that individuals voluntarily choose employment in the so called informal sec- tor, when they perceive competitive opportunities there (Gindling, 1991; Magnac, 1991; Maloney, 2004).7 In this paper, we do not distinguish between the formal and the informal sectors for two reasons. First, Maloney (2004, p.1159) notes that, “as a first ap- proximation we should think of the informal sector as the unregulated, devel- oping country analogue of the voluntary entrepreneurial small firm sector found in advanced countries, rather than a residual comprised of disadvantaged work- ers rationed out of good jobs.” As most empirical research on the determinants of self-employment is based on data from the developed economies, the results will stand comparable to the results of earlier studies if we consider both the sectors together and treat the informal sector akin to the entrepreneurial small firm sector of the developed countries. Second, the other main purpose of the paper is to examine the determinants of self-employment choice in agriculture and nonagriculture in India through the lens of economic geography. Though the characteristics of the informal sector in a developing country are well debated in the literature, examining the determinants of self-employment in this light is an interesting avenue for future research. 2.2.3 Hypotheses: Determinants of Self-employment Though there are compelling reasons to posit that there are sectoral differences in self-employment choice, male, married and older individuals are more likely 6Blau (1985) positively tests for competitive labor markets in the nonagriculture sector in LDCs but finds negative selection into self-employment based on managerial ability in the farm sector. His results suggest that self-employed earn more than wage employees in urban areas whereas in rural areas the self-employed earn much less than the wage employees. 7More recently, a growing body of literature attempts to capture the heterogeneity within the informal sector. This strand of literature argues that the informal sector is a blend of both disadvantaged and competitive sectors (Cunningham and Maloney, 2001; Fields, 2005; Günther and Launov, 2006) and claims simultaneous presence of disadvantaged “lower” and voluntary “upper” tiers within the informal sector. Pratap and Quintin (2006) do not find any evidence for segmented labor markets in Argentina. Yamada (1996) finds evidence of voluntary self-selection and higher earnings in self-employment in informal sector in Peru. 8 This suggests that returns to salaried employment increase faster than returns to entrepreneurship as the per-capita income grows, and this makes individuals more risk averse and decreases their willingness to become entrepreneurs (also see Lucas, 1978). Thus, there are compelling reasons to posit that individuals who are more educated opt for salaried employment relative to self-employment in an LDC context (see Sluis et al., 2005, for a survey). Hence, we hypothesize that individuals with greater human capital might prefer salaried employment as opposed to self-employment. Another determinant of self-employment that is discussed in the literature is wealth. Wealth possessed by the individuals provides a degree of security for entering self-employment and helps them to ease their credit constraints.10 As Boháček (2006, p.2196) notes, In order not to default on loan contracts, entrepreneurs can borrow only limited amounts secured by collateral. This collateral (accumu- lated assets) guarantees not only the repayment of the loan but also positive consumption of the entrepreneur in the case of a project’s failure. As the financial constraint is endogenously related to a bor- rower’s wealth, entrepreneurship becomes positively correlated with wealth. Households with very high levels of wealth have a higher propensity to take risk (Carroll, 2000). Hurst and Lusardi (2004) argue that as households with higher levels of wealth have a higher tolerance for risk, they are most likely to be busi- ness owners.11 Blanchflower and Oswald (1998) find that inheritance increases the probability of self-employment. Banerjee and Neuman (1993) argue that wealth distribution determines the occupational structure. For these reasons, we hypoth- esize a positive relationship between household wealth and the entrepreneurship choice. Borjas and Bronars (1989) present differences in self-employment rates amongst racial minorities in US. They show that consumer discrimination af- 10Lindh and Ohlsson (1996) test if the presence of credit constraints inhibit people from becoming self-employed. Many other studies also find that credit constraints act as barriers to entry of individuals into self-employment (Evans and Jovanovic, 1989; Evans and Leighton, 1989b; Blanchflower and Oswald, 1998). 11However, Hurst and Lusardi (2004) find that the relationship between wealth and en- trepreneurship is flat over the majority of the wealth distribution. They discover a positive relationship only after the ninety-fifth percentile. They argue that the reason could be that capital needed for a start-up in the United States is relatively low (also see Bhidé, 2000). 11 fects the earnings of self-employed blacks and other minority communities, mak- ing them less likely to select into self-employment relative to whites. Some other studies find that self-employment is higher in minority communities (Clark and Drinkwater, 1998). In an Indian context, the presence of caste system leads us to hypothesize that individuals of the backward classes may have a lesser propensity to be self-employed. Based on insights from the theory of new economic geography (Krugman, 1991; Fujita and Krugman, 2003), we hypothesize that individuals in neighbor- ing regions exhibit similar occupational preferences and in some neighborhoods individuals are more likely to be self-employed than in others and that this effect is non-linear in shaping economic outcomes over space. The presence of many self-employed people in a wealthy neighborhood may induce others to choose self-employment. Thus, it may have an inducement effect on the local popula- tion. People in such regions are likely to be more entrepreneurial and risk loving. However, presence of many self-employed people in poor neighborhoods indicates that dearth of viable employment opportunities compells people to select into self-employment in such neighborhoods. 2.3 Bayesian Semiparametric Methodology Semiparametric regression technique based on Bayesian P-Splines and geoaddi- tive models is used for the empirical analysis. The methodology allows for the estimation of non-linear effects of the continuous variables and the neighborhood effects of spatial units on the probability of individuals selecting self-employment. A brief outline of the method is presented here.12 2.3.1 Geoadditive Models Let (yi, xi, vi) for i in {1,2,...N} describe a dataset of N observations. Let yi be the response variable and xi be a m-dimensional vector of continuous covariates and 12This section draws on Lang and Brezger (2004) and Brezger and Lang (2005). This method- ology has been applied earlier by Kandala et al. (2001) and Kandala et al. (2002) to examine the determinants of under-nutrition in African countries. 12 vi be a vector of categorical variables.13 Assume yi are independent and Gaussian with mean ηi = f1(xi1) + .... + fp(xip) + viγ, and a common variance σ2. If fi are unknown smooth functions of the continuous variables and viγ corresponds to the parametric part of the regression, the regression model is called the Additive Model or a Semiparametric regressor. Eilers and Marx (1996) use polynomial regression splines that are parameterized in terms of B-Spline basis functions, the P-Splines, in the context of an Additive Model, to estimate the smooth functions within the semiparametric framework. Fahrmeir and Lang (2001a,b) use simple random walk priors in a bayesian version of the Additive Model. Kammann and Wand (2003) introduce Geoadditive models within the Additive Mixed Model framework to deal with unobserved heterogeneity across different spatial units.14 Furthermore, Lang and Brezger (2004) and Brezger and Lang (2005) generalize the work of Fahrmeir and Lang (2001a,b) and develop the Bayesian version of the P-Spline approach of Eilers and Marx (1996), Bayesian P-Splines.15We use these methods in the empirical analysis. Assume that the unknown functions fj can be approximated by a l degree spline with equally positioned knots in the domain of xj (Eilers and Marx, 1996). By writing such a spline in the form of a linear combination of k B-Spline basis functions, Bjk, where k is equal to the number of knots plus the degree of the spline, fj(xj) = ΣβjkBjk and, in matrix notation, η = ΣXjβj + V γ. By defining a roughness penalty based on the differences of adjacent B-Spline coefficients, for ensuring smoothness of the estimated functions, the penalized likelihood assumes the form: L = l(y, β1, ....., βp, γ)− λ1Σ(4kβ1) 2 − .......λpΣ(4kβp) 2 (2.1) 13We first present the case of the gaussian response distribution and then show how the family of binomial probit models can be generalized to the family of gaussian response, using a link function. 14Generalized Additive Mixed Models (Lin and Zhang, 1999) for cases with unobserved het- erogeneity are extensions of Generalized Additive Models (Hastie and Tibshirani, 1990). For an overview of semiparametric regressions, see Fahrmeir and Tutz (2001). Additive Mixed Mod- els in the Bayesian framework have also been considered by Hastie and Tibshirani (2000) and Fahrmeir and Lang (2001a,b) but these approaches do not consider the unobserved heterogene- ity, the spatially correlated random effects. 15The difference penalties are replaced by Gaussian (intrinsic) random walk priors that serve as smoothness priors for the unknown regression coefficients. A related approach is the Bayesian smoothing splines methodology of Hastie and Tibshirani (2000). 13 tial patterns can be explained using one of the following econometric approaches. A simple strategy is to regress the mean residual spatial effects on the regional variables. Thus, after estimating the geoadditive model, the total spatial effect of each region is explained by regressing the posterior mean of the estimated spatial residual effect on the regional variables. However, this empirical strategy does not consider the estimated posterior variance of spatial effects. In order to overcome this problem, a discrete choice model of the 95% or 80% spatial effects can be estimated. In this case, a variable is constructed that takes a value of (-1) when the region has a significant negative effect, takes a value of (0) if the effect is insignificant and takes a value of (1) if the effect is significant and positive. This leads to a straightforward multinomial specification. This variable is then regressed on the regional variables. We employ both strategies to examine the determinants of the residual spatial patterns. 2.4 Data The data used for the analysis is the 60th round employment-unemployment sur- vey of the National Sample Survey Organization (NSSO) of India conducted in 2004. As the focus of the paper is on economically active individuals, we restrict the sample to those who are older than 15 years but younger than 70 years. This reduces the sample size from 303,811 to 204,298.16 While the principal economic activity of this sample ranges from domestic duties to full time employment (in the form of salaried employment, self-employment, casual labor or unemploy- ment), 17% of the individuals in this sample are engaged in subsidiary activities. For the rest of the analysis, we consider the principal economic activity alone for two reasons. First, all individuals are not engaged in subsidiary activities. Second, as less than one sixth of the entire sample are engaged in subsidiary activities, considering such activities would further complicate the analysis when individu- als report as both self-employed and paid employees. Furthermore, the principal economic activity is the activity to which the individuals devote most of their time. For these reasons, we consider only the primary occupation for classifying workers into self-employment and paid employment. Table 2.1 lists the number of 16We drop 17 individuals who adhere Zoharastrianism for reasons of consistency with the next chapter. 16 individuals in different occupational categories. We also drop individuals who are unpaid family workers, students, workers involved in domestic duties, pensioners, those who are unable to work due to disabilities and people who reported to belong to the occupational class ‘other’. This reduces the final sample to 88,623 economically active individuals.17 We thus only consider those who have reported their primary occupation as self-employed (includes own account workers and em- ployers), salaried employees, casual laborers, or unemployed.18 The descriptive statistics in Table 2.2 show that 65% percent of the individuals have attended at least primary school, 65% live in rural areas and 40% are in the agricultural sector. Table 2.3 presents the descriptive statistics of self-employed and others in agricultural as well as nonagricultural sectors. Self-employed are older in both sectors. 13% of the self-employed in nonagriculture have university education compared to 3.7% of those who are self-employed in agriculture. A higher proportion of educated individuals are self-employed in agriculture and a higher proportion of educated individuals are salaried employees in nonagricul- ture. In the absence of an appropriate measure for wealth, we proxy it using the land-possed by the household. We thus posit that individuals who own large areas of land are more likely to be self employed. While in agriculture, land enables self-employed farming, and this makes people to choose self-employment over other modes of occupation, in the nonagricultural sector, land serves as potential collateral to obtain credit for starting an enterprise.19 These descriptive tables also show that more than 50% of individuals in agri- 1721.91% of these individuals are engaged in some subsidiary economic activity but for reasons listed earlier, we only consider the primary occupation in classifying individuals as self-employed workers or paid employees. 18We merge the occupations into self-employment and paid-employment for the rest of the analysis in this chapter. In the next chapter, we consider the four occupational categories as distinct classes. 19On the one hand, self-employed individuals in agriculture may possess more land as they need it for agricultural purposes. On the other hand, only those who possess land may be able to choose self-employment. Thus, the land possessed is also likely to determine the self-employment status. Hence the problem of endogeneity with respect to land even in the agricultural sector may not be so severe. The dataset has some information on the purchases made on the some durable commodities for some households. However, the information is missing for a number of households and for a number of items in the representative consumption bundle. Hence, we are not in a position to use this data. Furthermore, as income data is not available for the majority of individuals in the sample, we are not able to instrument the land possessed using income data. 17 culture are self-employed in comparison to a relatively lower proportion in nona- griculture. The presence of agricultural sector in the data poses several problems in analyzing the determinants of self-employment. The farm sector is usually found in rural areas with mainly farmers as self employed individuals. There are compelling reasons to posit that they are different from self-employed individuals in nonagriculture. As some scholars have noted before, the process of economic development reduces participation in farm sector and this induces a bias when analyzing the changes in self-employment rates with time if the agricultural sector is included in the analysis (Parker, 2004).20 Researchers have usually analyzed the determinants of self-employment only in the non-farm sector in order to get around these problems. As the farm sector is very important in a developing country like India, we also study self-employment in this sector. 2.5 Empirical Analysis In order to use the entire data set on hand and to make robust inferences on the determinants of self-employment, three different models are estimated. 2.5.1 Aggregate Model In the first model, participation in the agricultural sector is controlled using a dummy variable. The following semiparametric geoadditive probit model is estimated: η = γconst + γfemale + γmarital_status + γeducation_general + γeducation_technical + γwealth+γurban+γagri+γhindu+γbackward+fage+fspatial(district)+frandom(district) The non-linear effect of age is modeled as third degree P-Spline with second order random walk penalty.21 Figure 2.1(a) shows that the probability of being 20However, as our study is cross-sectional and does not analyze self-employment rates over time, this limitation does not apply here. Furthermore, we analyze the determinants of self- employment in agriculture and nonagriculture separately. 21The number of equidistant knots is assumed to be 20. The structured spatial effects are estimated based on Markov random field priors and random spatial effects are estimated with gaussian priors. The variance component in all the cases are estimated based on inverse gamma priors with hyperparameters a=0.001 and b=0.001. The number of iterations is set to 110000 with burnin parameter set to 10000 and the thinning parameter set to 100. The autocorrelation files and the sampling paths show that the MCMC algorithm has converged. These plots are available from the author. 18 2.5.2 Sector Specific Models Agricultural and Nonagricultural Self-employment The first model assumes that the determinants of self-employment are same for all self-employed individuals in agricultural as well as nonagriculture. In order to examine the differences in the two sectors, the following semiparametric model is estimated for individuals in agricultural and nonagricultural sectors separately: η = γconst + γfemale + γmarital_status + γeducation_general + γeducation_technical + γwealth + γurban + γhindu + γbackward + fage + fspatial(district) + frandom(district) The parameters for a, b, the number of iterations, burnin, and the thinning parameter are set equal to the first model’s parameters.29 The relationship of age with self-employment is very close to being linear in the agricultural sector, as seen in Figure 2.1(e), while in the nonagricultural sector, as Figure 2.1(c) shows, the age function increases at a decreasing rate until the age of 55 years and then increases at an increasing rate. Table 2.5 and Table 2.6 show considerable differences in relative human capital endowments of self-employed individuals in the two sectors. While in the agricultural sector, those who are endowed with higher levels of human capital (proxied by age and education) are more likely to be self employed, in the nonagricultural sector such individuals are more likely to be salaried employees. Belonging to a backward class is significantly negatively related to being self-employed in both the sectors, and being a Hindu has a significant negative relationship only in nonagriculture. For people in nonagriculture, as maps in Figure 2.3 suggest, the north-south divide seen in the spatial effect on the self-employment choice for individuals in the aggregate model is less pronounced. People of Kerala and some districts of Tamil Nadu in the south, Maharastra and Madhya Pradesh in western and central parts of India, and the majority of districts in the north-eastern states are less likely to be self-employed. People living in Uttar Pradesh, Bihar, Rajasthan, some districts of Andhra Pradesh, and West Bengal are more likely to be self-employed. The maps of spatial effects in agriculture in Figure 2.4 show that the result of north-south spatial divide observed in the first model can be attributed mainly to such a phenomenon in the agricultural sector. In sharp contrast to some districts in the western and the northern parts of India, people are very less likely to be 29The autocorrelation files and plots of the sampling paths show that sufficient convergence is achieved in these models also. 21 self-employed in agriculture in southern and central states. As Figures 2.3(b) and 2.4(b) demonstrate, the unstructured random effects are negligible compared to the structured spatial effects. The confidence interval plots for the random spatial effects also show that the local effects are small and insignificant compared to the effects of structured spatial effects in all the three estimated models.30 2.5.3 Determinants of Residual Spatial Patterns The presence of spatial patterns, as shown by the empirical analysis, suggests that it is not just personal characteristics of individuals that totally explain their occupational choice. As discussed below, regional characteristics also play an important role in determining self-employment choice. In particular, financial constraints, level of economic development, unemployment and small business employment are found to influence the self-employment rates in a region by earlier studies. Hence, we hypothesize that these variables can explain the residual spatial patterns. We follow the empirical approach described in subsection 2.3.3. Holtz-Eakin et al. (1994) test the role of liquidity constraints in the formation of new enterprises. Their analysis suggests that the size of inheritance has an effect on entrepreneurial choice and also on investment in the capital of a new enterprise. Many studies find that credit constraints are barriers to entry for individuals into self-employment (Evans and Jovanovic, 1989; Evans and Leighton, 1989b; Blanchflower and Oswald, 1998). Lindh and Ohlsson (1996) test for the presence of credit constraints as inhibitors to self-employment, by seeing if those who win a lottery are more likely to enter self-employment. They also find that such individuals start firms with higher capital. Cabral and Mata (2003) find that the presence of binding financial constraints inhibit firms from growing to their optimal size. Hence, we hypothesize that the level of financial development in the region, measured by the per-capita credit or the credit-deposit ratio in a district can explain the residual spatial pattern. Lucas (1978) predicts that entrepreneurship decreases with economic devel- opment. Calvo and Wellisz (1980) show that the growth rate of total stock of knowledge requires greater ability of the marginal entrepreneur in a steady state equilibrium. This suggests that, given a fixed ability distribution in a population, the number of entrepreneurs decreases and average firm size increases with tech- 30These plots are available from the author. 22 nological progress. Empirical studies of Acs et al. (1994) and Fölster (2002) find that per-captia gross net product (GNP) is negatively related to self-employment. Acs et al. (1994) argue that self-employment decreases in the early stages of de- velopment as technological change shifts output from agriculture and small scale industry to large scale manufacturing. We thus hypothesize that level of economic development determines the propensity to be self-employed in a region. Cross-sectional evidence gives a mixed impression about the effect of unem- ployment on the propensity to be self-employed. The recession-push hypothesis claims that high unemployment decreases the probability of getting paid employ- ment and thus pushes individuals into self-employment. However, the prosperity- pull hypothesis suggests that high unemployment reduces demand for goods and services of the self-employed, leading to a reduction in self-employment. Many cross-sectional studies find a negative relationship between unemployment and the probability of self-employment (Taylor, 1996; Blanchflower and Oswald, 1998). However, many studies also indicate that the self-employed experience a spell of unemployment (Evans and Leighton, 1989b; Blanchflower and Meyer, 1994). As Storey (1991) notes, time series studies show a positive relationship but cross-sectional studies suggest a negative relationship. Hence we hypothesize that unemployment could explain the residual self-employment pattern. We also introduce a number of demographic controls. In particular, we control for size of the district and the population density. Armington and Acs (2002) sug- gest that these factors play an important role in explaining the spatial patterns of new firm formation. We also control for agglomeration, measured by the density of firms in the region, as presence of a large number of firms in the neighbor- hood is likely to result in spillovers that induce new firm formation. As Krugman (1991, p. 484) notes, “the concentration of several firms in a single location offers a pooled market for workers with industry-specific skills, ensuring both a lower probability of unemployment and a lower probability of labor shortage.” Further- more, as Armington and Acs (2002, p.38) argue, “informational spillovers give clustered firms a better production function than isolated producers have. The high level of human capital embodied in their general and specific skills is another mechanism by which new firm start-ups are supported.” Thus regions with high agglomeration are more likely to be associated with higher probability of people entering self-employment. 23 the determinants in urban and rural areas. We control for regional effects using a set of state level regional dummies. We estimate this for the sub-sample of individuals in the nonagricultural sector alone.33 We also check the robustness of the estimates, with respect to the presence of land variables, by running separate regressions with and without land variables. We estimate the regressions with the land variables excluded in the first specification and land variables included in the second specification (Table 2.10). However, the regression estimates for the two specifications are not very different. It can be argued that in the Indian context, wealth plays a definite role in self-employment choice. As argued earlier, this is possible if credit is rationed in favor of individuals possessing assets such as land. We interpret the results of the specification with the land variables, as Table 2.10 suggests that the estimates of models with and without them are similar. The results are broadly consistent with results of the semi-parametric estima- tion. The estimated signs of higher education variables are negative in rural as well as urban areas. The absolute value of the coefficients are, however, slightly higher in the rural areas suggesting that educated people in the rural areas have a still lower propensity for self-employment. The returns to self-employment in ru- ral areas may be lower in comparison to the returns to self-employment in urban areas and this could explain this result. This issue is analyzed more extensively in chapter 5. While technical education is insignificant in rural estimations, it is significant and negative in urban regressions. The land variables are positive and increase the propensity to be self-employed in rural and urban areas. However, the coefficients are larger in urban areas, indicating that people in urban areas with more land have a higher propensity to choose self-employment. This may be because land in urban areas is more expensive relative to land in rural areas. This has a direct implication for obtaining credit from financial institutions. The esti- mates of the religion and caste variables are consistent with the semi-parametric model for the nonagricultural sector estimated earlier and the coefficients are significant and negative. The absolute value of the coefficient of the ‘Hindu’ vari- able is larger in the urban regression than in the rural regression equation. This is counter intuitive to some degree, because cultural institutions responsible for lower likelihood of Hindus and individuals of backward classes to be self-employed are expected to be stronger in rural areas. A plausible explanation is that individ- 33As the agricultural sector is mostly found in the rural areas only, we restrict the urban-rural analysis to the nonagricultural sector. 26 uals of other religions face greater discrimination in urban areas when it comes to wage-employment. Thus the probability of Hindus entering wage-employment may be higher in urban areas. 2.6 Conclusion The field of entrepreneurship in economics provides insights into the individual determinants of the self-employment choice in developed countries. We contribute to one aspect of this literature that remained neglected for a long time. We use recent advances in Bayesian semiparametric methodologies to examine the spa- tial as well as individual determinants of self-employment choice in a developing country, India. Consistent with studies based on datasets from developed coun- tries, we find age to have a non-linear relationship with the probability to be self-employed, particularly in nonagriculture. A clear jump after the age of 55 is noticed, which could be a direct result of the retirement effect. The effect is linear and monotonically increasing in agriculture. Married individuals are more likely to be self-employed in both sectors. In nonagriculture, educated people are less likely to be self-employed while in agriculture, they are more likely. The results are consistent with empirical studies of developed economies and also shed light on the unexplored agricultural self-employment in a developing country context. The analysis further suggests that in the nonagriculture, self-employed people are more or less uniformly distributed across different spatial units but in agri- culture self-employed individuals are concentrated in certain geographic pockets. In both sectors, the regions with the highest propensity of self-employment are the states of Uttar Pradesh and Bihar. While it can be argued that these regions are more entrepreneurial, these regions are also the poorest regions in India, in terms of per-capita income and human development. This leads to an important conclusion that self-employment in Indian context may actually support the view that self-employment in a fast growing economy like India continues to be the main occupational option in the poorest neighborhoods and not for individuals with high human capital. Furthermore, an analysis of the determinants of nona- gricultural self-employment in rural and urban areas suggests that in rural areas educated individuals have still lower propensity to become self-employed. 27 Ta bl e 2. 1: D is tr ib ut io n of O cc up at io n To ta lN um be r P er ce nt ag e C um ul at iv e Se lf- em pl oy ed (O w n A cc ou nt W or ke rs ) 37 ,1 97 18 .2 1 18 .2 1 Se lf- em pl oy ed (E m pl oy er s) 92 2 0. 45 18 .6 6 H ou se ho ld H el pe rs (U np ai d Fa m ily W or ke r) 23 ,5 05 11 .5 1 30 .1 6 Sa la ri ed E m pl oy ee s 21 ,2 23 10 .3 9 40 .5 5 C as ua lL ab or (P ub lic ) 31 0 0. 15 40 .7 0 C as ua lL ab or (O th er ) 23 ,8 23 11 .6 6 52 .3 6 U ne m pl oy ed 5, 14 8 2. 52 54 .8 8 St ud en ts 25 ,8 53 12 .6 5 67 .5 4 O nl y D om es ti c D ut ie s 40 ,8 94 20 .0 2 87 .5 6 D om es ti c D ut ie s an d C ol le ct io n of W oo d et c. 18 ,0 45 8. 83 96 .3 9 P en si on er s 2, 64 5 1. 29 97 .6 8 N ot w or ki ng du e to di sa bi lit y 1, 38 1 0. 68 98 .3 6 B eg ga rs an d P ro st it ut es 33 52 1. 65 10 0 To ta l 20 4, 29 8 10 0 28 15 28.5 42 55.5 69 -1.04 -0.5 0.05 0.59 1.13 Effect of age age (a) Posterior mean of the non-linear ef- fect of ‘age’ together with 95% and 80% pointwise credible intervals in the Ag- gregate Model. 15 28.5 42 55.5 69 -0.023 0.01 0.043 0.075 0.108 Derivative of Effect of age age (b) Derivative of the posterior mean of the ‘age’ function with 95% and 80% pointwise credible intervals in the Ag- gregate Model. 15 28.5 42 55.5 69 -0.95 -0.41 0.13 0.67 1.21 Effect of age age (c) Posterior mean of the non-linear ef- fect of ‘age’ together with 95% and 80% pointwise credible intervals in Nonagri- culture. 15 28.5 42 55.5 69 -0.055 -0.0010 0.053 0.106 0.16 Derivative of Effect of age age (d) Derivative of the posterior mean of the ‘age’ function with 95% and 80% pointwise credible intervals in Nona- griculture. 15 28.5 42 55.5 69 -1.27 -0.64 -0.01 0.61 1.24 Effect of age age (e) Posterior mean of the non-linear ef- fect of ‘age’ together with 95% and 80% pointwise credible intervals in Agricul- ture. 15 28.5 42 55.5 69 -0.041 -0.0050 0.031 0.067 0.102 Derivative of Effect of age age (f) Derivative of the posterior mean of the ‘age’ function with 95% and 80% pointwise credible intervals in Agricul- ture. Figure 2.1: Non-linear Effects of Age on Self-employment 31 Table 2.4: Determinants of Self-employment Variable Mean Std. Dev. 2.5%-Qt. 97.5%-Qt. Personal Characteristics Female -0.398 0.014 -0.426 -0.372 Married 0.175 0.018 0.141 0.211 Divorced 0.317 0.029 0.259 0.376 General Education Informal 0.265 0.019 0.227 0.304 Primary School 0.332 0.014 0.304 0.360 High School 0.193 0.016 0.163 0.224 University -0.181 0.020 -0.218 -0.141 Technical Education Technical Degree -0.127 0.057 -0.232 0.016 Technical Diploma -0.117 0.026 -0.168 -0.068 Land Possessed 0.2< Land <0.4 Hectares 0.149 0.014 0.120 0.176 0.4< Land < 2 Hectares 0.791 0.017 0.758 0.824 Land > 2 Hectares 1.180 0.024 1.132 1.226 Location Urban 0.253 0.013 0.227 0.279 Agriculture 0.336 0.013 0.312 0.361 Religion & Social Group Hindu -0.205 0.014 -0.233 -0.179 Backward -0.183 0.012 -0.206 -0.160 Constant -0.545 0.027 -0.599 -0.492 N 86140 Deviance(Mean) 93422.587 Std. Dev. 36.196992 deviance(µ̄) 92973.92 pD 448.66642 DIC 93871.253 Notes: Dependent variable is binary self-employment status of the indi- vidual. Base categories for marital status, general education, technical education, land dummies are unmarried, no general education, no tech- nical education and less than 0.2 hectares of land respectively. 32 Table 2.5: Determinants of Self-employment in Nonagriculture Variable Mean Std. Dev. 2.5%-Qt. 97.5%-Qt. Personal Characteristics Female -0.256 0.018 -0.290 -0.221 Married 0.203 0.019 0.165 0.240 Divorced 0.218 0.042 0.137 0.298 General Education Informal 0.141 0.028 0.085 0.195 Primary School 0.130 0.021 0.086 0.169 High School -0.039 0.022 -0.078 0.004 University -0.349 0.024 -0.395 -0.301 Technical Education Technical Degree -0.109 0.057 -0.217 0.004 Technical Diploma -0.134 0.025 -0.183 -0.084 Land Possessed 0.2< Land <0.4 Hectares 0.151 0.015 0.122 0.181 0.4< Land < 2 Hectares 0.112 0.022 0.070 0.153 Land > 2 Hectares 0.160 0.033 0.097 0.222 Location Urban 0.029 0.015 0.001 0.059 Religion & Social Group Hindu -0.180 0.016 -0.213 -0.149 Backward -0.150 0.014 -0.179 -0.121 Constant -0.222 0.031 -0.282 -0.163 N 51674 Deviance(Mean) 60166.724 Std. Dev: 34.978124 deviance(µ̄) 59807.524 pD 359.20045 DIC 60525.925 Notes: Dependent variable is binary self-employment status of the indi- vidual. Base categories for marital status, general education, technical education, land dummies are unmarried, no general education, no tech- nical education and less than 0.2 hectares of land respectively. 33 -0.745864 0 0.681658 (a) Structured Non-linear Effect of ‘District’. Shown are the posterior means. -0.0960833 0 0.112093 (b) Unstructured Random Effect of ‘District’. Shown are the posterior means. (c) Non-linear Effect of ‘District’. Pos- terior probabilities for a nominal level of 95%. Black denotes regions with strictly negative credible intervals, white denotes regions with strictly pos- itive credible intervals. (d) Non-linear Effect of ‘District’. Posterior probabilities for a nominal level of 80%. Black denotes regions with strictly negative credible inter- vals, white denotes regions with strictly positive credible intervals. Figure 2.3: Spatial Effects in ‘Nonagriculture’ 36 -1.89128 0 2.67315 (a) Structured Non linear Effect of ‘District’. Shown are the posterior means. -0.240888 0 0.202827 (b) Unstructured Random Effect of ‘District’. Shown are the posterior means. (c) Non–linear Effect of ‘District’. Posterior probabilities for a nominal level of 95%. Black denotes regions with strictly negative credible inter- vals, white denotes regions with strictly positive credible intervals. (d) Non–linear Effect of ‘District’. Posterior probabilities for a nominal level of 80%. Black denotes regions with strictly negative credible inter- vals, white denotes regions with strictly positive credible intervals. Figure 2.4: Spatial Effects in ‘Agriculture’ 37 Ta bl e 2. 7: D et er m in an ts of Sp at ia lP at te rn s in F ig ur e 2. 2, F ig ur e 2. 3 an d F ig ur e 2. 4 A ll N on ag ri cu lt ur e A gr ic ul tu re F in an ci al D ev el op m en t P er -c ap it a C re di t 0. 00 62 2 -0 .0 18 3 0. 02 75 (0 .0 16 ) (0 .0 12 ) (0 .0 44 ) C re di t- D ep os it R at io -0 .1 02 ** * 0. 04 36 ** -0 .4 02 ** * (0 .0 23 ) (0 .0 18 ) (0 .0 61 ) E co no m ic D ev el op m en t P er -C ap it a N SD P -0 .3 10 ** * -0 .2 68 ** * -0 .2 91 ** * -0 .3 25 ** * -0 .4 18 ** * -0 .2 53 ** * (0 .0 34 ) (0 .0 30 ) (0 .0 26 ) (0 .0 23 ) (0 .0 91 ) (0 .0 78 ) U ne m pl oy m en t -0 .0 60 3* ** -0 .0 47 1* ** 0. 04 06 ** * 0. 03 69 ** * -0 .2 91 ** * -0 .2 39 ** * (0 .0 16 ) (0 .0 15 ) (0 .0 12 ) (0 .0 12 ) (0 .0 41 ) (0 .0 40 ) D em og ra ph ic s M id Si ze D is tr ic t 0. 00 32 5 0. 01 41 0. 08 69 ** * 0. 08 19 ** * -0 .1 91 ** -0 .1 47 * (0 .0 30 ) (0 .0 29 ) (0 .0 22 ) (0 .0 22 ) (0 .0 79 ) (0 .0 76 ) La rg e D is tr ic t 0. 02 80 0. 03 05 0. 07 50 0. 07 19 -0 .1 76 -0 .1 61 (0 .0 91 ) (0 .0 90 ) (0 .0 68 ) (0 .0 68 ) (0 .2 4) (0 .2 3) P op ul at io n D en si ty -0 .0 18 9 -0 .0 18 3 0. 05 94 ** * 0. 05 54 ** * -0 .1 29 ** * -0 .1 29 ** * (0 .0 15 ) (0 .0 14 ) (0 .0 11 ) (0 .0 11 ) (0 .0 40 ) (0 .0 37 ) A gg lo m er at io n In de x F ir m D en si ty -0 .0 07 67 -0 .0 02 13 -0 .0 02 90 -0 .0 08 74 -0 .0 17 9 0. 00 45 3 (0 .0 13 ) (0 .0 12 ) (0 .0 09 4) (0 .0 09 0) (0 .0 33 ) (0 .0 31 ) C on st an t 2. 92 6* ** 2. 53 4* ** 2. 61 8* ** 2. 74 9* ** 4. 28 0* ** 2. 77 8* ** (0 .3 5) (0 .3 6) (0 .2 7) (0 .2 7) (0 .9 5) (0 .9 3) O bs er va ti on s 53 4 53 4 53 1 53 1 53 2 53 2 R 2 0. 20 0. 23 0. 40 0. 40 0. 16 0. 22 F 19 .0 8 22 .4 6 49 .4 3 50 .3 0 14 .1 1 21 .4 2 R 2 A dj us te d 0. 19 2 0. 22 0 0. 39 0 0. 39 4 0. 14 7 0. 21 2 N ot es : *S ig ni fie s p< 0. 05 ; ** Si gn ifi es p< 0. 01 ; ** * Si gn ifi es p< 0. 00 1. St an da rd er ro rs ar e re po rt ed in pa re nt he se s. D ep en de nt va ri ab le is th e m ea n sp at ia le ffe ct pe r di st ri ct af te r es ti m at io n of th e ge oa dd it iv e m od el s. 38 Table 2.10: Self-employment in Nonagriculture Rural and Urban Regressions Model I Model II Independent Var. Rural Urban Rural Urban Personal Characteristics Age 0.0298*** 0.0332*** 0.0294*** 0.0335*** (0.0052) (0.0049) (0.0052) (0.0049) Age Square -0.0224*** -0.0229*** -0.0221*** -0.0239*** (0.0064) (0.0059) (0.0065) (0.0060) Female -0.232*** -0.275*** -0.231*** -0.276*** (0.027) (0.024) (0.027) (0.024) Married 0.252*** 0.298*** 0.255*** 0.302*** (0.028) (0.027) (0.028) (0.027) Divorce/Widow 0.376*** 0.250*** 0.380*** 0.268*** (0.061) (0.053) (0.061) (0.053) General Education Informal Education 0.175*** 0.0874** 0.170*** 0.0799** (0.038) (0.040) (0.038) (0.040) Primary School 0.159*** 0.0759*** 0.155*** 0.0614** (0.027) (0.028) (0.027) (0.029) High School -0.0540* -0.0248 -0.0567** -0.0510* (0.028) (0.029) (0.029) (0.030) Diploma/University Education -0.410*** -0.278*** -0.412*** -0.317*** (0.036) (0.032) (0.036) (0.032) Technical Education Technical Degree 0.168 -0.211*** 0.164 -0.220*** (0.12) (0.063) (0.12) (0.063) Technical Diploma 0.0251 -0.205*** 0.0262 -0.208*** (0.042) (0.033) (0.042) (0.033) Household Controls 0.2< Land <0.4 Hectares 0.117*** 0.166*** (0.027) (0.018) 0.4< Land <2 Hectares 0.0603** 0.226*** (0.030) (0.043) Land >2 Hectares 0.113*** 0.344*** (0.041) (0.066) Hindu -0.128*** -0.237*** -0.128*** -0.238*** (0.024) (0.020) (0.024) (0.020) Backward -0.117*** -0.157*** -0.119*** -0.157*** (0.021) (0.018) (0.021) (0.018) Total Observations 23916 28611 23895 28589 Log Likelihood -14191 -16930 -14169 -16865 LR (χ2) 2472 2685 2492 2789 Degrees of freedom 47 47 50 50 Pseudo R2 0.0801 0.0735 0.0808 0.0764 Notes: Probit estimation. *Signifies p< 0.05; ** Signifies p<0.01;*** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is ‘selfemployed’. State dummies are included in all the regressions and are not reported here. The coefficients of the constant are not reported. 41 Chapter 3 Religion and Entrepreneurship While considerable concern has emerged about the impact of religion on economic de- velopment, little is actually known about how religion impacts the decision making of individuals. This chapter examines the influence of religion on the decision for people to become an entrepreneur. Based on a large-scale data set of nearly ninety thousand workers in India, this chapter finds that religion shapes the entrepreneurial decision. In particular, some religions, such as Islam and Christianity, are found to be more con- ducive to entrepreneurship than Hinduism. In addition, the caste system is found to influence the propensity to become an entrepreneur. Individuals belonging to a back- ward caste exhibit a lower propensity to become an entrepreneur. Thus, the empirical evidence suggests that both religion and the tradition of the caste system influence entrepreneurship, suggesting a link between religion and economic behavior. 3.1 Introduction Religion and economics have had a tenuous relationship. On the one hand, schol- ars dating back at least to Adam Smith and Max Weber have argued that religion plays a fundamental role in shaping economics.1 On the other hand, only scant attention has recently been given as to how and why religion might influence eco- nomics. The omission of religion as a determinant of economic activity is startling, given the recent suggestion by Iannaccone (1998, pp. 1492) that “the economics of religion will eventually bury two myths - that of homo economicus as a cold 1Anderson (1988, p. 1068) notes, “In Wealth, Smith was not interested in theological issues or even in the nature of religious belief. Rather, he was concerned with two basic problems: (1) the economic incentives involved in the individual’s decision to practice religion and (2) the economic effects of different systems of religious belief as reflected in individual behavior. He did not attempt to develop an economic theory of the emergence of religious beliefs... Smith attempted the more limited task of defining the logical economic consequences of certain kinds of religious beliefs.” 42 creature with neither need nor capacity for piety, and that of homo religiosus as a benighted throwback to pre-rational times.” Moreover, as Edmund Phelps argues, “values and attitudes are as much a part of the economy as institutions and policies are. Some impede, others enable.”2 In India, for instance, Hinduism is strongly associated with the emergence of the caste system. Although some aspects of the caste system such as untoucha- bility, were abolished by the government, it remains formidable and imposing in practice. There remains a heated public debate in India on the impact of the caste system on the economic status of what is widely referred to “backward classes”. For example, in an article announcing, “Indian College Quota Law Suspended”, The New York Times reports that, “Caste discrimination is outlawed but contin- ues to persist in obvious and subtle ways, and the contest over the latest university admissions quotas revolve around how to best redress an entrenched and often ugly social bias.”3 Recent studies suggest the existence of a relationship between religion and economic performance (Barro and McCleary, 2003; McCleary and Barro, 2006b; Guisa et al., 2006). For example, Barro and McCleary (2003) estimate the im- pact of adherence to religious beliefs on economic performance using international survey data on religiosity. They find that increases in church attendance tend to reduce economic growth while increases in the belief in hell and an afterlife in- crease economic growth. These empirical findings raise several important but unanswered questions: (1) What are the channels by which religion influences economic activity? and (2) Is the impact of religion on economic activity homo- geneous across all religions? The purpose of this paper is to shed light on these questions by examining whether religion has any impact on one particular channel of economic decision- making influencing economic growth – the decision to become an entrepreneur. Recent studies suggest that entrepreneurship may be a key factor generating growth and development (Baumol, 2002). As Lazear (2002, p. 1) concludes, “The Entrepreneur is the single most important player in a modern economy.” Lazear’s conclusion is supported by considerable theoretical and empirical literature link- ing entrepreneurship to economic growth.4 2“It’s All About Attitude,” Newsweek International Edition, 30 April, 2007. 3“India College Law Suspended,” The New York Times, 29 March, 2007. 4See for example the studies by Holtz-Eakin and Kao (2003) and Audretsch et al. (2006). 43 in India. Compared to the other main religions of India, Hinduism provides little encouragement or value to change one’s situation in terms of material well being (Singer, 1966). According to Uppal (2001, p. 20), “The people of South Asia are deeply religious and all facets of their lives including their endeavors to achieve material advancement are affected greatly by religious beliefs and values.”7 According to Hinduism every human being is Amrutasya Putraha, a child of immortality and a spark of divinity. The purpose of life is to attain liberation which essentially is freedom from re-birth and the chain of cause and effect. One should live to understand reality and not for transitory material pursuits. Dharma Righteousness, Artha Earnings, Kama Desire, Moksha Liberation are supposed to guide the lives of Hindus. The scriptures ordain individuals to follow righteousness, perform duties and earn their livelihood, satisfy their desires and finally seek liberation. Dharma, Artha, Kama, Moksha can also be interpreted differently: one should righteously earn his livelihood and desire only for liberation (also referred to as self-realization). An individual has to do his duty as dictated by the scriptures and should not loose himself in material pursuits. Varna refers to classification of individuals into different classes, categories or castes. Historically Hindus were classified into four major castes. Initially their occupation determined their caste and caste affiliation akin to the religious iden- tity was passed on to their progeny. Brahmins were scholars, priests, advisors to kings, intelligentsia of the community. Kshatriyas were kings and noblemen. Their duties involved protection of the community from enemies and adminis- tration. Traders, businessmen and entrepreneurs were Vyshyas and people of all other occupations were classified as Shudras. Thus the Varna System that ini- tially categorized individuals into different classes persisted across generations and later determined the occupations of Hindus to a great extent. In his third major work on the sociology of religion, Weber (1958, pp. 103-104) states that “If the stability of the caste order could not hinder property differ- entiation it could at least block technological change and occupational mobility, which from the point of view of caste were objectionable and ritually danger- ous.” In summary, he claims that the impact of caste system on the economy is essentially negative (Medhora, 1965). In one of the few studies analyzing the effects of the caste system, Munshi and 7Uppal (2001) also provides an excellent overview of the philosophy of Hinduism. 46 Rosenzweig (2006) examine the influence of the caste within the context of an educational choice model in Bombay. They find that lower caste boys are more likely to study in schools where the medium of instruction is the local language and not English. This is very likely to lead them into traditional occupations as defined by the caste structure. Munshi and Rosenzweig (2006, p. 1230) note, “caste networks might place tacit restrictions on the occupational mobility of theirs members to preserve the integrity of the network” and “although these restrictions might have been welfare enhancing and indeed equalizing when they were first put in place, such restrictions could result in dynamic inefficiencies when the structure of the economy changes.” The clear demarcation of occupations based on castes, the persistence of oc- cupation decisions across generations and the other tenets that entail Hindus not to live a life of material pursuits, lead us to hypothesize that these factors might continue to influence the occupational choices of Hindus, and in particular inhibit the propensity to become an entrepreneur. We have no strong predictions how other religions in India, like Islam or Christianity, might influence an individual’s entrepreneurial decision. It is likely, however, that the impact of the caste system on economic behaviors is stronger for Hindus as compared to non-Hindus. In the following sections we will analyze whether Hinduism, as well as belong- ing to a lower caste, will influence the propensity to become an entrepreneur. 3.3 Data The main source of data to link religion and caste affiliation to entrepreneurship is the National Sample Survey Organization (NSSO) of India. We use the NSSO’s 60th round Employment-Unemployment Survey. This household level survey was conducted in 2004. Almost three hundred thousand individuals in sixty thousand households were questioned about their economic status, religious affiliation and personal background. The households were selected based on a stratified sampling methodology. Since the focus of this paper is on economically active individuals, we only consider those who have reported to be: self employed (includes own account workers and employers), salaried employees, casual laborers and unem- ployed. For similar reasons, we restrict our sample to those who are older than 15 years but younger than 70 years. We thus exclude from our analysis family 47 members who assist household enterprises, such as children and the elderly, as well as people classified into other miscellaneous occupational categories. These individuals can also be located according to their region. The final sample consists of 87,181 individuals. Table 3.1 provides the means and standard deviations of the independent vari- ables. 79% of the final sample are Hindus, 11.2% are Muslims, 5.6% are Christians, 1.4% are Sikhs, 0.3% are Jains, 1% are Buddhists and 1.1% are individuals of other religions or without religion. This roughly corresponds to the distribution of religion within the overall population of India.8 66.5% of Jains in the sample are self-employed, 50.4% of Christians and 48.6% of Muslims, 41% of Hindus and Sikhs and 38% of Buddhists. (Figure 3.1 and Table 3.2). Individuals included in the database are also classified according to class affil- iation. They belong to either one of the three backward classes (Schedule Castes, Schedule Tribes, Other Backward Classes) or to the forward castes. 12.5% of the sample belong to schedule castes, 18% to schedule tribes, 36.8% to other backward classes. These three classes combine to account for 67.5% of the entire sample. It should be emphasized that although the caste system is a distinct feature of Hin- duism and the Constitution of India (Schedule Castes) Order, 1950 notes that, “no person who professes a religion different from the Hindu, the Sikh or the Buddhist religion shall be deemed to be a member of a Scheduled Caste”, almost 66% of Christians are classified in the Schedule Caste. As Table 3.3 suggests, the other religions also have a share of their population that claims to be backward. While in Christianity this may be the result of conversion of individuals of the lower castes of Hinduism, in other religions this possibly reflects the economic backwardness rather than social backwardness. The presence of caste system, a characteristic of Hinduism, is also reflected in other religions in India. Within Islam certain sects are considered to be nobler than others. In Christianity, con- verts from lower castes of Hindu society are treated as lower caste members of Christianity. We cannot rule out conversions into Christianity giving rise to this phenomena. Also, we cannot rule out the possibility of the caste system diffusing into other religions in India. When we examine class based occupational behavior specifically in Hinduism, 8According to the 2001 Census, the religious composition of population in India is as follows: 80.9% are Hindus, 12.9% are Muslims, 2.4% are Christians, 1.9% are Sikhs, 0.4% are Jains, 0.8% are Buddhists, and 0.7% are others. See Premi (2004, p. 4294). 48 backward class but not members of other religions. One might therefore argue that the reservation system enables Hindu backward class to favor salaried em- ployment instead of self employment whereas members of other religions choose self employment. However, the values of estimated marginal effects suggest that the positive coefficients for salaried employment category are negligible compared to the negative coefficients in the self-employment category. This suggests that the effect of caste system in inhibiting Hindus from selecting self-employment is significant. In fact, the backward class Hindus have a higher propensity to be casual laborers. In order to focus on the impact of caste system we estimate the model based on the sample of Hindus only (Table 3.9) . The strong presence of class struc- tures within Hinduism leads us to posit that Hindu individuals belonging to the backward class might have a lower propensity to become an entrepreneur than Hindus belonging to the forward class. Thus, the impact of both religion and caste system, by being both a Hindu and a member of the backward class on the decision to become an entrepreneur was estimated and the results are presented in Table 3.9. The evidence suggests that a Hindu who is a member of the back- ward class scheduled caste is almost 14.6% and backward class scheduled tribe is 18% less likely to be self employed than a forward class Hindu. The four estimated models confirm our hypotheses that Hindus are less likely to be entrepreneurs than are individuals of other religions. This leads us to the last question. How does the propensity to enter into entrepreneurship compare between the non-Hindu and the Hindu religions? Thus, the results included in Table 3.10 take Hinduism as the base class and show the marginal effect on the probability to be self employed for individuals of other religions. The results sug- gest that Muslims are 7.9%, Christians 2.9%, and Jains 27% more likely to be self employed compared to Hindus. By contrast, individuals of other minor reli- gions and those without religion are almost 13.4% more likely to be entrepreneurs compared to Hindus. Buddhists and followers of Sikhism are pretty much in the same boat as Hindus. As a further check of the robustness of the results, we estimate a model by considering the self-employed separated as employers and only self-employed people (Table 3.11). It is startling to observe that the coefficients of the Hindu variable and the backward class variable are significant and negative even for 51 the employer group. This suggests that the Hindus have a lesser propensity to be entrepreneurs.13 An important qualification of the results is that the self- employed includes both agricultural and non-agricultural self-employed people. However, when the sample is restricted to non-agriculture (Table 3.12), the results confirm that there is virtually no difference. It is important to note that minority communities are associated with higher self-employment rates even in the developed countries (Clark and Drinkwater, 1998). However, the insight from our analysis is that even when we consider the Hindus alone, the caste system has an effect on the propensity to be self-employed. This supports our theory that the caste-system continues to exert an influence on the occupational choice of Hindus. 3.5 Conclusion Religion is rarely attributed to shaping economic phenomena. So it is with the de- cision to become an entrepreneur. While a rich and robust literature has emerged identifying a number of important characteristics and factors alternatively con- ducive to or impeding entrepreneurship, religion has been noticeably absent. The results of this paper suggest that religion matters. While India is rich with diverse religions, some of them, such as Islam and Christianity, are conducive to entrepreneurship. By contrast, others, and in particular Hinduism, inhibit entrepreneurship. We control for regional specific effects by introducing state level dummies and the results are robust to these controls as well. Similarly, the caste system is found to influence the propensity to become an entrepreneur. In particular, belonging to a backward caste inhibits entrepreneur- ship. The least entrepreneurial people tend to be Hindus in the lower class. One reason for this might be the long shadow of caste system that persists and limits the freedom of occupational choice to some extent not only to all individuals of backward classes but to Hindus in particular. Hence, the results of this paper suggest that elements of religion and the caste system need to be explicitly considered in understanding what influences important economic phenomena, such as entrepreneurship. Just as religion plays 13The marginal effects are very small but this could partly be attributed to the very small number of employers in the sample. 52 a major role in influencing entrepreneurial activity, so too does the caste system. At least in the case of India, Max Weber’s insight is found to hold - religion is an important influence on economic behavior. It may be fruitful for future research to consider not just the impact of religion on economic activity, such as entrepreneurship, but also the conditioning effect of the particular locational context. One clue about the importance of location is provided by the results of studies showing that Indian and other Asian im- migrants in the United Kingdom and North America actually exhibit a greater propensity for entrepreneurship (Clark and Drinkwater, 1998). While the specific religion of the immigrants is not explicitly identified, the inhibiting impact of a specific religion and particular caste may, in fact, disappear along with the change in location and institutional context. Without the painstaking future research, however, such a conjecture will remain simply that, a conjecture. 53 Table 3.2: Religion and Occupational Choice (Descriptives) Religion Self Salaried Casual Unemployed Total Employed Employee Labor Hinduism 41.30 23.90 28.99 5.81 100 Islam 48.62 20.92 24.28 6.17 100 Christianity 50.43 30.01 13.58 5.98 100 Sikhism 41.00 30.53 22.2 6.26 100 Jainism 66.54 28.08 4.23 1.15 100 Buddhism 37.97 26.00 32.15 3.88 100 Others 69.69 16.45 9.70 4.16 100 Total 43.01 23.95 27.23 5.81 100 Table 3.3: Religion and Caste System (Descriptives) Religion Backward Backward Backward Forward Total Caste(SC) Tribe(ST) Other(OB) Caste Hinduism 8.84 21.28 40.06 29.82 100 Islam 2.98 0.99 35.67 60.37 100 Christianity 66.24 3.69 11.60 18.47 100 Sikhism 0.56 31.56 19.90 47.98 100 Jainism 7.31 0.00 2.69 90.00 100 Buddhism 39.27 50.81 5.83 4.10 100 Others 85.36 1.30 11.68 1.67 100 Total 12.52 18.17 36.88 32.43 100 For explanation on SC, ST, OB see notes of Table 3.1. 56 Table 3.4: Caste System and Occupation (Descriptives) Social Self Salaried Casual Unemployed Total Group Employed Employee Labor Backward Caste(SC) 46.91 18.69 29.77 4.62 100 Backward Tribe(ST) 28.32 18.72 47.39 5.57 100 Backward Other(OB) 45.75 21.59 27.50 5.17 100 Forward Caste 46.62 31.59 14.66 7.13 100 Total 43.01 23.95 27.23 5.81 100 For explanation on SC, ST, OB see notes of Table 3.1. Table 3.5: Caste System and Occupation in Hinduism (Descriptives) Social Self Salaried Casual Unemployed Total Group Employed Employee Labor Backward Caste(SC) 36.10 13.72 45.70 4.48 100 Backward Tribe(ST) 28.78 18.29 47.45 5.47 100 Backward Other(OB) 45.67 21.44 27.84 5.05 100 Forward Caste 45.90 34.23 12.43 7.44 100 Total 41.3 23.9 29 5.8 100 For explanation on SC, ST, OB see notes of Table 3.1. 57 Table 3.6: Hinduism and Entrepreneurship (Marginal Effects after Multinomial Probit Estimation) Independent Self Salaried Casual Unemployed Employed Employee Labor Religion: Hinduism -0.0861*** 0.0293*** 0.0534*** 0.00346*** (0.0052) (0.0044) (0.0042) (0.00088) Personal Characteristics: Age 0.0123*** 0.00758*** -0.0160*** -0.00397*** (0.0011) (0.0010) (0.00093) (0.00031) Agesq/100 -0.00424*** -0.00834*** 0.00939*** 0.00318*** (0.0013) (0.0012) (0.0011) (0.00040) Female -0.133*** 0.0630*** 0.0425*** 0.0272*** (0.0055) (0.0052) (0.0048) (0.0019) Married 0.0883*** -0.0445*** 0.000897 -0.0447*** (0.0066) (0.0058) (0.0056) (0.0026) Divorced 0.106*** -0.0540*** -0.0375*** -0.0149*** (0.012) (0.0096) (0.0089) (0.0011) General Education: Informal Education 0.0308*** 0.0721*** -0.102*** -0.000700 (0.0084) (0.0087) (0.0045) (0.0026) Primary School 0.0148** 0.170*** -0.202*** 0.0171*** (0.0060) (0.0060) (0.0035) (0.0022) High School -0.0763*** 0.312*** -0.286*** 0.0499*** (0.0065) (0.0066) (0.0029) (0.0037) University -0.226*** 0.426*** -0.297*** 0.0958*** (0.0070) (0.0081) (0.0022) (0.0066) Technical Education: Technical Degree 0.0139 0.0930*** -0.107*** 0.000122 (0.025) (0.021) (0.027) (0.0033) Technical Diploma -0.00744 0.105*** -0.111*** 0.0134*** (0.010) (0.0090) (0.0084) (0.0021) Household Characteristics: Urban 0.0439*** 0.171*** -0.218*** 0.00384*** (0.0047) (0.0042) (0.0033) (0.00088) 0.2<Land<0.4 Hectares 0.0730*** -0.0762*** 0.00339 -0.000272 (0.0054) (0.0044) (0.0045) (0.0010) 0.4< Land < 2 Hectares 0.325*** -0.146*** -0.176*** -0.00309*** (0.0055) (0.0045) (0.0039) (0.0011) Land > 2 Hectares 0.397*** -0.154*** -0.237*** -0.00606*** (0.0053) (0.0047) (0.0026) (0.0012) Observations 87181 Notes: *Signifies p< 0.05; ** Signifies p<0.01;*** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is primary occupation of the individ- ual. Base categories for marital status, general education, technical education, land dummies are unmarried, no general education, no technical education and less than 0.2 hectares of land respectively. Full set of state level regional dummies are also included in the regression. 58 Table 3.9: Backward Classes and Entrepreneurship (Only Hindus) (Marginal Effects after Multinomial Probit Estimation) Independent Self Salaried Casual Unemployed Employed Employee Labor Religion and Class: Hindu SC -0.146*** -0.0331*** 0.183*** -0.00331** (0.0084) (0.0078) (0.0090) (0.0016) Hindu ST -0.181*** -0.0415*** 0.222*** 0.000495 (0.0063) (0.0054) (0.0067) (0.0012) Hindu OBC -0.0446*** -0.0425*** 0.0926*** -0.00547*** (0.0057) (0.0048) (0.0055) (0.0010) Controls: Personal Characteristics YES General Education YES Technical Education YES Household Characteristics YES Regional Dummies YES Observations 69705 Notes: *Signifies p< 0.05; ** Signifies p<0.01; *** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is primary occupation of the individual. Base category for the Hindu caste is Hindu Forward. Set of state level regional dummies that have nonzero observations in all the four categories are included in the regression. 61 Table 3.10: Religion and Entrepreneurship (Marginal Effects after Multinomial Probit Estimation) Independent Self Salaried Casual Unemployed Employed Employee Labor Religion and Class: Muslim 0.0792*** -0.0475*** -0.0271*** -0.00462*** (0.0063) (0.0052) (0.0052) (0.00098) Christian 0.0290** 0.0200** -0.0490*** -0.0000146 (0.012) (0.010) (0.0090) (0.0020) Sikh 0.00315 -0.0224 0.0145 0.00476 (0.021) (0.016) (0.020) (0.0048) Jain 0.271*** -0.132*** -0.124*** -0.0155*** (0.029) (0.018) (0.027) (0.00094) Buddhist -0.0194 0.0350* -0.0111 -0.00444 (0.021) (0.018) (0.016) (0.0031) Others 0.134*** -0.0493** -0.0827*** -0.00196 (0.022) (0.019) (0.017) (0.0044) Backward Class -0.0778*** -0.0150*** 0.0941*** -0.00126 (0.0047) (0.0041) (0.0039) (0.00087) Controls: Personal Characteristics YES General Education YES Technical Education YES Household Characteristics YES Regional Variables YES Observations 87175 Notes: *Signifies p< 0.05; ** Signifies p<0.01; *** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is primary occupation of the individual. Base category for religion is Hindu. 62 Ta bl e 3. 11 :S el f-e m pl oy ed an d E m pl oy er s (M ar gi na lE ffe ct s af te r M ul tin om ia lP ro bi t E st im at io n) In de pe nd en t Se lf E m pl oy er Sa la ri ed C as ua l U ne m pl oy ed E m pl oy ed E m pl oy ee La bo r R el ig io n an d C la ss : H in du is m -0 .0 72 0* ** -0 .0 01 61 ** 0. 01 09 ** * 0. 05 88 ** * 0. 00 38 6* ** (0 .0 04 8) (0 .0 00 72 ) (0 .0 04 2) (0 .0 03 9) (0 .0 00 87 ) B ac kw ar d C la ss -0 .0 72 7* ** -0 .0 07 31 ** * -0 .0 18 2* ** 0. 10 00 ** * -0 .0 01 68 * (0 .0 04 4) (0 .0 00 83 ) (0 .0 03 8) (0 .0 03 7) (0 .0 00 88 ) C on tr ol s: P er so na lC ha ra ct er is ti cs Y E S G en er al E du ca ti on Y E S Te ch ni ca lE du ca ti on Y E S H ou se ho ld C ha ra ct er is ti cs Y E S R eg io na lD um m ie s Y E S O bs er va ti on s 87 17 5 N ot es : E m pl oy er s ar e tr ea te d as a se pa ra te cl as s he re . *S ig ni fie s p< 0. 05 ; ** Si gn ifi es p< 0. 01 ; ** * Si gn ifi es p< 0. 00 1. St an da rd er ro rs ar e re po rt ed in pa re nt he se s. D ep en de nt va ri ab le is pr im ar y oc cu pa ti on of th e in di vi du al .B as e ca te go ry fo rr el ig io n is no n- H in du an d fo rc as te is no n- ba ck w ar d cl as s. Se t of st at e le ve lr eg io na ld um m ie s th at ha ve no nz er o ob se rv at io ns in al lt he fiv e ca te go ri es ar e in cl ud ed in th e re gr es si on . 63 The few studies on start-up size show that the industry characteristics (Mata and Machado, 1996; Mata, 1996) and human capital of entrepreneurs (Astebro and Bernhardt, 2005; Colombo et al., 2004; Colombo and Grilli, 2005), determine the start-up size of new firms. However, the role of spatial location on the start-up size has never been studied although the economic geography literature empha- sizes the geographic location as an important determinant in shaping economic activity (Krugman, 1991; Fujita and Krugman, 2003). This paper contributes to the growing literature on the start-up size by highlighting that the firm size distribution of start-ups (FSDS) is not independent of the spatial context. Using recent methodological advances in spatial econometrics and a dataset of 150,000 firms that registered as small firms in India from 1998-2000, we find that the FSDS is remarkably spatially skewed and displays distinct spatial patterns. The paper consists of five sections. In the next section, we discuss the theoret- ical framework and present the hypotheses on the FSDS in an Indian context. In the third section, we present the geoadditive modeling techniques with Bayesian inference based on Monte Carlo Markov Chain(MCMC) methods. In the fourth section, we give the empirical results linking the region with the FSDS. In the final section, we provide the conclusions and summary and present possible avenues for future research. 4.2 The Start-Up Size One of the stylized facts in the industrial dynamics literature is that the magni- tude of firm entry, across industries, time periods, and regions is quite startling. Firm size distribution is skewed and the majority of entrants are small (Cable and Schwalbach, 1991). The likelihood of survival for new entrants is low and those that do survive grow at a higher rate than the incumbents. Firms that have a higher start-up size have a higher likelihood of survival (Dunne et al., 1989; Guimaraes et al., 1995).3 Many empirical studies categorically reject the Gibrat’s Law which, in essence, claims that the firm growth is independent of size. Three 3However, there are some exceptions. Agarwal and Audretsch (2001) show that the entry size is more important in the early stages of the industry life cycle but not in the mature stages. Audretsch et al. (1999), however, find that there is no relationship between start-up size and firm survival in a sample of Italian firms. They also find that growth rates are even neagtively correlated with initial size. 66 important surveys (Geroski, 1995; Sutton, 1997; Caves, 1998) summarize these and other major findings of the literature on entry, growth, survival and exit of firms. While the effects of entry are extensively discussed, the determinants of the start-up size have received little attention. As Colombo et al. (2004, p. 1184) note, “if a larger start-up size positively affects the likelihood of survival of new firms and if surviving new firms that started operations at smaller scale struggle to catch up, the question arises why there are firms with small initial size.” The few empirical studies on the determinants of the start-up size of firms include Mata and Machado (1996), Mata (1996), Görg et al. (2000), Görg and Strobl (2002), Astebro and Bernhardt (2005), Colombo et al. (2004), Colombo and Grilli (2005). These studies examine the role of industry characteristics such as the minimum efficient scale (MES) of the industry, industry growth, effects of operation at suboptimal scale (defined as the proportion of those employed in firms that are operating at sub-optimal scale), impact of market size, role of human capital characteristics of founders, such as previous work experience and education, and credit constraints, on the initial size of firms. As Mata and Machado (1996, p. 1321)4 note, “entry on a relatively large scale in each industry is much more sensitive to the minimum efficient scale and to the extent of firm turnover in the industry than entry in small scale. Put differently, it seems that small new firms appear everywhere, while relatively large ones only appear where economies of scale make it crucial, or where sunk costs are low, therefore leading to low losses in case of failure.” A similar study on Irish firms shows comparable results, but finds a negative effect of industry size and positive effect of industry growth on start-up size (Görg et al., 2000). The start-up size increases with age and education of the founder, and is higher in industries with higher minimum efficient scale (MES), greater turbulence, and in industries where few suboptimal firms operate (Mata, 1996). Industry-specific professional knowledge and managerial and entrepreneurial experience have been found to have a greater positive impact than education and working experience on the start-up size (Colombo et al., 2004).5 4Mata and Machado (1996) analyze a sample of 1079 new firms from Portugal. In their sample, not more than 25% have greater than the average size of 17 employees, and 50% of the firms employ less than 10 people. 5Colombo et al. (2004) investigate start-up size of 391 technology based young Italian firms in both manufacturing and services. 67 Görg and Strobl (2002) find that the presence of multinationals negatively effects the size of domestic Irish entrants. Astebro and Bernhardt (2005) show that entrepreneurial human capital of founders co-determines their household wealth and the firms start-up capital. According to (Colombo and Grilli, 2005), firms receiving external private equity financing have greater start-up size. Advertising costs and R&D expenditures are important in determining the start-up size of large firms than small firms (Arauzo-Carod and Segarra-Blasco, 2005). Nurmi (2006) studies sectoral differences in start-up size in Finland and finds that results for manufacturing and service sectors are very similar. In addition, some studies show that start-up size is higher when entrepreneurs receive inheritances (Holtz- Eakin et al., 1994). Evans and Jovanovic (1989) discover the presence of binding liquidity constraints that limit start-up capital of entrepreneurs. They find that “entrepreneurs are limited to a capital stock that is no more than about one and one-half times of their wealth.” Thus, almost all entrepreneurs in their sample “devote less capital to their business than they would like to.” (p. 825) As mentioned earlier, we hypothesize that start-up size is not independent of the geographic region. The growing literature of economic geography (Krugman, 1991; Fujita and Krugman, 2003) gives us compelling reasons to hypothesize that the spatial location should play an important role in determining the size of new start-ups. In particular, there are compelling reasons to posit that some regions give birth to firms with a greater start-up size while others lead to creation of very small firms. We also hypothesize that initial knowledge endowments of the firm and the ownership structure influence the start-up size. Entrepreneurs who possess technical knowhow are more likely to start with larger firms. Firms that have single proprietary ownership are more likely to be small compared to those that have partnership or co-operative ownership structures. 4.3 Geoadditive Models We use semiparametric regression techniques based on Bayesian P-Splines and geoadditive models for the empirical analysis. The method allows estimating the non-linearities of continuous variables and the neighborhood effects on the start- up size of new firms.6 A brief outline of the methodology is presented here. 6This section draws from Lang and Brezger (2004); Brezger and Lang (2005). 68 This would show the regional patterns of start-up size without controlling for firm characteristics. However, when firm characteristics (also called fixed effects) are also introduced into the geoadditive model, the resulting spatial pattern shows the residual spatial pattern after these characteristics are controlled for. Thus, the spatial patterns estimated in this paper are the residual spatial patterns, as we simultaneously introduce firm characteristics and the spatial components in the geoadditive framework. These estimated residual spatial patterns can be explained using one of the following econometric approaches. A simple strategy is to regress the mean residual spatial effects on the regional variables. Thus, after estimating the geoadditive model, the total spatial effect of each region is explained by regressing the posterior mean of the estimated spatial residual effect on the regional variables. However, this empirical strategy does not consider the estimated posterior variance of spatial effects. In order to overcome this problem, a discrete choice model of the 95% or 80% spatial effects can be estimated. In this case, a variable is constructed that takes a value of (-1) when the region has a significant negative effect, takes a value of (0) if the effect is insignificant and takes a value of (1) if the effect is significant and positive. This leads to a straightforward multinomial specification. This variable is then regressed on the regional variables. We employ both strategies to examine the determinants of the residual spatial patterns. 4.4 Data The main source of data for linking the geographic location of the firm with the start-up size is the Ministry of Small Scale Industries in India. We use firm level data from the third census of registered small scale firms. This census was conducted in 2001. We consider manufacturing firms that have started producing in 1998, 1999 or 2000 as new start-ups for the analysis following Audretsch and Keilbach (2004), who also consider the three year period, as new start-ups are subject to a very high degree of stochastic disturbance if only a very short period is considered. This rich dataset of entrants consists of 149,708 firms. Each such start-up was asked the set of initial conditions under which it was founded (like the original value of its plant and machinery, its year of initial production, the sector, the source of its technical knowledge, its spatial location). We use this 71 data to test the hypothesis that directly follows from our theoretical analysis. As the dataset is of small firms, we do not have information of large entrants. This limitation of the dataset, however, does not pose serious problems for testing our hypothesis as the theory of firm size distribution suggests that majority of en- trants are small and numerous. Furthermore, if few large entrants are also present in the dataset, they would at best be outliers and introduce heterogeneity.7 As the descriptive tables in section 4.6 suggest 88.5% of the sample consists of firms that are started by proprietors. 6.8% of the firms are owned by two or more partners and are referred to as partnerships. Firms having other ownership struc- tures such as co-operatives are 4.7% of the sample. 15.8% of the firms are managed by women. 73.8% are small scale industrial units. Thus, 26.2% of the firms are small scale business enterprises, primarily consisting of repairing, servicing and maintenance units. More than 14% of the firms have reported that they have technical knowledge. While only 0.94% of the firms in the sample have reported to have obtained knowledge from sources outside India, as many as 6.6% have their technical knowledge from other firms and 6.67% from universities. 20.9% of the firms are in the industrial sub-sector of apparels manufacture and 19.2% are firms dealing with food products. With 11.67% of all the firms, the next largest group comprises of firms in the industrial sub-sector of fabricated metals. 4.5 Empirical Analysis Geoadditive models are estimated to examine the effect of the geographic location on the start-up size. Two measures of start-up size are used. In the first model, the dependent variable, start-up size, is measured using initial employment of a firm. In the second model, initial value of fixed assets is used as a measure of start-up size. The following geoadditive models are estimated: η = γconst +γProprietaryOwnership+γWoman +γTechnicalKnowledge+γIndustrialSector + fspatial(district) + frandom(district) The structured spatial effects are estimated based on Markov random field 7There is compelling evidence that entry takes place in the form of new small firms (Au- dretsch, 1995; Dunne et al., 1989). This is one of the main reasons for the studies on the start-up size to use quantile regressions (Mata and Machado, 1996; Görg et al., 2000; Colombo et al., 2004). 72 priors and random spatial effects are estimated with gaussian priors.8 Table 4.4 suggests that the ownership type has an influence on the start- up employment. Firms that are started by single proprietors and women have a smaller start-up size. The estimation results suggest that start-ups by proprietary owners have a start-up size that is that is 66% smaller than the average size, and start-ups by women have a size that is 18% smaller than the average size, ceteris paribus. Firms that have a different ownership structures, such as partnerships, and firms that have technical know-how are more likely to have a higher start-up size. In particular, technical knowledge from abroad increases the start-up size by 15%, technical knowledge from other firms increases the start-up size by 7.8% and from universities by 8.3%. Thus firms that have technical knowledge at the start-up phase tend to have higher start-up size than firms that do not have any technical knowledge. Furthermore, firms that are located in urban regions are more likely to have a larger start-up size. Table 4.5 shows that these findings are robust to an alternate specification, with initial size measured by the initial value of fixed assets. Proprietary own- ers are found to start with lower levels of initial assets and so are women en- trepreneurs. It is also seen that technical knowledge and urban location also positively effect the start-up size. Figure 4.1 shows a clear presence of neighborhood effects on the start-up size, measured by initial employment. The structured spatial effects plotted in Figure 4.1(a) show that start-ups in northern regions of Uttaranchal, Uttar Pradesh, Bihar, Madhya Pradesh, western regions of Gujrat and Rajasthan and southern regions of Kerala, Karnataka, and Tamil Nadu are likely to be smaller. While Uttaranchal, Uttar Pradesh, Bihar, and Madhya Pradesh are poorer states, Gu- jrat, Kerala, Karnataka, and Tamil Nadu are richer regions. However, as the 95% confidence map in Figure 4.1(c) suggests, the negative effect of size that is seen in the richer southern regions is insignificant. Many districts of Uttar Pradesh and Rajasthan become insignificant in the 95% confidence map, as seen in Figure 4.1(c). Figure 4.1(a) suggests that the start-ups in Maharastra, Andhra Pradesh in the south, West Bengal in the east, the northeastern states, and Punjab in the 8The variance components in all the cases are estimated based on inverse gamma priors with hyperparameters a=0.001 and b=0.001. The number of iterations is set to 120000 with burnin parameter set to 20000 and the thinning parameter set to 100. The autocorrelation files and the sampling paths show that the MCMC algorithm has converged. These plots are available from the author. 73 cial development play an important role in determining the start-up size of new firms. The financial development is measured by per-capita credit flows in the re- gion, credit-deposit ratio and density of banking facilities. The estimates suggest that the financial-development has a positive impact on the start-up size. The per-capita net state domestic product and the literacy rate in the region have sig- nificant positive effects while unemployment has a negative effect on the start-up size. Agglomeration index is significantly negative throughout. The demographic variables are mostly insignificant. Thus, the estimation results suggest that eco- nomic and financial development are more important determinants of start-up size. Table 4.8 explains the spatial effects of the second model, shown in Figure 4.2(a). The results confirm the effects of financial and economic development on start-up size. In the estimations in Table 4.7 and Table 4.9, the 95% significant spatial effects in Figure. 4.1(c) and Figure. 4.2(c) are explained using multinomial logit models. The inferences from these tables are consistent with the estimations having the mean spatial effects as the dependent variable, though there are some deviations. For instance, start-ups in large districts have a higher initial employ- ment in Table 4.7 and start-ups in mid-sized districts have higher initial fixed effects. 4.6 Conclusion A growing body of literature examines the determinants of the start-up size of firms. These few studies mainly focus on the industry characteristics and person- ality traits of the entrepreneurs. Using a new database of entrants in India, this paper examines geography and location as determinants of start-up size. Our contribution is threefold: First, we show that the spatial location is a micro-determinant of start-up size of entrants. In particular, spatial neighbor- hood effects exert strong influence on firm size at entry. Second, we show that the ownership structure and initial knowledge endowments determine the firm size distribution of new start-ups. Third, we provide first insights into the deter- minants of the start-up size in a developing economy. The results also suggest that financial and economic development of a region can explain, to some degree, the spatial patterns that remain after controlling for the firm level effects. 76 Table 4.1: Characteristics of Start-ups (Descriptives) Log(Employment) 1.0768 Std. Dev. (0.8382 ) Log(Value of Plant and Machinery) 10.5329 Std. Dev. (1.9202 ) Proprietary 0.8850 Partnership 0.0678 Other Ownership 0.0472 Managed by Woman 0.1581 Small Scale Industry (SSI) 0.7382 Small Scale Business Enterprise (SSBE) 0.2618 Tech Knowledge (Foreign) 0.0097 Tech Knowledge (Firm) 0.0659 Tech Knowledge (University) 0.0667 Food Products 0.1922 Tobacco 0.0013 Textiles 0.0490 Apparels 0.2090 Leather 0.0226 Wood 0.0391 Paper 0.0112 Printing 0.0382 Coke 0.0046 Chemicals 0.0368 Rubber 0.0432 Minerals 0.0651 Basic Metals 0.0168 Fabricated Metals 0.1167 Machinery 0.0290 Computing Machinery 0.0021 Electric Machinery 0.0233 Communication Equipment 0.0051 Precision Instruments 0.0036 Motor Vehicles 0.0062 Transport Equipment 0.0028 Furniture 0.0814 Recycling 0.0004 77 Table 4.2: Model I Diagnostics Scale Parameter(Mean) 0.398845 Std. dev. 0.00145918 Unstandardized Saturated Deviance (Mean) 287268.13 149733.13 Std. dev 35.141 547.965 deviance(µ̄) 286731 149195 pD 537.121 535.54219 DIC 287805.25 150271.25 Table 4.3: Model II Diagnostics Scale Parameter (Mean) 1.83737 Std. dev. 0.00694745 Unstandardized Saturated Deviance (Mean) 504292.27 146514.26 Std. dev. 34.018535 519.3417 deviance(µ̄) 503744.24 145965.31 pD 548.03044 548.94972 DIC 504840.3 147063.21 78 -0.728659 0 0.806542 (a) Structured Non linear Effect of ‘Dis- trict’. Shown are the posterior means. -0.236363 0 0.273669 (b) Unstructured Random Effect of ‘Dis- trict’. Shown are the posterior means. (c) Non–linear Effect of ‘District’. Poste- rior probabilities for a nominal level of 95%. Black denotes regions with strictly nega- tive credible intervals, white denotes regions with strictly positive credible intervals. (d) Non–linear Effect of ‘District’. Poste- rior probabilities for a nominal level of 80%. Black denotes regions with strictly nega- tive credible intervals, white denotes regions with strictly positive credible intervals. Figure 4.1: Spatial Effects in Model I 81 -1.76545 0 1.97014 (a) Structured Non linear Effect of ‘Dis- trict’. Shown are the posterior means. -0.904058 0 0.724814 (b) Unstructured Random Effect of ‘Dis- trict’. Shown are the posterior means. (c) Non–linear Effect of ‘District’. Poste- rior probabilities for a nominal level of 95%. Black denotes regions with strictly nega- tive credible intervals, white denotes regions with strictly positive credible intervals. (d) Non–linear Effect of ‘District’. Poste- rior probabilities for a nominal level of 80%. Black denotes regions with strictly nega- tive credible intervals, white denotes regions with strictly positive credible intervals. Figure 4.2: Spatial Effects in Model II 82 Table 4.6: Determinants of the Mean Spatial Effects in Figure 4.1 (Start-up Size given by initial employment) Independent Model I Model II Model III Financial Development Per-capita Credit 0.0567*** (0.014) Credit-Deposit Ratio 0.112*** (0.019) Per-Capita Bank Offices 0.0385 (0.027) Economic Development Per-Capita NSDP 0.243*** 0.231*** 0.285*** (0.035) (0.034) (0.033) Unemployment -0.0517*** -0.0726*** -0.0564*** (0.013) (0.013) (0.013) Literacy Rate 0.00331*** 0.00530*** 0.00404*** (0.0011) (0.0011) (0.0012) Demographics Mid Size District 0.0396 0.0319 0.0470* (0.025) (0.024) (0.025) Large District 0.0740 0.0704 0.0830 (0.075) (0.074) (0.076) Population Density -0.00749 0.00320 0.00582 (0.013) (0.012) (0.012) Agglomeration Index Firm Density -0.179*** -0.173*** -0.171*** (0.011) (0.010) (0.011) Constant -4.634*** -3.870*** -4.037*** (0.32) (0.33) (0.45) Observations 534 534 534 R2 0.44 0.46 0.43 F 52.32 56.09 49.36 R2 Adjusted 0.435 0.453 0.421 Notes: *Signifies p< 0.05; ** Signifies p<0.01; *** Signifies p<0.001. Standard errors are reported in parentheses. Dependent variable is the mean spatial effect per district after estimation of the geoaddi- tive models. 83