The Protestant ethic and entrepreneurship: Evidence from religious minorities in the former Holy Roman Empire

Author’s Accepted Manuscript The Protestant Ethic and Entrepreneurship: Evidence from Religious Minorities in the Former Holy Roman Empire Luca Nunziata, Lorenzo Rocco www.elsevier.com

PII: DOI: Reference:

S0176-2680(16)30314-7 http://dx.doi.org/10.1016/j.ejpoleco.2017.04.001 POLECO1639

To appear in: European Journal of Political Economy Received date: 7 December 2016 Revised date: 1 April 2017 Accepted date: 1 April 2017 Cite this article as: Luca Nunziata and Lorenzo Rocco, The Protestant Ethic and Entrepreneurship: Evidence from Religious Minorities in the Former Holy Roman Empire, European Journal of Political Economy, http://dx.doi.org/10.1016/j.ejpoleco.2017.04.001 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting galley proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

The Protestant Ethic and Entrepreneurship: Evidence from Religious Minorities in the Former Holy Roman Empire∗ Luca Nunziata†1,2 and Lorenzo Rocco‡1 1

University of Padua 2 IZA April 8, 2017

We investigate the effect of Protestantism versus Catholicism on the decision to become an entrepreneur in former Holy Roman Empire regions. Our research design exploits religious minorities’ strong attachment to religious ethic and the predetermined historical determination of religious minorities’ geographical distribution in the 1500s as a result of the “cuius regio eius religio” (whose realm, his religion) rule. We find that today Protestantism increases the probability to be an entrepreneur by around 5 percentage points with respect to Catholicism, a result that survives to a battery of robustness checks. We explicit the assumptions underlying the identification strategy and provide an extensive testing of their validity by making use of several European datasets. Keywords: Entrepreneurship, Religion, Culture, Protestantism, Catholicism. JEL Codes: J24, Z12, J21, Z13.

∗

We thank the participants of the Association for the Study of Religion, Economics, and Culture Conference in Arlington, the Society of Labor Economists Conference in New York, the European Society for Population Economics Conference in Essen, the Italian Congress of Econometrics and Empirical Economics in Pisa, the Brucchi Luchino Conference in Padua and the First Workshop Padua-Linz for valuable comments on the current and previous versions of this paper. The usual disclaimer applies. † [email protected] ‡ [email protected]

1

1

Introduction Religion has often been investigated as a potential driver of economic behaviour and eco-

nomic outcomes. In “The Protestant Ethic and the Spirit of Capitalism”, Weber (1904) argued that the ethical principles embodied by Protestantism were instrumental to the early development of capitalism by persuading individuals in pre-capitalistic societies to engage in secular activities. The debate about the correctness of Weber’s thesis has been renewed by recent contributions, with mixed results (Becker and Woessmann, 2009; Arrunada, 2010; Andersen, Bentzen, Dalgaard, and Sharp, 2013; Basten and Betz, 2013; Cantoni, 2015; Spenkuch, 2017). A common problem in the study of the economic effects of religion, especially in contemporary Europe, is that religious affiliation is measured with error because individuals typically report the religion they were raised in, regardless of the intensity of their current adhesion to its ethical principles. More specifically, many people who report of being Catholic or Protestant are in fact agnostic or follow a set of norms and beliefs that does not correspond to those of the canon. In a recent work (Nunziata and Rocco, 2016), we provide a novel way to address this problem and to test Weber’s thesis, by exploiting the fact that the measurement error in religious affiliation is arguably smaller among minorities. Indeed, empirical evidence suggests that members of minority religions adhere more tightly to their denominations’ underlying ethical principles than do members of larger religious groups, because religion is felt to be part of their worth preserving identity. In Nunziata and Rocco (2016) we focus on religious minorities and we compare Protestant and Catholic minorities in Switzerland, using census data and exploiting the historically determined geographical distribution of religious denominations. We find that Protestantism (in its Calvinist and Zwinglian branches) is associated with a significantly higher probability for an individual to be an entrepreneur than is Catholicism.1 The estimated difference ranges between 1.5 and 3.2 percentage points, it is larger the smaller the size of the religious minority and it is mainly driven by prime age males. One question that remains open is whether our findings that stand for Switzerland can be generalized to other areas and to other branches of Protestantism such as Lutheranism. The 1

We broadly define entrepreneurs to include all those individuals that risk on their own, self-organize their work schedule, often manage dependent employees and are residual claimants of the results of their ventures, in good and bad times. Hence, the entrepreneurs are captains of industry, business leaders in any economic sector but also professionals and small business owners.

2

present paper focuses on the European areas subject to the Holy Roman Empire of the German Nation (HRE henceforth) in the sixteenth and seventeenth centuries that host today a variety of combinations of Protestant and Catholic minorities and majorities whose presence is historically determined. The geographic distribution of minorities of both denominations largely reflects the diffusion of Protestantism in Europe in that period (Cantoni, 2015; Spenkuch, 2010, 2017). In particular, in the former HRE regions, the distribution of minorities is the result of the “cuius regio eius religio” (whose realm, his religion) rule, which granted the rulers the right to decide the official and unique religion of their territories after the Peace of Augsburg in 1555. This paper contributes to the literature in several ways. First, it presents new empirical findings on the effect of Protestantism on entrepreneurship beyond the effect found in Switzerland by Nunziata and Rocco (2016). It therefore provides new evidence on the effect of Lutheranism, as opposed to Calvinism and Zwinglianism in Switzerland, on entrepreneurship. Second, it provides an analytical derivation of the empirical model that makes explicit the identification assumptions in our research design and the extent of the bias in our estimates. Third, differently from the previous contribution, by controlling for entrepreneurial family background, it focuses on the effect of religious ethic on the autonomous individual decision to be an entrepreneur netting out the effect of inheriting a family business. Moreover, exploiting several datasets and the much richer information available at the European level compared to the Swiss level, we are able to examine more in depth to what extent religious minorities are stricter observants of religious principles than larger religious groups and to what extent Protestant and Catholic minorities are comparable on non religious factors that may be more or less conductive to entrepreneurship, such as the intensity of their social networks, whether minorities are subject to discrimination, the role of risk aversion. Finally, we provide a detailed analysis on the geographical persistence of religious minorities in the former HRE regions of Germany, building on previous work by Spenkuch (2010). Our empirical estimates indicate that adhesion to Protestant ethical principles increases the probability to be an entrepreneur by around 5 percentage points compared to adhesion to Catholic ethical principles. The point estimates are in line with those obtained on Swiss cantons data. We provide a number of robustness checks of our findings, including a placebo test, and show that the empirical results are not spurious or driven by specificities related to areas where 3

each denomination has been historically predominant, or by migration patterns. The paper is organized as follows. Section 3 presents our research design and provides a discussion of our identification strategy. The data are presented in section 4 and the empirical findings in section 5. Section 6 provides an extensive discussion of our identification assumptions, section 7 concludes. Further materials and analysis are available as online appendices.

2

Religious principles and entrepreneurship In this section we briefly summarize the key differences between Catholicism and Protes-

tantism, the consequence of the internalization of these principles to individual psychology, world-view and motivation, and discuss the hypothesized relation between religious ethic and the propensity to entrepreneurship. Catholicism and Protestantism share the same fundamental articles of faith, but they differ substantially in how religion should be practically lived. According the religious historians Johann Jakob Herzog and Phillip Shaff, Protestantism brought the believer into direct relation and union with Christ and set aside the doctrines of sacerdotal and saintly mediation and intercession. (“Protestantism” in Herzog and Schaff, 1908, vol. IX). The three fundamental principles of Protestantism are the supremacy of the Word, the supremacy of the grace of Christ, and universal priesthood of believers. The supremacy of the Word states that the scriptures are the only source of faith and establishes the right to individual interpretation. This is in sharp contrast with Catholicism which grounds faith on both the Bible and tradition and makes the clergy the only legitimate interpreter of the Word. The supremacy of the grace of Christ affirms that salvation comes only and directly from Christ and it is a “free” grace. In Catholicism, salvation depends instead on both faith and good works, stressing the role of works. The universal priesthood of believers states that all Christians have the right and the duty to read the Bible and proclaim it in public. This contrasts with Catholicism, which is based on exclusive priesthood authority, with priests being the necessary mediators between God and the people (Herzog and Schaff, 1908). These principles influence the psychology of believers, and are associated to the work ethic, and the appreciation of worldly success. From a psychological perspective, the Catholic should have a more “collectivistic” personality that values social connections and group affiliation,

4

given that Catholicism emphasises the sense of community, the heard, and group rituality. Conversely, the principles of personal relation with God and that of being the unique legitimate interpreter of the Scriptures imply that the Protestant should have an individualistic personality. The Protestant believes in the autonomy of his or her conscience, which is the ultimate judge of his or her actions (Cohen and Hill, 2007). As regards the work ethic, in continuation with the medieval perception that work is a punishment for man’s original sin (Tilgher, 1958), in the Catholic tradition work remains associated with toil and difficulty. In contrast, Protestantism, in both its streams of Calvinism and Lutheranism, perceives work of whatever kind as honourable, necessary, and a service to God. Luther contended that all persons should see their labour as their vocation (Eaton, 2013). Finally, regarding the attitude towards the worldly success, the Catholic Church tends to orient its members toward the hereafter, so that worldly successful has generally little importance. Differently, Protestants believe that the attainment of material possession, status, and the prestige that is associated with upward social mobility are signals of salvation and of the favour of God (Mayer and Sharp, 1962).2 On these grounds, Protestants should be more determined to achieve success, more willing to work hard, more disciplined, and more focused on their own self-realisation than are Catholics. According to Furnham (1987), these high achievers are likely to become successful entrepreneurs. In the following we empirically test such an intriguing hypothesis.3 Our work is related to the literature on the effect of personality and psychological traits on entrepreneurship, since we investigate one of the deep cultural determinants of the choice of becoming an entrepreneur. Zhao and Seibert (2006) compare entrepreneurs and managers as regards the so called Big Five personality traits (Costa and McCrae, 1992),4 and conclude that 2

This is especially the case in Calvinism. According to the Calvinist concept of predestination, God decides a person’s salvation or damnation before his or her birth, but it is hidden to the person, who can do nothing to alter his or her fate. Worldly success is regarded as a sign of being among the select group that will be saved from damnation by God (Becker and Woessmann, 2009; Arrunada, 2010). 3 For a comprehensive survey of the causes and consequences of the Protestant reformation see Becker, Pfaff, and Rubin (2016). 4 The Big Five personality traits are extraversion, emotional stability, openness to experience, conscientiousness, and agreeableness. Extraversion is the extent to which individuals are assertive, dominant, ambitious, energetic, and seek leadership roles; emotional stability measures how much individuals are self-confident, relaxed, and able to tolerate stressful situations; openness to experience refers to the individual ability to seek new experiences and to explore new ideas; conscientiousness is the extent to which individuals are achievement oriented, hard workers, efficient, and dutiful; finally agreeableness accounts for the forgiving and trusting nature of individuals and their being altruistic and flexible (Judge, Higgins, Thoresen, and Barrick, 1999; Caliendo, Fossen, and Kritikos, 2014).

5

entrepreneurs are more emotionally stable, open to experience and conscientious, as well as less agreeable, than managers. As regards extraversion entrepreneurs and managers are similar. By exploiting longitudinal data, Caliendo, Fossen, and Kritikos (2014) estimate the influence of the Big Five and other psychological traits, including the locus of control, risk tolerance and reciprocity, on the condition of being self-employed and the decision to enter in and exit from entrepreneurship. As regards the condition of being self-employed, they find that positive determinants are extraversion, openness to new experience, high risk tolerance and a strong belief that individual future outcomes depends entirely on individual’s own actions (internal locus of control). The traditional economic literature on entrepreneurship suggests that the decision to become an entrepreneur might depend on the individual’s exogenously distributed entrepreneurship ability (Lucas, 1978), on the probability that a particular new venture will survive (Holmes and Schmitz, 1990), or on the individual’s level of risk aversion, which influences how the trade-off between risk and returns is resolved (Kilhstrom and Laffont, 1979). Hence, the kind of soft skills that Protestantism is supposed to favour might be accommodated by all these models. For instance, the probability of a new venture’s survival might depend on individual determination and willingness to succeed, or the individual disposition to work hard might be central in the decision to adopt strategies of self-insurance, such as long hours of work to limit the risk of failure (Parker and Belghitar, 2006). A more recent literature suggests that an important determinant for the decision to become an entrepreneur is the ”entrepreneurial culture” prevailing at the local level. Rentfrow, Gosling, and Potter (2008) show that the Big Five personality traits are geographically clustered and that there is a clear overlap with the distribution of some individual behaviors that can be immediately attributed to the Big Five. Similarly, Obschonka, Schmitt-Rodermund, Silbereisen, Gosling, and Potter (2013) document the existence of a geographic clustering of entrepreneurship-prone personality profiles and that such pattern coincides with that of entrepreneurial activity. Obschonka, Stuetzer, Gosling, Rentfrow, Lamb, Potter, and Audretsch (2015) further argue that a high prevalence of entrepreneurship-prone personalities at the local level could push individuals to become entrepreneurs and could explain the geographic clustering of entrepreneurship. On the other hand, Stuetzer, Obschonka, Audretsch, Wyrwich, Rentfrowg, Coombes, Shaw-Taylor, and Satchell (2016) show that the historical presence of 6

large-scale industries is negatively correlated with current entrepreneurship and entrepreneurship culture, because an abundance of dependent work opportunities discouraged the formation of human capital and entrepreneurial skills, and favoured the development of formal and informal institutions that inhibited entrepreneurship. Our analysis suggests that the geographical distribution of religious beliefs might contribute to explain why entrepreneurship-prone personalities are more concentrated in certain areas than others.

3

Research Design A major issue in dealing with the ethical and cultural content of religions is that self-

identification with a certain religious creed does not necessarily imply internalization of its religious ethical principles, so the implications derived from a simple comparison of individuals’ reported affiliations may be misleading. Most of the religiousness-intensity indicators suggested by the literature to measure individual attachment to religious beliefs (McCleary and Barro, 2006), such as the frequency of attendance of religious services, weekly prayers, and donation of money and time to religious organizations, are likely to be endogenous to labour market outcomes. To overcome this problem, Nunziata and Rocco (2016) propose an indirect measure of attachment to religious principles: the condition of belonging to a religious minority. The rationale is that members of minority religions are more fervent believers than are members of majority religions, for three primary reasons. First, according to Bisin and Verdier (2000), Bisin and Verdier (2001), and Bisin, Topa, and Verdier (2004), religion, along with language, history and culture, is an important element of people’s identity that minorities seek to defend. Second, Stark, Finke, and Iannaccone (1995), Finke and Stark (1998), and Stark (1998) suggest that a minority religion’s clergy works hard to preserve its followers from the constant pressure of the surrounding dominant religion. Third, because of this pressure, only people who are strongly attached to their faith can resist the influence of the majority religion and maintain their attachment to the minority denomination. We investigate whether European data on religiosity confirm these hypothesis. We examine the behaviour of Protestant and Catholic minorities in Europe using ISSP Religion III data and find that members of religious minorities declare themselves to be more religious and to pray

7

more often and more regularly than their non-minority counterparts do. Minority Protestants tend to agree more with the principles of direct and unmediated relationship with God (one tenet of Protestantism) than other Protestants do, and minority Catholics believe more than other Catholics that salvation can be achieved by means of good works and that individual fate is not predetermined (one tenet of Catholicism). (See Online Appendix A for details.) Crucially, being born into a given religious minority can be considered predetermined because the rise and geographic distribution of minorities of both denominations in the regions of the former HRE follow the equilibrium found at the end of the religious wars of the sixteenth and seventeenth centuries that depended mainly on contingent historical conditions (Cantoni, 2015, Cantoni, 2012). The resulting distribution has persisted over the centuries, so it can be considered predetermined to individual labour market choices in current times. (See Online Appendix B for an extensive discussion.) These considerations justify our strategy of estimating the differential effect of Protestant versus Catholic ethic on entrepreneurship by comparing minorities. Below we detail the conditions under which such comparison is meaningful. We consider a sample of self-declared Protestant and Catholic individuals5 and our purpose is that of estimating the parameters of the following model Yirc = α0 + α1 Pirc + α2 Airc + α3 Airc Pirc + γ1 Xirc + µirc + εirc

(1)

where Yict is propensity for entrepreneurship of individual i living in region r of country c; Pirc is a dummy indicating whether i is Protestant; and A is i’s degree of adhesion to his or her religion. The model includes a set of observable controls Xirc — which include individual, regional and country controls that we specify below — and a set of unobservable controls µirc , such as social ties and networks, level of risk aversion, preferences and features of the residence place that contribute to the individual propensity for entrepreneurship. Without loss of generality, we let Aict take only two values, 0 corresponding to no adhesion and 1 to full adhesion to the ethical principles of i’s creed6 and we assume that:

5

We exclude atheists because they are likely to differ from religious individuals in some key respects, such as risk aversion, as we discuss in section 6.4 and Online Appendix D. We also exclude other religious denominations since their fellowship is very small in our data 6 We relax this assumption in the discussion in Online Appendix G.

8

Assumption 1. E(µirc |Pirc = 1, Airc = 1, Xirc ) = E(µirc |Pirc = 0, Airc = 1, Xirc ) = ω1

(2)

E(µirc |Pirc = 1, Airc = 0, Xirc ) = E(µirc |Pirc = 0, Airc = 0, Xirc ) = ω0 meaning that the expected contribution of unobservables, conditional to Aict and Xict is the same among Protestant and Catholics, and equal to ωA for A = 0, 1. Under assumption 1, the differential effect of Protestantism compared to Catholicism on the propensity for entrepreneurship is given by: E(Yirc |Pirc = 1, Airc = 1, Xirc ) − E(Yirc |Pirc = 0, Airc = 1, Xirc ) = α1 + α3

(3)

When the degree of adhesion to one’s religious ethic is zero, the likelihood that one will choose to be an entrepreneur is unaffected by religious affiliation and the reported religious affiliation carries no specific religious content and has no consequence on entrepreneurship. We refer to this feature as the “common intercept hypothesis” which can be empirically tested by checking whether the following expression E(Yirc |Pirc = 1, Airc = 0, Xirc ) − E(Yirc |Pirc = 0, Airc = 0, Xirc ) = α1

(4)

is not significantly different from zero. Since the degree of adhesion is unobservable, model (1) cannot be directly estimated. To operationalize it, we substitute the degree of adhesion with the condition of belonging to a minority, i.e. a religious group with large proportions of individuals with strong religious feelings, or a majority, i.e. a religious group with large proportions of people indifferent to the ethical principles of their confession. The precise empirical definitions of religious minority and majority are given in the next section. For now, it is sufficient to say that individual i belongs to a minority or a majority (or neither) depending on the proportion of individuals who reside in the same region as individual i and follow his or her religion. Accordingly, the condition of belonging to a minority or a majority is an individual characteristic that depends on the individual’s own religion. The resulting model can be empirically estimated but permits only a partial identification of 9

the parameters of interest αk , k = 1, 2, 3, that is however sufficient to draw conclusions on the effect of interest. More formally, we reformulate model (1) as: Yirc = β0 + β1 Pirc + β2 Mirc + β3 mirc + β4 Mirc Pirc + β5 mirc Pirc + δ1 Xirc + µirc + θirc

(5)

where the degree of adhesion A is replaced by Mirc , a dummy that equals one if individual i is part of any religious majority, and mirc , a dummy that equals one if individual i is part of any religious minority. Mirc Pirc and mirc Pirc are interactions between the majority/minority dummies and the Protestant dummy, and θirc is an IID error term. The pair (mirc , Mirc ) fully determines which religious group the individual belongs to: (1, 0) indicates a minoritarian, (0, 1) a majoritarian and (0, 0) an individual whose religion is neither majority nor majority in his region of residence. Obviously, mirc and Mirc cannot be simultaneously equal to 1 as one cannot be both minoritarian and majoritarian. Although minority and majority conditions are correlated with the degree of adhesion, they do not perfectly reflect it. Nonetheless, the estimates of the parameters βk are closely related to the parameters of interest αk . Let the proportion of individuals that display full adhesion to their religious ethic be πP in Protestant minorities, πC in Catholic minorities and ΠP and ΠC in Protestant and Catholic majorities, respectively, with πP > ΠP and πC > ΠC . Assume also that: Assumption 2. πP = πC = π

(6)

ΠP = ΠC = Π i.e. that there is the same distribution of adhesion to ethical principles among Protestant and Catholic minorities (resp. majorities). Furthermore, restate (and strengthen) assumption 1 as follows:

10

Assumption 3. E(µirc |Pirc = 1, Airc = 1, mirc , Mirc , Xirc ) = E(µirc |Pirc = 0, Airc = 1, mirc , Mirc , Xirc ) = ω1m,M (7) E(µirc |Pirc = 1, Airc = 0, mirc , Mirc , Xirc ) = E(µirc |Pirc = 0, Airc = 0, mirc , Mirc , Xirc ) = ω0m,M Compared to Assumption 1, Assumption 3 explicits that the means of unobservables may vary between minorities and majorities, but, strengthening the requirement of symmetry, need to be the same across denominations. For instance, religious discrimination or the intensity of social networks could be stronger among minorities than among majorities, but Assumption 3 requires that they contribute to the propensity to become an entrepreneur in the same manner in Protestant and Catholic minorities (resp. majorities). Assumptions 2 and 3 jointly formalize the identification hypothesis that Protestant and Catholic minorities (resp. majorities) are comparable. We will extensively discuss the validity of these assumptions in Section 6. Given model (1), by applying the law of total expectations, we obtain E(Yirc |Pirc = 1, mirc = 1, Mirc = 0, Xirc ) =

(8)

= π × E(Yirc |Pirc = 1, Airc = 1, mirc = 1, Mirc = 0, Xirc ) + +(1 − π) × E(Yirc |Pirc = 1, Airc = 0, mirc = 1, Mirc = 0, Xirc ) = = α0 + α1 + π(α2 + α3 ) + πω11,0 + (1 − π)ω01,0 + δ1 Xirc for Protestant minorities and similar expressions for the other religious groups. Finally, combining these quantities, we explicit the link between the expected difference in the propensity for entrepreneurship between Protestant and Catholic minorities (resp. majorities) derived from model (5) and the parameters of interest included in model (1).

E(Yirc |Pirc = 1, mirc = 1, Mirc = 0, Xirc ) − E(Yirc |Pirc = 0, mirc = 1, Mirc = 0, Xirc ) = = β1 + β5 = = α1 + πα3

11

(9)

E(Yirc |Pirc = 1, mirc = 0, Mirc = 1, Xirc ) − E(Yirc |Pirc = 0, mirc = 0, Mirc = 1, Xirc ) =

(10)

= β1 + β4 = = α1 + Πα3 Rearranging equations (9) and (10) we obtain: α1 = β1 + β4 −

α3 =

(β5 − β4 )Π π−Π

β5 − β4 π−Π

(11)

(12)

and the effect of Protestantism α1 + α3 is equal to: α1 + α3 = β1 + β5 + (1 − π)α3

(13)

Indeed, the observable quantity β1 + β5 , which is the differential effect between Protestant and Catholic minorities, as determined from model (5), captures the effect of Protestantism up to the bias −(1 − π)α3 . The latter tends to disappear as π approaches 1, that is, the better the condition of belonging to a minority proxies for a strong attachment to religion. Moreover, given equation (12), the sign of α3 coincides with the sign of β5 − β4 . The specification test statistics becomes α1 = β1 + β4 − Πα3

(14)

and the observable quantity β1 + β4 , which is the differential effect between Protestant and Catholic majorities according to model (5), captures the parameter of interest α1 up to the bias Πα3 , which tends to disappear as Π approaches 0, that is, the better the condition of belonging to a majority proxies for a weak attachment to religion. In both cases the sign of the bias depends on the sign of α3 that can therefore be inferred using β5 − β4 .

12

4

The Data Our sample includes the first four waves of the European Social Survey (ESS henceforth)

data, collected every other year, in 2002, 2004, 2006, and 2008.7 Selecting all geographic regions from the former HRE and excluding atheist and all non-Christian religious minorities left us with 19,964 individuals. We then focused on those who said they were active in the labour market and ended up with a sample of 8,966 individuals in 8 countries and 59 regions. The regions selected for our estimation sample belongs to Austria, Belgium, France, the Czech Republic, Germany, Luxembourg, Poland, and Slovenia and are highlighted in Figure 1. Table 1 shows the composition of our estimation sample. Regions are defined according to their populations. Depending on the country, ESS data provide different levels of regional aggregation, NUTS 1, NUTS2, and NUTS 3, the last of which is available only for a few countries. TABLE 1 AROUND HERE Once the regions in our data are outlined, we can define minorities and majorities. Our definitions respond to two considerations. On the one hand, a religious minority should collect a share of believers small enough to let its members perceive that a certain effort is required to preserve their religious identity and to internalize the religion’s ethical principles. On the other hand, a religious majority should be large enough to let its members perceive that they are part of an undisputedly dominant group. In practice, these requirements need to be matched with the data at use. Thus, our baseline specification assumes that minorities are the smallest religious group in a given region and, in any case, collect less than 25 percent of the residents. Symmetrically, majorities have to collect more than 50 percent of the residents in a region.8 We define minorities and majorities at the most disaggregate regional level available for each country but checked to ensure our results are robust to changes in the adopted criteria.9 The minority threshold corresponds to that in Nunziata and Rocco (2016) while the majority one is less stringent (i.e. 50 percent instead of 60 percent) because of the larger proportions of atheist 7

We decided to leave aside the waves collected in 2010 and 2012 because they attain to the years of the Great Recession whose effects on entrepreneurship might confound our findings. 8 These proportions are always defined with reference to the total population, atheists included. 9 See section 5.4 for a broader discussion on how results are sensitive to the choice of different regional aggregation levels.

13

in former HRE regions in early 2000 compared to Switzerland in 1990. Nevertheless, we test the validity of our findings for different thresholds in the robustness section and find that our empirical results are remarkably stable when we adopt other sensible thresholds.10 The outcome variable we consider in the paper is the individual entrepreneurial status, and more precisely whether the individual, being self-employed, risks on his or her own, selforganizes his or her working schedule, often manages dependent employees and is the residual claimant of the results of his or her ventures, in good and bad times (as in Audretsch, Bnte, and Tamvada, 2013 and Nunziata and Rocco, 2016). These business owners are distributed across all sectors, and their occupations are especially concentrated among directors and chief executives, wholesale and retail managers, professionals, crop and animal producers, restaurant and hotel managers, shop keepers, metal workers and builders. Most of the entrepreneurs in the sample are medium or high skilled, with only 9 percent being low skilled.11 The probability of being an entrepreneur at a given point in time depends on the probability of starting up a business sometime in the past and the business’ survival probability. Both events depend on individual characteristic that might be affected by religious ethics.12 In our sample, 18.4 percent of religious individuals are Protestants and 81.6 percent are Catholics. The Protestants who reside in regions where they are in the minority are 20.6 percent of the total; the corresponding figure for Catholics is around 3.9 percent. Therefore, minority Protestants and Catholics are 3.8 percent and 3.2 percent of our sample, respectively. Note that our sample includes also Protestants and Catholics that reside in regions where they do not satisfy neither the definition of majority nor that of minority. FIGURES 1, 2 AND 3 AROUND HERE 10

See Online Appendix E for additional details. Low skilled entrepreneurs are classified under ISCO categories 5, i.e. service workers and shop and market sales workers, and 9, elementary occupations. 12 Given the cross-sectional nature of European Social Survey data, alternative measures of entrepreneurship, such as entry in and exit from entrepreneurship (Caliendo, Fossen, and Kritikos, 2014), are not available. In addition, the entrepreneurial-prone personality discussed by Obschonka, Schmitt-Rodermund, Silbereisen, Gosling, and Potter (2013) cannot be defined because the ESS does not include questions on the Big-Five personality traits. Other measures of entrepreneurship have been proposed by the so called macro approach to the research on entrepreneurship that investigates the environmental and institutional determinants of entrepreneurship rather than the individual decision to become an entrepreneur. These are the startup activity rate, i.e. number of start-ups normalized by population (Obschonka, Stuetzer, Gosling, Rentfrow, Lamb, Potter, and Audretsch, 2015; Zelekha, Avnimelech, and Sharabi, 2014; Fritsch and Falck, 2007) or the local self-employment rate, i.e. the share of self-employed among the employed population (Stuetzer, Obschonka, Audretsch, Wyrwich, Rentfrowg, Coombes, Shaw-Taylor, and Satchell, 2016). 11

14

Figures 2 and 3 display the geographic distribution of Catholicism and Protestantism across former HRE regions (at NUTS 2 when available), while Figures 4 and 5 report the geographic distribution of Catholic and Protestant minorities. Catholicism is mostly prevalent in Austria, Belgium, the Czech Republic, Luxembourg, Poland, and Slovenia, while Protestantism is mostly concentrated in Germany, where the two confessions are almost identically represented. Europe has undergone a fast process of secularization in the last few decades. According to our ESS sample, in most regions of France and northern Europe in the period covered by our data, more than half of the residents declared that they were atheist. Similar patterns are provided by ISSP religion surveys collected in 1991, 1998 and 2008 that were especially designed to provide detailed information about various aspects of individual religiosity. FIGURES 4 AND 5 AROUND HERE Figure 6 displays the percentage of self-employed individuals on total employment in former HRE regions. The Czech Republic, France, and Slovenia have the smallest number of self-employed individuals. Predominantly Catholic countries like Poland have the largest proportion of self-employed individuals. Of course, such a simple correlation is not particularly informative since it is driven by unobservable country-specific factors that affect both religion and entrepreneurship. In our data, entrepreneurs are typically richer, as they are 7 percentage points more likely to living comfortably on present income (as opposed to coping, or finding it difficult or very difficult). They are also more likely to be represented on the top quartile of the income distribution. FIGURE 6 AROUND HERE We estimate model (5) by adopting a linear probability model specification. We are interested in the average effect of religion on entrepreneurship, so the linear estimator is preferable to Probit or Logit since it is consistent and it does not require a specific distributional assumption of the error term.13 Individual controls include age, gender, whether the individual is a foreigner, years of education, marital status, and a wealth variable that indicates whether the individual’s main source of income is financial. We also control for entrepreneurial family background, that is, whether the individual’s father was self-employed. This information was absent 13

For a similar approach see Angrist and Evans (1998).

15

in the Swiss census data used by Nunziata and Rocco (2016), even though family background is of particular relevance in our setting in order to distinguish between individuals who inherited a business (or perhaps a propensity to be self-employed) from those who did not have such an advantage. In addition, we include a full set of area (country or regional) and time fixed effects in order to control for unobservable and/or institutional country (or region) characteristics and common cyclical factors. In all specifications we include time-varying regional controls provided by Eurostat. These include regional GDP growth, population density, unemployment rate, number of doctors per capita (as a proxy for social development), educational attainment at the regional level, and the extension of motorways (as a proxy for infrastructure development). Summary statistics are reported in Table A.1 in Online Appendix F.

5

Empirical Findings

5.1

Baseline Estimates

We present our empirical findings starting from the estimates of our baseline specification of equation (5). This specification accounts for differences in country-specific institutions, regional and individuals characteristics that may affect entrepreneurship.14 Table 2 displays our estimates, reporting the differential effect of Protestant minority with respect to Catholic minority (β1 + β5 ) from equation (5) and the differential effect of Protestant majority with respect to Catholic majority (β1 + β4 ), i.e. our test of the “common intercept hypothesis”. According to our baseline estimates (column 1), the probability that an individual is an entrepreneur is 5.4 percentage points higher among Protestant minorities than it is among Catholic minorities. This is a sizeable effect, as only about 13 percent of working individuals in former HRE are self-employed. The estimated effect is close but somehow larger than what Nunziata and Rocco (2016) find for Switzerland when their geographical unit of reference are Swiss cantons, i.e. the geographical level closest (but still much smaller) to that available in our former HRE data. Our estimated 14

Standard errors are clustered at the regional level. Having 59 regions in our sample, we do not risk to over reject the null hypothesis (Cameron, Gelbach, and Miller, 2008). Nevertheless, our findings do not change when using robust standard errors.

16

coefficient is however almost identical to what Nunziata and Rocco find in their estimation limiting the sample to the region of Espace-Mitteland, at the canton level. Table A.2 in the Online Appendix reports the individually estimated parameters β of model (5). We note that β5 − β4 is positive and statistically significant,15 suggesting that α3 is also positive, given the derivation of section 3. We conclude that there is a significant and positive differential effect of Protestant ethic on entrepreneurship, and the differential effect of the Protestant minority that we have estimated is a lower bound of it (see equation 13). Similarly to Nunziata and Rocco (2016)’s findings, our specification test indicates that the differential effect of Protestantism versus Catholicism is not present when we compare religious majorities. The differential effect of the Protestant majority is positive but not significantly different from zero. Since this test statistics is the sum of α1 and a small upward bias (equation 14), we conclude that also α1 is not significantly different from zero.16 The baseline result is robust to the exclusion of individual level economic controls (e.g., education, marital status, whether the main source of wealth is financial and whether the respondent’s father was selfemployed) in column 2. In principle, these controls may also be affected by Protestantism and therefore they could confound the interpretation of our results in the setting of our research design. However, the point estimates are almost identical. TABLE 2 AROUND HERE Column 3 of Table 2 shows similar results from the estimated model using German regions only, i.e. the regions where Protestantism and Catholicism are almost evenly represented. Despite a much smaller sample, the estimated differential effect is still significant and the point estimate is only marginally smaller than the baseline. Similarly, our findings are confirmed when we exclude all formerly Communist regions (column 4),17 i.e. those areas that, for historical reasons, may be characterized by different institutions and entrepreneurial culture and a large share of atheists. 15

One-sided tests not reported but available upon request Π (β5 − β4 ). It follows that α1 is not More precisely, from equation (14) we obtain α1 = (β1 + β4 ) − π−Π Π significantly different from zero if π−Π is small enough. According to our estimates, this is the case for all Π π−Π < 6, i.e. whenever π > 1.17 · Π, a condition which looks quite plausible since it requires only a little higher proportion of individuals that display full adhesion to their religious ethic among minorities then among majorities. 17 These are the regions belonging to the Czech Republic, East Germany, Poland, and Slovenia. 16

17

In addition, we check whether our estimates are affected by a possible heterogeneity in the selection rate into atheism among individuals coming from Protestant and Catholic family backgrounds (Spenkuch, 2017).18 In section 6.5 below, we show that the proportion of individuals born within religious minorities who retain their parents’ religion is not statistically different between Protestant and Catholic minorities, and no significant difference in the degree of religious adhesion between minorities is detected. Nevertheless, in column 5 we exclude all regions where the share of atheists is higher than one third of the resident population.19 This amounts to include only those areas where the selection into atheism is weaker, and therefore the resulting Protestant and Catholic minorities are more likely to be homogeneous. Still, our point estimates are basically unchanged. Taken together, our findings indicate that religious ethic has a significant impact on the individual propensity for entrepreneurship when adhesion to religion is strong, as it is among minorities. The effect is absent when the adhesion is less strong, as it is among majorities. Interestingly, our findings are reinforced when we drop from the sample all individuals characterized by low skilled occupations (ISCO categories 5, service workers and shop and market sales workers, and 9, elementary occupations) in column 5. In this case, the point estimate is larger, indicating that the effect of Protestantism is particularly strong among high skilled workers.20 Regarding the effects of controls, the probability that an individual is an entrepreneur increases by almost 15 percentage points if his or her father was an entrepreneur, by 6 percentage points if the individual is male, and it is generally increasing in age and education. The effect of being a foreigner and whether the individual’s main source of income is financial are not statistically significant. Our estimates hence confirm the existence of a positive and statistically significant effect of Protestantism on entrepreneurship, also when the branch of Protestantism that is considered 18

Azzi and Ehrenberg (1975), Iannaccone (1992) and Spenkuch (2017) also show that economically successful individuals may be more likely to opt out of religion. However, in our setting this would constitute a problem only if the rate of opting out among successful individuals would be different among Protestant and Catholic minorities. 19 The regions in reduced sample are Austria excluding Wien, Luxembourg, Baden-W¨ urttemberg, Bayern, Hessen, Nordrhein-Westfalen, Rheinland-Pfalz, Saarland in Germany and Dolno´slaskie, Lubuskie, Zachodniopomorskie in Poland. 20 If we use a looser definition of low skilled workers, i.e. we also include craft and related workers and plant and machine operators and assemblers, then our sample is reduced by more than a half and the point estimate is even larger (not reported, but available upon request).

18

is Lutheranism. The next sections provide a set of robustness checks of the validity of our estimates in former HRE regions.

5.2

Robustness to Migration Patterns

This section addresses the potential impact of recent waves of migration on the geographic distribution of religious beliefs across Europe. We claimed that the geographic pattern of Christian minorities across the regions of the former HRE is historically determined since it follows the events generated by the Reformation during the sixteenth and seventeenth centuries.21 A comparison of historical information with the actual distribution of minorities in selected areas of the HRE seems to exclude that cumulative historical migration patterns may have significantly altered the map of the incidence of denominations and minorities across Europe. (See the evidence discussed in section B.2 of the Online Appendix.) However, we cannot exclude a priori the possibility that contemporary migration patterns could be endogenously driven by religious factors as well as by the individual propensity to start or export a business abroad. If such is the case, the estimated relationship between religion and entrepreneurship may be biased. We check for the effect of migration by excluding all individuals who are immigrants (i.e., either non-nationals or those born abroad) and recalculating our minority indicators. Table 3 shows the new estimates in columns 1 and 2. The differential effect of Protestantism is still positive and significant and, at 4.8 percentage points, is close to what we found in our baseline estimation. Again, we find no significant effect when we compare majority individuals. In addition, when we compare the immigrants’ religion with the predominant religion of the host country, we find no clear correlation, a further indication that our findings are not confounded by migration-related factors. TABLE 3 AROUND HERE

21

See the discussion in Online Appendix B.

19

5.3

Sectoral Composition and Unpaid Family Workers

Table 3 also checks whether sectoral composition affects our estimates. We control for sectoral composition at the individual level because of the possibility that entrepreneurship patterns differ across industries. Accounting for the fifteen main NACE sectors of activity, we estimate a positive and significant differential effect of Protestantism that is only marginally smaller than that found previously, that is, around 4.5 percentage points (column 3). No significant effect is found for majorities. Special attention should be paid to agriculture, as the agricultural sector is traditionally characterized by a comparatively high concentration of self-employed individuals, and agricultural communities are traditionally more religious than urban communities are. When we exclude from our sample all individuals who work in agriculture, which amounts to around 3.3 percent of our observations, our point estimate remains strongly statistically significant and close to our previous estimate, at 5.1 percentage points (column 4). Next, we exclude self-employed individuals who are unpaid family workers, that is, those whose compensation is in a form other than wages, such as an increase in business value. In our sample only around 1 percent of respondents (around 400) fit this category, and the effect of excluding them is marginal (column 5).

5.4

Robustness to Regional Fixed Effects and Alternative Calculations of Minorities

Table 4 shows that our results hold when we include NUTS 1 regional fixed effects instead of country fixed effects (column 1) as well as when we consider only regions in which respondents from both religious denominations are represented (column 2). When minorities are calculated at a NUTS 1 regional level for all countries,22 the estimated effect does not change (column 3), and similar findings are estimated when minorities are calculated at the NUTS 2 level (column 4).23 TABLE 4 AROUND HERE 22

Whenever ESS provides information on respondents’ regions of residence at the NUTS 2 and 3 levels, we aggregate at NUTS 1 and calculate which religious denomination, if any, is the minority. 23 The level of geographical aggregation remains at NUTS 1 for those countries that do not report NUTS 2, while it is aggregated to NUTS 2 for those countries reporting NUTS 3.

20

Finally, we estimate a model on individuals who are employed instead of just active. Around 4.6 percent of individuals in our baseline sample are unemployed. Since we know their occupation and employment status (if any) during their last employment, and since we are interested in the causal effect of religious ethic on occupational outcomes, our baseline estimations include respondents who were unemployed at the time of the survey. However, in principle, unemployed individuals may differ from employed individuals based on unobservable dimensions. Nevertheless, when we exclude unemployed respondents, our point estimates are very similar (column 5). In addition, all of our findings remain valid when we estimate our baseline model on a larger sample that includes retired individuals for whom we know their last job before retiring (not reported).

6

Discussion of Our Identification Assumptions In this section we discuss the validity of the assumptions (2) and (3) at the basis of our

identification strategy.

6.1

The Geographic Distribution of Minorities.

Suppose that the territories that are more inclined to commercial activities may have turned to Protestantism in anticipation of its supposed growth-promoting potential. Or, on the other hand, they may have remained Catholic to avoid the costs and risks of an institutional transition. Alternatively, suppose that nowadays the regions where Protestant and Catholic minorities reside differ in terms of institutional characteristics that are more or less favourable to entrepreneurship. If these hypothesis were true, then by comparing Protestant and Catholic minorities we would capture also such differences and our results would be spurious. To address these concerns, in this section we investigate whether the regions where Protestant and Catholic minorities live are similar in all respects relevant to entrepreneurship. An important argument in favour of comparability is that religious minorities in regions of the former HRE are historically determined and that their rise is related to the spread of Protestantism between the sixteenth and seventeenth centuries.24 24

Online Appendix B reports extensive evidence confirming that the current distribution of religious minorities closely reflects that which emerged at the end of the religious wars, complementing the discussion in Cantoni (2015) and Spenkuch (2010).

21

Moreover and more importantly, according to Cantoni (2012, 2015) the only variables that significantly predict a territory’s adoption of Protestantism are the distance from Wittenberg, home of the Reform, and whether the territory was formerly an ecclesiastical domain.25 Cantoni then argues that the determinants of conversion to Protestantism were mainly contingent on the strategic conditions at the time of the Reform, rather than on economic considerations, and concludes that Protestant and Catholic territories were similar in terms of their economic potential at the time of the Reform. Confirming this conclusion, if pre-Reform-era characteristics mattered, we should observe a differential propensity for entrepreneurship also between religious majorities. For instance, if the territories that were more inclined to commercial activities had turned to Protestantism for the most part, we should observe a positive effect of being a majority Protestant on entrepreneurship. However, this is not what we find in our estimates, as we find only a difference between Protestant and Catholic minorities and no difference between majorities. As a further check, we performed a placebo test, by randomly reshuffling individual religious affiliation within each region. This reshuffling preserves the market shares of the religions and the geographic distribution of majorities and minorities. We estimated model (5) on the reshuffled data and checked whether a significant effect of religion resulted, which would indicate that the effect we obtained from the baseline model is due to confounders, not to religion. We repeated this operation 1000 times, and the effect of Protestantism was significantly different from zero in less than four percent of the repetitions. Therefore, we can reject the possibility of a spurious effect.

6.2

Minorities’ Social Networks and Entrepreneurship

Since the condition of being in a minority generates stronger networks and higher social capital than does the condition of being in the majority, it could also favour entrepreneurship through these channels rather than through religious ethic. Social networks tend to be stronger among minorities because cooperation is easier to achieve among smaller communities, which tend to share similar values and cultural traits (McPherson, Smith-Lovin, and Cook, 2001). Minority religions are typically over-represented among entrepreneurs, as suggested by the 25

The distance from Wittenberg is understood as distance from Saxony, the powerful German state that first supported Luther’s ideas and guaranteed military protection against possible reactions by the HRE.

22

“middleman theory” (Bonacich, 1973): either minorities act as mediators between other social groups, or they demand mediation. Historically, Jews developed commercial and financial networks in the small communities scattered among European cities (Botticini and Eckstein, 2005), but it was not the Jewish religion alone that favoured entrepreneurship. The conditions that the religion is likely to have favoured, such as close connection and trust among its members, also played a part (Dana, 2006). Dana (2009) reviewed several examples of financial, employment, and information networks that emerge between people of the same religion, while Ellison, Krause, Shepherd, and Chaves (2009) and the references therein suggested that small congregations are much more likely than large communities to provide their members with support and protection in the case of negative shocks. Therefore, a strong social connection could be the key to successful entrepreneurship. We use data from the ISSP survey on Social Networks II collected in 2001 to determine whether there is a systematic difference between Catholic and Protestant minorities in terms of social networking. After considering a battery of social network indicators related to respondents’ network size, participation in voluntary associations, potential for financial exchange with relatives and friends, social support, and generalized trust, we find no evidence of either a systematic difference between minorities in terms of social connection, or even a difference between minorities and non-minorities, as detailed in Online Appendix C.

6.3

Entrepreneurship as a Way out of Discrimination

Intense social ties within minority groups can favour entrepreneurship, but entrepreneurship could also be a way out of discrimination. In this subsection we consider whether religious majorities discriminate against religious minorities. It is commonly maintained that the Catholic Church is intransigent in condemning norms, behaviours, and conducts that are contrary to its principles, whereas the less hierarchical and more dispersed Protestant churches are considered to be much more open and inclusive.26 If 26

However, the supposed higher level of tolerance of Protestant churches and Protestant people is doubtful. According to Kaplan (2007), extreme intolerance, which was equally common across denominations and countries, lasted for at least 150 years after the end of the religious wars. Protestants were often deeply intolerant not only of Catholics but also of each other (as confirmed by the burning of Servetus, a Spanish Protestant, in Calvin’s Geneva). In Britain and Ireland, the civil war, the spread of Puritanism, the Catholic intolerance of James II, and the anti-Catholic Gordon riots are examples of intolerance from both sides.

23

such were actually the case, Catholic individuals should be more intolerant toward other faiths or beliefs than Protestants are. Protestant minorities that are surrounded by Catholic majorities should then be over-represented among entrepreneurs, since entrepreneurship may represent an option against discrimination and minority Catholics who live alongside open-minded and tolerant Protestant majorities should be similarly over-represented as non-entrepreneurs. As a result, there should be comparatively more entrepreneurs among Protestant minorities than among Catholic minorities, a situation that would have less to do with religious ethic being conducive to entrepreneurship than with the majorities’ attitude towards tolerance. In addition, we would not find differences between religious majorities in terms of entrepreneurship because neither would be discriminated against. In order to analyse patterns of religious discrimination in the present day, we use data from all European countries where clear patterns of Protestant and Catholic majorities and minorities exist. Our analysis of ISSP Religion III data reveals that 82 percent of Catholics and 75 percent of Protestants in Europe agree with the principle that all religions deserve respect, and about 80 percent of the members of both religions would accept the idea of a close relative marrying a person from another religion. The ESS data has specific information about discrimination, as respondents report whether they are part of a group that experiences discrimination and whether that discrimination is due to religious reasons. We perform a set of independent-sample t-tests that compare the degree of discrimination between Protestants and Catholics in our estimation sample. The tests are performed to compare all Protestants and all Catholics, as well as to compare minority and majority Protestants and Catholics only. The p-values of the tests, shown in Table A.6 in the Online Appendix, indicate that there are no significant differences between denominations in the degree of discrimination perceived by the survey respondents, even among minorities and majorities.

6.4

Risk Aversion as an Explanatory Factor

Another issue worth investigating is the role of risk aversion, which is an important ingredient in entrepreneurship. (See Kihlstrom and Laffont, 1979 and Ekelund, Johansson, Jarvelin, and Lichtermann, 2005, among many others.) In principle, being Catholic, Protestant, or atheist could be correlated with several attributes of individual preferences because religion and 24

some character traits are learned within the family. For instance, the growing literature of endogenous preferences suggests that time preferences (Doepke and Zilibotti, 2005) and trust (Guiso, Sapienza, and Zingales, 2008) are partly decided and shaped by parents, so it may be reasonable to suppose that risk aversion is also learned in the family. If religious affiliation and the propensity to choose entrepreneurship are affected by a predetermined level of risk aversion, then our estimates may be more difficult to interpret than they would be otherwise. We focus on Germany and use SOEP 2004 data to determine whether there are differences between Catholics and Protestants in terms of risk aversion. Our choice of German data is motivated by the fact that the SOEP provides several measures of individual risk aversion, along with information on respondents’ religions and family religious backgrounds. The overall message of the analysis, discussed in detail in Online Appendix D, is that there are no systematic differences in Catholics’ and Protestants’ levels of risk aversion. This result is reassuring, as any difference in the level of entrepreneurship between the two religions cannot be ascribed to omitting the attitude toward risk.

6.5

Degree of Adhesion among Minorities and Majorities

The identification assumption is that Protestant and Catholic minorities’ average degrees of adhesion to their underlying religious ethical tenets are similar. A number of indirect pieces of evidence suggest that this assumption is corroborated by the data. First, the proportion of individuals born within religious minorities who retain their parents’ religion (i.e., those who resist the pressure of the surrounding religious majority) is similar between Protestant and Catholic minorities (69 percent for Catholics and 66 percent for Protestants, a difference that is not statistically significant).27 Second, using ISSP Religion III data, we tested whether Protestant and Catholic minorities differ in terms of level of adhesion based on certain religious outcomes. We find no significant difference in the degree of religious adhesion between minorities.28 Third, again with ISSP Religion III data, we estimated the proportions of individuals who reported believing in life after death, the main tenet of Christianity, among those individuals who reported being Protestant or Catholic and belonging to either minorities or majorities.29 These proportions should serve as valid proxies for the parameters π and 27

See also the discussion in Online Appendix B.1. See Table A.8 in the Online Appendix. 29 These figures are calculated using ISSP population weighted data.

28

25

Π introduced in section 3. Our results show that 47.9 percent and 48.0 percent of minority Protestants and minority Catholics, respectively, believe in life after death, compared to 18.5 percent and 25.8 percent among majority Protestant and Catholics, respectively. When asked about whether they believe in hell, an aspect of Christian theology that lost considerable attention over the last century, 29.1 percent and 28.7 percent of respondents responded affirmatively among Protestant and Catholic minorities, respectively, against 9.0 percent and 8.0 percent among Protestant and Catholic majorities, respectively. We observed a clear alignment in the proportions of Protestant and Catholic majorities who responded that they believed in heaven (16.5% and 13.0%, respectively) and some discrepancy among Protestant and Catholic minorities (37.0% and 53.9%, respectively).

7

Conclusions We analyse the effect of Protestantism on entrepreneurship in former HRE regions, following

the approach suggested in Nunziata and Rocco (2016) for Switzerland. We identify the implications of alternative ethical norms embodied by religious denominations on entrepreneurship on the basis of the minority status of religious denominations across those regions. We exploit religious minorities’ stronger degree of attachment to religious ethic compared to that of religious majorities and the predetermined historical determination of the distribution of religious denominations that took place at the time of the Reformation in the HRE. In addition, we rely on the homogeneous general economic and cultural environment to which Protestant and Catholic minorities are subject to in those regions. Our findings suggest that religious ethic may play a significant role in explaining a facet of economic performance. We identify a significant and positive effect of Protestantism on entrepreneurship. Protestants are found 5 percentage points more likely than Catholics to be entrepreneurs. This effect is even larger than that found in Nunziata and Rocco (2016) on Swiss census data, confirming that the effect of Protestantism on entrepreneurship is not limited to Switzerland but can also be detected in the larger area of the former HRE, at the core of central Europe. In addition, we show that the effect of Protestantism on entrepreneurship is not limited to its Calvinist and Zwinglian branches mostly common in Switzerland, but it is also shared by its Lutheran branch typical of former HRE regions.

26

This effect is not the result of a correlation between religious denomination and socioeconomic characteristics, such as education, economic development, sectoral composition, or institutions at the regional or country level. More importantly, our ESS data enable to show that our results are robust to the inclusion of entrepreneurial family background. This is an important addition, since we are able to distinguish between the autonomous decision to be an entrepreneur from the effect of inheriting a family business. Our findings are in line with Arrunada (2010), who shows that Protestants work longer hours and have a stronger sense of the rule of law than Catholics do, Basten and Betz (2013), who find that Protestants have a stronger work ethic than Catholics, and Hillman and Potrafke (2016), who find a correlation between the country aggregate share of Protestants and economic freedom. More importantly, our estimates confirm the findings by Spenkuch (2017) who, exploiting the historical persistence of religious denominations in Germany, shows that religious differences in individuals’ values can account for most of the positive effects of Protestantism on hours worked and self-employment. Our findings are also robust across alternative specifications and to a number of robustness checks, including a placebo test, and they do not depend on the inclusion of specific geographic clusters in the sample. We provide a broad discussion of the validity of our model by producing a series of empirical tests of the assumptions behind our identification strategy: we test whether Protestant and Catholic minorities are characterized by social networks of different intensity, whether they are subject to religious discrimination, the role of risk aversion, and how the degree of adhesion differs among minorities and majorities. In addition, we expand Spenkuch’s (2010) analysis by documenting the historical stability of religious minorities in the former HRE regions of Germany. Our findings on members of majority religions suggest that we can exclude any systematic difference between individuals who generically define themselves as Protestants or Catholics but are less observant than are individuals who practice minority religions, as no difference is found in any of our specifications between Protestant and Catholic majorities after controlling for a long battery of individual and regional covariates. Only when internalization of ethical norms and values is high, as is often the case among minorities, does a significant difference in the impact of the two religions’ ethical norms emerge. This result suggests that a historical, long-lasting religious tradition does not influence 27

entrepreneurship per se but that a genuine individual attachment to a specific religious culture inherited through the family (in addition to all the other elements we highlighted in our analysis, including entrepreneurial family background) is what matters in whether an individual will choose entrepreneurship. What are the specific normative channels through which Protestantism is more favourable to entrepreneurship than Catholicism? We can only speculate at this stage. The emphasis of Protestantism on the individual, unmediated relationship with God seems to be correlated to the emergence of a strong sense of self and self-esteem, two important engines of entrepreneurship. Moreover, the Protestant ethic may favour the moral approbation toward success and wealth accumulation that could push individuals toward entrepreneurship. Since social approbation develops only when the closely surrounding community shares the same Protestant ethic, this interpretation fits well with the finding that Protestantism is more conducive to entrepreneurship within minority denominations, where people tend to be more strictly observant than they are in majority groups.

28

References Andersen, T. B., J. Bentzen, C.-J. Dalgaard, and P. Sharp (2013): “Pre-Reformation Roots of the Protestant Ethic,” Working Paper 137, Centre for Competitive Advantage in the Global Economy, Department of Economics, The University of Warwick. 1 Angrist, J., and W. Evans (1998): “Children and Their Parents Labor Supply: Evidence from Exogenous Variation in Family Size,” American Economic Review, 88, 450–477. 13 Arrunada, B. (2010): “Protestants and Catholics: Similar Work Ethic, Different Social Ethic,” Economic Journal, 120, 890–918. 1, 2, 7 Audretsch, D. B., W. Bnte, and J. P. Tamvada (2013): “Religion, social class, and entrepreneurial choice,” Journal of Business Venturing, 28, 774789. 4 Azzi, C., and R. Ehrenberg (1975): “Household Allocation of Time and Church Attendance,” Journal of Political Economy, 83, 27–56. 18 Basten, C., and F. Betz (2013): “Beyond Work Ethic: Religion, Individual, and Political Preferences,” American Economic Journal: Economic Policy, 5(3), 67–91. 1, 7 Becker, S., and L. Woessmann (2009): “Was Weber Wrong? A Human Capital Theory of Protestant Economic History,” The Quarterly Journal of Economics, 124(2), 531–596. 1, 2 Becker, S. O., S. Pfaff, and J. Rubin (2016): “Causes and consequences of the Protestant Reformation,” Explorations in Economic History, 62(C), 1–25. 3 Bisin, A., G. Topa, and T. Verdier (2004): “Religious Intermarriage and Socialization in the United States,” Journal of Political Economy, 112(3), 615–664. 3, A Bisin, A., and T. Verdier (2000): “Beyond The Melting Pot: Cultural Transmission, Marriage, And The Evolution Of Ethnic And Religious Traits,” Quarterly Journal of Economics, 115(3), 955–988. 3, A (2001): “The economics of cultural transmission and the dynamics of preferences,” Journal of Economic Theory, 97, 298–319. 3

29

Bonacich, E. (1973): ““A Theory of Middleman Minorities”,” American Sociological Review, 38(5), 583–594. 6.2 Botticini, M., and Z. Eckstein (2005): “Jewish Occupational Selection: Education, Restrictions, or Minorities?,” The Journal of Economic History, 65(04), 922–948. 6.2 Caliendo, M., F. Fossen, and A. S. Kritikos (2014): “Personality characteristics and the decisions to become and stay self-employed,” Small Business Economics, 42, 787814. 4, 2, 12 Cameron, A. C., J. B. Gelbach, and D. L. Miller (2008): “Bootstrap-Based Improvements for Inference with Clustered Errors,” The Review of Economics and Statistics, 90(3), 414–427. 14 Cantoni, D. (2012): “Adopting a New Religion: the Case of Protestantism in 16th Century Germany,” The Economic Journal, 122(560), 502–531. 3, 6.1 (2015): “The economic effects of the Protestant Reformation: Testing the Weber hypothesis in the German Lands,” Journal of the European Economic Association, forthcoming. 1, 3, 24, 6.1, B.1 Cohen, A., and P. Hill (2007): “Religion as Culture: Religious Individualism and Collectivism Among American Catholics, Jews, and Protestants,” Journal of Personality, 75(4), 709–742. 2 Costa, P. T., and R. R. McCrae (1992): Revised NEO Personality Inventory (NEO PI-R) and NEO Five-Factor Inventory (NEO-FFI). Psychological Assessment Resources, Odessa. 2 Dana, L. P. (2006): “A historical study of the traditional livestock merchants of Alsace,” Food Journal, 108(2), 586–598. 6.2 (2009): “Religion as an explanatory variable for entrepreneurship,” Entrepreneurship and Innovation, 10(2), 87–89. 6.2 Doepke, M., and F. Zilibotti (2005): “Social Class and the Spirit of Capitalism,” Journal of the European Economic Association, 3(2-3), 516–524. 6.4 30

Eaton, D. (2013): “The Economists of the Reformation: An Overview of Reformation Teaching Concerning Work, Wealth, and Interest,” SAGE Open, pp. 1–9. 2 Ekelund, J., E. Johansson, M.-R. Jarvelin, and D. Lichtermann (2005): “Selfemployment and risk aversion–evidence from psychological test data,” Labour Economics, 12(5), 649–659. 6.4 Ellison, C. G., N. M. Krause, B. C. Shepherd, and M. A. Chaves (2009): “Size, Conflict, and Opportunities for Interaction: Congregational Effects on Members’ Anticipated Support and Negative Interaction,” Journal for the Scientific Study of Religion, 48(1), 1–15. 6.2 Finke, R., and R. Stark (1998): “Reply to Olson: Religious choice and competition,” American Sociological Review, 63, 761–766. 3, A Fritsch, M., and O. Falck (2007): “New Business Formation by Industry over Space and Time: A Multidimensional Analysis,” Regional Studies, 41(2), 157–172. 12 Furnham, A. (1987): “The protestant work ethic: a review of the psychological literature,” European Journal of Social Psychology, 14, 87–104. 2 Guiso, L., P. Sapienza, and L. Zingales (2008): “Alfred Marshall Lecture Social Capital as Good Culture,” Journal of the European Economic Association, 6(2-3), 295–320. 6.4 Herzog, J. J., and P. Schaff (1908): The New Schaff-Herzog Encyclopedia of Religious Knowledge. Funk and Wagnalls Company. 2 Hillman, A. L., and N. Potrafke (2016): “Economic Freedom and Religion: An Empirical Investigation,” Public Finance Review, forthcoming. 7 Holmes, T. J., and J. A. Schmitz (1990): “A theory of entrepreneurship and its application to the study of business transfers,” Journal of Political Economy, 87, 265–294. 2 Iannaccone, L. R. (1992): “Sacrifice and Stigma: Reducing Free-riding in Cults, Communes, and Other Collectives,” American Journal of Sociology, 100, 271–291. 18

31

Judge, T. A., C. A. Higgins, C. J. Thoresen, and M. R. Barrick (1999): “The big five personality traits, general mental ability, and career success across the life span,” Personnel Psychology, 52(3), 621–652. 4 Kaplan, B. J. (2007): Divided by Faith: Religious Conflict and the Practice of Toleration in Early Modern Europe. Harvard University Press, Cambridge, Mass and London. 26 Kihlstrom, R., and J. Laffont (1979): “A General Equilibrium Theory of Firm Formation Based on Risk Aversion,” Journal of Political Economy, 87, 719–748. 6.4 Kilhstrom, R. E., and J. Laffont (1979): “A general equilibrium entrepreneurial theory of firm formation based on risk aversion,” Journal of Political Economy, 87, 719–749. 2 Lucas, R. (1978): “On the size distribution of business firms,” Bell Journal of Economics, 9, 508–523. 2 Mayer, A., and H. Sharp (1962): “Religious Preference and Worldly Success,” American Sociological Review, 27(2), 218–227. 2 McCleary, R. M., and R. J. Barro (2006): “Religion and Economy,” Journal of Economic Perspectives, 20(2), 49–72. 3, 33 McPherson, M., L. Smith-Lovin, and J. M. Cook (2001): “Birds of a Feather: Homophily in Social Networks,” Annual Review of Sociology, 27(1), 415–444. 6.2 Nunziata, L., and L. Rocco (2016): “A Tale of Minorities: Evidence on Religious Ethic and Entrepreneurship,” Journal of Economic Growth, 27, 189–224. 1, 3, 4, 5.1, 7, A Obschonka, M., E. Schmitt-Rodermund, R. K. Silbereisen, S. D. Gosling, and J. Potter (2013): “The Regional Distribution and Correlates of an Entrepreneurship-Prone Personality Profile in the United States, Germany, and the United Kingdom: A Socioecological Perspective,” Journal of Personality and Social Psychology, 105(1), 104122. 2, 12 Obschonka, M., M. Stuetzer, S. D. Gosling, P. J. Rentfrow, M. E. Lamb, J. Potter, and D. B. Audretsch (2015): “Entrepreneurial Regions: Do Macro-Psychological Cultural Characteristics of Regions Help Solve the Knowledge Paradox of Economics?,” PLoS ONE, 10(6), e0129332. 2, 12 32

Parker, S., and Y. Belghitar (2006): “What happens to nascent entrepreneurs? An econometric analysis of the PSED,” Small Business Economics, 27, 81–101. 2 Rentfrow, P. J., S. D. Gosling, and J. Potter (2008): “A Theory of the Emergence, Persistence, and Expression of Geographic Variation in Psychological Characteristics,” Perspectives on Psychological Science, 3(5), 339–369. 2 Spenkuch, J. L. (2010): “The Protestant Ethic and Work: Micro Evidence from Contemporary Germany,” SOEPpapers on Multidisciplinary Panel Data Research 330, DIW Berlin, The German Socio-Economic Panel (SOEP). 1, 24, 7, B.2, 39 (2017): “Religion and Work: Micro Evidence from Contemporary Germany,” Journal of Economic Behavior and Organization, 135, 193–214. 1, 5.1, 18, 7 Stark, R. (1998): “Catholic contexts: Competition, commitment and innovation,” Review of Religious Research, 39, 197–208. 3, A Stark, R., R. Finke, and L. Iannaccone (1995): “Pluralism and piety: England and Wales, 1851,” American Sociological Review, 34, 431–444. 3, A Stuetzer, M., M. Obschonka, D. B. Audretsch, M. Wyrwich, P. J. Rentfrowg, M. Coombes, L. Shaw-Taylor, and M. Satchell (2016): “Industry structure, entrepreneurship, and culture: An empirical analysis using historical coalfields,” European Economic Review, 86, 52–72. 2, 12 Tilgher, A. (1958): Homo faber: Work through the ages. Regnery, Chicago. 2 Weber, M. (1904): The Protestant Ethic and the Spirit of Capitalism, vol. reprinted 2001. Roxbury Publishing Company. 1 Zelekha, Y., G. Avnimelech, and E. Sharabi (2014): “Religious institutions and entrepreneurship,” Small Business Economics, 42, 747767. 12 Zhao, H., and S. E. Seibert (2006): “The Big Five Personality Dimensions and Entrepreneurial Status: A Meta-Analytical Review,” Journal of Applied Psychology, 91(2), 259271. 2

33

Figure 1: Regions of the former Holy Roman Empire Former Holy Roman Empire regions 0− 1−

NUTS1 and NUTS 2

34

Figure 2: Geographic distribution of Catholicism across regions of the former Holy Roman Empire % Catholics 0− 4− 8− 12− 16− 20− 24− 28− 32− 36− 40− 44− 48− 56− 60−

Source: authors’ elaboration on ESS data, 2002-2008. NUTS1 and NUTS 2

35

Figure 3: Geographic distribution of Protestantism across regions of the former Holy Roman Empire % Protestants 0− 4− 16− 20− 24− 28− 32− 40− 44− 48−

Source: authors’ elaboration on ESS data, 2002-2008. NUTS1 and NUTS 2

36

Figure 4: Geographic distribution of Catholic minorities across regions of the former Holy Roman Empire % Minority Catholics 0% 3−4% 4−5% 6−7% 7−8% 9−10% 18−19% >20%

Source: authors’ elaboration on ESS data, 2002-2008. NUTS1 and NUTS 2

37

Figure 5: Geographic distribution of Protestant minorities across regions of the former Holy Roman Empire % Minority Protestants 0% 0−1% 1−2% 2−3% 3−4% 5−6% 6−7% 16−17% 19−20%

Source: authors’ elaboration on ESS data, 2002-2008. NUTS1 and NUTS 2

Figure 6: Percentage of self-employed across regions of the former Holy Roman Empire % Self−Employed 2− 4− 6− 8− 10− 12− 14−

Source: authors’ elaboration on ESS data, 2002-2008. NUTS 1 and NUTS 2

38

Table 1: Countries of the Former Holy Roman Empire and baseline model sample size. Country 2002 2004 2006 2008 Total Austria 758 691 893 0 2342 Belgium 344 302 288 307 1241 Czech Republic 118 275 0 0 393 Germany 643 629 700 687 2659 France 0 0 48 45 93 Luxembourg 300 398 0 0 698 Poland 117 95 93 87 392 Slovenia 270 337 266 275 1148 Total 2550 2727 2288 1401 8966 Source: European Social Survey data, 2002-2008.

39

Table 2: Religious denomination and the propensity for entrepreneurship Min Prot - Min Cath s.e. Maj Prot - Maj Cath s.e.

(1) 0.054*** (0.011) 0.000 (0.022)

(2) 0.051*** (0.012) 0.000 (0.021)

(3) 0.050*** (0.007) 0.005 (0.022)

(4) 0.052*** (0.011) -0.010 (0.022)

(5) 0.053*** (0.011) -0.017 (0.026)

(6) 0.063*** (0.013) 0.008 (0.023)

Economic Controls Demographic Controls Regional Controls Country Fixed Effects Year Fixed Effects

YES YES YES YES YES

NO YES YES YES YES

YES YES YES YES YES

YES YES YES YES YES

YES YES YES YES YES

YES YES YES YES YES

Skills Regions

ALL ALL

ALL ALL

ALL GERM.

ALL WEST

Obs R-squared

8966 0.071

8966 0.039

2659 0.080

6520 0.083

ALL MED/HIGH LOW ATH. ALL 5124 0.072

8146 0.072

Clustered standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01. Dependent variable: dummy variable equal to one if the respondent is an entrepreneur. The table presents the estimated differential effect on the propensity for entrepreneurship of being a minority Protestant compared to being a minority Catholic (Min Prot-Min Cath). This quantity corresponds to the sum of parameters β1 + β5 in equation (5). Column 1 is the baseline model; column 2 excludes individual level economic controls; column 3 is estimated on German regions only; column 4 excludes former ex-Communist regions; column 5 include those regions where the atheist share is lower than one third of the population; column 6 excludes low skilled occupations (ISCO categories 5, service workers and shop and market sales workers, and 9, elementary occupations). All columns are estimated by a linear probability model and include country and time fixed effects. Standard errors are clustered at the regional level. Individual level controls include whether the respondent is a foreign national, age, gender, whether the respondent’s father was self-employed, years of education, marital status, and whether the respondent’s main source of income is financial. Regional controls include unemployment rate, GDP growth, the extent of motorways, population density, number of doctors per thousand residents, proportion of college graduated residents. Maj Prot-Maj Cath is the test of whether there is a significant difference in the propensity for entrepreneurship between majorities. This quantity corresponds to the sum of parameters β1 + β4 in equation (5).

40

Table 3: Religious denomination and the propensity for entrepreneurship: robustness checks 1. Min Prot - Min Cath s.e. Maj Prot - Maj Cath s.e.

(1) 0.048*** (0.012) -0.003 (0.020)

(2) 0.048*** (0.015) -0.004 (0.023)

(3) 0.045*** (0.015) -0.022 (0.019)

(4) 0.051*** (0.012) -0.028 (0.028)

(5) 0.053*** (0.011) -0.001 (0.022)

Economic Controls Sector of Activity Demographic Controls Regional Controls Country Fixed Effects Year Fixed Effects

YES NO YES YES YES YES

YES NO YES YES YES YES

YES YES YES YES YES YES

YES NO YES YES YES YES

YES NO YES YES YES YES

NO (BA) YES YES ALL

YES YES YES ALL

YES NO YES ALL

YES YES NO ALL

8213 0.072

7875 0.168

8671 0.058

8855 0.073

Immigrants NO (NN) Agricultural Sector YES Unpaid Family Workers YES Regions ALL Obs R-squared

8508 0.072

Clustered standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01. Dependent variable: dummy variable equal to one if the respondent is an entrepreneur. The table presents the estimated differential effect of being a minority Protestant compared to being a minority Catholic, on the propensity for entrepreneurship (Min Prot-Min Cath). This quantity corresponds to the sum of parameters β1 + β5 in equation (5). Column 1 excludes all non-national immigrants; column 2 excludes all born-abroad immigrants; column 3 controls for sector of activity at the individual level; column 4 excludes individuals whose main activity is in agriculture; and column 5 excludes unpaid family workers. All columns are estimated by a linear probability model and include country and time fixed effects. Standard errors are clustered at the regional level. Individual level controls include whether the respondent is a foreign national, age, gender, whether the respondent’s father was self-employed, years of education, marital status, and whether the respondent’s main source of income is financial. Regional controls include unemployment rate, GDP growth, the extent of motorways, population density, number of doctors per thousand residents, proportion of college graduated residents. Maj Prot-Maj Cath is the test of whether there is a significant difference in the propensity for entrepreneurship between majorities. This quantity corresponds to the sum of parameters β1 + β4 in equation (5).

41

Table 4: Religious denomination and the propensity for entrepreneurship: robustness checks 2. Min Prot - Min Cath s.e. Maj Prot - Maj Cath s.e.

(1) 0.054*** (0.012) 0.156 (0.038)

(2) 0.054*** (0.011) -0.004 (0.021)

(3) 0.053*** (0.011) -0.005 (0.019)

(4) 0.054*** (0.011) -0.000 (0.021)

(5) 0.058*** (0.013) 0.002 (0.022)

Economic Controls Sector of Activity Demographic Controls Regional Controls Country Fixed Effects NUTS 1 Fixed Effects Year Fixed Effects

YES NO YES YES NO YES YES

YES NO YES YES YES NO YES

YES YES YES YES YES NO YES

YES NO YES YES YES NO YES

YES NO YES YES YES NO YES

Regions Labour Market Status

ALL ACT

BOTH REL ACT

ALL ACT

ALL ACT

ALL EMPL

Minorities

NUTS 2/3

NUTS 2/3

NUTS 1

NUTS 2

NUTS 2/3

Obs R-squared

8966 0.072

8563 0.072

8978 0.071

8966 0.071

8561 0.073

Clustered standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01. Dependent variable: dummy variable equal to one if the respondent is an entrepreneur. The table presents the estimated differential effect of being a minority Protestant compared to being a minority Catholic, on the propensity to be an entrepreneur (Min Prot-Min Cath). This quantity corresponds to the sum of parameters β1 + β5 in equation (5). Column 1 includes NUTS 1 regional fixed effects instead of country fixed effects; column 2 is estimated only on regions where both religions are represented; minorities in column 3 are identified at the NUTS 1 regional level instead of NUTS 2 and NUTS 3; whenever possible, minorities in column 4 are identified at the NUTS 2 regional level instead of NUTS 3; and column 5 includes employed individuals only instead of active individuals. All columns are estimated by a linear probability model and include country and time fixed effects. Standard errors are clustered at the regional level. Individual level controls include whether the respondent is a foreign national, age, gender, whether the respondent’s father was self-employed, years of education, marital status, and whether the respondent’s main source of income is financial. Regional controls include unemployment rate, GDP growth, the extent of motorways, population density, number of doctors per thousand residents, proportion of college graduated residents. Maj Prot-Maj Cath is the test of whether there is a significant difference in the propensity for entrepreneurship between majorities. This quantity corresponds to the sum of parameters β1 + β4 in equation (5).

42