Don t Touch My Road.

Don t Touch My Road. Evidence from India on Segregation and Affirmative Action. In progress - Please do not quote or cite without permission VICTOIRE GIRARD October 2015 Abstract Inter-group relations may take the form of segregation, with public goods turned into club goods as a result. Caste-based discrimination is a striking example of this segregation process. In the Hindibelt, the heartland of India, 44.5% households members of the marginalized castes labeled Scheduled Castes (SCs) declared that some streets were off-limits due to their caste in the 2006 survey used in the article. The exclusion rate was 20 percentage points higher 10 years before. This article investigates whether affirmative action in the form of mandated political representation has played a part to decrease exclusions. The identification strategy takes advantage of the Indian system of seat reservation for SC members in local assemblies, which changes SC s local power. The effect of political reservations for SC s access to streets appears to be large and significant, both statistically and economically. However, the effect does not last after the end of the reserved term. This evidence underlines important consequences of political reservation in the short run. However, it also questions legislators goals when using positive action as a transitory tool, to be abandoned once practices and beliefs have been altered. Keywords: Roads, Public Goods, India, Inequality, Caste, Political Reservation, Discrimination JEL D63, D74, J15, O12, O53 LEO, Univerite d Orleans and CES, Universite Paris 1 Pantheon Sorbonne. Contact: victoire.girard@univ-paris1.fr. I thank Andrew Foster for providing the data. 1

1 Introduction Inter-group relations may take the form of segregation. Caste-based discrimination offers a striking example of segregation. In rural India, caste groups can define the group of people entitled to access a good. As a result, some public goods are turned into caste club goods. In 2006, 44.5% household members of the marginalized scheduled castes (SCs, formerly called untouchable ) declared some streets were offlimits due to their caste in states of the Hindi belt. 1 Streets are usually considered as a non-excludable public good. But castes make things more complex in India. The Indian Constitution of 1949 abolished untouchability and universal access to public goods is a cornerstone of this abolition. Yet, news reports abound on how SC households may be punished for breaking a caste-based rule. 2 Caste-based exclusions, from streets or other public goods, are not leftovers from old habits that would mechanically fade away. 3 This paper investigates whether a mandated political representation can help to improve marginalized communities access to public goods. I am interested in the effectiveness of political quotas to improve low castes access to public goods. SCs and other low castes benefit from quotas under the form of seat reservations in Indian local political assemblies. I evaluate whether quotas decrease street exclusion. Street exclusion is a striking example of caste-based discrimination, I exploit a survey asking households whether they are excluded from some streets because of their caste at the moment of the survey and 10 years before. Caste quotas could help to compensate for and eradicate a legacy of discrimination in several manners. Low-caste representatives can -and do- favor their peers in the law and allocation of funds they implement (beneficiary selection or location of the public goods to be maintained or built). However the legacy of discrimination goes beyond an unfair targeting of public money beneficiaries. It appears in everyday interactions. Proponents of positive action in India have justified it as a transitory tool. One 1 States where Hindi is widely spoken. In this study: Bihar, Haryana, Madhya Pradesh, Rajasthan, and Uttar Pradesh. 2 A special law, the Prevention of Atrocities Act, is dedicated to protect SCs and STs against caste-based offenses. However, the Union Minister for Home Affairs declared in 2010: The number of cases of atrocities against the Scheduled Castes registered in 2006 was 26,665. [...] Many of the cases are simply not registered. [...] What makes it doubly painful is that there is rise in atrocities, but when you try to prosecute and convict, the conviction rate is only 30%. It was 28%, 31.4% and 32% [in 2006, 2007 and 2008]. Not only are acquittals very high; pendency is about 80%. Moreover, privatized public goods can trigger such atrocities. In Bros and Couttenier (2015), open wells trigger more atrocities against SCs than sources that are believed to be less prone to ritual pollution. 3 Households considered untouchable could not access public roads in 18% of the villages in 2001-2002 (Shah, Mander, Thorat, Deshpande, and Baviskar, 2006). The rate of exclusion varies with the type of public good denied: water facilities were privatized in 50% of the villages, primary health centers in 18% of the villages. 2

of the purposes is to change everyday interactions, otherwise once reservation is repealed, the country will be back to a pre-reservation (discriminatory) state. The main aim of this article is to check whether reservation changes interactions. I additionally discuss which channels are are consistent with the pattern of results. My identification strategy relies on the way quotas are attributed. Quotas are attributed to villages within each state by the state administration. The attribution rule is state-specific. It can be random or depend upon village-level characteristics. I identify the effect of political quotas on SC households access to streets through within-village variation, which allows me to clear out the effect of village-level characteristics. I document a large and significant (both statistically and economically) effect of political quotas on low-caste members access to streets. The more time since the election and the quota implementation, the bigger the effect. However, this effect does not last after the end of the reserved term, and does not extend to the private sphere (measured by perceived discrimination on the job market). Reservation can improve access to streets during the reserved term through several channels, detailed in the next section. The pattern of results in the article is consistent with an essential part played by the SC leader while in office. This article is an extension of the literature on the impact of political representation and quotas. Its distinction is the focus on interactions. Quotas are implemented to compensate for, and eventually eradicate, a discriminatory legacy. Most of the literature on minority representation in India studies whether the leader s financial or voting choices are a function of the leader s identity. 4 However, laws and public fund allocation are not the only outcomes affected by an elected representative s identity. For instance, a female leader increases the proportion of female candidates at the next election, because it decreases both voters stereotypes on gender roles in the public and private sphere and parties biases against women candidates (respectively Beaman et al., 2009; Bhalotra et al., 2013). Gender quotas have also led to a tremendous increase in the number of crimes against women (Iyer, Mani, Mishra, and Topalova, 2011). This increase is due to an increase in crime reporting - not in crime incidence. Iyer et al. (2011) interpret this increase as a signal that women have a better access to police stations. The literature on the impact of political representation by caste on constituents actions is scarcer. Gille (2014) documents that having an Other Backward Caste (not SC) leader in office makes OBC constituents more likely to apply for a 4 An exhaustive list is beyond the scope of this introduction, seminal references are Chattopadhyay and Duflo (2004) for women and Besley, Pande, Rahman, and Rao (2004) for castes. Articles in this literature often use minorities representation to evaluate whether the leader s identity matters at all, an observation in contradiction with the Downsian voting model. 3

reserved job. Closer to the perspective of the present article, Chauchard (2014) assesses the consequence of reservation on everyday interactions. He finds that SC quotas decrease the enforcement of caste-based discrimination even if stereotypes are not affected. He attributes this finding to the emergence of new norms of interaction between castes thanks to reservation. I complete the above-mentioned literature in two respects. Firstly, I provide additional evidence on the effect of quotas. Quotas are still a controversial policy tool, in particular political quotas which temporarily prevent the usual candidates to run from office. In Rohini Pande s words, at Independence, the Indian state committed to use public policy to end caste-based discrimination, and to improve the economic status of disadvantaged groups. (Pande, 2003). Quotas are a centerpiece of this commitment. The impact of quotas on street access encompasses both objectives cited by Pande (2003): it is a question of caste-based discrimination; and freedom to move is a pre-requisite of economic empowerment. The combination of both objectives makes road access a great candidate to assess the consequence of reservations. Secondly, most of the economic literature on quotas, and in particular the literature on caste-based quotas, asks whether the way funds are allocated is a function of the leader s caste. A fund re-allocation requires diverting resources from some projects to put these resources into other projects (favored by the minority group). I ask whether quotas can improve access to already-existing public goods. Improved access is a matter of overcoming traditional restrictions to the free use of pre-existing public goods. Improved access does not require any resource re-allocation. 5 To my knowledge, the question has not yet been addressed in the economic literature. The most relevant precedent I am aware of is Chauchard (2014). However, my study differs from his in at least two important respects. A first distinction is technical. Because of his data design, 6 Chauchard (2014) cannot assess the effect of reservation over time. A second distinction pertains to the question tackled. Chauchard (2014) relies on declarations of discriminatory intentions by non-scs. But surveying discriminatory intentions is delicate. Non-SCs may manipulate their declarations, and action may differ from intention. In this article, I study the same question but from a different angle: I use SCs declarations after non-scs actions have taken place. In the next section, I present the background of this study with basic caste features, the reservation 5 One could also expect improved access to signal a change in the nature of everyday interactions. In the long run, this change could be a signal of successful quotas. However the evidence available in this article does not signal any long-lasting change in the nature of everyday interactions. 6 He uses first-hand data from a sample of never reserved and first time reserved villages. 4

system, and possible channels through which reservation can affect caste-based discrimination. Section 3 lays out the empirical strategy. I present the data and results in sections 4 and 5 respectively. Section 6 concludes. 2 The Indian Context 2.1 Castes and Segregation Castes have shaped the Indian social setting for more than 3,000 years. Three key features of castes are important to keep in mind for this study. First, castes are hereditary, exclusive and virtually unchangeable at the household level. 7 Second, castes are ordered on a social status ladder, which matches a purity ladder, where so-called ritual pollution may happen between people of different purity statuses if some rules to prevent this pollution are not followed. 8 Third, and closely linked to the second aspect, caste groups are segregated: this has led to forms of spatial segregation, and to historically very strict matrimonial segregation. These theoretical features translate into preferences or rules affecting everyday life interactions. Up until now, castes have structured business networks (Munshi, 2011), and severely curbs mate selection (Banerjee, Duflo, Ghatak, and Lafortune, 2013), while revealing subjects castes in an experiment significantly affects their performances (Hoff and Pandey, 2006). Castes are recorded in the Census through four broad groups, Scheduled Castes (SC), Scheduled Tribes (ST), Other Backward Castes (OBC) and Other Castes (OC). Scheduled Caste households, which belong to a list of particularly backward castes and used to be considered untouchables, still suffer from caste-based discrimination. This translates in a variety of ways, including exclusion from market, or systematic deviation from market price at their disadvantage, being sold/rented out products at higher prices, and bought/rented in products at lower prices than non-sc households (Thorat, Mahamanlik and Sadana s survey results in Thorat and Newman, 2010). Spatial segregation is also observed, as in a traditional Indian village SCs live in a separate neighborhood or hamlet (Deliège, 2004). Lastly, SCs suffer from widespread exclusion. Exclusion can be formal, through clear-cut access denial as outlined in the introduction, or more pernicious, through a differential treatment, which has been widely 7 Even if some cases of identity manipulation have been recorded, these cases are marginal, and correspond to group-level strategies, to improve the status of a caste in society, or on the contrary benefit from some positive action policies (Deliège, 2004; Cassan, 2013). 8 Ritual pollution is believed to happen for example if a member of a high caste performs ritually dirty actions like scavenging or shares some intimacy (direct touching or sharing food or drinks) with a member of a low caste. In the case of a failure to comply with the rule, being polluted has spiritual consequences, but time or de-polluting activities usually allow to get one s purity back. 5

documented in the case of schooling. Low-caste students can be ignored or physically punished by teachers (PROBE report, 1999), or receive less good grades for similar answers (Hanna and Linden, 2012). Beside the moral aspect of the question, universal access to public goods is central for SCs economic empowerment, and thereby for India to reduce poverty. 2.2 Reservations In 1993, the 73rd amendment to the Constitution of India instituted local political councils called Gram Panchayats (GP hereafter). GPs are elected. GPs have decision power over maintaining and building local public goods such as roads or water devices. GPs also decide who are the households entitled to social programs. GPs are composed of ward representatives and headed by a pradhan. Pradhans have agenda-setting power in panchayat meetings, but no veto power. They are the only council members working full time. Pradhans are elected either directly by GP constituents, or indirectly by members of the GP council. The 1993 reform is also important because it implemented quotas, as a tool of positive discrimination. Both the pradhan and the ward representative seats can be reserved to low-caste members (SC, ST or OBC) or women. Seats are reserved for one term at a time. Reserved seats rotate between villages.the proportion of caste quotas varies within each state: each caste quota is proportional to the weight of the caste in the state population. 9 The rotation rule differs from one state to another. 10 Some states allocate reserved seats in a purely random way (e.g. West Bengal, as exploited by Bardhan et al., 2010), others list villages according to the proportion of their population which is SC/ST/OBC, and use these lists to attribute reservations (e.g. Rajasthan, as exploited by Chauchard, 2014). What is essential for the identification strategy in this article is that the rotation is decided by the state administration. The decision can be a function of village-level attributes, but which villages are concerned by the pradhan seat reservation during a given term remains an administrative decision. Gender quotas are imposed in a third of all constituencies, and the rotation of gender quotas is random. Reservation can improve access to streets during the reserved term through several -non exclusivechannels. An SC leader could trigger or catalyze a reaction among SCs to voice their rights. In Hirschman (1970) s framework, a leader can help dissatisfied individuals to voice their concern (rather than follow 9 The 73rd amendment of the Constitution specifies that "the number of offices of Chairpersons reserved for the Scheduled Castes and Scheduled Tribes in the Panchayats at each level in any State shall bear, as nearly as maybe, the same proportion to the total number of such offices in the Panchayats at each level as the population of the Scheduled Castes in the State or of the Scheduled Tribes in the State bears to the total population of the State." 10 The precise rule is not public in all states 6

the exit option, e.g. migration, or the loyalty option, accepting the discriminatory situation). SCs collective consciousness can play an important part since the anticipation of discrimination can on the contrary lead to a vicious circle of low achievement for the low castes (Hoff and Pandey, 2006). Alternatively, non-scs may update their priors -stereotypes- about SCs, thanks to exposure to an SC leader. Reservation ensures that a member of the marginalized SCs has both social visibility, and responsibilities. This contact channel is in line with Allport (1954) s theory, partly confirmed for women s reservation in Beaman et al. (2009). However, non-scs could also change attitudes toward SCs independently from their priors. This change could be consistent with two stories. In the first story, the low-caste leader, while in a position of power, acts to ensure better living conditions for other low-caste members. The leader can either negotiate this improvement with multi-caste partners, or enforce this improvement against the other castes (Chauchard, 2014). 11 In the second story, the social norm changes, because non-scs perception of the norm what is socially acceptable evolves away from segregation. In this story, suggested by Chauchard (2014), non-scs are less reluctant to everyday interactions with SCs because they see other non-scs interact with SCs (yet the non-scs stereotypes are unaffected). 12 3 Estimation Strategy The aim of this article is to assess the link between reservation of the pradhan s seat for an SC and SC households access to streets. Although the dependent variable is binary, I rely on linear probability models because of the important number of fixed effects included in the second specification. 13 First, I estimate the following equation in cross-section to investigate street access in 2006: 11 As outlined earlier in this Introduction, the Prevention of Atrocities Act exists to protect low castes from the leftovers of a discriminatory legacy. But low castes do not systematically take advantage of this law, because they have difficulties accessing officials (if only because 18% of them cannot access a police station Shah et al., 2006), and because even once cases are reported, the judiciary system can let them go unsolved. One can expect a parallel with Iyer et al. (2011) s observation that women more easily file complaints when a woman is elected in local political bodies. Lastly, (Chauchard, 2014) outlines that the high caste know that the SC leader interacts with higher rank officials. 12 Although (Chauchard, 2014) s work is not on roads, he notes that As the [SC leader] and their entourage walk through village streets to assess various public works, villagers see members of the SCs on streets on which they otherwise dared not venture. 13 The use of Linear Probability Models has spread since Angrist (2001) s plea that quantities of interest are, ultimately, marginal effects. 7

Street_access iv = 1 + 2 Reserv_SC v SC i + 3X 0 i + v + " iv (1) where Reserv_SC is a village dummy equal to one when the village pradhan s seat is reserved to a SC. SC is a dummy for Scheduled Castes households. The coefficient of interest is 2, it tells whether SC households access to streets in 2006 is affected by the fact that their pradhan is an SC, elected after a caste quota. X i is a vector of household-level controls which can be related to the household s social status in the village. X i contains : the household caste category, its exclusion from streets in 1996, the household head s sex, education level (above or below primary), age, income and agricultural occupation. The main challenge in this study is the endogeneity of the reservation timing. To allocate reserved seats between villages, state administrations can rely on village-level criteria. At a given moment, villages with reservations can therefore differ from villages without reservations and the impact of having an SC pradhan on SCs cannot be identified. I follow Besley et al. (2004) s strategy to tackle this issue: I account for village characteristics through village fixed effects v. v accounts for all the village-level unobservables which may be correlated to the reservation status of the village, or to the probability that SCs are excluded in the village given their share in the village population. This strategy allows me to identify the impact of reservation through within-village variations. The error term, " iv is clustered at the level of the variable of interest, namely caste categories within villages. 14 The issue of the above specification is that I cannot rule out a declaration bias, by SC respondents, correlated with quotas. SCs may be over-optimistic when one of them is the pradhan: they may feel more legitimate and consequently declare less segregation although nothing has changed. This overoptimism would lead to an amplification bias in the estimation of 2. On the contrary, SCs may become over-sensitive to discrimination, and complain more easily about it, during an SC reserved term. This over-pessimism would lead to an attenuation bias in the estimation of 2. To deal with this concern, I follow two strategies. Firstly, I take advantage of the fact that both street access and reservation are observed at two different points in time, both at the time of the survey and 10 years before, through a recall question. I can therefore control for all time invariant characteristics at the household level through household fixed effects. These fixed effects can account for any systematic declaration bias, which could be correlated with reservation (including over-optimism or pessimism). I implement this strategy in equation 2 below. However, I cannot rule out that the bias is time variant. SC quotas may bias SCs declarations only at the moment when they apply to the village. To check for this 14 Results are robust to clustering at the village level. 8

possibility, I can check the impact of reservation on another outcome than street access. I describe the implementation of this test later in this section. The panel specification is: Street_access ivt = Reserv_SC vt SC i + vt + gt + i + ivt (2) The coefficient of interest is, it has the same interpretation as 2 in the first specification. i is household fixed effects to account for time invariant unobservables at the household level. The household caste is absorbed by these fixed effects. vt accounts for possible changes over time at the village level, which may be correlated to the reservation status of the village. gt accounts for the general trend of improvement observed at the group level, which can be independent from village-level reservations. The error term ivt is clustered two ways, to account for shocks both at the household level and at the caste group level within each village and year (Cameron, Gelbach, and Miller, 2011). The scale of the second cluster is the scale of the interaction term of interest. 15 The panel estimation relies on households which declare a change in their access to streets between 1996 and 2006. is significant only if the change in street access is different between SCs and others in their village when there is a reservation, and if this change is different from the general trend of improvement observed for SCs in the whole sample between 1996 and 2006. Results are consistent with a positive impact of SC quotas on the probability that SCs can access every street in their village. I additionally control for the interaction term of SC household and SC pradhan in all specifications, but the two most parsimonious ones. SC pradhans elected outside caste quotas seem to worsen SCs access to streets: the coefficient of the interaction term is positive (it increases street exclusion). Because elections outside quotas are endogenous, and mix several situations, the coefficient is hard to interpret. Elections outside quotas mix the effect of an election contested between people of different castes, the leader s gender identity (since I outline in the next section that most elected SC pradhans are elected to seats reserved for women), and unobservables. Because of the descriptive statistics presented in Section 3, in sub-section 5.2 I control for gender quotas, and an interaction term between gender quotas and SC pradhans. I use these dummies to investigate the impact of a pradhan s caste and gender identities on SCs. The rest of the article focuses on the robustness, credibility, and interpretation of the results on SC quotas and SCs access to streets. 15 Results are robust to clustering at both the household and the year*village level. 9

I check the result validity against the sample definition. Firstly, I exclude villages where an SC quota has been imposed in the past. In this sample, the control group has never received any reservation treatment. With this sample I can estimate a standard difference-in-difference estimator. Secondly, I include all Indian states in the sample. SCs living outside the Hindi-belt do not face exclusion nearly as much as SCs living in the Hindi-belt. As a consequence, the dependent variable does not vary as much outside the Hindi-belt as inside the Hindi-belt. It is however interesting to check how the results can be transposed to the whole country. I check the effect of quotas over time. Firstly, I check whether the effect is a function of the time elapsed since the election. I consider the time elapsed between the latest panchayat election and the moment of measurement of street exclusion. Panchayats with an SC quota have been in place for 1 to 4 years. In order to allow any functional shape, I create four new dummy variables to replace the SC res_sc term, one dummy for each panchayat duration. SC res_sc_y 1 takes the value one for SC households living in one of the 4 villages where an SC pradhan has been elected after an SC quota, for more than one year but less than two years. I follow the same logic to create SC res_sc_y 2 (2 villages), SC res_sc_y 3 (7 villages), and SC res_sc_y 4 (11 villages). Secondly, I check whether reservation has a persistent effect. To do so, I add an interaction term for SC reservation during the previous panchayat term. The pradhan s seat was reserved to an SC in the term prior to the 1996 term in 1 village only, and in the term prior to the 2006 term in 7 villages. 16 Then, I test how quotas affect other groups. SC quotas have been designed to help SCs, but quotas for one low caste could have a spillover effect for other low castes. Firstly, I test whether the street access of STs, or OBCs, is a function of SC reservation. OBCs and SCs are usually more like rivals than friends, but it is interesting to check whether there are spillovers from SC reservation on other low castes. Secondly, I test the symmetric relation: whether OBC seat reservation improves SCs access to streets. Thirdly, I check whether Muslims access to streets is a function of SC reservation. I do not expect Muslim households to be affected by reservation for an SC, this specification is a placebo test. Muslim exclusions are grounded in a Hindu/Muslim divide, while reservations for low castes aim at reducing the low/high-caste gap. In the last section, I discuss the reliability and interpretation of the results. In a first step, I check 16 The Constitution was amended in 1993, but states knew before that year that they would have to implement it. And some states had pre-existing panchayats. Although most states had their first panchayat election in 1995 in the framework of the law, election timing varies across states. 10

the result against the concern of a time-varying declaration bias, induced by reservation. To do so, I test whether ongoing reservation makes SCs over-optimistic about caste-based discrimination in other domains than street access. The REDS2006 asks household heads whether they or a family member have been discriminated against, because of their caste or religion, while seeking a job. The question is asked separately for the household head and other household members, and for the time of the survey and 10 years before. Respondents and their relatives perceived discrimination on the job market decreased by one and three percent, respectively, over the 10-year period. These variables clearly rely on the interviewee s interpretation. If reservation leads to a time-varying declaration bias by SCs, I would expect this variable to be affected in the same direction as street access declarations. In a second step, I discuss which channels are consistent with the pattern of results. I consider the following channels, presented in the introduction: a collective action by SCs, the evolution of non-scs priors or perception of the norm, or SC pradhans actions. To assess the relevance of these channels, I take advantage of the fact that they should have different implications in the long run, when the reserved term has ended. 4 Data 4.1 Dataset Data are taken from the last round of the Rural Economic and Development Survey (REDS), undertaken in 2006 by the National Council for Applied Economic Research (NCAER). It contains both a survey at the village level and a complete census of households in the villages surveyed. The dataset contains information about 242 villages in 17 major Indian states, while the household census covers 115,000 households. This article uses the REDS2006. It is focused on the Hindi-belt states, namely the states of Bihar, Haryana, Madhya Pradesh, Rajasthan and Uttar Pradesh. The Hindi belt is the heartland of India, which encompasses on average poorer states and displays more tensed inter-caste relationships than southern India (Jaffrelot, 2002). In the villages surveyed, the 1996 exclusion figure for SCs in Hindi Belt states is 65% against 12% in non-hindi-belt states. Politics is also more caste-oriented in these states, where SCs have risen as a political group (Jaffrelot, 2002). The area is particularly suitable for studying how caste-based political reservation can affect caste-based exclusion. The final sample encompasses 48,219 households, spread over 89 villages. 11

4.2 Street Exclusions The household census contains information on whether certain village streets are off-limit for respondents due to their caste or religious identity. Respondents are asked one question about caste-based exclusion from streets, and another question about religion-based exclusion. Information is collected both for the time of the survey and 10 years before, through a recall question. Grounding street access in caste identity means that streets which are supposed to be public are privatized. Access to a supposed public good becomes a function of the household caste. High castes can enjoy the good, but prevent low castes from enjoying it. Traditional Indian villages are spatially segregated: high castes live in one neighborhood and low castes in another neighborhood, or a separate hamlet. Street exclusion declarations by SCs correspond to the fact that they cannot enter some parts of the village. This spatial segregation is one among many consequences of the religious belief that high castes should avoid interaction with a lower castes. This idea is grounded in the belief that contact with a low caste could ritually pollute the high caste. Descriptive statistics on the extent of street exclusions and how they vary with group identities in Table 1 show that SCs, as expected, declare the bulk of street exclusions. STs declare fewer street exclusions, although their social status is comparable to the SCs status. While only 7% of STs suffered from street exclusion in 2006, the corresponding figure for SCs is 44.5%. But SCs and STs display very different settlement patterns. STs are descended from tribes and live in isolated autonomous villages. SCs on the contrary are part of traditional multi-caste villages. Historically, SCs handled jobs which higher castes did not want. As a result, STs more often than SCs live in places where they belong to the majority or are a very small minority: 60% of STs live in villages where STs represent less than 10% or more than 50% of the population, while only 10% of SCs live in villages where SCs make up less than 10% or more than 50% of the population. Table 1: Street access and group shares in the whole sample caste population share excluded share excluded number of households who group share 1996 2006 gained access from 1996 to 2006 SC 20.0% 65.0% 44.5% 2,076 ST 6.5% 31.4% 7.2% 779 OBC 50.8% 5.5% 4.0% 520 OC 22.4% 1.6% 1.3% 62 12

4.3 SC Pradhans and Reservations The study focuses on SC quotas, implemented in 24 of the 89 villages surveyed. SC reservation is observed only once in each village. Villages where the pradhan seats is reserved for a SC in 1996, which correspond to the first election round, have a higher proportion of SC households and are, if anything, poorer than unreserved villages (results of mean comparison tests are in appendix table 9). This is consistent with the selection of reserved villages through the share of SC households in some states. However, for the third election round of 2006, the repartition of reserved seats appears independent of villages wealth or power. As for ST reservation, the sample is too small to allow a proper study: only three villages display variations both in ST street access declarations and reservation status. 17 Out of the 178 pradhan terms considered in this article (two election terms in 89 villages), a total of 43 pradhans belong to a Scheduled Caste. In addition to the 24 SC pradhans elected in a reserved seat, 19 SC pradhans won an open election. The latter figure is surprisingly high. Political reservations were implemented precisely because of SC under-representation. Villages where an SC is elected in the absence of an SC quota are likely to have specific features. What makes villages where an SC pradhan is elected outside a caste quota so special? I check the share of the village population which is SC. I also check whether relations between castes in the village are particularly good; or whether these villages combine numerous SC households and bad relations between castes (leading to an SC collective action). The share of SC households in the village population, and the share of excluded households among SCs are known. Table 2 displays mean comparison tests on these variables. The first sample encompasses villages where an SC pradhan has been elected outside an SC quota, the second sample encompasses villages without SC quotas. SCs appear to be more numerous in villages where SC pradhans were elected. However they apparently remain a minority. 18 More importantly, there seems to be no difference between the two samples on street exclusions. This is essential, because street exclusion is the variable of interest in the study. A last characteristic I test is the part played by gender quotas, because gender quotas prevent the traditional elite from running for office. The prevalence of gender quotas differs significantly 17 At least one ST is present in 47 of the 89 villages in the study, but only 5 villages have experienced reservation of the pradhan seat to an ST over one of the two terms, and in two of these villages no ST ever declared any street exclusion. Four other villages have had an ST reservation in both terms, but the impact is not time variant in that case, so it would be captured by household-level fixed effects. 18 A panchayat is often constituted of more than one village, and it may be the case that the SC share in the GP differs from the share in the sampled village. However, I would not expect SCs to be the vast majority in many GPs because of traditional settlement patterns. 13

between samples. Table 2: Mean comparison tests, villages with versus without an SC pradhan (the sample is restricted to the 166 terms without an SC quota) SC Pradhan Difference Sign No SC Pradhan P> z SC population share 0.292 > 0.183 *** SC rate of street exclusion 0.468 < 0.514 women reservation 0.714 > 0.393 *** Looking more closely at the overlap between SC pradhans, SC quotas and gender quotas, out of 19 terms when an SC pradhan was elected outside an SC quota, 15 were elected on a gender quota, and 3 were elected after an SC quota in the previous term. Only one SC pradhan was elected outside any reservation treatment. Gender quotas change the political landscape in villages: Table 3 shows that pradhans caste distribution in women s reserved seats converges toward the population s caste distribution. In Table 3, the statistics are computed for the 139 terms where neither an SC, nor an ST, nor an OBC seat reservation was imposed. Outside gender quotas, OCs (the high castes) win twice as many pradhan seats as their share in the population, while SCs and STs are poorly represented. The contrast with caste distribution for women s reserved seats is striking. Caste distribution for women s reserved seats mirrors the population caste composition. Some women may be elected only because their husbands could not run for office. 19 But another parallel dynamic is at play: seat reservation for women disqualifies the traditional male, high-caste elite from running for office. It calls for reconsidering who is legitimate to run for office. Low-caste women, in particular SCs, appear to have taken advantage of this situation. This mirrors Buch s affirmation, after field research on panchayats in Madhya Pradesh, Rajasthan and Uttar Pradesh that Women who entered new panchayat showed a wider participation across social and economic class/caste. Their occupation, income levels and household assets indicated presence of hitherto unrepresented sections. (Buch, 2013, p. 52). I focus the empirical analysis on the impact of SC quotas. I follow this course to examine quotas as a political tool. Moreover, it allows me to exploit an exogenous source of variation in the leader s identity. However, in almost all specifications, I control for SC elections outside quotas. I also further explore the impact of gender quotas on SC elections and SC street access in Section 5.2. 19 This idea is widely spread, in particular among quota opponents, but some evidence shows that the election of women on reserved seats does change things for women, be it through a change in priorities or in attitudes (Chattopadhyay and Duflo, 2004; Beaman et al., 2009; Iyer et al., 2011). 14

Table 3: Caste composition of elected pradhans and population for the terms without any caste-based quota. unreserved seat reserved woman population share SC 8.3 22.4 19.3 ST 0 4.5 6.6 OBC 45.8 46.3 50.8 OC 45.8 26.8 22.7 total 100 100 100 5 Results 5.1 SC Reservations and SC Street Access: Baseline Results Table 4: Baseline results: street exclusions grounded in caste identity (1) (2) (3) (4) Cross Section Cross Section Panel Panel caste exclu caste exclu caste exclu caste exclu SC*res_SC -0.0852*** -0.126*** -0.0995*** -0.137*** (0.0325) (0.0472) (0.0183) (0.0253) SC*pradh_SC 0.0471 0.0453** (0.0399) (0.0225) SC 0.0877*** 0.0806*** (0.0168) (0.0178) Observations 40,047 40,047 79,972 79,972 village fixed effects X X O O household controls X X O O time*village fixed effects O O X X time*caste fixed effects O O X X household fixed effects O O X X R-squared 0.340 0.340 0.186 0.187 Robust standard errors in parentheses. Errors are clustered by village caste group in columns 1 and 2. Errors are clustered two ways for time*village*caste and household levels in columns 4 and 5. Controls included (columns 1 and 2): household caste category, household exclusion from streets in 1996, household head s sex, education level (above or below primary), age, income and occupation with a dummy for whether the primary income source comes from agriculture. SC reservation and SC pradhan main effects are absorbed by village fixed effects. *** p<0.01, ** p<0.05, * p<0.1. 15

Reservation raises SCs access to streets according to the results displayed in Table 4. In 2006, the effect offsets SCs initial disadvantage (column 1, Table 4). Moving to the panel specification, the effect magnitude is the same. For SC households, having an SC pradhan makes street exclusions 8.5% less likely (column 3, Table 4). This is almost half of the observed 20% point decrease in SC street exclusion declarations. However, the estimates in columns 1 and 3 leave 19 villages where a SC pradhan has been elected out of a reserved seat in the control group. SC pradhan elections outside reservation are controlled for in columns 2 and 4 (and in the rest of the article). The variable has an unexpected significantly positive sign: SC pradhans elected outside SC quotas seem to worsen SCs situation. I investigate this effect in Section 5.2. 16

5.2 SC reservation, SC Pradhans and Gender reservation Table 5: Caste-based street exclusions, with SC reservation, SC pradhans, or gender reservation, and by respendent s gender (1) (2) (3) caste exclu caste exclu caste exclu SC*res_SC -0.176*** -0.127*** -0.130*** (0.0396) (0.0419) (0.0256) SC*pradh_SC -0.00521-0.0793 0.0405* (0.0351) (0.0489) (0.0230) SC*res_woman 0.0482** 0.0192 (0.0240) (0.0276) SC*pradh_SC*res_woman 0.149*** (0.0527) -0.0708** (0.0333) 0.0578* (0.0314) Observations 79,972 79,972 79,972 time*village fix effects X X X time*caste fix effects X X X household fix effects X X X R-squared 0.134 0.137 0.187 Robust standard errors in parentheses are clustered two ways by time*village*caste and household levels. *** p<0.01, ** p<0.05, * p<0.1. In this section, I investigate further the detrimental effect of SC pradhans elected outside an SC reservation. Please recall from Section 4.3 that out of 19 SC pradhans elected outside a caste quota, 15 were elected on a gender quota. In Table 5, I decompose the effect of SC pradhans with respect to gender reservation. Female pradhans, elected after a gender quota, seem to worsen SCs situation (column 1). However, if I add an interaction term for SC pradhans elected on a gender quota, the entire positive effect of women reservation comes from SC women (column 2). Lastly, the worsening effect of SC pradhan holds independently of the gender of the respondent, if anything, it is amplified for the 3,326 female SC respondents (column 3). 17

Why should SC female pradhans have a different impact than male SC pradhans? Gender and caste may interact in unexpected ways. Caste-based and gender-based preferences may not converge. Clots-Figueras (2011) studies a situation which mirrors the one studied in this article: she shows that the gender effect is different for lowcaste women (elected on SC ST reserved seats in state elections) than for higher-caste women (elected on unreserved seats). Her interpretation relies on the idea that gender and class effects can be contradictory. Moreover, female pradhans, even if they are low caste, may prioritize gender-based requests over caste-based requests. I cannot compare the effect of SC female pradhans elected with and without a gender quota outside a caste quota. But if anything, being elected thanks to a gender quota should increase the likelihood that the woman pradhan will prioritize gender-based requests. 20 Another possible channel appears in Bardhan et al. (2010). They find that SC female pradhans target SC beneficiaries more than non-sc female pradhans do, but still target them less than either general pradhans or SC pradhans. Their interpretation relies on a lack of socialization of the newly elected female pradhan who does not know how to implement traditional patronage. Lastly, descriptive statistics in Table 3 show that elections with gender quotas are more contested. SC women elected on reserved seats are elected after a contest against other women, who may have been high-caste women. This competition between castes is not observed during elections for an SC reserved seat. High castes have to accept that an SC is to be pradhan when there is an SC quota. Please recall Table 3: outside reserved seats, high castes win nearly half of the elections, although they represent a fifth of the electorate, while on women s reserved seats, high castes win only a quarter of all elections. Women s reservation may favor the emergence of new political personalities (Buch, 2013) or alliances between lower castes to increase their access to power. The interaction of caste and gender effects deserves further investigation. However, this investigation is left for future work, with more complete data. The core of this article therefore focuses on SC elections on SC reserved seats, although I always control for the election of an SC pradhan outside SC quotas. 20 In the lab, insisting on Asian women s ethnicity versus their gender identity affects their math and verbal performances in opposite directions, each consistent with stereotypes (Shih et al., 2006). In the field, one can imagine that being elected on a seat made accessible due to one s gender may strengthen gender identity. Additionally, roads are a gendered concern in Chattopadhyay and Duflo (2004), although which gender is more preoccupied by roads varies in the two states studied. 18

5.3 Reservations and Street Access: Robustness I check the results in Table 4 with respect to several concerns: firstly, is the effect robust to the sample definition? Reassuringly, in Table 6, both the significance and the magnitude of the coefficient of interest stay constant across alternative samples. I consider two alternative samples. In the first sample, I omit villages with a past reservation, in the second sample I add non-hindi-belt states. Secondly, is the effect of reservation independent from how long reservation has been in place? Reservation has a significant effect on SCs access to streets only after the first year (7 column 1). The impact is bigger for 2-year-long reservations than for 3- or 4-year-long reservations. However, only two panchayats have been in place for two years. Because of the small sample size, and because effects in time are not contradictory, I focus on the average effect of SC quotas, independent from their age, in the rest of this article. Thirdly, does the effect last after the reservation has ended? To answer this question, I control for reservation in the previous panchayat term. The effect of reservation does not last after the end of the reserved term (Table 7, column 2). This observation points to a crucial part played by SC pradhans while in office. This observation can also be worrisome: it goes against the argument that reservation should be used as a transitory tool. However, the results should once again be interpreted with caution. Beaman et al. (2009) underline that gender quotas are more effective after a second reserved term than after only one reserved term. Since the law on reservation is recent, villages had been reserved at most once when the data was collected. Fourthly, do villages with SC reservations display a general improvement of relations between groups? I check this for relations between castes, because SC pradhans may care about other low castes in the village (although for OBCs the effect may be blurred by the fact that OBCs are the dominant high caste in some villages). Reservations for SCs do not affect either OBCs or STs street access (Table 7, column 3). Conversely, I find no impact of seat reservation for other low castes on SCs access to streets (column 5, Table 7). Moreover, if reservations for OBCs had an impact on SCs access to streets, it would be detrimental. 21 I focus on reservations for OBCs because of sample restrictions on reservations for STs. 21 This can be interpreted directly, if OBCs are willing to insist on their traditionally higher-caste status than SCs and enforce untouchability practices more once elected. This can also be interpreted indirectly as having the pradhan position reserved for an OBC prevents any SC from being elected and elected SCs improve SCs access to streets on average (even if the impact is 19

Lastly, I check whether Muslims street exclusions, grounded in their religious identity, react to SC quotas. It should not be the case since Muslims exclusions are grounded in a Hindu/Muslim divide, while reservations for low castes aim at reducing the low/high-caste gap. Reassuringly, column 5 in Table 7 confirms that the logic at play with SC reservations is only a caste logic. The percentage of Muslim households excluded from streets dropped from 27% to 16% over the period. But the increase in Muslim households access to streets is independent from SC quotas. In the end, SC households access to streets appears to be improved by reservation of pradhans seats for SCs. The effect is observed only within the SC group, other castes are not concerned. The effect increases over time, but does not last after the end of the reserved term. led by pradhans on SC reserved seats). In line with the latter interpretation, if the interaction term for SC seat reservation is omitted, the OBC reservation remains at 0.05 but becomes significant at 10% (result not included). 20