10ISSN 1183-1057 SIMON FRASER UNIVERSITY Department of Economics Working Papers 17-17 “Kinship, Fractionalization and Corruption” Mahsa Akbari, Duman BahramiRad, and Erik O. Kimbrough June 30, 2017 Kinship, Fractionalization and Corruption∗ Mahsa Akbari† Duman Bahrami-Rad‡ Erik O. Kimbrough§ This version: June 30, 2017 First version: October 3, 2016 Abstract By shaping patterns of relatedness and interaction, marriage practices influence the relative returns to norms of nepotism/favoritism versus norms of impartial cooperation. In-marriage (e.g. consanguineous marriage) yields a relatively closed society of related individuals and thereby encourages favoritism and corruption. Out-marriage creates a relatively open society with increased interaction between non-relatives and strangers, thereby encouraging impartiality. We report a robust association between in-marriage practices and corruption across countries and across Italian provinces. A stylized corruption experiment comparing subjects from two countries with divergent marriage patterns provides complementary evidence that the degree of impartiality varies with marriage patterns. JEL classifications: D7, D0, C9, J1 Keywords: corruption, fractionalization, institutions, mating patterns, consanguinity, experiments ∗ Kimbrough would like to thank the SSHRC Insight Grants Program (435-2015-0798) and the Simon Fraser University Teaching and Learning Centre Development Grants Program for funding this research. We would like to thank Quamrul Ashraf, Greg Dow, Simon Halliday, Alex Karaivanov, Jonathan Schulz, Bart Wilson, and participants at numerous seminars and conference presentations for helpful comments. We would also like to thank Yoram Halevy and Anujit Chakraborty for helping us to conduct some of our experiments at ELVSE at the University of British Columbia, as well as the Moasser Research Center, the Tehran Thought Club, the West Azerbaijan Ministry of Education and Urmia University for assistance with recruiting and collecting data in Iran. We also thank Babak Jahanshahi and Erasmo Papagni for helping us collect data on Italy. All remaining errors are our own. † PhD Candidate, Department of Economics, Simon Fraser University. ‡ PhD Candidate, Department of Economics, Simon Fraser University. § Corresponding Author: Associate Professor, Department of Economics, Simon Fraser University, email: ekimbrou@sfu.ca. “Now it appears, that in the original frame of our mind, the strongest attention is confin’d to ourselves; our next is extended to our relations and acquaintances; and ’tis only the weakest which reaches to strangers and indifferent persons. This partiality, then, and unequal affection, must not only have an influence on our behaviour and conduct in society, but even on our ideas of vice and virtue. . . ” (David Hume, 1740, A Treatise of Human Nature, Section 3.2.2.) 1 Introduction Norms of solidarity, fidelity and self-sacrifice in favor of kin, tribe and clan have often been praised as virtues, but these virtues may become vices when they conflict with the abstract rules and formal institutions of the modern political and economic system. In particular, favoring kin at the expense of others may lead to corruption, disrupting or subverting impartial institutions and hampering economic development. In this paper, we provide evidence that a history of consanguineous mating practices generates fractionalization between local, sub-ethnic groups, increases incentives for local favoritism, and thus encourages corruption. Previous studies of corruption and its effects on growth have explored the idea that ethnic heterogeneity (and concomitant fractionalization) may cause corruption when individuals favor members of their own ethnic group. Mauro’s (1995) influential study used ethnic fractionalization as an instrumental variable for corruption. Since then, several studies have investigated whether ethnic heterogeneity causes corruption, with mixed results. In support, Easterly and Levine (1997) found that ethnic fractionalization is positively correlated with corruption; La Porta et al. (1999), Treisman (2000) and Alesina et al. (2003) also found that fractionalization has a reduced-form relationship with corruption but reported non-robust results when controlling for other variables such as per capita income. However, Serra (2006) and Elbahnasawy and Revier (2012) found no significant effect of fractionalization on corruption. In addition to the cross-country studies, Glaeser and Saks (2006) and Dincer (2008) found a significant relationship between ethnic heterogeneity and corruption across US states. These contradictory results have encouraged skepticism (see e.g. Chuah et al., 2013). A typical regression model from the cross-country empirical studies is as follows: C = α + βEF + γX + u where C is a corruption index, EF is an ethnic fractionalization index1 and X is a set of in1 As a measure of heterogeneity within countries, empirical studies have used fractionalization indices from two sources: (I) the ethnolinguistic fractionalization index (often referred to as ELF) from the Atlas Narodov Mira compiled from sources in the former Soviet Union (Bruk and Apenchenko, 1964), and (II) the ethnic, linguistic, and religious fractionalization indices provided by Alesina et al. (2003). Both sources define fractionalization as the 1 dependent variables. The richest specifications in La Porta et al. (1999), and Alesina et al. (2003) include legal origins, religion, latitude, per capita income, country size and regional dummies. However, while the motivation for such analysis is intuitively appealing, there is no obvious theoretical justification for the model or for a causal effect of ethnic fractionalization on corruption. Why should individuals favor members of their ethnic group? We provide a framework, rooted in biological theory, that explicitly connects ethnicity, kinship and corruption. While the theory may be unfamiliar to economists, “the biologically based approach shares strengths in common with the best of economic theory; it is parsimonious, counter-intuitive, and falsifiable” (Cox and Fafchamps, 2007, p. 3759). The basic intuition comes from the biological notions of inclusive fitness and kin selection, which imply that genetic relatives, because they share genes with an interest in propagating themselves to the next generation, will sometimes be willing to incur costs to help one another (see e.g. Hamilton, 1964a,b). A model of kin selection implies that increases in relatedness can encourage corruption and favoritism. Since the relatedness of two randomly chosen co-ethnics is quite low, shared ethnicity per se is insufficient to foster corruption, helping explain the ambiguous findings noted above. Key to our argument is that different marriage practices will increase or decrease relatedness at a local level, thereby directly altering the incentives for corruption and favoritism. For instance, the offspring of a consanguineous marriage (e.g. marriage between two first cousins) share more genes with their parents, siblings and cousins than the offspring of a marriage between two unrelated individuals. As we illustrate in a simple model below (see Section 2), from the point of view of biological theory, all else equal, the former family has stronger incentives for favoritism and corruption at the expense of non-relatives than the latter. While the intuition is rooted in evolutionary theory, our argument is not that different groups have different genes and hence different behaviors.2 Instead, we assume that all humans possess the capacity for kin altruism as well as for cooperation with non-kin, and that societies differ in the degree to which operant social norms favor one or the other mechanism of cooperation. Since humans are social creatures reliant on cultural norms of cooperation and information-sharing for survival, selection pressures also operate on these norms, so that a society’s norms adapt to local conditions (Henrich, 2015). Our focus is on the effects of in- and out-marriage practices on norms related to corruption. probability that two randomly drawn individuals from a country’s population belong to two different groups. 2 In biology, models of kin selection are typically used to explain the evolution of altruism or cooperation, by showing that “genes for” such behaviors may provide a selective advantage to their carriers. While it is possible, in principle, that persistent differences in marriage patterns between populations could eventually lead to genetic differences between those populations (increasing the relative frequency of “genes for” kin altruism due to differential selection pressures; see Hamilton (1975) for a theoretical treatment, and see hbdchick (2014) for a provocative argument that this might apply to human populations), in practice, genetic change (or genetic difference) is not necessary to explain how marriage practices can alter incentives for kin altruism (and consequent corruption). 2 Our view is that the incentives produced by different marriage practices have influenced the evolution of social norms. Endogamous and consanguineous marriage (in-marriage) directly increase local relatedness, increasing the relative returns to local favoritism and corruption; while exogamous marriage practices (out-marriage) reduce local relatedness, increasing the relative returns to impartiality. Social norms are constructed on top of this lattice of biological relatedness, and thus changes in relatedness due to marriage practices can directly influence the evolution of norms.3 Moreover, on top of these direct effects via the mechanism of relatedness, there are also intuitive indirect mechanisms by which marriage practices can influence the relative returns to norms of favoritism and corruption vis-a-vis norms of impartial cooperation. In particular, different marriage patterns also typically result in different patterns of social interaction, reflecting complementarities between marriage practices and other social structures. Out-marriage encourages (indeed, requires) interaction with non-kin and strangers. As we discuss below, historical religious bans on consanguineous marriage were an important cause of migration, especially in agricultural societies where one male child inherited the family’s land. Siblings without property, especially females, had to migrate to find eligible marriage partners (Cavalli-Sforza et al., 2004), and we argue that this exchange of people across distances encouraged the development of norms of impartial cooperation with non-kin. On the other hand, in-marriage is associated with extensive interaction among local (and more closely related) in-group members (e.g. kin, clan, tribe), and relatively less interaction with strangers. Seen in this light, in-marriage practices generate another kind of fractionalization at a level of granularity finer than the ethnic or linguistic group. We refer to this as sub-ethnic fractionalization, and we argue that patterns of increased local relatedness and concomitant intense local interaction diminished the impetus to develop norms of impartial cooperation, instead favoring the development of norms of local favoritism (which may manifest as corruption).4 To see how sub-ethnic fractionalization can help account for some puzzling observations on 3 Many groups extend altruistic norms to affines, friends and other less-related individuals by adopting cultural practices that create fictive kinship, and it has been argued that these practices piggy-back on, or hijack, the evolved mechanisms for kin cooperation and extend them to non-kin (see Henrich, 2015, for a summary). Incest taboos present another example of how social norms can harness biological mechanisms: “Incest taboos are social norms that evolved culturally to regulate sex and pair-bonding between non-close relatives by harnessing innate intuitions and emotional reactions that originally arose via genetic evolution to suppress sexual interest among close relatives, especially siblings. By harnessing innate incest aversion and labeling distant relatives as ‘brothers’ and ‘sisters,’ cultural evolution seized a powerful lever to control human behavior, since incest taboos can strongly influence mating and marriage, and kin-based altruism can be extended through social norms. If you control mating and marriage, you get a grip on much of the larger social structure, and even aspects of people’s cognition and motivation” (Henrich, 2015, p. 153). See also Jones (2016) for on feedback between social norms and kin altruism. 4 Todd and Garrioch (1985) make a similar point: while exogamous marriage in Europe served as a model for impersonal bureaucratic relations by creating links between unrelated individuals, endogamy impedes the creation of impersonal relationships by cutting “horizontally across the vertical edifice of the state, undermining the system and producing what in conventional administrative terms is called corruption” (p. 146). 3 the relationship between ethnic (and linguistic) fractionalization and corruption, consider Table 1 derived from Alesina et al. (2003) which lists a few countries with low ethnic and linguistic heterogeneity, but relatively high levels of corruption.5 Although Yemen, Tunisia, Saudi Arabia and Bangladesh are relatively homogeneous in terms of ethnicity and language, they are highly fractionalized due to the presence of and competition between other close-knit kin-based and local groups such as extended families, tribes, clans, and religious groups (see e.g. Lewis, 2014, on clan structures in the Gulf of Aden). We argue that distinctive family structures and mating patterns generate sub-ethnic fractionalization (e.g. the preference for in-marriage in many African and Asian countries) and can help account for corruption in many societies, even those that are ethnically homogeneous. Fractionalization index Corruption index 0-5% Yemen Tunisia 0-20% ... 5-15% Saudi Arabia Bangladesh 20-40% ... 15-35% ... 35-55% ... 55-75% ... 40-60% Saudi Arabia Tunisia 60-80% Bangladesh 80-100% Yemen Table 1: Misspecification due to omission of sub-ethnic fractionalization. A second puzzle for the view that heterogeneity per se causes corruption can be seen in Table 2 (also derived from Alesina et al., 2003), which lists a number of countries, all of which are highly ethnically and linguistically fractionalized: Fractionalization index Corruption index 0-5% ... ... 5-15% ... ... 15-35% ... ... 0-20% Canada Luxembourg Switzerland 20-40% Belgium 40-60% 35-55% Switzerland Belgium Iraq Uzbekistan 60-80% Iran Pakistan 55-75% Canada Luxembourg Iran Pakistan 80-100% Iraq Uzbekistan Table 2: Misspecification due to conflation of ethnic and sub-ethnic fractionalization. Although, e.g., both Canada and Pakistan are ethno-linguistically heterogeneous, Canada has effective, impartial institutions; while Pakistan is quite corrupt. As above, the countries differ in the importance of sub-ethnic groups such as extended family, tribe and clan to social and political life. Pashtuns, one of the largest ethnic groups in Afghanistan and Pakistan “are said to having [sic] developed the world’s largest tribal society, . . . [with] sub-tribes, clans and sub-clans down to the local lineages and families” (Glatzer, 2002, p. 3). Similar arguments contrast Switzerland, Luxembourg, and Belgium on one hand to Iran, Iraq, and Uzbekistan on the other.6 Viewed 5 The fractionalization index in the table is the simple average of ethnic fractionalization and linguistic fractionalization from Alesina et al. (2003), and the range of the fractionalization index in the table is the same as Mauro (1995)’s ethnolinguistic fractionalization table. For the corruption index in the table, we used the 2014 Corruption Perception Index provided by Transparency International (http://www.transparency.org/research/cpi/overview). 6 See also Sailer (2004) on clans, corruption and state-building in Iraq in light of cousin marriage practices. 4 through the lens of marriage practice-driven sub-ethnic fractionalization, these observations can be understood. To make our argument about the incentives produced by in-marriage more precise, we introduce a stylized bribery model where a private agent offers a bribe to an official, and if the official accepts the bribe and makes a corrupt effort, a negative externality is imposed on third parties called citizens. This is a well-known model of corruption used in a number of laboratory studies.7 We employ a utility function that embeds the implications of inclusive fitness to illustrate how increases in relatedness (as would be produced by in-marriage) may encourage corruption. In a one-shot game, the subgame perfect equilibrium of the bribery game for payoff maximizing agents involves neither bribery nor corruption (nor the associated negative externalities), but when the private agent and the official are sufficiently related to one another, both bribery and corruption can be supported in equilibrium. The model thus shows how relatedness can influence the returns to corruption, and we use that intuition to motivate an empirical analysis of the relationship between in-marriage practices and corruption. Our empirical analysis combines data from population genetics, corruption, and comparative development studies to test the hypothesis that marriage practice-driven sub-ethnic fractionalization causes corruption using both cross-country and within-country regression analysis. As a measure of sub-ethnic fractionalization, we collect data on national and regional (within-Italy) rates of consanguineous (cousin) marriage. We find that consanguinity rates have a substantial and positive association with corruption, both across countries and within Italy, even after controlling for other “deep” determinants of comparative development. To complement our regression analyses and provide a test of the proposed mechanism, we also design a cross-cultural lab experiment comparing the bribery and corruption behavior of strangers, co-ethnics and kin in Canada and Iran, exploiting the fact that the two countries are both ethnically (and linguistically) fractionalized by standard measures but vary substantially in their degree of sub-ethnic fractionalization, due to cultural differences in family structure. 849 students from different ethnic origins in Canada and Iran participated in a bribery game, in which the first mover chooses whether to offer a bribe and the second mover chooses to accept or reject it. If he accepts the bribe, the second mover also decides whether to make a corrupt effort to benefit the first mover, thereby imposing a negative externality on a passive third player. Subjects play the three-player bribery game with one unrelated person and one co-ethnic (or sibling, in the Kin treatment). Three possible assignments of roles to two co-ethnics (kin) create three treatments through which we explore the effect of co-ethnicity (kinship) on corrupt acts. Our design allows us to compare the frequency of bribery and corruption in treatments with in-group members (kin or co-ethnics) as first and second movers to treatments with one in-group 7 E.g. see Abbink et al. (2002); Cameron et al. (2009); Alatas et al. (2009); Barr and Serra (2010); Rivas (2013). 5 member as the first or second mover and the other as the passive third party. We can also test for differences in the strength of norms favoring each kind of in-group within-country and for differences across countries that reflect differences in norms (plausibly related to differences in sub-ethnic fractionalization). We find evidence of favoritism in both countries, but among coethnics the pattern is more pronounced in Iran. Robustness checks include treatments in which the first and second-movers are close friends and treatments varying the incentives; increased favoritism among friends in Iran compared to Canada provides further support for cross-country normative differences in the degree of in-group favoritism. Overall, our findings suggests that marriage practice-driven sub-ethnic fractionalization is an alternative but important channel through which history can explain variation in present day institutional quality. Two further considerations suggest that using variation in consanguinity to study the effects of sub-ethnic fractionalization is a reasonable approach. First, consanguinity has a direct impact on local relatedness (and thus, the relative returns to norms of local favoritism) since the offspring of a consanguineous marriage will be more closely related to their kin than the offspring of a randomly mating pair. Second, although some of the variation in consanguineous marriage rates has been traced to variation in historical modes of subsistence, inheritance rules, geographical constraints, parasite threats, and etc., (see e.g. Goody, 1983; Cavalli-Sforza et al., 2004; Hoben et al., 2010; Walker and Bailey, 2014), two other factors loom large as deep-rooted causes that explain a large portion of the observed variation: Christianization and the Arab conquests in the early centuries of Islam. The Catholic Church has restricted cousin marriage since 500AD, at times extending a ban as far as sixth cousins. However, the Church prohibited not only consanguineous marriages to blood relatives, but also to affinal kin, (e.g. a dead brother’s widow), to spiritual kin (e.g. godchildren) and to fictional kin (e.g. adoptees), “producing a vast range of people, often resident in the same locality, that were forbidden to marry” (Goody, 1983, p. 56). Consanguineous marriages “had historically provided one means of creating and maintaining kinship groups–such as clans, lineages, and tribes” (Greif, 2006, p. 309). The Church’s restrictions made in-marriage virtually impossible, fundamentally altering social organization by slowly dissolving clan, lineage, and tribe boundaries in Europe, and thereby created pressures favoring the development of impartial norms, large-scale cooperation, and eventually modern institutions (see Korotayev (2003) and Greif (2006) for further discussion and references). Going the other direction, a preference for (a particular kind of) cousin marriage seems to have spread with the Arab conquests at the dawn of Islam, despite cousin marriage being neither explicitly encouraged nor prohibited by the religion (Korotayev, 2000). Indeed, Italy is a uniquely interesting case in that by the 20th century (when we observe consanguinity rates) it was a nearly 100% Catholic country, but some southern parts of Italy, where cousin marriage is 6 most common in our data, were also part of the Caliphate for more than a century from roughly 850AD to 1000AD. Therefore, consanguinity rates today also partly reflect historical military and cultural conquests which carried with them different attitudes towards family and marriage patterns, and in our view, the impact of this variation in marriage practices on patterns of relatedness has helped shape the extent of corruption and the quality of institutions today. While we cannot allay all endogeneity concerns with our regression analyses, we also report instrumental variables estimates to provide some evidence that the relationship we identify is likely to be causal. In our cross country regressions, we use linguistic variation as an IV which captures whether the language spoken in an area distinguishes various types cousins from each other. It has been suggested that differentiated cousin-terms are associated with the prevalence of consanguineous marriage, since kin terms may serve a “classificatory” function, dividing (real and fictive) kin into those who are and are not eligible for marriage (see Garth, 1944; Fox, 1967, for an example and an overview). Schulz (2016) also uses this empirical approach, and we adopt his method here. In our within Italy regression analysis, we construct an instrument that captures province-level variation in exposure to the Catholic Church’s consanguinity bans: the number of years with an active Catholic archdiocese in each province. This variation has two sources: variation in the initial expansion of the Catholic Church’s dominion and suppression of archdioceses in some provinces in Southern Italy during centuries of Arab domination of the region. Both sets of IV estimates provide evidence consistent with our reduced form analysis, suggesting that in-marriage practices are an important potential cause of variation in corruption. Our argument is related to the literature that distinguishes between generalized morality and limited morality (or amoral familism) (Banfield, 1958; Platteau, 2000; Tabellini et al., 2008). Limited morality is the extreme reliance on a narrow circle of family, friends or relatives; outside this circle, harming and cheating are allowed and frequent. In this narrow circle, people are raised to trust in-group members only. They are also taught to distrust people outside the circle, which hampers cooperation and exchange with strangers and outsiders, and as a result, impedes the development of formal institutions. Generalized morality is characterized by respect for abstract individuals and their rights, generalized trust and loyalty to general rules, which facilitates large-scale cooperation. The underlying mechanism that determines whether a society adopts limited morality has been attributed to strong versus weak family ties (Ermisch and Gambetta, 2010; Alesina and Giuliano, 2011, 2014), collectivism versus individualism (Yamagishi and Yamagishi, 1994; Yamagishi et al., 1998), and clan versus corporation (Greif and Tabellini, 2015; Greif, 2006). We also contribute to this literature highlighting another mechanism: in-marriage practices raise the relative returns to limited morality and therefore encourage corruption. Finally, our work is also related to a growing literature on the political and economic consequences of historical variation in social and familial organization (e.g. Woodley and Bell, 2013; 7 Schulz, 2016; Moscona et al., 2017a,b; Enke, 2017). Moscona et al. (2017a,b) provide evidence that other aspects of kinship and social organization help account for levels of inter-group conflict in Sub-Saharan Africa; in particular, they show that groups organized around segmentary lineages are prone to localized conflicts. The most closely related papers by Woodley and Bell (2013) and Schulz (2016) document a negative relationship between rates of consanguineous marriage and levels of democracy, with Schulz (2016) providing IV estimates suggesting the relationship is causal. Also related is Enke (2017), who develops a measure of historical “kinship tightness” along three dimensions: family structure (domestic organization, post-wedding residence), marriage patterns (cousin marriage, polygamy), and descent systems (lineages, segmented communities and localized clans). He reports a negative correlation between his measure of kinship tightness and cooperation with and trust of outsiders using a variety of datasets. Our paper contributes to this literature by highlighting an important channel by which marriage practices and kinship structures directly alter the incentives to develop norms of impartiality, which underlie both the degree of corruption (our focus) and the extent of democracy. 2 Theory and hypotheses Corruption can be defined as “abuse of public office for private gain” (World Bank, 1997, p. 8). Public office can be abused in hiring for governmental positions, manipulating government procurement or facilitating/limiting access to basic goods or services in places like hospitals, schools, police departments, etc. Private gain is often realized through bribery, with gifts, money, or similar benefits offered in exchange for official actions. However, enforceable contracts for such exchanges are impossible because corruption is typically illegal. Therefore, bribery necessitates implicit contracts which rely on trust and cooperation. 2.1 A basic model of bribery Seen from a game-theoretic perspective, bribery is a social dilemma much like a trust game (see e.g. Berg et al., 1995; Fehr and Fischbacher, 2003) where (i) a sequential exchange takes place in the absence of enforceable contracts, (ii) both players are better off exchanging their goods or favors, and (iii) there is also a strong temptation to cheat, e.g. by accepting the bribe and failing to reciprocate. However, as noted by Abbink et al. (2002), the trust game lacks two essential components of bribery: the possibility of negative externalities and the risk of penalty. Figure 1a shows our bribery game inspired by Abbink et al. (2002). Player 1 represents a private agent and player 2 represents a public official. Player 1 may offer a bribe (t) to player 2 in the hope that player 2 will misuse his office to benefit her (B). If player 1 offers a bribe, she also 8 incurs a small cost (c) of initiating the relationship with the official. The private agent’s benefit from the official’s corrupt effort, B is high enough that B > t + c. The official, player 2, has the option of accepting t but making no effort or making a corrupt effort and incurring a cost (e). The effort cost is low enough that e < t. If the official chooses to make the corrupt effort, there is a small probability (ǫ) of getting caught, where both private agent and official end up with zero payoffs. If the official is not caught, the negative externality of the official’s corrupt effort on citizens, who have no move in the game, is Xi , which is displayed below the payoff vectors whenever it occurs. Assuming ∑ Xi ≥ B, the game also captures another characteristic of corruption; it is inefficient. Private Agent No offer Private Agent No offer Offer t Offer t Official  P1 P2  Reject Official  Accept P1 P2  Reject Accept Official  P1 − c P2  No effort Official  Effort P1 − c P2 − r po c  No effort Effort Nature  P1 − c − t P2 + t  1−ǫ  P1 − c − t + B  P2 + t − e  -X ... -X  Nature  ǫ   0 0 (a) All players unrelated. P1 − c − (1 − r po )t P2 + (1 − r po )t − r po c  1−ǫ  P1 − c − (1 − r po )t − r po e + B  P2 + (1 − r po )t − r po c + r po B − e -X ... -X  ǫ   0 0 (b) Private Agent and Official are relatives. Figure 1: A bribery game between strangers and relatives, highlighting the implications of inclusive fitness. In the unique subgame perfect equilibrium of the one-shot game, the official accepts the offer but makes no effort, and the private agent chooses not to offer a bribe to the official. However, from field observations and experimental studies, we know that “corruption exists, bribes are paid, and favors are reciprocated” (Lambsdorff, 2012, p. 280). One possible source of observed corruption is nepotism, and its antecedent biological or kin altruism.8 8 Of course we do not claim it is the only source - repeat interaction, reciprocity and threats may also facilitate corruption, even among non-kin. The purpose of the one-shot game described here is merely to provide a simple framework in which to highlight the role of kinship as another important potential causal factor. 9 2.2 Inclusive fitness, kin altruism, and corruption Sequential social dilemmas such as the bribery game allow agents to engage in altruistic behavior: one party may incur a cost in order to provide a larger benefit to another (in this case at a cost to third parties). In general, costly, altruistic behaviors like self-sacrifice, non-reciprocal help and subordination of private interests for the good of the group are all commonly observed among kin. Costly altruism seems to contradict both models of individual self-interest and Darwinian natural selection, because behaving altruistically is disadvantageous for the altruist, by definition. Intuitively, individuals that incur costs in order to provide fitness benefits to others will have lower fitness than free-riders, and hence, prima facie, should have their numbers dwindle. However, Hamilton (1964a)’s kin selection theory provides a simple and empirically successful argument explaining how such altruistic behavior could evolve under natural selection. Hamilton solved the problem by focusing on selection at the level of the gene rather than the individual. We can imagine “a gene which causes its bearer to behave altruistically towards other organisms, e.g. by sharing food with them” (Okasha, 2013). We expect the altruist gene to be eliminated because it is disadvantageous for the altruist. But what if altruists share food only with those with whom they also share genes? Since there is a certain probability that the recipients of the food will also carry copies of that gene, the altruistic gene can in principle spread by natural selection. Thus altruistic behavior may increase the number of copies of the altruistic gene in the next generation, and thus “the incidence of the altruistic behavior itself” (Ibid.). Hamilton demonstrated that an altruistic gene will be favored by natural selection and will spread in the population when a certain condition, known as Hamilton’s rule, is satisfied. According to Hamilton’s rule, a donor provides an altruistic act if rB > C, where C is the cost of the altruistic act to the donor, B is the benefit of the act to the recipient, and r is the coefficient of relatedness between the donor and the recipient. This rule is based upon expected costs and benefits in terms of inclusive fitness which represents one’s own fitness9 plus the weighted sum of relatives’ fitness, where the weights are the coefficients of relatedness. Then from a gene’s eye view an individual benefits not only through personal reproduction, but also by helping the reproduction of others who share some of their genes (Okasha, 2013; Cox and Fafchamps, 2007). Therefore, if we assume that all people have some propensity toward kin altruism, all else equal, more closely related individuals have stronger incentives to behave altruistically towards one another. There is a lot of supporting evidence for this claim in other contexts: kinship patterns correlate with within-household violence, allocation of food, provision of childcare, and safeguards against infanticide, as well as migrant workers’ remittances to their families, willingness to murder political rivals and form stable alliances, taking sides in disputes, emotional and 9 Fitness should be thought of as reproductive success, e.g. as the expected number of progeny. 10 material support within social networks, cooperation under catastrophic circumstances, membership in cooperative labor units, organ donation rates, etc. (see e.g. Cox and Fafchamps, 2007; Madsen et al., 2007; Bowles and Posel, 2005; Fellner and Marshall, 1981). The relatedness of two individuals can be approximated by the coefficient of relatedness, that is, the expected fraction of identical by descent genes that are shared between two individuals in a randomly mating population. The value of the relatedness coefficient for identical twins is 1, for full siblings and fraternal twins 1/2, for parents and offspring 1/2, for grandparents and grandchildren 1/4, for first cousins 1/8, and so-on to a randomly chosen pair who have a relatedness coefficient of 0 (Okasha, 2013).10 “J. B. S. Haldane once remarked, it would make sense to dive into a river to save two drowning siblings or eight drowning cousins” (Siegfried, 2006, p. 85). See Appendix A for further detail. Bribery game with inclusive fitness Suppose the payoffs in the bribery game are in units of biological reproductive fitness.11 According to inclusive fitness theory, if players in the bribery game are related, their payoffs should include not only the fitness effects on themselves but also on the other parties involved. In particular the benefits and costs to others enter into in the players’ payoffs weighted by the coefficient r, of relatedness between them. Let r po represent the relatedness of the private agent and the official. Also, let r pc = 0 be the sum of relatedness of the private agent to citizens, and let roc = 0 be the sum of relatedness of the official to citizens. Then, the payoffs to the bribery game are modified as shown in Figure 1b. In the bribery game with genetically related players, the subgame perfect equilibrium can be characterized as follows, by backward induction: (I) If accepting the offer, the official honors the trust of the private agent and makes a corrupt P +(1−r )t−r c and effort on her behalf with a unique equilibrium strategy if: (1 − ǫ) > P +(12−r )t−por c+po r B−e 2 po po po he accepts the offer if: (1 − ǫ) > P +(1−r )tP−2r c+r B−e . po po po 2 (II) Assume that both aforementioned conditions hold so that the official accepts the offer and exerts the corrupt effort as his unique equilibrium strategy. The private agent foresees the 10 In reality, even in a random mating population, a parent might share more than half of her genes with her offspring; “half those genes are surely identical because they came from the parent, while gene sharing with the other half of the child’s genome is just what is shared with any random member of the population.” Hence, a more precise way to think of relatedness is to “think of gene sharing in excess of random gene sharing” (Harpending, 2002, p. 142), and the coefficient of relatedness is more properly defined as r = ( Q − Q)/(1 − Q), where Q is the relatedness of the two individuals, while Q is the average relatedness in the population (Nowak et al., 2010, p. 1059). 11 Of course, the analogy is imprecise in the sense that corruption is not transacted in units of fitness. However, in many cases, corruption influences the allocation of large quantities of resources (monetary and otherwise), which are correlated with reproductive success. In an extreme case, if a corrupt act results in one individual living to reproduce and another dying before reproduction, the effects are direct in fitness terms. 11 optimal strategy of the official; therefore she places trust and offers t in a unique equilibrium strategy if: (1 − ǫ) > P −c−(1−rP1 )t−r e+ B . 1 po po Implication 1: All else equal, the official is more likely to accept a bribe and make a corrupt effort as r po increases. Implication 2: All else equal, the private agent is more likely to offer a bribe as r po increases. Note that while our example sets roc and r pc equal to 0, if we allow roc to vary, the official is less likely to accept a bribe and make a corrupt effort as roc increases, and similarly, if we allow r pc to vary, the private agent is less likely to offer a bribe as r pc increases. The analysis so far highlights the role of relatedness in shaping the incentives for corruption. In-group favoritism among co-ethnics was Mauro’s (1995) motivating example for using ethnic fractionalization as an instrumental variable for corruption: “bureaucrats may favor members of their same group”(Ibid., p.693). However, as we noted in the introduction, empirical estimates suggest that relatedness among co-ethnics (relative to neighboring groups) is often not far above zero (Cavalli-Sforza et al., 1994), which helps explain the weak association between ethnic fractionalization per se and corruption. Our focus is instead on sub-ethnic fractionalization, driven by marriage patterns (and associated variation in patterns of social interaction), which causes increased relatedness and thereby alters the returns to norms of favoritism and corruption. 2.3 Consanguineous marriage, sub-ethnic fractionalization, and corruption Underlying the coefficients of relatedness for kin reported above is a crucial assumption: random mating, “that mates are chosen with complete ignorance of their genotype (at the locus under consideration), degree of relationship, or geographic locality” (Gillespie, 2004, p. 13). The key to our argument about the role of sub-ethnic fractionalization in corruption is that in many cases, the assumption of a randomly mating population is violated. When mating is non-random, so that members of some local groups are more likely to mate within the group than outside the group (e.g. consanguinity due to geography, culture, etc.), the expected relatedness of kin and group members is higher than under random mating.12 For example, two offspring of a first-cousin marriage have a relatedness higher than 1/2 (r = 1/2 + 1/2 × 1/8): with probability 1/2 they inherit a gene from the same parent at each locus, and with probability 1/2 they each inherent a gene from a different parent, in which case the probability of gene sharing is just the relatedness between their parents, 1/8. If this pattern repeats over generations, local relatedness only grows (see Hamilton, 1975, for a theoretical 12 Related literature suggests that assortative matching (and mating) can encourage the evolution of (local) cooperation and favoritism. See e.g. Bergstrom (2003); Grimm and Mengel (2009) for theoretical and experimental evidence on cooperation, and also Hammond and Axelrod (2006); Efferson et al. (2008); Fu et al. (2012) on favoritism. 12 analysis and Appendix A for some empirical evidence). In a cross-cultural ethnographic tabulation due to Murdock (1967) and Gray (1998), a total of 476 out of 1024 societies for which we have data either permitted or favored first and/or secondcousin marriage, and estimates suggest that roughly 10% of marriages around the world today are consanguineous (Bittles, 2012).13,14 The wide diversity in attitudes towards consanguinity in human societies partially originates in religious beliefs due to the common jurisdiction of religious institutions over marriage. Table 3 shows some of the diversity of religious attitudes toward cousin marriage around the world, and Appendix B summarizes the history of Christian and Islamic attitudes toward consanguinity. Of particular note are the fact that the Catholic Church has placed restrictions on cousin marriage since at least 500AD (sometimes extending these bans out to sixth cousins, as well as to fictive and affinal kin) and the fact that a persistent preference for cousin marriage can be seen in many Islamic countries, reflected in contemporary rates as high as 50% of marriages. Attitude toward Cousin Marriage Religion Sect Judaism Ashkenzi Sephardi Coptic Orthodox Eastern Orthodox Protestant Roman Catholic Sunni Shia Indo-European Dravidian Christianity Islam Hinduism Buddhism Sikhism Confucian/Taoist Zoroastrian/Parsi Permitted Permitted Permitted Proscribed Permitted Diocesan approval req. Permitted Permitted Proscribed Permitted Permitted Proscribed Partially permitted Permitted Table 3: Religious attitudes to consanguineous marriage (Bittles, 2012, p. 14). In fact, marriage norms may persist, even as religious attitudes change. As one example, consanguineous marriage has long been prevalent in parts of Italy, despite it being an almost entirely Catholic country. Cavalli-Sforza et al. (2004) suggest this may be a result of persistent cultural norms imported during the Arab conquest of southern Italy over 1000 years ago. Going the other direction, majority-Protestant countries mostly legalized cousin marriage after centuries of living under the Catholic ban. Nevertheless, consanguinity remains rare in those countries. In the United States, cousin marriage is illegal in 25 states, though its frequency remains low 13 While the negative health effects of consanguinity are well-known (e.g. increased risk of autosomal-recessive disorders), some believe that there are countervailing positive effects as well (Jaber et al., 1998). 14 In a small sample, genomic estimates of the “inbreeding coefficient”, which measures “the proportion of a genome that is ‘autozygous’ - homozygous for alleles inherited identically by descent from a common ancestor,” are correlated as expected with consanguineous marriage rates (r = 0.349, p-value = 0.04, N = 26). The correlation would likely be higher except that the genomic measures also capture background (so-called “cryptic”) inbreeding due to geographical population division (Pemberton and Rosenberg, 2014, p. 38). See Appendix A for more information. 13 even where it is legal. Of course, not all consanguineous marriage is driven by religious preference. For instance, consanguineous mating can be caused by population division due to geography. As populations migrated around the world historically, they became isolated from one another due to vast distances and geographic barriers such as mountains, deserts and oceans that were only recently broken down by transportation technologies. Due to isolation, some groups accumulated relatively high local relatedness. Others have argued that consanguinity is a cultural adaptation to social and ecological circumstances. In his history of the family, Goody (1983) suggests that consanguinity may be a property and wealth-preserving response to gender-egalitarian inheritance rules, which would result in the diffusion of property under out-marriage. A few studies have examined the causes and consequences of consanguinity in small-scale societies using detailed genealogies to directly measure relatedness. Walker and Bailey (2014) show that among forager peoples, such marriages are rare due to norms of exogamy and fission/fusion dynamics that disperse kin across groups, but among agropastoralists, particularly those that practice polygyny, the practice is more common, with average spousal relatedness rising as high as r = 0.18 (almost 50% greater than first cousins, r = 0.125). Using log(surviving children) as a measure of fitness, estimates in Bailey et al. (2014) suggest that these marriage practices may be adaptive, with fitness maximized for moderate consanguinity among agropastoralists and with minimal consanguinity among foragers. Other evidence from Hoben et al. (2010) suggests that consanguinity may be more prevalent near the equator since it can raise the frequency of homozygosity for adaptive recessive mutations that defend against diseases and parasites, which are also more prevalent in warmer climes. Regardless of its origins, wherever it is practiced, consanguineous marriage directly increases local relatedness and encourages sub-ethnic fractionalization, thereby altering the returns to norms of favoritism and norms of impartial cooperation. Thus, variation in consanguinity rates facilitates a test of our main hypothesis: that sub-ethnic fractionalization causes corruption. 3 Empirical Strategy and Results To provide evidence for a relationship between sub-ethnic fractionalization and corruption, we present data from cross-country regressions, within-country regressions and laboratory experiments. Each of the methods has limitations, and none of them can wholly address identification or endogeneity concerns, but our goal is that by approaching the problem at various levels of granularity we can provide a set of robust, complementary tests of the main hypothesis: that sub-ethnic fractionalization causes corruption. In our cross-country and within-country analyses, we employ data on consanguineous mar14 riage as our measure of sub-ethnic fractionalization and we regress measures of corruption on consanguinity. As discussed below and detailed in Appendix C, Bittles and Black (2015) have collected data on rates of consanguineous marriage (among 2nd cousins or closer relatives) at the country level from a multitude of sources, including surveys, public health studies, and church records. While our cross-country consanguinity sample is neither random nor representative, it covers the large majority of the global population. For our within-country analysis, the data on consanguinity has more complete coverage, as Cavalli-Sforza et al. (2004) have collected data on consanguinity in Italian provinces. Finally, we report the results of a laboratory corruption experiment in Iran and Canada where our treatments directly manipulate relatedness, bringing strangers, co-ethnics and kin (siblings) into the lab and varying their assignment to roles in the game. These two countries are similarly ethnically fractionalized, but Iran has much higher rates of consanguinity. The design allows us to test for the effects of relatedness by comparing across treatments within each country and for the effects of marriage patterns on social norms by comparing across countries. Robustness checks in which the corruption game is played with friends and/or the social and private cost of corruption is varied provide further support for an interpretation of cross-country differences in social norms rooted, at least in part, in distinct marriage patterns. 3.1 Cross-country analysis Figure 2 displays average ICRG institutional quality data from 1984-2011 alongside consanguinity data from Bittles and Black (2015). ICRG indices are widely used in corruption studies since they capture many kinds of corruption and have wide coverage. Grey colored areas indicate missing data. Although we have consanguinity data for 72 countries, in our primary analyses, there are 67 countries for which we have the full set of covariates used in our main regression analyses. In this sample, we find a negative and highly significant correlation between institutional quality and consanguinity (Spearman’s ρ = −0.56, p-value < 0.001, N = 67). While the correlational evidence is strong, we also conducted a series of OLS regressions controlling for relevant confounds and alternative explanations.15 Our empirical strategy follows La Porta et al. (1999) and Alesina et al. (2003) who attempted to address the endogeneity of corruption by focusing on “(reasonably) exogenous sources of variation” (La Porta et al., 1999, p. 223) in the economic, geographical, political and cultural characteristics of countries. Hence, we do not include contemporary political variables or variables capturing public policy. Instead, our analysis focuses on “more fundamental, or at least historically predetermined” variables (LaPorta et al., 1999, p. 230; see also Treisman 2000, p. 409). 15 Results are robust to using Tobit to account for the dependent variable being restricted to the interval [0,6]. 15 (a) Corruption (b) Consanguinity 1 65.8 6 0 Figure 2: Corruption and consanguinity around the world. The most obvious economic heterogeneity across countries that can affect corruption is economic development, but development is almost certainly endogenous to corruption. While there is evidence that poor countries are perceived to be more corrupt than rich ones, so that per capita income is a potential determinant of corruption, corruption itself can reduce per capita income (see e.g. Mauro, 1998; Tanzi and Davoodi, 1998; Campos et al., 1999; Lambsdorff, 2003). As noted by Treisman (2000), one exogenous variable that is correlated with economic development but is unaffected by corruption is a country’s latitudinal distance from the equator (indeed, log GNI per capita averaged over 1984-2011 correlates with latitude, Spearman’s ρ = 0.59, p-value < 0.001, N = 67).16 In addition to latitude, we also include regional dummies from Alesina et al. (2003) for Sub-Saharan Africa, East Asia and Pacific, Latin America and Caribbean and country size (population). Following the literature, we also include ethnic fractionalization and legal origins as sources of exogenous variation in country-level political characteristics, and we report heteroskedasticity robust standard errors. After reporting a basic model using the variables described above, our main analysis relies on cross-country variation in cultural traits. First we consider the effects of a cultural preference for (and prohibitions of) consanguineous mating practices by adding consanguinity rates to the basic model, and then, in a series of regressions we allow consanguinity to compete with alternative cultural traits believed to influence corruption in the literature. Finally, we conclude with an instrumental variables approach intended to provide evidence that the connection between marriage practice-driven sub-ethnic fractionalization and corruption is causal. 16 One possible indirect connection is through an effect of latitudinal distance from the equator on consanguinity rates. As noted above, Hoben et al. (2010) argue that relative parasite prevalence near the equator may raise the returns to consanguinity by raising the frequency of homozygosity for adaptive recessive mutations (e.g. parasite immunities). We find a high correlation between distance from the equator and consanguinity which provides further reason to control for latitude in our regressions (Spearman’s ρ = -0.45, p-value < 0.001, N = 67). 16 Main findings Table 4 displays our first set of regressions, with ICRG institutional quality as the dependent variable. A full description of the variables is presented in Appendix C, Table C1, and the omitted legal origin dummy in the regressions is the British one. Column (1) presents our basic regression model, inspired by La Porta et al. (1999) and Alesina et al. (2003), which includes a set of historically predetermined and exogenous economic, geographical and political variables. In column (2), we run the same regression using only the sample of countries for which consanguinity data exists and find qualitatively similar results. In column (3), we include consanguinity rates to account for sub-ethnic fractionalization, and the estimated coefficient is significant at the 1% level, increases the R2 by 50%, and remains significant in alternative specifications using different measures of economic development. These estimates imply that a 1 standard deviation increase in consanguinity is associated with a reduction in quality of institutions (i.e. increase in corruption) by about 0.7 standard deviations. (1) Basic model VARIABLES (2) Basic model restricted sample (3) and Consanguinity (4) Income instead of Latitude (5) Latitude as instrument (6) both Income and Latitude (7) without Income and Latitude -0.222 (0.670) -4.463*** (0.769) 0.524 (0.514) -3.664*** (0.905) 0.0742 (0.460) -2.513* (1.331) -0.123 (0.417) -3.566*** (0.884) 0.226 (0.479) -5.087*** (0.772) 0.319 (0.555) yes 67 0.441 yes 67 0.668 yes 67 0.695 yes 67 0.654 yes 67 0.704 yes 67 0.632 Consanguinity Ethnic fractionalization -0.070 (0.470) Additional controls yes Observations 134 R-squared 0.527 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 4: Regression analysis of the relationship between consanguinity and corruption. Higher values of the dependent variable imply lower corruption. Additional controls include legal origins dummies, region dummies, latitude and log population. Full specification reported in Appendix Table E1. Latitude remains a significant determinant of corruption. This provides strong evidence that whatever the effect of corruption on growth, higher economic development is associated with lower corruption, as noted by Treisman (2000). When we include income per capita in the regression instead of latitude, in column (4), consanguinity is still highly significant. To address the endogeneity of income per capita more formally, we also used latitude as an instrument for income per capita in column (5) which yields similar results; here we cannot reject the null hypothesis of no endogeneity, suggesting the point estimates are consistent. From significance at 1%, latitude becomes insignificant when income per capita is also included in the regression in column (6). The most plausible interpretation is that latitude affects corruption only through income per capita which indicates that latitude is a reasonable proxy for income per capita in the regressions. Compared to those in common law countries, governments in countries of socialist legal 17 origin are more interventionist, and thus the observed negative effect of socialist legal origin on institutional quality in all regressions is consistent with previous findings. While Alesina et al. (2003) note the difficulty of disentangling the independent effect of ethnic fractionalization from income per capita and latitude because all are highly correlated, their ethnic fractionalization index is insignificant in all specifications, even excluding controls for income per capita and latitude as in column (7). As another robustness check, we replicate two specifications from Alesina et al. (2003), restricting the sample to countries for which we also have consanguinity data (see Appendix D), and our results continue to hold. As the authors along with Treisman (2000) note, ethnic fractionalization has a reduced form relationship with corruption and is not typically significant after controlling for per capita income and latitude, while consanguinity remains significant in columns (3)-(7), with or without one or both of income per capita and latitude. Alternative interpretations and confounds Given concerns about the endogeneity of income to corruption, we will retain latitude as our proxy for economic development in subsequent regressions. From here on, we build upon specification (3) which controls for historical and predetermined variables. To assure the robustness of our interpretation, we compare our preferred cultural metric of consanguinity against alternative theories of culture and also consider possible confounds. Religion. Religion is the most important historical cultural factor affecting institutions. Religion is potentially relevant to our analysis in two ways: through an indirect effect on corruption via consanguinity and through a direct effect on corruption as discussed in previous work (see La Porta et al., 1999; Alesina et al., 2003; Treisman, 2000). A preference for consanguineous marriage is common in the Islamic world, and thus consanguinity rates and the percent of the country practicing Islam are highly correlated (Spearman’s ρ = 0.73, p-value < 0.001, N=67).17 In contrast, Catholicism has imposed a long-standing ban on consanguineous marriage, and this is evinced by a strong negative correlation between the share of a country practicing Catholicism and consanguinity (Spearman’s ρ = -0.57, p-value < 0.001, N=67). Finally, while Protestant religions do not officially ban consanguineous marriage, the frequency is quite low, and we find a large negative correlation between a country’s share of Protestants and consanguinity (Spearman’s ρ = -0.53, p-value < 0.001, N=67). La Porta et al. (1999) and Treisman (2000) offered reasons why Protestantism may be a cultural deterrent to corruption, beyond its relationship to consanguinity. They argue that Protestantism 17 Note, however, that we find a significant correlation between consanguinity and corruption even if we focus only on minority-Muslim countries (Spearman’s ρ = -0.43, p-value = 0.002, N=47). 18 is associated with attitudes such as Weber (1958)’s “Protestant work ethic” and a separation of church and state, both of which may have been conducive to growth and quality government, while countries that were predominantly Catholic or Muslim during this period were relatively more insular, hierarchical and interventionist (see La Porta et al., 1999, p. 229). In column (2) of Table 5, we include variables indicating the share of each country’s population that practices Protestantism, Catholicism, and Islam; the excluded category is “other religions”. We confirm La Porta et al. (1999) and Treisman’s (2000) finding that the proportion of Protestants in a country’s population is associated with lower corruption relative to other religious groups. Moreover, Islam and Catholicism are both associated with higher corruption. Note that we also controlled for consanguinity, and though the coefficient is smaller, it remains highly significant. The significant coefficients on religion in our regressions confirm previous findings and suggest an additional effect of religion that is independent of its influence on consanguinity traditions. Lipset and Lenz (1999) argue that “Protestantism reduces corruption, in part, because of its association with individualistic, non-familistic relations” (Treisman, 2000, p. 428). Family ties is the next cultural trait that we discuss. Family ties. As Cavalli-Sforza et al. (2004) note, consanguinity “may be especially attractive where family values are especially important, the size of extended families is large, and social contacts are much more frequent with close relatives” (p. 287). This suggests that sub-ethnic fractionalization (measured as consanguinity) may simply reflect the relative importance of family ties across countries, since “strong and stable social relations (such as family ties and group ties) promote a sense of security within such relations but endanger trust that extends beyond these relations” (Yamagishi et al., 1998, p. 166-8). Several studies confirm the negative correlation of strong family ties with general trust (Yamagishi and Yamagishi, 1994; Yamagishi et al., 1998; Fukuyama, 1995; Ermisch and Gambetta, 2010; Alesina and Giuliano, 2011). Moreover, there is evidence that the strength of family ties contributes to the explanation of heterogeneity in corruption (among other macroeconomic variables, see e.g. Alesina and Giuliano, 2010, 2014), and we find a large correlation between their measure of family ties and consanguinity (Spearman’s ρ = 0.58, p-value < 0.001, N=45). Nevertheless, in column (3) which includes data on family ties, as with religion, consanguinity remains a highly significant predictor of corruption, despite losing nearly 1/3 of our observations to missing data on family ties; moreover, the coefficient on family ties is also significant, suggesting that the variables’ effects on corruption are independent and do not necessarily capture the same underlying factors. Trust. In contrast to the partiality that may be engendered by family ties, generalized trust is often considered to be a foundation of impartial institutions, and consistent with Yamagishi 19 VARIABLES Consanguinity Ethnic fractionalization (1) Basic model (2) and Religion (3) and Family ties (4) and Trust (5) and Genetic diversity (6) and Geography -4.463*** (0.769) 0.524 (0.514) -2.411** (1.176) 0.451 (0.458) 1.553* (0.791) -0.904** (0.370) -1.492*** (0.459) -3.853*** (0.722) 0.969 (0.667) -4.663*** (0.842) 0.846* (0.486) -3.494*** (0.952) 0.274 (0.516) -3.874*** (0.999) -0.140 (0.447) Protestant Catholic Muslim Family ties -1.261** (0.528) General trust 1.562 (1.105) Genetic diversity 48.933 (75.326) -40.988 (56.274) Genetic diversity squared Geographical variables Additional controls yes Observations 67 R-squared 0.668 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 yes 67 0.762 yes 45 0.783 yes 56 0.725 yes 67 0.689 yes yes 65 0.770 Table 5: Regression analysis of the relationship between consanguinity and corruption: potential confounds. N varies due to missing data for some countries. See Appendix E.1 for analogous estimates of column (1) for each subsample. Additional controls include legal origins dummies, region dummies, latitude and log population.Full specification reported in Appendix Table E2. et al. (1998) and others, trust and family ties are negatively related in the sample for which we also have consanguinity data (Spearman’s ρ = -0.52, p-value < 0.001, N=45). As measured in the World and European Values Surveys, trust has been shown to correlate with institutional quality and economic development (Uslaner, 2005; Francois and Zabojnik, 2005; Tabellini, 2010). One possible concern is that consanguinity-driven sub-ethnic fractionalization is an endogenous response to lack of trust, creating binding ties to encourage and enforce cooperation in the absence of other means. Perhaps surprisingly, we find no significant relationship between generalized trust and consanguinity (Spearman’s ρ = -0.19, p-value = 0.16, N=56).18 Nevertheless, when we control for trust in the regression analysis in column (4), it is insignificant; while consanguinity remains a highly significant predictor of corruption. Genetic diversity. Empirically, the genetic diversity of indigenous populations around the world (measured as mean expected heterozygosity) decreases with geographic (great circle) dis18 Note that this insignificant relationship holds if we also restrict the sample to countries for which we have family ties data (p-value = 0.27). This may reflect an ambiguity in the question as it is asked in the WVS and EVS; in particular, while previous analyses have typically interpreted the question as referring to trust of strangers (see e.g. Johnson and Mislin, 2012, for a discussion), the question as asked does not make this distinction, leaving it up to the respondent to determine the reference group. In societies with high levels of sub-ethnic fractionalization, “most people” [with whom you interact] may refer to a different reference set than in societies with low levels of sub-ethnic fractionalization. 20 tance from Ethiopia (Ramachandran et al., 2005; Pemberton et al., 2013). This correlation has been argued to reflect the prehistoric “out of Africa” exodus of Homo sapiens to settlements around the world. These migrations, which happened over thousands of years, resulted in a “serial founder effect”, in which small founding populations at each new settlement carried with them only a portion of the genetic diversity of the source population.19 Ashraf and Galor (2013) show that country-level predicted genetic diversity has an inverted U-shaped relationship with economic development, which suggests that we could include genetic diversity and diversity squared as an additional control for economic development. In column (5) of Table 5, we control for predicted genetic diversity and predicted genetic diversity squared, following Ashraf and Galor (2013) (see Appendix C for details on the variables). Although consanguinity and predicted genetic diversity are highly correlated (Spearman’s ρ = 0.50, p-value < 0.001, N=67), the coefficient on consanguinity remains sizable and highly significant. In an unreported regression, we instead include migratory distance from East Africa as an instrument for genetic diversity, and the results are qualitatively unchanged. Geographical factors. Cavalli-Sforza et al. (2004) report that “following the national trend” most consanguineous marriages were celebrated in the mountains “with a clear decreasing trend moving down” to the hills, in the plain and finally in the city (p. 37-38). This suggests that cross-country differences in consanguinity rates may be a consequence of heterogeneity in geographical barriers to migration and exogamy. As a proxy for geographical barriers, we use the terrain ruggedness index from Nunn and Puga (2012). As explained in the data description in Appendix C, terrain ruggedness captures the average elevation differences of adjacent lands in each country. Perhaps surprisingly, ruggedness is not significantly correlated with consanguinity (Spearman’s ρ = -0.14, p-value = 0.270, N = 67). In column (6) we control for ruggedness and also include a variety of other geographic controls that are plausibly related to economic development, taken from Ashraf and Galor (2013): soil suitability for agriculture, mean elevation, mean temperature, mean precipitation, percentage of the population living in tropical and subtropical zones, percentage of population living in temperate zones, and percentage of land near a waterway. Consanguinity again remains a highly significant predictor of corruption. Additional robustness checks. To assuage concerns about the changing number of observations in Table 5 as we include additional variables for which we have limited data, in Table E5 in Appendix E.1, we compare the basic specification including only consanguinity from column (1) and the specification including each cultural variable, restricting to the sample for which we have data on both measures. Finally, since the data on consanguinity were collated from 19 Evidence of subsequent admixture complicates this view; see e.g. Lazaridis et al. (2016). 21 448 studies over many decades, we report additional analysis in Table E6 in the appendix that addresses data collection dates. Our results remain robust in both tables. Instrumental variable: cousin terms While the evidence is compelling, such cross-country analyses can never allay all endogeneity concerns. Indeed, ethnic conflict resulting from ethnic fractionalization, greater intensity of local group interaction due to family ties, and in-group loyalty all might encourage consanguinitydriven sub-ethnic fractionalization (or vice versa) for various reasons. Moreover, corrupt, lowquality governance may cause all of these variables indirectly. In many countries with weak institutions “social safety nets are incomplete or nonexistent and households must cope with an unforgiving environment of severe poverty and shocks to economic and physical well-being [. . . ] especially against a backdrop of inadequate formal credit and insurance markets and a minimal welfare state” (Cox and Fafchamps, 2007, p. 3714). In such an environment, kin-based, tribal and ethnic networks may play an important role in helping households to manage shortages and uncertainty by supporting informal exchanges including gifts, feasts, rotating saving, informal loans, intermarriage and arranged marriages. These networks rely on in-group trust and reciprocity to enforce informal contracts and provide families with risk-sharing, insurance, and information - all functions which, in developed countries, are typically carried out by markets. Kinship, tribal and ethnic networks also help to organize the provision of public goods, a role that, in developed countries, is usually performed by government. But unlike governments, these networks do not have the power to tax or mobilize resources. Therefore, the provision of public goods relies on local, informal exchange of favors (see Cox and Fafchamps, 2007; Alesina and La Ferrara, 2005; Johnson and Earle, 2000). Thus, there is reason to believe that weaker institutions may also increase sub-ethnic fractionalization, raising the relative returns to kin-based, clan and tribal organization and concomitant norms of local favoritism. To attempt to address these concerns and provide evidence for a causal relationship between sub-ethnic fractionalization and corruption, we seek an instrument that is correlated with consanguineous marriage practices, but unlikely to be correlated with corruption. A valuable recent approach uses historical ethnographic data from the Ethnographic Atlas to aggregate information about the historical cultural characteristics of resident ethnic groups into country-level measures. We follow the method introduced by Alesina et al. (2013) to produce such measures. As an estimate of the geographic distribution of ethnicities across the globe, we use Language locations from WLMS (World Language Mapping System) 19th version (2017) which corresponds to the 19th edition (2016) ISO 639-3 standard, and the 16th Edition of the Ethnologue (Languages of the World). WLMS provides a shape file that divides the world into polygons indicating the locations around the world where 7650 languages are spoken. Each of these lan22 guages are matched to one of the 1291 ethnic groups included in the extended Ethnographic Atlas provided by D-PLACE.20 We also use the LandScan 2014 database which reports the world’s population in 2014 for 30 arc-second by 30 arc-second grid cells globally to construct the country-level population weighted averages of measures from the Ethnographic Atlas. See Alesina et al. (2013) for more details of the methodology used in the construction of the country-level data. In Appendix E.2, we replicate their measure of historical plow use, using the new data; see Figure E1. The key for our study is to find a measure that is related to consanguineous marriage (and hence sub-ethnic fractionalization) but unlikely to be directly related to corruption. While the Ethnographic Atlas contains some measures of consanguineous (cousin) marriage practices, path dependence in the development of norms and institutions might make such measures subject to the same endogeneity concerns that arise in using contemporary marriage practices.21 We follow a clever strategy employed by Schulz (2016), that exploits a connection between kinship terminology and marriage patterns. Anthropologists distinguish six systems of kin terminology into which languages can be classified, following Morgan (1871). Eskimo (which includes English) and Hawaiian kin terminologies do not distinguish first cousins from one another.22 Sudanese, Iroquois, Omaha, and Crow terminologies either fully or partially distinguish cousins from one another. Anthropologists have long argued that differentiated cousin-terms are associated with the prevalence of consanguineous marriage, since kin terms may serve a “classificatory” function, dividing (real and fictive) kin into those who are and are not eligible for marriage (Morgan, 1871; Tylor, 1889; Garth, 1944; Fox, 1967; Goody, 1976). Variable 27 in the Ethnographic Atlas records kinship terminology used to refer to cousins and thus provides a measure that can be used as an instrument for cousin marriage rates. Following Schulz (2016), we created a dummy variable which equals 1 if the language spoken in a society fully or partially distinguishes first cousins from each other (Sudanese, Iroquois, Omaha, Crow), and equals 0 otherwise (Eskimo or Hawaiian). Using Alesina et al. (2013)’s method explained above, first we assign this measure to all ethnic groups (see the left panel of Figure 3). 20 The extended version of the Ethnographic Atlas by D-PLACE (Database of Places, Language, Culture, and Environment) (Kirby et al., 2016) has several advantages over last version (Murdock, 1962–1971; Gray, 1999) used in previous studies such as Alesina et al. (2013); 1- Original Ethnographic Atlas includes 1267 societies with three duplicate observations of which two were discovered and removed by Alesina et al. (2013). D-PLACE added data for 27 societies in Eurasia, recently coded by Korotayev et al. (2004); Bondarenko et al. (2005), which fill important regional gaps in northern Eurasia; 2- Each data point in Ethnographic Atlas is linked to one or more of the 3502 ethnographic sources that were consulted in coding the data; 3- Errors in the years in which societies were sampled were identified and replaced; 4- The latitude-longitude data of societies were corrected and improved. 21 We discuss other issues with the marriage practice variables in Appendix E.2. 22 Eskimo terminology further distinguishes siblings from their cousins. According to Morgan (1871) and Goody (1976), accumulation of property and its inheritance by individuals (i.e. direct inheritance) were crucial in the development of such “individualizing” kin terms and the “isolation of the nuclear family” (Goody, 1976) in Eskimo terminology. 23 Then, we construct a population-weighted average of the dummy variable for all ethnic groups living in a country. Therefore, our country-level measure is the fraction of the population that speaks a language which differentiates among cousins (see the right panel of Figure 3). (a) WLMS polygons. (b) Country-level means. Figure 3: Kin terms partially or fully distinguishing first cousins (Descriptive/Crow/Iroquois/Omaha/Sudanese) versus not distinguishing any first cousins (Eskimo/Hawaiian), by ethnic/linguistic group and country. Results. Tables 6 and 7 employ the country-level cousin terms measure in both reducedform and IV analysis of the specifications reported in Table 5. In all specifications in Table 6, we see a negative and significant relationship between the share of the population using kin terms that distinguish cousins and institutional quality. This relationship holds controlling for the set of confounds introduced in the last section. Similarly, when cousin terminology is used as an IV for consanguinity in Table 7, we observe a large, negative impact of consanguinity rates on institutional quality in the baseline specification. This result is robust to including our other controls, except in the case of column (2), which introduces controls for religion. This is unsurprising since the Church’s influence on family structure also led, e.g. the English language to shed excess kin terms (Mitterauer, 2010, p. 69). In all specifications we cannot reject the null hypothesis of exogeneity. We report additional robustness checks in Appendix E.2. Summary. Overall, our cross-country results are consistent with the idea that marriagepractice driven sub-ethnic fractionalization, by increasing the relative returns to norms of favoritism, is an important cause of corruption. To further test this hypothesis, we next report analysis exploiting historical variation in consanguinity within a single developed country. 3.2 Within-country analysis (Italy) To further test the hypothesis that sub-ethnic fractionalization causes corruption, while controlling for country-specific institutional and cultural factors, we collected data on corruption 24 VARIABLES Cousin term measure Ethnic fractionalization (1) Basic model (2) and Religion (3) and Family ties (4) Trust (5) and Genetic diversity (6) and Geography -0.733*** (0.181) 0.044 (0.453) -0.323* (0.190) 0.283 (0.402) 0.639 (0.520) 0.367 (0.255) -0.895*** (0.250) -0.706** (0.317) 0.295 (0.548) -1.039*** (0.247) 0.010 (0.440) -0.519*** (0.183) 0.020 (0.427) -0.551** (0.230) 0.120 (0.507) Protestant Catholic Muslim Family ties -0.580 (0.620) General trust 1.540* (0.898) Genetic diversity 23.798 (57.826) -24.128 (41.920) Genetic diversity squared Geographical variables Additional controls yes Observations 128 R-squared 0.619 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 yes 127 0.702 yes 69 0.726 yes 86 0.728 yes 127 0.652 yes yes 122 0.703 Table 6: Regression analysis of the relationship between cousin term measure and corruption (reduced form). N varies due to missing data for some countries. Additional controls include legal origins dummies, region dummies, latitude and log population. Full estimates reported in Appendix Table E3. VARIABLES Consanguinity Ethnic fractionalization (1) Basic model (2) and Religion (3) and Family ties (4) Trust (5) and Genetic diversity (6) and Geography -5.048*** (1.157) 0.586 (0.532) -0.814 (4.631) 0.193 (0.604) 1.996 (1.571) -0.715 (0.602) -1.722** (0.835) -3.498*** (1.219) 0.869 (0.570) -5.610*** (1.313) 1.057* (0.527) -4.503** (1.896) 0.406 (0.587) -5.804*** (1.860) -0.136 (0.542) Protestant Catholic Muslim Family ties -1.097* (0.586) General trust 1.513 (1.127) Genetic diversity -6.551 (90.365) 1.878 (68.444) Genetic diversity squared Geographical variables Additional controls yes Observations 64 R-squared 0.667 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 yes 64 0.792 yes 44 0.801 yes 53 0.709 yes 64 0.687 yes yes 63 0.747 Table 7: Regression analysis of the relationship between consanguinity and corruption, with cousin term measure as an instrument for consanguinity. N varies due to missing data for some countries. Additional controls include legal origins dummies, region dummies, latitude and log population. Full estimates reported in Appendix Table E4. 25 and consanguinity across Italian provinces. Our consanguinity data are from Cavalli-Sforza et al. (2004) and are based on records of Papal dispensations for consanguineous marriages kept in the Vatican archives. Because consanguinity was officially banned by the church, couples who wished to circumvent the ban had to request approval from their local diocese. Detailed records of these marriages were preserved by the church and compiled in province-level statistics. We report the consanguinity rate by province, measured as the share of all marriages in the province that were between first cousins, or closer relatives (e.g. uncle-niece marriages) for the period 1945-1964, since data from this period are available for all provinces and do not include sample variation due to either World War.23 The underlying data, however, date back to 1911, and Cavalli-Sforza et al. (2004) provide evidence that heterogeneity in consanguinity rates across Italy at the start of their time period was unrelated to a large vector of demographic, social and economic variables (e.g. birth/death rates, infant mortality, population density, immigration/emigration, literacy, and industrialization). Moreover, their evidence suggests that trends in consanguinity in Italy were similar across the entire country, despite substantial differences in levels (see Appendix E.4). Data on actual corruption crimes in Italy are not available for the provinces. Therefore, exploiting a known link between corruption and associative crime (i.e. criminal conspiracy and mafia association, see Fiorino et al., 2012), we use the latter as a proxy for corruption. The link is straightforward: associative crime reflects charges leveled at members of criminal organizations that seek to enrich themselves at the expense of others and of the government. Corruption is essential for such criminal organizations because it facilitates the operation of illegal markets for goods and services such as cigarettes, drugs, prostitution, gambling, as well as activities such as car-theft, extortion, tax evasion, etc. Corruption allows criminal organizations to obtain information about potential attempts to subvert their operations (by law enforcement or competitors), supports the deletion, falsification or destruction of incriminating evidence, and may be used to neutralize judges, prosecutors, police or experts who might interfere with their plans.24,25 23 Due to high consanguinity rates in Sicily and Sardinia, the church relaxed its rules such that dispensations were not required for marriages between relatives more distant than first cousins in those regions, though dispensations for such marriages remained mandatory elsewhere. Thus to ensure comparability of the data across provinces, we restrict attention to first cousin marriages and closer. Note that the coefficient of relatedness between first cousins (in a randomly mating population) is r = 1/8 vs. r = 1/32 for second cousins. 24 A model by Kugler et al. (2005) connects corruption and associative crimes, based on criminal organizations’ attempts to avoid punishment by bribing law enforcement and otherwise engaging in local corruption. 25 The proposed connection is also borne out in the available data. Data on corruption crimes exist at the region level (N = 20), so we report region-level correlations between consanguinity and corruption crimes in Appendix E.4, and the results are consistent with the province-level analysis. Moreover, aggregating our proxy measure of corruption, “associative crime”, to the region level, the two measures are highly correlated, even with a small number of observations, suggesting that associative crime is a reasonable proxy for corruption (Spearman’s ρ = 0.49, p-value = 0.03). See also Gounev and Bezlov (2010) who analyzed links between organized crime and corruption for The European Commission. Their analysis includes case studies on several European countries including Italy, 26 Thus, we proxy for province-level corruption by using the number of associative crimes per 100,000 inhabitants of the province (via ISTAT). Our data on associative crimes and control variables cover the period 2000-2013. All variables are described in detail in Appendix C, Table C2. Because of missing data on consanguinity or corruption (partly due to changes in the number of provinces since the year 2000), our analysis is based on data from 101 provinces. Figure 4 displays our measures of corruption and consanguinity by province in Italy. We use the log transformation of consanguinity rates to bring the contrast into sharper relief for the figure, but our statistical analysis uses the raw rate of consanguinity as in the cross-country analysis. Grey-colored regions in the figures have missing data. Overall, we find a positive and highly significant correlation between consanguinity rates and our measure of corruption in Italian provinces (Spearman’s ρ = 0.55, p-value < 0.001, N=101). Figure 4: Corruption and consanguinity in Italy. As in the cross-country analysis, we regress our measure of corruption on a number of controls and include consanguinity rates as our measure of sub-ethnic fractionalization. In such analyses, per capita income and the relative size of the agricultural sector “are often used as proxy variables for the level of development” (Del Monte and Papagni, 2007, p. 390). Again, due to the likely endogeneity of income per capita to corruption, we prefer to use the share of agriculture in the regression, though we report specifications including both. In fact, the two variables are highly correlated (Spearman’s ρ = -0.37, p-value < 0.001, N=101), reiterating that share of agriculture is a good proxy for per capita income. We also control for the population of the provinces. These variables are averaged over 2000-2013. where corruption and organized crime “are closely intertwined” (p. 157) and “criminal organizations such as the mafia are the most visible in terms of exercising power” (p. 163). 27 In addition to our correlational estimates using consanguinity rates, we also attempt to address causality by exploiting plausibly exogenous variation in historical exposure to the marriage laws of the Catholic church due to Arab domination of southern Italy from the 7th-10th centuries. This historical episode led to suppression of the Catholic church in some regions and thus facilitates an instrumental variables approach that provides evidence of a causal relationship between consanguinity and corruption. Main findings Table 8 displays the results of our first set of regressions. Our baseline specification in column (1) reveals a large and significant coefficient on agricultural share of income, indicating that less developed parts of Italy also exhibit lower institutional quality. In column (2), we introduce our measure of sub-ethnic fractionalization: the consanguinity rate. From the estimates in column (2), the effect of consanguinity on corruption is quite large: a 1 standard deviation increase in consanguinity rate is associated with a roughly 0.6 standard deviation increase in associative crimes per hundred thousand inhabitants in the province. Moreover, the 55 percentage point difference in consanguinity involved in moving from the least to the most consanguineous region would predict an increase of roughly 2.8 associative crimes per hundred thousand, about 45% of the difference between the most and least-corrupt region. (1) Basic model (2) and Consanguinity (3) Income instead of Share of agriculture (4) Share of agriculture as instrument (5) both Income and Share of agriculture (6) without Income and Share of agriculture 5.371*** (1.117) 4.984*** (1.167) 0.451 (1.654) 0.763 (1.505) -0.011 (0.132) -0.034 (0.083) 1.162 (1.607) -0.161 (0.220) -0.264 (0.273) 2.274 (2.074) 5.155*** (1.115) 3.705 (3.550) 0.035 (0.133) 0.018 (0.092) 0.662 (1.624) 5.427*** (1.081) 12.707*** (4.408) 0.069 (0.124) 5.144*** (1.099) 3.472 (3.243) 0.022 (0.115) 101 0.430 101 0.424 101 0.396 101 0.430 101 0.423 VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Constant Observations 101 R-squared 0.106 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.011 (0.116) 1.000 (1.521) Table 8: Regression analysis of the relationship between consanguinity and corruption in Italy. In columns (3) - (6), we show that our results are robust whether we control for share of agriculture or income, or both or neither. Contrary to the cross-country analysis, population, income and its proxy are not significant in the regressions. This suggests that between-province variability of population or the degree of development is too low to capture its effect on corruption. Consanguinity is associated with significantly higher corruption in all specifications. 28 Table 9 displays the results of our first set of regressions including additional geographic and climatic controls: latitude, mean annual temperature and precipitation, soil suitability for agriculture, distance to the coast, average elevation, average slope, and ruggedness. (1) Basic model (2) and Consanguinity (3) Income instead of Share of agriculture (4) Share of agriculture as instrument (5) both Income and Share of agriculture (6) without Income and Share of agriculture 4.506* (2.421) 4.440* (2.470) -0.359*** (0.130) -0.271* (0.147) -0.032 (0.124) 0.029 (0.081) -0.278* (0.148) -0.067 (0.181) -0.024 (0.228) -0.269* (0.148) 4.472* (2.460) 0.991 (3.631) -0.021 (0.132) 0.042 (0.088) -0.273* (0.149) 4.470* (2.395) 1.537 (3.469) -0.020 (0.123) 4.451* (2.443) 0.366 (3.385) -0.050 (0.113) -0.273* (0.146) yes 101 0.502 yes 101 0.502 yes 101 0.501 yes 101 0.503 yes 101 0.502 VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Latitude Additional controls yes Observations 101 R-squared 0.459 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.051 (0.111) Table 9: Replication of Table 8 including controls for climate and geography.Additional controls include mean temperature, annual precipitation, soil suitability for agriculture, distance to the coast, elevation, slope, and ruggedness. Full estimates reported in Appendix Table E9. In column (1) of Table 9, which is analogous to column (1) of Table 8, we observe a negative and highly significant effect of latitude, consistent with the perception that northern Italy is less corrupt. This difference is usually attributed to cultural differences between the south and north (Banfield, 1958; Putnam et al., 1994). As shown in the table, the effect of latitude becomes smaller and its standard errors become larger once we include consanguinity in columns (2)-(6), which are analogous to the same columns in Table 8. In these specifications, the consanguinity rate remains a statistically significant determinant of corruption, though its magnitude is slightly smaller and the associated standard errors slightly larger than above, likely due to the correlation between latitude and consanguinity shown in Figure 4b. IV Regressions. These findings suggest that some of the modern cultural differences between northern and southern Italy might be driven, in part, by historical differences in marriage patterns between the two regions, perhaps “as a remote consequence of Arab domination in Sicily and southern Italy in the eighth to the eleventh centuries” (Cavalli-Sforza et al., 2004, p. 3). The possibility of differences in mating patterns due to a historical event such as Arab domination provides a potential exogenous variation that is worthwhile to explore. Since the Arab conquest also brought new religious authorities, it introduced exogenous variation in the exposure to the Catholic church (and its policies on legitimate marriage). There are 42 Roman Catholic ecclesiastical provinces in Italy. Each ecclesiastical province is served by a metropolitan archdiocese. Historical data on the all dioceses and archdioceses of Italy is available from their 29 date of establishment. For each ecclesiastical province, we calculated the number of active years of its archdioceses from the date of establishment of the first archdiocese to the present (see the appendix for more details). This measure captures two sources of variation. First, the establishment dates of archdioceses vary from 100 AD to 1100 AD. Second, archdioceses in five ecclesiastical province (covering 10 administrative provinces today) in southern Italy were suppressed for more than 200 years between the 7th and 10th centuries, coinciding with the historical Arab domination of southern Italy. We matched the data on the number of active years of archdioceses in ecclesiastical provinces to today’s administrative provinces. This measure is strongly and negatively correlated with consanguinity rates in the 20th century (Spearman’s ρ = -0.390, p-value < 0.000, N=101). (1) Basic model (2) Income instead of Share of agriculture (3) both Income and Share of agriculture (4) without Income and Share of agriculture -0.002*** (0.000) 10.365** (4.181) 0.128 (0.120) -0.002*** (0.000) -0.002*** (0.000) 0.014 (0.150) -0.136 (0.111) 3.915** (1.823) -0.002*** (0.000) 10.778** (4.559) 0.152 (0.156) 0.032 (0.113) 2.212 (1.915) 101 0.129 101 0.186 101 0.119 VARIABLES No. of active years Share of agriculture Log population Log value added per capita Constant 2.363 (1.766) Observations 101 R-squared 0.186 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.111 (0.120) 3.458* (1.765) Table 10: Reduced form regressions with active years of archdioceses. Table 10 reports reduced form regression results showing that provinces with a higher number of active years of archdioceses exhibit lower rates of corruption. Table 11 reports regression results using this measure as an instrumental variable for consanguinity and supports a causal interpretation; in all cases, we fail to reject the null hypothesis of exogeneity. In Table E10 we also check the robustness of the results to including geographic and climatic controls. Controlling for geography, the results remain robust. Latitude is highly correlated with active years of archdioceses (Spearman’s ρ = 0.434, p-value < 0.000, N=101). When it is included in the regression, consanguinity becomes insignificant. This provides further evidence that differences in mating patterns are mediated by differences in latitude, which makes sense; the Church’s influence was weaker in southern regions because of its geographic distance from Rome and due to historical Arab, Norman, and Spanish dominations of these regions. Robustness checks. As in the cross-country analysis, there are a few other potential confounds which, if addressed, would increase confidence in our results. In Appendix E.4, we report 30 (1) Basic model (2) Income instead of Share of agriculture (3) both Income and Share of agriculture (4) without Income and Share of agriculture 8.306*** (2.396) -2.205 (5.420) -0.006 (0.146) 8.576*** (2.329) 8.658*** (2.538) -1.731 (5.552) 0.045 (0.156) 0.077 (0.094) 0.530 (1.973) 8.035*** (1.891) VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Constant Observations R-squared 0.955 (1.899) 0.066 (0.172) 0.102 (0.116) 0.291 (2.160) 101 101 101 0.307 0.287 0.281 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.001 (0.145) 0.811 (1.913) 101 0.326 Table 11: Active years of archdioceses as an instrument for consanguinity. further analyses based on column (2) of Tables 8 and 9 and including additional variables that have been suggested to mediate institutional quality (hence corruption); these include measures of civil society, civic engagement, family types, family ties, and historical political domination by various groups (see e.g. Putnam et al., 1994; Todd, 1990; Alesina and Giuliano, 2014; Di Liberto and Sideri, 2015). The relationship between consanguinity rates and corruption remains robust; see Tables E11 and E12. Our results are also robust to altering cutoff dates for computing consanguinity rates from 1945-1964 to 1950-1964, 1955-1964, or 1960-1964; see Table E13. Since associative crime is not a measure of actual corruption, but only of the crimes reported and detected by the police, it may underestimate the true phenomenon because of judicial inefficiency. Moreover, the most corrupt regions may be those in which such crimes are least likely to be reported or detected, which further suggests that we may underestimate the relationship between consanguinity and corruption. Thus, in Appendix E.4, we show that our regression results are robust to using an alternative measure of corruption from Golden and Picci (2005) based on computing the ratio of the value of existing physical infrastructure stocks (in 1998) to the expected value of infrastructure given government expenditures over the period 19541997. “The intuition underlying this measure is that, all else equal, governments that do not get what they pay for are those whose bureaucrats and politicians are siphoning off more public monies in corrupt transactions” (Golden and Picci, 2005, p. 41). The measure is available for 90 provinces for which we also have consanguinity data, and it is highly correlated with consanguinity (Spearman’s ρ = -0.63, p-value < 0.001, N=90). See Figure E2 and Table E14. Summary. In our baseline specifications and controlling for a variety of possible climatic, geographical, and historical confounds, consanguinity is a significant predictor of corruption. Moreover, our IV results provide some evidence that the relationship is causal. As in the crosscountry analysis, these data are highly consistent with a model in which corruption is in part 31 driven by sub-ethnic fractionalization, as reflected in consanguinity.26 3.3 Laboratory Experiments The third prong of our approach employs experiments to directly compare norms of favoritism and corruption in a stylized laboratory game conducted in two countries with similar levels of ethnic fractionalization, but very different marriage practices. Subjects play a one-shot, threeperson bribery game, in which Person A may attempt to bribe Person B, who can choose whether to incur a corruption cost, thereby helping A and harming Person C. In each triplet, there are two “related” people (either co-ethnics or kin) and one unrelated person. By varying assignment of people to roles (e.g. A related to B vs. A related to C), we can test how nepotism and corruption vary within-country across different relationships (as e.g. implied by Hamilton’s theory), and by comparing across countries we can see whether norms vary with marriage practices as predicted. We conducted the experiments with students in two large, ethnically heterogeneous cities: Vancouver, Canada and Urmia, Iran.27 Importantly for our motivation, both Iran (0.67) and Canada (0.71) are similarly ethnically fractionalized according to the measure developed in Alesina et al. (2003). Vancouver is the largest city in the province of British Columbia and has seen a large influx of people from East and South Asia in the last 30 years. Today, the two most common ethnic backgrounds are English and Chinese. Urmia is the capital of West Azerbaijan Province in Iran, and the city has been home to numerous ethnic groups during its long history. Today, Azeri Turks and Kurds are the two main ethnic groups in the city. While Vancouver and Urmia are both multi-ethnic metropolises with populations around 600,000, they differ sharply in the importance and structure of kin networks. Extended family is very important in Urmia, and clans and tribes continue to have influence in the region.28 Moreover, consanguineous marriages are very common in Urmia; consanguinity rates for Azeri Turks and Kurds in Iran are 32% and 40%, while the rate for Iran as a whole is 32% (Bittles and Black, 2015). On the contrary, family structure in Vancouver is mostly nuclear and consanguinity is very rare; in Canada as a whole, the rate is 1.5% (Bittles and Black, 2015). In other words, Urmia exhibits substantial sub-ethnic fractionalization; while Vancouver does not, possibly yielding differences in norms of favoritism and corruption. 26 One further implication of our theory that our current data does not allow us to test is that local crime, in which neighbors are targets, should be less common in regions with high consanguinity rates. Buonanno and Vanin (2015) provide related evidence consistent with our findings and with this hypothesis as well. Using the distribution of surname frequencies from Italian phonebooks, they show that municipalities with more surname concentration (i.e. with more in-marriage) exhibit both more tax evasion (of a federal tax) and less local crime. 27 Our experiments in Iran were conducted in collaboration with the Moaser Research Center which possesses a permit from the Ministry of Science, Research and Technology to conduct research in Economics and Management. The center took full responsibility for planning, ethical review and official approvals to run experiments in Iran. 28 http://www.iranicaonline.org/articles/kurdish-tribes 32 One might argue that experiment results in Vancouver and Urmia cannot be readily generalized to Canada and Iran, but nevertheless the cross-cultural comparison provides suggestive evidence that speaks to our hypotheses. Moreover, subsequent robustness checks conducted in Tehran (described below) indicate no difference across those cities in Iran. The game. Subjects participated in the one-shot bribery game shown in Figure 5, which is a simplified version of the game described in Section 2. Unlike above, there is no risk of punishment. We remove this possibility to eliminate a source of noise, as the impact of punishment has been investigated in previous experimental studies (e.g. Abbink et al., 2002). For simplicity, we also assume that only one citizen suffers the negative externality of corruption. We chose our parameters so that bribery (A choosing “Transfer”) and corruption (B choosing “Accept/Right”) do not occur in the subgame perfect Nash equilibrium with unrelated, payoffmaximizing agents. Our design also ensures that the payoffs resulting from successful bribery and corruption are both inequitable and inefficient relative to the status quo, so that distributional and social welfare preferences do not predict bribery/corruption. However, if A and B are sufficiently related, bribery and corruption can occur in equilibrium (via Hamilton’s rule). More broadly, if we observe bribery and corruption, these must be driven by factors other than selfish payoff maximization, (pro)social preferences or concerns for efficiency. Our preferred interpretation is that such behavior reflects background social norms related to favoritism and corruption. Moreover, if bribery and corruption rates are higher in Iran than in Canada, this would provide evidence of normative differences that are at least correlated with observed higher levels of sub-ethnic fractionalization in Iran. Because of the negative connotation attached to words like “bribe”, “corrupt effort”, and “negative externality”, we applied non-normative language in the experiment using words like “Transfer/Not-transfer”, “Reject/Accept”, “Right/Left”, “payoff added/deducted”.29 In order to make full use of our limited sample, we elicited Person B’s strategies using the strategy method, so that we know what he/she would have chosen, had Person A offered the bribe. Payoffs were shown in Experimental Currency Units, and our conversion rates were designed to assure that the stakes were purchasing power-equivalent across the societies. We used local pizza prices as our measure of students’ purchasing power since both cities have many pizza shops, and pizza is popular among students. The price of a medium pizza in Vancouver including tax is around 15 CAD and in Urmia around 15000 Tomans.30 In Vancouver, we paid $7 for arriving on time and converted ECU at a rate of 10 ECU = $1. In Urmia, we paid 29 While Abbink and Hennig-Schmidt (2006) and Barr and Serra (2009) find no effect of framing in laboratory corruption games, their experiments were run with a single sample. To avoid risk of culture-specific framing effects, we erred on the side of caution. 30 Although the Rial is the official currency of Iran, Iranians employ the term ‘Toman’, meaning 10 rials 33 7000 Tomans for arriving on time and paid 10 ECU = 1000 Tomans. At the conclusion of the experiment a subset of subjects completed a post-experiment questionnaire (see Appendix F.3). Person A Not transfer Transfer 40 Person B  100 100 160  Reject Accept Person B   95 100 160 Left   55 140 160 Right   160 135 55 Figure 5: Bribery game in the experiment. Each terminal node shows the payoffs for Person A, Person B and Person C, from top to bottom. Design and treatments We employ a one-shot, between-subject design. Our treatments vary 1) whether the related pair of subjects in each triplet are related by kinship (K) or co-ethnicity (C), and 2) the assignment of subjects to roles. This generates the following treatments, where S refers to the Stranger: KKS/CCS. Persons A and B are kin/co-ethnics. A(B) knows that B(A) is kin/co-ethnic and also knows that B(A) knows that A(B) is kin/co-ethnic. No information regarding the ethnicity of Person C is given to A or B, and C has no information about the ethnicity of the other players or their being related by kin or co-ethnicity. KSK/CSC. Persons A and C are kin/co-ethnics. A(C) knows that C(A) is kin/co-ethnic and also knows that C(A) knows that A(C) is kin/co-ethnic. No information regarding the ethnicity of Person B is given to A or C, and B has no information about the ethnicity of the other players or their being related by kin or co-ethnicity. SKK/SCC. Person B and C are kin/co-ethnics. B(C) knows that C(B) is kin/co-ethnic and also knows that C(B) knows that B(C) is kin/co-ethnic. No information regarding the ethnicity of Person A is given to B or C, and A has no information about the ethnicity of the other players 34 or their being related by kin or co-ethnicity. We randomly assigned one of the three treatments at the kin/ethnic level to each triple of subjects. Subjects were matched in triplets with their kin/co-ethnic based on their self-reported kinship/ethnic origin in a pre-experiment questionnaire (see Appendix F.2). Pre-experiment questionnaires were collected from subjects online or in paper, prior to the experiment. In the questionnaire, in addition to ethnic origin, we collected demographic information such as age group, gender, degree, and field of study to avoid highlighting the aim of our research. Before subjects learn their roles and information about subjects in the other roles, we mentioned that “you might observe some background information from the pre-experiment questionnaire about participants in the other roles.” Also, we always included age-group information for other players in addition to ethnic origin information. We chose 18-30 as the age group to present in the experiment because it covered all the subjects in our sample; therefore, age information was the same for all treatments. We were hoping that these cautions along with the between-subject design would minimize any possible experimenter effect.31 The instructions and more detailed procedures of the experiment are presented in Appendix F, including sample pages showing how we exchanged information between subjects in the three-player game. Subject pool. For the co-ethnic treatments, our subjects in Vancouver were 180 Canadianborn undergraduate students with English or Chinese origins from the University of British Columbia and Simon Fraser University, both located in the Vancouver area. The subjects in the ethnic treatment in Urmia, Iran consisted of 180 Iranian-born undergraduate students with Azeri or Kurdish origins, taking courses at Urmia University during summer 2015. From each city, we collected data from 20 triplets in each of the three ethnic-level treatments (CCS, CSC, SCC). For the kin treatments, we collected data on all three matching schemes in Urmia and only one matching scheme in Canada (KKS) since recruiting subjects for the Kin treatment in Canada was extremely difficult. For these treatments, we asked students whether their sibling would like to participate in the experiment, and if they answered with “Yes”, we also asked the occupation and age group of their sibling. Then we invited those pairs of siblings who both were 18-30 years old and students. For each pair of siblings, another randomly chosen student was invited to participate in the three-person game. In Urmia 180 subjects (60 sibling pairs + 60 others) participated in the three kin level treatments (KKS, KSK, SKK), with 20 triplets per treatment. In 31 When we began our experiments in Vancouver, to present ethnic origin information, we used the word “ethnic origin” on the information page. Later, we dropped this word for the rest of the experiments in Vancouver and all the experiments in Iran, considering that it might affect subjects’ choices due to the salience of “ethnicity”. However, the results of experiments in Vancouver indicate that using the word ”ethnic origin” had no effect on behavior. 35 Vancouver, 39 subjects (13 sibling pairs + 13 others) participated in the KKS treatment.32 Finally, we conducted two robustness checks: 1) a “Friend” treatment in both countries, in which subjects were asked to bring a close friend to the laboratory and they participated together in the role of persons A and B. The Friend treatment is designed to test whether observed corruption among kin is driven by familiarity. In particular, the possibility in our design of ‘gains from exchange’ in corruption suggests that trusted friends might also be prone to cooperate at the expense of unknown 3rd parties; 2) a “High Cost” treatment involving variants of the Kin, Friend and Stranger treatments in which we increase the cost of corruption by player 2 to eliminate the mutual gains from corruption (such that player 2 is worse off by engaging in corruption than if player 1 had not offered the bribe). By varying the cost of helping, we can test the strength of norms of favoritism within and across countries. In total we collected data on an additional 90 subjects in Canada (10 triplets in the FFS treatment and 20 triplets in the FFS High treatments) and on 180 additional subjects in Tehran, Iran (12 triplets each in the SSS, SSS High, FFS High, KKS High and KKS High Cousins treatments). We conducted the SSS treatments to check for baseline differences in corruption among strangers between Tehran and Urmia. Hypotheses. One direct implication of the theory in Section 2 is that the frequency of offering a bribe by Person A and the frequency of accepting the bribe and making the corrupt effort by Person B are positive functions of their relatedness to one another and negative functions of their relatedness to Person C. More important for our purposes, comparing behavior across countries we can test whether norms of favoritism and corruption are more prominent in Iran where we observe increased sub-ethnic fractionalization. While we would expect kin to engage in favoritism in both countries, there is scope for differences in the other treatments. In sum: Hypothesis 1a: Bribery is (weakly) increasing in relatedness of A to B and (weakly) decreasing in relatedness of A to C. Hypothesis 1b: Corruption is (weakly) increasing in relatedness of A to B and (weakly) decreasing in relatedness of B to C. Hypothesis 2: Bribery/Corruption among non-kin are higher in Iran than in Canada, due to increased sub-ethnic fractionalization. Experimental findings First we report within-country comparisons testing for the effects of relatedness on behavior, and then we report between-country comparisons, which provide suggestive evidence for the effects 32 We exclude from the sample one father-daughter pair since all other kin observations were on siblings. Interestingly, the father played the role of person B and was one of very few subjects to reject the bribe; his explanation for his decision indicated that he planned to compensate his daughter for her loss outside of the experiment. 36 of sub-ethnic fractionalization. We conclude with a discussion of our Friend treatment, which highlights our interpretation of the findings. The experimental results from our primary treatments are presented in Table 12. Each entry in the table shows the fraction of subjects choosing to engage in bribery or corruption, by treatment and matching scheme. Relatedness. Let µk be the relative frequency of bribery in matching scheme k, with k ∈ {KKS, KSK, SKK } in the Kin treatment and k ∈ {CCS, CSC, SCC } in the Co-ethnic treatment. Similarly, let νk be the relative frequency of corruption under matching scheme k. From the discussion in Section 2 and Appendix A it follows that kin are more related than co-ethnics and that co-ethnics are more related (in expectation) than strangers. Iran A & B related A & C related B & C related Kin Bribery Corruption 18/20 19/20 1/20 7/20 8/20 1/20 Canada Co-Ethnic Bribery Corruption 17/20 16/20 10/20 10/20 14/20 9/20 Kin Bribery Corruption 13/13 12/13 NA NA NA NA Co-Ethnic Bribery Corruption 7/20 9/20 6/20 7/20 9/20 6/20 Table 12: Relative frequency of bribery and corruption by treatment in Urmia, Iran and Vancouver, Canada. Thus, hypothesis 1a has the following implication: µKKS ≥ µCCS ≥ µSKK/SCC ≥ µCSC ≥ µKSK . Using the data from Iran, a Cochran-Armitage test rejects the null hypothesis of equal relative frequency across the treatments in favor of the ordered alternative (technically, the alternative is that the ordering is weak, with at least one relationship strict, Z = 5.85, p-value < 0.001). With our Canadian data, we can only test a portion of the hypothesis, namely µKKS ≥ µCCS ≥ µSCC ≥ µCSC , but again a Cochran-Armitage test rejects the null hypothesis (Z = 3.19, p-value = 0.001). Similarly, hypothesis 1b implies that νKKS ≥ νCCS ≥ νKSK/CSC ≥ νSCC ≥ νSKK . Using Iranian data, a Cochran-Armitage test rejects the null hypothesis of equal relative frequency in favor of the ordered alternative (Z = 6.08, p-value < 0.001). Again, with the Canadian data, we can only test a portion of the hypothesis, namely νKKS ≥ νCCS ≥ νCSC ≥ νSCC , and a Cochran-Armitage test rejects the null hypothesis (Z = 3.28, p-value = 0.001). Finding 1: In both countries, the data are consistent with the relatedness hypothesis (1a and 1b). Cross-country comparisons. Table 12 reveals evidence of kin favoritism in both countries. We focus on the top row of the table since we have data in the kin treatment only in the KKS cell in Canada. Pooling over the decisions of Persons A and B in the KKS treatment, subjects took 25/26 (96%) corrupt actions in Canada and 37/40 (93%) corrupt actions in Iran; the degree of kin favoritism is virtually indistinguishable across countries (two-tailed proportions test, χ2 = 0.006, p-value = 0.94). 37 However, we see some differences in the degree of ethnic favoritism. Again pooling over the decisions of Persons A and B, and focusing on the CCS treatment, we see subjects took 16/40 (40%) corruption actions in Canada and 33/40 (83%) corrupt actions in Iran. A two-tailed proportions test confirms that these differences are statistically significant (χ2 = 13.5, p-value < 0.001). Using rows 2 and 3 of the table, we can also test for differences in the willingness to harm, rather than help, a co-ethnic by comparing bribery rates when A and C are co-ethnics and corruption rates when B and C are co-ethnics. Here we see 12/40 (30%) such actions taken in Canada and 19/40 (45%) such actions taken in Iran; the difference is not significant (two-tailed proportions test, χ2 = 1.90, p-value = 0.17). Thus, we see that Iranian subjects are substantially more willing to help and no more willing to harm co-ethnics than are Canadian subjects. Using the other cells of Table 12, we can also test for differences in the frequency of corrupt actions taken on behalf of strangers. In particular, the decision of Person A to offer a bribe the SCC/SKK treatments was made knowing nothing about the counterpart, as was the decision of Person B to accept and reciprocate (i.e. corruption) in the CSC/KSK. These decisions thus capture the rate of corrupt actions undertaken with strangers. Pooling over the decisions of A and B, we see 16/40 (40%) corrupt actions in Canada and 39/80 (49%) corrupt actions in Iran, and thus we have no evidence of significant differences in corrupt behavior when interacting with strangers (two-tailed proportions test, χ2 = 0.51, p-value = 0.48). Finding 2: Iranian subjects exhibit more ethnic favoritism than Canadians, though both countries show similar levels of kin favoritism and similar behavior toward strangers. This is consistent with our hypothesis that increased sub-ethnic fractionalization in Iran strengthens norms of local favoritism. Robustness: the Friend and High Cost treatments. One limitation of our basic design is that the effect of kinship cannot be separated easily from that of familiarity. Kin, especially those of similar ages who are willing to attend a laboratory experiment together, are likely to have a good relationship built on reciprocity and generosity that might be reflected in their willingness to cooperate with one another (at a third party’s expense). The Friend treatment allows us to highlight an important aspect of our interpretation of the findings: that differences in behavior reflect differences in norms of local favoritism, which we believe can be explained in part by difference in marriage practices. In the FFS treatment, conducted only in Canada, we observe bribery/corruption in 19/20 (95%) decisions, indicating that corrupt acts in these small stakes decisions are just as likely among friends as among kin and substantially more so than among co-ethnics (two-tailed proportions tests, χ2 = 0.00, p-value = 1 and χ2 = 14.4, p-value < 0.001, comparing FFS to KKS 38 and CCS, respectively). In almost all societies, norms of favoritism exist among friends, and like marriage norms, these may constitute an instance in which the biological machinery for kin altruism is co-opted to support cooperation among non-relatives.33,34 Canada Iran Kin (low cost) Bribery Corruption 13/13 12/13 18/20 19/20 Friend (low cost) Bribery Corruption 9/10 10/10 NA NA Stranger (low cost)a Bribery Corruption 9/20 7/20 30/52 22/52 Canada Iran Kin (high cost)b Bribery Corruption NA NA 24/24 22/24 Friend (high cost) Bribery Corruption 13/20 9/20 11/12 10/12 Stranger (high cost) Bribery Corruption NA NA 1/12 1/12 a Includes b Includes 40 pairs from Urmia in the SKK and SCC treatments. 12 sibling and 12 cousin pairs. Table 13: Summary of relative frequency of bribery and corruption when A & B are related. The adage that “blood is thicker than water” may nevertheless apply to our bribery game, and thus we hypothesized that the behavior of kin and friends might differ if there were no gains from exchange in our experiment (i.e., if the payoffs to A and B after Transfer→Accept→Right were 160 and 90, respectively). To test this conjecture we also conducted the High Cost variants of the treatment where player A and B are related as friends in both countries (N = 20 triplets in Canada and N=12 in Iran). In Iran, we also conducted versions of the High Cost treatment where players A and B are related as siblings and as cousins (N = 12 for both).35 As seen in Table 13, we did not conduct a full factorial experiment due to difficulties recruiting subjects (we struggled to recruit kin in Canada) and to obviousness of the results (in case of the Low Cost FFS treatment in Iran, which we had no reason to think would differ from the High Cost FFS treatment). The new experiments are described in more detail in appendix F.4. Consistent with our conjecture, the key finding here is that in Canada, high effort cost significantly decreases the frequency of bribery and corruption among friends relative to the experiment with low effort cost among either kin or friends. Pooling over the decisions of A and B in Canada, the frequency of corrupt acts is 22/40 (55%) in the High Cost FFS treatment versus 19/20 (95%) in the Low Cost FFS treatment and 25/26 (96%) in the Low Cost KKS treatment (two-tailed proportions tests, χ2 = 8.1, p-value = 0.004, and χ2 = 11.1, p-value = 0.001, comparing High Cost FFS to Low Cost FFS and Low Cost KKS, respectively). 33 But there is evidence that friends are more closely related than random individuals; see Appendix A. co-optation is reflected in the use of kinship words, e.g. ‘brother’ and ‘sister’, to refer to close friends. 35 In Iran, these followup experiments were conducted in Tehran instead of Urmia. One reason to do this was to see if our basic results were unique to Urmia or generalized to the larger and more cosmopolitan city of Tehran. We ran a low cost SSS treatment in Tehran and the frequency of bribery (8/12) and corruption (5/12) were not significantly different from the SKK and SCC treatments conducted in Urmia (22/40 and 17/40, p-values > 0.7, two-tailed proportions tests). In addition, as seen in Table 13, experiments with siblings and cousins in the treatment with high effort cost in Tehran rules out the possibility of weak kin and relative ties in Tehran relative to Urmia. 34 This 39 Perhaps surprisingly, in Iran, the frequency of bribery and corruption among friends in our high effort cost treatment is not distinguishable from that observed among kin, in either the Low or High cost treatment. Pooling over the bribe/accept decisions in Iran, the frequency of corrupt acts in the High Cost FFS is 21/24 (88%) versus 37/40 (93%) in the Low Cost KKS and 46/48 (96%) in the High Cost KKS, which includes both siblings and cousins since they were indistinguishable (two-tailed proportions tests, χ2 = 0.05, p-value = 0.82 and χ2 = 0.67, p-value = 0.41, comparing High Cost FFS to Low Cost KKS and High Cost KKS, respectively). In fact, when we compare the High Cost FFS treatment across countries, we see a significantly higher rate of corrupt acts in Iran than in Canada among friends (two-tailed proportion test, χ2 = 5.79, p-value = 0.016). In sum, in our High Cost treatment, our Iranian subjects are equally willing to engage in corruption on behalf of kin and friends, but in Canada, the High Cost treatment significantly decreases the willingness to benefit friends, as compared to kin. Finding 3: Iranian subjects exhibit more favoritism toward friends than Canadian subjects. This suggests that, in a consanguineous society—where the gains from local altruism are higher—norms of favoritism among friends tend to be stronger too. 4 Conclusion Countries around the world exhibit vast differences in levels of corruption, and understanding the sources of these differences is crucial to improving governance. We provide evidence from cross-country, within-country and experimental data that in-marriage and concomitant sub-ethnic fractionalization is an important determinant of corruption. In regions with high subethnic fractionalization, corruption is relatively prevalent, even after controlling for previously studied deep determinants of corruption. Motivating our argument with the notion of inclusive fitness from biology, we argue that differences in mating practices and family structure provide the source of this correlation. In particular in-marriage practices, which increase relatedness between individuals at the local level, raise the relative returns to norms of kin altruism, nepotism, and favoritism; while, out-marriage practices raise the relative returns to norms of impartial, generalized cooperation. While we focus on the example of cousin marriage, we would expect the same logic to apply to other local in-marriage practices (e.g. locally endogamous marriage within a caste). Our findings suggest that historical differences in mating practices (due to religion, geography, and local circumstance) may have had a powerful influence on today’s norms. 40 References A BBINK , K. and H ENNIG -S CHMIDT, H. (2006). Neutral versus loaded instructions in a bribery experiment. Experimental Economics, 9 (2), 103–121. —, I RLENBUSCH , B. and R ENNER , E. (2002). An experimental bribery game. Journal of Law, Economics, and Organization, 18 (2), 428–454. A LATAS , V., C AMERON , L., C HAUDHURI , A., E RKAL , N. and G ANGADHARAN , L. (2009). Subject pool effects in a corruption experiment: A comparison of indonesian public servants and indonesian students. Experimental Economics, 12 (1), 113–132. A LESINA , A., D EVLEESCHAUWER , A., E ASTERLY, W., K URLAT, S. and WACZIARG , R. (2003). Fractionalization. Journal of Economic growth, 8 (2), 155–194. — and G IULIANO , P. (2010). The power of the family. Journal of Economic growth, 15 (2), 93–125. — and — (2011). Family ties and political participation. Journal of the European Economic Association, 9 (5), 817–839. — and — (2014). Family ties. In P. Aghion and S. N. Durlauf (eds.), Handbook of Economic Growth, vol. 2A, North Holland: Netherlands, pp. 177–215. —, — and N UNN , N. (2013). On the origins of gender roles: Women and the plough*. The Quarterly Journal of Economics, p. qjt005. — and L A F ERRARA , E. (2005). Ethnic diversity and economic performance. Journal of Economic Literature, 43, 762– 800. A SHRAF, Q. and G ALOR , O. (2013). The’out of africa’hypothesis, human genetic diversity, and comparative economic development. The American Economic Review, 103 (1), 1. B AILEY, D. H., H ILL , K. R. and WALKER , R. S. (2014). Fitness consequences of spousal relatedness in 46 small-scale societies. Biology letters, 10 (5), 20140160. B ANFIELD , E. C. (1958). The moral basis of a backward society. New York: Glencol, 3, 4. B ARR , A. and S ERRA , D. (2009). The effects of externalities and framing on bribery in a petty corruption experiment. Experimental Economics, 12 (4), 488–503. — and — (2010). Corruption and culture: An experimental analysis. Journal of Public Economics, 94 (11), 862–869. B ERG , J., D ICKHAUT, J. and M C C ABE , K. (1995). Trust, reciprocity, and social history. Games and economic behavior, 10 (1), 122–142. B ERGSTROM , T. C. (2003). The algebra of assortative encounters and the evolution of cooperation. International Game Theory Review, 5 (03), 211–228. B ITTLES , A. and B LACK , M. H. (2015). Global patterns & tables of consanguinity. B ITTLES , A. H. (2012). Consanguinity in context, vol. 63. Cambridge University Press. B ONDARENKO , D., K AZANKOV, A., K HALTOURINA , D. and K OROTAYEV, A. (2005). Ethnographic atlas xxxi: Peoples of easternmost europe. Ethnology, pp. 261–289. B OWLES , S. and P OSEL , D. (2005). Genetic relatedness predicts south african migrant workers’ remittances to their families. Nature, 434 (7031), 380–383. 41 B RUK , S. and A PENCHENKO , V. (eds.) (1964). Atlas Narodov. Moscow: Miklukho-Maklai Ethnological Institute at the Department of Geodesy and Cartography of the State Geological Committee of the Soviet Union. B UONANNO , P. and VANIN , P. (2015). Social closure, surnames and crime, quaderni-Working Paper DSE. C AMERON , L., C HAUDHURI , A., E RKAL , N. and G ANGADHARAN , L. (2009). Propensities to engage in and punish corrupt behavior: Experimental evidence from australia, india, indonesia and singapore. Journal of Public Economics, 93 (7), 843–851. C AMPOS , J. E., L IEN , D. and P RADHAN , S. (1999). The impact of corruption on investment: Predictability matters. World development, 27 (6), 1059–1067. C AVALLI -S FORZA , L. L., M ENOZZI , P. and P IAZZA , A. (1994). The history and geography of human genes. Princeton university press. —, M ORONI , A. and Z EI , G. (2004). Consanguinity, Inbreeding, and Genetic Drift in Italy. No. 39 in Monographs in Population Biology, Princeton University Press: Princeton. C HUAH , S.-H., FAHOUM , R. and H OFFMANN , R. (2013). Fractionalization and trust in india: A field-experiment. Economics Letters, 119 (2), 191–194. C OX , D. and FAFCHAMPS , M. (2007). Extended family and kinship networks: economic insights and evolutionary directions. Handbook of development economics, 4, 3711–3784. D EL M ONTE , A. and PAPAGNI , E. (2007). The determinants of corruption in italy: Regional panel data analysis. European Journal of Political Economy, 23 (2), 379–396. D I L IBERTO , A. and S IDERI , M. (2015). Past dominations, current institutions and the italian regional economic performance. European Journal of Political Economy, 38, 12–41. D INCER , O. C. (2008). Ethnic and religious diversity and corruption. Economics Letters, 99 (1), 98–102. E ASTERLY, W. and L EVINE , R. (1997). Africa’s growth tragedy: policies and ethnic divisions. The Quarterly Journal of Economics, pp. 1203–1250. E FFERSON , C., L ALIVE , R. and F EHR , E. (2008). The coevolution of cultural groups and ingroup favoritism. Science, 321 (5897), 1844–1849. E LBAHNASAWY, N. G. and R EVIER , C. F. (2012). The determinants of corruption: Cross-country-panel-data analysis. The Developing Economies, 50 (4), 311–333. E NKE , B. (2017). Kinship systems, cooperation and the evolution of culture. E RMISCH , J. and G AMBETTA , D. (2010). Do strong family ties inhibit trust? Journal of Economic Behavior & Organization, 75 (3), 365–376. F EHR , E. and F ISCHBACHER , U. (2003). The nature of human altruism. Nature, 425 (6960), 785–791. F ELLNER , C. H. and M ARSHALL , J. R. (1981). Kidney donors revisited. In E. J. P. Rushton and R. M. Sorrentino (eds.), Altruism and Helping Behavior: Social, Personality and Developmental Perspectives, Lawrence Erlbaum Associates: New Jersey. F IORINO , N., G ALLI , E. and P ETRARCA , I. (2012). Corruption and growth: Evidence from the italian regions. European Journal of Government and Economics, 1 (2), 126–144. F OX , R. (1967). Kinship and Marriage. Penguin Books: Middlesex, UK. 42 F RANCOIS , P. and Z ABOJNIK , J. (2005). Trust, social capital and economic development. Journal of the European Economic Association, 3 (1), 51–94. F U , F., TARNITA , C. E., C HRISTAKIS , N. A., WANG , L., R AND , D. G. and N OWAK , M. A. (2012). Evolution of in-group favoritism. Scientific Reports, 2 (460). F UKUYAMA , F. (1995). Trust: The Social Virtues and the Creation of Prosperity. New York: Free Press. G ARTH , T. R. (1944). Kinship terminology, marriage practices and behavior toward kin among the atsugewi. American Anthropologist, 46 (3), 348–361. G ILLESPIE , J. (2004). Population genetics: a concise guide. Johns Hopkins University Press. G IULIANO , P. and N UNN , N. (2013). Ancestral characteristics of modern populations. G LAESER , E. L. and S AKS , R. E. (2006). Corruption in america. Journal of Public Economics, 90 (6), 1053–1072. G LATZER , B. (2002). The pashtun tribal system. Concept of tribal society, pp. 265–282. G OLDEN , M. A. and P ICCI , L. (2005). Proposal for a new measure of corruption, illustrated with italian data. Economics & Politics, 17 (1), 37–75. G OODY, J. (1976). Production and reproduction: a comparative study of the domestic domain. 17, Cambridge University Press. — (1983). The development of the family and marriage in Europe. Cambridge University Press. G OUNEV, P. and B EZLOV, T. (2010). Examining the links between organised crime and corruption, center for the Study of Democracy. G RAY, J. P. (1998). Ethnographic atlas codebook. World Cultures, 10 (1), 86–136. — (1999). A corrected ethnographic atlas. World Cultures, 10 (1), 24–85. G REIF, A. (2006). Family structure, institutions, and growth: the origins and implications of western corporations. The American economic review, pp. 308–312. — and TABELLINI , G. (2015). The clan and the corporation: Sustaining cooperation in china and europe. Available at SSRN 2565120. G RIMM , V. and M ENGEL , F. (2009). Cooperation in viscous populations—experimental evidence. Games and Economic Behavior, 66 (1), 202–220. H AMILTON , W. D. (1964a). The genetical evolution of social behaviour. i. Journal of Theoretical Biology, 7 (1), 1–16. — (1964b). The genetical evolution of social behaviour. ii. Journal of Theoretical Biology, 7 (1), 17–52. — (1975). Innate social aptitudes of man: an approach from evolutionary genetics. Biosocial anthropology, 133, 155. H AMMOND , R. A. and A XELROD , R. (2006). The evolution of ethnocentrism. Journal of Conflict Resolution, 50 (6), 926–936. H ARPENDING , H. C. (2002). Kinship and population subdivision. Population and Environment, 24 (2), 141–147. (2014). Big summary post on the hajnal line. https://hbdchick.wordpress.com/2014/03/10/bigsummary-post-on-the-hajnal-line/. HBDCHICK 43 H ENRICH , J. (2015). The Secret of Our Success: How Culture Is Driving Human Evolution, Domesticating Our Species, and Making Us Smarter. Princeton University Press: Princeton. H OBEN , A. D., B UUNK , A. P., F INCHER , C. L., T HORNHILL , R. and S CHALLER , M. (2010). On the adaptive origins and maladaptive consequences of human inbreeding: Parasite prevalence, immune functioning, and consanguineous marriage. Evolutionary Psychology, 8 (4), 658 – 676. H UME , D. (1740). A Treatise of Human Nature. Oxford: Oxford University Press, (2003) edn. J ABER , L., H ALPERN , G. J. and S HOHAT, M. (1998). The impact of consanguinity worldwide. Community Genetics, 1 (1), 12–17. J OHNSON , A. W. and E ARLE , T. K. (2000). The evolution of human societies: from foraging group to agrarian state. Stanford University Press. J OHNSON , N. D. and M ISLIN , A. (2012). How much should we trust the world values survey trust question? Economics Letters, 116 (2), 210–212. J ONES , D. (2016). Socially enforced nepotism: How norms and reputation can amplify kin altruism. PLoS ONE, 11 (6), 1–13. K IRBY, K. R., G RAY, R. D., G REENHILL , S. J., J ORDAN , F. M., G OMES -N G , S., B IBIKO , H.-J., B LASI , D. E., B OTERO , C. A., B OWERN , C., E MBER , C. R. et al. (2016). D-place: A global database of cultural, linguistic and environmental diversity. PLoS One, 11 (7), e0158391. K OROTAYEV, A. (2000). Parallel-cousin (fbd) marriage, islamization, and arabization. Ethnology, 39 (4), 395–407. —, K AZANKOV, A., B ORINSKAYA , S., K HALTOURINA , D. and B ONDARENKO , D. (2004). Ethnographic atlas xxx: peoples of siberia. Ethnology, 43 (1), 83–92. K OROTAYEV, A. V. (2003). Unilineal descent organization and deep christianization: A cross-cultural comparison. Cross-Cultural Research, 37 (1), 133–157. K UGLER , M., V ERDIER , T. and Z ENOU , Y. (2005). Organized crime, corruption and punishment. Journal of Public Economics, 89 (9), 1639–1663. L A P ORTA , R., L OPEZ - DE S ILANES , F., S HLEIFER , A. and V ISHNY, R. (1999). The quality of government. Journal of Law, Economics, and organization, 15 (1), 222–279. L AMBSDORFF , J. G. (2003). How corruption affects persistent capital flows. Economics of Governance, 4 (3), 229–243. — (2012). Behavioural and experimental economics as a guidance to anticorruption. In D. Serra and L. Wantchekon (eds.), New advances in experimental research on corruption, vol. 15, Emerald Group Publishing, pp. 279–300. L AZARIDIS , I., N ADEL , D., R OLLEFSON , G., M ERRETT, D. C., R OHLAND , N., M ALLICK , S., F ERNANDES , D., N OVAK , M., G AMARRA , B., S IRAK , K., C ONNELL , S., S TEWARDSON , K., H ARNEY, E., F U , Q., G ONZALEZ F ORTES , G., J ONES , E. R., R OODENBERG , S. A., L ENGYEL , G., B OCQUENTIN , F., G ASPARIAN , B., M ONGE , J. M., G REGG , M., E SHED , V., M IZRAHI , A.-S., M EIKLEJOHN , C., G ERRITSEN , F., B EJENARU , L., B L ÜHER , M., C AMPBELL , A., C AVALLERI , G., C OMAS , D., F ROGUEL , P., G ILBERT, E., K ERR , S. M., K OVACS , P., K RAUSE , J., M C G ETTIGAN , D., M ERRIGAN , M., M ERRIWETHER , D. A., O’R EILLY, S., R ICHARDS , M. B., S EMINO , O., S HAMOON -P OUR , M., S TEFANESCU , G., S TUMVOLL , M., T ÖNJES , A., T ORRONI , A., W ILSON , J. F., Y ENGO , L., H OVHANNISYAN , N. A., PATTERSON , N., P INHASI , R. and R EICH , D. (2016). Genomic insights into the origin of farming in the ancient near east. Nature, 536 (7617), 419–424. L EWIS , A. (2014). Security, Clans and Tribes: Unstable Governance in Somaliland, Yemen and the Gulf of Aden. Palgrave Macmillan: London. 44 L IPSET, S. M. and L ENZ , G. S. (1999). Corruption, culture, and markets, george Mason University. M ADSEN , E. A., T UNNEY, R. J., F IELDMAN , G., P LOTKIN , H. C., D UNBAR , R. I., R ICHARDSON , J.-M. and M C FAR LAND , D. (2007). Kinship and altruism: A cross-cultural experimental study. British Journal of Psychology, 98 (2), 339–359. M AURO , P. (1995). Corruption and growth. The Quarterly Journal of Economics, pp. 681–712. — (1998). Corruption and the composition of government expenditure. Journal of Public economics, 69 (2), 263–279. M ITTERAUER , M. (2010). Why Europe?: The Medieval Origins of Its Special Path. University of Chicago Press. M ORGAN , L. H. (1871). Systems of consanguinity and affinity of the human family, vol. 218. Smithsonian institution. M OSCONA , J., N UNN , N. and R OBINSON , J. A. (2017a). Keeping It in the Family: Lineage Organization and the Scope of Trust in Sub-Saharan Africa. Tech. rep., National Bureau of Economic Research. —, — and — (2017b). Social structure and conflict: Evidence from sub-saharan africa. M URDOCK , G. P. (1962–1971). Ethnographic atlas, installments i-xxvii. Ethnology, pp. 1–10. — (1967). Ethnographic Atlas. University of Pittsburgh Press. N OWAK , M. A., TARNITA , C. E. and W ILSON , E. O. (2010). The evolution of eusociality. Nature, 466 (7310), 1057– 1062. N UNN , N. and P UGA , D. (2012). Ruggedness: The blessing of bad geography in africa. Review of Economics and Statistics, 94 (1), 20–36. O KASHA , S. (2013). Biological altruism. In E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy, fall 2013 edn., Standford University Press. P EMBERTON , T. J., D E G IORGIO , M. and R OSENBERG , N. A. (2013). Population structure in a comprehensive genomic data set on human microsatellite variation. G3: Genes— Genomes— Genetics, pp. g3–113. — and R OSENBERG , N. A. (2014). Population-genetic influences on genomic estiamtes of the inbreeding coefficient: A global perspective. Human Heredity, 77 (1-4), 37–48. P LATTEAU , J.-P. (2000). Institutions, social norms, and economic development, vol. 1. Psychology Press. P UTNAM , R. D., L EONARDI , R. and N ANETTI , R. Y. (1994). Making Democracy Work: Civic Traditions in Modern Italy. Princeton University Press: Princeton. R AMACHANDRAN , S., D ESHPANDE , O., R OSEMAN , C. C., R OSENBERG , N. A., F ELDMAN , M. W. and C AVALLI S FORZA , L. L. (2005). Support from the relationship of genetic and geographic distance in human populations for a serial founder effect originating in africa. Proceedings of the National Academy of Sciences of the United States of America, 102 (44), 15942–15947. R IVAS , M. F. (2013). An experiment on corruption and gender. Bulletin of Economic Research, 65 (1), 10–42. S AILER , S. (2004). Cousin marriage conundrum. In T. Folger and S. Pinker (eds.), The Best American Science and Nature Writing 2004, Mariner Books. S CHULZ , J. F. (2016). The churches’ bans on consanguineous marriages, kin-networks and democracy. S ERRA , D. (2006). Empirical determinants of corruption: A sensitivity analysis. Public Choice, 126 (1-2), 225–256. 45 S IEGFRIED , T. (2006). A Beautiful Math:: John Nash, Game Theory, and the Modern Quest for a Code of Nature. National Academies Press. TABELLINI , G. (2010). Culture and institutions: Economic development in the regions of europe. Journal of the European Economic Association, 8 (4), 677–716. — et al. (2008). The scope of cooperation: Values and incentives. The Quarterly Journal of Economics, 123 (3), 905–950. TANZI , V. and D AVOODI , H. R. (1998). Roads to nowhere: how corruption in public investment hurts growth, vol. 12. International Monetary Fund. T ODD , E. (1990). L’invention de l’Europe [The invention of Europe]. Seuil. — and G ARRIOCH , D. (1985). The explanation of ideology: Family structures and social systems. Blackwell Oxford. T REISMAN , D. (2000). The causes of corruption: a cross-national study. Journal of public economics, 76 (3), 399–457. T YLOR , E. B. (1889). On a method of investigating the development of institutions; applied to laws of marriage and descent. The Journal of the Anthropological Institute of Great Britain and Ireland, 18, 245–272. U SLANER , E. M. (2005). Trust and corruption. In J. G. Lambsdorff, M. Taube and M. Schramm (eds.), The New Institutional Economics of Corruption, Routledge, pp. 76–93. WALKER , R. S. and B AILEY, D. H. (2014). Marrying kin in small-scale societies. American Journal of Human Biology, 26 (3), 384–388. W EBER , M. (1958). The protestant ethic and the spirit of capitalism, trans. Talcott Parsons (New York: Charles Scribner’s Sons, 1958), 182. W OODLEY, M. A. and B ELL , E. (2013). Consanguinity as a major predictor of levels of democracy: A study of 70 nations. Journal of Cross-Cultural Psychology, 44 (2), 263–280. W ORLD B ANK , W. (1997). Helping countries combat corruption: the role of the world bank. YAMAGISHI , T., C OOK , K. S. and WATABE , M. (1998). Uncertainty, trust, and commitment formation in the united states and japan 1. American Journal of Sociology, 104 (1), AJSv104p165–194. — and YAMAGISHI , M. (1994). Trust and commitment in the united states and japan. Motivation and emotion, 18 (2), 129–166. 46 Appendices to Akbari, Bahrami-Rad and Kimbrough (2017) A Quantitative description of genetic relatedness and distance First let us explain some vocabulary. A “locus” is a place on a chromosome where an “allele” resides. A locus is not a tangible object, it is a map describing where to find an allele, which is the piece of DNA in that location. Some books use gene as a synonym for an allele. An individual has two alleles at a particular locus, one from the mother and the other from the father. Alleles are identical by descent if they share a common ancestor allele in a relatively short time in the past, say, the past 10 generations (Gillespie, 2004, p. 6-8). Identity by descent is used as the basis of a quantitative description of relatedness. One simple measure is the “coefficient of kinship notated as f xy which is the probability that two alleles, one from individual X and one from individual Y, are identical by descent. This coefficient can be written as f xy = 1 1 p1 + p2 4 2 where p1 and p2 are, respectively, the probabilities of sharing one and two identical by descent alleles (p0 is the probability of sharing 0 identical by descent alleles). The intuition is that there are two mutually exclusive ways that two chosen alleles from X and Y might be identical by descent; they share exactly one allele or exactly two alleles. The term 14 p1 in the equation above is the probability that X and Y share exactly one identical by descent allele times the conditional probability that these two alleles are chosen from X and Y. The term 12 p2 is the probability that X and Y share exactly two identical by descent allele times the conditional probability that two identical alleles are chosen from X and Y. Relationship Identical twin (IT) Parent-offspring (PO) Full sibs (FS) First cousins (FC) p0 0 0 1/4 3/4 p1 0 1 1/2 1/4 p2 1 0 1/4 0 Table A1: Identity by descent. It follows that f PO = 1/4, f FS = 1/4 and f FC = 1/16. The “coefficient of relatedness”, r, is one-half of the mean number of shared alleles (p1 + 2p2 ), r = 21 r1 + r2 therefore, f xy = 21 r and r PO = 1/2, r FS = 1/2 and r FC = 1/8 (Gillespie, 2004, p. 121-123). It follows that f xy is defined over the the range of [0, 0.5]. It is important that the coefficient of kinship not be confused with the coefficient of relatedness, r. In a random mating population, the relationship between the two coefficients is simple: the coefficient of relatedness is just twice the coefficient of kinship, therefore in the range of [0, 1]. The coefficient of relatedness could be interpreted as the expected fraction of alleles that are shared identical by descent between two individuals (Harpending, 2002). Table A2 summarizes coefficients of relatedness for kin relationships. The expected coefficients of relatedness computed from Table A1 are valid only under the assumption of random mating. With consanguineous marriages, actual relatedness will exceed expected relatedness. For example, two offspring from a first-cousin marriage (drawn from a previously randomly mating population) have a relatedness higher than 1/2 (r = 1/2 + 1/2 × 1/8 = 0.5625): with probability 1/2 they inherit a gene from the same parent at each locus, and with probability 1/2 they each inherent a gene from a different parent, in which case the probability of gene sharing is just the relatedness between their parents, 1/8. Individuals born of consanguineous union have segments of their genomes that are homozygous as a result of inheriting identical-by-descent genomic segments through both parents. The extra term in the relatedness of offspring of a first-cousin marriage (i.e. 1/2 × 1/8) represents the expected 1 Relationship to you identical twin fraternal twin, parent, child, sibling grandparent, grandchild, aunt, uncle, niece, nephew great-grandparent, great-grandchild, great-aunt, great-uncle, great-niece, great-nephew, first-cousin second-cousin nth cousin a perfect stranger Relatedness coefficient 1 1/2 1/4 1/8 1/32 1/22n+1 0 Table A2: Expected relatedness of individuals under random mating. excess homozygosity in the genome of the offspring of a union between first cousins. The extra term is (1/2 × 1/32) for offspring of second-cousins; (1/2 × 1/4) for offspring of double-first cousins; and (1/2 × 1/2) for the offspring of sibling or parent-offspring (incestuous) unions. Therefore, “offspring of second cousins are expected to have children with 1/64 of their genome homozygous; offspring of first cousins, 1/16; offspring of double-first cousins, 1/8; and offspring of incestuous union, 1/4” (Woods et al., 2006, p. 889). Hamilton (1975) explores the consequences of extreme inbreeding, developing a model of endogamous colonies where the relatedness of all colony members can rise to the level of siblings under random mating (1/2). In such a world, “siblings, parents, and offspring will still be the individual’s closest relatives. Owing to inbreeding, their relatedness will be above the value of 1/2 that applies under random mating. Thus an individual should be more altruistic than usual to his immediate kin. But other neighbors who are not immediate kin are now also closely related, and it is this reduced contrast between neighbors and close kin that will give what is probably the most striking effect: we expect less nepotistic discrimination and more genuine communism of behavior. At the boundary of the local group, however, there is a sharp drop in relatedness, [. . . ] this drop may be such as to promote active hostility between neighboring groups” (Hamilton, 1975, p. 340). Empirically, persistent inbreeding has raised the relatedness of siblings as high as 7/10 in some samples (compared to 1/2 in a randomly mating population), and the average sibling relatedness among three persistently inbreeding groups was 6/10 (Woods et al., 2006). Note that cousins in such a society share almost 75% more genes than expected in a randomly mating population; their relatedness is (0.6 × 0.6 × 0.6 = 0.216), compared to (0.5 × 0.5 × 0.5 = 0.125) in a randomly mating population. The expected percentage of excess homozygosity arising from consanguineous mating is known as the inbreeding coefficient of the individual with respect to the local sub-population, or FIS . For example, for the offspring of first cousins, FIS is 1/16. Inbreeding coefficient (FIS ) of an individual is equal to kinship coefficient of the individual’s parents; as seen before, coefficient of kinship of the first cousin parents is also f FC = 1/16. In fact, inbreeding coefficient FIS measures homozygosity in individual genomes excess to the expected frequency under random mating in the sub-population. Therefore, inbreeding coefficient Hobs,i FIS can be calculated as 1 − Hexp,i , where Hobs,i is the observed heterozygosity of an individual, and Hexp,i is expected heterozygosity within the sub-population which is equal to Hobs,p ; the average of observed heterozygosity of individuals within the sub-population. Consanguineous mating is not the only way of producing excess homozygosity. If a sub-population is genetically isolated and thus “cryptically” inbred with respect to the total population, there will be excess average homozygosity in individual genomes of the sub-population compared to the expected frequency 2 under random mating across the total population. This expected increase in homozygosity arising from genetic isolation of the sub-population is called FST ; the average inbreeding coefficient of sub-populations relative to the total population. Pairwise FST measures population differentiation, producing higher values when two populations have large between-population differences but small within-population differences. InH breeding coefficient FST can be calculated as 1 − Hobs,p , where Hobs,p is the mean of observed heterozygosity exp within the sub-population, and Hexp is expected heterozygosity in the total population. Thus, if the alleles in an individual are identical by descent in excess of the expected frequency in the total population, and due to population inbreeding, further inbreeding by consanguineous marriages can not increase homozygosity in those alleles. The “inbreeding coefficient” or Fhbd captures both FIS and FST , and measures the inbreeding of an Individual relative to the total population. The inbreeding coefficient H , where Hobs,i is the observed heterozygosity of an individual, and Hexp Fhbd can be calculated as 1 − Hobs,i exp is expected heterozygosity within the sub-population. This is the coefficient reported in footnote 14. A.1 Ethnicity and relatedness Genomic methods allow us to measure relatedness between and within populations. The genetic distance of two populations can be measured by FST known as the coancestor coefficient: “the probability that two alleles at a given locus selected at random from two populations will be different, [. . . ] FST is strongly related to how long two populations have been isolated from each other. When two populations split apart, their genes can start to change as a result either of random genetic drift or natural selection” (Spolaore and Wacziarg, 2009, p. 481). As shown by Harpending (2002), for large populations, genetic distance between two populations implies genetic similarity within those populations. Therefore, FST also measures the coefficient of kinship between members of the same population; for a random mating population, FST is simply half of the coefficient of relatedness, r.1 Between some ethnic groups, empirical estimates suggest that relatedness is not far above zero, so that co-ethnics are unlikely to be sufficiently related for kin selection to substantially influence behavior. For example according to Cavalli-Sforza et al. (1994), the genetic distance between English and French populations is FST = 0.0024. Therefore, in a world consisting of only English people, the kinship of any randomly chosen pair is zero, but in a world consisting of both English and French populations, two random English (or French) people have a relatedness of only r = 0.0048 (in between the relatedness of 3rd and 4th cousins under random mating). Perhaps surprisingly, this is about how closely groups of friends are related to one another. Christakis and Fowler (2014) find that friends’ genotypes tend to be positively correlated, and the increase in similarity relative to strangers is at the level of fourth cousins. The authors were aware that some of the similarity in genotypes can be explained by “a simple preference for ethnically similar others” or “distant relatives” (Ibid., p. 10797). Therefore, they applied strict controls for such factors in their study. This suggests that there might be “some sort of kin detection system in humans [. . . ] such that, for each individual encountered, an unspecified system may compute and update a continuous measure of kinship that corresponds to the genetic relatedness of the self to the other individual” (Ibid., p. 10800). 1 An individual’s coefficient of kinship with someone randomly chosen from his own population is FST while his kinship with someone from the other population is − FST . “Negative relatedness implies that two individuals share fewer genes than average” (Gardner and Stuart, 2006, p. R663). 3 B Religion and Consanguinity As noted in the text and in Table 3, religious traditions have diverse relationships to cousin marriage practices. Below we summarize Christian and Islamic views on consanguinity. Catholic and Orthodox Christianity. In the early Christian era, consanguineous marriages were common,2 and there are few direct biblical prohibitions on marriage among close kin.3 The laws and customs of the Roman empire with respect to domestic life, “conformed to patterns that were wide-spread throughout Mediterranean and Middle East” that permitted and even encouraged consanguinity. Thus the Christian religion emerged in a setting in which such practices were tolerated. In other regions to which Christianity later spread such as Greece and Egypt, the earlier presence of close marriage was yet more marked (Goody, 1983, p. 39). However, in the centuries after the conversion of the Roman Empire to Christianity, radical changes occurred with respect to the issue of marriage to kin. For the first two-hundred years after Constantine, the legality of consanguinity was in flux.4 Roman Catholic authority stepped in to settle the issue. Pope Gregory I (AD 540-604) in a letter to Augustine, ‘the first Bishop and apostle of the English’ confirms that a certain secular law in the Roman state allows cousin marriage and advises that ‘it is necessary that the faithful should only marry relations three or four times removed, while those twice removed must not marry in any case’ (Goody, 1983, p. 36). This letter was significant because through it, the Church asserted (implicitly) its jurisdiction over marriage and the family, a position that Christian churches still maintain today. Biological relationship First cousin Second cousin Third cousin Genetic relationship Third degree Fifth degree Seventh degree Roman classification Fourth degree Sixth degree Eighth degree Germanic classification Second degree Third degree Fourth degree Table B1: Genetic and religious classification of consanguinity (Bittles 2012, p. 16). Within the Roman Catholic Church, the stringency of the restrictions continued to increase over time. 2 On one hand, it has been suggested that with the aim of favoring outbreeding, pre-Christian Roman law forbade all unions among people within the seventh degree of consanguinity, and the Church “initially followed the Roman” regulations on consanguineous marriages (Cavalli-Sforza et al., 2004, p. 29,31). On the other hand, some authors argue that there is not enough evidence suggesting prohibition of consanguineous unions at the very earliest stage in Roman history (Goody, 1983, p. 53). However, it seems that in the early Christian era, consanguineous marriages were common either due to “relaxed” (Cavalli-Sforza et al., 2004, p. 29) earlier prohibitions or because Roman “law had nothing to say against most forms of close marriage” (Goody, 1983, p. 39). For example, the first Christian emperor, Constantine the Great married his son and daughters to his half-brother’s children (Goody, 1983, p. 53). 3 The only types of forbidden relationships in the Bible are found in the Levitical prohibitions in The Old Testament. For example, a man could not marry his mother (Lev. 18:7), his sister (Lev. 18:12, 20:19), his father’s sister (Lev. 18:12, 20:19), his mother’s sister (Lev. 18:13, 20:19), or his son’s daughter (Lev. 18:10). A man could marry his niece (Cavalli-Sforza et al., 2004, p. 29). 4 Following the acceptance of Christianity as the official religion of the Eastern Roman Empire, Theodosius the Great (AD 378-395) condemned unions between first cousins in a law made in AD 384, although it was still possible to effect such marriages by imperial dispensation. In AD 405, his son, Arcadious (AD 378-408) legalized cousin marriages once again for the Eastern Roman Empire, however, his younger brother and the Western Roman emperor, Honorious (384-423), who himself married his dead wife’s sister, permitted marriages among cousins in AD 409 only if the parties obtained an imperial dispensation. It was not until later that under civil legislation they became freely permissible once again. In AD 533, the validity of cousin marriage in the secular law of the Eastern Roman Empire, the institutes of Justinian (482-565), was recognized as perfectly legitimate (Goody, 1983, p. 55; Cavalli-Sforza et al., 2004, p. 29-30; and Bittles, 2012, p. 16). 4 A canon attributed to various popes and embodied in a letter of Pope Gregory III (AD 731-741) forbade marriage to the seventh degree of consanguinity in AD 732 (i.e. to 3rd cousins, see Table B1), and confusion resulting from differences in the Germanic and Roman systems of consanguinity evaluation eventually led to the ban being extended all the way to 6th cousins in 1076 (7th degree in the Germanic system). Not only were these extended prohibitions attached to blood ties, but they were also assigned to affinal kinship such as marriage to the dead brother’s widow, spiritual kinship such as marriage to godchildren and fictional kinship such as adoption, “producing a vast range of people, often resident in the same locality, that were forbidden to marry” (Goody, 1983, p. 56). While the bans were later loosened and dispensations have typically been available in specific cases, prohibitions on first cousin marriage remain in effect today.5 Similarly, within the Greek Orthodox Church, first-cousin marriages were prohibited by AD 692, a policy which remains in place. It is worthwhile to briefly discuss the possible reasons behind Church prohibitions of consanguineous marriages. From Gregory’s letter to Augustine, it is clear that he understood the potential negative health consequences of inbreeding - ‘the offspring of such marriages cannot thrive’ - and that he had a moral opposition to incest, derived from his reading of scripture (Goody, 1983, p. 36-37). More importantly for our purposes, it seems that the potential impact on social behavior had not gone unnoticed by the Church. St. Augustine of Hippo, writing in the early 5th century, noted that restrictions on consanguinity expand the scope for mutually beneficial cooperation: “for affection is now given its proper place, so that men, for whom it is beneficial to live together in honorable concord, may be joined to one another by the bonds of diverse relationships: not that one man should combine many relationships in his sole person, but that those relationships should be distributed among individuals, and should thereby bind social life more effectively by involving a greater number of persons in them” (Augustine, 1998, Book 15, Ch. 16, italics added). Similar arguments were later echoed by Thomas Aquinas, who worried that “consanguineous marriages would ’prevent people widening their circle of friends’” (Goody, 1983, p. 57).6 This suggests an awareness of the sorts of favoritism that might generate corruption. It seems that as the Church extended prohibitions to remoter degrees of consanguineous marriage, large kinship groups such as clans, lineages, and tribes gradually disappeared throughout Europe. There is a large and significant negative correlation between the spread of Christianity (for at least 500 years) and the absence of clans and lineage groups in Europe (Greif, 2006, p. 309) which may have facilitated the emergence of impartial economic and political institutions, nuclear families and individualistic culture. Protestant and Anglican Christianity. As the Protestant Reformation emphasized a return to scripture, Martin Luther’s (1483-1546) views on consanguinity were based on the Levitical prohibitions and did not include a ban on cousin marriages; similarly, John Calvin (1509-1564) and his followers based their views on scripture, though they extended Levitical guidelines to a wife’s relatives so that affinity and consanguinity were treated equivalently. The Church of England grew directly out of a dispute between Henry VIII (1491-1547) and the Catholic Church about marriage law, and its creation led to changes in consanguinity law that allowed for first cousin marriage (Goody, 1983, p. 168-177). 5 The prohibition on consanguineous marriages was reduced to 4th degree relationships (third cousin) in AD 1215. The ban also reduced to 2nd degree relationships (first cousin or closer) for South American Amerindians in 1537, later for the indigenous population of The Philippines, and for black populations in 1897. After 1917, the consanguinity prohibition was reduced in all populations, initially to second cousins or closer and in 1983 to first cousins or closer, which remains current law (Cavalli-Sforza et al., 2004; Bittles, 2012; Goody, 1983). 6 Goody (1983) also suggests a number of other possible, complementary reasons for the bans, including attempts to increase fealty to the Church by diminishing the role of the family (rooted in Jesus’ entreaties to deny the family, e.g. Matthew 10: 34-37) and Church desire to acquire property by leaving the deceased without eligible heirs. The latter argument finds support in the tendency of the Church to oppose any practice that expanded the number of kin, e.g. divorce, polygamy, and adoption. 5 In the emergent Protestant denominations, marriages up to and including first-cousin unions were permitted under Reformed Church law.7 However, “many Protestant reformers discouraged marriages within the third degree. While Luther did not think they were positively harmful, he considered them to be ‘inexpedient on the ground that people would marry without love merely to keep property within family, while poor women would be left spinsters’” (Goody, 1983, p. 181-2). In practice, consanguinity rates remained low in the Protestant world, as seen in Figure 2 in Section 3 below, perhaps due to ingrained norms from centuries of prohibition. Islam. Today, consanguinity rates are high in Islamic countries, though “contrary to widespread Western opinion, there is no specific guidance in the Qur’an to encourage consanguinity” (Bittles, 2012, p. 22) and “marriage between cousins is not prescribed by the Qur’an” (Courbage and Todd, 2014, p. 33). The permitted degrees of consanguinity within Islam closely match the Levitical guidelines with one exception: a prohibition of uncle-niece marriage (Qur’an 4:23). In addition to the Qur’an, Muslims recognize two other sources of Islamic Law (Sharia) which bear on consanguinity: the Hadith (oral pronouncements of the Prophet Mohammad) and the Sunnah (the deeds of the Prophet).8 “The overall attitude to consanguineous marriage within Islam is somewhat ambiguous” (Bittles, 2012, p. 22) since a Hadith of the Prophet Muhammad stated: “Do not marry cousins as the offspring may be disabled at birth” (Akrami and Osati, 2007, p. 314); while, according to the Sunnah, two of Muhammad’s wives were his first cousins, and the Prophet Muhammad married his daughter, Fatima to his cousin Ali.9 “Thus, [despite the content of the Hadith] for Muslims, the practice of cousin marriage could potentially be interpreted as following the example provided by the Sunnah” (Bittles, 2012, p. 22). In any case, evidence suggests that cousin marriage customs in Islamic countries “probably antedated the spread of Arabs” (Cavalli-Sforza et al., 2004, p. 284) and “predated Islam” (Courbage and Todd, 2014, p. 33). A preference for consanguinity is reflected in “the well known Iranian proverb ‘the first cousin’s marriage contract has been recorded in heaven,’” though the preference “is merely a cultural and local custom rather than a religious belief” (Akrami and Osati, 2007, p. 315). Nevertheless, cousin marriage is not forbidden or clearly discouraged in Islam, and it is likely that “Islam served as a vehicle for the geographical spread of the endogamous practices [. . . ] the model was adopted not for religious reasons but because it was the practice of a prestigious group, the Arabs, bearers of the message of the Qur’an” (Courbage and Todd, 2014, p. 33-4). One explanation suggesting why the conversion to Islam may have encouraged the high prevalence of cousin marriages in Islamic countries is the Quranic law on the inheritance of property which entitled daughters to inherit half of the amount received by sons, and wives to receive a share from their husbands. A dowery (Mahr) also is specified in Islamic law as part of the marriage arrangement. Under these circumstances, consanguineous marriages could prevent part of the family wealth from leaving the clan (Goody, 1983, p. 32). Other religions. In Table 3, we summarize other religions’ views on consanguineous marriage based on Bittles (2012), p. 23-28. Suffice to say that there has historically been a great diversity of practices around the world, with many groups banning cousin marriage outright and certain groups favoring even closer marriages (e.g. Dravidian Hindu and Sephardic Jewish preference for uncle-niece marriages). 7 “An exception to this generalization is provided by the State Lutheran Church of Sweden, which until 1680 refused to recognize first-cousin unions, and from 1680 to 1844 approval was required from the King of Sweden before a first-cousin union could proceed” (Bittles, 2012, p. 20). 8 In Shi’a Islam, law also derives from the oral pronouncements and deeds of the twelve Imams. 9 The first Imam, in the Shi’a tradition, and the fourth of the Rashidun Caliphate according to Sunni tradition. 6 C Empirical data description Cross country regressions: Variable name Consanguinity Corruption Description and source As a working definition, unions contracted between persons biologically related closer than second cousins are categorized as consanguineous. Bittles and Black (2015) provided a compilation of the proportion of consanguineous marriages from 262 journal papers and book chapters which report consanguinity percentages of 448 samples of different sizes in different locations of 72 countries based on household surveys, Roman Catholic Church dispensations, parish records, civil registrations, marriage registrations and surveys on blood donors, obstetric inpatients, hospital outpatients, hospital births, etc that include information for around 9.8 million marriages. The sample of countries is non-random since the data were collected for other purposes (e.g. the public health studies disproportionately sample high consanguinity societies since these are the societies with higher rates of genetic disorders), but the coverage is broad. Data collection periods vary from 1922 to 2013, with around 90% after 1950, 75% after 1960, 50% after 1970, 40% after 1980, 30% after 1990 and 15% after 2000. We collected this data from www.consang.net in December 2015 and computed the mean percentage of consanguineous marriages for each country by weighting the individual estimates according to sample size. Note that (i) the data on Czechoslovakia in the period 1961-64 is used for both the Czech Republic and Slovakia; (ii) some sample studies were reported twice, once at the city level and once for the province, in which case we considered only the province level report; (iii) we also ignored the studies listed as “Minorities and Isolates” to avoid overweighting outliers. Source: Bittles and Black (2015). Measures financial corruption in the form of demands for special payments and bribes connected with import and export licenses, exchange controls, tax assessments, police protection, or loans, and actual or potential corruption in the form of excessive patronage, nepotism, job reservations, ‘favor-for-favors’, secret party funding, and suspiciously close ties between politics and business. In our tables, this measure is averaged over 1984-2011. La Porta et al. (1999) and Alesina et al. (2003) used the average of the months of April and October in the monthly index between 1982 and 1995, and we did so in the replication of their tables. Source: Political Risk Services, International Country Risk Guide (2012). 7 Ethnic fractionalization Ethonolinguistic fractionalization Family ties Measures ethnic fractionalization; the probability that two randomly selected individuals from a population belong to different ethnic groups. Source: Alesina et al. (2003). Average value of five different indices of ethonolinguistic fractionalization. Its value ranges from 0 to 1. The five component indices are: (1) index of ethonolinguistic fractionalization 1960, which measures the probability that two randomly selected people from a given country will not belong to the same ethnolinguistic group (the index is based on the number and size of population groups as distinguished by their ethnic and linguistic status); (2) probability of two randomly selected individuals speaking different languages;( 3) probability of two randomly selected individuals do not speak the same language; (4) percent of the population not speaking the official language; and (5) percent of the population not speaking the most widely used language. The data is collected using Easterly and Levine (1997). The sources of the components of the average index are (1) Bruk and Apenchenko (1964); (2) Muller (1964); (3) Roberts (1962); (4) and (5) Gunnemark (1991). Source: La Porta et al. (1999). Measures the strength of family ties by looking at three variables from the World Value Survey (WVS) and the European Social Survey (EVS) which capture beliefs regarding the importance of the family in the respondent’s life, the duties and responsibilities of parents and children, and the love and respect for one’s own parents. The first question asks how important the family is in one person’s life and can be answered with (i) Very important; (ii) Rather important; (iii) Not very important; (iv) Not at all important, which in our measure of family ties, take values of 4 to 1, respectively. The second question asks whether the respondent agrees with one of the two statements (taking the values of 2 and 1, respectively): (i) Regardless of what the qualities and faults of one’s parents are, one must always love and respect them; (ii) One does not have the duty to respect and love parents who have not earned it. The third question prompts respondents to agree with one of the following statements (again taking the values of 2 and 1, respectively): (i) It is the parents’ duty to do their best for their children even at the expense of their own well-being; (ii) Parents have a life of their own and should not be asked to sacrifice their own well-being for the sake of their children. Following Alesina and Giuliano (2010), we extract the first principal component from the whole data set with all individual responses for the original variables. Source: WVS (Six waves, 1981-2014) and EVS (four waves, 1981-2008). 8 Trust Cousin term measure Predicted genetic diversity Measures generalized trust by considering the following question from the World Value Survey (WVS) and the European Social Survey (EVS): “Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?” The answer could be either “Most people can be trusted” or “Can’t be too careful”, which in our measure of trust, take values of 2 and 1, respectively. Source: WVS (Six waves, 1981-2014) and EVS (four waves, 1981-2008). The fraction of the population of a country that speaks a language in which there are words used for distinguishing (at least some) first cousins from each other. Based variable 27 of the Ethnographic Atlas, a dummy variable is created which takes value 1 if the language of the ethno-linguistic group in the Ethnographic Atlas fully or partially distinguishes first cousins from each other (Sudanese, Iroquois, Omaha, Crow), and takes value 0 otherwise; for ethnicities which the language does not distinguish any first cousins (Eskimo or Hawaiian). Following Alesina et al. (2013)’s methodology, the country-level cousin term measure is created by population weighted average of the dummy variable for all ethnic groups living within a country. The expected heterozygosity (genetic diversity) of a given country as predicted by (the extended sample definition of) migratory distance from East Africa (i.e., Addis Ababa, Ethiopia). This measure is calculated by applying the regression coefficients obtained from regressing expected heterozygosity on migratory distance at the ethnic group level, using the worldwide sample of 53 ethnic groups from the HGDP-CEPH Human Genome Diversity Cell Line Panel. The expected heterozygosities and geographical coordinates of the ethnic groups are from Ramachandran et al. (2005). Expected heterozygosities are constructed by measuring actual heterozygosity within an ethnic group at a sample of selectively-neutral loci and averaging over the loci. Source: Ashraf and Galor (2013). 9 Geographical controls (i) Ruggedness Terrain ruggedness measures small-scale terrain irregularities. The ruggedness calculation takes a point on the earth’s surface and calculates the difference in elevation between this point and each of the points on the grid 30 arcseconds (926 meters on a meridian) in each of the eight major directions of the compass (north, northeast, east, southeast, south, southwest, west, and northwest). The terrain ruggedness index at the central point is given by the square root of the sum of the squared differences in elevation between the central point and the eight adjacent points. Then by averaging across all grid cells in the country not covered by water, each cell weighed by its latitude-varying sealevel surface, the average terrain ruggedness of the country’s land area is obtained. Ruggedness is measured in hundreds of meters of elevation difference for grid points 30 arc-seconds apart. Source: Nunn and Puga (2012). (ii) Soil suitability for agriculture The soil suitability component, based on soil carbon density and soil pH, of an index of land suitability for agriculture. The soil suitability data are reported at a half-degree resolution by Ramankutty et al. (2002a) and are aggregated to the country level by Michalopoulos (2012) by averaging across grid cells within a country. For additional details on the soil suitability component of the land suitability index, the interested reader is referred to the definition of the land suitability variable above. Source: Ashraf and Galor (2013). (iii) Mean elevation The mean elevation of a country in km above sea level, calculated using geospatial elevation data reported by the GECON project (Nordhaus, 2006) at a 1-degree resolution, which, in turn, is based on similar but more spatially disaggregated data at a 10-minute resolution from New et al. (2002). The measure is thus the average elevation across the grid cells within a country. The interested reader is referred to the G-ECON project website for additional details. Source: Ashraf and Galor (2013). 10 Latitude Log GNI per capita (iv) Mean temperature The intertemporal average monthly temperature of a country in degrees Celsius per month over the 1961-1990 time period, calculated using geospatial average monthly temperature data for this period reported by the G-ECON project (Nordhaus, 2006) at a 1-degree resolution, which, in turn, is based on similar but more spatially disaggregated data at a 10-minute resolution from New et al. (2002). The measure is thus the spatial mean of the intertemporal average monthly temperature across the grid cells within a country. See the G-ECON project website for additional details. Source: Ashraf and Galor (2013). (v) Mean precipitation The intertemporal average monthly precipitation of a country in mm per month over the 1961-1990 time period, calculated using geospatial average monthly precipitation data for this period reported by the G-ECON project (Nordhaus, 2006) at a 1-degree resolution, which, in turn, is based on similar but more spatially disaggregated data at a 10minute resolution from New et al. (2002). The measure is thus the spatial mean of the intertemporal average monthly precipitation across the grid cells within a country. The interested reader is referred to the G-ECON project web site for additional details. Source: Ashraf and Galor (2013). (vi) Percentage of population living in tropical/subtropical/temperate zones The percentage of a country’s population in 1995 that resided in areas classified as tropical/subtropical/temperate by the Köppen-Geiger climate classification system. This variable was originally constructed by Gallup et al. (1999) and is part of Harvard University’s CID Research Datasets on General Measures of Geography. Source: Ashraf and Galor (2013). (vii) Percentage of land near a waterway The percentage of a country’s total land area that is located within 100 km of an ice-free coastline or sea-navigable river. This variable was originally constructed by Gallup et al. (1999) and is part of Harvard University’s CID Research Datasets on General Measures of Geography. Source: Ashraf and Galor (2013). The absolute value of the latitude of the country, scaled to take values between 0 and 1. The data is collected from CIA (1996). Source: La Porta et al. (1999). Logarithm of GNI per capita in current US dollars averaged over the period 1984-2011. Source: World Bank Development Indicators (WDI), Data retrieved Online in December 2015. 11 Log GNP per capita Log population Log population (1960) Regional dummy variables Legal origin dummy variables Religion dummy variables Logarithm of GNP per capita in current U.S. dollars averaged over the period 1970-1995. The data is collected from WDI. Source: La Porta et al. (1999). Logarithm of population averaged over the period 19842011. Source: World Bank Development Indicators (WDI), Data retrieved Online in December 2015. Logarithm of population in 1960. This is the variable used in Alesina et al. (2003) for the country size. Our data source might be different. Source: Penn World Table, Data retrieved Online in December 2015. Dummy variable for (1) Sub-Saharan Africa, (2) East Asia Pacific, and (3) Latin America and Caribbean. Source: World Bank (http://www.worldbank.org/en/country). Identifies the legal origin of the 212 Company Law or Commercial Code of each country. There are five possible origins: (1) English Common Law; (2) French Commercial Code; (3) German Commercial Code; (4) Scandinavian Commercial Code; and (5) Socialist/Communist Laws. The data is collected using La Porta et al. (1998), American Association of Law Libraries and Flores (1989) and CIA (1996). Source: La Porta et al. (1999) Identifies the percentage of the population of each country that belonged to the three most widely spread religions in the world in 1980. For countries of recent formation the data is available for 1990-1995. The numbers are in percent (scale from 0 to 100). The three religions identified here are: (1) Roman Catholic; (2) Protestant and (3) Muslim. The residual is called ”other religions”. The data is collected using Barrett (1982), Worldmark (1995), Statistical Abstract of the World (1995), UN (1995), CIA (1996). Source: La Porta et al. (1999). Table C1: Description of the data for cross-country analysis 12 Within country (Italy) regressions: Variable name Consanguinity Corruption Share of agriculture Log value added per capita Log population Description and source The data reported by Cavalli-Sforza et al. (2004) on consanguineous marriages, for 5-year periods from 1910 to 1964, comes from the Vatican’s Secret Archives in which requests for dispensations from the consanguinity impediment, sent by the Bishops to the Sacred Congregation of the Sacraments in Rome, were recorded with information of the name of the diocese of marriage, the date and the degree of relationship between the spouses. They grouped 280 Italian dioceses into the provinces present in 1961, for which the number of total marriages was available from year to year. They report four major degrees of consanguinity; uncle-niece/aunt-nephew, first cousins, first cousins once-removed, second cousins. However, the bishops of the islands (Sardinia and Sicily) were granted the privilege to accord the dispensation for consanguineous unions beyond degree III, i.e. first cousins once-removed and second cousins were not recorded in the Vatican Archives, and therefore are not reported for Sicily and are obtained from another source for Sardinia. Thus, we have only considered uncle-niece/aunt-nephew and first cousin unions to compute consanguinity rates for Italian provinces. We have also chosen 5-year periods from 1945 to 1964 for which the data is available in all reported provinces. We computed the consanguinity rate for each province as the average of consanguinity percentages of four 5-year periods (1945-1949, 1950-1954, 1955-1959, 1960-1964) weighted by the number of marriages. For newly created provinces, we used the consanguinity rate of the province they belonged to in year 1961. This gave us consanguinity rates of 108 Italian provinces. Source: Cavalli-Sforza et al. (2004). Province-level number of associative crimes reported by the police to the court per 100,000 inhabitants averaged over 2000-2013. Associative crimes include criminal association (article 416: when three or more persons associate together in order to commit more than one crime) and Mafiatype association (article 416-bis: participating in a Mafia-type unlawful association including three or more persons). The data is available for 103 Italian provinces. Source: ISTAT. Province-level share of agriculture in total value-added averaged over 2000-2013. Source: ISTAT. Logarithm of province-level total value added (at current prices millions Euros) averaged over 2000-2013. Source: ISTAT. Logarithm of province-level population averaged over 2000-2013. Source: ISTAT. 13 Civic involvement Dominations Family types Voluntary organizations Family ties An integer index ranging from 1 to 9 which combines five variables observed in the late 19th century and early 20th century; (i) Membership in mutual aid societies (18731904); (ii) Membership in cooperatives (1889-1915); (iii) Strength of the mass parties (1919-1921); (iv) Turnout in the few relatively open elections (1919-1921) before Fascism brought authoritarian rule to Italy; (v) The longevity of local associations founded before 1860. Source: Putnam et al. (1994). A categorical variable that identifies, for each province, the administration that presided during the period of the Spanish domination in Italy, 1560-1659; Spanish, Papal, Austrian, Venetian, Sabaudian and Independent. Source: Di Liberto and Sideri (2015). A categorical variable that identifies, for each province, Todd (1990)’s classification of family types which is argued to be very similar to what the geography of family types would have been in the Middle Ages. The three family types common in Italy were the following; (i) incomplete stem family (characterized by an extended family with several generations living under one roof and the inheritance of the house and the land by one son – generally, the eldest – who stays at home); (ii) Communitarian family (characterized by an extended family in which all the sons can get married and bring their wives to the family home and equal division of inheritance among children); (iii) Egalitarian nuclear (characterized with total emancipation of children in adulthood to form independent families and equal division of inheritance among children). Source: Duranton et al. (2009). Region-level number of voluntary organizations established before 1965 per 100,000 inhabitants at year 2001. The variable takes the same value for all provinces within a region. Source: ISTAT. Region-level fraction of youth aged 18-34 living with at least one parent averaged over 2002-2009. The variable takes the same value for all provinces within a region. Source: ISTAT. 14 Number of active years of archdioceses Alternative corruption metrics Number of active years of Catholic archdioceses in Italian provinces. The raw data is for 42 Roman Catholic ecclesiastical provinces in Italy. Each ecclesiastical province is served by a metropolitan archdiocese. For each ecclesiastical province, we calculated the number of active years of its archdioceses from the date of establishment of the first archdiocese to the present, subtracting the number of years which there were no active archdiocese in the province (the archdiocese was suppressed). Some of archdioceses are branched from pre-existing archdioceses, in which case we considered the date of the establishment of the pre-existing archdioceses. Some archdioceses lost territory to other dioceses which were promoted as the new archdioceses, in which case we considered the archdiocese of the ecclesiastical province being active. We matched the data on ecclesiastical provinces to today’s administrative provinces. Source: www.gcatholic.org. (i) Corruption crimes (region level, N=20) Region-level number of corruption crimes convicted of felony by final judgment per 100,000 inhabitants averaged over 2000-2011. Corruption crimes include the following types of crimes defined under the title offenses against public administration: crimes of peculation, malversation, bribery, and corruption. Source: ISTAT. (ii) Infrastructure (province level, N=92, and region level, N=20) The difference between a measure of the value of existing physical quantities of public infrastructure and the cumulative price government has paid for public capital stocks. Where the difference is larger between the monies spent and the existing physical infrastructure, more money is being siphoned off to mismanagement, fraud, bribes, kickbacks, and embezzlement; that is, corruption is greater. The measure is created for Italy’s 92 provinces and 20 regions as of the mid-1990s, controlling at the regional level for possible differences in the costs of public construction. Inspecting the data, the province-level ratios reported in the appendix appear to be the inverse of the ratios reported in the text for regions, so we have taken the inverse to align the interpretation of the province and region-level measures. Source: Golden and Picci (2005). 15 Latitude Mean temperature (in Celsius degrees)/ Mean precipitation (in meters) Tropical climate (iii) European Quality of Governance Index (EQI 2010) (region level, N=20) This measure is constructed from recent surveys that elicit perceptions of and experience with governmental corruption, as well beliefs about impartiality and quality in the provision of government services. This research was funded by the EU Commission for Regional Development. Source: Charron et al. (2014), available here (https://nicholascharron.wordpress.com/european-qualityof-government-index-eqi/). (iv) Institutional performance (region level, N=20) This measure is an index constructed by combining measures of policy processes and internal operation of government (Cabinet Stability, Budget Promptness, Statistical and Information Services), policy content (Reform Legislation, Legislative Innovation), and policy implementation (Daycare Centers, Family Clinics, Industrial Policy Instruments, Agricultural Spending Capacity, Local Health Unit Expenditures, Housing and Urban Development, Bureaucratic Responsiveness) over the period 1978-1985. Source: Putnam et al. (1994), see Ch. 3. The absolute value of the latitude in the geographic center of each province. Geographic centroids of Italian provinces are generated using polygons map of Italian administrative divisions. Source: GADM database of Global Administrative Areas. The average of mean temperature/precipitation in the geographical area of each province. The means of the entire annual cycles of precipitation and temperature are constructed for the time period between 1901 and 2014 based on monthly global maps (0.5 by 0.5 degree cells), CRU-TS 3.23 Climate Database. Source: Harris et al. (2014). The indicator variable for “tropical climate” takes value 1 if the geographic center of a province is classified as being either tropical or subtropical, and zero otherwise. The data is constructed based on Thermal Climate Zones of the World, a global raster datalayer with a resolution of 5 arc-minutes, with each pixel containing a class value for the dominant thermal climate found in the pixel. Source: FAO’s Food Insecurity, Poverty and Environment Global GIS Database (FGGD). 16 Suitability for agriculture Distance to coast (in kilometers) Slope (in percent) Ruggedness (in 100 meters) Elevation (in 100 meters) The average of suitability for agriculture in the geographical area of each province. Suitability for agriculture represents the fraction of each grid cell that is suitable to be used for agriculture. It is based on the temperature and soil conditions of each grid cell. The data is construed based on the global map (0.5 by 0.5 degree cells) obtained from Suitability for Agriculture, Atlas of the Biosphere. Source: Ramankutty et al. (2002b). Distance of the geographic center of each province from the coast is constructed based on a coastline physical vector map in 1:10m resolution. Source: Natural Earth. The average of slope in the geographical area of each province. Slope measures the mean uphill slope percentage between each grid cell and its neighbours. The data is constructed based on the global map (30 by 30 arc-second cells) obtained from Grid-cell-level Data on Terrain Ruggedness. Source: Nunn and Puga (2012). The average of ruggedness in the geographical area of each province. Ruggedness measures the elevation distance of each grid cell and its neighbours. The data is constructed based on the global map (30 by 30 arc-second cells) obtained from Grid-cell-level Data on Terrain Ruggedness. Source: Nunn and Puga (2012). The average of elevation in the geographical area of each province. Elevation is constructed based on the global map (30 by 30 arc-second cells) obtained from Global 30 ArcSecond Elevation data set. Source: GTOPO30 data set. Table C2: Description of the data for within-country analysis (Italy) 17 D Replication and Extension of Alesina et al. (2003) Here we report the results of regression analysis using the data from Alesina et al. (2003) in concert with our data from Bittles and Black (2015). The analysis tested the robustness of the claim that ethnic fractionalization causes corruption. We first replicate their findings and then extend them by including our measure of consanguinity. First, we note that there is a sizable correlation between ethnic fractionalization and consanguinity (Spearman’s ρ = 0.50, p-value < 0.001, N = 72). Alesina et al. (2003) report find a significant correlation between their ethnic fractionalization index and corruption, even after controlling for legal origins, which we replicate in regression (1) of Table D1. In column (2), we replicate their analysis including only those countries for which we have consanguinity data. Including consanguinity in specifications (3) and (4) causes the effect of fractionalization to disappear while consanguinity is significant. Columns (5) and (6) show that the effect of consanguinity on corruption is robust to using the cousin term measure (our IV from Section 3.1) in reduced form and as an instrument for consanguinity. When we introduce additional control variables in Table D2 and include both measures in regression (3) or replace ethnic fractionalization with consanguinity in regression (4) of the table, consanguinity is still significant despite including additional control variables; the same is true when we use our IV in reduced form or as an instrument for consanguinity.10 (1) VARIABLES Ethnic fractionalization Alesina (2003) (2) Alesina (2003) restricted sample (3) With Consanguinity -2.498** (0.981) -2.777* (1.394) -0.183 (1.196) -10.38*** (1.579) Consanguinity (4) With Consanguinity Without EF(2003) Africa East Asia Latin America Constant 0.0764 (0.260) -0.902 (0.569) -1.650** (0.661) -2.127*** (0.496) 7.295*** (1.052) Observations 122 R-squared 0.278 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 (6) Cousin term as instrument -2.581** (1.027) -0.235 (1.433) -10.71*** (2.540) -10.49*** (1.274) Cousin term measure Log population (1960) (5) Cousin term reduced form 0.203 (0.340) 0.202 (1.061) -1.484 (1.355) -1.101** (0.523) 6.441*** (1.485) -0.259 (0.244) 1.475** (0.709) -2.249* (1.244) -2.888*** (0.578) 9.112*** (0.996) -0.253 (0.249) 1.423** (0.598) -2.280* (1.220) -2.921*** (0.523) 9.045*** (0.995) -1.175** (0.484) -0.0752 (0.252) -0.758 (0.525) -1.822** (0.702) -2.690*** (0.532) 8.499*** (1.048) 64 0.172 64 0.519 64 0.519 118 0.342 -0.297 (0.258) 1.547* (0.825) -2.268* (1.263) -2.962*** (0.703) 9.358*** (1.162) 61 0.508 Table D1: Model specification Table (13), column (2) from Alesina et al. (2003), including consanguinity. 10 Our replications of previous studies give slightly different results for Alesina et al. (2003) in regression (1) of Table D1 and Table D2, possibly because we used different data sources for population and regional dummies. 18 (1) VARIABLES Ethnic fractionalization Alesina (2003) (2) Alesina (2003) restricted sample (3) With Consanguinity -1.029 (0.760) -1.635 (0.982) -0.225 (0.963) -6.076*** (2.092) Consanguinity (4) With Consanguinity Without EF(2003) Log population (1960) Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant 1.028*** (0.176) 0.609*** (0.207) 1.224** (0.576) -0.375 (0.616) -0.672 (0.444) 1.065** (0.521) -0.0753 (0.373) 0.103 (0.638) 2.157*** (0.472) -3.978** (1.764) Observations 120 R-squared 0.564 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 (6) Cousin term as instrument -0.858 (0.852) 0.242 (1.159) -7.746** (3.420) -6.251*** (1.909) Cousin term measure Log GNP per capita (1970-95) (5) Cousin term reduced form 1.273*** (0.198) 0.796*** (0.287) 2.371** (0.988) -0.1000 (1.049) 0.304 (0.467) 0.942 (0.677) -0.331 (0.548) -1.008 (0.686) 1.482** (0.642) -6.617*** (2.412) 0.879*** (0.276) 0.347 (0.284) 2.344** (0.924) -0.983 (1.153) -1.230* (0.701) 0.546 (0.848) -0.181 (0.489) -0.0499 (0.609) 1.409*** (0.507) -1.137 (3.411) 0.868*** (0.274) 0.346 (0.281) 2.272*** (0.820) -1.040 (1.138) -1.285* (0.646) 0.578 (0.839) -0.175 (0.482) 0.0336 (0.513) 1.468*** (0.456) -1.092 (3.371) -0.884** (0.435) 0.918*** (0.191) 0.508** (0.212) 1.064* (0.575) -0.568 (0.685) -1.194** (0.507) 1.000* (0.585) 0.00392 (0.375) -0.0173 (0.616) 2.662*** (0.561) -2.397 (2.099) 62 0.623 62 0.695 62 0.695 116 0.583 0.777** (0.378) 0.240 (0.397) 2.345** (0.972) -1.218 (1.268) -1.627 (1.109) 0.478 (1.009) -0.132 (0.529) 0.246 (0.748) 1.437** (0.549) 0.194 (4.994) 59 0.682 Table D2: Model specification Table (13), column (3) from Alesina et al. (2003), including consanguinity. 19 E Additional analysis E.1 Additional cross-country analysis While in the paper we use the ICRG index as our measure of corruption, the basic pattern of correlation between corruption and consanguinity rates that we observe is similar using alternative measures of corruption. For instance, Transparency International’s 2013 Corruption Perception Index11 is highly correlated with consanguinity (Spearman’s ρ = −0.49, p-value < 0.001, N = 67) as is the measure “Prevalence of Rule Violations” from Gächter and Schulz (2016) (Spearman’s ρ = 0.59, p-value < 0.001, N = 66). In fact, consanguinity rates are also correlated with the pre-treatment number of parking tickets per UN diplomat in NYC, reported in Fisman and Miguel (2007) (Spearman’s ρ = 0.37, p-value = 0.002, N = 65). And despite the small sample size, consanguinity is highly correlated with the recent cross-country laboratory measure of cheating from Gächter and Schulz (2016) (Spearman’s ρ = 0.72, p-value = 0.003, N = 15). Full Tables Reported in Section 3.1 (1) Basic model VARIABLES (2) Basic model restricted sample (3) and Consanguinity (4) Income instead of Latitude (5) Latitude as instrument (6) both Income and Latitude (7) without Income and Latitude -0.222 (0.670) -0.132 (0.167) 3.586*** (1.255) -4.463*** (0.769) 0.524 (0.514) -0.175 (0.129) 1.898** (0.878) -3.664*** (0.905) 0.0742 (0.460) 0.105 (0.125) -2.513* (1.331) -0.123 (0.417) 0.297 (0.235) -5.087*** (0.772) 0.319 (0.555) -0.132 (0.127) 1.288** (0.567) 1.209** (0.586) -0.275 (0.408) -0.365 (0.437) -0.470 (0.361) -0.0114 (0.247) -0.619 (0.478) 0.984*** (0.269) -3.269 (3.904) -3.566*** (0.884) 0.226 (0.479) 0.0423 (0.130) 1.026 (0.847) 0.592** (0.269) 0.766* (0.451) -0.364 (0.416) -0.604* (0.317) -0.871** (0.347) -0.135 (0.225) -0.0436 (0.295) 0.878*** (0.313) 1.005 (1.789) 67 0.654 67 0.704 67 0.632 Consanguinity Ethnic fractionalization Log population Latitude -0.070 (0.470) -0.106 (0.105) 3.001*** (0.861) Log GNI per capita Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant -0.272 (0.266) -0.022 (0.310) -0.184 (0.254) -1.347*** (0.277) -0.250 (0.180) 0.571 (0.355) 1.123*** (0.330) 3.239*** (0.856) Observations 134 R-squared 0.527 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.068 (0.487) 0.340 (0.460) 0.451 (0.326) -0.903*** (0.329) -0.550* (0.302) 0.434 (0.408) 0.690 (0.416) 3.313** (1.464) 0.388 (0.376) -0.440 (0.472) -0.807** (0.327) -1.212*** (0.312) -0.239 (0.206) 0.446 (0.342) 0.787** (0.327) 4.643*** (1.018) 0.712*** (0.256) 0.765* (0.449) -0.592 (0.394) -0.788** (0.303) -0.706** (0.333) -0.0574 (0.223) -0.163 (0.276) 1.131*** (0.260) 0.495 (1.755) 67 0.441 67 0.668 67 0.695 0.217 (0.361) -0.983** (0.441) -1.310*** (0.279) -0.998*** (0.360) -0.114 (0.232) 0.401 (0.366) 1.313*** (0.321) 5.147*** (0.882) Table E1: Regression analysis of the relationship between consanguinity and corruption. Higher values of the dependent variable imply lower corruption. Table E1 reports the full regression table underlying Table 4 in Section 3.1. 11 See http://www.transparency.org/research/cpi/overview. 20 VARIABLES Consanguinity Ethnic fractionalization (1) Basic model (2) and Religion (3) and Family ties (4) and Trust (5) and Genetic diversity (6) and Geography -4.463*** (0.769) 0.524 (0.514) -2.411** (1.176) 0.451 (0.458) 1.553* (0.791) -0.904** (0.370) -1.492*** (0.459) -3.853*** (0.722) 0.969 (0.667) -4.663*** (0.842) 0.846* (0.486) -3.494*** (0.952) 0.274 (0.516) -3.874*** (0.999) -0.140 (0.447) -0.126 (0.182) -4.349** (2.092) 0.446 (0.365) -0.643 (0.433) -0.895** (0.369) -0.989*** (0.318) -0.047 (0.227) 0.122 (0.388) 0.916* (0.458) 9.296*** (1.833) yes 65 0.770 Protestant Catholic Muslim Family ties -1.261** (0.528) General trust 1.562 (1.105) Genetic diversity -0.284** (0.133) 1.695** (0.681) -0.232 (0.340) -0.497 (0.469) -0.790** (0.331) -1.115*** (0.234) 0.156 (0.202) 0.087 (0.277) -0.780 (0.664) 5.805*** (1.076) -0.329 (0.216) 1.138 (1.299) -0.037 (0.357) -0.464 (0.536) -0.776** (0.366) -1.440*** (0.430) -0.259 (0.286) 0.008 (0.572) 0.255 (0.431) 6.127*** (1.674) -0.257* (0.143) 1.531 (1.001) 0.680 (0.570) -0.526 (0.502) -0.631* (0.338) -1.054*** (0.371) -0.140 (0.301) 0.491 (0.336) 0.400 (0.444) 3.224** (1.496) 48.933 (75.326) -40.988 (56.274) -0.223* (0.133) 2.296** (0.894) 0.639 (0.397) -0.582 (0.517) -1.305** (0.558) -1.084*** (0.351) -0.130 (0.225) 0.151 (0.398) 0.802** (0.359) -9.197 (25.046) 67 0.762 45 0.783 56 0.725 67 0.689 Genetic diversity squared Log population Latitude Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant -0.175 (0.129) 1.898** (0.878) 0.388 (0.376) -0.440 (0.472) -0.807** (0.327) -1.212*** (0.312) -0.239 (0.206) 0.446 (0.342) 0.787** (0.327) 4.643*** (1.018) Geographical variables Observations 67 R-squared 0.668 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table E2: Regression analysis of the relationship between consanguinity and corruption: potential confounds. N varies due to missing data for some countries. Table E2 reports the full regression table underlying Table 5 in Section 3.1. 21 VARIABLES Cousin term measure Ethnic fractionalization (1) Basic model (2) and Religion (3) and Family ties (4) Trust (5) and Genetic diversity (6) and Geography -0.733*** (0.181) 0.044 (0.453) -0.323* (0.190) 0.283 (0.402) 0.639 (0.520) 0.367 (0.255) -0.895*** (0.250) -0.706** (0.317) 0.295 (0.548) -1.039*** (0.247) 0.010 (0.440) -0.519*** (0.183) 0.020 (0.427) -0.551** (0.230) 0.120 (0.507) -0.138 (0.105) -0.172 (1.601) -0.265 (0.266) -0.103 (0.315) -0.488* (0.270) -1.393*** (0.269) -0.066 (0.163) 0.193 (0.362) 1.352*** (0.364) 4.905*** (1.702) yes 122 0.703 Protestant Catholic Muslim Family ties -0.580 (0.620) General trust 1.540* (0.898) Genetic diversity -0.143 (0.093) 2.413*** (0.648) -0.638*** (0.227) -0.297 (0.326) -1.136*** (0.266) -1.386*** (0.216) -0.021 (0.157) 0.213 (0.251) 0.769 (0.475) 3.888*** (0.760) -0.242 (0.161) 3.207** (1.286) -0.060 (0.540) 0.136 (0.587) -0.168 (0.377) -1.719*** (0.364) -0.240 (0.311) -0.067 (0.484) 0.852 (0.597) 4.476*** (1.502) -0.251** (0.126) 2.821*** (0.807) 0.216 (0.342) -0.187 (0.440) -0.154 (0.313) -1.516*** (0.287) -0.190 (0.225) 0.120 (0.299) 1.024** (0.475) 2.762** (1.310) 23.798 (57.826) -24.128 (41.920) -0.140 (0.099) 3.076*** (0.634) 0.034 (0.218) -0.406 (0.339) -1.469*** (0.387) -1.232*** (0.227) 0.001 (0.151) 0.461 (0.351) 1.420*** (0.279) -1.029 (20.055) 127 0.702 69 0.726 86 0.728 127 0.652 Genetic diversity squared Log population Latitude Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant -0.128 (0.101) 3.012*** (0.734) -0.192 (0.224) -0.051 (0.337) -0.529** (0.253) -1.413*** (0.250) -0.119 (0.160) 0.337 (0.334) 1.450*** (0.307) 3.620*** (0.771) Geographical variables Observations 128 R-squared 0.619 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table E3: Regression analysis of the relationship between cousin term measure and corruption (reduced form). N varies due to missing data for some countries. Table E3 reports the full regression table underlying Table 6 in Section 3.1. 22 VARIABLES Consanguinity Ethnic fractionalization (1) Basic model (2) and Religion (3) and Family ties (4) Trust (5) and Genetic diversity (6) and Geography -5.048*** (1.157) 0.586 (0.532) -0.814 (4.631) 0.193 (0.604) 1.996 (1.571) -0.715 (0.602) -1.722** (0.835) -3.498*** (1.219) 0.869 (0.570) -5.610*** (1.313) 1.057* (0.527) -4.503** (1.896) 0.406 (0.587) -5.804*** (1.860) -0.136 (0.542) -0.174 (0.196) -5.262** (2.327) 0.622 (0.568) -0.822* (0.487) -1.304** (0.527) -1.089*** (0.336) -0.009 (0.278) 0.097 (0.441) 0.915* (0.536) 10.310*** (2.042) yes 63 0.747 Protestant Catholic Muslim Family ties -1.097* (0.586) General trust 1.513 (1.127) Genetic diversity -0.305** (0.128) 2.151** (0.910) -0.362 (0.711) -0.250 (0.630) -0.699* (0.397) -1.059*** (0.324) 0.237 (0.184) 0.163 (0.320) -1.105 (1.008) 5.596*** (1.287) -0.186 (0.212) 2.281 (1.720) 0.278 (0.448) -0.350 (0.570) -0.503 (0.506) -1.307*** (0.441) -0.126 (0.319) 0.131 (0.592) 0.281 (0.437) 4.420** (2.019) -0.253* (0.141) 1.453 (1.501) 0.833 (0.673) -0.628 (0.638) -0.856 (0.566) -1.070** (0.396) -0.007 (0.344) 0.563 (0.355) 0.436 (0.432) 3.291* (1.664) -6.551 (90.365) 1.878 (68.444) -0.210 (0.130) 2.098 (1.359) 0.617 (0.474) -0.504 (0.584) -1.352** (0.628) -1.182*** (0.402) -0.064 (0.238) 0.358 (0.425) 0.776** (0.346) 8.544 (29.887) 64 0.792 44 0.801 53 0.709 64 0.687 Genetic diversity squared Log population Latitude Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant -0.185 (0.130) 1.976* (1.106) 0.526 (0.423) -0.467 (0.582) -0.947* (0.483) -1.252*** (0.340) -0.120 (0.232) 0.473 (0.361) 0.734** (0.323) 4.713*** (1.195) Geographical variables Observations 64 R-squared 0.667 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table E4: Regression analysis of the relationship between consanguinity and corruption, with cousin term measure as an instrument for consanguinity. N varies due to missing data for some countries. Table E4 reports the full regression table underlying Table 7 in Section 3.1. 23 Re-analysis using restricted samples Table E5 report direct comparisons of our baseline regression and those regressions computing the effects of various cultural variables on corruption, restricting the sample to only those countries for which data exists on both metrics. Note that this is not necessary for the specification including genetic diversity, since we have genetic diversity data on all 67 countries in our baseline regression. The results are consistent with those reported in the body of the paper. VARIABLES Consanguinity Ethnic fractionalization (1) Basic model (2) restricted sample (3) and Family ties (4) restricted sample (5) and Trust (6) restricted sample (7) and Geography -4.463*** (0.769) 0.524 (0.514) -4.533*** (0.760) 0.797 (0.637) -3.853*** (0.722) 0.969 (0.667) -1.261** (0.528) -4.470*** (0.882) 0.837 (0.503) -4.663*** (0.842) 0.846* (0.486) -4.049*** (0.769) 0.561 (0.501) -3.874*** (0.999) -0.140 (0.447) -0.168 (0.134) 2.522*** (0.902) 0.448 (0.376) -0.247 (0.465) -0.596* (0.326) -1.195*** (0.318) -0.237 (0.222) 0.507 (0.348) 0.705** (0.327) 4.245*** (1.142) -0.126 (0.182) -4.349** (2.092) 0.446 (0.365) -0.643 (0.433) -0.895** (0.369) -0.989*** (0.318) -0.047 (0.227) 0.122 (0.388) 0.916* (0.458) 9.296*** (1.833) 65 0.677 yes 65 0.770 Family ties General trust Log population Latitude Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant -0.175 (0.129) 1.898** (0.878) 0.388 (0.376) -0.440 (0.472) -0.807** (0.327) -1.212*** (0.312) -0.239 (0.206) 0.446 (0.342) 0.787** (0.327) 4.643*** (1.018) Geographical variables Observations 67 R-squared 0.668 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.210 (0.223) 2.200* (1.160) 0.053 (0.459) -0.548 (0.598) -0.723* (0.366) -1.274*** (0.414) -0.305 (0.303) 0.462 (0.471) 0.652 (0.395) 4.798*** (1.749) -0.329 (0.216) 1.138 (1.299) -0.037 (0.357) -0.464 (0.536) -0.776** (0.366) -1.440*** (0.430) -0.259 (0.286) 0.008 (0.572) 0.255 (0.431) 6.127*** (1.674) -0.202 (0.133) 2.172* (1.092) 0.517 (0.554) -0.375 (0.522) -0.641* (0.369) -1.211*** (0.357) -0.291 (0.238) 0.547 (0.363) 0.759** (0.352) 4.649*** (1.124) 1.562 (1.105) -0.257* (0.143) 1.531 (1.001) 0.680 (0.570) -0.526 (0.502) -0.631* (0.338) -1.054*** (0.371) -0.140 (0.301) 0.491 (0.336) 0.400 (0.444) 3.224** (1.496) 45 0.747 45 0.783 56 0.705 56 0.725 Table E5: Regression analysis of the relationship between consanguinity and corruption: sample size restrictions. Higher values of the dependent variable imply lower corruption. N varies due to missing data for some countries. The issue of data collection dates Another concern is the heterogeneous data collection dates for our measure of consanguinity. As explained in the data description in Appendix C, our measure of consanguinity is the weighted average of 448 sample studies in different locations from 72 countries, with the data collection date varying from 1922 to 2013. To analyze the data in more detail, we categorize the 448 sample studies into five overlapping groups: studies conducted after 1922, after 1950, after 1975, after 1980, and after 1984. Nine studies are excluded from this categorization because their exact data collection dates are not reported in Bittles and Black (2015). 67 countries are covered by both studies after 1922 and after 1950, and the average consanguinity rate of 67 countries in the first group of studies (after 1922) is 13.651 and drops to 13.508 in the group of studies after 1950. 34 countries are covered by both studies after 1922 and after 1975, and the average consanguinity rate of 34 countries in the first group of studies (after 1922) is 23.23 and drops 24 to 22.847 in the group of studies after 1975. 31 countries are covered by both studies after 1922 and after 1980, and the average consanguinity rate of 34 countries in the first group of studies (after 1922) is 24.360 and drops to 23.698 in the group of studies after 1980. 26 countries are covered by both studies after 1922 and after 1984, and the average consanguinity rate of 26 countries in the first group of studies (after 1922) is 25.056 and increases to 25.683 in the group of studies after 1984. This information reveals that countries with relatively high consanguinity rates have been studied later. In fact, the correlation between average data collection date weighted by sample size and consanguinity rate is 0.48. This is consistent with the later studies being motivated by an increasing awareness of the genetic sources of certain health problems and attempts to determine whether consanguinity was increasing the frequency of deleterious recessive genes. Perhaps surprisingly, the trend of consanguinity is flat or weakly decreasing overall. The correlation between data collection date and consanguinity rate within countries that were sampled in multiple studies over the years is positive for some, and for others negative, but the average of the correlations for all countries is -0.14. As noted by Bittles and Black (2015), “past predictions of a rapid decline in the overall prevalence of consanguineous unions have proved to be largely incorrect. In fact, the recorded numbers of consanguineous unions appear to have grown at least in step with increasing national and regional populations, and in some economically less developed countries the proportion of marriages contracted between close biological kin has expanded. The simplest explanation for this observation is that as greater numbers of children survive to marriageable age, the traditional social preference for consanguineous unions can be more readily accommodated” (Bittles and Black, 2015, Summary). Therefore, it seems that consanguinity is quite stable over time and could be among the drivers of institutional differences across countries. Alesina and Giuliano (2014) shows that the same is true about family values. Since the overall consanguinity trend is weakly decreasing, if consanguinity data is collected earlier for relatively corrupt and low development countries, we might end up with spurious correlation between consanguinity and corruption. However, the correlation of average data collection date of countries weighted by sample size with average income per capita of the countries for 1984-2011 is -0.10 and with average corruption index of countries for 1984-2011 is -0.24. It follows that the consanguinity data is collected later for countries with relatively high corruption, low economic development and high consanguinity. This can only underestimate the consanguinity rates of relatively high-corruption countries and overestimate the consanguinity rates of relatively low-corruption countries and would lead us to underestimate the effect of the consanguinity on corruption in the regressions. Nevertheless, to further investigate the robustness of the results, we also include the variable of average data collection date of countries weighted by sample size to regression (3) of Table 4 (including basic regressors and consanguinity). The result is displayed in column (1) of Table E6 and shows that consanguinity is significant even after controlling for the data collection dates and the variable of data collection dates is not significant. Also, we have run the same regression with different consanguinity rates obtained from the five groups of studies (sample studies after 1922, after 1950, after 1975, after 1980 and after 1984) and the results support a highly significant effect of consanguinity. 25 VARIABLES Consanguinity Consanguinity data collection date Ethnic fractionalization Latitude Log population Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant (1) Full sample 1922-2013 (2) Data collection date 1950-2013 (3) Data collection date 1975-2013 (4) Data collection date 1980-2013 (5) Data collection date 1984-2013 -5.169*** (0.621) -5.108*** (0.641) -7.166*** (1.153) -6.422*** (0.872) -5.113*** (1.121) 0.704 (0.528) 1.915** (0.830) -0.150 (0.134) -0.546*** (0.200) -0.538 (0.459) -0.873*** (0.323) -1.387*** (0.299) -0.359* (0.211) 0.333 (0.362) 0.658** (0.319) 4.583*** (1.015) 2.331** (0.886) 1.522 (1.082) -0.209 (0.176) 1.956*** (0.669) 0.849 (1.011) -0.109 (0.162) 0.842 (0.692) 1.183 (1.757) -0.217 (0.203) -1.530*** (0.454) -0.525 (0.573) 0.332 (0.503) -0.561* (0.322) 0.304 (0.364) 0.569 (0.574) 5.083*** (1.276) -1.608*** (0.437) -0.486 (0.494) 0.220 (0.455) -0.715** (0.333) 0.191 (0.279) 0.807 (0.555) 4.665*** (1.243) -0.991 (0.708) -0.215 (0.721) 0.574 (0.693) -0.613 (0.357) 0.615 (0.629) 5.411** (1.841) 62 0.710 30 0.779 27 0.840 22 0.802 -0.002 (0.002) 0.913 (0.548) 1.656* (0.866) -0.181 (0.140) -0.507** (0.229) -0.702 (0.443) -1.001*** (0.337) -0.849* (0.467) -0.362* (0.210) 0.408 (0.365) 0.753** (0.357) 8.677* (4.490) Observations 58 R-squared 0.727 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table E6: Regression analysis of the relationship between consanguinity and corruption, controlling for date of data collection. Higher values of the dependent variable imply lower corruption. N varies due to missing data for some countries. 26 E.2 Measures of Consanguinity from the Ethnographic Atlas Variables 23-26 in the Ethnographic Atlas which all are constructed based on column 25 of Murdock (1967)’s codebook (Kirby et al., 2016) provide measures related to historical consanguineous marriage practices - in particular, cousin marriages. In column 25, Murdock listed types of cousins12 a man is permitted to marry. He also indicated with an extra letter those types which are “preferred rather than just permitted” (Gray, 1999; Kirby et al., 2016). To discuss possible measures, we categorized variables 24 and 26 of the Ethnographic Atlas, as shown in Table E7.13 Preferred (1) A type of first/second cousin marriage preferred (2) No preferred cousin marriage Total (1) Four of four first cousins 55 62 117 (2) Three of four first cousins 6 19 25 (3) Two of four first cousins 140 75 215 (4) One of four first cousins 30 19 49 (5) No first cousins (some or all second cousins) (6) No first/second cousins 14 340 354 0 282 282 Total 245 797 1024 Permitted Table E7: Intersection of ”cousin marriages permitted” and ”Cousin marriages preferred” from the Ethnographic Atlas. First, let’s consider the cousin marriage permitted. This variable doesn’t capture cousin marriage frequency especially in the country level. The reason is that some populations such as New Englanders permitted marriage with all four first cousins, same as the Middle Eastern populations of Syrians, Turks and Iranians. This is probably just because Protestantism allowed cousin marriages. This variable doesn’t capture the fact that Protestantism historically discouraged cousin marriages. This difference between permitting cousin marriage, and frequencies of cousin marriage is likely to be true about some other populations as well. However, New Englanders is the most problematic for our country level analyses, because it is the only English society with available data on cousin marriage permitted. Therefore, countries with majority English populations such as UK, US, Australia, New Zealand, Canada, and also English regions in many countries (such as Ireland and South Africa) pick the cousin marriage measure from New Englanders’ society in the Ethnographic Atlas, while according to historical records and consanguinity rates (Bittles and Black, 2015), there is a big difference between consanguinity rates of these countries, and those of the Middle Eastern and North African countries. 12 (i) Mother’s brother’s daughter, (ii) Father’s sister’s daughter, (iii) Mother’s sister’s daughter, (iv) Father’s brother’s daughter, (v) second cousins. Matrilateral cousin refers to (i) and (iii); Patrilateral cousin refers to (ii) and (iv); Cross cousin refers to (i) and (ii); Parallel cousin refers to (iii) and (iv). 13 Based on variable 24, we constructed a variable for cousin marriages permitted. The variable takes on integer values ranging from 1 to 6 where higher values indicate wider cousin marriage culture; value 1 is assigned if there no first or second cousin marriages are allowed (category 8), value 2 is assigned if only second cousin marriages are allowed (categories 5, 6 and 7, considering that categories 5 and 6 are not distinguishable from category 7), and values 3 to 6 is assigned if one, two, three, and four of four cousins are marriageable receptively (categories 4 to 1 respectively). Also based on variable 26, we constructed another variable for cousin marriages preferred. The variable takes value 1 if a first or second cousin is preferred spouse (categories 1-5), and takes value 0 if there is no preferred cousin marriage. 27 Now, let’s consider the cousin marriage preferred. This variable also doesn’t capture cousin marriage frequencies across societies, and in the country level. First, consider that there is no report of cousin marriage preferred if it is not permitted (see table E7). Therefore cousin marriage preferred is a subset of cousin marriage permitted. Second, there are 245 societies that permit cousin marriages but has no preference for any types of cousins. While 72% of the Ethnographic Atlas societies permitted cousin marriage, only 24% of societies are reported to have a marriage preference with a cousin. The reason seems to be that cousin marriage preferred is not defined versus no preference for cousin marriage, but versus indifference among permitted cousins for marriage. Therefore, no report of a cousin marriage preferred for a society does not necessarily imply cousin marriages to be uncommon.14 Instead, it might have some implications about the nature of the descent group. According to Murdock, “the worldwide incidence of such preferences is so low”, but still reveals that anthropologists “are correct in ascribing matrilateral preferences primarily to patrilineal societies and patrilateral preferences to matrilineal societies” (Murdock, 1957, p.687). Third, while 175 societies permit marriage with first cousins but has no preferred cousin marriage, there are 14 societies which do not permit any first cousin marriages but has a preference to marry a second cousin. If cousin marriage preferred is used as cousin marriage measure, it is not obvious why the latter societies should have higher cousin marriage frequencies, while the earlier societies provide more possibility for cousin marriage. E.3 Giuliano and Nunn’s (2013) Data (a) Alesina et al. (2013). (b) Our replication. Figure E1: Traditional plough use across ethnic/linguistic groups. Downloaded from http://qje.oxfordjournals.org/ at Simon Fraser University on April 2, 2016 To compare our matching of languages to the Ethnographic Atlas societies with Alesina et al. (2013), we replicated the ethnic-level map of traditional plough use. The result shows very high similarity except for the newly added polygons in Australasia, central America, and south America (since we used new versions of the data and polygons). Robustness checks. In our country-level cousin marriage measure, we ignored the ethnicities with missing data. In regressions (1) and (2) of Table E8, we replicated the basic regression for both reduced form and instrumental variable estimation using the newly released data by Giuliano and Nunn (2013). 14 For example, New Englanders, Turks, Syrians, and Iranians all reported to permit marriage with all four first cousins. However, New Englanders and Turks has no preferred cousin marriage, while Syrians and Iranians prefer marriage with father’s brother’s daughter. This simply implies that Turks were indifferent among their cousins when marrying one, but it doesn’t imply that they had no preference for cousin marriage in general same as New Englanders (as we know from Bittles and Black (2015)’s consanguinity rates). 28 Our measure of cousin terms is highly correlated with theirs for the countries for which we also have corruption data (Spearman’s ρ = 0.78, p-value < 0.001, N = 129). Moreover, in regressions (3) and (4) we dropped countries for which share of ethnicities with missing information is more than 50%. Regressions (5) and (6) includes countries for which share of ethnicities with missing information is zero. Therefore, our results are robust with respect to using alternative data from Giuliano and Nunn (2013), or whether we drop countries with missing information for cousin term measure on some or all ethnicities. VARIABLES Cousin term measure (1) Reduced form -0.849*** (0.169) Consanguinity Ethnic fractionalization Log population Latitude Africa East Asia Latin America Socialist legal origin French legal origin German legal origin Scandinavian legal origin Constant (2) Instrumental variable 0.251 (0.448) -0.200* (0.106) 2.824*** (0.716) -0.301 (0.241) -0.107 (0.346) -0.673** (0.260) -1.397*** (0.241) -0.139 (0.161) 0.345 (0.320) 1.017*** (0.297) 4.221*** (0.846) Observations 128 R-squared 0.624 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 (3) Reduced form (4) Instrumental variable -0.823*** (0.190) -4.845*** (1.164) 0.677 (0.533) -0.243 (0.163) 1.771* (0.991) 0.348 (0.433) -0.468 (0.566) -0.927** (0.457) -1.318*** (0.350) -0.279 (0.249) 0.462 (0.368) 0.718** (0.351) 5.255*** (1.215) 63 0.663 (5) Reduced form (6) Instrumental variable -1.139*** (0.289) 0.310 (0.477) -0.253** (0.120) 2.678*** (0.724) -0.336 (0.278) -0.142 (0.352) -0.794*** (0.277) -1.407*** (0.251) -0.083 (0.188) 0.376 (0.325) 1.018*** (0.302) 4.659*** (0.935) -4.556*** (1.255) 0.594 (0.603) -0.208 (0.198) 1.860* (1.049) 0.307 (0.424) -0.486 (0.581) -0.910* (0.459) -1.287*** (0.345) -0.236 (0.263) 0.420 (0.421) 0.720** (0.350) 4.957*** (1.364) 0.762 (0.638) -0.571*** (0.194) 2.574** (0.963) 0.311 (0.447) -0.194 (0.497) -1.230*** (0.355) -1.533*** (0.321) -0.110 (0.316) 0.415 (0.398) 0.852** (0.406) 6.987*** (1.397) -6.053*** (1.749) 1.954 (1.393) -0.497 (0.430) 2.029 (1.330) -1.124 (0.778) -0.924 (0.618) -1.342** (0.476) -1.930*** (0.658) -0.316 (0.466) 0.838 (0.718) 0.543 (0.656) 6.910** (2.699) 112 0.633 57 0.666 59 0.718 29 0.705 Table E8: Regression analysis of the relationship between consanguinity and corruption, with cousin term measure in reduced form and as the instrument for consanguinity. N varies due to missing data for some countries. Regressions (1) and (2) ignore the ethnicities with missing data. Regressions (3) and (4) drop countries for which share of ethnicities with missing information is more than 50%. Regressions (5) and (6) includes countries for which share of ethnicities with missing information is zero. The data on cousin marriage measure in these specifications comes from the aggregation of the Ethnographic Atlas as reported in Giuliano and Nunn (2013). In addition, we replicated Alesina et al. (2003)’s regressions in Tables D1 and D2 using the cousin term measure in both reduced form and instrumental variable estimates; the connection between consanguineous marriage and corruption remains robust. See Appendix D. E.4 Additional within-country analysis (Italy) Table E9 reports the full regression table underlying Table 9 in Section 3.2. Table E10 reports robustness checks for the specifications reported in Table 11 in Section 3.2. Here we see that the effect of the instrument remains significant controlling for geography, but when we control for latitude (which is highly correlated with the instrument), standard errors become large and the coefficient is no longer significant. 29 (1) Basic model (2) and Consanguinity (3) Income instead of Share of agriculture (4) Share of agriculture as instrument (5) both Income and Share of agriculture (6) without Income and Share of agriculture 4.506* (2.421) 4.440* (2.470) -0.359*** (0.130) -0.125* (0.063) 0.697 (6.040) 0.437 (0.524) -0.001 (0.002) 0.018 (0.054) -0.449 (0.788) 1.387 (2.426) 17.200** (6.607) -0.271* (0.147) -0.120* (0.060) -4.032 (4.367) 0.362 (0.526) -0.002 (0.002) 0.026 (0.054) 0.580 (0.859) -1.896 (2.666) 13.886* (7.026) -0.032 (0.124) 0.029 (0.081) -0.278* (0.148) -0.121** (0.060) -4.075 (4.373) 0.349 (0.526) -0.002 (0.002) 0.027 (0.055) 0.629 (0.872) -2.051 (2.706) 14.120** (7.002) -0.067 (0.181) -0.024 (0.228) -0.269* (0.148) -0.119* (0.063) -3.967 (4.428) 0.375 (0.524) -0.002 (0.002) 0.026 (0.054) 0.583 (0.851) -1.907 (2.640) 13.953** (6.961) 4.472* (2.460) 0.991 (3.631) -0.021 (0.132) 0.042 (0.088) -0.273* (0.149) -0.122** (0.061) -4.143 (4.381) 0.339 (0.538) -0.002 (0.002) 0.026 (0.055) 0.575 (0.870) -1.877 (2.701) 13.771* (7.087) 4.470* (2.395) 1.537 (3.469) -0.020 (0.123) 4.451* (2.443) 0.366 (3.385) -0.050 (0.113) -0.273* (0.146) -0.120** (0.060) -4.017 (4.350) 0.363 (0.521) -0.002 (0.002) 0.026 (0.054) 0.604 (0.851) -1.973 (2.641) 14.030** (6.941) 101 0.502 101 0.502 101 0.501 101 0.503 101 0.502 VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Latitude Mean Temperature Mean precipitation Suitability for agriculture Distance to coast Elevation Slope Ruggedness Constant Observations 101 R-squared 0.459 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.051 (0.111) Table E9: Replication of Table 8 including controls for climate and geography. (1) Basic model (2) and geography (3) and latitude (4) both Income and Share of agriculture (5) and geography (6) and latitude 8.306*** (2.396) -2.205 (5.420) -0.006 (0.146) 9.535* (5.551) 0.227 (4.007) -0.064 (0.135) 7.634 (6.498) -0.472 (3.748) -0.071 (0.124) 8.658*** (2.538) -1.731 (5.552) 0.045 (0.156) 0.077 (0.094) 10.407* (5.820) 0.622 (4.246) -0.038 (0.148) 0.048 (0.104) -0.067 (0.054) -6.465 (7.998) 0.065 (0.569) -0.006*** (0.002) 0.083 (0.062) 0.950 (1.926) -3.209 (6.046) 2.407 (1.917) -0.208 (0.202) -0.116* (0.060) -7.412 (7.485) 0.308 (0.553) -0.003 (0.003) 0.031 (0.064) 1.316 (1.616) -4.243 (5.128) 11.516 (8.738) 0.530 (1.973) -0.072 (0.055) -7.736 (8.260) 0.042 (0.599) -0.006*** (0.002) 0.080 (0.065) 1.206 (2.006) -4.012 (6.300) 2.274 (1.988) 8.787 (6.663) 0.035 (3.984) -0.042 (0.138) 0.053 (0.096) -0.189 (0.207) -0.117* (0.061) -8.752 (7.471) 0.259 (0.590) -0.004 (0.003) 0.033 (0.068) 1.569 (1.672) -5.049 (5.305) 10.530 (8.884) 101 0.417 101 0.462 VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Latitude Mean Temperature Mean precipitation Suitability for agriculture Distance to coast Elevation Slope Ruggedness Constant Observations R-squared 0.955 (1.899) 101 0.307 101 101 101 0.435 0.480 0.281 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table E10: Active years of archdioceses as an instrument for consanguinity. 30 Time series of consanguinity across Italy 70 As described in Appendix C Table C2, we computed the consanguinity rate of Italian provinces using consanguinity percentages reported by Cavalli-Sforza et al. (2004) for the period 1945-1964. Although, consanguinity was measured for regions in the late 1940s, 1950s and early 1960s, if the regions share a common trend, it can be argued that consanguinity rates plausibly capture provincial differences in subsequent decades as well. 50 40 30 20 0 10 Consanguinity % 60 FRIULI VENEZIA GIULIA UMBRIA EMILIA ROMAGNA MARCHE VENETO VALLE D'AOSTA TOSCANA TRENTINO ALTO ADIGE PIEMONTE LOMBARDIA LAZIO PUGLIA ABRUZZO LIGURIA CAMPANIA BASILICATA SARDEGNA MOLISE CALABRIA SICILIA 19101914 19151919 19201924 19251929 19301934 19351939 19401944 19451949 19501954 19551959 19601964 Year Figure E2: Consanguinity trend for Italian regions. As noted by Cavalli-Sforza et al. (2004), trends of consanguinity in the 20 Italian regions shown in Figure E2 reveal “first, an increase, at first slow, then faster, which ends around 1915-1925 when a peak of consanguinity is reached, at a similar time for all Italian regions, and, second, a rapid descent begins, [. . . ] some of the possible explanations of the first phase, which are, in order of time: (1) changes of laws of inheritance; (2) decreased influence of the Roman Catholic Church; and (3) increase in relative abundance of cousins due to increase in population size, which increases disproportionately the abundance of relatives, though with a delay of one generation for first cousins and two for second cousins, [. . . ] The second phase of the phenomenon, the decrease of consanguinity, must be the same as that called the breakdown of isolates [. . . ] where it began in the second half of the nineteenth century. It is essentially due to an increase in individual mobility, tied to increased means and speed of transportation, and increased opportunities of work in specific industrial areas, favoring relocation of workers, [. . . ] The difference in the peak consanguinity values between north and south agrees with the lower growth rate of the northern populations, [. . . However] the difference in the time at which the peak appears seems relatively small in the various parts of the country, indicating that the breakdown of isolates took place at similar times all over Italy and was a national rather than a local phenomenon.” (CavalliSforza et al., 2004, p. 238-41). Therefore, the assumption of a common negative consanguinity trend across Italian provinces appears plausible. From post-war highs, consanguinity rates declined by an average of almost 50% across regions by 1964. Moreover, in unreported regression analysis, when we regress consanguinity rates as a percentage of the region-level average from 1945-1964 on region dummies and a region×year interaction, we find support for a negative and significant slope in each region over the period 1945-1964. However, a Wald test rejects the null hypothesis that the slope is the same in all regions (p-value < 0.01). 31 Province-level data: additional robustness checks This section reports additional regression analyses of the relationship between consanguinity and corruption in Italy, controlling for potential confounds. For clarity, we report our baseline specifications drawn from column (2) of Tables 8 and 9 side by side in columns (1) and (2) of Table E11. As we add controls for other confounding variables, we report specifications with and without controls for climate and geography. Civil society. One cultural difference across Italian regions that may influence the quality of institutions and the level of corruption is the extent of civil society (civicness). The intuition is that more civically active areas may have better developed a capacity for effective government as citizens acquire habits, skills and values through participation in non-governmental organizations. Putnam et al. (1994) contrasts cooperation rooted in civil society to that based on Banfield’s (1958) notion of ‘amoral familism’ in which heavy reliance on kinship ties results in lack of general trust and community cohesion. Our first measure of civicness is Putnam et al. (1994)’s civic involvement index, which is an aggregation of five indicators measured between 1860 and 1921, on a scale of 1 to 9. This measure of civicness is negatively correlated with consanguinity (Spearman’s ρ = -0.73, p-value < 0.00, N=101). This might suggest that consanguinity is just a product of lack of civicness. However, for civic involvement in columns (3) and (4) of Table E11, consanguinity remains highly significant, although the results also confirm Putnam et al. (1994)’s hypothesis in column (3); civicness exhibits a modest but significant negative relationship with corruption. As a second measure of civicness, in columns (5) and (6) we use the number of voluntary organizations (established before 1945) per 100,000 inhabitants, with qualitatively similar results. Family types, family ties. Consanguinity may be most attractive when extended families are close-knit and there are frequent interactions with kin, and strong family ties have been implicated as a source of corruption. Thus, the observed effect of consanguinity may instead simply proxy for family ties. Consistent with this idea, a number of studies have shown that medieval family types as classified by Todd (1990) are associated with current regional or cross-country differences in development and institutions (e.g. Duranton et al., 2009; Galasso and Profeta, 2011; Alesina et al., 2015). Todd’s classification of family types comes from the cohabitation patterns between generations within families (nuclear or extended) and inheritance practices (equal or unequal division of assets among children). Todd (1990) uses his classification of family types “to explain relative levels of diffusion or resistance to important societal changes such as Protestantism, secularism, or political ideology” (Alesina and Giuliano, 2014, p. 180), where e.g. nuclear family structures encourage children to the leave the home, weakening the influence of extended family on norms and behavior. According to Todd (1990), Italy had three family types; incomplete stem family (extended family, unequal inheritance), communitarian family (extended family, equal inheritance), and egalitarian nuclear family (nuclear family, equal inheritance). In columns (11) and (12) of Table E12, we controlled for family types where the omitted dummy is the incomplete stem family. Consanguinity remains highly significant in the regression. The communitarian family type, characterized by both cohabitation with extended family and egalitarian inheritance, is also significantly associated with higher corruption. As a second and more recent measure of family ties, following Alesina et al. (2015) we used the fraction of youth aged 18-34 living with at least one parent, averaged over the period 2002-2009 (from ISTAT). Controlling for this measure of family ties in columns (9) and (1), consanguinity remains significant. South and islands. A well-known fact about corruption in Italy is that the corruption level is higher in the south. This difference is usually attributed to cultural differences between the south and 32 (1) Basic model (2) and geography (3) Civic involvement (1860-1921) (4) and geography (5) Voluntary organizations (established before 1965) (6) and geography 5.144*** (1.099) 4.451* (2.443) 3.848*** (1.462) -0.092** (0.045) 4.317* (2.457) -0.025 (0.057) 4.972*** (1.138) 4.470* (2.450) -0.009** (0.004) 3.614 (3.271) 0.039 (0.117) 0.605 (1.520) -0.008 (0.005) 0.834 (3.388) -0.033 (0.118) -0.258* (0.148) -0.119* (0.061) -4.148 (4.389) 0.272 (0.536) -0.003 (0.002) 0.033 (0.053) 0.560 (0.866) -1.837 (2.688) 13.174* (7.109) 101 0.435 101 0.504 VARIABLES Consanguinity Civic involvement Voluntary organizations Share of agriculture Log population 3.472 (3.243) 0.022 (0.115) Latitude Mean Temperature Mean precipitation Suitability for agriculture Distance to coast Elevation Slope Ruggedness Constant 0.763 (1.505) Observations 101 R-squared 0.430 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.366 (3.385) -0.050 (0.113) -0.271* (0.147) -0.120* (0.060) -4.032 (4.367) 0.362 (0.526) -0.002 (0.002) 0.026 (0.054) 0.580 (0.859) -1.896 (2.666) 13.886* (7.026) 1.829 (3.613) 0.004 (0.114) 1.764 (1.613) 0.339 (3.409) -0.051 (0.113) -0.240 (0.167) -0.115* (0.060) -3.702 (4.386) 0.336 (0.525) -0.002 (0.002) 0.020 (0.055) 0.511 (0.885) -1.669 (2.756) 12.749* (7.654) 101 0.502 101 0.455 101 0.503 Table E11: Regression analysis of the relationship between consanguinity and corruption in Italy: controlling for civil society. north (Banfield, 1958; Putnam et al., 1994). Figure 4b shows high consanguinity rates in southern Italy. Therefore, the impact on corruption that we attribute to consanguinity might instead result from some other feature of southern Italy related to corruption. Therefore, we also included a dummy variable in the regressions for the provinces in so-called Mezzogiorno regions: southern Italy (Abruzzo, Molise, Campania, Puglia, Basilicata, Calabria) and the islands (Sicilia, Sardegna). Including this dummy variable assures a stringent test of our hypothesis because southern Italy and the islands also happen to have the highest consanguinity rates. Nevertheless, including a dummy for the south and islands, the relationship between consanguinity and corruption remains robust. Political domination. A historical path dependency in Italy may have arisen from colonization and political domination to which different parts of Italy were subjected in the Middle Ages, which carried with them cultural, political and institutional baggage that could influence outcomes today. Most relevant to our study is that the Papal States were territories under the sovereign direct rule of the pope, from the 8th century until Italian unification in 1871. Therefore, it is plausible to expect stricter historical emphasis on consanguinity bans, and this is consistent with the fact that we observe lower consanguinity rates in the Papal states than in the other provinces (means are 6% and 13%, respectively, t-test, two-sided p-value < 0.01, N=101). Moreover, southern territories under the Spanish or Norman rule have had lower institutional quality and higher corruption (Putnam et al., 1994; Di Liberto and Sideri, 2015). These are also the regions with the highest consanguinity rates. On the other hand, northern territories were historically under Austrian administration which was “usually portrayed as a good administrator that 33 VARIABLES Consanguinity South and Islands (7) South and islands (8) and geography (9) Family ties (10) and geography (11) Family types (Middle Ages) (12) and geography (13) Dominations (Middle Ages) (14) and geography 4.267** (1.681) 0.348 (0.352) 4.670* (2.527) -0.247 (0.399) 4.746*** (1.227) 4.411* (2.536) 5.508*** (1.227) 5.099* (2.687) 5.209*** (1.397) 5.383* (2.756) 0.032 (0.029) -0.005 (0.040) 0.427** (0.197) 0.091 (0.196) 0.459 (0.332) 0.405 (0.359) 0.381 (0.269) -0.636*** (0.240) -0.121 (0.253) -0.125 (0.275) 0.065 (0.249) 3.534 (3.348) 0.021 (0.118) 0.739 (1.612) 0.232 (0.329) -0.810* (0.483) -0.108 (0.309) 0.129 (0.333) 0.109 (0.260) 0.944 (3.169) -0.028 (0.119) -0.248 (0.181) -0.127* (0.069) -3.706 (5.136) 0.453 (0.643) -0.003 (0.003) 0.032 (0.058) 0.919 (0.887) -2.953 (2.752) 12.441 (8.436) 101 0.455 101 0.517 Family ties Communitarian family Egalitarian nuclear family Papal Austrian Venetian Sabaudian Independent Share of agriculture Log population 2.325 (3.622) 0.017 (0.116) Latitude Mean Temperature Mean precipitation Suitability for agriculture Distance to coast Elevation Slope Ruggedness Constant 0.855 (1.525) Observations 101 R-squared 0.440 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 0.425 (3.361) -0.051 (0.114) -0.326** (0.160) -0.128** (0.061) -4.594 (4.294) 0.387 (0.537) -0.002 (0.002) 0.030 (0.056) 0.721 (0.905) -2.348 (2.816) 16.355** (7.566) 101 0.505 2.149 (3.741) 0.032 (0.113) -1.154 (2.017) 0.415 (3.531) -0.051 (0.114) -0.285 (0.197) -0.121* (0.061) -4.306 (5.131) 0.380 (0.502) -0.002 (0.002) 0.026 (0.054) 0.621 (0.886) -2.022 (2.744) 14.800 (10.818) 3.608 (3.204) 0.024 (0.117) 0.500 (1.562) 0.828 (3.151) -0.019 (0.122) -0.217 (0.160) -0.131** (0.065) -2.925 (4.574) 0.481 (0.563) -0.004 (0.003) 0.033 (0.052) 0.649 (0.862) -2.114 (2.685) 10.638 (7.923) 101 0.444 101 0.502 101 0.451 101 0.512 Table E12: Regression analysis of the relationship between consanguinity and corruption in Italy: controlling for family structure and political history. did not implement exploiting or extracting policies” (Di Liberto and Sideri, 2015, p. 13). To capture the effect of these historical differences, we use dummy variables based on the map constructed by Di Liberto and Sideri (2015) that identify, for each province, the administration that presided continuously during the period 1560-1659: Spanish, Papal, Austrian, Venetian, Sabaudian and Independent.15 In columns (13) and (14) of Table E12, we include these variables as additional controls; the omitted category is for provinces under Spanish domination. As expected, medieval Austrian administration in northern Italy is associated with lower corruption today, and being in the Papal states at that time is not associated with higher corruption than southern territories under the Spanish rule. However, consanguinity remains significant in the regressions. 15 The Papal and Spanish dominations in Di Liberto and Sideri (2015)’s map are roughly the same regions with the Papal states and the Kingdom of Sicily in Putnam et al. (1994)’s map. However, contrary to Putnam et al. (1994)’s map, there are no missing regions in Di Liberto and Sideri (2015)’s map. In addition, Di Liberto and Sideri (2015)’s mapping of the missing regions in Putnam et al. (1994)’s map, and of the Papal states is consistent with the sources used for Putnam et al. (1994)’s map, such as Hyde (1973). 34 Data collection dates. In Table E13 we vary the cutoff dates for defining the consanguinity rate in Italy. The results are robust to restricting attention to more recent consanguinity estimates. VARIABLES Consanguinity Share of agriculture Log population Constant (1) Consanguinity sample of basic model (1945-1964) (2) Consanguinity sample restricted to 1950-1964 (3) Consanguinity sample restricted to 1955-1964 (4) Consanguinity sample restricted to 1960-1964 5.144*** (1.099) 3.472 (3.243) 0.022 (0.115) 0.763 (1.505) 5.504*** (1.178) 3.195 (3.280) 0.021 (0.113) 0.820 (1.479) 5.701*** (1.222) 3.014 (3.323) 0.023 (0.110) 0.822 (1.444) 5.950*** (1.308) 2.749 (3.336) 0.022 (0.110) 0.872 (1.446) 101 0.436 101 0.440 101 0.438 Observations 101 R-squared 0.430 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table E13: Robustness check for date of data collection, Italy. Alternative corruption metric. Figure E2 displays Golden and Picci’s (2005) measure of corruption based on infrastructure at the province level alongside the consanguinity rate (see Appendix C for details on the data). Here a higher score is associated with lower corruption (in contrast to our baseline measure which uses the number of crimes). This alternative measure of corruption is even more highly correlated with consanguinity than our primary measure (Spearman’s ρ = −0.63, p-value < 0.001, N = 90). Moreover, Table E14 reports replications of our province-level regressions from Tables 8 and 9 in the body of the paper using this alternative measure of corruption. Consanguinity remains a robust predictor of corruption in all specifications. However, as Golden and Picci (2005) note, their province-level ratios are not cost-adjusted due to lack of data (the region-level ratios are cost-adjusted), and this is why we focus our main analysis on associative crime. Finally, note also that our measures of share of agriculture in GDP and population come from the period 2000-2013 (and hence from a period after Golden and Picci’s data on corruption). We do not have access to earlier province-level statistics on GDP or population, so the reliability of these specifications depends on the assumption that provincial GDP (and share of agriculture therein) and population in 2000-2013 is highly correlated with the same in 1954-1997, which seems plausible. 35 Figure E2: Alternative measure of corruption from Golden and Picci (2005) and consanguinity in Italy: province level. Darker colored regions are more corrupt. Grey colored areas indicate missing data. (1) Basic model (2) and Consanguinity (3) Income instead of Share of agriculture (4) Share of agriculture as instrument (5) both Income and Share of agriculture (6) without Income and Share of agriculture -2.323*** (0.343) -2.307*** (0.388) 2.850*** (1.052) 2.162** (0.967) -0.110 (0.084) -0.066 (0.067) 2.583** (1.015) -0.103 (0.119) -0.056 (0.138) 2.536** (1.207) -2.312*** (0.363) -0.239 (2.985) -0.113 (0.091) -0.068 (0.076) 2.620** (1.091) -2.216*** (0.318) -4.646* (2.352) -0.121 (0.079) -2.283*** (0.359) 1.058 (2.649) -0.062 (0.072) 90 0.275 90 0.282 90 0.281 90 0.282 90 0.274 (2) and Consanguinity (3) Income instead of Share of agriculture (4) Share of agriculture as instrument (5) both Income and Share of agriculture (6) without Income and Share of agriculture -1.703** (0.853) 3.326 (3.020) -0.099 (0.073) -1.685** (0.823) -1.770** (0.835) 0.182** (0.073) -0.003 (0.035) 1.939 (2.399) -0.016 (0.443) 0.001 (0.002) 0.049 (0.047) -0.774* (0.405) 2.247* (1.217) -5.214 (3.474) 0.158** (0.074) -0.002 (0.034) 4.317* (2.313) -0.099 (0.451) 0.002 (0.002) 0.037 (0.049) -1.158** (0.453) 3.483** (1.375) -4.504 (3.462) -0.196** (0.083) -0.119* (0.065) 0.157** (0.076) 0.000 (0.035) 4.588* (2.342) -0.084 (0.434) 0.002 (0.002) 0.034 (0.053) -1.096** (0.438) 3.275** (1.332) -3.509 (3.471) -0.219* (0.118) -0.156 (0.135) 0.161** (0.078) 0.001 (0.036) 4.838* (2.433) -0.079 (0.433) 0.002 (0.002) 0.032 (0.053) -1.127** (0.441) 3.377** (1.337) -3.555 (3.485) -1.749** (0.834) 1.049 (3.493) -0.181* (0.093) -0.107 (0.078) 0.160** (0.077) 0.000 (0.035) 4.673* (2.353) -0.085 (0.440) 0.002 (0.002) 0.034 (0.052) -1.137** (0.446) 3.410** (1.353) -3.855 (3.610) -1.411* (0.799) 2.209 (2.880) -0.125 (0.079) 0.141* (0.076) -0.003 (0.035) 3.786* (2.214) -0.100 (0.444) 0.002 (0.002) 0.042 (0.049) -0.994** (0.439) 2.947** (1.332) -3.361 (3.484) 90 0.394 90 0.407 90 0.405 90 0.408 90 0.384 VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Constant Observations 90 R-squared 0.042 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 (1) Basic model VARIABLES Consanguinity Share of agriculture Log population Log value added per capita Latitude Mean Temperature Mean precipitation Suitability for agriculture Distance to coast Elevation Slope Ruggedness Constant Observations 90 R-squared 0.374 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.068 (0.071) 2.268** (0.942) -0.120* (0.071) Table E14: Replication of Tables 8 and 9 using the corruption measure from Golden and Picci (2005). 36 Region-level data To provide further evidence for Italy, we collected data on corruption and consanguinity in 20 regions of Italy. We use multiple available measures of corruption for Italian regions: (i) number of corruption crimes (such as peculation, malversation and bribery) per 100,000 inhabitants used in previous studies (e.g. see Del Monte and Papagni, 2001, 2007; Fiorino et al., 2012; Blackburn et al., 2015), (ii) the ratio of existing public infrastructure (in 1998) to expected infrastructure given past government spending (from 1954-1997) computed by Golden and Picci (2005), (iii) the European Quality of Government Index (EQI) computed from survey data by Charron et al. (2014), and (iv) a measure of regional institutional performance from Putnam et al. (1994). Note that the latter three measures assign higher scores to less corrupt regions, in contrast to the measures based on crime statistics. Overall we find a highly significant correlation between consanguinity and each of our corruption measures for Italy (i) crimes (Spearman’s ρ = 0.59, p-value = 0.001, N=20) (ii) infrastructure (Spearman’s ρ = -0.79, p-value < 0.001, N=20), (iii) EQI (Spearman’s ρ = -0.81, p-value < 0.001, N=20), and (iv) institutional performance (Spearman’s ρ = -0.91, p-value < 0.001, N=20). Moreover, each of these measures is strongly and negatively correlated with province-level associative crime, providing further evidence that associative crime is a reasonable proxy for corruption (Spearman’s ρ < −0.43, all p-values ≤ 0.05). 37 F Experiment materials and procedures F.1 Experiment instructions You are now participating in a decision making experiment. At the end of the experiment, you will be paid in cash based on your decisions. Please read the instructions carefully so you understand clearly how your payoff is determined. Please do not talk to other participants. If you have any questions, raise your hand and the experimenter will answer them privately. This experiment consists of only one round where you will make a decision in a three-person scenario. You will be assigned one of three possible roles in this scenario: A, B and C. Your role will be determined randomly. In addition to your $7 show up payment, you will earn Experimental Currency Units (ECU) during the experiment based on your decision. 10 ECU is worth $1. At the end of the experiment, we will convert your earnings to from ECU to dollars. Figure 1 You will see a graph similar to Figure 1 during the experiment. The graph will help you to see the possible outcomes of the three-person scenario based on your and other participants’ decisions. 38 Little blue circles show which person is making a choice. Two black lines exiting from each circle show the choices available to the person choosing. For example, the top circle and its lines show that Person A can choose Choice 1 or Choice 2. Little blue squares show all possible outcomes of the scenario and the payoffs for each person are shown by A= ... B= ... C= ... Payoffs are determined based on the decisions of Person A and Person B. As you can see Person C has no decision to make. The scenario: You will participate in a three-player scenario. This scenario is shown in the figure. Participants in the roles of A and B start with an initial endowment of 100 ECU and C starts with 160 ECU. The scenario works as follows: Stage 1 39 Stage 1: At the first stage, A decides whether to choose ”Transfer” or ”Not-transfer”. - If A chooses ”Not-transfer”, the round ends immediately, and all participants’ final payoffs will be equal to their initial endowment (i.e. A=100, B=100, C=160). - If A chooses to ”Transfer” 40 ECU to B, his/her endowment will be reduced by 5 ECU. Whether 40 ECU will actually be transferred to B or not and the final payoffs will depend on choices made by B at the next stages. Stage 2 Stage 2: Assuming that A has transferred 40 ECU, B chooses ”Accept” or ”Reject”. - If B chooses ”Reject”, the round ends immediately. The final payoff of A will be 100-5=95, and the final payoff of B and C will be their initial endowments. - If B chooses ”Accept”, 40 ECU will be deducted from A’s endowment and will be added to B’s endowment. Then, the experiment moves to stage 3. Stage 3: B who has accepted the transfer, now decides whether to choose ”Right” or ”Left”. - If B chooses ”Left”, the round ends immediately and the payoffs of participants are as follows: 40 Stage 3 A receives the initial endowment minus 5 ECU, minus the amount of transfer = 100 - 5 - 40 = 55 ECU. B receives the initial endowment plus the transfer = 100 + 40 = 140 ECU. C receives the initial endowment = 160 ECU. - If B chooses ”Right”, 5 ECU will be deducted from B, 105 ECU will be deducted from C, but 105 ECU will be added to A. Then the round ends, and the payoffs of participants are as follows: A receives the initial endowment minus 5 ECU, minus the amount of the transfer, plus 105 = 100 - 5 40 + 105 = 160 ECU. B receives the initial endowment plus the transfer, minus 5 = 100 + 40 - 5 = 135 ECU. C receives the initial endowment minus 105 = 160 - 105 = 55 ECU. 41 F.2 Recruitment survey Vancouver Our online pre-experiment questionnaire in Vancouver included the following questions. Pre-experiment Questionnaire Please enter your email: 1. How old are you? 18-22 23-25 26-30 more than 30 ◦ ◦ ◦ ◦ 2. What is your field of study? 3. What is your gender? Male Female ◦ ◦ 4. Which country are you born in? 5. Are you a Canadian Citizen? Yes No ◦ ◦ 6. What is your PRIMARY ethnic origin? Aboriginal, African, Arab, Caribbean, Chinese, Dutch, English, French, German, Indian, Iranian, Irish, Italian, Jewish, Norwegian, Polish, Portuguese, Russian, Scottish, Spanish, Swedish, Ukrainian, Welsh, Other If you chose ”other”, please specify: 7. Please specify time slots (as many as you can) when you are available this semester for the experiment at [lab address at UBC/SFU]. 8. Would you like to participate in our experiment with a parent or sibling (18 or older) in a public location (such as Starbucks, Tim Hortons, McDonalds, etc) around your neighbourhood? Yes. No ◦ ◦ If “Yes”, which family member of yours (parent or sibling) will participate in the experiment? How old is she/he? and please specify the name and postal code of a public location. 42 Urmia Our in-paper pre-experiment Persian questionnaire in Iran, titled “Pre-research Questionnaire”, included the following questions. After we finished collecting data for our kin treatments, we just kept questions 1-7. Pre-research Questionnaire 1. Choose your age group? 18-22 23-25 26-30 more than 30 2. Choose your gender: Male Female 3. Choose your occupation group: Student Private sector Government sector Unemployed 4. Choose your ethnicity: (written in Persian alphabetical order) Beluch Azeri Turk Arab Persian Kurd Gilak Lur Other 5. What is your latest degree (or current level of study, if student)? High school diploma Kardani [2-year university or college degree] Karshenasi (B.A) Karshenasi Arshad (M.A) 6. Which university did/do you study? ——————— 7. What is your field of study? ——————— 8. Would your brother or sister also like to participate in the research? (in this case, both you and your sibling will be separately paid around 20000 Tomans for your participation in the research session) No Yes. My brother would like to participate Yes. My sister would like to participate If your answer to question 7 is ”Yes”, please enter your brother or sister’s age group: 18-22 23-25 26-30 more than 30 If your answer to question 7 is ”Yes”, please enter your brother or sister’s occupation group: Student Private sector Government sector Unemployed 9. To be informed of location, dates and hours of the research sessions, please enter your cell phone number. ——————— ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ 43 F.3 Post-experiment questionnaire The following are the list of questions from post-experiment questionnaire. Questions 1 and 2 are the “Harm” and “Ingroup” questions from the Moral Foundations questionnaire (Graham et al., 2008). Questions 3-5 are used to measure “family ties” and are drawn from the World Values Survey. Questions 11, 12, 13, 15 and 16 were not included in our experiments in Iran. The variables were coded numerically as indicated below. Q1. When you decide whether something is right or wrong, to what extent are the following considerations relevant to your thinking? Please rate each statement using this scale: 0 not at all relevant 1 not very relevant 2 slightly relevant 3 somewhat relevant 4 very relevant 5 extremely relevant • Whether or not someone suffered emotionally • Whether or not someone cared for someone weak or vulnerable • Whether or not someone was cruel • Whether or not someone was good at math • Whether someones action showed love for his or her country • Whether or not someone did something to betray his or her group • Whether or not someone showed a lack of loyalty Q2. Please read the following sentences and indicate your agreement or disagreement: 0 Strongly Disagree 1 Moderately Disagree 2 Slightly Disagree 3 Slightly Agree 4 Moderately Agree 5 Strongly Agree • Compassion for those who are suffering is the most crucial virtue. • One of the worst things a person could do is hurt a defenseless animal. • It can never be right to kill a human being. • I am proud of my country’s history. • People should be loyal to their family members, even when they have done something wrong. • It is better to do good than to do bad. • It is more important to be a team player than to express oneself. Q3. How important is family in your life? (3) Very important (2) Rather important (1) Not very important (0) Not at all important Q4. Which of two statements do you agree more? • (1) Regardless of what the qualities and faults of one’s parents are, one must always love and respect them. • (0) One does not have the duty to respect and love parents who have not earned it. Q5. Which of two statements do you agree more? • (1) It is the parents’ duty to do their best for their children even at the expense of their own well-being. • (0) Parents have a life of their own and should not be asked to sacrifice their own well-being for the sake of their children. Q6. Which of two statements do you agree more? • (1) Regardless of what the qualities and faults of one’s elderly relatives (grand-father and -mother, un- 44 cles and aunts) are, one must always love and respect them. • (0) One does not have the duty to respect and love elderly relatives who have not earned it. Q7. Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people? (1) Most people can be trusted (0) Need to be very careful Q8. How much do you trust . . . (3) Trust completely (2) Somewhat (1) Not very much (0) No trust at all • your extended family? • people of same ethnicity? • people you meet for the first time? • your family? Q9. To which of these groups would you say you belong first? (0) Your country (1) Your ethnicity Q10. How important is for you that your sibling would not marry someone of a different ethnicity? (3) Very important (2) Rather important (1) Not very important (0) Not at all important Q11. How important is for you to not have people of a different ethnicity as neighbor? (3) Very important (2) Rather important (1) Not very important (0) Not at all important Q12. Do you agree that ethnic diversity erodes a country’s unity? (1) Agree (0) Disagree Q13. How many of your friends have the same ethnicity as yours? (4) All of them (3) Most of them (2)About half of them (1) A few of them (0) None of them Q14. People have different views about themselves and how they relate to the world. How strongly do you agree or disagree with each of the following statements about how you see yourself? (3) Strongly agree (2) Agree • I see myself as a/an [ethnicity]. • I see myself as a/an [Nationality]. • I see myself as a world citizen. (1) Disagree (0) Strongly disagree Q15. In your opinion, what are the odds that someone in Vancouver pay a bribe to get a job? (2) more than 50 percent (1) 50 percent (0) less than 50 percent Q16. In your opinion, what are the odds that someone in Vancouver gets punished if he/she pays a bribe to get a job? (0) more than 50 percent (1) 50 percent (2) less than 50 percent Q17. Some people have a stronger sense of belonging to some things than others. How strong is your 45 sense of belonging to . . . Not strong at all 1 Very strong 2 3 4 5 • your family? • your extended family? • your ethnicity? • your country? Q18. Do you know any cousins who are married to each other? (1) YES (0) NO Q19. Suppose your parent hit a pedestrian while exceeding maximum speed. You are the only witness and police knows that. The only way to save your parent from the serious consequences is that you lie that he/she was not exceeding maximum speed. Would you lie to the police? (1) YES (0) NO Survey results For completeness, below we report the survey results from the post-experiment questionnaire, indicating whether there are significant differences between the responses in the two countries via Wilcoxon RankSum tests. Note the sharp difference in the familiarity with cousin marriage (Q18), as well as differences in the WVS questions underlying studies of family ties (Q3 - Q5). Question Q1.1 Q1.2 Q1.3 Q1.4 Q1.5 Q1.6 Q1.7 Q2.1 Q2.2 Q2.3 Q2.4 Q2.5 Q2.6 Q2.7 Q3 Q4 Q5 Q6 Q7 Q8.1 N Obs. Canada 3.37 3.15 3.47 1.17 1.61 3.15 3.21 3.53 3.72 3.06 3.15 3.03 4.46 2.74 2.53 0.56 0.68 0.5 0.49 2.07 199 Iran 3.1 2.79 3.32 1.98 3 3.79 3.6 3.7 4.09 3.9 3.82 3.81 4.57 3.75 2.7 0.91 0.45 0.72 0.15 2.05 188 p-value 0.021 0.004 0.372 <0.001 <0.001 <0.001 0.009 0.026 <0.001 <0.001 <0.001 <0.001 0.029 <0.001 0.007 <0.001 <0.001 <0.001 <0.001 0.929 Question Q8.2 Q8.3 Q8.4 Q9 Q10 Q11 Q12 Q13 Q14.1 Q14.2 Q14.3 Q15 Q16 Q17.1 Q17.2 Q17.3 Q17.4 Q18 Q19 Canada 1.67 1.41 2.72 0.5 0.49 0.29 0.19 2.31 1.86 2.23 2.17 0.33 1.16 3.34 2.21 2.17 2.59 0.04 0.45 Iran 1.84 1.06 2.91 0.33 1.22 p-value 0.005 <0.001 <0.001 0.248 <0.001 1.89 2.52 2.44 0.444 <0.001 <0.001 3.18 2.06 2.06 2.86 0.86 0.36 0.03 0.129 0.327 0.002 <0.001 0.072 199 188 Table E14: Responses to post-experiment survey questionnaire, by country. 46 F.4 Experiment procedure in Urmia and Tehran Our experiments in Iran were conducted with collaboration of the Moaser Research Center16 which possesses a permit from the Ministry of Science, Research and Technology to conduct research in Economics and Management. The center took the full responsibility of planning, ethical review and official approvals to run experiments in Iran. We also had ethics approval for the Iranian experiment from Simon Fraser University’s Office of Research Ethics. Kin treatments in Urmia Recruiting: We recruited from undergraduate students who were registered in summer semester courses at various colleges and universities in Urmia. We hired local research assistants (a teacher and some students) to help us with the recruiting for kin treatments. Our research assistants went to different colleges and universities throughout the city and invited students to participate in the experiments and earn cash. Those who were interested to participate in the experiment were given forms titled “preresearch questionnaire” in Persian, translated in Appendix F.2. Based on their self-reported background, we chose subjects who were i) students, ii) less than 30 years old, iii) Azeri or Kurdish. Then our assistants contacted and asked them to choose one of the scheduled experiment sessions to attend. Those who had siblings interested in participating in the experiment (answered question 7 with “Yes”) and whose siblings were i) students and ii) less than 30 years old, were invited in pairs to play in the roles of two players with kin relation in our kin treatments. Subjects who participated alone played the stranger role in our kin treatments. Participants: In total 180 subjects participated in the kin treatments of our experiment (60 participants for each kin treatment; KKS, KSK, SKK). From 120 subjects in the roles of Person A and B, half were Azeri Turk and Half were Kurd. 57 of these subjects were female and 63 of them were male. Recruiting student siblings below 30 years old during our two months conducting experiments in Urmia was achievable for some reasons; (i) in Iran, most of the undergraduate students spend their summers with their families in their hometowns, because most of the universities and colleges in Iran do not offer dormitories to undergraduate students during summer semesters. Those who want to take summer courses can take general courses in local universities and colleges as a “guest student” in their hometown, (ii) youth mostly live in the same place with their families until they get married, (iii) the age of marriage in Iranian metropolises is relatively high (23-4 for women, 27-8 for men), (iv) the show-up fee (7000 Tomans for each, i.e. 14000 for a sibling) covers costs of commuting to the experiment location from all areas of the city by chartered cab (max 5000 Tomans for a one-way trip, i.e. 10000 Tomans for commuting of a sibling pair), (iv) Higher education is a trend in today’s Iran and there are variety of public and private colleges and universities. Many youth between 18-24 in cities are studying for a bachelor or master degree in some college or university. Conducting experiment sessions: We ran the kin treatment sessions in a high school with the official approval of the West Azerbaijan province’s Ministry of Education in June-July 2015. In the high school, we had access to three classrooms with twelve seats in each class. We labeled the classrooms with Persian letters “Aleph”, “Be”, “Jim” (same as A, B, C in English) and labeled seats with numbers 1 to 12. Subjects in the roles of Person A, Person B and Person C were directed to classes “Aleph”, “Be” and “Jim” respectively. Each subject was matched with the other participants seating in the seats with the 16 mem.ac.ir 47 same numbers in the other two classrooms. The number of subjects in each session varied between 12 (4 in each class) and 24 (8 in each class). Sessions lasted around an hour. Experiments were run using pen-and-paper. All forms and booklets were anonymous and include only the class and seat number, except for the consent form and receipt, on which participants printed their full name, signed and dated. Subjects had been asked to bring their National and Student ID Cards. A National ID Card includes full name, the date of birth and the name of father which allowed us to check whether pairs were actually siblings. Upon arrival, we handed subjects forms asking their background information (with the same questions 1-7 of pre-research questionnaire). After subjects filled the background information forms and signed consent forms, we handed them the instruction booklets, which were the same as in Appendix F.1 but translated to Persian, and with converted currency (see payments section). They were not allowed to talk to each other during experiments. They were asked to raise their hand to ask questions, and these were answered privately by experimenters. Two experimenters managed the “Aleph” classroom (subjects in the roles of Person A) and the “Be” classroom (subjects in the roles of Person B) and an assistant attended to the “Jim” classroom with the subjects in the passive role (C). After reading instructions, subjects received a decision booklet in three pages. The first page read (in Persian): Decision booklet Please note that: I. Participants in roles of A and B make choices at the same time. We will match their choices to determine the final outcome. You will see the outcome and your final payoff at the end of the experiment. I I. You might observe some background information from the pre-experiment question– naire about participants in the other roles. page 1 The second page of the decision booklet presented the information to the subjects about the other people in the three-person scenario (their kin and a stranger). About non-kin they saw only their age group, written as “18-30 years old” which was true about all subjects. This information was the same for all kin and ethnic treatments, therefore introduced no noise in the experiment. Below is the second page of decision booklet of player A in the KKS treatment who has a sister in the role of player B. The third page displayed the game tree, a question asking which action the player would like to take, and a box asking subjects to explain their choice. Below is the third page as seen by Person B in a KKS treatment. 48 Your Role: Person A The information provided to you and other participants: Person C’s background information Person B’s background information Your sister 18-30 years old Person B observes the same type of Person C observes the same type of information about you. information about you. page 2 Your Role: Person B Suppose Person A decided to Transfer 40 ECU to you. Please choose one of the following options: ◦ Accept, then Right ◦ Accept, then Left ◦ Reject page 3 49 After collecting the decision booklets, we distributed post-experiment questionnaires. While subjects were answering questionnaires, we matched the decisions of players in roles A and B and wrote down their payoffs. Subjects remained seated until they got paid with cash in an envelope. The collected data was converted to a computer file after experiment sessions, and all the forms and booklets filled by subjects during experiments are stored. Payments: In the kin level experiment, the minimum total payment was 13000 Tomans, the maximum total payment was 23000 Tomans and average payment to 180 subjects was 19000 Tomans which is equal to about 7 CAD based on free market rates. Ethnic treatments in Urmia We ran our ethnic treatments in the Faculty of Literature and Humanities of Urmia University in August 2015 with an approval from university officials. Urmia University is a public university and the oldest and largest university in the city. In summer 2015, the university offered general courses and also a range of courses in different fields of study (such as engineering and economics), but all classes were concentrated in the Faculty of Literature and Humanities. Participants: The total number of subjects in the ethnic treatments was 180 (60 participants for each ethnic treatment; CCS, CSC, SCC). All participants were undergraduate students (except few medical students taking general courses) and were below 30 years old. From 120 subjects in the roles of Person A and B, 80 were Azeri Turk and 40 were Kurd. 46 of these subjects were female and 74 of them were male. Recruiting, conducting experiment and payments: We had access to three classrooms with 20 seats each in the Faculty of Literature and Humanities. As in the kin treatment sessions, we labeled the rooms (“Aleph”, “Be” and “Jim”) and seats (from 1 to 20), and subjects played the three-person scenario with two other subjects sitting in the other classrooms in seats with the same numbers. The number of subjects in each session varied between 30 (10 in each class) and 60 (20 in each class). Each session lasted about an hour. To recruit from students in the Faculty of Literature and Humanities, we prepared a form including questions 1-7 of the pre-research questionnaire used in the kin treatment sessions. We printed 60 of these forms and wrote a number (1 to 60) in two corners of each form. One corner of the form with a number on it was cut and could be tear apart easily. At the beginning of the classes, we asked instructors to give us 5 minutes to invite students to our experiment. After a brief introduction, we handed them the forms and told them to tear apart the number in the corner of the form in order to participate in the experiment right after their class and provided them with the room number of class “Jim”. We had 1.5 hour to assign roles and seat numbers to the numbers on pre-experiment questionnaire forms. After students were dismissed from their classes, they started to show up and were directed to their seats based on the numbers in their hands. Payments were the same as in the kin treatments, with an average payment of about 19111 Tomans. Experiments were conducted exactly the same as the kin treatments except that the second page of the decision booklets presented the information of a co-ethnic and a stranger player. Below is the second page of decision booklet for Person A in CCS treatment. 50 Your Role: Person A The information provided to you and other participants: Person C’s background information Person B’s background information 18-30 years old Azeri Turk 18-30 years old Person B observes the same type of Person C observes the same type of information about you. information about you. page 2 Additional experiments in Tehran For the robustness checks in Iran, we ran experiments in Tehran with strangers in both low and high effort cost variants, and with siblings, cousins and friends in the high cost variant. Tehran is the capital city of Iran and is highly ethnically diverse. The results of the experiments with strangers and siblings are not distinguishable from those in Urmia which indicates that our results are good representative of the urban society in Iran. The game tree below was presented to subjects in the high cost variant of the experiment. The experiments in Tehran took place in August 2016 prior to a meeting of the Tehran Thought Club. The club hosts young adults (mostly students) interested in social sciences for various presentations and discussions. We collected the data from 12 triplets per treatment in our experiments in Tehran with almost 51 the same procedure in Urmia except that in pre-experiment questionnaire we added choices to participate in the experiment with a close friend or a cousin. We specified that those who participate in the experiment as close friends should not be relatives. References A KRAMI , S. M. and O SATI , Z. (2007). Is consanguineous marriage religiously encouraged? islamic and iranian considerations. Journal of Biosocial Science, 39 (02), 313–316. A LESINA , A., A LGAN , Y., C AHUC , P. and G IULIANO , P. (2015). Family values and the regulation of labor. Journal of the European Economic Association. —, D EVLEESCHAUWER , A., E ASTERLY, W., K URLAT, S. and WACZIARG , R. (2003). Fractionalization. Journal of Economic growth, 8 (2), 155–194. — and G IULIANO , P. (2010). The power of the family. Journal of Economic growth, 15 (2), 93–125. — and — (2014). Family ties. In P. Aghion and S. N. Durlauf (eds.), Handbook of Economic Growth, vol. 2A, North Holland: Netherlands, pp. 177–215. —, — and N UNN , N. (2013). On the origins of gender roles: Women and the plough*. The Quarterly Journal of Economics, p. qjt005. A MERICAN A SSOCIATION OF L AW L IBRARIES , A. and F LORES , A. A. (1989). Foreign law: Current sources of codes and basic legislation in jurisdictions of the world. Littleton, Colo.: FB Rothman. A SHRAF, Q. and G ALOR , O. (2013). The’out of africa’hypothesis, human genetic diversity, and comparative economic development. The American Economic Review, 103 (1), 1. A UGUSTINE , S. (1998). The City of God Against the Pagans (trans: R. W. Dyson). Cambridge University Press. B ANFIELD , E. C. (1958). The moral basis of a backward society. New York: Glencol, 3, 4. B ARRETT, D. B. (1982). World Christian encyclopedia: a comparative study of churches and religions in the modern world, AD 1900-2000. Oxford University Press, USA. B ITTLES , A. and B LACK , M. H. (2015). Global patterns & tables of consanguinity. B ITTLES , A. H. (2012). Consanguinity in context, vol. 63. Cambridge University Press. B LACKBURN , K., N EANIDIS , K., R ANA , M. P. et al. (2015). Organized Crime, Corruption and Growth: Theory and Evidence. Tech. rep., Economics, The Univeristy of Manchester. B RUK , S. and A PENCHENKO , V. (eds.) (1964). Atlas Narodov. Moscow: Miklukho-Maklai Ethnological Institute at the Department of Geodesy and Cartography of the State Geological Committee of the Soviet Union. C AVALLI -S FORZA , L. L., M ENOZZI , P. and P IAZZA , A. (1994). The history and geography of human genes. Princeton university press. —, M ORONI , A. and Z EI , G. (2004). Consanguinity, Inbreeding, and Genetic Drift in Italy. No. 39 in Monographs in Population Biology, Princeton University Press: Princeton. C HARRON , N., D IJKSTRA , L. and L APUENTE , V. (2014). Regional governance matters: quality of government within european union member states. Regional Studies, 48 (1), 68–90. 52 C HRISTAKIS , N. A. and F OWLER , J. H. (2014). Friendship and natural selection. Proceedings of the National Academy of Sciences, 111 (Supplement 3), 10796–10801. CIA, E. (1996). The world factbook (published online). Central Intelligence Agency, Washington, DC. C OURBAGE , Y. and T ODD , E. (2014). A convergence of civilizations: The transformation of Muslim societies around the world. Columbia University Press. D EL M ONTE , A. and PAPAGNI , E. (2001). Public expenditure, corruption, and economic growth: the case of italy. European Journal of Political Economy, 17 (1), 1–16. — and — (2007). The determinants of corruption in italy: Regional panel data analysis. European Journal of Political Economy, 23 (2), 379–396. D I L IBERTO , A. and S IDERI , M. (2015). Past dominations, current institutions and the italian regional economic performance. European Journal of Political Economy, 38, 12–41. D URANTON , G., R ODR ÍGUEZ -P OSE , A. and S ANDALL , R. (2009). Family types and the persistence of regional disparities in europe. Economic Geography, 85 (1), 23–47. E ASTERLY, W. and L EVINE , R. (1997). Africa’s growth tragedy: policies and ethnic divisions. The Quarterly Journal of Economics, pp. 1203–1250. F IORINO , N., G ALLI , E. and P ETRARCA , I. (2012). Corruption and growth: Evidence from the italian regions. European Journal of Government and Economics, 1 (2), 126–144. F ISMAN , R. and M IGUEL , E. (2007). Corruption, norms, and legal enforcement: Evidence from diplomatic parking tickets. Journal of Political economy, 115 (6), 1020–1048. G ÄCHTER , S. and S CHULZ , J. F. (2016). Intrinsic honesty and the prevalence of rule violations across societies. Nature, 531 (7595), 496–499. G ALASSO , V. and P ROFETA , P. (2011). When the state mirrors the family: the design of pension systems, cEPR Discussion Paper No. DP8723. G ALLUP, J. L., S ACHS , J. D. and M ELLINGER , A. D. (1999). Geography and economic development. International Regional Science Review, 22 (2), 179–232. G ARDNER , A. and S TUART, A. (2006). Spite. Current Biology, 16 (17), R662–R664. G ILLESPIE , J. (2004). Population genetics: a concise guide. Johns Hopkins University Press. G IULIANO , P. and N UNN , N. (2013). Ancestral characteristics of modern populations. G OLDEN , M. A. and P ICCI , L. (2005). Proposal for a new measure of corruption, illustrated with italian data. Economics & Politics, 17 (1), 37–75. G OODY, J. (1983). The development of the family and marriage in Europe. Cambridge University Press. G RAHAM , J., H AIDT, J. and N OSEK , B. (2008). The Moral Foundations Quiz. www.yourmorals.org. G RAY, J. P. (1999). A corrected ethnographic atlas. World Cultures, 10 (1), 24–85. G REIF, A. (2006). Family structure, institutions, and growth: the origins and implications of western corporations. The American economic review, pp. 308–312. G UNNEMARK , E. (1991). Countries, peoples, and their languages: The geolinguistic handbook. summer institute of linguistics. Inc., Dallas, Texas. 53 H AMILTON , W. D. (1975). Innate social aptitudes of man: an approach from evolutionary genetics. Biosocial anthropology, 133, 155. H ARPENDING , H. C. (2002). Kinship and population subdivision. Population and Environment, 24 (2), 141–147. H ARRIS , I., J ONES , P., O SBORN , T. and L ISTER , D. (2014). Updated high-resolution grids of monthly climatic observations–the cru ts3. 10 dataset. International Journal of Climatology, 34 (3), 623–642. H YDE , J. K. (1973). Society and politics in medieval Italy: the evolution of the civil life, 1000-1350. Not Avail. K IRBY, K. R., G RAY, R. D., G REENHILL , S. J., J ORDAN , F. M., G OMES -N G , S., B IBIKO , H.-J., B LASI , D. E., B OTERO , C. A., B OWERN , C., E MBER , C. R. et al. (2016). D-place: A global database of cultural, linguistic and environmental diversity. PLoS One, 11 (7), e0158391. L A P ORTA , R., L OPEZ - DE S ILANES , F., S HLEIFER , A. and V ISHNY, R. (1999). The quality of government. Journal of Law, Economics, and organization, 15 (1), 222–279. —, S HLEIFER , A. and V ISHNY, R. (1998). Law and finance. Journal of Political Economy, 106, 1113–1155. M ICHALOPOULOS , S. (2012). The origins of ethnolinguistic diversity. The American Economic Review, 102 (4), 1508. M ULLER , S. H. (1964). The World’s Living Languages: Basic Facts of Their Structure, Kinship, Location, and Number of Speakers. Ungar. M URDOCK , G. P. (1957). World ethnographic sample. American Anthropologist, 59 (4), 664–687. — (1967). Ethnographic Atlas. University of Pittsburgh Press. N EW, M., L ISTER , D., H ULME , M. and M AKIN , I. (2002). A high-resolution data set of surface climate over global land areas. Climate Research, 21 (1), 1–25. N ORDHAUS , W. D. (2006). Geography and macroeconomics: New data and new findings. Proceedings of the National Academy of Sciences of the United States of America, 103 (10), 3510–3517. N UNN , N. and P UGA , D. (2012). Ruggedness: The blessing of bad geography in africa. Review of Economics and Statistics, 94 (1), 20–36. P UTNAM , R. D., L EONARDI , R. and N ANETTI , R. Y. (1994). Making Democracy Work: Civic Traditions in Modern Italy. Princeton University Press: Princeton. R AMACHANDRAN , S., D ESHPANDE , O., R OSEMAN , C. C., R OSENBERG , N. A., F ELDMAN , M. W. and C AVALLI S FORZA , L. L. (2005). Support from the relationship of genetic and geographic distance in human populations for a serial founder effect originating in africa. Proceedings of the National Academy of Sciences of the United States of America, 102 (44), 15942–15947. R AMANKUTTY, N., F OLEY, J. A., N ORMAN , J. and M C S WEENEY, K. (2002a). The global distribution of cultivable lands: current patterns and sensitivity to possible climate change. Global Ecology and Biogeography, 11 (5), 377–392. —, —, — and — (2002b). The global distribution of cultivable lands: current patterns and sensitivity to possible climate change. Global Ecology and biogeography, 11 (5), 377–392. R OBERTS , J. (1962). Sociocultural change and communication problems. Study of the role of second languages in Asia, Africa, and Latin America, pp. 105–23. S POLAORE , E. and WACZIARG , R. (2009). The diffusion of development. The Quarterly Journal of Economics, 124 (2), 469–529. T ODD , E. (1990). L’invention de l’Europe [The invention of Europe]. Seuil. 54 UN (1995). Demographic Yearbook. United Nations: New York. W OODS , C. G., C OX , J., S PRINGELL , K., H AMPSHIRE , D. J., M OHAMED , M. D., M C K IBBIN , M., S TERN , R., R AYMOND , F. L., S ANDFORD , R., S HARIF, S. M. et al. (2006). Quantification of homozygosity in consanguineous individuals with autosomal recessive disease. The American Journal of Human Genetics, 78 (5), 889–896. W ORLDMARK (1995). Worldmark encyclopedia of the nations. Gale Cengage. 55