From Rags to Riches: Does Culture Aect Entrepreneurial Activity?

Transcription

1 From Rags to Riches: Does Culture Aect Entrepreneurial Activity? Christian Busch, Andrea Lassmann Preliminary Draft, October 2009 Abstract Entrepreneurial activity diers substantially across countries. While cultural dierences across countries have often been proposed as explanations, measuring a country's cultural characteristics suers from various problems. This paper oers new evidence on the relative importance of the cultural determinants of economic activity. We test the hypothesis that cultural factors inuence entrepreneurial behavior. To achieve this, we look at dierences in self-employment rates between immigrant groups within the same market, which allows holding constant the institutional environment. Using U.S. Census data for the year 2000, we nd signicant dierences in the propensity to become self-employed across immigrants from 148 countries which is in line with previous ndings. However, previous studies could not relate self-employment shares to selfemployment shares in the immigrants' home-countries which contradicts a cultural explanation. We improve over the existing literature by additionally accounting for determinants of self-employment in the immigrants' home countries. We nd evidence of a signicantly positive relationship between self-employment rates of U.S. immigrants and entrepreneurial activity in their respective countries of origin that is robust to the inclusion of further variables. Our ndings suggest that cultural factors are to some degree an expression of the behavior acquired under dierent institutional environments. We show that the dierences can be explained by JEL classication: J21, J61, L26 Keywords: entrepreneurship, culture, immigrants 1 Introduction The variation in entrepreneurial activity across countries is substantial. For instance, looking at self-employment rates a common proxy for entrepreneurship KOF Swiss Economic Institute, ETH Zurich, Switzerland KOF Swiss Economic Institute, ETH Zurich, Switzerland 1

2 , values range from a share of 0.17% in Cambodia to 22% in Italy. According to data collected from the World Bank, entry rates of new businesses vary between more than 19% in New Zealand and a mere 2.4% in Haiti. 1 What explains these vast dierences? Most of the literature dealing with entrepreneurship has focused on individual characteristics of entrepreneurs and has identied strong determinants such as age, gender, education, and economic endowments. More recently, with improved cross-country data becoming available, there has been an increased eort to identify more fundamental explanations including institutions, geography and culture (e.g., Klapper et al. (2007); Glaeser (2007); Guiso et al. (2006)). However, while geography and institutional aspects, such as corruption or the protection of property rights can be measured to some degree, culture is essentially unobservable. The analysis of separate inuences of culture and institutions is additionally complicated by the fact that both are correlated and proxies of entrepreneurial activity may be endogenous to the level of development or to other (often unmeasurable) characteristics across countries. In this paper, we try to shed light on the relative importance of institutions and culture in shaping entrepreneurial activity. Our approach is to estimate cultural dierences in entrepreneurial activity by observing the probability of immigrants from dierent nations in the United States to become self-employed. This approach by measuring dierences of observed outcomes within the same market allows holding constant a number of alternative explanatory variables such as the institutional environment. In a second step, we test whether the cultural component of entrepreneurial activity, which we estimated from the selfemployment shares of immigrants in the US, can be related to similar measures in their countries of origin. 2 There is a rich literature showing persistent and signicant dierences in selfemployment rates across immigrant groups in the United States. Since these studies are able to account for most of the determinants of entrepreneurial activity at the individual level, the remaining dierences have often been attributed to cultural dierences across immigrants from dierent nations. The argument holds that since all immigrants face the same institutional framework, the differences between immigrant groups are likely to capture after accounting for personal characteristics cultural characteristics across nationalities. However, while the early work of Yuengert (1995) nds a positive correlation with the home-country self-employment rate, in the more recent literature, the results indicate that aggregate self-employment shares of immigrants are uncorrelated with average self-employment shares in their home countries (Akee et al. (2007)), or even negatively correlated (Fairlie and Meyer (1996); Tubergen (2005)). 3 While most of the literature has tried to explain this puzzle by resort to ob- 1 Sources: International Labour Oce database of the ILO Bureau of Statistics and The World Bank Group Entrepreneurship Database. 2 I.e. we use a concept of culture that relates to dierences across nations. 3 In addition, Fairlie and Meyer (2003) argue that Yuengert's results are due to the dierence between immigrants and US-natives. After accounting for these dierences, they found no signicant eect. 2

3 servable dierences across immigrants, we believe it is a promising strategy to additionally account for dierences in the country of origin. Yet, we try to account for this puzzle by extending this research in several respects. First, we argue that when using census data, the focus on overall selfemployment rates as a proxy for entrepreneurship is awed. While certainly many of the individuals who report to be self-employed correspond to the notion of entrepreneurs, others, mainly in service sectors, the transport industry, and retail services, often do not have a choice of not being self-employed. The use of self-employment shares thus understates the role of push-factors. In some countries, people working in certain professions may be self-employed while in others, the same jobs are mainly done by paid workers, further complicating cross-country analysis. In addition, if we have in mind a notion of entrepreneurs as building rms, growing through capital accumulation and investment, and creating employment, which seems to be the case in much of the literature, we certainly have to exclude most of these micro-entrepreneurs. For these reasons, we focus on a more narrow denition of entrepreneurs, by accounting for differences among incorporated self-employed rather than for all people reporting to be self-employed (this measure has also been used, for instance, by Borjas (1986); Evans and Leighton (1989); Lofstrom and Wang (2006), however, in a dierent context). Second, while previous research has chiey explained immigrant dierences by resort to observable characteristics of immigrants, we argue that a promising strategy is to add information about the country of origin. Even though high quality census data is available, it is dicult to evaluate informal and job-specic educational aspects of immigrants (Hendricks (2002); Akee et al. (2007)). Also, wealth is only incompletely measured in most of these datasets, but may be important to account for unequal distribution in initial conditions between immigrants from dierent countries. However, there are good reasons to believe that wealth diers systematically across countries of origin. This means that there may be eects beyond the individual immigrant's unobservable endowments and that could be explained more systematically by resort to dierences in immigrants' countries of origin. Thus, in addition to explaining dierences in self-employment rates between immigrants with their observable characteristics in the new environment, we extend this approach by additionally accounting for dierences across their home countries. For instance, immigrants' institutional knowledge may dier, which could explain why immigrants from a given country may be more familiar with the institutional environment in the United States. Or local institutions in foreign countries may distort labor-market decisions. We thus attempt to bridge the micro-level literature with the cross-country literature that explains entrepreneurial activity by performing a multilevel analysis of the determinants of self-employment. The use of a narrower concept of self-employment also mitigates the inuence of push-factors (such as unemployment and taxes) that are likely to be driven by institutional factors and thus should facilitate the comparison with countries of origin. However, the drawback of this measure is that the number of 3

4 self-employed immigrant individuals is much smaller (about half) of what it is for the case of all self-employed. Yet, the application of multilevel analysis requires the number of clusters to be large (see Moulton (1990); Milyo et al. (2006)). We thus need an encompassing dataset in order to evaluate the differences across countries. We therefore use data from the 2000 U.S. Census (Ruggles et al. (2008)). This dataset contains information about whether the respondent worked for her own enterprise or as an employee. This variable splits self-employment into employers and other self-employed individuals that own an incorporated businesses or farm and those whose businesses and farms are not incorporated. Furthermore, the data includes information about the country of origin of immigrants. Containing about 14 million respondents, of which 820 thousand are self-employed, the 5% sample is large enough to analyze the choice of self-employment according to country of origin. With regard to the cross-country analysis, our main explanatory variable relates to employers as a share of the total work force, rather than to also include own-account workers as is usually done. 4 This concept does not correspond entirely to the variable using incorporated self-employed individuals generated for US-immigrants. However we believe that such a comparison is nevertheless an improvement over the existing literature. Looking at the distribution across countries, our variable seems to be more sensible as a proxy for entrepreneurship. Third, an important nding that emerged from the micro-level literature is that besides persistent dierences between immigrants, the dierence between all immigrants and US-citizens fades over time (Borjas (1985, 1986)). There are various explanations for this assimilation eect, relating to income, legal and language issues, or adaptation to the environment. We thus believe that in order to relate immigrant self-employment rates to home country eects, one has to account for these dierences in timing. We assume that cultural beliefs and attitudes, i.e., unobserved characteristics, are imported in the country of immigration and for this reason show some degree of persistence even if the institutional environment changes. However, with the adjustment to the new institutional environment the cultural eect is likely to fade over time as immigrants generate knowledge about local institutions. To the degree that immigrants bring with them certain cultural habits, we should therefore look at dierences within a relatively short time span upon arrival. The paper is organized as follows: Section 2 gives a short overview on the literature related to entrepreneurial culture and compares existing arguments that institutions and culture are fundamental determinants of entrepreneurial activity. Section 3 describes the data methods used, followed by a presentation of the empirical results, including robustness checks in Section 4. The nal Section concludes. 4 The variable is calculated from the ILO Laborsta database. 4

5 2 Related Literature 2.1 The Eect of Culture versus Institutions on Economic Outcomes The notion of culture is so broadly dened and vague that economists have long been reluctant to rely on culture as an explanation for economic phenomena. There are several reasons why cultural explanations are much more dicult to test in general, and to separate from other institutional explanations in particular. First, culture is much harder to measure than other determinants of economic activity, including institutions. Second, culture can be dened in many ways, and the channels through which culture may aect economic outcomes are manifold such that it is dicult to state refutable hypotheses (Guiso et al. (2006)). Similarly, there are no clear hypotheses that give guidance as to which cultural traits or clusters of cultural traits should matter. Third, attempts to measure culture by resort to survey data suer from severe cognitive biases (Halo eects). Fourth, culture often shows a remarkable degree of persistence, but eventually is not immutable over time. This complicates the use of historical events as instrumental variables since cultural change is potentially endogenous to the institutional environment. Finally, culture may not have an independent role in determining economic outcomes. E.g., Greif (1993, 1994) demonstrates that institutions and culture are two closely related dimensions. In his analysis of the dierent societal organization of the Genoese and Maghribi traders, culture drives the selection between multiple equilibria and through this determines the informal institutional mechanisms that govern exchange. There is a rich literature demonstrating that cultural inuences often change very slowly over time. Various explanations have been proposed for these ndings, in particular for those cultural norms that persist even if they become economically inecient over time (Greif, 1993; Guiso et al. (2006)). 5 Whatever the explanation for this persistence, it enables us to study the inuences that cultural norms exhibit in a changing environment. A few studies attempted to exploit these quasi-experiments. E.g., Guiso et al. (2004) show by exploiting social capital dierences within Italy that migrants transport their culture to new locations. Similarly, according to Guiso et al. (2003) common economic attitudes of religiously raised individuals do not change if they reject religious denomination later on. Guiso et al. (2006) show that there is a strong positive correlation between trust in ancestor's country of origin and trust of immigrants. 5 Yet, Fisman and Miguel (2007) show that unpaid parking violations of UN diplomats positively correlate with high-corruption in the diplomat's country of origin. However, these violations sharply dropped after the right to conscate diplomatic license plates of violators was introduced. 5

6 2.2 Determinants of Entrepreneurship In studying the determinants of entrepreneurship at the aggregate level, we can distinguish between determinants at the level of the individual and determinants at the macro level. The latter is covered by institutional, technological, demographic and cultural conditions as well as economic and political development (see Stel et al. (2003)). The impact of institutional, i.e., political, legal and regulatory variables on entrepreneurship, rm entry and survival has been frequently studied. Klapper et al. (2006) nd signicant adverse eects of entry regulation on the creation of rms. In countries with high entry costs, new rms are larger and existing rms grow slowly. High entry costs matter most in developed countries and in countries with low corruption. 6 According to Desai et al. (2003), institutional factors such as property rights protection and fairness have a positive eect on entry and survival in Europe and lower average rm size, particularly in the less developed Eastern European countries. More recently, research in economic geography explored the eect of location and industry characteristics on entrepreneurship. For instance, Glaeser (2007) shows that entrepreneurial activity is correlated with the presence of input suppliers. Finally, Wadhwa et al. (2007) nd that immigrant entrepreneurs tend to be concentrated in technology clusters. Due to the relatively early availability of Census data in several countries, entrepreneurship has also been extensively studied at the micro level during the last 25 years. A large number of studies seek to explain dierences in the choice of self-employment among individuals of dierent race and ethnic groups or from dierent countries conditional on socio-economic variables for the U.S., Germany and the United Kingdom. The explanatory variables are largely driven by the availability of data and include age, gender, education, marital status, language prociency, income, capital assets, and health limits. In general, selfemployment is more prevalent among the male, the older, the better educated and the married population (see, e.g., Blanchower et al. (2001)). Also, a larger holding of assets increases the likelihood that an individual chooses selfemployment (see, e.g., Blanchower and Oswald (1998); Evans and Jovanovic (1989); Evans and Leighton (1989)). Immigrant self-employment typically exceeds native-born self-employment. Furthermore, self-employment rates dier by race and ethnic groups (see, e.g., Borjas and Bronars (1989); Fairlie and Meyer (1996); Rees and Shah (1986); Clark and Drinkwater (2000)). The literature identies various potential reasons for the signicant dierences in the probability to become self-employed among dierent groups or vis-à-vis native individuals. First, using dummies for immigrant cohorts, Borjas (1985) and Borjas (1986) show that strong assimilation eects of immigrant self-employment and earnings exist. After 10 years of residence in the U.S., the probability to become self-employed is at least as large for immigrants as for natives. This nding gives support to the fact that setting up an own business requires assets that are not needed to enter paid employment. Second, a set of factors may push 6 See also Djankov et al. (2002) for a discussion on cross-country entry regulation. 6

7 immigrants out of paid employment into self-employment. According to the disadvantage theory, factors like discrimination and poor language skills lower the relative returns to paid employment and therefore push immigrants into selfemployment. While Clark and Drinkwater (2000) nd evidence for an inuence of wage dierentials in the United Kingdom, Fairlie and Meyer (1996) show that self-employment rates in the U.S. are higher among ethnic groups with higher estimated wage earnings. Also, poor language skills are empirically reected in lower self-employment probabilities. Moreover, restrictions related to the legal status of the immigrant (i.e., citizenship, visa) may inuence self-employment in the country of immigration as shown by Constant and Zimmermann (2006) and Uwaifo Oyelere and Belton (2008). Third, factors such as the existence of ethnic enclaves may pull immigrants into self-employment. Enclaves provide a self-sustaining environment creating a comparative advantage in catering group-specic needs. Further, large enclaves potentially provide less expensive co-ethnic labor supply. For these reasons, they may foster the likelihood of selfemployment of immigrants (see, e.g., Aldrich and Waldinger (1990); Fairlie and Meyer (1996)). Using data at the very disaggregated geographic level, dierences in self-employment rates of immigrants can be partly explained by enclave eects. However, while Borjas (1986) nds positive evidence for Hispanics in the U.S., Clark and Drinkwater (2000) show that ethnic enclaves decrease the probability to become self-employed in the United Kingdom. 7 While the literature described above focuses on reasons for dierences in selfemployment that lie within the new environment, determinants of dierences in self-employment across immigrant groups may also be related to the country of origin for several reasons. These approaches are driven by the idea that immigrants transport cultural and economic endowments into their new environment. First, higher aggregate self-employment in the home-country may increase the likelihood to become self-employed in the country of immigration. The sign and signicance of an eect of home-country self-employment rates dier however. Yuengert (1995) nds a positive and statistically signicant effect of self-employment rates relative to U.S. rates. Tubergen (2005) shows that home-country self-employment is positive but statistically insignicant. According to Akee et al. (2007), home-country self-employment rates have a signicantly negative eect on U.S. self-employment. Relating home-country self-employment rates to the predicted self-employment of ethnic groups in the U.S., Fairlie and Meyer (1996) nd that the coecient on U.S. self-employment is insignicant and small in magnitude. Second, since higher assets increase the probability to become self-employed, income in the home-country may matter for self-employment in the country of immigration. Uwaifo Oyelere and Belton (2009) show that the immigrants from developed countries have higher self-employment probabilities in the U.S. A counter-argument is provided by Tubergen (2005): since skills (i.e., qualications) from lower-income countries may not be accredited in the country of immigration, self-employment in the 7 Further hypotheses are mentioned in the literature, including the sectoral choice model, see, e.g., Fairlie and Meyer (1996), or the tax avoidance hypothesis, see Yuengert (1995). 7

8 new environment may be inversely related to home-country economic status. GDP per capita yields a positive and signicant eect on self-employment. One strand in the literature tests the eect of home-country self-employment rates on the probability to become self-employed in the country of immigration. The notion behind this approach is that immigrants from countries where selfemployment is more common might have developed larger entrepreneurial skills that will inuence their occupational choice in the new environment. A further strand of studies examines cultural dierences in latent entrepreneurship, i.e. the probability of preferring to be self-employed, in dierent countries (see Grilo and Thurik (2005) and Blanchower et al. (2001)). These studies compare latent preferences for entrepreneurial activity in varying institutional settings to actual entrepreneurial activity. Our work analyzes the behavior of immigrants from dierent nations within a single market and thus excludes dierences in institutional settings that are present in cross-country studies. This allows us to study whether the cultural background of emigrants is transmitted into the new environment, holding other characteristics constant. We then return to cross-country data in order to take into account various institutional and noninstitutional dimensions that determine entrepreneurship across countries and compare these to the cultural component that we extracted within a constant institutional setting. 3 Data and Empirical Strategy Attempts to disentangle the contribution of entrepreneurial culture on economic outcomes from alternative explanations suer mainly from the problem of measuring countries' cultural attributes. In this paper, we explore a dierent source of variation. We estimate cultural dierences in entrepreneurial activity by observing the inclination of immigrants from dierent nations to become selfemployed. We thus look at national culture as the relevant cultural concept. This is in line with related literature that suggests to proxy culture by measures that are largely invariant over time in order to isolate cultural determinants of economic outcomes. 8 We choose an approach that has frequently been applied in previous literature (e.g. Yuengert (1995), Fairlie and Meyer (1996), Carroll et al. (1994) or Hendricks (2002) 9 ). The idea is to measure dierences in observed outcomes within the same market. Hence, this approach allows holding constant a number of competing explanatory variables such as the institutional environment. The market we look at in this paper is the United States. Our main variable of interest is self-employment. It is equal to 1 if an individual is self-employed 8 See Guiso et al. (2006). 9 Hendricks (2002) estimates dierences in human capital endowments by measuring differences in labor earnings across U.S. immigrants with identical skills using uses 1990 U.S. Census data. Similarly, and Carroll et al. (1999) study cultural dierences in savings patterns using immigrant data for Canada and the U.S. 8

9 and 0 otherwise. Although this measure of entrepreneurial activity has some limitations (e.g., it does not capture the size of the rm, the year the business was started, etc.), it may reasonably proxy the propensity to become an entrepreneur and is a measure commonly used in the literature (for a discussion see Blanchower (2000), Glaeser (2007)). We use U.S. Census data, specically, the 5% sample of the Integrated Public use Microdata Series (ipums) of the Minnesota Population Center covering the year 2000 (Ruggles et al. (2008)). 10 This dataset has a number of advantages. First, it covers a large number of immigrants. This is particularly important as the number of self-employed workers with respect to the overall population is very small. Having only a few conrming observations is a common problem in binary data regressions and possibly results in severe biases. The total number of immigrants in the sample is over 1.2 million, covering more than 98 thousand foreign-born self-employed individuals. For technical reasons, it is restricted to include 100'000 randomly selected U.S. native citizens. Second, the dataset covers a wide range of demographic, social, and economic variables which have been found to explain the probability of being self-employed. These include age, gender, education, marital status, the number of children, prociency in English, industry of employment and U.S. state. This allows us to control for most systematic bias resulting from dierences in educational background across immigrant groups or due to immigrant-self selection. Third, we can distinguish between incorporated and unincorporated self-employment, i.e., employers and other self-employed individuals that own an incorporated businesses or farm and those whose businesses and farms are not incorporated. 11 We follow the literature and omit all individuals younger than 20 and older than 69 years. We drop the 55 thousand observations from the sample that report to be unemployed and self-employed at the same time for quality reasons. 12 In the regressions using individual country information, we also exclude countries with less than 30 reported individuals being self-employed because self-employment rates calculated from few observations may be less reliable. A few other variables mentioned in the literature that may determine individual probability of becoming self-employed, such as inheritance, access to funding etc., are not available in the ipums dataset and are thus not accounted for. In the unrestricted sample, including the unincorporated and the incorporated self-employed, the share of self-employed individuals among non-native Amer- 10 The same dataset is used by Michelacci and Silva (2007) to analyze how local entrepreneurship contributes to business creation; Fairlie (2004); Fairlie and Woodru (2006); Lofstrom and Wang (2006) 11 The demographic variables include age, gender, education, marital status and number of children in the household. In addition we obtained the U.S. state of residence, country of origin, duration of stay in the U.S. in years, prociency in English, income and industry of employment from the same source. The total sample size is about 14 million of which 820 thousand are self-employed. The overall share of self-employment in the sample is thus around 6%. 12 Since the majority of these are immigrants, these observations most likely exhibit a quality bias. 9

10 icans ranges from 2.8% for immigrants from Cape Verde to close to 20% for Greek immigrants. In a rst stage, we estimate a logit function, regressing self-employment on the control variables mentioned above and include a dummy variable equal to one for each of the countries of origin. A common assertion states that immigrants arrive with a set of cultural values and behaviors dierent from those in the destination country. This is reected in the possibility of a non-zero value of the country dummies in this stage. Over time, those transplanted behaviors fade, and institutional factors and cultural norms in the country of immigration become more important. We thus also examine the evolution of entrepreneurial activity of immigrants over time using interaction terms for each country dummy with the duration of residence in the U.S. We nally also repeat these regressions with an immigrant dummy and continental dummies instead of country dummies. In order to exclude the very small businesses, these regressions are repeated with a restricted sample, including the incorporated self-employed instead of the sum of the incorporated and unincorporated self-employed as the dependent variable. The self-employment share then spans from 0.5% for immigrants from Cape Verde to 10% for immigrants from Greece. To anticipate the ndings from the rst stage, we nd a substantial and statistically signicant amount of variation across countries. Since the institutional environment is the same for all immigrant groups, and we are able to control for demographic and educational background, we interpret these results as evidence for cultural explanations of entrepreneurial activity. Self-employment rates across countries can be due to diverse circumstances. While in some countries, a legal and nancial environment that fosters the start-up of businesses may further the existence of entrepreneurs, people may be forced into self-employed work as a result of high unemployment, corruption, lack of paid work opportunities. If people with similar abilities emigrate from dierent countries, focusing on the country of origin may similarly well explain the dierences between immigrant groups. Instead of focusing on immigrant differences solely, we thus try to explain the dierences by including determinants on self-employment rates in the immigrants' home countries. Thus, in the second stage, we test whether entrepreneurial activity in the country of origin aects the propensity to become self-employed in the U.S. As proxies for entrepreneurial activity in dierent countries, we calculate selfemployment ratios using data from the International Labor Organization. Since the self-employment ratios from ILO refer to employers, the incorporated selfemployed of the Census sample may better t our approach. We include both of them in our baseline regressions. 10

11 4 Empirical Results 4.1 Dierences in Self-Employment Across Immigrant Groups in the U.S. Table 1 presents the coecients of logit estimates with self-employment and incorporated self-employment of immigrants to the US and US-citizens as the dependent variable. All regressions include age, gender, education, marital status, the number of children, prociency in English, and industry of employment as control variables. The latter are of particular importance, as self-employment is systematically higher in some professions than others. We constructed these variables by aggregating dummy variables for those professions that yielded the largest fraction of self-employed persons. In particular, our dummy variables indicate whether a person works in one of the following branches: agriculture, building and construction, retail, services, transport and medical. Finally, we also included a dummy variable for household work. 11

12 Table 1: Probability of Being Self-Employed by Immigrant Status and Region, Logit Regressions Title: Probability of Being Self-employed by Immigrant Status and Country Group of Origin, Logit Regressions (1) (2) (3) (4) (5) (6) (7) (8) Overall Inc Overall Inc Overall Inc Overall Inc Immigrant *** *** *** (0.014) (0.023) (0.016) (0.027) Immigrant x Years US 0.008*** 0.009*** (0.000) (0.001) Middle East & North Africa 0.474*** 0.576*** ** (0.022) (0.033) (0.035) (0.051) Sub-Saharan Africa *** *** *** *** (0.031) (0.049) (0.050) (0.078) Western Europe 0.180*** 0.326*** 0.148*** 0.311*** (0.015) (0.025) (0.025) (0.039) Latin America & The Caribbean *** *** *** *** (0.015) (0.026) (0.019) (0.033) North America 0.264*** 0.312*** 0.216*** 0.321*** (0.022) (0.036) (0.039) (0.061) Eastern Europe & Central Asia *** *** *** (0.020) (0.033) (0.027) (0.043) South Asia *** *** *** (0.022) (0.033) (0.034) (0.049) East Asia & Pacific *** *** *** (0.016) (0.027) (0.022) (0.036) Oceania * * (0.042) (0.071) (0.077) (0.131) Middle East & North Africa x Years US 0.020*** 0.021*** (0.001) (0.002) Sub-Saharan Africa x Years US 0.027*** 0.027*** (0.003) (0.004) Western Europe x Years US (0.001) (0.001) Latin America & The Caribbean x Years US 0.017*** 0.022*** (0.001) (0.001) North America x Years US 0.004*** 0.007*** (0.001) (0.001) Eastern Europe & Central Asia x Years US (0.001) (0.002) South Asia x Years US 0.038*** 0.038*** (0.002) (0.002) East Asia & Pacific x Years US 0.017*** 0.014*** (0.001) (0.001) Oceania x Years US (0.003) (0.005) R-squared Number of Observations Notes: All regressions above are logit regressions for the year 2000 calculated with robust standard errors. The dependent variable is a dummy for being self-employed, where Overall includes the unincorporated and the incorporated self-employed and Inc includes the incorporated self-employed. The estimations included the following variables not shown in the Table above: age, age squared, gender, number of children, dummies for marital status, education, proficiency in english and industry of employment. *** significant on the 1%-level, ** significant on the 5%-level, * significant on the 10%-level. In Columns (1) and (2) of Table 1 we include a dummy variable which takes a value of 1 if the individual is an immigrant in the U.S. and 0 otherwise. The results indicates that the immigrant's overall probability of being self-employed is not signicantly dierent from the probability of US-citizens, conditional on all the control variables mentioned above. However, the overall probability of being self-employed within an incorporated organization is signicantly higher. In Columns (2) and (3) we present one explanation for the lack of signicance of 12

13 overall self-employment. We add an interaction term of the immigrant dummy with the years that have passed since immigration. The intuition for using an interaction term is that immigrants are assumed to arrive with a set of cultural values and behaviors that are dierent from US-citizens. Over time, immigrants assimilate their behavior, and factors that determine self-employment of all inhabitants in the US equally become more important. In addition, immigrants possess less knowledge and often less money which may lower their immediate probability to become self-employed. The individual coecient on immigrants now presents the eects of immigrants' probability of being self-employed immediately after arrival. The regression results in Columns (3) and (4) that allow for assimilation eects clearly show that the immediate probability is signicantly lower compared with an average US-citizen. Yet, this eect fades over time. This eect is is depicted in Figure 1 for overall self-employment which indicates that after approximately 15 to 20 years the probability of immigrants' being self-employed has converged to those of US-citizens (indicated by the solid line) and is slightly higher afterwords. Figure 1: Convergence of Immigrant Self-Employment Rates Over Time In the remaining columns of Table 1 we include variables indicating the region of the immigrants' countries of origin. In Columns (5) and (6), where we do not account for assimilation eects, three points stand out. First, the dierences in the coecients are much larger than for the overall immigration eect shown in Columns (1) and (2). Second, the probability for immigrants from Western Europe, Canada, and Middle East and North Africa is signicantly above average while immigrants from Latin America and Sub-Saharan Africa 13

14 have a signicantly lower probability of being self-employed for both measures. Third, for the incorporated, all regional dummies except for Oceania are signicant, while Eastern Europe, South Asia, and East Asia are not signicant for all self-employed. Finally in Columns (7) and (8) we account for regionspecic eects of the speed of assimilation. Again, the eects are much more pronounced with the immediate eects being lower, in general. Together with the interaction terms, the regression results indicate that the dierences across regions are larger if we account for dierences in the speed of convergence. The results show that only immigrants from western Europe and from Canada have higher probability of being self-employed than US-citizens even within relatively short time since immigration. Next, we delve beyond dierences in regional aggregates and include dummy variables for each country of origin. Figure 2 shows estimated coecients and condence intervals of the dummy variables for each of the major regions. In addition, the solid lines indicate each of the estimated regional coecients. Figure 2 shows that there is a substantial amount of variation within each region. This variation explains the relatively low coecients found for regional aggregates. However, looking beyond the regional aggregates, only for some regions the ranking within each of the regions gives a more consistent picture. For instance, in Latin America, almost all of the Caribbean countries can be found below the regional average. 14

15 Figure 2: Estimated Dierences in Self-Employment Rates Across Countries of Origin Scotland Country Dummy Results - Western Europe Coefficients and 90% Confidence Intervals Portugal Azores Germany Spain Northern Ireland Norway United Kingdom Ireland Italy France Finland Belgium Austria Netherlands Sweden Wales Switzerland DenmarkGreece Country Dummy Results - East and South-East Europe Coefficients and 90% Confidence Intervals Bosnia Kosovo Albania Serbia Slovakia Yugoslavia Macedonia Latvia Poland Croatia Bulgaria Lithuania Cyprus Romania Czechoslovakia Hungary Czech Republic Country Dummy Results - South East Asia Coefficients and 90% Confidence Intervals Country Dummy Results - Middle East & North Africa Coefficients and 90% Confidence Intervals Philippines Laos Burma (Myanmar) Indonesia Thailand Malaysia Singapore Vietnam Cambodia Algeria Kuwait Saudi Arabia Morocco Egypt Iraq Jordan Iran Yemen Arab Republic (North) Lebanon Syria Israel Country Dummy Results - Latin America Coefficients and 90% Confidence Intervals Country Dummy Results - Sub-Saharan Africa Coefficients and 90% Confidence Intervals St. Kitts-Nevis Dominica Barbados Haiti St. Lucia St. Vincent Antigua-Barbuda Puerto Rico Grenada Guyana/British Guiana Jamaica Panama Bahamas U.S. Virgin Islands Trinidad and Tobago Belize/British Honduras Nicaragua Mexico Bolivia Honduras Ecuador El Salvador Dominican Republic Guatemala Peru Costa Rica Colombia Bermuda Cuba VenezuelaBrazil Chile Uruguay Argentina Paraguay Cape Verde Sudan Somalia Liberia Sierra Leone Ethiopia Cameroon Ghana Nigeria Eritrea Uganda Kenya Senegal Zimbabwe South Africa (Union of) Tanzania In the Appendix we show the corresponding regressions that include all of the country dummies, as well as the interaction terms jointly, analogous to the regressions shown in Table 1. Again, the regressions that include dummy variables for each country of origin indicate signicant dierences in the probability of being self-employed between countries as well as from U.S. citizens. Using interaction terms for each country dummy with the duration of residence in the U.S., we also nd strong convergence eects of immigrants towards the average self-employment ratio of U.S. citizens. 15

16 4.2 Cross-Country Dierences in self-employment and selfemployment in the U.S. The Graphs in 2 give an intuitive explanation for the relatively low coecients predicted in Table 1 for each of the regions. The reason is that there is still substantial variation within each of the regions. This indicates that we need to account for variation across self-employment shares at the country-level rather than at the regional level. This is in contrast with recent work that includes regional dummies while earlier work tried to account for country dierences but had only a limited number of countries available (Fairlie and Meyer (1996); Yuengert (1995)) In the following section, we rst try to account for whether the estimated dierences across immigrants from dierent countries are systematically related to aggregate levels of entrepreneurial activity found in the country of origin. In a second step, we add macro-level data for various determinants of entrepreneurship, including common institutional variables in the country of origin in order to explore the determinants of possible cultural dierences across immigrant groups. More specically, we are interested in testing whether institutional explanations can account for dierences in culture. Finally, we use the predicted self-employment values for each country as an explanatory variable using GLS, in order to verify the robustness of our results. In all of the specications comparing self-employment shares in the US with their respective counterparts in the country of origin, we cluster the standard errors to account for both a general form of heteroskedasticity as well as for any intra-country correlation (Moulton, 1990). As a rule of thumb, the number of clusters should be around 50 in multilevel analysis integrating microand macro-data (Primo, Jacobsmeier and Milyo, 2006). Table 2 contains the same control variables as before. However, accounting for the critique stated by Fairlie and Meyer (1996), we do not scale self-employment rates to those of UScitizens, but rather omit all US observations from the regressions. Column (1) of Table 2 reveals that after correcting the standard errors by clustering them at country level there is no signicant relationship between the self-employment share of immigrants in the US with the aggregate self-employment shares of the respective countries of origin. Without accounting for intra-cluster correlation the relationship is actually negative and highly signicant. 16

17 Table 2: U.S. Self-employment and Home-country Shares, Logit regressions Title: Home-Country Self-Employment and Assets, Logit Regressions (1) (2) (3) (4) (5) (6) (7) (8) Total self-employment Incorporated All Immigrants Immigrant < 10 Years in US Immigrant > 10 Years in US Immigrant < 5 Years in US All Immigrants Immigrant < 10 Years in US Immigrant > 10 Years in US All Immigrants Log dividend income (8.75)*** (17.73)*** Self-employment share (-1.02) (1.89)* (-0.85) (-0.05) Employer share (2.70)*** (1.92)* (2.77)*** (2.36)** Pseudo R-squared Number of Clusters Number of Observations Notes: All regressions above are logit regressions for the year 2000 calculated with clustered standard errors. The dependent variable is a dummy for being self-employed (immigrants only). The estimations included the following variables not shown in the Table above: age, age squared, gender, number of children, dummies for marital status, education, proficiency in english and industry of employment. The home-country self-employment share is defined as the share of the sum of own-account workers and employers for regressions including overall US self-employment; as the share of employers for regressions including incorporated US self-employment (Source: ILO Laborsta Database). *** significant on the 1%-level, ** significant on the 5%-level, * significant on the 10%-level. 17

18 Column (2) and (3) account for the convergence hypothesis by separating immigrants to those within ten years since immigration and those that lived at least ten years in the US. Although the sample is signicantly smaller, in the specication focusing on recent immigrants only, the correlation with average self-employment rates in the home country becomes signicant yet remains negative. Using an interaction term (not shown) again conrms this result. This conrms our hypothesis that in order to evaluate a possible cultural inuence that is persistent across borders, it is important to account for the adjustment within the new institutional framework. However, the negative coecient remains a puzzling nding which we address in the remainder of the empirical sections. In columns (5) to (7) of Table 2, we rst analyze the dierences between overall self-employment ratios and our more narrow denition of entrepreneurs. The denitions are not entirely consistent across the two datasets, be we try to dene entrepreneurs in such a way that the two concepts correspond as closely as possible. While we calculate incorporated self-employment as the dependent variable from the ipums dataset, the cross-country variable calculated from ILO data refers to employers as a share of the total work force. Column (5) again accounts for the eect over all immigrant groups. Notice that in this specication, the dummy variable indicating household work in the US contains no self-employed individuals and thus has to be omitted. This further conrms the validity of our approach. As expected, the results in Column (6) yield a positive inuence of the share of employers in the country of origin on the share of incorporated self-employed immigrants in the US which is signicantly dierent from zero at the 1% level. This eect is robust over dierences in years since immigration, while the coecient even increases slightly over cohorts with a longer duration. These results clearly show that overall self-employment rates should be used cautiously as a proxy for entrepreneurship and that using a more narrow denition of entrepreneurs is preferable. One problem with the regression results in Table 2 may be that Immigrants from richer countries dier in their wealth and may thus be more inclined to become entrepreneurs, in particular within the rst years upon arrival. A common way to account for dierences in endowment is to include a variable that measures the income individuals earn from interests and dividends which should be highly correlated with the wealth they hold. However, this is only an indirect proxy and the data is likely to be relatively unreliable. For instance, we cannot separate individuals with no interest income and those who do not report. In our dataset, the number of individuals reporting positive interest income amounts to about 13% of the whole sample. Nevertheless, in Column (8) of Table 2, we repeat the estimation of incorporated self-employed with interest and dividend income as an additional control variable. Despite the largely reduced sample size that results from this additional restriction, the inuence of national background is robust to accounting for wealth. When we include our wealth proxy in the overall self-employment regressions from Columns (1) through (4), the home-country self-employment variable never turns out to be signicant. There are two ways to account for this discrepancy. The rst one is to account 18

19 for additional heterogeneity between immigrants. An alternative approach is to explain why average self-employment rates are heterogeneous across countries of origin. In the next section, we follow the second approach. 4.3 Institutional Determinants of Self-Employment The dierences in self-employment rates between immigrants from dierent countries have often been interpreted as cultural dierences. However, for culture to be relevant, we should observe a correlation with corresponding measures in the country of origin which recent literature could not conrm. In the previous section, we demonstrated that a part of this puzzle can be explained by resort to a dierent proxy for entrepreneurship. An alternative approach is to explain this puzzle not only by the observed dierences of immigrants but also with resort to systematic dierences in their home countries. For instance, if institutions determine entrepreneurial activity, the inuence of culture may only be observed if we control for the institutional environment. However, institutions may also shape culture over time. In this case, we might observe our institutional measure to dominate compared with the entrepreneurial measures used in the previous section. In Tables 3 and 4 we try to explain Dierences in US-immigrants' self-employment with systematic dierences in their countries of origin. In these estimations we focus on the immigrants within 5 years of migration for two reasons. First, even though we showed a persistent gap between immigrants and a correlation with the business self-employment in the home countries, the institutional set-up in the host country will become more important over time. Thus, long-term dierences may also be due to a more rapid accumulation of wealth of certain immigrant groups. By tracing the observed dierences to dierences in endowments across countries, we are ultimately interested in eects within the rst years of migration. Second, we try to explain the dierences between immigrants with dierences across countries. A systematic selection of indicators that measure the performance of the business climate across countries has been done only in the past few years. We, therefore, cannot explain long-term dierences between immigrants with this approach. However, and a bit surprisingly, the results we obtain also hold for the overall immigrant sample. One reason for this nding may be that the broad institutional environment across countries is remarkably persistent over time. In the following section, we use data from the beginning of the 1990s. For some variables, this is not possible. In these cases, instead, we use averages over the years 1990 to 2000, assuming that broad aspects of the institutional environment are highly persistent over time. As a baseline, we include several determinants of entrepreneurial activity across countries in our regressions shown in Table 4. A convenient way to account for systematic dierences across countries is to include levels of income per capita. Uwaifo Oyelere and Belton (2009) nd that immigrants in the US from developed countries have a much higher probability of being self-employed after 19