The Origins and Consequences of Kin Networks and Marriage Practices

Home , Andrey Korotayev, Cousin marriage, Daria Khaltourina, Endogamy, Matrilateral, Mosuo

The origins and consequences of kin networks and marriage practices

by Duman Bahramirad

M.Sc., University of Tehran, 2007 B.Sc., University of Tehran, 2005

Thesis Submitted in Partial Fulﬁllment of the Requirements for the Degree of Doctor of Philosophy

in the Department of Economics Faculty of Arts and Social Sciences

c Duman Bahramirad 2018 SIMON FRASER UNIVERSITY Summer 2018

Copyright in this work rests with the author. Please ensure that any reproduction or re-use is done in accordance with the relevant national copyright legislation. Approval

Name: Duman Bahramirad

Degree: Doctor of Philosophy (Economics)

Title: The origins and consequences of kin networks and marriage practices

Examining Committee: Chair: Nicolas Schmitt Professor

Gregory K. Dow Senior Supervisor Professor

Alexander K. Karaivanov Supervisor Professor

Erik O. Kimbrough Supervisor Associate Professor Argyros School of Business and Economics Chapman University

Simon D. Woodcock Supervisor Associate Professor

Chris Bidner Internal Examiner Associate Professor

Siwan Anderson External Examiner Professor Vancouver School of Economics University of British Columbia

Date Defended: July 31, 2018

ii Ethics Statement

iii

iii Abstract

In the first chapter, I investigate a potential channel to explain the heterogeneity of kin networks across societies. I argue and test the hypothesis that female inheritance has historically had a positive effect on in-marriage and a negative effect on female premarital relations and economic participation. In the second chapter, my co-authors and I provide evidence on the positive association of in-marriage and corruption. We also test the effect of family ties on nepotism in a bribery experiment. The third chapter presents my second joint paper on the consequences of kin networks. Taking a bigger-picture approach, we define a kinship intensity index based on basic elements of kinship systems such as marriage practices, residence patterns, and lineage organizations. Combining data on 20 psychological outcomes, we show that a significant portion of the existing psychological variation around the world originates in the differences of kin networks. Using historical measures of Church exposure, we also show that the variation in these differences arose historically from the Catholic Church’s marriage and family policies. Keywords: Female inheritance; Kin networks; Gender inequality; Cousin marriage; Endogamy; Corruption; Individualism

iv Dedication

To my wonderful wife, Mahsa.

v Acknowledgements

I would like to thank Greg Dow, Alexander Karaivanov, Erik Kimbrough, and Simon Woodcock for their guidance and support.

vi Table of Contents

Approval ii

Ethics Statement iii

Abstract iv

Dedication v

Acknowledgements vi

Table of Contents vii

List of Tables ix

List of Figures x

1 Introduction 1

2 Keeping It in the Family: Female Inheritance, In-marriage, and the Status of Women 3 2.1 Introduction ...... 3 2.2 Inheritance system as a determinant of kinship pattern ...... 7 2.2.1 Classiﬁcation of inheritance systems ...... 8 2.2.2 Origins and persistence of inheritance systems ...... 9 2.2.3 Patterns of inmarriage and female economic participation ...... 12 2.3 Conceptual framework ...... 13 2.3.1 Female inheritance and inmarriage ...... 15 2.3.2 Female inheritance and the status of women ...... 18 2.4 Empirical strategies and results using historical data ...... 20 2.4.1 Pre-industrial societies ...... 21 2.4.2 Mid-twentieth-century Italy ...... 25 2.5 Empirical strategies and results using contemporary data from developing countries 27 2.5.1 Individual-level analysis using Indonesian data ...... 28 2.5.2 Difference in differences analysis using Hindu Succession (Amendment) Act of India ...... 33

vii 2.6 Conclusion ...... 38

Bibliography Keeping It in the Family: Female Inheritance, In-marriage, and the Status of Women 40

3 Kinship, Fractionalization and Corruption 56 3.1 Introduction ...... 56 3.2 Theory and hypotheses ...... 63 3.2.1 A basic model of bribery ...... 63 3.2.2 Inclusive ﬁtness, kin altruism, and corruption ...... 65 3.2.3 Consanguineous marriage, sub-ethnic fractionalization, and corruption . . . 67 3.3 Empirical Strategy and Results ...... 69 3.3.1 Cross-country analysis ...... 69 3.3.2 Within-country analysis (Italy) ...... 75 3.3.3 Laboratory Experiments ...... 81 3.4 Conclusion ...... 88

Bibliography Kinship, Fractionalization and Corruption 89

4 Origins of WEIRD Psychology 105 4.1 Theory ...... 106 4.2 Methods ...... 108 4.3 Results ...... 112 4.3.1 Psychological variation across countries ...... 113 4.3.2 Psychological Variation within Europe ...... 119 4.3.3 The children of immigrants ...... 123 4.4 Discussion ...... 124

Bibliography Origins of WEIRD psychology 125

Appendix A Keeping It in the Family: Female Inheritance, In-marriage, and the Status of Women 127

Appendix B Kinship, Fractionalization and Corruption 164

Appendix C Origins of Weird Psychology 217

viii List of Tables

Table 2.1 Female inclusion and partibility of inheritance in the EA ...... 21 Table 2.2 Set of control variable used in the regression analyses with the EA data . . . 22 Table 2.3 First set of regression analyses with the EA data ...... 23 Table 2.4 Second set of regression analyses with the EA data ...... 24 Table 2.5 Regression analyses with data on Italian provinces ...... 27 Table 2.6 Ethnic-level regression analyses with IFLS data ...... 30 Table 2.7 Individual-level regression analyses with IFLS data ...... 31 Table 2.8 Diff-in-diff regression analyses with NFHS data ...... 36 Table 2.9 Diff-in-diff regression analyses with NFHS data, counterfactual trends . . . 37

Table 3.1 Misspecification due to omission of sub-ethnic fractionalization ...... 59 Table 3.2 Misspecification due to conflation of ethnic and sub-ethnic fractionalization . 59 Table 3.3 Religious attitudes to consanguineous marriage ...... 68 Table 3.4 Cross-country regression analysis of the relationship between consanguinity and corruption ...... 71 Table 3.5 Cross-country regression analysis of the relationship between consanguinity and corruption: potential confounds...... 73 Table 3.6 Regression analysis of the relationship between consanguinity and corruption in Italy ...... 78 Table 3.7 Regression analysis of the relationship between consanguinity and corruption in Italy controlling for climate and geography...... 78 Table 3.8 Reduced form regressions with active years of archdioceses ...... 79 Table 3.9 Active years of archdioceses as an instrument for consanguinity ...... 80 Table 3.10 Relative frequency of bribery and corruption by treatment ...... 85 Table 3.11 Summary of relative frequency of bribery and corruption ...... 87

Table 4.1 Psychological and behavioral measures ...... 112 Table 4.2 Baseline cross-country regressions for psychological outcomes ...... 119 Table 4.3 Regression of psychological outcomes on exposure to the medieval Western Church ...... 120 Table 4.4 Regression of psychological outcomes for the children of immigrants in Europe123

ix List of Figures

Figure 2.1 Average cousin marriage rates in India ...... 6 Figure 2.2 Worldwide distribution of pre-industrial inheritance systems ...... 9 Figure 2.3 Cousin marriage and female economic participation rates around the world 12 Figure 2.4 Cousin marriage and female economic participation rates in Italy . . . . . 13 Figure 2.5 Female inheritance and cousin marriage ...... 17 Figure 2.6 Cousin marriage rates among Hindu women across Indian states...... 33

Figure 3.1 A bribery game between strangers and relatives with inclusive ﬁtness . . . 64 Figure 3.2 Corruption and consanguinity around the world ...... 70 Figure 3.3 Corruption and consanguinity in Italy ...... 77 Figure 3.4 Bribery game in the experiment ...... 82

Figure 4.1 Kinship Intensity Index for ethno-linguistic populations around the globe . 109 Figure 4.2 Western Church exposure across European regions ...... 110 Figure 4.3 Individualism and independence, kinship intensity and Church exposure . 114 Figure 4.4 Conformity and obedience, kinship intensity and Church exposure . . . . 115 Figure 4.5 Impartiality, kinship intensity and Church exposure ...... 116 Figure 4.6 Impersonal cooperation and trust, kinship intensity and Church exposure . 117 Figure 4.7 Relationships between regional estimates of cousin marriage and psychological measures ...... 122

x Chapter 1

Introduction

The thesis includes three chapters on the origins and consequences of kin networks, marriage practices, and gender norms. In the first chapter, I study historical origins of existing differences in marriage practices and gender norms across societies. I argue that in patrilineal societies that man- date female inheritance, such as Islamic societies, cousin marriage and endogamy have developed to preserve property in the male lineage, prevent land fragmentation, and limit conflicts over inheritance. In these societies, female inheritance has also encouraged restrictions on women’s premarital sexual freedom in order to arrange cousin marriages and avoid out-of-wedlock children as potential heirs. These restrictions, such as through veiling and segregation from men, are incompatible with female participation in agriculture and have further influenced the historical gender division of labor. Using data on pre-industrial societies, Italian provinces, and Indonesian individuals, I find evidence consistent with these hypotheses: female inheritance is associated with higher cousin marriage, endogamy, and arranged marriage, and lower female economic participation and premarital sexual freedom. Finally, I use a 2005 reform of Indian inheritance laws to obtain causal estimates of the effect of female inheritance on the cousin marriage rate and the female premarital sex rate. In the second chapter, along with my co-authors Mahsa Akbari and Erik Kimbrough, I study consequences of marriage practices for quality of formal institutions. By shaping patterns of relatedness and interaction, marriage practices influence the relative returns to norms of nepotism/favoritism versus norms of impartial cooperation. Cousin 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. In the last chapter, along with my co-authors Jonathan Schulz, Jonathan Beauchamp, and Joseph Henrich, I study consequences of kinship patterns for social psychology. Recent research not only confirms the existence of substantial psychological variation around the globe but also highlights the peculiarity of populations that are Western, Educated, Industrialized, Rich and Democratic

1 (WEIRD). We propose that much of this variation arose as people psychologically adapted to dif- fering kin-based institutions—the set of social norms governing descent, marriage, residence and related domains. We further propose that part of the variation in these institutions arose historically from the Catholic Church’s marriage and family policies, which contributed to the dissolution of Europe’s traditional kin-based institutions, leading eventually to the predominance of nuclear families and impersonal institutions. By combining data on 20 psychological outcomes with historical measures of both kinship and Church exposure, we ﬁnd support for these ideas in a comprehensive array of analyses across countries, among European regions and between individuals with different cultural backgrounds.

2 Chapter 2

Keeping It in the Family: Female Inheritance, In-marriage, and the Status of Women

“This is what the LORD commands . . . Every daughter who inherits land in any Israelite tribe must marry someone in her father’s tribal clan, so that every Israelite will possess the inheritance of his fathers." (The Bible, Numbers 36)

2.1 Introduction

Modern economic literature widely assumes that female inheritance empowers women, increases their autonomy, and promotes gender equality. This is true in the modern world, where women’s inheritance rights enable them to control and exploit their property and therefore improve their social and economic prospects. However, in a patrilineal traditional society, female inheritance transmits property through women, not to women. Here, women function mainly as carriers of property from father to husband, and on to children, rather than as active managers of wealth [113, 157, 159, 147]. Therefore, under female inheritance, a woman’s marriage determines whom her family will have to share their land or herds with. Thus, her male relatives have the incentive to arrange her marriage to a cousin in order to keep her share of property among themselves. Arranging her marriage also requires controlling her premarital romantic and sexual relationships. These arrangements and controls are likely to negatively impact the woman’s role and participation in society. Despite inheritance being an important economic institution, the above paradigm from the social sciences has received little attention from economists. I attempt to fill this gap in the field. I propose that under female inheritance, patrilineal societies encourage inmarriage—in the form of cousin marriage or endogamy (marriage within the limits of a local community such as a clan or village)— and control over women’s sexuality to keep property within the male lineage, prevent property fragmentation, and limit conflicting claims on the estate. Using different data sources, I confirm that female inheritance is associated with more cousin marriage, endogamy and arranged marriage, as

3 well as with less female economic participation and premarital sexual freedom. I also provide causal evidence that female inheritance has a positive effect on the cousin marriage rate and a negative effect on the female premarital sex rate. I present a conceptual framework to analyze inmarriage and the status of women in both patrilineal pre-industrial societies and contemporary developing countries. In this framework, the objective of a family head in a patrilineal society is to preserve the patrilineal succession, that is, to minimize the probability of patrilineal extinction in order to perpetuate his male lineage. At the same time, the production technology exhibits increasing returns to scale up to a point, due to the size of the land required for plowing and harrowing, or to maintain a base of local political power (such as a feudal estate).1 I develop the conceptual framework based on three assumptions. First, I take patrilineal bias—in succession of names, rights, titles, properties, and so on—as given based on the fact that patrilineal systems have historically been widespread [103]. As Adam Smith noted, patrilineal systems incor- porate the patrilineal bias in order to preserve male lineages [236, 60] as their basic social units. Sec- ond, capital markets are imperfect, and most land is acquired through inheritance as a non-market mechanism of land transmission between close kin. This assumption characterizes pre-industrial agricultural societies and many contemporary developing countries [60, 213, 36, 152]. Third, I assume female inheritance was introduced exogenously to some patrilineal societies but not others. As I discuss in section 2.2, the literature suggests that the differences in traditional inheritance systems across societies are deep-rooted in geographical characteristics and the subsistence economy of the regions. Moreover, religions and legal systems also contributed to the persistence of inheritance systems throughout history. For example, Muslims have had to follow Islamic inheritance law because it is stated explicitly in the text of the Qur’an. The centuries-long legal recognition of entails in many European countries is another example. By an entail, the founder of an estate could prevent its sale or division by an indefinite and automatic chain of succession specified at his will. Because of its deep-rooted origins and its persistence through religious or secular laws or traditions, following the literature, we can reasonably consider inheritance an exogenous determinant of more flexible social institutions such as marriage practices and gender norms. Using insights from anthropological studies such as [113], I propose two hypotheses. The first hypothesis is that when women are included in inheritance, inmarriage is more frequent. Several possible mechanisms are involved. Cousin marriage keeps the land within the male lineage and therefore decreases the probability of patrilineal extinction. Inmarriage also provides a way to decrease property fragmentation: a man can marry a woman within the kin or community to pool land parcels and capital goods, which decreases land and capital fragmentation. Finally, cousin marriage decreases rivalry and conflict between siblings over inheritance by creating overlapping interests

1I base my arguments on inheritance of land only, both for the sake of clarity and because land was historically the most important form of property, factor of production, and source of wealth. However, my arguments and hypotheses may well apply to other property such as herds or commercial property.

4 among their offspring; a sister’s offspring will benefit from the property of a brother when it passes to his offspring. The second hypothesis is that under female inheritance, more restrictions are imposed on women’s premarital sexual behavior and their economic participation. Two potential mechanisms are involved. The first mechanism is through inmarriage. Female inheritance promotes cousin and endogamous marriages, which are usually arranged by parents. Arranging marriage requires controlling premarital courtship and sexual relationships through gender segregation and restrictions on contact between opposite sexes. Under these restrictions, young people are less likely to form romantic attachments with outsiders, which makes it easier to arrange marriage with an intended cousin. Fur- ther, under such restrictions, cousins are among the few people of the opposite sex whom young people are likely to meet and form a romantic relationship with. However, restrictions on contact between opposite sexes tend to disadvantage women more than men, because women’s sexual behavior can more easily be screened (through virginity and unwanted pregnancy). Gender segregation is also likely to lead to women being secluded at home and wearing the veil, which are incompatible with participation in agriculture and may intensify the pre-existing division of labor within the agricultural family. The second mechanism is through maternity certainty.2 Female sexual freedom is a threat to a family’s property because out-of-wedlock children are connected to the family due to maternity certainty and therefore are considered potential heirs. Avoiding illegitimate children by requiring women’s virginity before marriage protects the kin group from dispersion of property by illegitimate children and also limits the possibility of conflicting claims on the estate in which a woman has rights. In the empirical section, I use three different data sets (ethnicity-level data from the Ethno- graphic Atlas, province-level data from Italy, and individual-level data from Indonesia) to test the correlations predicted by the hypotheses. I also use individual-level data from India to provide evidence for the causal relationships predicted by the hypotheses. Regression analysis on the Ethnographic Atlas data confirms that female inheritance is associated with lower female participation in agriculture and higher cousin marriage, endogamy, and female premarital sex prohibitions. The reduced female participation in agriculture associated with female inheritance is of a comparable magnitude to that of plow agriculture. This is an important finding considering the famous study by [16] on the negative impact of plow agriculture on women’s participation in agriculture. As they argued, such a negative impact may carry over to beliefs about the role and participation of women in society generally. I also test the effects of traditional inheritance systems in Italy on cousin marriage and female economic participation rates. The regression analyses show that Italian provinces with an egalitarian inheritance system that included women in inheritance have had higher cousin marriage rates and lower female economic participation rates.

2It is always known who a child’s mother is, and the mother knows her children, since she produces them.

5 I also examine contemporary data from developing countries. Using the Indonesia Family Life Survey (IFLS) and the ethnographic data on inheritance practices of Indonesian ethnicities, I show that females of ethnicities that traditionally included women in inheritance are more likely to marry endogamously within the village, are more likely to be in arranged marriages, and are less likely to be self-employed. I also find that Indonesian women who inherit property—even after marriage— are more likely to be in endogamous and arranged marriages. This effect is obtained from a regression analysis controlling for individual characteristics and village-level, religion, and ethnicity fixed effects. This shows not only that different traditional inheritance systems have different consequences for marriage practices and social norms across societies—as established in previous empirical strategies—but also that different expectations of inheritance may create different marriage outcomes for individuals living in the same society. In my last empirical strategy, I use the amendment of the Hindu Succession Act in 2005—which substantially improved Hindu women’s inheritance rights on land—in a difference in differences approach to provide evidence of the causal impact of female inheritance on inmarriage and the status of women. The Hindu Succession Act applies only to Hindus (including Sikhs, Jains, and Buddhists), and explicitly exempts Muslims, Christians, Parsis, and Jews. For this analysis, I use data from the Indian National Family Health Survey (NFHS). I show that the cousin marriage rate was significantly higher and the premarital sex rate was significantly lower in the treated group, Hindu women married in or after 2005. I also provide tests to confirm the identical counterfactual trends of the two outcome variables for treatment and control groups (see Figure 2.1 for a visual inspection).

Figure 2.1: The average cousin marriage rate of women married in the five-year period (2000–2004) before 2005 amendment of the Hindu Succession Act compared with those married after the amendment increased significantly among Hindu women (from 7.93% to 8.72%, Wald test p-value<0.001), while it did not change significantly among Muslims and Christians (Wald test p-value=0.417 and 0.371 respectively).

These ﬁndings contribute to our knowledge of marriage and the status of young women in developing countries and highlight a potential unintended consequence of exogenously introduced policies to improve female inheritance in these countries. They also suggest that female inheritance affects marriage practices and the status of women even in the short term. However, this short- term mechanism has operated for a long time and for a larger share of the population in some

6 patrilineal societies that traditionally practiced female inheritance. The long-term practice of female inheritance in a patrilineal society may create persistent cultural traits and beliefs regarding marriage and the status of women that affect people—even those who do not receive inheritance—in an era in which, through industrialization, inheritance is no longer the only source of wealth and means of production. This has an important implication for the evolution of norms encouraging inmarriage and seclusion of women in Islamic societies where Sharia has mandated female inheritance. In the Qur’an, there is no speciﬁc guidance that encourages cousin marriage [43] and no explicit prescription on the veiling of women [5]. But, of all the economic rules in the Qur’an, the most detailed are those of inheritance [163]. The Qur’an, the main source of Islamic law, explicitly states the Islamic inheritance rules in such detail (the Qur’an 4:11) that no space is left for different interpretations regarding female inheritance. Islamic religious authorities often paid great attention to the observance of female inheritance, while similar legal rights for women did not exist in the West until the nineteenth century [159]. Islam, in fact, may be the only religion that formally speciﬁes women’s inheritance rights.3 In line with my arguments in the hypotheses, this may explain why cousin marriage (with mean 32%), gender segregation, seclusion of women inside homes, and the veiling of women are most common, and female economic participation (with mean 27%) is lowest, in the Middle East and North Africa (see Figure 2.3). The paper is organized as follows. Section 2.2 reviews the literature, introduces different inheritance systems, and discusses their origins and persistence. Section 2.3 develops the conceptual framework for the effect of female inheritance on inmarriage and the status of women. Section 2.4 and 2.5 describe the empirical strategies and present results using historical data, and contemporary data from developing countries respectively. Finally, section 2.6 presents my conclusions.

2.2 Inheritance system as a determinant of kinship pattern

Social scientists discuss marriage practices and the status of women in the larger context of kinship patterns, for which inheritance is considered an important determinant.4 Anthropological studies clearly emphasize that inheritance can affect marriage patterns, residence arrangements, family structures, patriarchy, courtship and sex, kinship terminology, and so on.5 To understand why the link between inheritance and kinship is important, we need to consider the institutional environment of pre-industrial societies and many contemporary developing coun-

3For example, according to the Bible, property is to be inherited by sons, and daughters inherit property only in the absence of sons (the Bible, Numbers 27).

4Max Weber perceived a kin group as “a group of expectant heirs" [257, p.365]. Lewis Morgan argued that the family grew out of the development of a knowledge of property and its transmission by inheritance, and that even in the face of other factors, “with more effective power the rights of property might inﬂuence the system of relationship" [189, p.14]. Jack Goody noted that inheritance is an institution “in which interpersonal relationships are structured" [113, p.1]. David Sabean suggested that “there is no system of obligations and duties" that is not mediated through property [225, p.171].

5See, e.g., [189, 76, 40, 116, 181, 182, 237, 232, 110, 159, 140, 234].

7 tries. In pre-industrial agricultural societies, which were characterized by imperfect capital markets [60], land as the basic source of wealth and means of production was universally transmitted between close kin by the process of inheritance [112, 237]. Even today in many developing countries, land sales are rare, and most land is acquired through inheritance as a non-market mechanism [213, 152]. It is not surprising that in such a world, “kinship and property are closely interlocked" [112, p.70]. I follow the literature and consider the inheritance system “as the independent (or exogenous) variable" [113, p.20] to inmarriage and the status of women. In the next subsections, after introduc- ing different inheritance systems, I discuss their origins and persistence, which reﬂect the chain of causation suggested in the literature.

2.2.1 Classiﬁcation of inheritance systems

It is important first to define what I refer to as "inheritance systems". I classify different inheritance systems using combinations of possible modes of property transmission. The first dimension involves lineal versus lateral inheritance systems. In lineal inheritance systems, property is transmitted vertically to children. In lateral inheritance systems, property is transmitted horizontally to siblings or indirectly vertically to siblings’ children. The second dimension involves impartible versus partible inheritance systems. In impartible inheritance systems, a land parcel is preserved intact from generation to generation, and only one lineal or lateral heir inherits property. The examples are primogeniture (inheritance by a senior child, sibling, or sibling’s child) and ultimogeniture (inheritance by the junior child, sibling, or sibling’s child). In partible inheritance systems, the land parcel is not preserved intact. Instead, each parcel is divided up lineally or laterally, among some or all of the children, siblings, or siblings’ children. The third dimension involves female inclusion versus female exclusion in inheritance. To see how different inheritance systems can be characterized by the combinations of these three categories, consider the following historical examples. Primogeniture, in practice carried out by preference given to the senior son—found in Japan, Korea, and Northwest Europe [198, 109, 245, 213, 186, 36]—, can be characterized as lineal and impartible, with female exclusion. The inheritance system of equal division of land property among sons—common in Eastern Europe, Russia, China, and South Asia [109, 213, 156]—can be characterized as lineal and partible, with female exclusion. Dividing inheritance among all sons and daughters—common in Mediterranean Europe, Latin America, and Islamic societies [109, 186, 163]—can be characterized as lineal and partible, with female inclusion. Lateral inheritance system was common in Africa south of the Sahara. In theory, this system could be partible or impartible, and with female inclusion or exclusion. But in practice, lateral inheritance in sub-Saharan Africa excluded women [113]. The mode of property

8 transmission can also be characterized by the absence of any rules of inheritance or any private property rights.6

Figure 2.2: Inheritance systems of 820 Ethnographic Atlas societies, distinguishing impartible and partible inheritance systems, and for the latter, distinguishing female inclusion and female exclusion in inheritance.

It might be helpful at this point to see the global distribution of inheritance systems based on the suggested classiﬁcation. Using data on pre-industrial societies from the Ethnographic Atlas (see section 2.4.1), Figure 2.2 displays traditional inheritance systems around the world. The distribution of inheritance systems in the ﬁgure coincides with the historical examples mentioned above. The Ethnographic Atlas data focuses as much as possible on the characteristics of the sample of societies prior to European contact. Hence it obviously provides scant observations for Western Europe. However, based on detailed studies and data on Europe, we know that impartible inheritance was more common in Northwest Europe than any other region in Eurasia [245, 248, 213, 36, 163]. See also maps from [245] and [248] in Figure A.2.1 of Appendix A.2.

2.2.2 Origins and persistence of inheritance systems

The differences between inheritance systems are thought to be deep-rooted in agricultural and political organization. Capital-intensive (e.g., involving plows), open-ﬁeld, and manorial agriculture might favor impartible inheritance due to economies of scale. However, agricultural organization itself is determined by geographic factors and political organization. For example, heavy soils usually required large plows pulled by several horses, which were expensive and practical only on large

6For example, in many hunting and gathering societies (such as Native American societies) individuals had little property except personal equipment, which was often destroyed at death [113]. Other societies had communal ownership of land, in which individuals inherited from their parents a general right of access to the whole of the community’s resources that continued to exist after the head of the family passed away. Communal inheritance could be patrilineal (from father to sons), such as the Russian peasantry, or matrilineal (from mother to daughters), such as among the Minang in Indonesia.

9 land holdings, while sandy light soils could be cultivated by handheld tools like the hoe and digging stick on small family farms. In terms of political organization, manorial agriculture, for example, was closely linked to feudalism.7 Despite the influence of agriculture, the literature suggests that inheritance systems are best explained by the political organization of societies. In societies with impartible inheritance, such as Japan or Northwest Europe, lands were controlled by powerful nobility whose interests were best served by maintaining their holdings intact through impartible inheritance because the political and military functions associated with the estate were indivisible [236, 213, 36]. Also, in the European countries, large estates came with seats on parliamentary bodies. Therefore, property became indivisible because the office was indivisible [36]. On the other hand, a necessary condition for partible inheritance was a strong central government [18]. In regions such as China, India, Russia, and the Mediterranean, inheritance rules were subject to the legislation of strong central bureaucracies with an interest in restricting the development of powerful landholding families by fragmenting their properties through partible inheritance [258, 109, 213, 163]. But contrary to patrilineal partible inheritance in China, India, and Russia, partible inheritance in the Mediterranean region (including the Middle East) included both sons and daughters. Inclusion of women in inheritance in the Mediterranean region had Roman-Byzantine roots [156]. Again, it seems that geographic factors had a role here. For example, [258] suggests that in regions such as the Middle East, states had centralized power by controlling large-scale irrigation systems essential to the agriculture. [37] provide general evidence on this account. [184, 185] argue and present evidence that a centralized Islamic state featuring redistributive principles such as partible inheritance emerged to address economic inequalities resulting from geographical features of the region—this is, unequal agricultural potential with few fertile places and a large share of arid lands—and their interaction with the diversion of trade routes in seventh-century Arabia. Consid- ering the incompatibility of female inheritance with a patrilineal system, the mere fact that female inheritance has been present within the patrilineal context of Islam and Islamic societies itself suggests that it was most likely introduced to Islam by an exogenous source. Whatever the deep-rooted sources are, once they determine the form of an inheritance system, its development “very much tends to follow the track that has been laid down, and is relatively independent of changing socio-economic conditions" [36, p.82]. One can find a strong continuity and a systematic pattern through all changes [116, 36]. Two important aspects of inheritance contributed to its persistence and path dependency. First, inheritance practices cannot be understood as purely individual decisions. Rather, they are regulated by secular or religious institutions and laws. Inher- itance laws frequently “continue in force long after the circumstances which first gave occasion to them" [236, p.305]. Second, inheritance is a non-market institution [213, 36].

7See [213] for a comprehensive discussion on different agricultural organizations and the involved political and geographic factors.

10 For example, primogeniture was legally recognized through entails8 which were “respected through the greater part of Europe" [236, p.384]. By entail, the testator not only determined the heir, but also decides to whom the land must be bequeathed after the death of the heir. If real property was entailed, it could not be sold by the heir, and it had to be passed on automatically from generation to generation according to the succession determined by the founder. Entails prevented the division of property through sale or inheritance. Therefore, an entailed property was removed from the market process. Aside from enjoying legal recognition for centuries—until 1780 in the United States, 1848 in France, 1919 in Germany, and 1925 in Britain [36]—, supporters of primogeniture collected more than a dozen biblical verses9 to give it a Christian foundation [163]. Under strict manorial controls in Europe, even peasants had no right to divide or alienate the land [213, 156]. On the other hand, inheritance has been subject to the partible Qur’anic inheritance law10 in Is- lamic societies. Islamic inheritance law clearly subordinated personal preferences and strengthened the inheritance rights of women. The law took shape in the Mediterranean Middle East region, in Syria and Iraq, which were already accustomed to partible inheritance practices and inclusion of women. However, by entering into the text of the Qur’an, the law became a path-dependent institution for all Muslim populations. Although in some Muslim populations women were excluded from inheritance, they were still more likely to inherit than their counterparts in Christian societies [49] because the Qur’an left no space for different interpretations regarding female inheritance. Finally, communal and joint property persists in many developing countries and even Europe.11 Under these “archaic regimes" [207], access to land is possible only through membership in a communal assembly or a joint family composed of several generations, and a single member cannot transfer or alienate their membership right. Thus, land sales and partitions are rare [207, 152, 55]. Due to its deep-rooted origins and its persistence, inheritance is a process critical to the reproduction of the social system itself. It is true that the differences in inheritance systems were a marked feature of the pre-industrial era. But, as [116] argued, “whatever the reasons, these differences have consequences for the position of women, the structure of social roles, the behaviour of kin, and the strategies of family organization" (p.35).

8Fideikommiss (in German), Substitutions and Majorats (in French).

9E.g. Isaac’s ﬁrst-born son, Esau, sold his “birthright" to his younger brother, Jacob, for a bowl of stew (The Bible, Genesis 25).

10Verses 11, 12, and 176 in the fourth chapter of the Qur’an, Surah An-Nisa. Islamic inheritance law limits an individual’s power of testamentary disposition to one-third of his estate, and two-thirds of the estate passes to the legal heirs of the deceased under the compulsory rules of inheritance. Legal heirs include children, spouses, parents, and siblings of both sexes. The females among these relatives only take half the share of the male relative of the same degree of relation to the decedent. However, a female must have her ﬁrm share of inheritance in all types of property left by her father.

11See, e.g., [4] for joint property in India and [188] for communal lands in some Italian towns.

11 2.2.3 Patterns of inmarriage and female economic participation

Because the authority of religious institutions over marriage is very common, religious attitudes explain a large share of the variation in inmarriage rates around the world today. Figure 2.3 (left) shows cousin marriage rates up to and including second cousins.12

Figure 2.3: Left: cousin marriage rates (up to and including second cousins) around the world [42]. Right: Labor force participation rate of women above 15 years old, 2010–2016 estimates by the International Labour Organization, with darker colors indicating lower percentages.

Christian countries have the lowest rates of cousin marriage, as they have a long religious history of discouraging this practice. The Catholic Church has prohibited cousin marriage since the sixth century, at times extending the prohibitions as far as third cousins in the eighth century and sixth cousins in the eleventh century.13 The highest cousin marriage rates can be observed in Islamic countries, where pre-Islamic Arab culture and the Islamic Sunnah (the deeds of the Prophet) might have encouraged cousin marriages [67, 43]. Figure 2.3 confirms that religious history has been an important determinant of marriage patterns in terms of cousin marriages. However, to better introduce the first hypothesis of this study, we need to take a look within a religiously homogeneous country. Although the Church banned cousin marriage, it was still possible to marry a cousin by requesting a dispensation letter from a diocese. Figure 3.3 (left) shows the cousin marriage rates up to first cousins14 (based on the number of issued dispensation letters) within an almost 100% Catholic country: Italy from 1945 to 1965. As seen in the figure, cousin marriage rates varied considerably across the Italian provinces. According to my first hypothesis, female

12This also includes marriage between ﬁrst cousins once removed, and uncle-niece/aunt-nephew marriages.

13Prohibitions on cousin marriages were relaxed to third cousins in 1215, second cousins in 1917, and ﬁrst cousins in 1983 [115, 43]. The Church’s family and marriage regulations might have partly formed and evolved in response to inheritance rules. According to [115], an edict of a Roman emperor allowed the Church to possess properties that did not belong to any individual. Therefore, the Church could only beneﬁt by leaving a deceased without eligible heirs. It does not seem accidental that the Church would have condemned polygyny, cousin marriage, remarriage, adoption, and concubinage. By providing heirs, these practices could deprive the Church of property. The Church prohibited cousin marriages within seven degree of kinship because it was the degree considered for the purpose of inheritance by Roman law. Therefore, one could no longer marry anyone from whom one could have inherited. [121] also agrees that the Church’s marriage laws were “self-serving" (p.308).

14This also includes uncle-niece/aunt-nephew marriages.

12 inheritance could be the cause of the residual variation in cousin marriage rates after controlling for the religion (i.e., within the same religious denomination, as in the case of Italy).15

Figure 2.4: Left: Cousin marriage rates within Italy 1945-1965 [57]. Middle: Traditional inheritance systems within Italy [248]. Right: The rate of economically active women above 15 years old, 2011 census, the Italian National Institute of Statistics (ISTAT), with darker colors indicating lower percentages.

Figure 3.3 (middle) displays traditional inheritance systems for Italian provinces. Consistent with the hypothesis, regions in Italy characterized by egalitarian nuclear family type, where women were more likely to receive property, were also historically characterized by high cousin marriage rates.16 The same correlation can also be observed between cousin marriage rates for world regions (Figure 2.3, left) and the distribution of Ethnographic Atlas societies in the regions practicing female inheritance, that is, partible inheritance by both sexes (Figure 2.2). Interestingly, regions of the world that traditionally practiced female inheritance have also experienced lower female economic participation (see Figure 2.3, right). These are mostly Muslim countries also with norms favoring gender segregation, women wearing the veil, and seclusion of women in homes. The pattern also holds for Italy. Southern Italy has traditionally been associated with lower female participation and freedom than has the north17 (see also Figure 3.3, right). This correlation between female inheritance and female economic participation and freedom is a predic- tion of my second hypothesis.

2.3 Conceptual framework

An implicit assumption of studies on inheritance, such as in Adam Smith’s The Wealth of Nations, is an overlapping generations perspective, where the objective of the family head is to preserve the patrilineal succession and perpetuate his male lineage.18 This objective is embedded in the nature of

15As argued by [113], “in Europe, to some extent, the desire for like to marry like runs against the rules of the Catholic church against marrying close kin" (p.14).

16One piece of historical evidence that female inheritance can promote inmarriage is the increased cousin marriage rates among Italians with the introduction of the Napoleonic civil code in 1810, which emphasized equal inheritance rights between men and women. See, e.g., [95] (p.86), [63] (p.154), [57] (p.159).

17See, e.g., [157] (p.18) and [175] (p.51–2).

18Adam [236] suggested that primogeniture was introduced in Europe “to preserve a certain lineal succession [. . . ] and to hinder any part of the original estate from being carried out of the proposed line" (p.384). Similarly, [64] describe

13 patrilineal societies, where male lineages are the basic social and economic units, and the succession of names, lordly rights, titles, and other valuables tends to be passed on in the male line. Patrilineal systems have been historically widespread, such as in Western cultures [103].19 Many religions, specifically Abrahamic religions, offer an extensive demonstration of patrilineal relationships [see, e.g., 246]. Of course, a family head would also like to support all his sons and daughters, and their families. However, in a disorderly time (such as the pre-industrial era) or poor environment (such as in some developing countries), to divide the land among all children is to expose every parcel and family to extinction. On the other hand, leaving large land parcels for few children increases their families’ chances of survival and therefore continuation of the lineage for another generation. Three factors contribute to this trade-off between supporting all children and preserving the lineage: economies of scale, high mortality rates, and imperfect capital markets. First, division of land might decrease the total production of the offspring due to economies of scale in land size. Some forms of agricultural organization require larger land parcels. For example, plowing and harrowing can take place only on large parcels. Moreover, larger land parcels are associated with more political and bargaining power. For example, the size of a feudal estate determined the political and military power of the lord who owned it. Political power, such as a parliamentary seat, and military functions and obligations of lords were indivisible by nature [213, 36]. As Adam [236] noted, “to divide it [i.e. the land] was to ruin it, and to expose every part of it to be oppressed and swallowed up by the incursions of its neighbours" (p.383). Second, since land is the basic means of production, under high mortality rates a larger land parcel provides a family with more production and higher income which means better nutrition, hygiene, and health—in short, a better chance of survival. Third, imperfect capital markets imply rigid intergenerational mobility of income groups. Since land is the main source of wealth, a child who gets a large land parcel would stands a better chance of staying rich or even moving up the social ladder, increasing the family’s chance of survival. Emphasizing the role of high mortality rates and imperfect capital markets, [60] develops an economic model and simulations to show that even in the absence of economies of scale, the optimal strategy of a family head seeking to primogeniture as an attempt for “perpetuation of the family line" (p.176). [198] note that the purpose of primogeniture in Japan is to “keep the succession line firm" (p.11). [157] notes that in Medieval Italy, “the transmission of property through sons was in harmony with the desire to preserve the family estate intact, whereas property given to daughters was thought to be lost from the family of origin" (p.17). [239] reports that not having a son in an Indian villiage “meant the end of the lineage, a disaster that no one liked even to contemplate" (p.112). [3] presents historical examples of the “justifiable fear of extinction" of the family name among English aristocrats and the strategies that were devised to procure continuity, such as transmitting land to men prepared to trade their family names (p.43). In the case of female inheritance, strategies such as cousin marriage are described as the means “to preserve house and holdings in the male line" [63, p.154], to prevent “plots of land from escaping the control of an identifiable patrilineage" [95, p.86], and to “keep parcels of land intact and within families" [234, p.S169], which all appears to be linked to the ideology of the “family line" [232, p.64].

19“A patrilineal bias can coexist with cognatic descent groups and bilateral patterns of afﬁliation" [46, p.19].

14 perpetuate his lineage is primogeniture. However, he also acknowledges that economies of scale alone are enough to rationalize primogeniture.20 Therefore, a family head seeking to protect the welfare of all his children and preserve his lineage might find himself with insufficient resources to fulfill both goals. In this case, sacrificing the welfare of some of his children is the only way to reach the objective of perpetuating his lineage. Since in a patrilineal society “the male sex is universally preferred to the female" [236, p.383] and lineal succession “refers to the preservation of a family name by sons" [60, p.83], sons are preferred to daughters in the succession of land. Patrilineal primogeniture manifests the extent of the sacrifice of families to preserve male lineages: family heads attempt to prevent the extinction of the male lineage by leaving the land intact for only one son, knowing that this “beggars all the rest of the children" [236, p.384]. In imperial China, the whole clan, not just immediate family members, often pooled their money to subsidize the education of just one child, hoping that he would pass the civil service examination, and bring honor and prestige to the clan by becoming an official [60]. The patrilineal bias in succession persists even today in family farms of rural Europe [see, e.g., 211, Ch.5] and many developing countries [see, e.g., 152, for India].

2.3.1 Female inheritance and inmarriage

Bequeathing land to sons only (such as in primogeniture) is not always an option. As discussed earlier, female inheritance is mandated in some patrilineal societies—not to empower women, but most likely to restrict the development of local powers. In this section, I will describe three mechanisms through which inmarriage arises in this specific institutional environment characterized by the intersection of a patrilineal system and female inheritance. First, preserving land within the male lineage. Where female inheritance is mandated, marriage of a granddaughter to an outsider means that her share of land will eventually leave the male lineage. As anthropologists have noted,21 inmarriage provides a way to avoid this problem. If the granddaughter marries a son’s son (a first-cousin marriage), her share of inheritance will still serve to perpetuate the male lineage of the family head by increasing the chance of survival of the grand- son’s family. If there is no eligible partner among the family head’s descendants, marriage of his granddaughter to his brother’s son (a first-cousin-once-removed marriage) or his brother’s son’s son (a second-cousin marriage)—who are carriers of the same family name—could be arranged to keep

20Similarly, in an evolutionary model, [222] shows that under poor environment or positive correlation between earned and inherited wealth, the optimal strategy is to limit the number of heirs and maximize the wealth inherited by them. [135] and [228] argue that heritable wealth has reproductive value independent of number of offspring because it increases the reproductive chances of offspring by improving their quality as measured by status, health, nutrition, and so on. Therefore, a strategy, such as primogeniture, favoring offspring quality may better serve to enhance inclusive ﬁtness in the long run. In other words, inheritance of wealth in humans alters the equilibrium favored by solely genetic inheritance of reproductive strategies.

21See, e.g., [112, 146, 137, 159, 140, 234].

15 the parcels of lands under the family name and within the higher-level segment of the male lineage (all male descendants of the family head’s deceased father). Attempts to keep the land within the larger segments of the lineage (all male descendants of a remote common ancestor) lead to marriage between remote (and even unidentiﬁable) cousins. This creates endogamy within the lineage or clan. Of all ﬁrst-cousin marriages, marriage of a daughter to her father’s brother’s son is the most straightforward union to keep her inheritance within the male lineage. Her marriage to her father’s sister’s son or her maternal cousins will serve her father’s lineage only if those cousins are also members of the male lineage of the father.22 It is not surprising that marriage of a daughter to her father’s brother’s son—or equivalently marriage of a son to his father’s brother’s daughter— is the most preferential form of cousin marriage in Islamic countries, where female inheritance is mandated [146, 159].23 Following the arguments above, cousin marriage might even arise under impartible inheritance (such as primogeniture) or partible inheritance by males only, if daughters inherit property in the absence of male offspring. In his study of cousin marriage in Japan, [229] notes that a family head who has only daughters faces a problem because a wife takes the family name of her husband. “Pride in family name" and “attempts to ensure the perpetuation of the family name" leads him to “select her spouse from among her male relatives having the same family name, in which case the headship does not leave the family" (p.295). [229] presents statistics which show cousin marriage is higher among families with no male offspring. Using data from rural Bangladesh, [234] also report that women without brothers are more likely to be in cousin marriages. This calls to mind the biblical account about female inheritance in the absence of sons (the Bible, Numbers 27), in which the daughter is compelled to marry someone from the same clan as her father (the Bible, Numbers 36). Second, decreasing land fragmentation. Partible inheritance could fragment the land until it is no longer viable because of decreased economies of scale [213].24 Under the constraint of partible inheritance by both sexes, cousin marriage emerges as a solution to decrease land fragmentation. Avoiding fragmentation of land by the means of inmarriage is well documented in the literature.25 Cousin marriage provides the possibility for pooling farms and resources, and the continual recom-

22For example, if the father’s sister herself is married to a ﬁrst cousin within the male lineage, her son is also considered a member of the same male lineage—although not through his mother and as a ﬁrst cousin of the bride, but through his father and as her second cousin.

23Interestingly, it has been suggested that the practice of marrying a father’s brother’s daughter [159] and the veiling of women [5] were introduced into Islamic societies through the contact of Arabs with regions such as Syria and Palestine, the same regions that shaped Islamic inheritance law [163].

24There is supporting evidence in fact that partible inheritance fragments the land and that land fragmentation sacriﬁces economies of scale, for example in nineteenth-century France [36, p.45], and today’s China [200, 255, 242], Indonesia and India [261], and Africa [24].

25See, e.g., [116, 159, 140, 57, 234].

16 bining of portions by adding the claims of the groom and the bride in new conjugal estates. In the case of double cousin marriage—between a brother and sister of one family to a cousin sister and brother of another family—in fact no land changes hands at the marriage, which avoids land fragmentation with zero transaction costs. The same arrangement of exchanging daughters between unrelated but neighbor farmers within the same village leaves both lineages with unfragmented lands [210]. This creates endogamy within the village.

(a) (b) (c) (d)

Figure 2.5: Triangles, circles, lines, and “=" represent males, females, descent bonds, and marriage respectively. The ﬁrst generation starts with two brothers, and thereafter, every family has one son and one daughter. Black triangles and circles represent anyone within the male lineage. Chart (a): Impartible inheritance: primogeniture retains the family land intact, only at the cost of creating many landless offspring. Chart (b): Partible inheritance by both sexes, and outmarriage: female offspring marrying outsiders rapidly fragments the land and also diffuses land parcels out of the male lineage, since thereafter next generations are carrying a different family name. Chart (c): Partible inheritance by both sexes, and double cousin marriage: inmarriage decreases land fragmentation and, contrary to impartible inheritance, does so without leaving landless offspring. Chart (d): Partible inheritance by both sexes, double cousin marriage, and population growth: higher fertility (four sons and four daughters) creates the same land fragmentation in the third generation as the case with outmarriage (chart c), but still all land parcels remain under the family name.

Figure 2.5 shows that compared with outmarriage in chart (b), double cousin marriage in chart (c) keeps land parcels within the male lineage and decreases land fragmentation. To highlight the former mechanism, population growth is added in chart (d). Note that the dimension of inheritance systems relevant for inmarriage is inclusion of women, not partibility. Under partible inheritance by sons only, keeping land parcels within the male lineage is irrelevant, and there is no possibility to decrease land fragmentation by inmarriage. It is also important to recall the assumption of imperfect capital markets. Under this assumption, pooling land parcels (through markets) with landed outsiders is costly. Moreover, outmarriage might be associated with negative externalities for the community, such as the diminished stock of knowledge over time and conﬂict over inheritance (see below). Therefore, a marriage partner with a plot of land within the community is a better candidate to preserve the property of the male lineage than an outsider with the same size and quality of land plot. For the same reasons, other landed communities also prefer inmarriage. Thus, outmarriage most likely attracts landless outsiders and creates

17 immigration pressure [55] such as by non-inheriting children from communities under impartible inheritance [63]. Third, decreasing conflict over inheritance. Cousin marriage also reduces the rivalries and conflicts over inheritance. In fact, the “solution of inheritance fights" was one of the situations that could be used to support a dispensation request from the Catholic Church to marry a relative [57, p.37]. By excluding outsiders, cousin marriage reduces the number of claimants on property and increases biological and cultural ties among those involved in the division of land after the death of the family head. By cousin marriage, men also take advantage of existing relationships. They know each other and have a sense of each other’s personalities and how to work together, which is an advantage for the male lineage as a group of cooperators [234]. In contrast, outmarriage might diminish local skills and the stock of knowledge over time and across generations [41]. Cousin marriage also decreases the potential for conflict among siblings over inheritance by creating overlapping interests and doubly relating them to each other through the young couple and their grandchildren [234]. A sister’s offspring will benefit from a brother’s property when it passes to his offspring. In the case of double cousin marriage, siblings’ conflict over inheritance becomes irrelevant because in the next generation, the land will be reallocated only between their grandchildren (see Figure 2.5, chart (c)). Therefore, where women are legally entitled to a share of an inheritance—such as in Islamic societies—cousin marriage offers a way to deny them inheritance but at the same time keep them content to not take legal action, such as through Sharia courts. Even if a Muslim community disinherits women in practice—despite Islamic law for female inheritance— cousin marriage could at least alleviate the potential for the conflict stemming from the mandatory law.

2.3.2 Female inheritance and the status of women

An important line of research has recently developed to study how contemporary differences in gender norms are determined by various historical factors.26 This study aims to contribute to this literature by highlighting the role of female inheritance as another pre-industrial characteristic in- ﬂuencing gender norms. Some potential mechanisms are as follows. Arranged marriages. As discussed earlier, female inheritance promotes cousin marriage and endogamy. If not controlled, love and premarital sexual relationships might lead to marriage with outsiders because ties of descent are not central in these relationships [137, 187]. Therefore, cousin and endogamous marriages are generally arranged by parents controlling courtship and premarital sexual relationships of their children through gender segregation and restrictions on contact between opposite sexes. Under such restrictions, not only are young people less likely to form romantic relationships, have premarital sex, and have out-of-wedlock children—which makes arranging their

26Such as subsistence technology of agriculture and herding [215, 16, 102, 133, 34], language [227], geography [54], family structures [13, 250, 11], religion [127, 35, 203], historical shocks [203, 125, 264, 244, 52], and pre-industrial societal characteristics such as matrilineality [107, 111, 174], modes of residence after marriage [170], and dowry versus bride price [25]. For more references see [103].

18 marriages easier—but they are also much more likely to meet and thus form romantic attachments to their cousins, who are among the few young people of the opposite sex with whom it is appropriate for them to socialize [113, 234]. Therefore, in cousin marriage, men marry by guarding their sisters’ ’honor’ and staying on the right side of their uncles. Women are reduced to male property and their conformity and clan loyalty are prized [80]. Restrictions on contact between opposite sexes tend to disadvantage women for several reasons. First, sexual behaviors of women can easily be screened (through virginity and unwanted pregnancy) and punished. Therefore, male relatives restricting a woman’s premarital sexuality is a means of arranging a good marriage—one that would allow access to her inheritance. Female premarital sex is usually controlled by encouraging early marriage, stressing virginity at marriage, veiling, and chaperoning post-pubertal daughters [112, 137]. Second, men have higher productivity in traditional agriculture [16, 133]. Therefore, gender segregation likely leads to women’s seclusion and more involvement in activities within the home because men are needed outside the home in the fields. In other words, gender segregation intensifies the pre-existing division of labor within the agricultural family. Moreover, whenever women appear outside the home, restrictions on contact between opposite sexes require them to wear veils, which is also incompatible with strenuous manual work in agriculture. Illegitimate children. Female inheritance also exposes the kin group to another threat from female sexuality: any out-of-wedlock children of female members are undeniably tied to the kin group due to maternity certainty and therefore are considered potential heirs. Thus, maternity certainty creates unequal gender norms regarding premarital sexual contact. Historical evidence confirms that societies gave “the father the option of refusing to acknowledge an illegitimate child [which] was intended to protect the property of the legitimate paternal kin from the claims of illegitimate children" [36, p.102]. Until the early twentieth century, in France, Germany, and the United States, illegitimate children had the right to acknowledgement only in relation to the mother, took their mother’s name, and “were granted inheritance rights from the maternal side of the family much earlier than from the paternal side" [36, p.102]. This is still the law today in many Islamic countries.27 Any children born outside of regulated sexual relationships is to attribute free membership to the lineage [257]. Therefore, controlling women’s sexuality is a means of controlling arbitrary distribution of property. Conflict and fights over inheritance play an important role here as well. Avoiding illegitimate children by controlling female premarital sex also limits the possibility of conflicting claims on the estate in which a woman has rights [112]. [112] concluded that the link between female inheritance, family honor, and veiling that marks the Mediterranean world cannot be separated from the position of women as carriers of property. Anthropological evidence suggests that within

27According to Sunni law, illegitimate children do not bear the name of the father, but that of the mother’s family. They also inherit only from the mother’s side of the family and not from the father’s. According to Shia law, illegitimate children do not inherit from either parent. However, a child who is denied paternity by the father (who is married at the time to the mother) inherits from its mothers’ family [150].

19 the same community, women with lower status and without access to land have more sexual freedom [137, 210]. Female mobility and entrepreneurship. One famous argument about the effect of different inheritance systems on economic development is that impartible inheritance stimulated the development of new economic opportunities by encouraging mobility, emigration, and entrepreneurship among landless offspring. People who did not acquire landed property had to emigrate to cities to join the industrial labor force [213, 36]. Also, the lack of any future inheritance encouraged the need for education, training, mobility, and entrepreneurship [248, 78]. The same argument can apply to landless versus landed women. All else being equal, a woman with no prospect of inheritance had more incentive to be educated and independent, to migrate and work in industrial settings, to be an entrepreneur and create her own business. In fact, women in Europe migrated more often than men because men inherited land more often and had to stay in their home village [57]. Women’s informal learning, as servants or teachers, is considered a signiﬁ- cant contributor to the intergenerational transmission and accumulation of human capital during the industrial revolution [90]. These factors, as possible consequences of women’s exclusion from inheritance, might have contributed to women’s higher social and economic participation in Europe. In the Middle East, on the other hand, women as carriers of property were immobile, engaged in cousin marriages, and under the strict control of their male relatives. The arguments presented above on the relationship of female inheritance with inmarriage and the status of women can be summarized in two hypotheses. Hypothesis 1: Female inheritance has a positive effect on cousin marriage, endogamy, and arranged marriage. Hypothesis 2: Female inheritance has a negative effect on female economic participation and premarital sexual freedom.

2.4 Empirical strategies and results using historical data

In this section, I provide historical and big-picture evidence on the correlations predicted by my two hypotheses. In section 2.4.1, I describe historical data on pre-industrial societies from the Ethno- graphic Atlas to test the association of female inheritance with outcome variables. I show that female inheritance is associated with higher cousin marriage and endogamy (as stated in hypothesis 1) and lower female participation in agriculture and premarital sexual freedom (as stated in hypothesis 2). In section 2.4.2, I use mid-twentieth-century Italian province-level data to show the correlation of egalitarian inheritance system with higher cousin marriage rates and lower female economic participation rates.

20 2.4.1 Pre-industrial societies

The Ethnographic Atlas28 includes data on 1,291 pre-industrial societies distributed globally and mostly sampled between 1800 and 1950, ranging from societies with complex agricultural economies and political systems to small hunter-gatherer groups. Table B.3.1 of Appendix B.3 shows a detailed description of all variables used in this section. If a variable is defined in previous empirical studies using the Ethnographic Atlas, I have used exactly the same definitions to construct the variable, in which case the related study is also mentioned in the table. Inheritance systems. Earlier, I defined female inheritance as partible inheritance by both sexes. Therefore, I construct indicator variables for inheritance systems using entries in the Ethnographic Atlas on female inclusion in inheritance (EA074) and partibility of inheritance ( EA075). With the available data, we can clearly characterize the inheritance systems of 820 Ethnographic Atlas societies29 as summarized in Table 2.1.

EA074 (1) (2) (3) & (6) (4) & (5) (7) Absence of Younger Sons and EA075 private property Sisters’ sons brothers daughters Sons only Total Percent

(1) Equally distributed 0 10 20 86 191 307 %37.44

(2) Best qualiﬁed 0 0 9 3 8 20 %2.44

(3) Ultimogeniture 0 0 2 1 19 22 %2.68

(4) Primogeniture 0 11 107 8 122 248 %30.24

(9) Absence of 223 0 0 0 0 223 %27.20 private property

Total 223 21 138 98 340 820

Percent %27.20 %2.56 %16.83 %11.95 %41.46 %100.00

Table 2.1: Intersection of entries EA074 and EA075 of the Ethnographic Atlas data.

Based on the two entries, I construct dummy variables for the following categories: impartible inheritance (categories 2, 3, and 4 of EA075), partible inheritance by males only (intersection of category 1 of EA075 and categories 2, 3, 6, and 7 of EA074), partible inheritance by both sexes (intersection of category 1 of EA075 and categories 4 and 5 of entry EA074), and absence of private property (category 1 of entry EA074, or equivalently category 9 of entry EA075). Figure 2.2 seen earlier followed this categorization to portray the distribution of inheritance systems globally for Ethnographic Atlas societies. Dependent variables. As a measure of cousin marriage in the Ethnographic Atlas societies, I use entry EA023 on the rules governing cousin marriage. I construct a variable for cousin marriage that takes on integer values ranging from 1 to 4, where higher values indicate tighter cousin marriage culture. The highest value is assigned if ﬁrst-cousin marriage is allowed with a father’s brother’s daughter, since this is the ideal marriage to keep the property within the male lineage. Value 3 is

28[196, 33, 120, 160, 45, 158]

29EA074 is available only for 856 societies, and EA075 is available only for 820 societies.

21 assigned if marriage with any first cousins except a father’s brother’s daughter is allowed. Value 2 is assigned if only second-cousin marriages are allowed. Finally, value 1 is assigned if no first- or second-cousin marriages are allowed. To construct a variable for endogamy, I use entry EA015, which classifies the prevalence of local endogamy, agamy, and exogamy. I construct a variable for endogamy, which takes on integer values ranging from 1 to 3, where higher values indicate higher endogamy: value 1 is assigned for exogamous communities, value 2 is assigned for societies with no marked tendency toward endogamy or exogamy, and value 3 is assigned for endogamous communities. Entry EA078 classifies prevailing standards of sexual behavior for unmarried women. I construct an indicator variable for female premarital sex prohibition, which takes value 1 if premarital sex of unmarried women is precluded or prohibited, and value 0 if it is permitted. Finally, to construct a variable for female participation in agriculture, I follow [16] and construct a variable that takes on integer values ranging from 1 to 5, where higher values indicate more participation of women in agriculture. Control variables. Table 2.2 lists all control variables used in the regression analyses. To test hypotheses with the Ethnographic Atlas data, I report three sets of regressions, which in addition to dummy variables for inheritance systems include the following set of controls: first, the regression specification from [16] as the specification with minimal controls; second, a large set of ethnographic and geographic controls, including those from [16]; third, the set of robustness controls, including those mentioned as “historical controls" by [16]. See Table B.3.1 of Appendix B.3 for a detailed description of all dependent and control variables.

Minimal controls Traditional plough use, Settlement complexity, Political hierarchies, Presence of large animals, Tropical climate, Suitability for agriculture Maximal controls Ethnographic controls: Traditional plough use, Non-irrigated intensive agriculture, Irrigated intensive agriculture, Settlement complexity, Political hierarchies, Presence of large animals, Year society sampled, Patrilineal descent, Matrilineal descent, Dowry, Brice price Geographic controls: Latitude, Mean temperature, Temperature predictability, Mean precipitation, Precipitation predictability, Tropical climate, Suitability for agriculture, Distance to coast, Slope, Ruggedness, Elevation Robustness controls Population, Proportion of subsistence from herding, Proportion of subsistence from hunting, Proportion of subsistence from gathering, Patrilocal marriages, Matrilocal marriages, Extended family, Nuclear family

Table 2.2: Set of control variable used in different speciﬁcations for the Ethnographic Atlas data analysis.

The set of control variables controls for possible factors that can affect inheritance systems, inmarriage, and the status of women, including agricultural organization, economic and political development, subsistence economy, kinship structure, transfers at marriage, and a large set of geographic variables. Results. Table A.2.1 of Appendix A.2 reports descriptive statistics for all dependent and control variables based on available observations in the full sample. The number of observations varies

22 across regressions due to missing observations in the outcome variables. Tables A.2.2–A.2.9 of Appendix A.2 report descriptive statistics for the sample of societies in regression analyses of each outcome variable, as well as full regression results with minimal, maximal, and robustness controls. Since the results are robust to inclusion of robustness controls, here I report only the minimal and maximal regression results. Table 2.3 reports the results for cousin marriage, endogamy, and female premarital sex prohibition. The omitted category in the regressions is impartible inheritance. Consistent with the hypothesis, partible inheritance by both sexes has a significant positive correlation with cousin marriage, endogamy, and female premarital sex prohibition. The estimated coefficients of the dummy variable for partible inheritance by both sexes are large considering the standard deviations of dependent variables, and also comparing with the coefficient of the dummy variable for traditional plough—which is considered as an important determinant of pre-industrial characteristics of societies. The effect of partible inheritance (by males only) is insignificant in the regressions, confirming that the dimension of inheritance systems relevant for these outcomes is female inheritance, not partibility of inheritance.

Female premarital Cousin marriage Endogamy sex prohibition (SD=0.990) (SD=0.580) (SD=0.499) VARIABLES (1) (2) (3) (4) (5) (6)

Partible inheritance by males only 0.046 0.145 0.011 0.013 0.032 0.005 (0.098) (0.096) (0.056) (0.056) (0.062) (0.064) Partible inheritance by both sexes 0.284* 0.376** 0.365*** 0.233*** 0.258*** 0.238*** (0.151) (0.150) (0.078) (0.079) (0.081) (0.083)

Absence of private property yes yes yes yes yes yes

Minimal controls yes yes yes Maximal controls yes yes yes

Observations 651 651 686 686 427 427 R-squared 0.088 0.218 0.067 0.144 0.138 0.171

Table 2.3: OLS estimates are reported with robust standard errors in parentheses. ***, **, and * indicate signiﬁcance at the 1, 5, and 10% levels.

One concern with estimations based on Ethnographic Atlas data is spatial auto-correlation (across nearby units). Table A.2.10 of Appendix A.2 reports OLS estimates for maximal regression specification with [66] standard errors for spatial dependence with cutoffs of 60 decimal degrees. The coefficients remain significant at 1% and 5% levels. Since the outcome variables are ordinal and not cardinal, I also report regression results with maximal specifications using binary and ordered logit estimations in Table A.2.11 of Appendix A.2. The coefficients remain significant at 1% and 5% levels. Finally, I control for region fixed effects to address concerns about possible confounding with religion and culture. Christianity has a long history of discouraging inmarriage, which happens to coincide with patrilineal inheritance. In contrast, Islam has long encouraged both inmarriage and female inheritance. Therefore, all reported correlations above might just be the result of such historical coincidences. Based on the Ethnographic Atlas data, I created 17 subcontinent regions,

23 importantly including 7 regions within the Islamic world: 2 in its core (Middle East and Northern Africa) and 5 in its periphery (Caucasus, Middle Asia, Indian Subcontinent, Malesia and Papuasia, and Northern Tropical Africa) where Islam might be mixed with other religions or local traditions. I also defined 4 regions within the Christian world: Northwest Europe, Southeast Europe, Eastern Europe, and Siberia. Table A.2.12 of Appendix A.2 includes a map of the regions and reports the regression results using region fixed effects with standard errors clustered in the region level. Despite region fixed effects washing away most of the variation across societies, partible inheritance by both sexes is still significant in all regressions at 5% or 10% levels.

Female participation in agriculture (SD=1.018) (SD=1.036) (SD=0.985) (SD=0.985) (SD=0.985) (1) (2) (3) (4) (5) VARIABLES Alesina et al. (2013) Updated data Restricted sample and Inheritance Updated controls

Partible inheritance by males only -0.244** -0.190** (0.097) (0.092) Partible inheritance by both sexes -0.570*** -0.278** (0.128) (0.129) Traditional plough use -0.883*** -0.848*** -0.751*** -0.615*** -0.397*** (0.114) (0.107) (0.120) (0.119) (0.146)

Absence of private property yes yes

Minimal controls yes yes yes yes Maximal controls yes

Observations 660 713 489 489 489 R-squared 0.135 0.135 0.143 0.183 0.326

Table 2.4: OLS estimates are reported with robust standard errors in parentheses. ***, **, and * indicate signiﬁcance at the 1, 5, and 10% levels.

In Table 2.4, I focus on female participation in agriculture, which is highlighted in the famous study by [16]. They confirm the hypothesis that traditional plough use has decreased female participation in agriculture. Regression 1 of Table 2.4 replicates the regression result by [16] using the same specification and data.30 The resulting coefficient for plough agriculture is the same as in [16]. In column 2, I estimate the same regression specification, but using the version of the Ethnographic Atlas data used in this study.31 The coefficient is almost the same. In column 3, the sample is restricted to observations with available inheritance data. Plough agriculture remains significant, and the coefficient is not significantly different than column 2 (Wald test p-value=0.105). Column 4 introduces the controls for inheritance systems. Both plough use and partible inheritance by both sexes are significant. This decreases the coefficient on plough use (Wald test

30See regression 1 of Table 4 in [16]. The data is available online at http://scholar.harvard.edu/nunn/home.

31The new version of the Ethnographic Atlas used in this study includes more societies. Also, the data is improved by linking each data point to one or more of the 3,502 ethnographic sources. Moreover, data sources and radius around societies used for tropical climate and suitability for agriculture are different than those in [16]. With available data from [16], it is possible to run the same regressions with the minimal controls. Table A.2.14 of Appendix A.2 presents the regressions, indicating that results are robust across two versions of the Ethnographic Atlas.

24 p-value=0.000) relative to column 3. Moreover, the coefficient of partible inheritance by both sexes is not significantly different than the coefficient of plough use (Wald test p-value=0.814). Contrary to regressions on other outcome variables, here I am cautious not to attribute the full magnitude of the coefficient of partible inheritance by both sexes to the described mechanism in the conceptual framework. This is because partible inheritance by males only is also significant—although smaller than partible inheritance by both sexes (Wald test p-value=0.013) and insignificant when region dummies are included (see Table A.2.12 of Appendix A.2). This could indicate that a different mechanism also partly contributes to the difference in coefficients of impartible and partible inheritances.32 In regression 5, where I use maximal controls, the coefficient of partible inheritance by both sexes is still significant and not significantly different from the coefficient of plough use (Wald test p-value=0.566). For full regression results with minimal, maximal, and robustness controls, estimates with Conley standard errors, ordered logit model, and region fixed effects, see Tables A.2.8–A.2.13 of Appendix A.2. The results are robust. 33

2.4.2 Mid-twentieth-century Italy

Data on cousin marriage rates and inheritance systems of Italian provinces is available for 1950– 60s, as presented in Figure 3.3. For inheritance systems of Italian provinces, I use the most detailed data available for European regions, which is provided by [248] mostly based on censuses from Western European countries in the 1950–1960s. According to Todd, family types in Italian provinces can be classiﬁed as communitarian, incomplete stem, or egalitarian nuclear. While inheritance in communitarian and incomplete stem family types was patrilineal, in the egalitarian nuclear family type common in south and north west Italy, inheritance was equally divided among children, and women—at least in southern Italy—were likely to inherit.34 Table B.3.2 of Appendix B.3 presents a full description of the data and variables used in this section.

32One possible mechanism described by [250] is that under impartible inheritance, usually two generations live together (stem family). This gives wives more time to work on the farm because co-residence with the mother-in-law reduces the burden of household work, freeing up the wife’s time for non-domestic work. This requires the assumption that on the other hand, under partible inheritance, all sons start their own independent households.

33In the above analyses, I have not distinguished between lineal and lateral inheritance systems. For example, impartible inheritance category also includes primogeniture by sister’s sons and younger brothers. In Table A.2.15 of the appendix, I deﬁne separate indicator variables for lateral inheritance systems—that is, inheritance by matrilineal heirs (such as sister’s sons) and inheritance by patrilineal heirs (such as younger brothers). Regression analyses report qualitatively similar results for inheritance by both sexes.

34The Neapolitans are one pre-industrial society recorded in the Ethnographic Atlas within the borders of today’s Italy. Neapolitan was and still is the language of much of southern Italy, with 6 million native speakers today mostly in the Campania and Calabria regions [172]. The Ethnographic Atlas indicates that the Neapolitans practiced partible inheritance of land by both sexes. [109] also describe the inheritance system of southern Italy as bilateral inheritance (i.e., by both sexes). The Ancient Romans are the other Ethnographic Atlas society within the borders of today’s Italy. They also practiced partible inheritance by both sexes—in which the prevalence of female inheritance in the Mediterranean region was rooted [156].

25 To test whether traditional female inheritance can explain the variation of cousin marriage rates in Italy—as stated in hypothesis 1—I run a series of regressions controlling for economic, demographic, and geographic factors. Cousin marriage data is from [57] based on records of Papal dispensations kept in the Vatican archives. The data is available for all provinces between 1945 and 1964. Since the available historical demographic and economic data belongs to 1960s, I use only the average of the last five-year period (1960–64). I control for two variables as indicators of the economy of Italian provinces at the same time period: GDP per capita and share of agriculture in the GDP for the regions they belong to, both averaged over 1963 and 1964. I also control for the population of provinces using the available data on population of provinces in 1961. Similarly to the Ethnographic Atlas analysis, I also control for a wide range of geographic controls such as ab- solute latitude, mean temperature, mean precipitation, an indicator for tropical climate, suitability for agriculture, distance to coast, slope, ruggedness, and elevation. To make sure that the effect of inheritance systems is not due to confounding cultural and historical factors, I also control for two key variables. First, a well-known fact about Italy is the cultural differences between the south and north [29, 214]. Therefore, I include a dummy variable in the regressions for the provinces in so-called Mezzogiorno regions: southern Italy and the islands. In- cluding this dummy variable assures a stringent test of the hypothesis because southern Italy and the islands also happen to have the highest cousin marriage rates (mean cousin marriage rate of 1.377 compared to 0.306 in other regions, mean-comparison test p-value<0.001, N=101) and the egalitarian inheritance system. Second, regions in central Italy with the communitarian family type are also characterized by historical domination of the Papal States. The Papal States were territo- ries under the sovereign direct rule of the pope, from the eighth century until Italian unification in 1871. Therefore, it is plausible to expect stricter historical emphasis on cousin marriage bans and lower cousin marriage rates in the Papal States (0.412 compared to 0.976 in other regions, one-sided mean-comparison test p-value=0.063, N=101). This might confound with inheritance by sons in the communitarian family type in the region, which allows outmarriage, according to the conceptual framework. Columns 1 and 2 of Table B.5.13 show regression results on cousin marriage percentage. The egalitarian inheritance system is highly significant in both specifications, and it is associated with almost 0.4 higher percentage points of cousin marriage (by 35% of the standard deviation) in the full regression specification. The magnitude of the coefficient is not significantly different from the dummy variable for the Papal States (Wald test p-value=0.129). This confirms the hypothesis that where women are more likely to inherit, inmarriage is more likely. Columns 1 and 2 of Table B.5.13 show regression results on female participation percentage. Using data from census 2011, the variable is defined as the percentage of currently economically active women (either employed or unemployed) aged 15 or older from the total population of a province. For these regressions, I use province-level data on GDP per capita and share of agriculture averaged over 2000–2013. In the full regression specification, the egalitarian inheritance system is associated with 3.75 percentage points lower female participation (by 68% of the standard devia-

26 Cousin marriage percentage Female participation percentage (SD=1.149) (SD=5.529) VARIABLES (1) (2) (3) (4)

Geographic controls yes yes

Observations 101 101 101 101 R-squared 0.592 0.874 0.412 0.822

Table 2.5: OLS estimates are reported with robust standard errors in parentheses. ***, **, and * indicate signiﬁcance at the 1, 5, and 10% levels. Descriptive statistics and full regression results are reported in Table B.5.14 of Appendix A.2. tion). This is consistent with the historical accounts on the links between female inheritance, culture of honor, and lower female participation in southern Italy and the Mediterranean Europe in general [157, 112].

2.5 Empirical strategies and results using contemporary data from developing countries

In developed countries, economic empowerment of women increases their control over decisions and bargaining power within the household, and it creates positive outcomes such as reduced domestic violence against women [see, e.g., 6, 21]. However, it is not clear that this relationship exists in all developing countries. Some studies suggest that policies designed to empower women in developing countries, such as female inheritance, create positive outcomes for women [178, 69, 134, 224, 141, 19, 22]. However, others report unintended negative consequences, such as increased female child mortality [223], domestic violence and suicides [23, 126], and so on [for other studies see, e.g., 176]. In their patrilineal nature, many developing countries resemble the pre-industrial world and ﬁt the conceptual framework of my study. The effect of this patrilineal bias may be even more signiﬁ- cant for women’s premarital and marital outcomes—the focus of my study—than their postmarital outcomes. This is because women are likely to exhibit different household bargaining allocations depending on their roles as spouses or daughters. As spouses, women who inherit and own property might gain higher bargaining power against their husbands for household decisions such as children’s education [75]. However, women’s premarital and marital outcomes are formed in their

27 parental households when they are very young.35 As daughters, they either have not received inheritance yet—because their fathers are still alive—or they have received inheritance but have not much opportunity to gain bargaining power in premarital and marital decisions made for them in the paternal household. In this section, I present ﬁndings on women’s premarital and marital outcomes using individual- level data from Indonesia and India. In section 2.5.1, using data from Indonesian Family Life Sur- veys, I examine the association of female inheritance with endogamy and arranged marriage—as stated in hypothesis 1—and economic participation—as stated in hypothesis 2. Then, in section 2.5.2, I use the national amendment of the Hindu Succession Act in 2005 for a difference in differences analysis of the effect of the amendment on cousin marriage and premarital sex rates—as predicted by hypotheses 1 and 2 respectively—among Indian women.

2.5.1 Individual-level analysis using Indonesian data

The Indonesian Family Life Survey (IFLS) is a longitudinal survey conducted between 1993 and 2014 in Indonesia. The sample is representative of about 83% of the Indonesian population and contains over 30,000 individuals of 7,224 households living in 13 of the 26 provinces, which captures the cultural and socioeconomic diversity of Indonesia. I use data on male and female adults (household heads, their spouse, and selected seniors in the household) to construct outcome variables.36 The IFLS has no information on cousin marriages, but the data allows creation of a proxy for endogamy. In the case of female inheritance, we should expect women to marry their relatives or neighbours more often, which usually means marrying from the same village. In contrast, where women are excluded from inheritance, men inherit the land and stay in the same village, but landless women are likely to outmarry and join their husbands in other villages. Therefore, I construct a variable for endogamy that indicates staying within the same village after marriage. The constructed indicator variable for endogamy takes value 0 if the respondent moved to another village after the wedding in the latest marriage, and takes value 1 otherwise. I dropped a few (less than 0.02%) observations where the respondents didn’t start living with their spouse after the wedding.37 The second outcome variable indicates whether the respondent’s marriage was an arranged marriage (versus a love marriage). Respondents are asked who chose their spouse in their ﬁrst marriage. I create a dummy variable for arranged marriage that takes value 1 if parents or family members chose the respondent’s spouse, and takes value 0 if the respondent chose their spouse by them-

35In my study samples, most women in Indonesia and India married before age 20.

36Panel respondents were not asked the related questions in the IFLS (2000) if they had been asked in the IFLS (1997). Therefore, I had to merge IFLS (2000) with IFLS (1997) to have a full sample.

37In unreported regressions on endogamy, to make sure that moving to the new residence after wedding was not temporary, I dropped some observations (around 11%) where the respondent stayed less than a year in their ﬁrst residence (whether in another village or within the same village) after the wedding. The results do not change.

28 selves. It is positively correlated with the endogamy variable (Spearman’s r=0.048, p-value=0.000, N=13819). While 23% of women are engaged in arranged marriages, this percentage for men is only 13%. This difference—which also shows up in the coefficients of female inheritance in the regression analyses—reflects another aspect of unequal gender norms. Men are more likely to choose their wives and ask their parents to arrange the marriage. Therefore, what is a choice for the husband can be perceived as an arranged marriage by the wife. Another aspect of unequal gender norms in arranged marriages is the underage marriage (below age 16 in Indonesia). More than 90% of underage marriages involve girls, of which 37% are arranged marriages; among women married at the legal age, this percentage is only 20%. For economic participation as my third outcome variable, I create three indicator variables to capture any economic activity by the respondent. The first variable takes value 1 if the respondent is self-employed, and takes value 0 otherwise. The second variable takes value 1 if she is employed in public or private sectors (working for others), and takes value 0 otherwise. The third variable takes value 1 if she works unpaid for her family. To define these variables, I dropped observation for those above 60 years old.

Indonesian ethnic groups with female inheritance tradition: Here, I investigate the effect of traditional inheritance systems of Indonesian ethnicities on the outcome variables. Following the strategy in [25], I use ethnographic data from the Ethnographic Atlas and Ethnic Groups of Insular Southeast Asia [169] to link individuals to their traditional inheritance systems through their self- reported ethnicity in the IFLS. Then I construct an indicator variable for female inheritance, which takes value 1 if the ethnicity to which the individual belongs has traditionally included women in inheritance.38 In the regression analysis, following [25], I control for education (five categories), quadratic in marriage age, and ethnicity control variables, and I cluster standard errors at the ethnicity level. In addition to their ethnicity controls (traditional presence of female-dominated agriculture and traditional matrilinearity), I also control for bride price practice from their study.39 Moreover, I control for religion fixed effects (Muslim, Protestant, Catholic, Hindu, and other), quadratic in age, type of residence (urban or rural), and province fixed effects. The latter is relevant because ethnicities that traditionally excluded women from inheritance have a majority in three provinces only. Table 2.6 reports regression results for the sample of female and male adults separately. Table A.2.17–A.2.20

38The IFLS (2000) contains information on 23 ethnicities. As described in Table B.3.3 of Appendix B.3, 14 of the ethnicities (around 75% of the sample) can be directly matched to the ethnic groups for which the inheritance information is available from ethnographic records, meaning that the name of the society in ethnographic records and the name of the ethnicity in the IFLS are the same, and based on the coordinates in ethnographic records, the society is located in the same regions that the IFLS surveys took place.

39Following [25], female-dominated agriculture is a dummy variable that takes value 1 if female participation in agriculture is higher than male participation. The matrilineal dummy variable takes value 1 for Minang ethnicity (the only ethnicity in IFLS sample with matrilineal inheritance). The dummy variable for bride price takes value 1 if the ethnicity practices bride price.

29 of Appendix A.2 show the descriptive statistics of variables, full regression results, and logistic estimation with qualitatively similar results.

Economic participation Endogamy Arranged marriage Self Public/Private Family (SD=0.457) (SD=0.424) (SD=0.441) (SD=0.405) (SD=0.357) VARIABLES (1) (2) (3) (4) (5)

Panel A: Women’s sample Female inheritance 0.111*** 0.063*** -0.086*** 0.018 0.005 (0.014) (0.018) (0.011) (0.016) (0.036) Education -0.049*** -0.029** -0.018*** 0.035*** -0.018*** (0.004) (0.011) (0.004) (0.008) (0.003) Urban -0.098*** -0.053*** -0.005 0.049*** -0.132*** (0.009) (0.014) (0.021) (0.014) (0.016)

Observations 7,575 7,705 6,645 6,645 6,645 R-squared 0.058 0.231 0.074 0.063 0.089

Panel B: Men’s sample Female inheritance -0.018 0.023* -0.022 0.041 -0.006 (0.019) (0.012) (0.025) (0.029) (0.003) Education -0.041*** -0.019*** -0.056*** 0.056*** -0.002* (0.003) (0.005) (0.010) (0.011) (0.001) Urban -0.114*** -0.035*** -0.176*** 0.176*** -0.012** (0.009) (0.004) (0.019) (0.015) (0.004)

Observations 6,045 6,116 5,180 5,180 5,180 R-squared 0.083 0.130 0.123 0.126 0.016

Quadratic in age yes yes yes yes yes Quadratic in marriage age yes yes yes yes yes Ethnicity Controls yes yes yes yes yes Religion FE yes yes yes yes yes Province FE yes yes yes yes yes

Table 2.6: OLS estimates are reported with robust standard errors, clustered at the ethnicity level, in parentheses. ***, **, and * indicate signiﬁcance at the 1, 5, and 10% levels.

Consistent with the hypotheses, women from ethnicities with traditional female inheritance are more likely to be engaged in endogamous and arranged marriages. The coefficients are significant and large considering the standard deviation of dependent variables and also compared with coefficients of other important variables such as Catholicism, education, and urban residence. Women from ethnicities with traditional female inheritance are also less likely to be self-employed. In the men’s sample, female inheritance is insignificant in all regressions except for arranged marriage. However, the coefficient of arranged marriage is smaller than the one in the women’s sample (Wald test p-value=0.086), likely due to unequal power of sexes in their choice of spouse—as discussed before. Insignificance of the coefficient in the endogamy regression for the men’s sample might imply that an ethnicity with female inheritance is different from an ethnicity without female inheritance in migration rates of brides (not grooms) across villages. Since men inherit land under both regimes and stick to their land in the village, female inheritance should not affect the average endogamy rate of men across regimes.

30 Indonesian women receiving inheritance: Here, instead of traditions of the ethnic groups, I focus on the actual inheritance individuals received. In the first survey of the IFLS, adults are asked if either of their parents died, and if so, whether and what the parent bequeathed to them: house, land, livestock, jewellery, or money. Using this information, I construct an indicator variable for receiving inheritance that takes value 1 if the respondent inherited house, land, and livestock (properties subject to economies of scale) from either parents, and takes value 0 otherwise. Of the sample, 36% reported that they received an inheritance in the form of house, land, and livestock. To capture the effect of received inheritance on individual-level outcomes, I follow a strict strategy. The smallest administrative division in Indonesia (and in the IFLS data) is community (desa/kelurahan). Therefore, controlling for fixed effects of 312 IFLS communities, I make sure there is no confounding due to economic, geographic, and other differences across communities. In other words, the regressions capture only the variation of the inheritance indicator within communities, not across communities. Using fixed effects of religions and ethnicities, I also control for variation due to different religious and ethnic attitudes. In addition, I include controls from previous regressions: education, quadratic in age, and quadratic in marriage age. I also cluster standard errors at the community level.

Economic participation Endogamy Arranged marriage Self Private/Public Family VARIABLES (1) (2) (3) (4) (5)

Panel A: Women’s sample Inheritance dummy 0.050*** 0.035** 0.029 -0.006 0.007 (0.015) (0.015) (0.019) (0.013) (0.012) Education -0.034*** -0.050*** -0.029*** 0.054*** -0.020*** (0.011) (0.008) (0.009) (0.011) (0.006)

Observations 4,231 4,289 3,656 3,656 3,656 R-squared 0.179 0.382 0.161 0.194 0.338

Panel B: Men’s sample

Inheritance dummy 0.065*** 0.020 0.041** -0.038** 0.002 (0.015) (0.014) (0.019) (0.019) (0.003) Education -0.032*** -0.026*** -0.095*** 0.091*** -0.001 (0.009) (0.007) (0.010) (0.010) (0.001)

Observations 3,702 3,776 3,059 3,059 3,059 R-squared 0.180 0.301 0.348 0.346 0.121

Quadratic in age yes yes yes yes yes Quadratic in marriage age yes yes yes yes yes Religion FE yes yes yes yes yes Community FE yes yes yes yes yes

Table 2.7: OLS estimates are reported with robust standard errors, clustered at the community level, in parentheses. ***, **, and * indicate signiﬁcance at the 1, 5, and 10% levels. See Tables A.2.21 and A.2.23 for the regressions including ethnicity ﬁxed effects.

Since the ethnicity information was collected in 2000 for the first time, there is a rate of attrition. Table 2.7 presents results for both male and female adults without ethnicity fixed effects. The results are unchanged when ethnicity fixed effects are controlled for. Tables A.2.21–A.2.24 Appendix

31 A.2 show the descriptive statistics of variables, full regression results, including those with ethnicity fixed effects, and logistic estimations with qualitatively similar results. Unreported regression analyses show that the results are also unchanged if standard errors are clustered at the province level. Receiving inheritance from parents is highly significant in endogamy and arranged marriage regressions in the women’s sample, and still comparable with the coefficient of education. The coefficient of inheritance in the regression on endogamy in the men’s sample is not significantly different than in the women’s sample (Wald test p-value=0.456 and 0.861). This makes sense, since men and women are sampled from the same communities, and endogamy entails staying within the same community after marriage. However, the coefficients in arranged marriage regressions in the men’s sample are smaller and insignificant. Again, since both men and women are sampled from the same communities, this reflects the different perception men and women have about the nature of their marriages. Men have more freedom choosing the spouses who are arranged to marry them. Inheritance is insignificant in regressions on all three economic participation variables in the women’s sample. While inheritance is associated with higher self-employment and lower private/public sector employment in the men’s sample, it doesn’t show a positive association with self-employment in the women’s sample. Recall that in the ethnic-level regression analyses, we found a negative association between female inheritance practice and self-employment. Therefore, higher self-employment in the men’s sample may be considered the positive effect of inheriting property in a developing world, which is missing for women, likely due to the patrilineal restrictions on women’s empowerment.

Inheritance expectations: There is a possibility of reverse causality in the above analyses. That is, those who marry within the village or by their parents’ choice are more likely to inherit property from their parents just because they are closer to them physically or emotionally. To address this concern, I split the inheritance indicator into two separate variables: an indicator variable called post-marriage inheritance that takes value 1 if the respondent inherited house, land, and livestock after their marriage, and an indicator variable for pre-marriage inheritance which takes value 1 if the respondent inherited house, land, and livestock before their marriage. These categories are defined by comparing the respondent’s date of marriage and the date of death of the parent who left the inheritance, and are based on the assumption that the inheritance is transferred upon the death of the parent. Around 54% of respondents received inheritance post-marriage. Table A.2.25 of Appendix A.2 reports regression results for the full sample (including both men and women), first with the inheritance variable as defined in earlier analyses, then with pre- and post-marriage inheritance variables.40 Just as the inheritance variable, both post- and pre-marriage

40When pre- and post-marriage inheritance categories are used for male and female samples separately, I end up with around 700 observations for each inheritance-sex combination, which is not enough to get robust results. Therefore, here I ignore the male and female distinction, and instead focus on the pre- and post-marriage distinction.

32 inheritance indicators are significant in the regressions. The Wald test results indicate that the difference of the coefficients is not significant. This implies that expecting to inherit property has the same association with marriage outcomes as actually inheriting the property. The potential for future inheritance can be predicted based on ethnicity, religion, and local customs, and whether the parent has a property to bequest. Those with high potential to receive an inheritance are subject to the same marriage restrictions as they would be if they actually received the inheritance.

2.5.2 Difference in differences analysis using Hindu Succession (Amendment) Act of India

Since 1956, property rights for Hindus in India (including Sikhs, Jains, and Buddhists) have been governed by the Hindu Succession Act (HSA). The Act applies to all states except Jammu and Kashmir, and it explicitly exempts Muslims, Christians, Parsis, and Jews. As in traditional Hindu law, under the HSA women had no rights to joint family property (including land and other ancestral assets). Since 1956, some states have amended the Act so that both sons and daughters have a right to joint family property (Kerala in 1976; Andhra Pradesh in 1986; Tamil Nadu in 1989; Maharashtra and Karnataka in 1994). However, these amendments applied only to women who were not yet married at the time of the reform in their state. In the other 29 states,41 men remained the sole joint heirs of family property until 2005. Under the Hindu Succession (Amendment) Act of 2005, all daughters, including married daughters, are also joint heirs in family property such as agricultural land.42

Figure 2.6: Cousin marriage rates among Hindu women across Indian states.

41Excluding Jammu and Kashmir, which was exempt from the Hindu Succession Act, and Telangana, which was part of Andhra Pradesh until 2014.

42See [4] and [23] for more details on the HSA and its amendments in different states. The Hindu Succession (Amend- ment) Act was passed on September 5, 2005, and applied to any disposition, alienation, and partition of property that had taken place after December 20, 2004.

33 The states that passed amendments to the HSA before 2005 are the farthest southern states, whose traditional schools of law (the Madras and Bombay sub-schools) were more gender equal, and they agreed to include female inheritance rights when the HSA was passed in 1956. However, the northern states dismissed the idea by a majority vote, and the traditional laws of female exclusion in joint property were maintained until 2005 [23]. Interestingly, Hindus in southern India have historically experienced much higher cousin marriage rates compared with Hindus in the other states (see Figure 2.6); in the sample of those married in or after 2000, the rate was 24% in the states that passed local amendments in the past versus 6% in other 29 states. This historical difference in the marriage pattern of the southern and northern states might be a consequence of their different traditional attitudes to female inheritance. Therefore, there might be concerns regarding the endogeneity of the amendments in the five states. For three reasons, here I focus on other states and the 2005 amendment only. First, endogeneity of the amendments in the five states is possible, as discussed above. Second, under the past amendments in the five states, only unmarried women were eligible to inherit. Therefore, timing of marriages might have responded to the amendments. With the anticipation of the amendments, families who did not want their daughters to be eligible married them just before the law passed. This is not a concern for the 2005 amendment under which both married and unmarried women were eligible for inheritance. Third, past amendments not only were different from the 2005 amendment, but also might have been different from each other. For example, the amendment of the state of Kerala removed the legal status of the joint family altogether. I also drop some observations of respondents with unknown religions or who do not identify as Hindu (including Sikhs, Jains, and Buddhists), Muslim, or Christian. To estimate the impact of the 2005 amendment—which substantially improved female inheritance— on two outcome variables, I exploit a difference in differences approach using data collected from adult women (15–49 years old) by National Family Health Survey of India (repeated cross-sectional surveys conducted in four rounds between 1992 and 2016). The first outcome variable is cousin marriage, which takes value 1 if a woman’s husband in her first marriage is a blood relative (first cousin, second cousin, or other blood relatives such as uncle), and takes value 0 otherwise. The second outcome variable is premarital sex, which takes value 1 if a married woman had her first sexual intercourse before her first union with her first husband, and takes value 0 if she had her first sexual intercourse at or after her first union or if she has never had sex with her first husband.43 The first difference I use is the religion of the respondent. The amendment should have had an impact on Hindu women (Hindus, Sikhs, Jains, or Buddhists) but not women from the exempted

43Contrary to the cousin marriage variable, the sample for the premarital sex variable includes only those women who are currently married. This is because the age at first union was not asked of other groups, such as those who are widowed, divorced, or separated. I have created the premarital sex variable using information on age at first union and age at first sexual intercourse. The latter includes a category labelled as "at first union". In addition, whenever age at first union and age at first sexual intercourse are the same, because of strong norms against premarital sex, I considered it as sex at first union with husband. However, it is possible for age at first union and age at first sexual intercourse to be the same, but for first sexual intercourse to have occurred before the marriage.

34 religions (Muslims, Christians, Parsis, or Jews). The second difference I use is exposure to the amendment, as measured by the year of the first marriage. The decision to marry a relative could have been affected by the amendment only if the marriage took place in or after 2005.44 Therefore, the marriage decision of those Hindu women who married before 2005 could not have been affected by the amendment. Similarly, the premarital sex attitude of those Hindu women who married before 2005 could not have be affected by the amendment because by definition premarital sex takes place before the first marriage. I assume that an outcome y for woman i of religion r married for the first time in year t in state s is a function with the following form:

yirts = α + β1Tirt + β2γr + β3δt + β4θs + β5(θs × δt) + β6(θs × γr) + β7Xirts + eirts

Tirt captures a woman’s treatment status and takes value 1 if she is a Hindu (i.e., Hindu, Sikh,

Jain, or Buddhist) and married for the first time in or after 2005. The coefficient of interest is β1, which identifies the effect of exposure to the amendment. Religion dummy (γr) takes value 1 if a woman is a Hindu, and takes value 0 if she is a Muslim, Christian, Parsi, or Jew. It captures time- invariant characteristics of Hindus. Marriage year fixed effects (δt) control for time-series changes of the outcome variable. State fixed effects (θs) control for time-invariant characteristics of the states.

State-marriage year ﬁxed effects (θs × δt) control for state-speciﬁc changes over time. Religion-state

fixed effects (θs × γr) control for time-invariant characteristics of Hindus across the states. Finally, a vector of observable characteristics (Xirts) controls for respondent’s education (four categories); wealth quintile; dummy variables for whether she lives in an urban region, whether she is a member of a scheduled caste, and whether she is a member of a scheduled tribe; and finally fixed effects for rounds in which respondents are surveyed. Table 2.8 presents regression results for cousin marriage and premarital sex with a subset or all of the controls in the full specification. Being a Hindu is negatively associated with cousin marriage, since Muslims—which constitute 59% of the dropped category—have much higher cousin marriage rates (with 17.1% versus 4.7% among Hindus). Among Hindu women, exposure to the amendment significantly increased the likelihood of marrying with blood relative and decreased the likelihood of having premarital sex. The coefficients are large compared with the coefficients of individual- level controls and considering the mean of dependent variables for the Hindu sample. Table A.2.26 of Appendix A.2 presents the descriptive statistics and description of variables in the regressions. In Table A.2.27 of Appendix A.2, I run two sets of robustness regressions. First, instead of primary sampling units, I cluster the standard errors at the state level. Second, in addition to year and state-year fixed effects based on first marriage year, I also control for those based on birth year. I have not included birth year fixed effects in the baseline regressions because later I use them to create

44Here, I assume that before the amendment, the respondent (and her family) did not know that she would become a joint heir in the future. Later, I will address the potential endogeneity concern.

35 Cousin marriage Premarital sex (Hindu sample mean=0.047, SD=0.212) (Hindu sample mean= 0.090, SD=0.286) VARIABLES (1) (2) (3) (4) (5) (6)

Subject to amendment 0.013*** 0.012*** 0.012*** -0.035*** -0.012*** -0.012*** (0.003) (0.004) (0.004) (0.003) (0.003) (0.003) Hindu -0.119*** -0.099*** -0.099*** 0.003 0.002 0.004 (0.003) (0.009) (0.009) (0.002) (0.007) (0.007)

Caste 0.000 -0.000 (0.001) (0.001) Tribe 0.001 -0.000 (0.002) (0.002) Education -0.001 -0.005*** (0.000) (0.001) Wealth 0.001 -0.003*** (0.000) (0.001) Urban 0.000 -0.007*** (0.001) (0.002) Survey round FE yes yes

Marriage year FE yes yes yes yes yes yes State FE yes yes yes yes yes yes State × Marriage year FE yes yes yes yes State × Hindu FE yes yes yes yes

Observations 461,746 461,746 461,746 419,478 419,478 419,478 R-squared 0.043 0.056 0.056 0.061 0.070 0.071 Number of clusters 22751 22751 22751 25131 25131 25131

Table 2.8: OLS estimates are reported with robust standard errors, clustered at the level of the primary sampling unit, in parentheses. ***, **, and * indicate significance at the 1, 5, and 10% levels. instrumental variables. The results are robust to these specifications. In Table A.2.28 of Appendix A.2, I run two sets of regressions by changing the sample. First, I restrict the sample to only those five states that locally passed similar amendments in the past. The results indicate no significant effect on Hindu women married after 2005 in the five states. Second, I run the regressions with the full sample from all Indian states. The coefficients remain highly significant and in the predicted directions. The identifying assumption of the difference in differences approach is the identical counterfactual trends in treatment and control groups. Here, I provide some robustness and falsification tests on this account. In the first column of Table 2.9, I include an additional variable called "near-treated Hindus", which takes value 1 for Hindu women married for the first time within the two-year span (2004 and 2003) just before the amendment. Its coefficient is insignificant in the regressions on cousin marriage and premarital sex, meaning that there is no evidence that the trend in outcome variables for Hindu women had diverged from non-Hindu women even before the amendment. Next, I conduct a placebo test in time. This involves re-estimating the difference in differences model over the pre-treatment period, but with the assumption that the treatment took effect at an earlier date. Column 2 and 3 show the regression results assuming that the amendment took place in 2003 and 2002 respectively. The difference in difference estimator is statistically insignificant implying zero placebo effect. In the next regressions, I conduct a placebo test in space using un- treated groups. In column 4, I falsely pretend that Christian women have been treated, although they

36 Hindus before Amendment Amendment Christians Muslims the amendment in 2003 in 2002 treated treated VARIABLES (1) (2) (3) (4) (5)

Panel A: Cousin marriage Subject to amend. 0.012*** 0.003 0.003 0.006 -0.006 (0.004) (0.007) (0.006) (0.009) (0.009) Near-treated Hindus 0.004 (0.007) Hindu -0.099*** -0.109*** -0.109*** (0.009) (0.010) (0.010) Christian -0.029 (0.032) Muslim 0.029 (0.032)

Observations 461,746 302,565 302,565 80,768 80,768 R-squared 0.056 0.060 0.060 0.103 0.103 Number of clusters 22751 22684 22684 9584 9584

Panel A: Premarital sex Subject to amend. -0.012*** -0.002 -0.002 0.004 -0.004 (0.003) (0.006) (0.005) (0.011) (0.011) Near-treated Hindus -0.002 (0.006) Hindu 0.004 0.004 0.004 (0.007) (0.007) (0.007) Christian 0.011 (0.031) Muslim -0.011 (0.031)

Observations 419,478 266,923 266,923 71,669 71,669 R-squared 0.071 0.070 0.070 0.092 0.092 Number of clusters 25131 24996 24996 10381 10381

Marriage year FE yes yes yes yes yes State FE yes yes yes yes yes State × Marriage year FE yes yes yes yes yes State × Hindu FE yes yes yes State × Christian FE yes State × Muslim FE yes Individual-level controls yes yes yes yes yes

Table 2.9: OLS estimates are reported with robust standard errors, clustered at the level of the primary sampling unit, in parentheses. Individual-level controls include caste and tribe dummy, education, wealth, and urban dummy variables, and survey round fixed effects. ***, **, and * indicate significance at the 1, 5, and 10% levels. have not been treated, and I use Muslims as their control group. In column 5, I use Muslims and Christians as treatment and control groups respectively. Again, the results show zero placebo effect. Unlike the past amendments in the five states, under the 2005 amendment both unmarried and married women were eligible for inheritance. Therefore, endogeneity of the year of marriage is less of a concern for the 2005 amendment. As column 1 of Table 2.9 shows, there is no evidence that variables of interest were affected by the anticipation of the 2005 amendment. However, to address any such concerns and to show the robustness of the results, here I follow two strategies. First, in a similar approach to [141], I use an instrumental variable approach where woman’s treatment status

(Tirt)—which is a function of religion-year of marriage cells—is instrumented by ﬁxed effects for

37 each religion-year of birth cell, i.e. γr × λτ where γr is the Hindu dummy and λτ is year of birth fixed effects. Therefore, in this approach, exogenous variation in year of birth is used as a measure of exposure to the treatment. The second approach is similar to [77], [206], and [141]. I compare younger cohorts likely to be subject to the amendment with older cohorts who were likely to have been married by the time the amendment was passed and thus were probably not affected by the amendment in their marriage and premarital sex decisions. I define the treatment group to be Hindu women aged 14 or younger (the 10th percentile of the first marriage age distribution for females) in 2005—when the amendment was passed—and the control group to be women of all religion aged 24 or older (the 90th percentile of the distribution) in 2005. Defined based on the described age cohorts, 97.5% of the treatment group were actually exposed to the amendment, and 98% of the control group were not exposed to the amendment. Table A.2.29 of Appendix A.2 reports the results from the instrumental variable and age cohort comparison approaches. The direction and significance of the coefficients of the exposure to the amendment remain unchanged.

2.6 Conclusion

Theories on how inheritance systems shape family relationships, marriage patterns, and the status of women have been advanced over a century by social scientists. Following this literature, and standing on the shoulders of giants such as Adam Smith, Lewis Morgan, Max Weber, and Jack Goody, I developed two hypotheses with important implications for today’s cultural differences across societies. Due to deep-rooted differences in their geography, subsistence economy, and agricultural and political organization, patrilineal societies ended up with different prevailing inheritance systems. I argue that patrilineal societies that traditionally included women in inheritance have developed practices encouraging inmarriage and controlling women’s sexuality in order to preserve the property within the patrilineage, prevent its fragmentation, and limit conflicting claims on the estate. The findings of the study show that ethnicities, regions, and women whose ancestors practiced female inheritance have experienced higher cousin marriage, endogamy, and arranged marriages, and lower female economic participation and sexual freedom. I also find that women who receive inheritance are more likely to be in arranged and endogamous marriages. To show that these correlations can at least in part be attributed to the causal impact of female inheritance, I use a difference in differences approach to estimate the impact of a reform of inheritance regulations in India that substantially improved women’s rights on land. Consistent with the causal direction stated by the hypotheses, the results indicate that women exposed to the reform had higher cousin marriage and lower premarital sex rates. These findings have several important implications. First, female inheritance imposed by Sharia law might be a major historical factor explaining why today’s Islamic countries experience the highest cousin marriage rates and the lowest female participation rates in the world, and sustain their tribal social organization.

38 Second, there is growing evidence on how tight kinship systems impair the development of an individualistic social psychology; undermine social trust, large-scale cooperation, and democratic institutions; and encourage corruption and conﬂict.45 Cousin and arranged marriages, as means of creating and maintaining kin-based groups such as clans and tribes [121, 187], are considered as important elements of kinship patterns. Therefore, I also add to the literature by highlighting how inheritance systems contribute to the heterogeneity of kinship patterns across societies. For example, it has been suggested that the Catholic Church’s prohibitions on cousin marriages and promotion of consensual (or love) marriages played an important role in dismantling tribes and clans in Europe and stimulating its divergent development [115, 143, 82, 160, 121, 187, 231]. It is true that cousin and arranged marriage rates have been historically low in Christian countries. However, these low cousin and arranged marriage rates are also consistent with the fact that by putting no emphasis on female inheritance, European inheritance systems did not create economic incentives for cousin or arranged marriages, and provided an incentive-compatible institutional environment for the Church’s marriage policies. Finally, under patrilineal restrictions on women’s empowerment in developing countries, the mere enacting of female inheritance might create unintended consequences for young women’s sexual lives and marital choices.

45See, e.g., [266], [85], [14], [122], [7], [230], [84], [190], [231], [193]

39 Bibliography

[1] Klaus Abbink and Heike Hennig-Schmidt. Neutral versus loaded instructions in a bribery experiment. Experimental Economics, 9(2):103–121, 2006.

[2] Klaus Abbink, Bernd Irlenbusch, and Elke Renner. An experimental bribery game. Journal of Law, Economics, and Organization, 18(2):428–454, 2002.

[3] Mary Abbott. Family ties: English families 1540-1920. Routledge, 2013.

[4] Bina Agarwal. Gender and legal rights in agricultural land in india. Economic and Political Weekly, pages A39–A56, 1995.

[5] Leila Ahmed. Women and gender in Islam: Historical roots of a modern debate. Yale University Press, 1992.

[6] Anna Aizer. The gender wage gap and domestic violence. American Economic Review, 100(4):1847–59, 2010.

[7] Mahsa Akbari, Duman Bahrami-Rad, and Erik O Kimbrough. Kinship, fractionalization and corruption. SSRN Working Paper, 2847222, 2016.

[8] Seyed Mohammad Akrami and Zahra Osati. Is consanguineous marriage religiously encouraged? islamic and iranian considerations. Journal of Biosocial Science, 39(02):313–316, 2007.

[9] Vivi Alatas, Lisa Cameron, Ananish Chaudhuri, Nisvan Erkal, and Lata Gangadharan. Sub- ject pool effects in a corruption experiment: A comparison of indonesian public servants and indonesian students. Experimental Economics, 12(1):113–132, 2009.

[10] Alberto Alesina, Yann Algan, Pierre Cahuc, and Paola Giuliano. Family values and the regulation of labor. Journal of the European Economic Association, 2015.

[11] Alberto Alesina, Benedetta Brioschi, and Eliana La Ferrara. Violence against women: A cross-cultural analysis for africa. NBER Working Paper, 21901, 2016.

[12] Alberto Alesina, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. Fractionalization. Journal of Economic growth, 8(2):155–194, 2003.

[13] Alberto Alesina and Paola Giuliano. The power of the family. Journal of Economic growth, 15(2):93–125, 2010.

[14] Alberto Alesina and Paola Giuliano. Family ties and political participation. Journal of the European Economic Association, 9(5):817–839, 2011.

40 [15] Alberto Alesina and Paola Giuliano. Family ties. In Philippe Aghion and Steven N. Durlauf, editors, Handbook of Economic Growth, volume 2A, pages 177–215. North Holland: Nether- lands, 2014.

[16] Alberto Alesina, Paola Giuliano, and Nathan Nunn. On the origins of gender roles: Women and the plough. The Quarterly journal of economics, 128(2):469–530, 2013.

[17] Alberto Alesina and Eliana La Ferrara. Ethnic diversity and economic performance. Journal of Economic Literature, 43:762–800, 2005.

[18] Lee J Alston and Morton Owen Schapiro. Inheritance laws across colonies: Causes and consequences. The Journal of Economic History, 44(02):277–287, 1984.

[19] Soﬁa Amaral. Do improved property rights decrease violence against women in india? SSRN Working Paper, 2504579, 2017.

[20] AALL American Association of Law Libraries and Arturo A Flores. Foreign law: Current sources of codes and basic legislation in jurisdictions of the world. Littleton, Colo.: FB Rothman, 1989.

[21] Dan Anderberg, Helmut Rainer, Jonathan Wadsworth, and Tanya Wilson. Unemployment and domestic violence: Theory and evidence. The Economic Journal, 126(597):1947–1979, 2016.

[22] Siwan Anderson. Legal origins and female hiv. American Economic Review, forthcoming, 2018.

[23] Siwan Anderson and Garance Genicot. Suicide and property rights in india. Journal of Development Economics, 114:64–78, 2015.

[24] Kenneth RM Anthony. Agricultural change in tropical Africa. Cornell University Press, 1978.

[25] Nava Ashraf, Natalie Bau, Nathan Nunn, and Alessandra Voena. Bride price and female education. NBER Working Paper, 22417, 2018.

[26] Quamrul Ashraf and Oded Galor. The’out of africa’hypothesis, human genetic diversity, and comparative economic development. The American Economic Review, 103(1):1, 2013.

[27] St. Augustine. The City of God Against the Pagans (trans: R. W. Dyson). Cambridge Univer- sity Press, 1998.

[28] Drew H Bailey, Kim R Hill, and Robert S Walker. Fitness consequences of spousal relatedness in 46 small-scale societies. Biology letters, 10(5):20140160, 2014.

[29] Edward C Banﬁeld. The moral basis of a backward society. New York: Glencol, 3:4, 1958.

[30] Abigail Barr and Danila Serra. The effects of externalities and framing on bribery in a petty corruption experiment. Experimental Economics, 12(4):488–503, 2009.

[31] Abigail Barr and Danila Serra. Corruption and culture: An experimental analysis. Journal of Public Economics, 94(11):862–869, 2010.

41 [32] David B Barrett. World Christian encyclopedia: a comparative study of churches and religions in the modern world, AD 1900-2000. Oxford University Press, USA, 1982.

[33] Herbert Barry. Ethnographic atlas xxviii. Ethnology, 19(2):245–263, 1980.

[34] Anke Becker. Herding and male dominance. SSRN Working Paper, 3069684, 2017.

[35] Sascha O Becker and Ludger Woessmann. Luther and the girls: Religious denomination and the female education gap in nineteenth-century prussia. The Scandinavian Journal of Economics, 110(4):777–805, 2008.

[36] Jens Beckert. Inherited wealth. Princeton University Press, 2008.

[37] Jeanet Sinding Bentzen, Nicolai Kaarsen, and Asger Moll Wingender. Irrigation and autoc- racy. European Economic Association. Journal, 2015.

[38] Joyce Berg, John Dickhaut, and Kevin McCabe. Trust, reciprocity, and social history. Games and economic behavior, 10(1):122–142, 1995.

[39] Theodore C Bergstrom. The algebra of assortative encounters and the evolution of cooperation. International Game Theory Review, 5(03):211–228, 2003.

[40] Lutz K Berkner. The stem family and the developmental cycle of the peasant household: An eighteenth-century austrian example. The American Historical Review, 77(2):398–418, 1972.

[41] Chris Bidner and Mukesh Eswaran. A gender-based theory of the origin of the caste system of india. Journal of Development Economics, 114:142–158, 2015.

[42] Alan Bittles and M. H. Black. Global patterns and tables of consanguinity. 2015.

[43] Alan H Bittles. Consanguinity in context, volume 63. Cambridge University Press, 2012.

[44] Keith Blackburn, Kyriakos Neanidis, Maria Paola Rana, et al. Organized crime, corruption and growth: Theory and evidence. Technical report, Economics, The Univeristy of Manch- ester, 2015.

[45] Dmitri Bondarenko, Alexander Kazankov, Daria Khaltourina, and Andrey Korotayev. Ethno- graphic atlas xxxi: Peoples of easternmost europe. Ethnology, pages 261–289, 2005.

[46] Robert Borofsky. Making history: Pukapukan and anthropological constructions of knowledge. Cambridge University Press, 1987.

[47] Samuel Bowles and Dorrit Posel. Genetic relatedness predicts south african migrant workers’ remittances to their families. Nature, 434(7031):380–383, 2005.

[48] S.I. Bruk and V.S. Apenchenko, editors. Atlas Narodov. Moscow: Miklukho-Maklai Eth- nological Institute at the Department of Geodesy and Cartography of the State Geological Committee of the Soviet Union, 1964.

[49] Ulf Brunnbauer. Descent or territoriality: inheritance and family forms in the late ottoman and early post-ottoman balkans. Distinct Inheritances. Property, family and Community in a Changing Europe, pages 181–205, 2003.

42 [50] Paolo Buonanno and Paolo Vanin. Social closure, surnames and crime. Quaderni-Working Paper DSE, 2015.

[51] Lisa Cameron, Ananish Chaudhuri, Nisvan Erkal, and Lata Gangadharan. Propensities to engage in and punish corrupt behavior: Experimental evidence from australia, india, indonesia and singapore. Journal of Public Economics, 93(7):843–851, 2009.

[52] Pamela Campa, Michel Seraﬁnelli, et al. Politico-economic regimes and attitudes: Female workers under state-socialism. 2016.

[53] J Edgardo Campos, Donald Lien, and Sanjay Pradhan. The impact of corruption on investment: Predictability matters. World development, 27(6):1059–1067, 1999.

[54] Eliana Carranza. Soil endowments, female labor force participation, and the demographic deﬁcit of women in india. American Economic Journal: Applied Economics, 6(4):197–225, 2014.

[55] Marco Casari and Maurizio Lisciandra. Gender discrimination in property rights: six centuries of commons governance in the alps. The Journal of Economic History, 76(02):559– 594, 2016.

[56] Luigi Luca Cavalli-Sforza, Paolo Menozzi, and Alberto Piazza. The history and geography of human genes. Princeton university press, 1994.

[57] Luigi Luca Cavalli-Sforza, Antonio Moroni, and Gianna Zei. Consanguinity, inbreeding, and genetic drift in Italy, volume 39. Princeton University Press, 2004.

[58] Nicholas Charron, Lewis Dijkstra, and Victor Lapuente. Regional governance matters: quality of government within european union member states. Regional Studies, 48(1):68–90, 2014.

[59] Nicholas A Christakis and James H Fowler. Friendship and natural selection. Proceedings of the National Academy of Sciences, 111(Supplement 3):10796–10801, 2014.

[60] CY Cyrus Chu. Primogeniture. Journal of Political Economy, 99(1):78–99, 1991.

[61] Swee-Hoon Chuah, Reema Fahoum, and Robert Hoffmann. Fractionalization and trust in india: A ﬁeld-experiment. Economics Letters, 119(2):191–194, 2013.

[62] EUA CIA. The world factbook (published online). Central Intelligence Agency, Washington, DC, 1996.

[63] Nevill Colclough. Invisible patrimonies–capitoli matrimoniali in ancien régime ascoli. Dis- tinct inheritances: property, family and community in a changing Europe, pages 145–158, 2003.

[64] John W Cole and Eric R Wolf. The hidden frontier: ecology and ethnicity in an alpine valley. Univ of California Press, 1999.

[65] Robert K Colwell. Predictability, constancy, and contingency of periodic phenomena. Ecol- ogy, 55(5):1148–1153, 1974.

43 [66] Timothy G Conley. Gmm estimation with cross sectional dependence. Journal of economet- rics, 92(1):1–45, 1999.

[67] Youssef Courbage and Emmanuel Todd. A convergence of civilizations: The transformation of Muslim societies around the world. Columbia University Press, 2014.

[68] Donald Cox and Marcel Fafchamps. Extended family and kinship networks: economic insights and evolutionary directions. Handbook of development economics, 4:3711–3784, 2007.

[69] Klaus Deininger, Aparajita Goyal, and Hari Nagarajan. Women’s inheritance rights and intergenerational transmission of resources in india. Journal of Human Resources, 48(1):114–141, 2013.

[70] Alfredo Del Monte and Erasmo Papagni. Public expenditure, corruption, and economic growth: the case of italy. European Journal of Political Economy, 17(1):1–16, 2001.

[71] Alfredo Del Monte and Erasmo Papagni. The determinants of corruption in italy: Regional panel data analysis. European Journal of Political Economy, 23(2):379–396, 2007.

[72] Alfredo Del Monte and Erasmo Papagni. The determinants of corruption in italy: Regional panel data analysis. European Journal of Political Economy, 23(2):379–396, 2007.

[73] Adriana Di Liberto and Marco Sideri. Past dominations, current institutions and the italian regional economic performance. European Journal of Political Economy, 38:12–41, 2015.

[74] Oguzhan C Dincer. Ethnic and religious diversity and corruption. Economics Letters, 99(1):98–102, 2008.

[75] Matthias Doepke and Michele Tertilt. Does female empowerment promote economic development? Technical report, 2016.

[76] Gertrude E Dole. The lineage pattern of kinship nomenclature: Its signiﬁcance and development. Southwestern Journal of Anthropology, pages 36–62, 1965.

[77] Esther Duﬂo. Schooling and labor market consequences of school construction in indonesia: Evidence from an unusual policy experiment. American economic review, 91(4):795–813, 2001.

[78] Gilles Duranton, Andrés Rodríguez-Pose, and Richard Sandall. Family types and the persistence of regional disparities in europe. Economic Geography, 85(1):23–47, 2009.

[79] William Easterly and Ross Levine. Africa’s growth tragedy: policies and ethnic divisions. The Quarterly Journal of Economics, pages 1203–1250, 1997.

[80] Lena Edlund. Cousin marriage is not choice: Muslim marriage and underdevelopment. AEA Papers and Proceedings, 108:353–57, 2018.

[81] Charles Efferson, Rafael Lalive, and Ernst Fehr. The coevolution of cultural groups and ingroup favoritism. Science, 321(5897):1844–1849, 2008.

[82] Robert B Ekelund, Robert D Tollison, Gary M Anderson, FH Robert, Audrey B Davidson, et al. Sacred trust: the medieval church as an economic ﬁrm. Oxford University Press, 1996.

44 [83] Nasr G Elbahnasawy and Charles F Revier. The determinants of corruption: Cross-country- panel-data analysis. The Developing Economies, 50(4):311–333, 2012.

[84] Benjamin Enke. Kinship systems, cooperation and the evolution of culture. NBER Working Paper, 23499, 2017.

[85] John Ermisch and Diego Gambetta. Do strong family ties inhibit trust? Journal of Economic Behavior & Organization, 75(3):365–376, 2010.

[86] Ernst Fehr and Urs Fischbacher. The nature of human altruism. Nature, 425(6960):785–791, 2003.

[87] Carl H. Fellner and John R. Marshall. Kidney donors revisited. In Eds. J. Philippe Rushton and Richard M. Sorrentino, editors, Altruism and Helping Behavior: Social, Personality and Developmental Perspectives. Lawrence Erlbaum Associates: New Jersey, 1981.

[88] Nadia Fiorino, Emma Galli, and Ilaria Petrarca. Corruption and growth: Evidence from the italian regions. European Journal of Government and Economics, 1(2):126–144, 2012.

[89] Raymond Fisman and Edward Miguel. Corruption, norms, and legal enforcement: Evidence from diplomatic parking tickets. Journal of Political economy, 115(6):1020–1048, 2007.

[90] James Foreman-Peck and Peng Zhou. Late marriage as a contributor to the industrial revolution in england. The Economic History Review, 2017.

[91] Robin Fox. Kinship and Marriage. Penguin Books: Middlesex, UK, 1967.

[92] Patrick Francois and Jan Zabojnik. Trust, social capital and economic development. Journal of the European Economic Association, 3(1):51–94, 2005.

[93] Feng Fu, Corina E Tarnita, Nicholas A Christakis, Long Wang, David G Rand, and Martin A Nowak. Evolution of in-group favoritism. Scientiﬁc Reports, 2(460), 2012.

[94] Francis Fukuyama. Trust: The Social Virtues and the Creation of Prosperity. New York: Free Press, 1995.

[95] Donna R Gabaccia and Franca Iacovetta. Women, gender and transnational lives: Italian workers of the world. University of Toronto Press, 2002.

[96] Simon Gächter and Jonathan F. Schulz. Intrinsic honesty and the prevalence of rule violations across societies. Nature, 531(7595):496–499, 03 2016.

[97] Vincenzo Galasso and Paola Profeta. When the state mirrors the family: the design of pension systems. CEPR Discussion Paper No. DP8723, 2011.

[98] John Luke Gallup, Jeffrey D Sachs, and Andrew D Mellinger. Geography and economic development. International Regional Science Review, 22(2):179–232, 1999.

[99] A. Gardner and A.W. Stuart. Spite. Current Biology, 16(17):R662–R664, 2006.

[100] Thomas R. Garth. Kinship terminology, marriage practices and behavior toward kin among the atsugewi. American Anthropologist, 46(3):348–361, 1944.

[101] JH Gillespie. Population genetics: a concise guide. Johns Hopkins University Press, 2004.

45 [102] Paola Giuliano. The role of women in society: from preindustrial to modern times. CESifo Economic Studies, 61(1):33–52, 2015.

[103] Paola Giuliano. Gender: A historical perspective. NBER Working Paper, 23635, 2017.

[104] Paola Giuliano and Nathan Nunn. Ancestral characteristics of modern populations, 2013.

[105] Edward L Glaeser and Raven E Saks. Corruption in america. Journal of Public Economics, 90(6):1053–1072, 2006.

[106] Bernt Glatzer. The pashtun tribal system. Concept of tribal society, pages 265–282, 2002.

[107] Uri Gneezy, Kenneth L Leonard, and John A List. Gender differences in competition: Evi- dence from a matrilineal and a patriarchal society. Econometrica, 77(5):1637–1664, 2009.

[108] Miriam A Golden and Lucio Picci. Proposal for a new measure of corruption, illustrated with italian data. Economics & Politics, 17(1):37–75, 2005.

[109] Walter Goldschmidt and Evalyn Jacobson Kunkel. The structure of the peasant family. Amer- ican Anthropologist, 73(5):1058–1076, 1971.

[110] Melvyn C Goldstein. When brothers share a wife. Natural History, 96(3):109–112, 1987.

[111] Binglin Gong and Chun-Lei Yang. Gender differences in risk attitudes: Field experiments on the matrilineal mosuo and the patriarchal yi. Journal of economic behavior & organization, 83(1):59–65, 2012.

[112] Jack Goody. Inheritance, property, and marriage in africa and eurasia. Sociology, 3(1):55–76, 1969.

[113] Jack Goody. Production and reproduction: A comparative study of the domestic domain. Cambridge Studies in Social Anthropology, 17, 1976.

[114] Jack Goody. Production and reproduction: a comparative study of the domestic domain. Number 17. Cambridge University Press, 1976.

[115] Jack Goody. The development of the family and marriage in Europe. Cambridge University Press, 1983.

[116] Jack R Goody, Joan Thirsk, and Edward Palmer Thompson. Family and inheritance. Past and Present Publications (USA), 1976.

[117] Philip Gounev and Tihomir Bezlov. Examining the links between organised crime and corruption. Center for the Study of Democracy, 2010.

[118] Jesse Graham, Jonathan Haidt, and Brian Nosek. The Moral Foundations Quiz. www.yourmorals.org, 2008.

[119] J Patrick Gray. Ethnographic atlas codebook. World Cultures, 10(1):86–136, 1998.

[120] J Patrick Gray. A corrected ethnographic atlas. World Cultures, 10(1):24–85, 1999.

[121] Avner Greif. Family structure, institutions, and growth: the origins and implications of western corporations. The American Economic Review, 96(2):308–312, 2006.

46 [122] Avner Greif and Guido Tabellini. The clan and the corporation: Sustaining cooperation in china and europe. SSRN Working Paper, 2565120, 2015.

[123] Avner Greif and Guido Tabellini. The clan and the corporation: Sustaining cooperation in china and europe. Available at SSRN 2565120, 2015.

[124] Veronika Grimm and Friederike Mengel. Cooperation in viscous populations—experimental evidence. Games and Economic Behavior, 66(1):202–220, 2009.

[125] Pauline Grosjean and Rose Khattar. It’s raining men! hallelujah? SSRN Working Paper, 2445285, 2015.

[126] Eleonora Guarnieri and Helmut Rainer. Female empowerment and male backlash. CESifo Working Paper, 7009, 2018.

[127] Luigi Guiso, Paola Sapienza, and Luigi Zingales. People’s opium? religion and economic attitudes. Journal of monetary economics, 50(1):225–282, 2003.

[128] EV Gunnemark. Countries, peoples, and their languages: The geolinguistic handbook. summer institute of linguistics. Inc., Dallas, Texas, 1991.

[129] W. D. Hamilton. The genetical evolution of social behaviour. i. Journal of Theoretical Biology, 7(1):1–16, 1964.

[130] William D Hamilton. The genetical evolution of social behaviour. ii. Journal of Theoretical Biology, 7(1):17–52, 1964.

[131] William D Hamilton. Innate social aptitudes of man: an approach from evolutionary genetics. Biosocial anthropology, 133:155, 1975.

[132] Ross A Hammond and Robert Axelrod. The evolution of ethnocentrism. Journal of Conﬂict Resolution, 50(6):926–936, 2006.

[133] Casper Worm Hansen, Peter Sandholt Jensen, and Christian Volmar Skovsgaard. Modern gender roles and agricultural history: the neolithic inheritance. Journal of Economic Growth, 20(4):365–404, 2015.

[134] Mariaﬂavia Harari. WomenâA˘ Zs´ inheritance rights and bargaining power: Evidence from kenya. Department of Economics of MIT, pages 1–47, 2014.

[135] Henry Harpending and Alan Rogers. Fitness in stratiﬁed societies. Ethology and sociobiology, 11(6):497–509, 1990.

[136] Henry C Harpending. Kinship and population subdivision. Population and Environment, 24(2):141–147, 2002.

[137] Stevan Harrell. Human families. Westview Press, 1997.

[138] IPDJ Harris, PD Jones, TJ Osborn, and DH Lister. Updated high-resolution grids of monthly climatic observations–the cru ts3. 10 dataset. International Journal of Climatol- ogy, 34(3):623–642, 2014.

47 [139] hbdchick. Big summary post on the hajnal line. https://hbdchick.wordpress.com/2014/03/10/big-summary-post-on-the-hajnal-line/, March 2014.

[140] Patrick Heady and Hannes Grandits, editors. Distinct inheritances: property, family and community in a changing Europe. LIT Verlag Münster, 2003.

[141] Rachel Heath and Xu Tan. Intrahousehold bargaining, female autonomy, and labor supply: Theory and evidence from india. University of Washington, 2016.

[142] Joseph Henrich. The secret of our success: how culture is driving human evolution, domesti- cating our species, and making us smarter. Princeton University Press, 2015.

[143] David Herlihy. Medieval households. Harvard University Press, 1985.

[144] Benedikt Herrmann, Christian Thöni, and Simon Gächter. Antisocial punishment across societies. Science (New York, N.Y.), 319(5868):1362–1367, 2008.

[145] Ashley D. Hoben, Abraham P. Buunk, Corey L. Fincher, Randy Thornhill, and Mark Schaller. On the adaptive origins and maladaptive consequences of human inbreeding: Para- site prevalence, immune functioning, and consanguineous marriage. Evolutionary Psychol- ogy, 8(4):658 – 676, 2010.

[146] Ladislav Holy.` Kinship, honour, and solidarity: cousin marriage in the Middle East. Manch- ester University Press, 1989.

[147] Martha C Howell. Commerce before capitalism in Europe, 1300-1600. Cambridge University Press, 2010.

[148] David Hume. A Treatise of Human Nature. Oxford: Oxford University Press, (2003) edition, 1740.

[149] John Kenneth Hyde. Society and politics in medieval Italy: the evolution of the civil life, 1000-1350. Not Avail, 1973.

[150] Shabnam Ishaque. Islamic principles on adoption: examining the impact of illegitimacy and inheritance related concerns in context of a child’s right to an identity. International Journal of Law, Policy and the Family, 22(3):393–420, 2008.

[151] Lutﬁ Jaber, Gabrielle J Halpern, and Mordechai Shohat. The impact of consanguinity worldwide. Community Genetics, 1(1):12–17, 1998.

[152] Tarun Jain. Where there is a will fertility behavior and sex bias in large families. Journal of Human Resources, 49(2):393–423, 2014.

[153] Allen W Johnson and Timothy K Earle. The evolution of human societies: from foraging group to agrarian state. Stanford University Press, 2000.

[154] Noel D Johnson and Alexandra Mislin. How much should we trust the world values survey trust question? Economics Letters, 116(2):210–212, 2012.

[155] Doug Jones. Socially enforced nepotism: How norms and reputation can amplify kin altruism. PLoS ONE, 11(6):1–13, 06 2016.

48 [156] Karl Kaser. Power and inheritance: Male domination, property, and family in eastern europe, 1500–1900. Distinct Inheritances. Property, family and Community in a Changing Europe, pages 53–63, 2003. [157] David I Kertzer. The family in Italy from antiquity to the present. Yale University Press, 1993. [158] Kathryn R Kirby, Russell D Gray, Simon J Greenhill, Fiona M Jordan, Stephanie Gomes-Ng, Hans-Jörg Bibiko, Damián E Blasi, Carlos A Botero, Claire Bowern, Carol R Ember, et al. D-place: A global database of cultural, linguistic and environmental diversity. PLoS One, 11(7):e0158391, 2016. [159] Andrey Korotayev. Parallel-cousin (fbd) marriage, islamization, and arabization. Ethnology, pages 395–407, 2000. [160] Andrey Korotayev, Alexander Kazankov, Svetlana Borinskaya, Daria Khaltourina, and Dmitri Bondarenko. Ethnographic atlas xxx: peoples of siberia. Ethnology, 43(1):83–92, 2004. [161] Andrey V Korotayev. Unilineal descent organization and deep christianization: A cross- cultural comparison. Cross-Cultural Research, 37(1):133–157, 2003. [162] Maurice Kugler, Thierry Verdier, and Yves Zenou. Organized crime, corruption and punishment. Journal of Public Economics, 89(9):1639–1663, 2005. [163] Timur Kuran. The long divergence: How Islamic law held back the Middle East. Princeton University Press, 2012. [164] Rafael La Porta, Florencio Lopez-de Silanes, Andrei Shleifer, and Robert Vishny. The quality of government. Journal of Law, Economics, and organization, 15(1):222–279, 1999. [165] Rafael La Porta, Andrei Shleifer, and RW Vishny. Law and ﬁnance. Journal of Political Economy, 106:1113–1155, 1998. [166] Johann Graf Lambsdorff. How corruption affects persistent capital ﬂows. Economics of Governance, 4(3):229–243, 2003. [167] Johann Graf Lambsdorff. Behavioural and experimental economics as a guidance to anticor- ruption. In Danila Serra and Leonard Wantchekon, editors, New advances in experimental research on corruption, volume 15, pages 279–300. Emerald Group Publishing, 2012. [168] Iosif Lazaridis, Dani Nadel, Gary Rollefson, Deborah C. Merrett, Nadin Rohland, Swa- pan Mallick, Daniel Fernandes, Mario Novak, Beatriz Gamarra, Kendra Sirak, Sarah Con- nell, Kristin Stewardson, Eadaoin Harney, Qiaomei Fu, Gloria Gonzalez-Fortes, Eppie R. Jones, Songül Alpaslan Roodenberg, György Lengyel, Fanny Bocquentin, Boris Gasparian, Janet M. Monge, Michael Gregg, Vered Eshed, Ahuva-Sivan Mizrahi, Christopher Meik- lejohn, Fokke Gerritsen, Luminita Bejenaru, Matthias Blüher, Archie Campbell, Gianpiero Cavalleri, David Comas, Philippe Froguel, Edmund Gilbert, Shona M. Kerr, Peter Kovacs, Johannes Krause, Darren McGettigan, Michael Merrigan, D. Andrew Merriwether, Seamus O’Reilly, Martin B. Richards, Ornella Semino, Michel Shamoon-Pour, Gheorghe Stefanescu, Michael Stumvoll, Anke Tönjes, Antonio Torroni, James F. Wilson, Loic Yengo, Nelli A. Hovhannisyan, Nick Patterson, Ron Pinhasi, and David Reich. Genomic insights into the origin of farming in the ancient near east. Nature, 536(7617):419–424, 08 2016.

49 [169] FM LeBar and GN Appell. Ethnic groups of insular southeast asia. new haven. Human Relations Area Files Press, 1972.

[170] David Levine and Michael Kevane. Are investments in daughters lower when daughters move away? evidence from indonesia. World Development, 31(6):1065–1084, 2003.

[171] Alexandra Lewis. Security, Clans and Tribes: Unstable Governance in Somaliland, Yemen and the Gulf of Aden. Palgrave Macmillan: London, 2014.

[172] M. Pau Lewis, Gary F. Simons, and Charles D. Fennig. Ethnologue: Languages of the World, Nineteenth edition. Dallas, Texas: SIL International, 2016.

[173] Seymour Martin Lipset and Gabriel Salman Lenz. Corruption, culture, and markets. George Mason University, 1999.

[174] Sara Lowes. Kinship systems, gender norms, and household bargaining: Evidence from the matrilineal belt. Working Paper, Harvard mimeo, 2016.

[175] Katherine A Lynch. Individuals, families, and communities in Europe, 1200-1800: the urban foundations of Western society, volume 37. Cambridge University Press, 2003.

[176] Ramzi Mabsout and Irene Van Staveren. Disentangling bargaining power from individual and household level to institutions: Evidence on womenâA˘ Zs´ position in ethiopia. World Development, 38(5):783–796, 2010.

[177] Elainie A Madsen, Richard J Tunney, George Fieldman, Henry C Plotkin, Robin IM Dun- bar, Jean-Marie Richardson, and David McFarland. Kinship and altruism: A cross-cultural experimental study. British Journal of Psychology, 98(2):339–359, 2007.

[178] Aparna Mathur and Sita Slavov. Escaping domestic violence: Empowering women through employment, earnings and wealth in india. AEI Economic Policy Working Paper Series, 2013.

[179] Paolo Mauro. Corruption and growth. The Quarterly Journal of Economics, pages 681–712, 1995.

[180] Paolo Mauro. Corruption and the composition of government expenditure. Journal of Public economics, 69(2):263–279, 1998.

[181] Hans Medick. The proto-industrial family economy: The structural function of household and family during the transition from peasant society to industrial capitalism. Social History, 1(3):291–315, 1976.

[182] Hans Medick and David Warren Sabean. Interest and emotion: essays on the study of family and kinship. Cambridge University Press, 1984.

[183] Stelios Michalopoulos. The origins of ethnolinguistic diversity. The American Economic Review, 102(4):1508, 2012.

[184] Stelios Michalopoulos, Alireza Naghavi, and Giovanni Prarolo. Islam, inequality and pre- industrial comparative development. Journal of Development Economics, 120:86–98, 2016.

[185] Stelios Michalopoulos, Alireza Naghavi, and Giovanni Prarolo. Trade and geography in the spread of islam. The Economic Journal, 2017.

50 [186] Michael Mitterauer. European kinship systems and household structures: medieval origins. Distinct inheritances. Property, family and community in a changing Europe, pages 35–52, 2003.

[187] Michael Mitterauer. Why Europe?: The Medieval Origins of Its Special Path. University of Chicago Press, 2010.

[188] Anna Momigliano. The town in italy where most women are barred from inheriting land. The Atlantic, 2016.

[189] Lewis Henry Morgan. Systems of consanguinity and afﬁnity of the human family, volume 218. Smithsonian institution, 1871.

[190] Jacob Moscona, Nathan Nunn, and James A Robinson. Keeping it in the family: Lineage organization and the scope of trust in sub-saharan africa. American Economic Review Papers and Proceedings, 107(5):565–71, 2017.

[191] Jacob Moscona, Nathan Nunn, and James A Robinson. Keeping it in the family: Lineage organization and the scope of trust in sub-saharan africa. Technical report, National Bureau of Economic Research, 2017.

[192] Jacob Moscona, Nathan Nunn, and James A Robinson. Social structure and conﬂict: Evi- dence from sub-saharan africa. 2017.

[193] Jacob Moscona, Nathan Nunn, and James A Robinson. Social structure and conﬂict: Evi- dence from sub-saharan africa. NBER Working Paper, 24209, 2018.

[194] Siegfried Hermann Muller. The World’s Living Languages: Basic Facts of Their Structure, Kinship, Location, and Number of Speakers. Ungar, 1964.

[195] George Peter Murdock. World ethnographic sample. American Anthropologist, 59(4):664– 687, 1957.

[196] George Peter Murdock. Ethnographic atlas, installments i-xxvii. Ethnology, pages 1–10, 1962–1971.

[197] George Peter Murdock. Ethnographic atlas: a summary. Ethnology, 6(2):109–236, 1967.

[198] Chie Nakane et al. Kinship and economic organization in rural japan. Kinship and economic organization in rural Japan., 1967.

[199] Mark New, David Lister, Mike Hulme, and Ian Makin. A high-resolution data set of surface climate over global land areas. Climate Research, 21(1):1–25, 2002.

[200] Tin Nguyen, Enjiang Cheng, and Christopher Findlay. Land fragmentation and farm productivity in china in the 1990s. China Economic Review, 7(2):169–180, 1996.

[201] William D Nordhaus. Geography and macroeconomics: New data and new ﬁndings. Pro- ceedings of the National Academy of Sciences of the United States of America, 103(10):3510– 3517, 2006.

[202] Martin A Nowak, Corina E Tarnita, and Edward O Wilson. The evolution of eusociality. Nature, 466(7310):1057–1062, 2010.

51 [203] Nathan Nunn, Emmanuel Akyeampong, Robert Bates, and James A Robinson. Gender and missionary inﬂuence in colonial africa. African Development in Historical Perspective, 2011.

[204] Nathan Nunn and Diego Puga. Ruggedness: The blessing of bad geography in africa. Review of Economics and Statistics, 94(1):20–36, 2012.

[205] Samir Okasha. Biological altruism. In Edward N. Zalta, editor, The Stanford Encyclopedia of Philosophy. Standford University Press, fall 2013 edition, 2013.

[206] Una Okonkwo Osili and Bridget Terry Long. Does female schooling reduce fertility? evidence from nigeria. Journal of development Economics, 87(1):57–75, 2008.

[207] Elinor Ostrom and Charlotte Hess. Private and common property rights. Encyclopedia of law and economics, 2:53–106, 2000.

[208] Trevor J Pemberton, Michael DeGiorgio, and Noah A Rosenberg. Population structure in a comprehensive genomic data set on human microsatellite variation. G3: Genes| Genomes| Genetics, pages g3–113, 2013.

[209] Trevor J Pemberton and Noah A. Rosenberg. Population-genetic inﬂuences on genomic estiamtes of the inbreeding coefﬁcient: A global perspective. Human Heredity, 77(1-4):37– 48, 2014.

[210] Francis Pine. Reproducing the house: kinship, inheritance, and property relations in highland poland. Distinct Inheritances: Property, Family and Community in a Changing Europe, pages 279–295, 2003.

[211] Leendert Plas. Rural gender studies in Europe. Van Gorcum, 1994.

[212] Jean-Philippe Platteau. Institutions, social norms, and economic development, volume 1. Psychology Press, 2000.

[213] Jean-Philippe Platteau and Jean-Marie Baland. Access to land, rural poverty, and public action (Impartible Inheritance Versus Equal Division: a Comparative Perspective Centered on Europe and Subsaharan Africa). Oxford University Press, 2001.

[214] Robert D Putnam, Robert Leonardi, and Raffaella Y Nanetti. Making democracy work: Civic traditions in modern Italy. Princeton university press, 1994.

[215] Nancy Qian. Missing women and the price of tea in china: The effect of sex-speciﬁc earnings on sex imbalance. The Quarterly Journal of Economics, 123(3):1251–1285, 2008.

[216] Sohini Ramachandran, Omkar Deshpande, Charles C. Roseman, Noah A. Rosenberg, Mar- cus W. Feldman, and L. Luca Cavalli-Sforza. 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, 2005.

[217] Navin Ramankutty, Jonathan A Foley, John Norman, and Kevin McSweeney. The global distribution of cultivable lands: current patterns and sensitivity to possible climate change. Global Ecology and biogeography, 11(5):377–392, 2002.

52 [218] Navin Ramankutty, Jonathan A Foley, John Norman, and Kevin McSweeney. The global distribution of cultivable lands: current patterns and sensitivity to possible climate change. Global Ecology and Biogeography, 11(5):377–392, 2002.

[219] Auke Rijpma and Sarah G Carmichael. Testing todd and matching murdock: Global data on historical family characteristics. Economic History of Developing Regions, 31(1):10–46, 2016.

[220] M Fernanda Rivas. An experiment on corruption and gender. Bulletin of Economic Research, 65(1):10–42, 2013.

[221] Janet Roberts. Sociocultural change and communication problems. Study of the role of second languages in Asia, Africa, and Latin America, pages 105–23, 1962.

[222] Alan R Rogers. Evolutionary economics of human reproduction. Ethology and Sociobiology, 11(6):479–495, 1990.

[223] Daniel Rosenblum. Unintended consequences of womenâA˘ Zs´ inheritance rights on female mortality in india. Economic Development and Cultural Change, 63(2):223–248, 2014.

[224] Sanchari Roy. Empowering women? inheritance rights, female education and dowry pay- ments in india. Journal of Development Economics, 114:233–251, 2015.

[225] David W Sabean. Young bees in an empty hive: relations between brothers-in-law in a south german village around 1800. Interest and Emotion: Essays on the Study of Family and Kin- ship. Cambridge: Cambridge University Press. p, pages 171–186, 1984.

[226] Steve Sailer. Cousin marriage conundrum. In Tim Folger and Steven Pinker, editors, The Best American Science and Nature Writing 2004. Mariner Books, 2004.

[227] Assaf Sarid, Oded Galor, and Omer Ozak. Geographical origins and economic consequences of language structures. CESifo Working Paper, 6149, 2017.

[228] Walter Scheidel. Sex and Empire. New York: Oxford University Press, 2009.

[229] William J Schull. Empirical risks in consanguineous marriages: sex ratio, malformation, and viability. American journal of human genetics, 10(3):294, 1958.

[230] Jonathan F Schulz. The churches’ bans on consanguineous marriages, kin-networks and democracy. SSRN Working Paper, 2877828, 2016.

[231] Jonathan F Schulz, Duman Bahrami-Rad, Jonathan Beauchamp, and Joseph Henrich. The origins of weird psychology. SSRN Working Paper, 3201031, 2018.

[232] Martine Segalen. Historical anthropology of the family. Cambridge University Press, 1986.

[233] Danila Serra. Empirical determinants of corruption: A sensitivity analysis. Public Choice, 126(1-2):225–256, 2006.

[234] Mary K Shenk, Mary C Towner, Emily A Voss, and Nurul Alam. Consanguineous marriage, kinship ecology, and market transition. Current Anthropology, 57(S13):S167–S180, 2016.

[235] Tom Siegfried. A Beautiful Math:: John Nash, Game Theory, and the Modern Quest for a Code of Nature. National Academies Press, 2006.

53 [236] Adam Smith. An Inquiry into the Nature and Causes of the Wealth of Nations. Vol. II of The Glasgow Edition of the Works and Correspondence of Adam Smith, 1776.

[237] RM Smith. Land, kinship and life-cycle. Cambridge (UK) Cambridge Univ. Press, 1984.

[238] Enrico Spolaore and Romain Wacziarg. The diffusion of development. The Quarterly Journal of Economics, 124(2):469–529, 2009.

[239] Mysore Narasimhachar Srinivas. The remembered village. Univ of California Press, 1980.

[240] Guido Tabellini. Culture and institutions: Economic development in the regions of europe. Journal of the European Economic Association, 8(4):677–716, 2010.

[241] Guido Tabellini et al. The scope of cooperation: Values and incentives. The Quarterly Journal of Economics, 123(3):905–950, 2008.

[242] Shuhao Tan, Nico Heerink, and Futian Qu. Land fragmentation and its driving forces in china. Land use policy, 23(3):272–285, 2006.

[243] Vito Tanzi and Hamid Reza Davoodi. Roads to nowhere: how corruption in public investment hurts growth, volume 12. International Monetary Fund, 1998.

[244] Edoardo Teso. The long-term effect of demographic shocks on the evolution of gender roles: Evidence from the transatlantic slave trade. SSRN Working Paper, 2927221, 2016.

[245] Joan Thirsk. The european debate on customs of inheritance, 1500-1700. Family and inheritance, pages 177–191, 1976.

[246] Todne Thomas, Asiya Malik, and Rose Wellman. New Directions in Spiritual Kinship: Sacred Ties Across the Abrahamic Religions. Springer, 2017.

[247] Emmanuel Todd. The explanation of ideology: Family structures and social systems. Black- well Oxford, 1985.

[248] Emmanuel Todd. L’invention de l’Europe. Seuil, 1990.

[249] Daniel Treisman. The causes of corruption: a cross-national study. Journal of public economics, 76(3):399–457, 2000.

[250] Ana Tur-Prats. Family types and intimate-partner violence: A historical perspective. Eco- nomics Working Papers 1486, Universitat Pompeu Fabra, 2015.

[251] Edward B Tylor. 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, 1889.

[252] UN. Demographic Yearbook. United Nations: New York, 1995.

[253] Eric M Uslaner. Trust and corruption. In Johann Graf Lambsdorff, Markus Taube, and Matthias Schramm, editors, The New Institutional Economics of Corruption, pages 76–93. Routledge, 2005.

[254] Robert S Walker and Drew H Bailey. Marrying kin in small-scale societies. American Journal of Human Biology, 26(3):384–388, 2014.

54 [255] Guang H Wan and Enjiang Cheng. Effects of land fragmentation and returns to scale in the chinese farming sector. Applied Economics, 33(2):183–194, 2001.

[256] Max Weber. The protestant ethic and the spirit of capitalism, trans. Talcott Parsons (New York: Charles Scribner’s Sons, 1958), 182, 1958.

[257] Max Weber. Economy and society: an outline of interpretive sociology. Vol. 1. University of Calif. Press, 1978.

[258] Karl August Wittfogel. Oriental despotism: A comparative study of total power. Yale Uni- versity Press, 1959.

[259] Michael A. Woodley and Edward Bell. Consanguinity as a major predictor of levels of democracy: A study of 70 nations. Journal of Cross-Cultural Psychology, 44(2):263–280, 2013.

[260] C Geoffrey Woods, James Cox, Kelly Springell, Daniel J Hampshire, Moin D Mohamed, Martin McKibbin, Rowena Stern, F Lucy Raymond, Richard Sandford, Saghira Malik Sharif, et al. Quantiﬁcation of homozygosity in consanguineous individuals with autosomal recessive disease. The American Journal of Human Genetics, 78(5):889–896, 2006.

[261] 1978 World Bank. World development report 1978. Oxford University Press, New York, 1978.

[262] WB World Bank. Helping countries combat corruption: the role of the world bank, 1997.

[263] Worldmark. Worldmark encyclopedia of the nations. Gale Cengage, 1995.

[264] Melanie Meng Xue. High-value work and the rise of women: The cotton revolution and gender equality in china. SSRN Working Paper, 2389218, 2016.

[265] Toshio Yamagishi, Karen S Cook, and Motoki Watabe. Uncertainty, trust, and commitment formation in the united states and japan 1. American Journal of Sociology, 104(1):AJSv104p165–194, 1998.

[266] Toshio Yamagishi and Midori Yamagishi. Trust and commitment in the united states and japan. Motivation and emotion, 18(2):129–166, 1994.

55 Chapter 3

Kinship, Fractionalization and Corruption

(with Mahsa Akbari and Erik Kimbrough)

“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 148, A Treatise of Human Nature, Section 3.2.2.]

3.1 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, [79] found that ethnic fractionalization is positively correlated with corruption; [164], [249] and [12] 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, [233] and [83] found no significant effect of fractionalization on corruption. In addition to the cross-country studies, [105] and [74] found a

56 signiﬁcant relationship between ethnic heterogeneity and corruption across US states. These con- tradictory results have encouraged skepticism [see e.g. 61]. 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 independent variables. The richest specifications in [164], and [12] 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" [68, 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. 129, 130]. 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 3.2), 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

1As 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 [48], and (II) the ethnic, linguistic, and religious fractionalization indices provided by [12]. Both sources deﬁne fractionalization as the probability that two randomly drawn individuals from a country’s population belong to two different groups.

2In 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 [131] for a theoretical treatment, and see hbdchick (2014) for a provocative argument that this might apply to human

57 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 [142]. Our focus is on the effects of in- and out-marriage practices on norms related to corruption. 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 [57], 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 di- 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).

3Many 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 142, 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" [142, p. 153]. See also [155] on feedback between social norms and kin altruism.

58 minished 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 the relationship between ethnic (and linguistic) fractionalization and corruption, consider Table 3.1 derived from [12] 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. 171, 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 0-5% 5-15% 15-35% 35-55% 55-75% Yemen Saudi Arabia ...... Tunisia Bangladesh Corruption index 0-20% 20-40% 40-60% 60-80% 80-100% ...... Saudi Arabia Bangladesh Yemen Tunisia

Table 3.1: 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 3.2 [also derived from 12], which lists a number of countries, all of which are highly ethnically and linguistically fractionalized:

Fractionalization index 0-5% 5-15% 15-35% 35-55% 55-75% ...... Switzerland Canada ...... Belgium Luxembourg Iraq Iran Uzbekistan Pakistan Corruption index 0-20% 20-40% 40-60% 60-80% 80-100% Canada Belgium Iran Iraq Luxembourg Pakistan Uzbekistan Switzerland

Table 3.2: 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

4[247] 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).

5The fractionalization index in the table is the simple average of ethnic fractionalization and linguistic fractionalization from [12], and the range of the fractionalization index in the table is the same as [179]’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).

59 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" [106, p. 3]. Similar arguments contrast Switzerland, Luxembourg, and Belgium on one hand to Iran, Iraq, and Uzbekistan on the other.6 Viewed 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.

6See also [226] on clans, corruption and state-building in Iraq in light of cousin marriage practices.

7E.g. see [2, 51, 9, 31, 220].

60 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 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 co-ethnics 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. 115, 57, 145, 254], 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" [115, p. 56]. Consanguineous marriages “had historically provided one means of creating and maintaining kinship groups–such as clans, lineages, and tribes" [121, 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 [161] and [121] 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 [159]. 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 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

61 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 may be causal. In particular, for our Italian data, 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. The IV estimates provide evidence consistent with our reduced form analysis, suggesting that in-marriage practices are an important potential cause of variation in corruption.8 Our argument is related to the literature that distinguishes between generalized morality and limited morality (or amoral familism) [29, 212, 241]. 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 [85, 14, 15], collectivism versus individualism [266, 265], and clan versus corporation [123, 121]. 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. 259, 230, 191, 192, 84]. [191, 192] 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 [259] and [230] document a negative relationship between rates of consanguineous marriage and levels of democracy. Also related is a recent paper by [84], 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, seg- mented 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

8In Appendix B.5.2 we also report an instrumental variables strategy for our cross-country analysis using an instrument based on differences in kin terminology, which also provides evidence consistent with our primary analyses.

62 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.

3.2 Theory and hypotheses

Corruption can be defined as “abuse of public office for private gain" [262, p. 8]. Public office can be abused in hiring for governmental positions, manipulating government procurement or facilitat- ing/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.

3.2.1 A basic model of bribery Seen from a game-theoretic perspective, bribery is a social dilemma much like a trust game [see e.g. 38, 86] 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 [2], the trust game lacks two essential components of bribery: the possibility of negative externalities and the risk of penalty. Figure 3.1a shows our bribery game inspired by [2]. 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 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 (e) 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.

63 Private Agent

No offer Offer t

Ofﬁcial P1 P2 Reject Accept

Ofﬁcial P1 − c P2 No effort Effort

Nature P1 − c − t P2 + t 1 − e e

  P1 − c − t + B 0  P2 + t − e  0 -X ... -X

(a) All players unrelated.

Private Agent

No offer Offer t

Ofﬁcial P1 P2 Reject Accept

Ofﬁcial P1 − c P2 − rpoc No effort Effort

Nature P1 − c − (1 − rpo)t P2 + (1 − rpo)t − rpoc 1 − e e

  P1 − c − (1 − rpo)t − rpoe + B 0 P2 + (1 − rpo)t − rpoc + rpo B − e 0 -X ... -X

(b) Private Agent and Ofﬁcial are relatives.

Figure 3.1: 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

64 favors are reciprocated" [167, p. 280]. One possible source of observed corruption is nepotism, and its antecedent biological or kin altruism.9

3.2.2 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, [129]’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" [205]. 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 fitness10 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 [205, 68]. 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 material support within social networks, cooperation under catastrophic circumstances, membership in cooperative labor units, organ donation rates, etc. [see e.g. 68, 177, 47, 87]. 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

9Of 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.

10Fitness should be thought of as reproductive success, e.g. as the expected number of progeny.

65 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 [205].11 “J. B. S. Haldane once remarked, it would make sense to dive into a river to save two drowning siblings or eight drowning cousins" [235, p. 85]. See Appendix B.1 for further detail.

Bribery game with inclusive fitness Suppose the payoffs in the bribery game are in units of biological reproductive fitness.12 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 rpo represent the relatedness of the private agent and the official. Also, let rpc = 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 3.1b. 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 effort on her behalf with a unique equilibrium strategy if: (1 − e) > 2 po po and he P2+(1−rpo)t−rpoc+rpo B−e accepts the offer if: (1 − e) > P2 . P2+(1−rpo)t−rpoc+rpo B−e (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 optimal strategy of the official; therefore she places trust and offers t in a unique equilibrium strategy if: (1 − e) > P1 . P1−c−(1−rpo)t−rpoe+B Implication 1: All else equal, the official is more likely to accept a bribe and make a corrupt effort as rpo increases.

Implication 2: All else equal, the private agent is more likely to offer a bribe as rpo increases.

Note that while our example sets roc and rpc 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 rpc to vary, the private agent is less likely to offer a bribe as rpc 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

11In 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" [136, 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 [202, p. 1059].

12Of 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.

66 that relatedness among co-ethnics (relative to neighboring groups) is often not far above zero [56], 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.

3.2.3 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 con- sideration), degree of relationship, or geographic locality" [101, 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.13 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 131, for a theoretical analysis and Appendix B.1 for some empirical evidence]. In a cross-cultural ethnographic tabulation due to [197] and [119], a total of 476 out of 1024 societies for which we have data either permitted or favored first and/or second-cousin marriage, and estimates suggest that roughly 10% of marriages around the world today are consanguineous [43].14,15 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 3.3 shows some of the diversity of religious attitudes toward cousin marriage around the world, and Appendix B.2 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. 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

13Related literature suggests that assortative matching (and mating) can encourage the evolution of (local) cooperation and favoritism. See e.g. [39, 124] for theoretical and experimental evidence on cooperation, and also [132, 81, 93] on favoritism.

14While the negative health effects of consanguinity are well-known (e.g. increased risk of autosomal-recessive disor- ders), some believe that there are countervailing positive effects as well [151].

15In 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 [209, p. 38]. See Appendix B.1 for more information.

67 Attitude toward Religion Sect Cousin Marriage

Judaism Ashkenzi Permitted Sephardi Permitted Christianity Coptic Orthodox Permitted Eastern Orthodox Proscribed Protestant Permitted Roman Catholic Diocesan approval req. Islam Sunni Permitted Shia Permitted Hinduism Indo-European Proscribed Dravidian Permitted Buddhism Permitted Sikhism Proscribed Confucian/Taoist Partially permitted Zoroastrian/Parsi Permitted

Table 3.3: Religious attitudes to consanguineous marriage [43, p. 14].

Catholic country. [57] 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. Nev- ertheless, consanguinity remains rare in those countries. In the United States, cousin marriage is illegal in 25 states, though its frequency remains low 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, [115] 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. [254] show that among forager peoples, such marriages are rare due to norms of exogamy and fis- sion/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 [28] 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 [145] 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.

68 3.3 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 marriage as our measure of sub-ethnic fractionalization and we regress measures of corruption on consanguinity. As discussed below and detailed in Appendix B.3, [42] 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 [57] 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.

3.3.1 Cross-country analysis Figure 3.2 displays average ICRG institutional quality data from 1984-2011 alongside consanguinity data from [42]. 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.16 Our empirical strategy follows [164] and [12] who attempted to address the endogeneity of corruption by focusing on “(reasonably) exogenous sources of variation" [164, 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). 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 in-

16Results are robust to using Tobit to account for the dependent variable being restricted to the interval [0,6].

69 (a) Corruption

(b) Consanguinity

65.8

Figure 3.2: Corruption and consanguinity around the world. come is a potential determinant of corruption, corruption itself can reduce per capita income [see e.g. 180, 243, 53, 166]. As noted by [249], 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).17 In addition to latitude, we also include regional dummies from [12] 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.

17One possible indirect connection is through an effect of latitudinal distance from the equator on consanguinity rates. As noted above, [145] 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).

70 Main findings Table 3.4 displays our first set of regressions, with ICRG institutional quality as the dependent variable. A full description of the variables is presented in Appendix B.3, Table B.3.1, and the omitted legal origin dummy in the regressions is the British one. Column (1) presents our basic regression model, inspired by [164] and [12], 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) (2) (3) (4) (5) (6) (7) Basic model Basic model and Income instead Latitude as both Income without Income VARIABLES restricted sample Consanguinity of Latitude instrument and Latitude and Latitude

Additional controls yes yes yes yes yes yes yes Observations 134 67 67 67 67 67 67 R-squared 0.527 0.441 0.668 0.695 0.654 0.704 0.632 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.4: 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 B.5.1.

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 [249]. 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 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 [12] 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 [12], restricting the sample to countries for which we also have consanguinity data (see Appendix B.4), and our results continue to hold. As the authors along with [249] note, ethnic fractionalization has a reduced form relationship with corruption and is not typically significant after controlling for per capita

71 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. Re- ligion 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 164, 12, 249]. 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).18 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). [164] and [249] offered reasons why Protestantism may be a cultural deterrent to corruption, beyond its relationship to consanguinity. They argue that Protestantism is associated with attitudes such as [256]’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 164, p. 229]. In column (2) of Table 3.5, 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 [164] 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. [173] argue that “Protestantism reduces corruption, in part, because of its association with individualistic, non-familistic relations" [249, p. 428]. Family ties is the next cultural trait that we discuss.

Family ties. As [57] 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

18Note, 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).

72 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" [265, p. 166-8]. Several studies confirm the negative correlation of strong family ties with general trust [266, 265, 94, 85, 14]. 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. 13, 15], 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.

(1) (2) (3) (4) (5) (6) VARIABLES Basic model and Religion and Family ties and Trust and Genetic diversity and Geography

Geographical variables yes Additional controls yes yes yes yes yes yes Observations 67 67 45 56 67 65 R-squared 0.668 0.762 0.783 0.725 0.689 0.770 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.5: Regression analysis of the relationship between consanguinity and corruption: potential confounds. N varies due to missing data for some countries. See Appendix B.5.1 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 B.5.2.

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 [265] 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 [253, 92, 240]. 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 gener-

73 alized trust and consanguinity (Spearman’s ρ = -0.19, p-value = 0.16, N=56).19 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) distance from Ethiopia [216, 208]. 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.20 [26] 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 3.5, we control for predicted genetic diversity and predicted genetic diversity squared, following [26] (see Appendix B.3 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. [57] 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 [204]. As explained in the data description in Appendix B.3, 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 [26]: 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 3.5 as we include additional variables for which we have limited data, in Table B.5.5 in Appendix B.5.1, 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 448 studies

19Note 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. 154, 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.

20Evidence of subsequent admixture complicates this view; see e.g. [168].

74 over many decades, we report additional analysis in Table B.5.6 in the appendix that addresses data collection dates. Our results remain robust in both tables.

Caveats. 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 consanguinity-driven sub-ethnic fractionalization (or vice versa) for various reasons. Moreover, corrupt, low-quality 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" [68, 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 or- ganize 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 68, 17, 153]. 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.

Summary. Overall, our cross-country results are consistent with the idea that marriage-practice driven sub-ethnic fractionalization, by increasing the relative returns to norms of favoritism, is an important cause of corruption. Nevertheless, our cross-country regressions cannot establish causality; in Appendix B.5.2 we report an instrumental variables analysis using a strategy developed by [230] that relies on linguistic differences in kin terminology. The evidence from the IV is consistent with that reported above, though since language and other cultural practices co-evolve, the assumption of exogeneity is a strong one. To further test our hypothesis, we next report analysis exploiting historical variation in consanguinity within a single developed country.

3.3.2 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 and consanguinity across Italian provinces. Our consanguinity data are from [57] 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

75 for all provinces and do not include sample variation due to either World War.21 The underlying data, however, date back to 1911, and [57] 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 B.5.4). 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 88], 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.22,23 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 B.3, Table B.3.2. 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 3.3 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). 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

21Due 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.

22A model by [162] connects corruption and associative crimes, based on criminal organizations’ attempts to avoid punishment by bribing law enforcement and otherwise engaging in local corruption.

23The 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 B.5.4, 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 [117] who analyzed links between organized crime and corruption for The European Commission. Their analysis includes case studies on several European countries including Italy, 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).

76 Figure 3.3: Corruption and consanguinity in Italy. for the level of development" [71, 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. 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 3.6 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 difference in consanguinity between the least and the most consanguineous province is associated with an increase of roughly 2.8 associative crimes per hundred thousand, about 45% of the difference between the most and least-corrupt region. 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.

77 (1) (2) (3) (4) (5) (6) Basic model and Consanguinity Income instead of Share of agriculture both Income and without Income and VARIABLES Share of agriculture as instrument Share of agriculture Share of agriculture

Consanguinity 5.144*** 5.371*** 4.984*** 5.155*** 5.427*** (1.099) (1.117) (1.167) (1.115) (1.081) Share of agriculture 12.707*** 3.472 3.705 (4.408) (3.243) (3.550) Log population 0.069 0.022 -0.011 -0.161 0.035 0.011 (0.124) (0.115) (0.132) (0.220) (0.133) (0.116) Log value added per capita -0.034 -0.264 0.018 (0.083) (0.273) (0.092) Constant 0.451 0.763 1.162 2.274 0.662 1.000 (1.654) (1.505) (1.607) (2.074) (1.624) (1.521)

Observations 101 101 101 101 101 101 R-squared 0.106 0.430 0.424 0.396 0.430 0.423 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.6: Regression analysis of the relationship between consanguinity and corruption in Italy.

Table 3.7 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) (2) (3) (4) (5) (6) Basic model and Consanguinity Income instead of Share of agriculture both Income and without Income and VARIABLES Share of agriculture as instrument Share of agriculture Share of agriculture

Additional controls yes yes yes yes yes yes Observations 101 101 101 101 101 101 R-squared 0.459 0.502 0.502 0.501 0.503 0.502 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.7: Replication of Table 3.6 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 B.5.11.

In column (1) of Table 3.7, which is analogous to column (1) of Table 3.6, 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 [29, 214]. 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 3.6. 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 3.3b.

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" [57, p. 3]. The possibility of differences in mating

78 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 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). We use this variable as an instrument for cousin marriage, but we do so with the caveat that the Church clearly had other influences on social, economic and political life, and so even if we cannot reject exogeneity, there are reasons to remain skeptical of the causal claim.

(1) (2) (3) (4) Basic model Income instead of both Income and without Income and VARIABLES Share of agriculture Share of agriculture Share of agriculture

No. of active years -0.002*** -0.002*** -0.002*** -0.002*** (0.000) (0.000) (0.000) (0.000) Share of agriculture 10.365** 10.778** (4.181) (4.559) Log population 0.128 0.014 0.152 0.111 (0.120) (0.150) (0.156) (0.120) Log value added per capita -0.136 0.032 (0.111) (0.113) Constant 2.363 3.915** 2.212 3.458* (1.766) (1.823) (1.915) (1.765)

Observations 101 101 101 101 R-squared 0.186 0.129 0.186 0.119 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.8: Reduced form regressions with active years of archdioceses.

Table 3.8 reports reduced form regression results showing that provinces with a higher number of active years of archdioceses exhibit lower rates of corruption. Table 3.9 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 B.5.12 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 B.5.4, we report

79 (1) (2) (3) (4) Basic model Income instead of both Income and without Income and VARIABLES Share of agriculture Share of agriculture Share of agriculture

Consanguinity 8.306*** 8.576*** 8.658*** 8.035*** (2.396) (2.329) (2.538) (1.891) Share of agriculture -2.205 -1.731 (5.420) (5.552) Log population -0.006 0.066 0.045 0.001 (0.146) (0.172) (0.156) (0.145) Log value added per capita 0.102 0.077 (0.116) (0.094) Constant 0.955 0.291 0.530 0.811 (1.899) (2.160) (1.973) (1.913)

Observations 101 101 101 101 R-squared 0.307 0.287 0.281 0.326 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

Table 3.9: Active years of archdioceses as an instrument for consanguinity. further analyses based on column (2) of Tables 3.6 and 3.7 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. 214, 248, 15, 73]. The relationship between consanguinity rates and corruption remains robust; see Tables B.5.13 and B.5.14. 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 B.5.15. 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 inefﬁ- ciency. 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 B.5.4, we show that our regression results are robust to using an alternative measure of corruption from [108] 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 1954-1997. “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" [108, 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 B.5.4 and Table B.5.16.

Summary. In our baseline specifications and controlling for a variety of possible climatic, geographical, and historical confounds, consanguinity is a significant predictor of corruption. More- over, our IV results may provide some evidence that the relationship is causal. As in the cross- country analysis, these data are highly consistent with a model in which corruption is in part driven by sub-ethnic fractionalization, as reflected in consanguinity.24

24One 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. [50] 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.

80 3.3.3 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, three-person 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: Van- couver, Canada and Urmia, Iran.25 Importantly for our motivation, both Iran (0.67) and Canada (0.71) are similarly ethnically fractionalized according to the measure developed in [12]. 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.26 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% [42]. 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% [42]. In other words, Urmia exhibits substantial sub-ethnic fractionalization; while Vancouver does not, possibly yielding differences in norms of favoritism and corruption. 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 3.4, which is a simpliﬁed version of the game described in Section 3.2. 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. 2]. 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, payoff- maximizing 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

25Our 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.

26http://www.iranicaonline.org/articles/kurdish-tribes

81 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".27 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.28 In Vancouver, we paid $7 for arriving on time and converted ECU at a rate of 10 ECU = $1. In Urmia, we paid 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 B.6.3).

Person A

Not transfer Transfer 40

Person B 100 100 Reject Accept 160 Person B  95  100 Left Right 160

 55  160 140 135 160 55

Figure 3.4: Bribery game in the experiment. Each terminal node shows the payoffs for Person A, Person B and Person C, from top to bottom.

27While [1] and [30] 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.

28Although the Rial is the ofﬁcial currency of Iran, Iranians employ the term ‘Toman’, meaning 10 rials

82 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 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 B.6.2). Pre-experiment ques- tionnaires 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.29 The instructions and more detailed procedures of the experiment are presented in Appendix B.6, 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 Canadian- born 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

29When 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.

83 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 Vancouver, 39 subjects (13 sibling pairs + 13 others) participated in the KKS treatment.30 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 3.2 is that the frequency of offer- ing 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.

30We 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.

84 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 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 3.10. 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 3.2 and Appendix B.1 it follows that kin are more related than co-ethnics and that co-ethnics are more related (in expectation) than strangers.

Iran Canada Kin Co-Ethnic Kin Co-Ethnic Bribery Corruption Bribery Corruption Bribery Corruption Bribery Corruption A & B related 18/20 19/20 17/20 16/20 13/13 12/13 7/20 9/20 A & C related 1/20 7/20 10/20 10/20 NA NA 6/20 7/20 B & C related 8/20 1/20 14/20 9/20 NA NA 9/20 6/20

Table 3.10: 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 3.10 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). 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

85 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 3.10, 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 somewhat more ethnic favoritism than Canadians, though both countries show similar levels of kin favoritism and similar behavior toward strangers. This suggests there are normative differences between the countries; we explore these differences further with subsequent treatments.

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 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.31,32 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).33 As seen in Table 3.11, we did not

31But there is evidence that friends are more closely related than random individuals; see Appendix B.1.

32This co-optation is reﬂected in the use of kinship words, e.g. ‘brother’ and ‘sister’, to refer to close friends.

33In 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

86 Kin (low cost) Friend (low cost) Stranger (low cost)a Bribery Corruption Bribery Corruption Bribery Corruption Canada 13/13 12/13 9/10 10/10 9/20 7/20 Iran 18/20 19/20 NA NA 30/52 22/52

Kin (high cost)b Friend (high cost) Stranger (high cost) Bribery Corruption Bribery Corruption Bribery Corruption Canada NA NA 13/20 9/20 NA NA Iran 24/24 22/24 11/12 10/12 1/12 1/12 aIncludes 40 pairs from Urmia in the SKK and SCC treatments. bIncludes 12 sibling and 12 cousin pairs.

Table 3.11: Summary of relative frequency of bribery and corruption when A & B are related. 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 B.6.4. 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). 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. 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 3.11, 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.

87 3.4 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 sub-ethnic 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, outmarriage 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.

88 Bibliography