Ethnic Diversity and Ethnic Discrimination: Explaining Local Public Goods Provision

Abstract

Numerous studies have found that ethnic diversity is negatively associated with the provision of local public goods. However, these accounts neglect both the strong role of central institutions in the provision of many “local” public goods and the frequently positive correlation between diversity and the presence of less politically powerful ethnic groups. These factors suggest that existing diversity findings may be explained in some cases by central governments discriminating against areas inhabited by less powerful groups. This hypothesis is tested using data in village-level public goods provision in Northern India, supplemented by data on service provision in Kenyan villages and American cities. While there is evidence that the presence of socially powerful groups is positively associated with service provision, evidence for the diversity hypothesis is weak. The results suggest that failures of public services in diverse areas may reflect larger inequalities within the political system rather than local problems in cooperation.

Keywords

race ethnicity and politics politics of growth/development

Introduction

In both wealthy and poor countries, many areas lack access to public services such as roads, schools, health clinics, and electricity. This lack is made particularly troubling by close association between these services, economic growth, and social outcomes. While there are many explanations for poor provision of public services, one in particular has received scholarly attention: the level of ethnic diversity, usually operationalized using a Herfindahl fractionalization index. The negative association between ethnic diversity and public services has been attested for a wide variety of nations, goods, and levels of aggregation (Alesina, Baqir, & Easterly, 1999; Algan, Hémet, & Laitin, 2011; Banerjee & Somanathan, 2007; Easterly & Levine, 1997; Miguel & Gugerty, 2005) though some recent studies have also questioned the relationship, especially at the national level (Gao, 2016; Gerring, Thacker, Lu, & Huang, 2015; Gisselquist, Leiderer, & Niño-Zarazúa, 2016; Singh & vom Hau, 2016; Soifer, 2016; Wimmer, 2016), and others have argued that the relationship is conditional on the salience of ethnic divisions, which may be a product of national policies, segregation, and between-group economic differences (Baldwin & Huber, 2010; Miguel, 2004; Trounstine, 2015). This relationship is usually traced to the inability of people from different ethnic backgrounds to cooperate with each other, a failure variously traced to differences in preferences, lack of a common language, and an inability to sustain cooperative equilibria due to social sanctioning (Habyarimana, Humphreys, Posner, & Weinstein, 2007).

Relative to many other hypotheses in political economy, the diversity–public goods relationship is both empirically well-supported and theoretically grounded. There are, however, two reasons to think that cooperation-based theories do not adequately capture the real life process by which public goods are allocated at many times and in many countries.

First, many local public services are provided through processes that involve national or regional governments much larger than the unit of observation, either through the direct provision, the distribution of money to local governments, or the staffing of facilities constructed by local governments. In the United States, for instance, local schools are constructed by local governments, but often a large portion of the funds used come from the state and federal governments. Indeed, in many nations there are considerable practical difficulties in identifying “pure” locally provided public goods outside of an experimental setting, and this has led many tests of the diversity hypothesis to focus on allocation processes in which nonlocal actors play an important role.

Second, most diversity theories implicitly assume that groups are politically equal conditional on local group size. However, in many countries, some ethnic groups are more politically and socially powerful than others. In many cases the powerful groups are also the national or regional majority group. In France, for instance, white French people are generally more influential than those of Middle Eastern descent. However, the powerful group may also be a small group that for historical reasons is overrepresented in higher education, land ownership, or the bureaucracy. Rwandan Tutsis, for instance, are both a minority group and are (today) the most politically powerful group in the country.

Taken together, these implied scope conditions of diversity models suggest an alternative, well-established, linkage between identity and public goods provision—the tendency of central policymakers to provide goods to areas inhabited by social powerful groups. If policymakers are disproportionately drawn from these groups, this preference may reflect the well-established tendency of politicians to favor coethnics in service provision (Burgess, Jedwab, Miguel, Morjaria, & i Miquel, 2015; Franck & Rainer, 2012; Hodler & Raschky, 2014). However, members of powerful social groups may have superior access to public goods even when their coethnics are not in elected office, due to more subtle processes of social discrimination.

Why, then, have so many studies found a negative diversity effect at the local level? This article will suggest that at least some of these findings are the result of collinearity. In countries or regions where the powerful group is a majority of the population, the local presence of these groups is highly collinear with ethnic fractionalization. In the United States, for instance, cities with high levels of diversity are overwhelmingly those with few Whites: fractionalization is negatively correlated with percent White at .86. It is thus often difficult, if not impossible, to separate any disadvantage stemming from diversity per se from the effects of discrimination against non-Whites. In many countries, this will mean that more diverse areas have worse public services even if this difference stems from discrimination by central institutions rather than a lack of cooperation among local people.

To properly test both the discrimination and diversity hypotheses, it is thus necessary to examine a case where (a) the literature has found a negative diversity effect for the types of goods in question, (b) we should expect cooperation problems between ethnic groups, and (c) levels of fractionalization are uncorrelated with the presence of the powerful group. This article implements this strategy using data on the provision of village-level public goods in the Northern Indian states of Bihar and Uttar Pradesh. Not only are these goods very similar to those examined in other parts of the world, but existing studies strongly support the idea that caste diversity is associated with low public goods provision in India (Banerjee & Somanathan, 2007) and that caste groups have difficulty cooperating with each other (Fehr, Hoff, & Kshetramade, 2008). But the most important advantage of Northern India as a case is that social diversity is uncorrelated with the presence of the most politically powerful groups. While members of four high-status caste groups have traditionally been disproportionately powerful in the politics and society of the study region, the small size of these groups mean that their presence is not associated with village-level caste fractionalization (ρ = −.01).

The Indian data provides strong support for the ethnic discrimination hypothesis. While the presence of upper castes has a strong positive association with the most important types of public services provision, there is no evidence of a negative association between diversity and public goods, even for goods with little nonlocal involvement. These results hold even when accounting for variation in the salience of caste identities. Like nearly all studies of the identity–public goods relationship, this study is potentially confounded by geographical and economic differences between areas inhabited by different types and numbers of ethnic groups. To attempt to deal with this issue, a variety of alternative models include controls for spatial differences between villages, village-level wealth, general and intergroup economic inequality, and patronage networks.

These results might be dismissed as a product of some unique aspect of the Indian case. Perhaps India’s public goods provision process is too centralized for the diversity hypothesis to hold, but the hypothesis still applies everywhere else? While weighing the relative importance of these two mechanisms on a global scale is impossible, this study provides additional evidence that even in cases cited as evidence for the diversity approach, the evidence for a diversity effect is limited once we account for discrimination. To do this, the study reanalyzes data on public goods provision taken from two influential studies: Alesina et al. (1999) (on American cities) and Miguel (2004) (on villages in western Kenya). In both studies, ethnic diversity is highly correlated with the presence of the regional or national majority ethnic group. As a consequence, while both ethnic diversity and the presence of the majority ethnic group are associated with public goods provision, the variables are so collinear that both relationships disappear when the two are included in the same model. These results demonstrate the difficulty of separating the power and diversity hypotheses in many cases, and the fact that the discrimination hypothesis is fully consistent with the data even in cases cited as evidence for a diversity effect.

Ethnic Politics and Public Goods: Existing Accounts

Governments have the potential to provide a wide variety of goods and services to their citizens that may both have direct welfare benefits and enhance economic growth. Classic examples include education, health care, transportation infrastructure, electricity, and water. In political economy discussions, such broadly beneficial goods are often referred to as public goods, even though they do not fit the classic definitions of the concept in economic theory, as many of these goods are in practice both rival and excludable. The underprovision of these goods represents a puzzle for scholars because it is, at least in a societal sense, irrational, as providing them could enhance overall welfare. A variety of theories of this failure have been developed, focusing on factors such as low state capacity, social capital, and self-interested behavior by politically powerful elites. One factor, however, has received a particularly large share of scholarly attention: The role of ethnic diversity.

The Diversity Hypothesis

Theoretical accounts of the diversity mechanism differ in details, but share certain broad features. Most importantly, they begin by assuming the public goods provision requires cooperation between a large number of individuals within a locality. This cooperation can take the form either of assembling a minimal winning coalition to impose taxes to pay for the services (Alesina et al., 1999), or encouraging a large number of people to voluntarily contribute labor and money in defiance of incentives to free ride (Miguel & Gugerty, 2005). If this cooperation succeeds, the good is provided, while if it fails, it is not.

The central contention of the diversity hypothesis is that cooperation is more likely to occur when the individuals involved are from the same ethnic group. A variety of mechanisms have been proposed to explain this pattern. Coethnics may prefer similar types of goods, may find cooperation easier due to shared languages and social networks, or may be more easily able to sanction free-riding. Since Habyarimana et al.’s (2007) influential contribution, this last mechanism has been the most popular, though all proposed mechanisms give a broadly similar prediction about the relationship between diversity and public goods, and link it to failures in local-level cooperation.

This relationship is supported by a wide variety of empirical analyses, showing diversity to be negatively associated with public goods across a wide variety of local units, including countries (Easterly & Levine, 1997), American cities (Alesina et al., 1999), African villages (Miguel, 2004), French housing blocks (Algan et al., 2011), and Indian districts (Banerjee & Somanathan, 2007). Supplementing the observational work, another literature has shown that non-coethnics are less likely to cooperate with each other or give to each other in laboratory settings (Habyarimana et al., 2007; Whitt & Wilson, 2007).

A well-known critique of these findings is that the ethnic divisions on which they focus are both the product of endogenous social processes and vary in their importance across social contexts. In response, scholars have proposed alternative formulations of the diversity argument, holding that ethnic divisions are only significant when they are made salient by factors such as national policy and history (Miguel, 2004; Singh & vom Hau, 2016), interethnic inequality (Baldwin & Huber, 2010), and residential segregation (Trounstine, 2015).

A small but growing literature has begun to question the validity of the diversity–public goods relationship at the subnational level, where it appears that the association is zero or positive in many contexts (Gao, 2016; Gerring et al., 2015; Gisselquist et al., 2016). However, these accounts generally do not question the theoretical mechanism that public goods are cooperatively provided and ethnic differences impede cooperation. Instead, they argue that problems of cooperation at the local level may be ameliorated by the availability of individuals with different sets of skills (Gerring et al., 2015), because monitoring and social sanctioning are easier in a smaller communities than in larger ones, or because diversity may enhance political competition (Gao, 2016).

Another, even more forceful, critique of the diversity hypothesis comes from scholars who argue that ethnic diversity is the result of historical processes of state building and national building (Singh & vom Hau, 2016; Soifer, 2016; Wimmer, 2016). State strength and regional and national identity are thus omitted variables in any analysis of the relationship between ethnic diversity and public goods provision: The states with the ability, time, and inclination to form homogenous societies are precisely those with the ability, time, and inclination to provide public goods. However, this literature does not directly address why provision might vary within states or regions with similar levels of capacity and similar nation-building strategies. These authors also do not explain why within-region variation in public goods might be correlated with ethnic diversity, though Singh and vom Hau (2016, p. 1322) do suggest some considerations that might influence politicians in these situations.¹ While this critique may apply to the cross-national variation or interregional variation in a large country, it thus leaves a large body of diversity-level diversity results unexplained.

Criticisms of the Diversity Hypothesis

Despite the status of the vadiversity hypothesis as “one of the most powerful hypotheses in political economy” (Banerjee, Iyer, & Somanathan, 2005, p. 639), there are several reasons to believe that in many circumstances the ethnic cooperation mechanism does not adequately capture the process by which public goods are produced and consumed in many countries. Two such critiques are detailed below. In common with the literature, these critiques assume that public goods are nonrival and nonexcludable; however, a third potential critique, the de facto excludability of many local “public” goods, is discussed in an online appendix.

Centralized provision

The diversity hypothesis assumes that public goods are local not just in their benefits but in their origin, and are the product of taxation and spending decisions by the communities that enjoy them. While this pattern may describe some public goods in some countries (and, by construction, laboratory public goods games), it does not describe public goods provision in many real-world situations, as some recent work has pointed out (Ejdemyr, Kramon, & Robinson, 2015; Singh & vom Hau, 2016). Many public goods are constructed by central or regional governments with little reference to the local community. Highways in the United States, for instance, are usually constructed and maintained by the state governments rather than the municipalities through which they run.

In many other cases, while the goods are provided by a local entity, a large portion of that entity’s funding is provided by higher levels of government. In the U.S. highway example, for instance, much of the money spent by the state government is in fact provided by the federal government. Finally, in some other cases, provision may require independent efforts by both local and supralocal governments. Local governments may build schools but require the central government to staff them with teachers, or regional governments may build power lines but require the central government to build the power plants to make them effective. These types of provision processes complicate an attempt to bifurcate “state” and “local” provision in a binary way.

While the role of nonlocal actors is widely acknowledged, with some exceptions authors have tended to treat it as an empirical obstacle to identifying local diversity effects rather than an object of theorizing in itself. To minimize its importance, studies often focus on services in which local initiative is thought to be especially important, or decisions with a specifically local component, such a budget distribution. Some have even argued that the diversity effect holds even for centrally provided goods, as lobbying the nonlocal government has similar dynamics to cooperative provision (Banerjee & Somanathan, 2007).

The existence of centralized provision, of course, does not mean that local provision does not exist. What it does mean is that finding a situation with “pure” local provision is surprisingly challenging in practice, and that many (if not most) existing observational studies of the diversity effect study processes in which supralocal actors play a role. Alesina et al. (1999), for instance, studies public goods provision in American cities while acknowledging that the state and federal actors provide considerable funding to these governments, while Banerjee and Somanathan (2007) studies provision in Indian districts while acknowledging that many important distributional decisions are made by state governments.

Powerful groups

In the classic diversity hypothesis, all groups are modeled as being similar, with only the local sizes of the groups being important. However, different groups may have vastly different political positions within the country as a whole—what Tilly (1998) calls “durable inequalities.” Such differences in ethnic power within countries are widely acknowledged, and now can even be measured cross-nationally, through the Ethnic Power-Relations data set (Cederman, Wimmer, & Min, 2010).

Many groups gain disproportionate power through their national or regional numerical size (which may give advantages in winning elections in a democratic or quasi-democratic context). In other contexts, the position of the powerful groups may be a result of historical social differences (such as colonial favoritism, Lee, 2017) or early immigration (Portes & Rumbaut, 2006) that left certain groups with legacy of access to land, military service, education, and government employment that they have used to obtain political power, and have since retained by “opportunity hoarding.”

In the United States, both factors are in play: Whites are not only the majority group, but have historically been overrepresented among nonelected officials and campaign donors and have generally had higher levels of political involvement and awareness than members of other groups. In India, by contrast, upper caste members have historically been heavily overrepresented among bureaucrats and the politically aware, but in most states are not the plurality group or social category. Similarly, while the most politically powerful group is often the wealthiest group, there are numerous examples of wealthy but powerless ethnic groups, such as the Chinese in Malaysia, or large and influential poor groups.

The extreme form of such differences in group political power would be a state run on completely ethnic lines, with all power monopolized by members of a single group and members of the less powerful groups having limited civil and political rights. Such ethnic exclusion has been shown to be both empirically common and associated with a variety of unpleasant social and political outcomes other than public goods, including armed revolt (Cederman et al., 2010; Wimmer, 2002, 2017). However, inequality in political power across ethnic groups is empirically common even in democracies and societies with accommodationist approaches to ethnic identity. In particular, much of the literature on ethnic diversity and ethnic favoritism discussed above come from polities where ethnic discrimination is subtle and noninstitutionalized.

The diversity thesis implies differences in group power, however strong, should not matter. An area with a minority from the powerful group and a majority from another group (for instance, a city that is 90% Black and 10% White) will be expected to have an identical probability of goods provision from an area with a majority from the powerful group and a minority from another group (for instance, a city that is 90% White and 10% Black), due to their identical probability of interethnic cooperation. In many countries, this seems a challengeable contention.

Empirically, the problem of differences in group power is compounded by the fact that in many (though certainly not all) cases, the presence of the largest group is associated with fractionalization at the local level. When the powerful group has the largest share of the population and the number of other groups is small, the correlation between the two measures will be very high by construction, as the largest component of Herfindahl is the square of the largest group’s population share. In practice, it might therefore be impossible to distinguish between the negative effects of diversity and the positive effects of the presence of a majority group, due to collinearity.²

Ethnic Politics and Public Goods: An Alternative Hypothesis

Bias in Central Government Provision

How might unequally powerful ethnic groups and provision by central institutions be incorporated into a theory of public goods provision? Consider the following simple scenario. Services, or funds that can be used to purchase services, are provided to local communities by a central government. While the decision-maker is disproportionately likely to be from the powerful group (or is easily influenced by that group), members of that group make up a varying proportion of local communities. The decision-maker, in the presence of a budget constraint, must decide which communities receive goods and funds. There are several reasons to expect that the central government will tend to allocate resources disproportionately to areas inhabited by the powerful group. Two of these are mentioned below:

Intraethnic favoritism

Decision-makers (who are disproportionately drawn from more powerful groups) may wish to provide resources to members of their own group. This may be because they have some preference for coethnic welfare, face social sanctions when they exclude their own coethnics, or tend to attribute positive qualities to members of their own group. All of these have some support from the experimental literature on discrimination and interethnic interaction, which has shown that subjects tend to prefer coethinics in contexts such as voting, even when they are not familiar with the candidate Adida (2015). However, one can obtain a similar prediction from a purely rationalistic theory. In many countries, ethnicity is closely associated with parties and the political system, and politicians use their ethnic identity as a way of getting votes. In Chandra’s (2007) influential formulation, in such a system ethnicity serves as a tool for distributing goods to supporters, by serving as a visible signal of political affiliation. This echoes other findings on the association between ethnic politics and clientelism. Kitschelt and Wilkinson (2007), broader theories of the relationship between voter choice and distribution (Dixit & Londregan, 1996), and Singh and vom Hau (2016) emphasize on the importance of “political survival considerations” in structuring responses by state officials to ethnic mobilization.

The empirical literature on coethnic favoritism in the provision of public services is very extensive. Much of it has focused on the ethnicity of Africa, generally using overtime changes in leader ethnicity as an identification strategy (Bates, 1983; Burgess et al., 2015; Franck & Rainer, 2012; Kramon, 2013; Lindberg, 2010), though similar studies have focused on India (Besley, Pande, Rahman, & Rao, 2004) and the world as a whole (Hodler & Raschky, 2014). This work has shown with some consistency that politicians target resources to members of their own ethnic group, and that voters in turn associate coethnic politicians with superior provision (Carlson, 2015)

Note that ethnic favoritism arguments that ignore intergroup inequality (like most of those above) do not contradict the diversity hypothesis—in fact, being concerned with distribution, they make no predictions about aggregate provision. If the political class is fairly representative of the population, ethnic favoritism will lead to temporal or spatial differences in provision, but in the aggregate each group will get their share, either in districts where their coethnic is elected or in years when their coethnic is elected. In fact, favoritism could provide an alternative causal pathway for the negative effect of diversity, by making homogenous communities more attractive targets (Ejdemyr et al., 2015). However, if (as section “Ethnic Politics and Public Goods” proposed) decision-makers are always more likely to come from powerful groups, we should expect areas inhabited by these groups to always have better access to public goods.

Other forms of ethnic advantage

However, the advantage enjoyed by politically powerful groups may extend beyond ethnic favoritism. Members of the most powerful ethnic group may (a) have access to a larger and more powerful social network (as social networks often follow ethnic lines; Anderson, Francois, & Kotwal, 2015). They may also (b) benefit from positive stereotypes about their behavior that were created to justify the status quo power distribution (Jost, Kivetz, Rubini, Guermandi, & Mosso, 2005). Finally, (c) members of particular ethnic groups may have higher levels of political participation and political awareness even after accounting for their higher socioeconomic status (Leighley & Vedlitz, 1999). If any of these factors hold, members of powerful groups may be more likely to gain resources even when a coethnic is not the decision-maker, either because they are more likely than members of other groups to be able to influence an influencer of the decision-maker, be perceived positively by the decision maker, or simply more likely to know how to lobby effectively.

These mechanisms are not mutually exclusive, and they may well act in concert with each other. All of them, however, produce the same basic empirical prediction: When the central government plays a large role in public goods provision, areas with a high proportion of the population from politically powerful groups will have higher levels of public goods provision than other areas. Diversity, on the other hand, should have little relationship to goods provision in this situation, as a decision-maker bent on benefiting coethnics will not be sensitive to the internal composition of the non-coethnic groups.³

Discussion

In cases where the powerful group is also the largest group, the ethnic discrimination hypothesis described in section “Bias in Central Government Provision” produces predictions very similar to the diversity model. The areas populated by the most powerful group, which will also tend to be the less diverse areas, will have higher levels of public goods provision. The theoretical implications, however, are quite different, as one emphasizes central discrimination and the other local cooperation. To understand the differences, consider the case of the city of Flint, Michigan, which combines a non-White population and fractionalization indices that are high relative to other cities in Michigan with a set of public services (most notoriously, its water system) that are of much lower quality. From the diversity perspective, this pattern would be attributed to the inability of actors within Flint to cooperate to provide high-quality services. From the discrimination perspective, this pattern would be attributed to a pervasive set of biases within the state and federal government that lead to lower levels of resources being provided to poor cities in general, and poor non-White cities in particular.⁴

This does not necessarily mean that all goods will be provided in a discriminatory fashion. For many goods in many countries, there are strong technocratic criteria that guide distribution. If these programs operate as designed, they may even lead central transfers to flow disproportionately to disadvantaged communities. For other goods, however, particularly those provided through less transparent or more politicized procedures, the rules might give considerable discretion for central decision-makers to favor the powerful.

These explanations are also not mutually exclusive: The discrimination logic might apply to cases where goods are provided by the central government and there are strong power asymmetries within groups, while the diversity logic might apply in cases where groups are relatively equal and local governments relatively autonomous. However, given the prevalence of ethnic inequalities and overcentralized systems of local government in both the world in general and developing countries in particular, the discrimination perspective is worthy of sustained empirical investigation.

The Indian Case: Data

Caste in India

To separate these two mechanisms, it is necessary to examine a case where the proportion of individuals from the most powerful groups is uncorrelated with the fractionalization. For this reason, the main analysis will focus on public goods provision in rural Northern India. All regions of peninsular India have some degree of exposure to the caste system, a ranked system of identities which, at least in theory, assigns relative status to India’s hundreds of endogamous caste groups, or jatis. In popular discussion, jatis are often discussed in terms of the categories to which they are assigned for the purpose of affirmative action: “upper”’ or “twice born” castes; “intermediate” or “peasant” castes; “other backward classes”; and “untouchables” or “scheduled castes.” The “scheduled tribes” of the hills and the Muslim minority are somewhat outside of this system, but are usually considered to be poor and politically marginalized. These levels of social status are associated with, though not identical to, levels of wealth, education, and traditional political power. Certain groups that combine relatively high status, large size, and control over land are referred to as “dominant castes,” and exercise disproportionate political power within specific regions (Lee, 2016; Srinivas, 1987).

Caste differs in several important respects from ethnic identities in other parts of the world, especially in its embrace to ranking. However, none of these differences lead to violations of the basic assumptions about ethnic divisions that underlie the diversity hypothesis: that goods preferences be heterogeneous across groups, that social sanctioning be more effective within groups, and that identities be commonly known. In fact, experimental evidence in India shows that there are great difficulties in cooperation between castes and low levels of other-regarding preferences across castes (Fehr et al., 2008). Similarly, Banerjee and Somanathan (2007) replicate the diversity finding across Indian districts, leading many to cite India as support for the diversity hypothesis.⁵

There is considerable evidence that caste plays a role in public service distribution decisions in India. Not only does the caste identity of officials appear to effect who received services (Besley et al., 2004), but upper caste individuals in particular appear to have strong preferences against any type of redistribution to the lower castes (Fehr et al., 2008). A perception of discrimination by the upper castes has also been at the center of the movements for lower caste empowerment that have flourished since the early 20th century (Jaffrelot, 2003).

Importantly, the presence of upper and intermediate castes is uncorrelated with the overall fragmentation of the caste system. Not only are Hindu upper and intermediate castes a relatively small proportion of the population (27%, according to the 2005 National Family Health Survey), but the category is internally divided into a variety of jatis, such as Brahmins, Rajputs, Jats, and Marathas. Overall, no Indian jati comprises more than 38% of the population of a major state (Lee, 2016). It is thus very possible for an area to be either very diverse with a high upper caste percentage (if several high status jatis are present) or homogenous with a low upper caste proportion (if one of the several relatively large low-status jatis is present).

The Data

The data used in this article are drawn from the UP-Bihar Survey of Living Conditions (UPBSLS), conducted by the World Bank in northern and central Bihar and eastern Uttar Pradesh in 1997 and 1998. These two states are among the poorest areas of India, with reputations for low public service quality. The data thus show considerable variation in provision of even very basic public goods. These areas are also well known for the high salience of caste in politics, making both the diversity and discrimination mechanisms potentially relevant. There is also much less variation in historical experience within this region than within India as a whole, particularly for such well-known factors as colonial land tenure system.

The UPBSLS covered 120 villages, evenly divided across the two states, selected using a stratified random design built upon an earlier qualitative survey. For each village, the enumerators collected a large amount of information on overall access to public services and social conditions, while also conducting a detailed individual survey of 30 (in some villages, 15) randomly chosen households. The UPBSLS was used here despite its age because the village-level data includes the population breakdown of castes in the village as a whole. This makes the survey a valuable rarity among village-level studies in India, given the government’s reluctance to collect data on caste.

The village-level caste enumeration departed in two key ways from full enumeration, as it only included the seven most populous castes in the village (for reasons of space on the form) and enumerated households rather than individuals. To deal with this problem, the main models use as a population measure the proportion of households of each caste within the total number of households from the seven largest castes. A variety of alternative approaches to this issue are discussed in an online appendix, and shown to have no impact on the results.

Unequal Caste Power

In the study area, both the anthropological literature and the available data show that four castes have traditionally had disproportionate levels of social and political influence: The Brahmins, Bhumihars, Rajputs, and Kayasths, derisively referred to by their opponents by the partial acronym bhurabal. All four groups are of high ritual status (“twice born”), and are of much higher ascribed status than any other local group. These four groups are also notable for their economic and political influence. Collectively, they comprise 15% of the population of Bihar, but own 48% of the land and provide 44% of the state’s college graduates (Lee, 2016). They have also traditionally been powerful within the political system and the bureaucracy. In 2005, they provided 26% of Bihar’s state legislators (Jaffrelot & Kumar, 2012). In 2012, these four castes provided 77% of Bihar senior civil servants, despite strict quotas (Witsoe, 2013, p. 85). Qualitative accounts of the politics and society of Bihar and UP also emphasize the enduring political and social importance of these groups, founded on their high levels of wealth and education (Chakravarti et al., 2001; Jha, 1972; Witsoe, 2013).

Is this regional dominance likely to be helpful to the relatively poor and rural members of the upper castes in our sample? Evidence from the Indian Human Development Survey (2005) shows that in these regions even ordinary upper caste villagers are much more likely to have useful contacts in the state apparatus than their lower caste neighbors. Appendix Table A.9 shows that relative to other caste members, upper caste respondents are more than twice as likely to know a government employee and three times as likely to know a government officer. Some, though not all, of this difference seems to be driven by caste-specific social networks, as the contacts of upper caste members are 25 percentage points more likely to come from their own caste as the contacts of others.

Upper caste power in these regions has been so marked that it has led to a compensating movement for lower caste self-assertion. Riding this wave, lower caste politicians such as Mayawati, Laloo Prasad Yadav, and Mulayam Singh held the highest political offices in both states during periods of the 1990s. However, there are a variety of reasons to doubt that this change, still somewhat novel and insecure in 1997, would have eliminated the relationship between upper caste interests and distributional patterns. First, the majority of public goods at the time of the survey would have been constructed before 1990, under the influence of upper caste politicians. Second, upper caste political involvement remains very high, both within the civil service and their role in the newly important lower-caste-led political parties (Jaffrelot & Kumar, 2012). Finally, most accounts of the these lower caste politicians have emphasized the degree to which they neglected public goods provision in favor of rent-seeking and patronage distribution, to a degree even more notable than their upper caste predecessors (Thakur, 2000; Witsoe, 2013).

However, despite their political power, the small size of the twice born groups means that their presence in villages is unassociated with diversity: Overall, the Herfindahl index of caste diversity of villages in the sample has a correlation with percent upper caste of ρ = −.01.⁶

The Dependent Variable

The UPBSLS has information on the actual incidence of a wide variety of public amenities available in each village. Four are the focus of the empirical analysis here: one continuous measure (the number of hours of electricity per day the village receives in a “good month”)⁷ and three binary measures (the presence of a public primary school, public primary health clinic, and landline telephone access. All are, at least potentially, broadly beneficial to residents of a village that has them and similar (if not identical) to the types of local public goods discussed in the broader literature. All are also provided by the state in India, with no possibility (by definition, in most cases) of private substitution. All are present in more than 10% and less than 90% of villages, and in no case is there provision contingent on the presence of another public good.

In addition to possessing these properties, these particular goods were selected because of their prominence in existing accounts of distributive politics in India. Schools, clinics, and telephones were all used as outcomes in Banerjee and Somanathan (2007), the most influential Indian formulation of the diversity hypothesis. Hours of electricity was chosen because Min (2015) provides one of the best existing accounts of the microprocesses by which state officials in the study area (in this case power engineers) distort public goods provision to favor areas with more political and economic clout. Min notes that

politicians routinely interfere in the operation of the state electricity board, from patronage transfers of employees, interventions in the selection of villages for electrification projects, and the assertion of influence on when, where, and how, power cuts are timed and distributed. (Min, 2015, p. 130)

In addition to these four measures, seven additional variables in the UPBSLS might be plausible measures of local public goods provision. In the appendix, Section A.8 and Table A.12 discuss the deficiencies of these alternate measures and report the results of models that use them, and the results of a multivariate regression model using all 11 outcomes.

The Distributional Process

The political and economic order that supplies these goods represents a mix of state and local institutions and pressures. Power lines, schools, and health clinics may be constructed by either the state government or the village (panchayat) government. The state government provides staff and funding for clinics and schools, and service, maintenance, current for phones and electricity, and a large share of the funding of the local panchayat. The panchayat is responsible for the maintenance of many of these same facilities. The local community, in addition, must lobby hard to receive state resources, and much of the activity of the panchayat government is taken up in this type of maneuvering. Banerjee and Somanathan (2007) suggest that cooperative lobbying is an important conduit for the diversity hypothesis.

This complex distributional arena might encourage proponents of both hypotheses discussed above. While Banerjee and Somanathan (2007) treated the provision process as being fundamentally local (and thus a good place to test the diversity hypothesis), others might argue that the power of the state government (and the power of the upper castes in that government) would make discrimination the obvious mechanism. While the Indian pattern differs considerably from the ideal types of either central or local provision, it is typical of the type of messy, multilevel distributional process common in many parts of the world, and of the types of goods discussed in the existing literature. In the section “A “More Local” Public Good,” we will examine a good where the local role is more pronounced than for these four goods.

The Indian Case: Results

The Diversity Hypothesis

The determinates of public goods provision in UP and Bihar are examined in Table 1, which shows the results of a series of regression models, using linear regressions for the number of hours of electricity and logistic regressions for the other three goods. All models control for the population of the village, as population is an obvious “technocratic” predictor of whether a village should be targeted for service delivery by the state. Population is also mentioned by Gerring et al. (2015) as a key factor in explaining variation in the effect of ethnicity on goods provision. Panel B of Table A.7 shows that the results are robust to not including this variable. In the appendix, a variety of alternative models are reported, including models using district fixed effects and bootstrapped standard errors.

Table 1.

The Diversity Hypothesis: Public Goods in India.

Variables	(1)	(2)	(3)	(4)
Variables	Elect. hours	Telephone	Public primary school	Health center
Panel A: Fractionalization
Village caste fract.	1.451	3.578*	−0.841	11.97**
Village caste fract.	(2.397)	(1.903)	(1.258)	(5.334)
Village pop.	0.00316	0.00157	0.0108***	0.00309*
Village pop.	(0.00289)	(0.00131)	(0.00311)	(0.00159)
Constant	2.417	−4.038***	−0.300	−11.66***
Constant	(1.628)	(1.345)	(0.871)	(4.081)
Observations	120	119	119	119
Panel B: Controls
Village caste fract.	0.594	4.542*	−0.555	14.24**
Village caste fract.	(2.411)	(2.330)	(1.418)	(6.265)
Village pop.	0.00507*	0.000576	0.0116***	0.00320
Village pop.	(0.00296)	(0.00148)	(0.00335)	(0.00199)
Bihar	−1.476	0.758	2.024**	1.217
Bihar	(1.138)	(0.603)	(0.786)	(0.949)
% House. with off-farm jobs	0.0194	0.000621	0.00704	0.0191
% House. with off-farm jobs	(0.0180)	(0.0106)	(0.00967)	(0.0170)
Village Road Access	2.692*	1.350	−0.676	−1.034
Village Road Access	(1.440)	(1.114)	(0.933)	(1.038)
Village prop. landless	−0.00690	0.00794	−0.0218	−0.00333
Village prop. landless	(0.0270)	(0.0141)	(0.0179)	(0.0231)
Irrigation % scoring	0.614	0.126	0.388*	0.692*
Irrigation % scoring	(0.392)	(0.229)	(0.225)	(0.409)
Avg. price no irrigated land	9.82e-06	2.39e-06	−2.49e-06	2.83e-06
Avg. price no irrigated land	(1.26e-05)	(6.49e-06)	(7.28e-06)	(9.02e-06)
Constant	−2.635	−7.037***	−1.929	−16.84***
Constant	(2.702)	(2.309)	(1.709)	(5.425)
Observations	116	115	115	115

Standard errors in parentheses. Column 1 shows the results of linear regression models with the number of hours of electricity a village gets in a “good” month as the dependent variable. Columns 2-4 are logistic regressions with a binary measure of whether a village has a particular facility as the dependent variable.

p < .1. **p < .05. ***p < .01.

Panel A of Table 1 examines the association between caste fractionalization in a village and a village’s level of public services. There is no evidence of a strong negative relationship between fractionalization and public services. In fact, if anything, the results go the other way, with the effect of fractionalization being statistically significant and positive for both telephones and health centers. This provides some limited evidence that, as the discrimination hypothesis would predict, local diversity is not a negative factor for obtaining services in areas where diversity and powerful groups’ presence are not correlated. In fact, there is some limited evidence for a positive (though not necessarily causal) association between diversity and local public goods.

One longstanding critique of the diversity hypothesis is that is does not account for unobserved differences in geography and economics between diverse and homogenous places. Panel B of Table 1 includes a vector of controls designed to capture such differences. These include dummy variables for whether the village was in Bihar or had road access (paved or unpaved), the proportion of the land that was irrigated, the average price of unirrigated land, and the proportion of households that were landless and had off the farm jobs. The inclusion of these controls does not alter the basic results: Caste fractionalization has no negative association with public goods provision.

The Discrimination Hypothesis

Table 2 examines the effect of the proportion of upper caste households on service provision. Panel A tests the simple association, conditional on village population. The proportion of upper castes has a strong, statistically significant positive association with public goods provision. For a village with the median population, the increase from an entirely lower caste village to an entirely upper caste one would increase the predicted number of hours of electricity a village receives in a day by 4 hr (a whole standard deviation) while the same change would increase its predicted probability of having a public primary school from 77% to 97%.

Table 2.

The Discrimination Hypothesis: Public Goods in India.

Variables	(1)	(2)	(3)	(4)
Variables	Elect. hours	Telephone	Public primary school	Health center
Panel A: Upper caste population
Village pop.	0.00387	0.00243*	0.0120***	0.00419***
Village pop.	(0.00278)	(0.00127)	(0.00328)	(0.00152)
Village prop. upper castes	4.345*	2.307**	3.447**	2.146*
Village prop. upper castes	(2.274)	(1.014)	(1.670)	(1.290)
Constant	2.577***	−2.200***	−1.438**	−3.627***
Constant	(0.917)	(0.476)	(0.659)	(0.675)
Observations	120	119	119	119
Panel B: Upper caste population and fractionalization
Village pop.	0.00348	0.00164	0.0125***	0.00300*
Village pop.	(0.00286)	(0.00135)	(0.00338)	(0.00161)
Village prop. upper castes	4.346*	2.916**	3.685**	3.580**
Village prop. upper castes	(2.281)	(1.154)	(1.801)	(1.799)
Village caste fract.	1.456	4.393**	−1.025	13.09**
Village caste fract.	(2.371)	(2.113)	(1.316)	(5.426)
Constant	1.736	−5.101***	−0.908	−13.07***
Constant	(1.649)	(1.565)	(0.944)	(4.239)
Observations	120	119	119	119
Panel C: Controls
Village pop.	0.00586**	0.00174	0.0130***	0.00505**
Village pop.	(0.00284)	(0.00143)	(0.00363)	(0.00197)
Village prop. upper castes	5.033**	2.061*	3.265*	2.592*
Village prop. upper castes	(2.263)	(1.110)	(1.739)	(1.531)
Controls	Yes	Yes	Yes	Yes
Constant	−3.313	−4.511***	−2.749*	−7.162***
Constant	(2.308)	(1.558)	(1.617)	(2.083)
Observations	116	115	115	115
Panel D: Other caste categories
Village pop.	0.00397	0.00247*	0.0126***	0.00466***
Village pop.	(0.00284)	(0.00129)	(0.00345)	(0.00159)
Village prop. upper castes	4.230*	2.230**	2.996*	2.957*
Village prop. upper castes	(2.467)	(1.117)	(1.702)	(1.587)
Village prop. scheduled castes/tribes	0.189	0.0119	0.00353	2.564
Village prop. scheduled castes/tribes	(2.454)	(1.269)	(1.219)	(1.746)
Village prop. Muslim	−0.465	−0.286	−1.499	0.776
Village prop. Muslim	(2.299)	(1.239)	(1.210)	(1.742)
Constant	2.571*	−2.170***	−1.310	−4.741***
Constant	(1.358)	(0.697)	(0.861)	(1.137)
Observations	120	119	119	119

p < .1. **p < .05. ***p < .01.

Panel B adds measures of caste fractionalization to the model. While caste fractionalization still does not have a negative effect on goods provision, upper caste presence retains a statistically significant and positive effect. At least within India, the presence of specific groups appears more closely associated with public services than overall diversity.

Panels C and D test the robustness of the association between upper caste presence and public goods. Panel C includes the standard vector of economic controls, which has little effect on the upper caste coefficient. Panel D controls for the presence of two especially disadvantaged types of social groups, Muslims and scheduled castes and tribes, with castes classified as other backward classes (OBCs) as the excluded category. The presence of neither of these groups has an effect on public goods provision statistically different than the OBCs, confirming that the results are capturing an upper caste advantage rather than the disadvantages of some other particular category.

A “More Local” Public Good

One implication of the discrimination hypothesis is that in a democratic context, the proportion of people from the advantaged group should only matter for public goods where nonlocal politicians or administrators play a large role. Within the location of a nonexcludable good, members of all groups will have an incentive to see the good provided (since they will benefit from them personally). In fact, this combination of purely local provision and democratic or cooperative local decision-making is exactly what the diversity hypothesis assumes.

Testing this implication is complicated by the difficultly identified in the section “Ethnic Politics and Public Goods”: While the distinction between locally and centrally provided funds may be clear in theory, in practice fund transfers to local governments and divided administrative responsibilities make “pure” local public goods difficult to identify in practice, tempting scholars to examine outcomes where a nonlocal actor plays a role. However, in the Indian data one can identify one service that, if not completely local, at least involves outsiders less that schools or electricity: sewage. Waste disposal is a core function of Indian village government, and one of the few functions guaranteed to panchayats rather than delegated to them by the state government. Moreover, the construction of open earth drains is simple and requires little capital investment or technical expertise, though it would certainly be helped by money or advise from state agencies.⁸ Drains thus should represent an easy case for the diversity hypothesis, since they are the good for which the distribution process most closely approximates the theoretical conditions to which the hypothesis is thought to apply

Table 3 shows that, as expected, upper caste presence is not robustly correlated with the provision of drains, though the coefficient is positive. Even more interestingly, the association between caste diversity and drains is positive and statistically significant at the 10% level, the opposite of what the diversity hypothesis would predict. These results weaken somewhat after controls for wealth and inequality are introduced, but do not become negative. While the idiosyncrasies of particular goods make generalization perilous, these results certainly provide no evidence for a negative diversity effect on precisely the type of good for which we should expect such an effect to be strongest.

Table 3.

Local Public Goods.

Variables	(1)	(2)	(3)
Variables	Drains	Drains	Drains
Village caste fractionalization	2.958*		3.090*
Village caste fractionalization	(1.711)		(1.753)
Village pop.	0.000711	0.00143	0.000788
Village pop.	(0.00133)	(0.00127)	(0.00134)
Village prop. upper castes		0.961	1.180
Village prop. upper castes		(1.049)	(1.156)
Constant	−3.363***	−1.691***	−3.643***
Constant	(1.195)	(0.452)	(1.261)

Observations	119	119	119

Standard errors in parentheses. The models are logistic regressions with a binary measure of whether a village has drainage ditches as the dependent variable.

p < .1. **p < .05. ***p < .01.

Identity Salience and Social Inequality

Many formulations of the diversity hypothesis have rejected the idea that identity fractionalization is always associated with bad outcomes, arguing instead that this effect is conditional of the effects of ethnic identity being salient. Alternatively, ethnicity might become a focus of conflict if there are high levels of economic inequality between groups, making differences in behavior or redistributive preferences especially large. In the Indian context, this would mean that the relevant variable for the diversity hypothesis is not caste fractionalization but social distance between castes, intercaste inequality, or the interaction of the two with fractionalization. Section “Endogenous Identity: Salience, Segregation and Inequality” tests these hypotheses, and finds that even after accounting for the endogenous salience of social difference, the presence of upper cases has a strong influence on the presence of public goods.

Alternative Hypotheses: Wealth, Social Capital, and Patronage

Wealth

Perhaps the most obvious alternative explanations for the positive relationship between public goods is economic differences between villages. Wealthier villages might both have more upper caste residents and be better able to afford high-quality public services without central assistance. While the geographical controls in Tables 1 and 2 address some of these concerns about unobserved differences between villages, they do not measure differences in wealth and income directly.

Measuring wealth in rural India is difficult, due to the prevalence of noncash income and transfers. To deal with this problem, it is customary to estimate income using expenditures or household stocks of durable goods. Table A.10 includes three such measures, each calculated from the household sample within each village: average aggregate household expenditure in the previous 30 days (with noncash expenditures estimated), average values of the scores of the first dimension of a factor analysis of the possession of 15 household durable goods, and the proportion of households living in “pucca” [well-constructed] houses, as judged by the survey taker. While wealth is generally positively associated with the availability of public goods, the inclusion of the variable does not affect the positive, statistically significant effect of upper caste presence, except for electric power in some models.⁹

Intraelite cooperation and social capital

Section “Ethnic Politics and Public Goods” suggested that the association between the presence of powerful groups and public goods is driven by the superior ability of these groups to obtain favors from higher levels of government. There is, however, an alternative explanation for these findings—that members of politically powerful groups are better at providing public goods for themselves, due to higher levels of social capital and social skills than other groups. While this pathway would be distinct from the diversity hypothesis, it would emphasize the role of local initiative and superior ability to resist free-riding, rather than nonlocal policy. This mechanism would be even more likely if we allow the possibility (considered in the appendix) that local public good can be made excludable.

There is reason to doubt that higher within-group cooperation is driving the results. Anthropological accounts of village India stress the importance of factional divisions within upper caste groups, and the fact that these divisions that are often more politically relevant than caste divisions (Miller, 1965).

Testing the social capital measure directly is hampered by the UPBSLS’s not asking questions that directly pertain to intracaste cooperation. To gain some perspective on this issue, Table A.9 presents levels of a variety of social capital measures taken from the 2005 Indian Human Development Survey, with the sample truncated to correspond to the UPBSLS’s sampling area. There appears strong evidence that conventional measures of social capital are not higher among the upper castes. Membership in caste associations is actually higher among the lower castes, as are (by a smaller margin) membership in other privates associations and generalized preferences for local collective action. Interestingly, the only type of associational membership more common among the upper castes is the only one sponsored by the state, cooperative societies, which in this region serve as a way for channeling cheap loans to the politically well-connected (Witsoe, 2013).

Patronage

A third alternative hypothesis is that the poor public goods are a result of widespread patron–client networks, and the desire of patrons to maintain these networks. Patrons (who, in the South Asian context, are likely to be upper caste) use their control over land and the risk-avoidant behavior of the poor to keep their clients in a dependent position. Public services such as education, which expands the outside options of clients, are thus against their interests, and powerful patrons will prevent their implementation, leading to suboptimal outcomes (Anderson et al., 2015; Shami, 2012).

While this is a plausible and well-attested mechanism, there are reasons to think that it cannot explain the strong association between upper caste presence and public goods. First, we should expect any patronage effect to be concentrated among goods specifically targeting the poor, and affecting wages: In fact, Anderson et al. (2015) find no association between their clientelism measure and the broadly beneficial goods discussed here. Second, if we assume that villages with an upper caste presence are more likely to exhibit clientelistic tendencies, we should expect these villages should receive less public goods rather than more.

Table A.11 examines the association between upper caste population and public goods while conditioning two direct measures of upper caste patronage taken from the household part of the UPBSLS. The first is the proportion of loans to lower caste households given by upper caste individuals other than shopkeepers, and the second is the proportion of lower caste households where the primary source of income was casual agricultural labor, usually identified as being the groups most vulnerable to exploitation by patrons. Neither measure has a noticeable effect on public goods provision, while the effect of the village’s upper caste population remains positive.

Non-Indian Cases

Given the prevalence of results showing a strong connection between diversity and public goods, the results in the section “The Indian Case: Results” showing no connection in Northern India might be thought to stem from some idiosyncratic feature of the case, such as the relatively small role of local initiative in public goods and the hierarchical nature of the caste system. To examine the wider generalizability of these findings, this section examines this hypothesis, using data from two influential studies of the diversity hypothesis: Alesina et al. (1999) and Miguel (2004). The results support the claim made in the section “Ethnic Politics and Public Goods”: The multicollinearity makes it difficult to separate the diversity and power hypotheses in most cases.

Urban America: Replicating Alesina et al. (1999)

Alesina et al. (1999) [ABE], in one of the first studies of the ethnicity on public goods spending, found a negative relationship between racial fractionalization and spending on several categories of public services in American cities and counties. They attribute this problem to heterogeneous preferences across racial groups, which lead many voters to have preferences over goods very different from the median voter, making them unwilling to tax themselves to fund the median voter’s preferred projects.

In many ways, urban America should be a more favorable case for the diversity hypothesis than rural India, given the relative autonomy of American local institutions. However, even in the United States, a large proportion of local budgets come from federal and state funds, funds that often come with constraints on how the money may be spent. While ABE attempt to account for this spending in their models, it is unclear if they can properly separate formulaic transfers to poorer communities from more discretionary favoritism. It is particularly notable that ABE’s strongest results come from a sector (roads) widely reputed to involve “pork” transfers, and that they do not find a direct effect of diversity on local taxes.

Another feature of race in America is that, relative to caste in India, the number of groups is smaller and the population of the most politically powerful group, White Americans, is much larger. This means that the correlation between diversity and the population of White Americans is very high: In ABE’s city sample, ρ = .86. The consequences of this close association are apparent in Table 4. Panel A reproduces ABE’s core results for their sample of U.S. cities: Expenditure on roads and sewage is negatively correlated with diversity, while expenditure on police rises. Panel B shows that the same associations hold for public goods and the proportion of White city residents. However, when both variables are included together (Panel C), the results become less clear. The estimated effect of fractionalization on relative road spending remains statically significant and negative, while the effect of White population remains statistically significant and positive. However, for relative sewage and per capita roads expenditure both estimated coefficients are diminished and statistically insignificant.¹⁰

Table 4.

Replicating Alesina, Baqir, and Easterly (1999): Ethnicity and Public Goods in U.S. Cities.

Variables	(1)	(2)	(3)	(4)
Variables	Prop. police	Prop. roads	Prop. sewage	Road exp. PC
Panel A: Racial
Fractionalization	0.0566***	−0.0962***	−0.0488***	−35.50***
Fractionalization	(0.0118)	(0.0114)	(0.0160)	(9.027)
Constant	0.139***	0.137***	0.143***	91.00***
Constant	(0.00403)	(0.00388)	(0.00546)	(3.079)
Observations	1,020	1,020	1,020	1,020
R ²	.022	.066	.009	.015
Panel B: Majority ethnic group
Prop. White	−0.0478***	0.101***	0.0432***	39.36***
Prop. White	(0.0123)	(0.0118)	(0.0166)	(9.341)
Constant	0.193***	0.0285***	0.0941***	49.34***
Constant	(0.00999)	(0.00958)	(0.0135)	(7.596)
Observations	1,020	1,020	1,020	1,020
R ²	.015	.067	.007	.017
Panel C: ELF and majority ethnic group
ELF	0.0777***	−0.0468*	−0.0587	−7.227
ELF	(0.0268)	(0.0257)	(0.0363)	(20.43)
Prop. White	0.0244	0.0570**	−0.0114	32.64
Prop. White	(0.0277)	(0.0267)	(0.0376)	(21.16)
Constant	0.113***	0.0768***	0.155***	56.79**
Constant	(0.0293)	(0.0282)	(0.0398)	(22.39)
Observations	1,020	1,020	1,020	1,020
R ²	.023	.070	.009	.017

Standard errors in parentheses. The models report the results of a series of linear regression models. Compare to Alesina et al. (1999), Table 4.

p < .1. **p < .05. ***p < .01.

These results do not show that the effect of the presence of a socially powerful group (Whites) is more important than fractionalization. Rather, they show that the two measures are so highly collinear that it is impossible to separate their effects. This emphasizes the importance of the research design used in this article, where diversity can be separated from the presence of a single group. Combined with what we know about the political economy of local finance in the United States, it suggests that differences between White and non-White cities are at least potentially significant, even in cases cited as evidence for the diversity hypothesis.

Rural Kenya: Replicating Miguel (2004)

Miguel (2004) found that village-level ethnic diversity had strong effects on the quality of public services in rural Kenya (where ethnicity is politically salient) but had no such effect in otherwise comparable parts of Tanzania (where ethnicity is less salient). As Miguel only finds a diversity effect in Kenya, this section only reproduces the Kenyan results. Like ABE, Miguel notes that the provision of public services in Kenya is a complicated, multilevel affair, involving both local communities and the central government: In fact, Miguel states that 90% of funding for schools (his central outcome) comes from the central government, though local funding might be important on the margins, as “local inputs and teachers are complements in educational production.” However, given the strong role of nonlocal actors in the production of public services, noncooperative mechanisms certainly seem worth examining in the Kenyan context.

As in the United States, the areas of Western Kenya in this study have a majority group, the Luhya. While the Luhya are neither the largest nor wealthiest group in the country as a whole, they are considered the “dominant” group locally, have been associated with the region for longer than the other major groups, and many local representatives and bureaucrats come from this group (Miguel, 2004). Overall, Luhya presence and ethnic fractionalization are highly correlated, at ρ = .64.

Panel A of Table 5 replicates Miguel’s key findings: Ethnic diversity is negatively associated with the provision of a variety of public services, a relationship that is statistically significant for school funding, school desks, and well maintenance. However, Panel B shows that very similar results can be obtained by substituting the proportion of Luhya in the village population, which is positively associated with the same goods. When the two variables are included in the same model (Panel C) both sets of coefficients diminish sharply, while standard errors become much larger. Both these effects, however, are less pronounced for the Luhya coefficient estimates, which remain statistically significant in the combined model.

Table 5.

Replicating Miguel (2004): Ethnicity and Public Goods in Rural Kenya.

Variables	(1)	(2)	(3)	(4)	(5)
Variables	School funds	Desks per pupil	Latrines per pupil	Classrooms per pupil	Prop. working wells
Panel A: ELF
ELF	−3.528**	−0.291***	−0.0152	−0.00981	−0.325***
ELF	(1.739)	(0.0864)	(0.0106)	(0.0113)	(0.110)
Constant	4.278***	0.276***	0.0196***	0.0329***	0.632***
Constant	(0.473)	(0.0235)	(0.00289)	(0.00307)	(0.0300)
Observations	84	84	84	84	84
R ²	.048	.121	.024	.009	.096
Panel B: Majority ethnic group
Prop. Luhya	1.753***	0.153***	0.00136	0.00475	0.133***
Prop. Luhya	(0.659)	(0.0313)	(0.00415)	(0.00433)	(0.0422)
Constant	2.314***	0.109***	0.0151***	0.0275***	0.470***
Constant	(0.487)	(0.0231)	(0.00306)	(0.00320)	(0.0312)
Observations	84	84	84	84	84
R ²	.079	.226	.001	.014	.108
Panel C: ELF and majority ethnic group
ELF	−1.215	−0.0792	−0.0206	−0.00368	−0.184
ELF	(2.160)	(0.102)	(0.0134)	(0.0142)	(0.137)
Prop. Luhya	1.469*	0.134***	−0.00347	0.00389	0.0899*
Prop. Luhya	(0.833)	(0.0395)	(0.00518)	(0.00548)	(0.0529)
Constant	2.784***	0.139***	0.0231***	0.0289***	0.541***
Constant	(0.968)	(0.0458)	(0.00601)	(0.00637)	(0.0614)
Observations	84	84	84	84	84
R ²	.083	.231	.030	.015	.127

Standard errors in parentheses. The models report the results of a series of linear regression models. Compare to Miguel (2004), Table 2. ELF = ethnic franctionalization.

p < .1. **p < .05. ***p < .01.

As with the ABE results, collinearity makes it difficult to separate the effect of diversity from the presence of the local majority group in the Kenyan case. They do, however, suggest that the discrimination hypothesis is just as consistent with the data as the diversity hypothesis, and in fact seems to fit the variation somewhat better.

Conclusion

The results presented in the sections “The Indian Case: Results” and “Non-Indian Cases” challenge the idea that the link between ethnic identity lies in the failure of ethnic groups to cooperate. Instead, they suggest that certain politically powerful ethnic groups are better able than others to obtain services for their communities than others, whether because of their coethnic links to higher-level political actors or the generally stronger social position of group members. While these results suggest a different mechanism from that outlined in the diversity literature, they are consistent with the data from much of this literature, due to the collinearity between the presence of powerful groups and standard diversity measures.

This is not to say that cooperation, and failures of cooperation caused by ethnic divisions, does not occur. In some situations, such as the provision of goods in highly decentralized countries with no asymmetrically powerful groups, the cooperation mechanism may very accurately capture the process by which public services are supplied. However, where political power is centralized and ethnic power relations are unequal, bias toward powerful groups is an important factor in distribution, consistent with the large existing literature on ethnic favoritism. As both centralized distribution (Singh & vom Hau, 2016) and ethnic inequality in political power (Cederman et al., 2010) are very common, this later combination may be more the rule than the exception.

As they focus on distributional decisions within countries, these results cannot speak to the relationship between diversity and public goods provision at the national level. However, the results complement recent work that shows that country-level interethnic inequalities are associated with failures of public goods provision (Baldwin & Huber, 2010) and that historical factors may explain the country-level between public goods and diversity (Singh & vom Hau, 2016; Wimmer, 2016) by showing how national-level inequalities can influence local-level variation.

These results have implications for policy. Many development interventions, motivated by the diversity literature, have sought to improve public services by improving within-village cooperation (Fearon, Humphreys, & Weinstein, 2009). These findings suggest, by contrast, that a focus on the policies of regional or national political actors is often appropriate, and attempts to empower politically weak ethnic groups may prove an effective strategy to change inequalities in public goods provision. This is not to suggest the behaviors of these actors will be easy to change: Bias toward the powerful is deeply seated in most political systems. However, a good place to start is by accepting that the problems of diverse communities are, at least in some cases, beyond their control.

Footnotes

Acknowledgements

The author would like to thank Jack Paine, Henk Goemens, Dan Posner, and participants at the Rochester Faculty Working Group, Juan March Institute Seminar, American Political Science Association annual meeting, Midwest Political Science Association annual meeting for their comments on the manuscript.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Supplemental Material

Supplementary material for this article is available online at the CPS website .

Notes

Author Biography

Alexander Lee is an assistant professor of Political Science at the University of Rochester. His research focuses on the politics of South Asia, historical political economy, and underdevelopment.

References

Adida

C. L.

(2015). Do African voters favor coethnics? Evidence from a survey experiment in Benin. Journal of Experimental Political Science, 2, 1-11.

Alesina

Baqir

Easterly

(1999). Public goods and ethnic divisions. The Quarterly Journal of Economics, 114, 1243-1284.

Algan

Hémet

Laitin

(2011). The social effects of ethnic diversity at the local level: A natural experiment with exogenous residential allocation. Journal of Political Economy 124, 696-733.

Anderson

Francois

Kotwal

(2015). Clientelism in Indian villages. The American Economic Review, 105, 1780-1816.

Baldwin

Huber

J. D.

(2010). Economic versus cultural differences: Forms of ethnic diversity and public goods provision. American Political Science Review, 104, 644-662.

Banerjee

Iyer

Somanathan

(2005). History, social divisions, and public goods in rural India. Journal of the European Economic Association, 3, 639-647.

Banerjee

Somanathan

(2007). The political economy of public goods: Some evidence from India. Journal of Development Economics, 82, 287-314.

Bates

R. H.

(1983). Modernization, ethnic competition, and the rationality of politics in contemporary Africa. In Rothchild

D. S.

Olorunsola

V. A.

(Eds.), State versus ethnic claims: African policy dilemmas (Vol. 152, p. 171). Boulder, CO: Westview Press.

Besley

Pande

Rahman

Rao

(2004). The politics of public good provision: Evidence from Indian local governments. Journal of the European Economic Association, 2, 416-426.

10.

Burgess

Jedwab

Miguel

Morjaria

i Miquel

G. P.

(2015). The value of democracy: Evidence from road building in Kenya. The American Economic Review, 105, 1817-1851.

11.

Carlson

(2015). Ethnic voting and accountability in Africa: A choice experiment in Uganda. World Politics, 67, 353-385.

12.

Cederman

L.-E.

Wimmer

Min

(2010). Why do ethnic groups rebel? New data and analysis. World Politics, 62, 87-119.

13.

Chakravarti

. (2001). Social power and everyday class relations: Agrarian transformation in North Bihar. Thousand Oaks, CA: Sage.

14.

Chandra

(2007). Why ethnic parties succeed: Patronage and ethnic head counts in India. Cambridge, UK: Cambridge University Press.

15.

Clarke

K. A.

Primo

D. M.

(2012). A model discipline: Political science and the logic of representations. Oxford University Press.

16.

Dixit

Londregan

(1996). The determinants of success of special interests in redistributive politics. The Journal of Politics, 58, 1132-1155.

17.

Easterly

Levine

(1997). Africa’s growth tragedy: Policies and ethnic divisions. The Quarterly Journal of Economics, 112, 1203-1250.

18.

Ejdemyr

Kramon

Robinson

A. L.

(2015). Segregation, ethnic favoritism, and the strategic targeting of distributive goods1 (APSA Annual Meeting Paper). Comparative Political Studies, September 25, 2017. Retrieved from http://stanford.edu/~ejdemyr/docs/EKR-20151103.pdf

19.

Fearon

Humphreys

Weinstein

J. M.

(2009). Can development aid contribute to social cohesion after civil war? Evidence from a field experiment in post-conflict Liberia. The American Economic Review, 99, 287-291.

20.

Fehr

Hoff

Kshetramade

(2008). Spite and development. The American Economic Review, 98, 494-499.

21.

Franck

Rainer

(2012). Does the leader’s ethnicity matter? Ethnic favoritism, education, and health in sub-Saharan Africa. American Political Science Review, 106, 294-325.

22.

Gao

(2016). Tribal mobilization, fragmented groups, and public goods provision in Jordan. Comparative Political Studies, 49, 1372-1403.

23.

Gerring

Thacker

S. C.

Huang

(2015). Does diversity impair human development? A multi-level test of the diversity debit hypothesis. World Development, 66, 166-188.

24.

Gisselquist

Leiderer

Niño-Zarazúa

(2016). Ethnic heterogeneity and public goods provision in Zambia: Evidence of a subnational “diversity dividend.” World Development, 78, 308-323.

25.

Habyarimana

Humphreys

Posner

D. N.

Weinstein

J. M.

(2007). Why does ethnic diversity undermine public goods provision? American Political Science Review, 101, 709-725.

26.

Hodler

Raschky

P. A.

(2014). Regional favoritism. The Quarterly Journal of Economics, 129, 995-1033.

27.

Jaffrelot

. (2003). India’s silent revolution: The rise of the lower castes in North India. New York, NY: Columbia University Press.

28.

Jaffrelot

Kumar

(2012). Rise of the plebeians? The changing face of the Indian legislative assemblies. Delhi, India: Routledge.

29.

Jha

S. S.

(1972). Political elite in Bihar. Bombay, India: Vora.

30.

Jost

J. T.

Kivetz

Rubini

Guermandi

Mosso

(2005). System-justifying functions of complementary regional and ethnic stereotypes: Cross-national evidence. Social Justice Research, 18, 305-333.

31.

Kitschelt

Wilkinson

. (2007). Patrons, clients, and policies: Patterns of democratic accountability and political competition. Cambridge, UK; New York, NY: Cambridge University Press.

32.

Kramon

(2013). Vote buying and accountability in democratic Africa (Doctoral dissertation). Stanford University, CA.

33.

Lee

(2016). Development in multiple dimensions: Social power and regional policy in India. Unpublished manuscript. Retrieved from https://www.rochester.edu/college/faculty/alexander_lee/wp-content/uploads/2014/09/chapter1revised.pdf

34.

Lee

. (2017). Redistributive colonialism: The long term legacy of international conflict in India. Politics & Society, 45(2), 173-224.

35.

Leighley

J. E.

Vedlitz

(1999). Race, ethnicity, and political participation: Competing models and contrasting explanations. The Journal of Politics, 61, 1092-1114.

36.

Lindberg

S. I.

(2010). What accountability pressures do MPs in Africa face and how do they respond? Evidence from Ghana. The Journal of Modern African Studies, 48, 117-142.

37.

Miguel

(2004). Tribe or nation? Nation building and public goods in Kenya versus Tanzania. World Politics, 56, 328-362.

38.

Miguel

Gugerty

M. K.

(2005). Ethnic diversity, social sanctions, and public goods in Kenya. Journal of public Economics, 89, 2325-2368.

39.

Miller

D. F.

(1965). Factions in Indian village politics. Pacific affairs, 38, 17-31.

40.

Min

(2015). Power and the vote: Elections and electricity in the developing world. Cambridge, UK: Cambridge University Press.

41.

Portes

Rumbaut

R. G.

(2006). Immigrant America: A portrait. California, UK: University of California Press.

42.

Shah

(2006). Untouchability in rural India. Thousand Oaks, CA: Sage.

43.

Shami

(2012). Collective action, clientelism, and connectivity. American Political Science Review, 106, 588-606.

44.

Singh

vom Hau

. (2016). Ethnicity in time: Politics, history, and the relationship between ethnic diversity and public goods provision. Comparative Political Studies, 49(10), 1303-1340.

45.

Soifer

H. D.

(2016). Regionalism, ethnic diversity, and variation in public good provision by national states. Comparative Political Studies, 49, 1341-1371.

46.

Srinivas

M. N.

(1987). The dominant caste and other essays. Delhi, India: Oxford University Press.

47.

Thakur

(2000). The making of Laloo Yadav: The unmaking of Bihar. Delhi, India: HarperCollins India.

48.

Tilly

(1998). Durable inequality. California, UK: University of California Press.

49.

Trounstine

. (2016). Segregation and inequality in public goods. American Journal of Political Science, 60(3), 709-725.

50.

Whitt

Wilson

R. K.

(2007). The dictator game, fairness and ethnicity in postwar Bosnia. American Journal of Political Science, 51, 655-668.

51.

Wimmer

(2002). Nationalist exclusion and ethnic conflict: Shadows of modernity. Cambridge, UK: Cambridge University Press.

52.

Wimmer

. (2016). Is diversity detrimental? Ethnic fractionalization, public goods provision, and the historical legacies of stateness. Comparative Political Studies, 49(11), 1407-1445.

53.

Wimmer

(2017). Nation building: Why some countries come together while others fall apart. Princeton, NJ: Princeton University Press.

54.

Witsoe

(2013). Democracy against development: Lower-caste politics and political modernity in postcolonial India. Chicago, IL: University of Chicago Press.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.58 MB

0.00 MB