Politics as Usual? Local Democracy and Public Resource Allocation in South India

Politics as Usual? Local Democracy and Public Resource Allocation in South India Timothy Besley LSE and CIFAR Rohini Pande Harvard University Revised September 2007 Vijayendra Rao World Bank Abstract This paper uses data on elected village councils in South India to examine the politics of public resource allocation. We stress two facets of the political process access to political authority, and the use of political power. We nd compelling evidence that the pattern of policy re ects politicians self-interest. Further, the extent and type of political opportunism in resource allocation is responsive to the design of political institutions. Thus, local democracy in India displays all the hallmarks of politics as usual. We thank Lupin Rahman, Radu Ban, Siddharth Sharma and Jillian Waid for research assistance, and IMRB sta for conducting the survey. Useful comments have been received from seminar audiences at Oxford and at a meeting of the CIFAR Institutions, Organizations and Growth program. We are grateful to the World Bank s Research Committee and the South Asia Rural Development Unit for nancial support. The opinions in the paper are those of the authors and do not necessarily re ect the points of view of the World Bank or its member countries. 1

1 Introduction Decentralization as a means of empowerment has become the mantra of e orts to improve governance in the developing world. 1 Yet, we are still learning about how decentralization works in practice. While the basic principle underlying decentralization is simple, a whole host of di erent institutional details underlie any e ort to increase local authority in allocating public resources. It is very likely that the precise choice of these institutions have substantial implications for the way public resource allocation works on the ground. This paper examines three facets of the local democratic process in South India. First, who gains access to political authority. Second, how do they use that authority to a ect public resource targeting across villages and across households within villages. And nally, how do speci c political institutions a ect this political authority. We show that political resource allocation works in a predictable way, once the incentives in the system are properly understood. Moreover, the extent and type of political opportunism in resource allocation is responsive to the design of political institutions Thus local democracy in India bears all the hallmarks of politics as usual. Our context is rural India where a 1993 constitutional amendment mandated the introduction of village-level self government, Gram Panchayats (GP). We exploit within- and cross- state variation in the way decentralization was implemented to study the impact of speci c institutions of decentralization on public resource allocation. The studied institutions are the political geography of GPs, and the form of selection of the GP head, the Pradhan. Here, we focus on whether the Pradhan is indirectly or directly elected and whether the Pradhan post is reserved for traditionally disadvantaged groups. We provide a simple theoretical model for understanding public resource allocation which is based on the idea that holding political o ce provides access to policy discretion which can be used to further the interests of particular groups. It is this discretionary political authority which creates a return to holding political o ce. First, there is a private return to politics in the form of greater access to targeted resources. Second, there is a village return which re ects the structure of political power within the GP. We show that the characteristics of those who access public 1 For broad discussions of the issues, see Bardhan (2002), Crook and Manor (1998), Triesman (2007). 2

o ce are determined at the individual-level, while the Pradhan s are determined by village characteristics. Political reservation, by changing who can get into o ce, in uences both. 2 A theme in our analysis is selection into political o ce. There are a variety of dimensions along which political selection could be important selection could be based on honesty or integrity or ethnicity or wealth. in which the spoils of public o ce are allocated. This may alter the way It may also a ect the quality of policy making. The Downsian model of political economy which dominated thinking in economics for over a generation put very little weight on this aspect of political life. However, the more recent citizen-candidate approach of Besley and Coate (1997) and Osborne and Slivinski (1996) puts this at center-stage in understanding political competition. This approach makes sense in a world in which the types of politicians has an impact on policy outcomes. 3 The main institutional constraints on selection that we study are political reservation and whether the Pradhan is directly or indirectly elected. Each state is required to reserve a certain fraction of Pradhan posts for women and low castes, and the precise allocation of reserved Pradhan posts within a state is random (see Chattopadhyay and Du o, 2004). Two of our sample states, Kerala and Karnataka, chose to enforce indirect elections to the post of Pradhan. The second main theme is the conduct of distributive politics the problem of allocating public resources across competing ends. There are two main issues. First, how far do politicians exercise self-interest in their allocation of public bene ts. Second, how does the geographical pattern of representation matter. Both issues have been discussed extensively in the political economy literature. 4 The question of whether politics yields private bene ts is at the heart of debates about political corruption. The usual concern is with politicians self-dealing. To gauge this in an Indian context, we look at an important targeted public transfer program whose administration has been transferred to the village level the below poverty line card (or BPL card) system. Instituted in 1997, this program entitles 2 Chattopadhyay and Du o (2004a) and Ban and Rao (2007) examine the impact of women s reservation on spending patterns in West Bengal and the South Indian states respectively. Chattopadhyay and Du o (2004b) exploits the practice of political reservation in favor of low castes to examine whether resource allocation at the village level is sensitive to low caste representation and Pradhan residence. Pande (2003) looks at the impact of state-wide reservation for Scheduled castes and Scheduled tribes on patterns of policy making. 3 See also Lee et al (2004) and Besley (2005) for general discussion. 4 See, for example, Weingast, Shepsle and Johnson (1981) and Baqir (2002). 3

households to buy food below market prices. The government uses the Panchayat administration to identify bene ciary households based on need. Studying the allocation of such cards provides a clean insight into the working of decentralized government and allows us to look at household level access to transfers. The issue of geographic targeting of public programs has spawned extensive empirical and theoretical literatures. Weingast, Shepsle and Johnsen (1981) emphasize the possibility of a cooperative outcome (universalism) in which there is a relatively equal allocation of public resources. Others, such as Riker (1962) and Baron (1991), emphasize the value of agenda control. Even if agenda control is important, there is an issue of how it is used for narrow self-interest or to serve broader ends. In examining the consequences of political geography, we exploit the fact that each state sets a separate minimum population size which would constitute a GP, and this led to within and cross state variation in the number of villages per GP. We exploit di erences in the number of villages per GP to explore the problem of distributive politics across villages. Speci cally, we examine whether there is a resource advantage to those who reside in the Pradhan s village. 5 More generally, our paper contributes to an emerging body of work on decentralization in the developing world. 6 Faguet (2004) examines whether decentralization led to improved targeting in Bolivia. In the Indian context, Bardhan and Mookherjee (2003) examine the role of elected village councils in a ecting land reform. Chattopadhyay and Du o (2004) examine public good allocation in West Bengal villages. Finally, Foster and Rosenzweig (2001) examine how decentralization interacts with land ownership patterns to a ect public good outcomes. The remainder of the paper is organized as follows. In Section two, we describe the institutional setting and in Section three we discuss some theoretical issues. Section four discusses the data and empirical analysis while Section ve concludes. 2 Background In India, a 1993 constitutional amendment made a three-tier elected local government obligatory throughout the country. Our focus is on the lowest tier of local 5 A number of contributions have tried to test the empirical relevance of universalism using U.S. local government data. The ndings are mixed see, for example, Baqir (2002), Bradbury and Stephenson (2003) and MacDonald and Sass (2005). 6 Besley, Pande, Rahman and Rao [2004a] and Besley, Pande, Rao. [2005] look at various aspects of resource allocation in Pachayats using the data from this study. 4

self-government. This is a popularly elected village council the Gram Panchayat (GP hereafter). The Panchayat Act of every Indian states mandates the population criteria to be followed in that state. Our empirical study focusses on the four South Indian states of Andhra Pradesh, Karnataka, Kerala and Tamil Nadu. In Kerala, where villages are large, each village is mandated as a separate GP (and the average population per GP is over 20,000). All other states in our sample use a population criterion to demarcate GPs. 7 Each GP consists of up to twenty wards. 8 Elections are at the ward-level, and the elected ward members constitute the GP council. The head of this council is the Pradhan. In two of our sample states (Andhra Pradesh and Tamil Nadu) the Pradhan is directly elected, while in Karnataka and Kerala he/she is nominated from the pool of elected ward members. The 73rd constitutional amendment mandated political reservation of a certain fraction of the Pradhan positions in a state in favor of the historically disadvantaged scheduled castes and tribes (SC/ST) and women. Two of our sample states also extended Pradhan reservation to other backward castes. Political reservation for a group implies that only individuals belonging to that group can stand for election to the reserved position. The 73rd constitutional amendment also requires that the extent of caste reservation in a state re ect the SC/ST population share, and that one-third of Pradhan positions in every state be reserved for women. No GP can be reserved for the same group in two consecutive elections. It also required the constitution of a village-level body consisting of persons registered in the electoral rolls of a GP, the Gram Sabha. This is intended to be a supervisory body which meets multiple times in a year to audit and regulate the functioning of the GP. A GP has responsibilities of civic administration with limited independent taxation powers. 9 While Panchayat legislation requires that the Pradhan decide the choice of bene ciaries and public good allocation in consultation with villagers and ward members, nal decision-making powers in a GP remain vested with the Pradhan. Though the precise scope of GP policy activism varies across states, the GP 7 The average population per GP is 1650 in Andhra Pradesh, 6500 in Karnataka and 4000 in Tamil Nadu. 8 In our sample states the population per ward varies between 300 and 800. 9 On average, roughly 10 percent of a GP s total revenue come from own revenues with the remainder consisting of transfers from higher levels of government. 5

is typically responsible for bene ciary selection for government welfare schemes and the construction and maintenance of village public goods such as street-lights, roads and drains. Since 1997 the Indian government has used a targeted public food distribution system which provides Below Poverty Line (BPL) card holders subsidized food while charging a near market price for the others. In 2000-01, for our sample states, the annual income gain from BPL card was roughly 5% of an agricultural labor household s annual expenditure. 10 In addition to subsidized food, most GP administered welfare schemes, e.g. employment and housing schemes, restrict eligibility to BPL households, and in some cases SC/ST households. To identify BPL households in a village the GP, together with state government o cials, conducts a census every 2-3 years. GP politicians bear substantial responsibility for conducting this survey. 11 They choose the village surveyors, and prepare a preliminary BPL list of recipients. This list is supposed to be nalized at a Gram Sabha meeting. However, in reality GP o cials enjoy substantial discretion in selecting BPL households. An important way of achieving this is to limit the powers of Gram Sabha only 76% of the villages we surveyed had held a Gram Sabha in the past year, and only 20% of households report ever having attended a Gram Sabha. Moreover, bene ciary selection was reported as discussed in only 22% of Gram Sabha meetings ( Besley, Pande and Rao (2005)). Our survey included politician surveys only 9% of the 540 surveyed politicians stated that the Gram Sabha decided the nal BPL list; in contrast, 87% believed that this power lay with a Panchayat o cial. 3 The Model In this section, we provide a simple theoretical structure to mirror the institutional context that we are studying. This yields predictions about how we expect politics 10 Under the public food distribution system 20 kg of food grains per month is provided at 50% economic cost to BPL households. Planning Commission, (2005) calculations suggest that the e ective annual income gain was Rs. 1025 in Andhra Pradesh, Rs. 520 in Karnataka, Rs. 1414 in Kerala and Rs. 809 in Tamil Nadu. We use data from the 1999 National Sample Survey to compute the implied gain for a agricultural household. 11 There is an GP level quota on number of BPL cards. The central government uses the Planning Commission s poverty estimates(based on the National Sample Survey) to release total foodgrains. Based on this, the state government allocates district-wise BPL card quota Within a district, a BPL quota is determined at the GP level. 6

to a ect public resource allocation based on rst principles assuming that politicians are self-interested. Political authority in our villages is in the hands of GP politicians. The GP is headed by a chief executive, the Pradhan, whom we model as an agenda setter. We focus on how the selection procedure for politicians, and institutional constraints on politicians discretionary policy powers, a ects public resource allocation. Our analysis looks at household level transfers within a village and public good allocation across villages within a GP. 3.1 Framework Assume there are V villages in a GP labelled v = 1; :::; V. Let n v be the number of households in village v whom we index i = 1; :::; n v. Each individual i in village v is either rich, R, or poor P. Let (i) 2 fp; Rg denote the type of individual i, and v the fraction of type P citizens in village v. While a gross simpli cation, studying two types su ces to make the main theoretical points needed to underpin the empirical analysis. The GP allocates a budget of size B across the villages. Denote village public expenditures by G v. Thus: VX G v = B: v=1 Resource allocation is controlled by a village council with a set of representatives one for each village whose identities are denoted by a vector = f 1 ; 2 ; :::; V g where v = i denotes the identity of the individual selected to represent the village in the council and ( v ) denotes their type. The council also controls household access to transfers from the state BPL cards in our case. Each household can receive a transfer i 2 f0; T g at a cost c ( i ; (i) ; v ). Costs should be interpreted quite widely to include e orts of those who wish to receive transfers as well of those who are charged with giving them out. We allow the cost of gaining this transfer to depend on who is elected to represent the village. 12 This could be because elected representatives exercise self-interest or a group preference targeting more resources towards certain groups. The preferences of citizen i are 12 In principle, this could depend upon the whole vector of council representatives rather than just the identity of the politician in the individual s village. However, this would complicate the political game laid out below considerably. 7

quasi-linear in private goods and denoted by: V ( (i) ; G v ) + i c ( i ; (i) ; v ) E ective targeting of private transfers towards the poor in village v requires: c (T; R; v ) > T > c (T; P; v ) : This says that only poor households nd it worthwhile to claim transfers. Let v ( (i) ; v ) = max f c (; (i) ; v )g 2f0;T g be the utility from household transfers as a function of the individual s type and the identity of the village representative. 3.2 The Policy Process Within each GP, one elected representative is the Pradhan and possesses agenda setting power. There are two methods of Pradhan selection discussed in greater detail below direct election from the GP as a whole and election from among the elected representatives elected from each village (indirect election). Following the citizen-candidate approach to elections due to Besley and Coate (1997) and Osborne and Slivinski (1996), we model elections in three stages. At the rst stage, any villager can stand as a candidate to represent the village. At stage two, the citizens vote among the self-declared candidates and at the third stage, policies are chosen by the council. We look for an equilibrium of this game in which voting is strategic at stage two (with voting decisions forming a Nash equilibrium) and entry decisions also forming a Nash equilibrium. We suppose that if nobody is elected then v =? and c (T; P;?) > T. There is a small cost of > 0 of standing as a candidate. We study the model in reverse order, beginning with policy determination. A.Policy Determination Suppose that there is a vector of elected representatives in the council including the Pradhan. Let v = denote the Pradhan s village. Thus, the outcome of the elected process is denoted by f; g. Without loss of generality, let the Pradhan s village be village 1. The policy process has two parts which can be studied separably. 8

The allocation of household transfers depends solely on v as the budget for this is not set by the GP. This was studied above. villages depends upon. It is this that we study now. The allocation of expenditures across The Pradhan will propose an allocation to other council members. be agreed to by a majority of council members in order to be accepted. 13 This must If the village council cannot agree to a public good allocation, then the status quo is that each district gets at least G and the Pradhan s village gets B G. Then, each village representative faces a status quo utility level of V ( ( v ) ; G) + v ( ( v ) ; v ) which can depend on the type of the politician in the village. This de nes a simple bargaining game between the Pradhan and other elected representatives. The (V 1) Pradhan knows that he can o er G to (V 1) =2 of the villages and get T G 2 for himself. The remaining villages get nothing which exceeds what he would get in the status quo. While this is simple and extreme, it is indicative of what will happen in a wide variety of circumstances where there is a xed agenda power. 14 Thus, resource allocation n the agenda setting model has: 8 (V 1) >< B G if v = 2 G v = G if v is in the winning coalition >: 0 otherwise. The key observation from the agenda setter model of a GP is the resource advantage for the Pradhan s village. This de nes the allocation as a function f; g. Observe this model does not predict any link between the Pradhan s type and the allocation. It also does not predict which villages will be in the minimum coalition. return to this below. We will Let v be an indicator variable that is equal to one in the Pradhan s village, i.e. if v = and zero otherwise. Also, let v be the probability that village v is the winning coalition if the village is not the Pradhan s village. Putting this together with the allocation of household transfers above, we can determine the overall allocation of transfers. The expected utility of household i in village v as a 13 The classic analysis of agenda setting is by Romer and Rosenthal (1978). Riker (1962) rst proposed the importance of minimum winning coalitions in legislative bargaining. 14 Things are more complex in models such as Baron (1992) where agenda setting power varies randomly over time. 9

function of f v ; v ; v g as: (V 1)! iv ( v ; v ; v ) = v V (i) ; B G 2 + (1 v ) [ v V ( (i) ; G) + (1 v ) V ( (i) ; 0)] + v ( (i) ; v ) : Observe that the payo of household i in village v depends solely on the type of the elected representative in their own village, i:e: v. B.Voting To study the election outcome, we make two natural assumptions: Assumption 1: (Political advantage) c (T; (i) ; i) < c (T; (i) ; k) for k 6= i: That is, relative to non-politicians, politicians nd it easier to access public transfers. This both leads us to expect politicians to claim public transfers more often than non-politicians and implies that there is rent to holding public o ce. We also make: Assumption 2: (Type advantage) c (T; (i) ; `) < c (T; (k) ; `) if (i) = (`) 6= (k). This says that households nd it easier to gain a transfer when a politician of their own type is in o ce and gives a reason for them to prefer politicians of their own type to hold o ce. 15 Suppose now that in village v a set of candidates C v have put themselves forward for o ce. Then all voters (weakly) prefer a candidate of their own type if there is one available and are indi erent between candidates in terms of the outcome with respect to G v which depends on v and v. There are typically many Nash equilibria in the voting game since voters are indi erent between voting for many candidates. We will pick the Nash equilibrium in which voters coordinate on one candidate of the majority preferred type. Thus, if v 1 < 1 2 2 and a poor (rich) candidate stands, then a poor (rich) candidate is elected. 15 There are other simple extensions of the model which deliver this. For example, politicians could have some in uence over which kinds of public projects G v is spent on. This would only strengthen our results. 10

C.Entry of Candidates We now consider an individual s decision to stand as candidate for public o ce, and the impact of reservation on this. A candidate will stand for o ce only if there is a bene t of doing so in terms of private goods. If T > c (T; (i) ; i), then there is no bene t to standing for o ce. There are two cases to consider. Case 1: T > c (T; (i) ; i) for (i) = R. In this case, there is no incentive for a rich candidate to run for o ce even if the type R s are a majority. Then, political competition will be among the poor candidates. Thus, we have: Proposition 1 In case 1, in any Nash equilibrium of the citizen-candidate entry game, all candidates are of type P. The equilibrium number of candidates is strictly positive for small enough and is given by: c (T; P; k) c (T; (i) ; i) n = int 1: In this world, candidate entry serves purely as a rent dissipation device su - cient entry by candidates reduces the private bene ts to zero. Note, however, that who holds political o ce does not a ect targeting since type P candidates would have claimed a transfer for their household anyway. In this equilibrium, reservation for group P will also not alter the outcome as they have su cient incentive to run for o ce. Case 2: c (T; (i) ; i) > T > c (T; R; k) for k 6= i and (i) = R: In this case, a rich candidate who holds public o ce will claim a transfer that he would not have claimed otherwise. This gives an incentive for a type R candidate to run for o ce. However, he or she will only have a chance of being elected against 11

a type P candidate if type R s are in a majority. If type P s are in a majority, then the outcome is as in Proposition 1. However, if type R s are in a majority, we have: Proposition 2 In case 2, in any Nash equilibrium of the citizen-candidate entry game and v < 1, all candidates are of type R. The equilibrium number of candidates is strictly positive for small enough and is given 2 by: T c (T; (i) ; i) n = int This case predicts two things. Politicians are an economic elite. However, they are motivated to run for o ce by the spoils that they obtain once they are there. This, we should expect to observe that households in which politicians live are more likely to receive a BPL card all else equal. 1: In this case, we should also expect to see an impact of reservation for type P s on transfers and on who holds o ce. The example studied here on candidate entry views the only bene t from holding public o ce as being through changing the costs of obtaining a BPL card which is patently simplistic. However, even in a richer model with a wider array of motives to stand for o ce, we would still expect the e ects that we have identi ed reduced cost of access to BPL cards by politicians and the e ect of reservation on policy outcomes to be at work on the margin. In this spirit, we will test for the impact of holding o ce on household access to BPL cards below. Proposition 2 is a case where the economic elite are in o ce based on their numerical superiority. However, there are other possible reasons why this could be the case. One is to suppose that entry costs into political o ce by the poor group are prohibitive so that only rich candidates stand. It may also be that the rich group is able to nd some other means of monopolizing the candidate entry process. In this case, Proposition 2 will be valid more widely, i.e. even in the case where v > 1=2. 3.2.1 The Pradhan s Village Finally, we consider determination of v and v. The model as stated is largely silent on this. Moreover in the data, we only observe which village has the Pradhan and not the structure of any winning coalition. That said, it is useful that, in the framework that we have laid out, that the determination of the Pradhan s village can be studied separably from determining which type (rich or poor) will hold o ce. 12

In practice, there are two institutional arrangements direct election of the Pradhan and election from among the GP members. As a benchmark, we consider a simple model where the Pradhan is chosen from among a set of candidates, each one representing the households of each village. In the case of direct election, then it is clear that the largest village in the GP would have an advantage. In the case of indirect election, it is the number of wards that should matter with the village with the most wards having the Pradhan s o ce being allocated to it. Since the number of wards is essentially determined by population size, it should not make much di erence which institutional arrangement is used. The bottom line is that this simple benchmark should lead to larger villages having an electoral advantage. 16 This simple model does, however, ignore the possibility of coalition formation during the electoral process. A candidate in one village may withdraw from the race for Pradhan and deliver the votes from his village to another candidate in exchange for belonging to a winning coalition ex post. For example with three villages of equal size, a candidate from one village could drop out with a coalition of two thirds of the voters forming behind a remaining candidate. This would be credible if the winning candidate could reward the village whose candidate dropped out. coalition proof equilibrium would then be one where there is no candidate who could drop out of the race and bene t in this way. Following this logic, we should expect each Pradhan to assemble a minimal winning coalition in which he gets (just over) half the support of either the voters or the ward members in a GP. There are typically many possible winning coalitions that could form for any given allocation of population across villages. A For example, in the case of three villages with a third of the population each, there are six possible winning coalitions each containing two thirds of the population. Each village is the Pradhan s village in two out of these six coalitions. But there is no obvious reason to expect any one of these coalitions to prevail in practice. In order to remain agnostic about which coalition will form, we choose an ex ante measure of the each village s power by computing the fraction of winning coalitions (i.e. with more than half the population) formed from among all villages in a GP in which that village is decisive to maintaining a coalition containing 50% of the GP population.a coalition with more than 50% of the GP population is assumed to be winning with the Pradhan being 16 If there were a citizen-candidate game with candidates running for the job of Pradhan, then with costly entry (although with a small enough cost) there is an equilibrium where a candidate from the largest village is elected unopposed. 13

chosen randomly from among the coalition partners. In an ex ante sense, we would expect villages with a larger power score of this kind to have a greater chance of being the Pradhan s village ex post. A village is more powerful if there are more coalitions in which is decisive. On this basis, any village that has more than 50% of the population in a GP gets a power score of one. In a one or two village GP only one village is powerful. The interesting cases arise for GPs with more than two villages in which case power is a non-linear function of the vector of village populations in a GP. We conjecture that the power variable is a determinant of v and v, and will explore empirically whether a village s power score predicts whether it will beome the Pradhan s village. We can also test whether its impact di ers between direct and indirect election states, and also whether power independently in uences nal resource allocation (independently of the pattern of political control). 4 Data Our analysis uses household survey and village meeting data which we collected between September and November 2002. Our sample includes over 500 villages in the four southern states of India Andhra Pradesh, Karnataka, Kerala and Tamil Nadu. 17 The sample villages are distributed across nine boundary districts in these states. Within each district we sampled three blocks (the administrative unit below the district), and within each block we randomly sampled six GPs. In GPs with less than four villages, we sampled all villages; otherwise, we sampled the Pradhan s village and two randomly selected villages. 18 In every sampled village we conducted an audit of public good facilities, a village meeting at which we obtained information on village demographics and public good provision, and a household survey with an elected Panchayat o cial. If the Pradhan lived in the village he/she was interviewed, otherwise a randomly selected village councillor (also called ward member). In a random sub-sample of three GPs per block (259 villages) we conducted household interviews. We administered surveys to twenty households in each village. Household selection was random, and we alternated between male and female respondents. In every village, we required that 17 At least one year had lapsed since the last GP election in every surveyed state. 18 In Kerala to account for the higher GP population we sampled 3 GPs per block and 6 wards per GP the Pradhan s ward and ve randomly selected wards. 14

four of the sampled households be SC/ST households. Our nal household sample size is 5180. Table 1 provides descriptive statistics. While the average respondent has over four years of education, politicians are signi cantly more educated. Average land holdings are 2.2 acres; however, among politicians this gure rises to 5.7 acres. Politicians elected from non-reserved seats are signi cantly more landed than politicians elected from reserved seats. Over sixty percent of our respondents are either SC/ST or female, and therefore eligible to contest election in reserved seats. terms of political experience, seven percent of our respondents come from a family where someone held a political position. For politician households this number jumps up to 25%. Finally, twenty-one percent of our households, and a quarter of the politician households possess a BPL card. In These ndings are consistent with the view that, for the most part, politicians are from the political and economic elite and yet have a greater chance of having a BPL card than a randomly selected non-politician household. Respondents are on-the-whole critical of their local politicians less than 40% believe the Pradhan looks after village needs or keeps election promises. Less than 10% of the respondents believe that their village facilities are better than in neighboring villages. Our analysis examines how politics a ects the targeting of BPL cards and public good activism. To measure public good activism we use information on the number of investments in di erent public services which was collected during a village meeting. We collected this data in the following categories: roads, village transport, water, sanitation, irrigation, electricity, education and health. For each category, we construct a count variable denoting how many investments were reported in the village meeting in each category. We then construct a standardized investment measure (z-score) by subtracting the mean for non-pradhan villages and dividing by the corresponding standard deviation. The explanatory variables of interest include political reservation of the Pradhan s position. Roughly 16% of our GPs each are (separately) reserved for women and SC/STs (this translates into 54% of the sampled villages having a reserved Pradhan). Within a block, reservation of the Pradhan position and of wards within a village, is by rotation and is exogenous to village characteristics. 19 19 No political position can be reserved for the same group for two consecutive elections. In Besley, Pande, Rao and Rahman (2004) we show that public good provision in 1991 was statistically 15

For reasons that we outlined above, we are also interested in how a village s relative population share within the GP a ects its political clout, or power. To de ne the power of the village in a GP with n villages we consider all coalitions of size< n with population half the GP population as winning coalitions. The own coalition of village v is the number of winning coalitions that include i and no longer remains winning when i is removed. The power of a village v is the ratio of own coalition v to the total number of winning coalitions in the GP. We assume that a tie means the coalition is winning. The average village in our sample belongs to 39% of the winning coalitions in the GP. We label this variable as power in the ensuing analysis. 5 Empirical Analysis The model identi es policy outcomes with the exercise of political authority. The result is a probability distribution over the village level allocation of public goods fg 1 ; ::; G V g and the targeting of transfers f 1v ; :::; nvvg. Our model assumes exercise of political authority implies: The Pradhan s village will receive a larger share of resources from the Panchayat than the other villages in the GP. Politicians are more likely to have a BPL card. Households are more likely to have a BPL card when a politician of their group is in o ce We next explore these issues. As background, we rst examine the correlates of being a politician and of being the Pradhan s village. We then look at whether the structure of political authority a ects individuals and villages propensity to receive state provided goods. 5.1 Selection A. Holding Political O ce We start by using our household data to ask whether individual characteristics a ect the likelihood that the respondent is an elected politician in our survey. We therefore indistinguishable in GPs with and without a reserved Pradhan. 16

estimate a linear probability model of the form p iv = v + x iv + " iv : (1) where p iv is a dummy variable for whether respondent i is a politician in village v, v is a village xed e ect and x iv is a vector of individual and household characteristics. Thus, the e ect of household and individual characteristics are estimated based on within-village variation. Standard erros are estimated allowing for clustering at the household level. Table 2 reports the results. In column (1) we see that an additional year of education, and owning an additional acre of land, each increase the likelihood that the respondent is a politician by roughly 0.8%. A respondent belonging to a family with a history of political participation is 12% more likely to be a politician. 20 In columns (2) and (3) we separately examine the propensity of being elected to a reserved and unreserved position. In both cases we observe positive selection on education and family political history. 21 However, reserved politicians are not more landed than the average citizen. Such politicians are, however, signi cantly more likely to be female or SC/ST re ecting the fact that these are the two groups which should bene t from political reservation. In columns (4)-(6) we restrict the sample to Pradhan villages, and the dependent variable to whether the respondent is the Pradhan. We observe very similar patterns of selection. These con rm the impression formed in the raw data reported in Table 1 that politicians are from a political and economic elite. weaker for politicians who have obtained reserved positions. 22 This conclusion is somewhat In terms of the theory above, the results point to a political equilibrium of the kind studied in Proposition 2 above where candidates for political o ce are mostly not poor. It will, therefore, be interesting to see whether politicians do use their access to public o ce to award themselves BPL cards. 20 We have estimated these regressions including party a liation variables. A respondent a liated with the party in power in the state is roughly 7% more likely to be a politician. 21 Further disaggregation shows that family political history is positively correlated with selection only for women. The absence of a political history e ect for SC/STs re ects the recent entry of these groups in politics on the back of reservation. 22 In our village meetings we also collected data on the extent of economic and political oligarchy enjoyed by the Pradhan. Reservation signi cantly reduces the likelihood that the Pradhan is an oligarch. 17

B. Selection of Pradhan Village To examine this empirically we estimate the following village-level linear probability model: P vgb = b + 1 X vgb + vgb P vgb is a dummy variable for village v in GP g in block b which is equal to one if the Pradhan lives in that village. We use b to denotes block xed e ects and X vb is a vector of village characteristics. These regressions exclude Kerala since all GPs in this state contain one village and hence are, by de nition, the Pradhan s village. Standard errors are estimated allowing for clustering at the GP level. The results are in Table 3. in competition for the Pradhan s chair. As we have conjectured, larger villages fair well Column (1) shows that a 1% increase in village population increases the probability that the village is the Pradhan s village by 0.24%. In column (2) we see that the e ect of village population does not vary across states with indirect and direct elections. In column (3) we include other measures of a village s political power - whether the village is the GP Headquarter and the number of wards in the village (both variables are positively correlated with population). While both measures are signi cant determinants of being the Pradhan s village population remains a signi cant predictor of Pradhan village. In columns (4)-(6) in Table 3 we investigate the importance of a village s relative population share within a GP. Arguably this is more relevant as a determinant of the villages likelihood of becoming the Pradhan s chair. In a rst past the post electoral system we expect the relative population share of a village within a GP to matter. The simplest way to see this is to estimate the above regression with GP xed e ects. In column (4) we see that, within a GP, a 1% increase in village population increases the likelihood that it is the Pradhan s village by 0.27%. In columns (5) and (6) we consider more structured forms of estimating the impact of a village s population share. In column (5) we see that a 1% increase in the share of GP population living in a village increases its likelihood of being the Pradhan s village by 0.6%. In column (6) we measure a village s population in uence within a GP by its power as de ned above the percentage of winning coalitions in the GP that a village belongs to. This variable positively predicts Pradhan village, and the village own population variable is no longer signi cant. The e ect of power also seems large with a move from a power of one to a power of one third reducing the probability of being the Pradhan s village by roughly 25%. Finally, the e ect of 18

power is similar across states with direct and indirect elections. 23 Overall, these results suggest that political geography is important in the location of the Pradhan s village. At the very least, it will be important to control for these when we investigate whether living in the Pradhan s village yields a bene t in terms of policy activism. 5.2 Policy Choice We now examine how political forces shape policy outcomes both across households within a village and across villages in a GP. A. Targeting Transfers between Households The basic idea of the BPL card program is to help disadvantaged groups. The fact that politician households are wealthier than non-politician households (Tables 1 and 2) ought therefore to imply that politician households are less likely to have a BPL card. To investigate this empirically, let b iv be an indicator variable denoting whether household i in village v has a BPL card. We estimate a linear probability model: b iv = v + x iv + iv (2) where x iv is a vector of household characteristics that are relevant to whether the household is needy. We also include an indicator variable for whether the household has a history of political involvement. The standard errors are estimated allowing for clustering at the village level. All village level characteristics are subsumed in a village xed e ect v so this equation only considers targeting of BPL cards exploiting within village variation in individual and household characteristics. The results are in Column (1) of Table 4. There is some evidence that, as we would expect, BPL cards are targeted towards disadvantaged groups. An SC/ST household is 16% more likely to get a BPL card and a landless household 7% more likely. Households with a more educated respondent are less likely to get a BPL card. Households in which at least one member holds or has previously held political o ce are no more likely to have BPL card. 23 This speci cation is available from the authors. 19

To test the role of political control, column (2) includes as a regressor whether household contains a currently elected GP politician. Consistent with the view that holding public o ce reduces the cost of access to such cards for politicians, we nd that politician households are roughly 8% more likely to have a BPL card. In column (3) we nd that this e ect is essentially limited to unreserved politicians, and there is no signi cant di erence in the propensity of village councillors and Pradhans to have a BPL card. Reserved politicians are more likely to be new entrants to politics, and are politically and economically less advantaged than unreserved politicians. These results are consistent with the political equilibrium being something along the lines of Proposition 2 above with an elite bene ting from control of public resources in unreserved seats. It is, however, interesting to note that political opportunism does, in column (3), appear to be lower when politicians are more educated. Finally, in column (4) we control only for block xed e ects and nd that living in the Pradhan s village does not a ect a household s propensity to receive a BPL card. 24 This is in line with our conjecture in the theory section which allowed us to treat the two di erent aspects of public resource allocation household and village separably. The evidence in Table 4 suggests that while the BPL program successfully targets the relatively disadvantaged households (as measured by SC/ST and landless status), politician households also bene t from this program. The theory suggested that the cost of targeting and hence its e ectiveness might depend on politicians characteristics. In Table 5 we further investigate this by looking at how village and politician characteristics in uence targeting to disadvantaged households. We do so by interacting Pradhan and village characteristics with being either an SC/ST or a landless household in the targeting equation. The rst three columns of Table 5 examine how Pradhan characteristics a ect BPL card allocation. For the average landless household, the likelihood of having a BPL card is lower when the Pradhan has a BPL card. This suggests that there could be a cost to political opportunism with opportunistic politicians targeting landless households less. Landless households appear to bene t from having a more educated Pradhan. This is further evidence that having educated politicians improves household level targeting in this setting. Having a reserved politician 24 Estimating this speci cation as a probit leaves the results unchanged. 20

does not impact a landless household s propensity to get a BPL card. It does, however, signi cantly increase the likelihood that a SC/ST household has a BPL card (column (3)). This con rms the idea that individuals do tend to bene t when there are politicians in o ce whose characteristics are more similar to their own. Finally, in columns (4) we examine Pradhan village e ects. Living in the Pradhan s village leaves a household s propensity to receive a BPL card una ected again con rming the idea that the Pradhan village e ect ought not to be important for this level of targeting. B. Allocation between Villages We now examine resource allocation across villages. Let Y vgk be the standardized measure of GP activism for activity k in village v in GP g. We estimate a regression of the form: Y vgk = b + P vgk + X vgk + " vgk (3) where b are block xed e ects, P vgk is an indicator variable equal to one if the observation is for the Pradhan s village and X vgk are the village control variables used to explain the Pradhan s village above. Standard errors are estimated allowing for clustering at the GP level. The categories of activism that we study are roads, transport, water, education, health, sanitation, electricity, and irrigation. Our standardized measure the construction of which was discussed in the data section above allows us to compare results across subcategories. Finally, following Kling et al (2007), we obtain an overall index by taking the average of equally-weighted standardized components of these activism measures. To estimate the covariance matrix (for both subcategories and the overall index) we use the seemingly unrelated regression (SUR) model. The results are reported in Tables 6a and 6b. Column (1) of Table 6a shows that, as predicted by an agenda setting model, being the Pradhan s village is positively correlated with overall GP activism. However, to conclude that this is an e ect of political control, it is necessary to be sure that the Pradhan village e ect is not just proxying for those characteristics that increase the probability of securing the Pradhan s position observed in Table 3. To gain some reassurance on this, column (2) estimates this regression with GP xed e ects and includes as covariates other determinants of Pradhan village location 21

within the GP. Encouraging to the view that we are picking up the e ect of political control, the Pradhan village e ect remains positive and signi cant. It is also the case that the population variables are jointly signi cant determinants of activism. The Pradhan village e ect, however, remains evident also when, as reported in column (3), we include the village power variable. It is striking that, although this variable strongly predicts which village will be the Pradhan s village, it does not appear to in uence policy outcomes. Columns (4) and (5) provide two further checks on this nding. In column (4) we report results which exclude Kerala without any material impact on the ndings. In column (5) we exclude the Pradhan village variable. The population e ect is signi cant, and it appears that including the Pradhan village e ect is (as expect) partially picking up the electoral in uence of population. Columns (1)-(8) of Table 6b report results for di erent categories of GP activism. These suggest that the Pradhan village e ect is mainly being driven by activism in roads and water two important areas of investment by GPs. In no case does the power variable predict GP activism (and nor does being the GP headquarter). However, there does appear to be an e ect of village size over and above the Pradhan village e ect in the case of roads. Overall, the results in Tables 6a and 6b are consistent with the Pradhan s village enjoying a policy advantage in the GPs that we are studying. Since we only have cross-sectional data we cannot directly compare public good provision in 2002 with that before the Panchayat system was instituted. However, as a baseline in Appendix Table 1 we consider a set of 1961 and 1991 village public goods as measured in the censuses taken in these years. For consistency, we construct standardized z-scores for each subcategory following the procedure outlined above and estimate the regressions in a SUR framework. In no case, do we nd that the Pradhan village is doing better. Instead the main positive predictor of public good provision appears to be village population. This further supports the notion that the Pradhan village e ect is picking up something about the contemporary level of government activism. In Table 7 we examine whether the Pradhan village e ect is in uenced by either Pradhan or village characteristics. This is interesting as means of seeing whether there are some particular types of Pradhan who are driving the result. The results show no evidence that Pradhan characteristics, as measured by whether he/she has a BPL card, years of education or reservation status, in uence public good allocation. The method of election of the Pradhan also seems not to have any 22