Local funds and political competition: Evidence from the National Rural Employment Guarantee Scheme in India

European Journal of Political Economy 41 (2016) 14–30

Contents lists available at ScienceDirect

European Journal of Political Economy journal homepage: www.elsevier.com/locate/ejpe

Local funds and political competition: Evidence from the National Rural Employment Guarantee Scheme in India☆ Bhanu Gupta a, Abhiroop Mukhopadhyay b,⁎ a b

Department of Economics, University of Michigan, Ann Arbor, USA Economics and Planning Unit, Indian Statistical Institute (Delhi), 7 SJS Sansanwal Marg, New Delhi 110016, India

a r t i c l e

i n f o

Article history: Received 13 February 2015 Received in revised form 13 October 2015 Accepted 14 October 2015 Available online 10 November 2015 JEL classification: D72 J08 H53 H75 Keywords: Political economy Local elections NREGS

a b s t r a c t The National Rural Employment Guarantee Scheme (NREGS) in India is one of the largest public employment programs in the developing world. It was introduced by the central government led by Indian National Congress (INC). While it's implementation is, in principle, based on demand for work from households, we investigate how political competition affects intra district allocation of funds under the scheme. Using longitudinal data on funds allocated to blocks and elections held for block councils, we find that greater amount of funds were allocated to blocks where INC had lower seat share. Further, we address the issue of endogeneity by focusing on a subsample of blocks where the aggregate vote share of INC was close to that of it's rivals. Our results suggest that 1.5 percentage point more funds were approved for blocks that had 1 percentage point lower seat share for INC. We also provide a mechanism for the effect by showing that the results are only true when the MP of the district, a member of the body that approves the block fund allocation, is from INC. © 2015 Elsevier B.V. All rights reserved.

1. Introduction Central governments, all over the world, often introduce flagship public schemes that not only have large budgetary outlays, but lead people to identify the scheme with a particular political regime. For example, Bolsa Familia in Brazil, is often identified with the Lula administration and is believed to have resulted in his victory in presidential elections in 2006. Similarly, the National Rural Employment Guarantee Scheme (NREGS), which guarantees 100 days of employment to rural households in India, is a flagship program of the Indian National Congress party (INC) and was touted to be one of the main reasons for INC getting re-elected to the central government in 2009 (Zimmerman, 2012). In the context of developing countries, the NREGS is an interesting experiment in policy implementation since it requires active participation of elected local representative bodies in rural areas (called the panchayati raj institutions: PRI). While such decentralization, in principle, may lead to better implementation, it also lends itself to local capture. These can often take the shape ☆ This project is jointly funded by the Effective States and Inclusive Development Research Centre (University of Manchester), the European Union's Seventh Framework Programme for research, technological development and demonstration under grant agreement no 290754 (NOPOOR), the Centre de Science Humaines and the Planning and Policy Research Unit (Indian Statistical Institute, Delhi). We wish to thank the participants at the DIAL conference (Paris 2015), 9th Annual Growth and Development Conference (Delhi, 2013), ESID Workshop (Delhi, 2013), Conference on MGNREGS (Mumbai, 2014) and seminar participants at the Institute of Developing Economies (Tokyo, 2014), CASI (U. Penn, 2014), SAU (U. Heidelberg, 2014) and HRI (U. Conn. 2014). We also thank Rajiv Dehejia, Stefan Dercon, Bhaskar Dutta, Himanshu, Dilip Mookherjee, Nishith Prakash, Kunal Sen and two anonymous referees for insightful comments. The views expressed in this publication are the sole responsibility of the authors and do not necessarily reflect the views of the funding agencies. ⁎ Corresponding author. E-mail addresses: [email protected] (B. Gupta), [email protected] (A. Mukhopadhyay).

http://dx.doi.org/10.1016/j.ejpoleco.2015.10.009 0176-2680/© 2015 Elsevier B.V. All rights reserved.

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

15

of elites getting disproportionate share of benefits from a scheme, especially when the beneficiaries are uninformed about the scheme (Bardhan and Mookherjee, 2000). At the same time, policy implementation can also be affected by local political competition: in particular competition between parties in local elections. Political will to implement the scheme can, in principle, be driven by ideologies of parties (as captured by Candidate-Citizen models of Besley and Coate, 1997). However, recent evidence finds that political opportunism can often dictate how policies get implemented. For example, Bardhan and Mookherjee (2010), in the context of West Bengal in India, find that areas which are subject to close legislative assembly elections often see better implementation of land reforms. They find that the relation between implementation and political strength (in terms of seats) is an inverted U, with parties not implementing the reform policy if they have a very low or very high representation in an assembly constituency. Similarly, there can be an interplay of party politics with clientalism. These would involve transfer of public resources to individuals/specific groups associated with the ruling political party (Grossman and Helpman, 1996). In the context of NREGS, there is no major ideological difference between the major parties about the scheme per se1; the difference in posture, if any, has more to do with the fact that the rural voters may identify the scheme with INC since it is one of its flagship programs. This may decrease the will of other political parties to implement the scheme. This leakage of benefits (or lack of it) when parties implement policies has been studied in the context of center-state transfers. For example, Arulampalam et al. (2009) study the impact of national and state assembly compositions on center-state transfers. In their context, the goodwill from center to state transfers is lost to “leakage” if the government at the state and center are from different parties. This affects the transfers the center is willing to make to the state. The case of NREGS is similar. While the scheme is largely funded by the center, the funds are channeled through local bodies that may have key political personnel who are not aligned to the party at the center. Hence, this paper explores whether the funds allocated at the local level are affected by local political competition. The analysis presented in the paper uses data from two waves of block council elections (in 2005 and 2010) and NREGS fund allocation to all blocks for the financial years 2009–10, 2011–12 and 2012–13 in Rajasthan, a state in India. Confounding determinants of demand for funds are controlled for by using block level data from 2001 and 2011 census. Moreover, we carry out block level fixed effects estimation and allow for appropriate trends. We model the funds allocated to a block as a function, among other things, of the existing seat share of the Indian National Congress (INC) in each block council. We find a negative relationship between INC seat share and NREGS fund sanctioned. To allay fears of endogeneity, we focus on a subset of blocks with close vote share difference between INC (BJP) and the largest other party in the 2005 and 2010 elections. To further elaborate, each block has multiple wards and elections are held for each ward. However, since party vote shares are reported for the whole block and not for each ward, we look at the aggregate vote share at the block level. We refer to these blocks as close blocks. In particular, close blocks are defined in terms of aggregate vote margins of no more than 4 percentage points difference between the vote share of INC (BJP) and the closest rival.2,3 We use the INC seat share variation within the close block councils after controlling for the aggregate vote share of INC. The resultant variation in seat shares comes from the varying distribution of INC votes across wards within the block. Using this variation, we find that, for close blocks, the funds sanctioned are 1.5 percentage higher if the seat share is 1 percentage point lower. Thus there is evidence that funds are targeted to areas where INC has lower seat share. This may be feasible within close blocks because voters are not necessarily biased towards any one party. Our analysis using the sample of close blocks does not go as far as proving that the relationship is causal. Concerns may still remain due to lack of data to carry out a more robust evaluation design. However a battery of robustness checks points out that the negative relationship between funds and seat share for INC is very strong and robust. Moreover, we provide further proof that these outlays may reflect political strategies by INC. The result that there are higher funds to low INC seat share blocks is only obtained when the district member of parliament (MP) is from INC. The MP is part of the district council, the body that approves the block plans and is a key political personnel in the district. We find no such result for BJP thus pointing out that, perhaps, BJP does not find it optimal to use NREGS funds since it is identified with the INC led central government, especially post general elections in early 2009. The paper contributes to three strands of the literature: It contributes to the empirical literature on the impact of local political competition on public policy implementation. It gives further evidence that political opportunism guides how parties act on policies. After 2008, INC was in power both at the center and the state. Hence, we are able to abstract away from any center-state issues and focus narrowly on local elections. This analysis is also unique in that we consider fund flow for a policy at the block level. Similar information at this level of disaggregation for implementation of policies are tough to get, especially in developing countries and most papers on developing countries focus mostly on district allocations. What is also useful about this exercise is that it is clear how political parties can affect outcomes, since political appointees have a declared role in fund allocation decisions. These results add to the literature on how political parties allocate funds. Dixit and Longdregan (1996) point out that parties transfer funds to swing voters if the voters are not ideologically inclined to any party, whereas they transfer funds to their core support group if they are better at targeting funds to their own supporters. We look within, what can be considered, swing areas to see how funds are allocated further. We do find a positive impact of vote share of a party (INC) and the funds it gets. However, we point out that within areas with similar vote share, funds are allocated to places with lower seat share. Our results are in contrast to empirical results that find evidence of political patronage in local politics (Besley et al., 2004). 1 The major parties of India are largely left of centre, especially in the context of the rural economy. The differences in rhetoric come largely from posturing during elections. For an interesting take on this issue, refer to http://debrajray.blogspot.in/2013/08/namomania.html. 2 The choice of 4% is adhoc. 4% is the lowest margin difference we can use for this paper due to sample size issues. We show robustness results with lower vote margin differences, but with fewer controls. In particular, this is not an optimal bandwidth, since we do not have data for each ward election. 3 The focus on blocks rather than wards is consistent with the NREGS fund allocation since money is transferred to blocks and not to wards.

16

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

This paper is similar in spirit to Bardhan and Mookherjee (2010) which show that party seat shares matter for policy implementation. However, some of the context is different in this paper and this paper uses a subset of close blocks for identification in contrast to using vote shares in other general elections as instruments. The difference between the two exercises is commented on in a later section of the paper. These results are also in contrast to the literature that points out that pre-election transfer of funds is only useful in getting voters to election booths and not for affecting their voting choice (Cox and Kousser, 1981). The second strand of literature for which this paper is relevant is the role of local politics in affecting economic outcomes in developing countries. Recent work on India, by Cole (2009) and Novosad and Asher (2013), show how local elections and politicians can affect farm credit and employment respectively. Since NREGS funds affect employment rates and have also been found to have impacts on poverty (Ravi and Engler, 2015; Klonner and Oldiges, 2014); by providing some evidence on how politics affect NREGS funds, our paper is indicative of a path for how politics and economic outcomes are connected. These studies, in the specific context of developing countries, add to the larger literature which look at the impact of political competition on economic growth. For example, Besley et. al (2010), in the context of the states of the United States, find that the lack of political competition may lead to policies that hinder economic growth. The third strand of literature that this paper contributes to is the nascent evidence on NREGS. The scheme is one of the largest public policies in a developing country context. With an allocation of Rs. 396.54 Billion in 2012–2013 (around 6.42 Billion USD at PPP), it is bigger than PROGRESSA and has the potential to change the lives of an unprecedented number of people. Studies looking at its impact (Azam, 2012; Imbert and Papp, 2015) are besotted with identification issues since the intensity of the program in any area and over time is not random. In providing a political explanation for funds allocated, this paper provides a potential identification channel to examine its impact.4 In Section 2, we describe the institutional setting of the political structure and the fund allocation procedure across administrative units and how they are related to the local political structure. Section 3 provides description of the data. In Section 4, we lay out an empirical model and describe variables used in a multivariate panel regression model. Further, we describe our identification strategy. Section 5 describes results while Section 6 offers an explanation for the results obtained. Robustness checks are discussed in Section 7 and we conclude in Section 8. 2. Institutional setting In this paper, since we discuss the fund allocation procedure and the impact of local political institutions, it is important to characterize both to understand the interplay. We start with the description of the political background. 2.1. Political Election are held every five years at various levels in India. First and foremost are parliamentary elections that are held for the national parliament. The areas covered by parliamentary constituencies are usually mapped to administrative districts though this mapping is often not exact. Each constituency elects one member to the national parliament (called Member of Parliament: MP). The national government is formed by the party which has majority (or can guarantee that the majority of members of parliament support it). These elections were held in India in 2004 and 2009 and the Indian National Congress (INC) formed the government both times as the leading partner in UPA (United Progressive Alliance). Analogously, voters in each state of India elect members, one from each legislative assembly constituency. Each legislative constituency is however much smaller than a district. They often straddle multiple districts and demarcation of these constituencies generally have a very bad correspondence to the administrative sub-district level unit: the block, which makes it hard to understand allocation decisions, which are often based on census definitions of areas. The government is formed by the party that garners the majority support of the elected members (called the Member of Legislative assembly: MLA). The election for the state government, relevant for this analysis, was held in 2008 and INC formed the state government, taking over from the BJP. In order to decentralize power to the grassroots, India has come up with a parallel system of “Panchayati Raj Institutions” (PRI). These exclude all areas that come under municipalities. Hence, these operate at the level of villages, rural blocks and rural districts. The elections for all PRI in Rajasthan were held in 2005 and 2010. To elaborate further, at the lowest tier of governance, a group of villages is carved out into wards, which then elects one member from each ward. These elected members form the Gram Panchayat, a council, which is headed by a Sarpanch, who is elected directly by voters from each ward (there were 9166 gram panchayats covering 104,937 wards in 2010 elections in Rajasthan). These elections are fought with no formal party affiliation and hence we don't study them.5 The next higher level for PRI is called “Panchayat Samities”. Each census block, within a district, has a panchayat samiti, or the block council. The number of seats in each council are determined by the voter population of the block. A block having population up to 100,000 consists of 15 wards; in case the population exceeds 100,000 then for every 15,000 or part thereof in excess of 100,000, the number of wards increases by two.6 In Rajasthan, the 248 blocks in 2010 were divided into 5273 wards. Since we study fund allocations at the block level, these elections are the focus of this paper.7 Two things are important to note. First, 4

Needless to say, this is contextual, as for many outcome variables, the exclusion criterion may not be met if political competition affects them directly. There is no uniformity in this procedure across states. In some states, the elected council themselves select the head of the council. In some states, these elections are fought with formal party affiliation. 6 This is not a proportional representation system. Hence it is not possible to use some of the identification techniques discussed, for example in Folke (2014). 7 There is no obvious mapping between village wards or gram panchayats and the wards at the block level. 5

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

17

the composition of the block council is the outcome of a separate election for each ward seat. Second, a higher proportion of seats to a party in this council implies a larger control of wards within the block; however there are no further considerations, unlike the case of federal and state government, of forming a local government.8 Apart from these elected members, the block council also consists of the members of legislative assemblies the block council wards overlap with. The block council also contains the head of the village councils, but they have no formal affiliation to any party. The highest level of PRI is the Zilla Panchayat, or the district council, which has elected members from district level constituencies. The number of constituencies within the district council is analogously determined based on voter population cut offs (the threshold is 400,000). These elections are held at the same time as block council elections. In 2010, in Rajasthan, there were 33 district councils which consisted of 1013 constituencies. This council also consists of the head of the block councils and the district member of parliament. Hence the PRI institutes form a parallel political universe, which is largely unrelated to the formation of the central and state governments. They are however important barometers of party support among the voters and are important to have grass root control over villages in rural India. We describe the outcomes of these local elections in our summary section.

2.2. NREGS fund allocation The National Rural Employment Guarantee Act (NREGA) provides a legal guarantee for at least one hundred days of employment in every financial year to adult members of any rural household willing to do unskilled manual work at the notified wage. The National Rural Employment Guarantee Scheme (NREGS), which operationalized the act, started in the financial year 2005– 2006 and was rolled out in phases. Initially restricted to 200 “ poorest” districts of India (February 2006), it was first extended to 130 more districts in phase II (May 2007) and to all districts by 1st April 2008. The legal entitlement of work implies that NREGS is, in principle, a demand based scheme. Thus, various modus operandi are laid out on how demand from households is to be registered and how funds will flow through the system (Mukhopadhyay, 2012). A gram panchayat is responsible for identification of projects in the area under its jurisdiction (through local meetings called Gram Sabha meetings). The plans are then sent to the block level (the next higher tier) before the start of the financial year (this is often referred to as a “labor projection” as well as “suggested shelf of works”). All project proposals received are integrated into the Block Plan. The block council, along with a block level administrative officer ( called the Program Office9) vets the block level plan, and forwards it to the district council for final approval. Based on these plans, funds are approved for blocks, and funds then flow to villages and subsequently to households that work on NREGS projects. While NREGS is, in principle, a demand based scheme, there is overwhelming evidence that the scheme is supply driven. Based on a village survey of 320 villages in Rajasthan, Himanshu et al. (2015) find that around 52% of villages believe that households only get work when there is some project available and not based on their demand.10 Moreover, Imbert and Papp (2015) report that “ many people are unaware of their full set of rights under the program”; “ in practice, very few job card holders formally apply for work while the majority tend to wait passively for work to be provided.” Other research on Andhra Pradesh (Ravi and Engler, 2015; Afridi et al., 2013) also indicate that the program is supply rather than demand driven.11 While fund allocations may not be completely demand driven, it is implausible to think that they are random. Given the various levels of local political institutions involved in the collation of demand requests, it is possible that they can influence the funds that are finally allocated. While there can be political forces at play that decide funds at the district level and at the state level, we focus, in this paper, on the intra district allocation of funds (that is, to blocks).12 Further, we look at the relationship between seat share of each party in the block council to subsequent block level approved funds. Recall that the block council elections are the lowest tier of local elections, for which seat shares are recorded party wise (by the state election commission). In addition, we look at the influence of the MP, who is a member of the district council, a body that finally approves block plans.13 While other layers of politics can matter for allocation of funds under NREGS, what makes the particular context we examine useful, is that the political structure at higher tiers of governance stayed the same during the period of our study. Both the central government and the state governments were headed by the same party: INC.

8

The block council does elect a head. The head usually belongs to the party with the highest number of seats in the council. The Block Development Officer (BDO) is often appointed the program officer. The Program Officer provides preliminary approval based on verification of maintenance of 60:40 ratio of wage to materials in terms of cost. 10 This is based on a focus group discussion in each village. 11 The beauracratic response to how fund allocation happens is mechanical. It is a view stated by almost all officials that the fund allocation is based only on the labour projection budgets and the shelf of works. However, it is hard to see how this is consistent with people reporting their demand for work is not met. 12 Once funds are approved for Gram Panchayats, there can be further local political forces at play. For example, Himanshu et al. (2015) find, that in multi-village Gram panchayats, the village of the head of the Gram Panchayat (called the Sarpanch) gets more NREGS work for his village. 13 We do not look at the party composition of the district council since the members are elected at the same time as the block council members. The MP is elected through a national election which was held separately. 9

18

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

3. Data & descriptives This analysis uses data from Rajasthan, a northern state of India. Rajasthan is touted as a success story in terms of the implementation of the scheme since funds have been used to provide employment in this state, in contrast to most other states of India, where its implementation has been poor. 14 We seek to investigate whether NREGS fund allocation to blocks, in a financial year, depend on the existing seat share of each political party within the block council.15 We exploit the fact that elections for block councils took place in the years 2005 and 2010, which led to a change in the seat share of each party, and examine the fund allocations in the financial years 2009–2010, 2011–12 and 2012–2013. The choice of the years is dictated by the fact that NREGS was implemented in all districts of India (and consequently all blocks of Rajasthan) by mid-2008. Hence 2009–2010 is the first financial year for which we have data for all districts (and blocks).16 The exclusion of the financial year 2010–11 was dictated by the fact, that given the complicated machinery of NREGS, it is plausible that it would take time for the newly elected local politicians to learn about how NREGS funding works. Indeed, 2010–2011 showed a sharp dip in total NREGS funds for the state. We also consider those two years so as to ensure that the unspent balances from previous years that often get extended to the funds available in the next financial year belong to the same political regime (post 2010). Our qualitative results stay the same even if we look at fund allocations in 2011–2012 and 2012–13 separately. The block level approved funds for NREGS for a financial year include fresh funds sanctioned as well as outstanding balance from the previous year.17 Data on these are sourced from the official website of the Government of India.18 The data are obtained for 219 blocks for financial years 2009–2010, 2011–12 and 2012–2013 (for ease of presentation, we refer to them as 2009, 2011 and 2012 respectively).19 The average block level funds for Rajasthan for the year 2009 was Rs. 173.36 million whereas the average annual block funds for the years 2011–2013 was Rs. 90.49 million (Table 1.A). Data on seat share for each party are obtained from the state election commission website.20 Data are obtained on block council elections held in 2005 and 2010. As explained earlier, each block is divided into wards and one member is elected from each ward. Each ward is accorded one seat in the council. The number of wards in each block council varies depending on population and cut offs described in the previous section. We divide the total number of seats a party gets by the total number of seats in the block council to calculate a party's seat share. Rajasthan politics is dominated by the two main national parties of India: the Indian National Congress (INC) and Bharatiya Janata Party (BJP). The average seat share of INC in 2005 was 44.1% while it increased to 48.8% in 2010. The BJP's seat share decreased from 41.7% in 2005 to 34.7% in 2010. The two seat shares together account for around 80% of the seats. The other parties play a very minor role. Put together, they get more than 50% seat share for only 5% of all blocks over the two years. Among close blocks, they are in majority for only 2 blocks. Figs. 1 and 2 show the spatial distribution of INC seat shares across the state for the election years 2005 and 2010 respectively. As can be seen, there is fair heterogeneity in seat shares for both years. It is also important for our analysis that even within a district, there is a fair degree of heterogeneity across blocks in seat share. The striped portions reflect close blocks, that is where INC vote margin (for the whole block) was less than equal to 4 percentage points.21 As can be gleaned from the figures, close blocks are not concentrated in any particular region. A comparison of Figs. 1 and 2 also shows that the seat shares have temporal variation.22 The block level funds are matched to block councils. As noted before, we are able to match these perfectly for 219 blocks and use this subsample for our analysis. Figs. 3 and 4 show the spatial distribution of sanctioned NREGS funds. The unconditional correlation between INC seat share and funds, after pooling the data for the two years, is 0.18 while that for BJP seat share and funds is much weaker at 0.02. However, these correlations could also be driven by other factors: those that affect the household demand for work. Intra-district analysis alleviates some of these concerns. The presence of confounding factors, however requires that we model the correlation between funds and seat shares in a multivariate framework. The data on demographic variables are sourced from Census 2001 and 2011. Rainfall data is available only at the district level and is sourced from the Indian Meteorological Department. Rainfall shocks are derived by taking deviations from a 20 year average for each district. The descriptive statistics for these variables are summarized in Table 1.A for all blocks and in Table 1.B for close blocks for both INC and BJP.

14 The total funds for Rajasthan for the years 2009 and 2012 were Rs. 82,027.25 million and Rs 37,757.78 million respectively. The state government, in many press releases, has claimed that there is decreasing demand for NREGS which needs to be investigated. The drop in overall funds for NREGS in Rajasthan has also been noted by Mukhopadhyay (2012). 15 We choose to look at fund allocations instead of expenditures because the latter is subject to issues of corruption and village politics, issues which are not relevant for testing our hypothesis. The correlation between funds and expenditure is 0.87, which reflects that higher funds do lead to higher expenditures. 16 It may be argued that results are affected by the inclusion of the election year. The choice of 2009 is dictated by the spread of the program. By 2009, all the districts (and blocks) had NREGS running in full swing. A previous year would have meant this allocation would be zero for many districts and blocks which were dealing with the scheme for the first time. Moreover, the national elections were held in early 2009. Hence 2008–09 allocations is likely to be equally distorted by the national elections. However, we do provide evidence in a robustness section that the results are true if we exclude the election year from our sample. 17 The proportion of Outstanding balance to total funds was 0.22, 0.20 and 0.19 for the years 2009, 2011 and 2012 respectively. 18 http://nrega.nic.in/netnrega/home.aspx. 19 There are 248 blocks in total. We drop blocks which could not be mapped onto panchayati samitis, the area delimited for election purposes. 20 http://www.rajsec.rajasthan.gov.in. 21 Vote shares of each party are not available for each ward. However, in so far as wards often cut through different villages, the block is the lowest level where funds can be mapped onto party affiliation of the elected members, for the state of Rajasthan. Parties are not officially registered, for example, at the level of the gram panchayat. 22 INC is relatively weaker in the north eastern blocks. However, even there, there is intra district variation in seat shares of INC.

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

a)

19

b)

Fig. 1. (a) INC seat shares (2005). (b) INC seat shares (2010).

4. Empirical model and identification Our main hypothesis is that, controlling for other factors that affect demand for funds, political factors, such as party politics, have a role to play in fund allocations across blocks. In particular, we test what the nature of this role is. It is not clear a priori what the relation should be. For example, models of patronage imply that funds should be transferred, where it is possible to do so, to where the vote bank of parties are. Alternatively, it may be optimal, in some contexts, to transfer funds to swing areas where the marginal impact of fund transfers on votes is the highest. In other contexts still, greater funds may be transferred (if such transfer is possible) to constituencies where a party has fewer representatives, if vote buying is cheap and funds are not constrained. To fix ideas, let b stand for the block; let d refer to the district where b is situated. The dependent variable in this analysis is the log of funds (Ln_fundsbdt); where t takes the value 0 for the year 2009, 1 for 2011 and 2 for 2012. In our main regression, we take the percentage of seats won by INC in a block council as our main political economy variable (INC_seatsharebdt). Since share of seats of all parties within a block council add up to 100, the marginal effect of INC_seatshare measures the impact of a marginal change of seat share of INC relative to other parties, in particular the BJP. The variation in seat shares may however reflect a systematic impact due to the difference in the total number of seats (wards). The number of wards may then be correlated to the fund allocation. Hence we control for the number of wards in a block (wardsbdt). Since the number of wards are typically a function of population, the number of wards varies over time because of the growth of population, even though the demarcation of a block does not change. The difference in seat shares may also reflect voter preferences for a policy. As a proxy for any bias of voters for any party that may correlate to fund allocation, we include the vote share margin between INC and the next highest party as a explanatory variable. To eliminate the impact of demand on NREGS funds, we control for variables that may affect the demand. We posit that the demand for NREGS funds depends on rainfall shock (rain_devdt) as NREGS has been put in place to mitigate effects of droughts. Moreover, funds allocated may depend on the population of a block popbdt. One would expect more funds would be allocated to areas where there was a higher proportion of the relatively less prosperous communities. Hence the proportion of Scheduled Castes (SCbdt) and Scheduled Tribes (STbdt) in the block are included as control variables. Moreover, the labor force participation of women in NREGS has been large in Rajasthan. Hence we include the proportion of females in the population ( fembdt) as an

20

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

a)

b)

Fig. 2. (a) Block-wise NREGS funds (2009–10). (b) Block-wise NREGS funds (2012–13).

explanatory variable. Further, to measure underdevelopment at the block level, which may lead to a higher NREGS demand, we take into account the illiteracy rate ILLbdt. To alleviate concerns that unobserved variables may influence fund allocations, we include block fixed effects (δbd) to take into account block level idiosyncrasies, for example, its geographic location. Moreover, we allow for time dummies, (δt), one each for 2011 and 2012, to take into account the aggregate change in funds for NREGS in Rajasthan. We also include district trends (ρdt) over the period to take into account trends in alternative employment opportunities (wages) at the district level. In addition we

Fig. 3. INC seat share and vote share: close blocks.

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

21

Fig. 4. Variation of INC seat share by number of wards in close blocks.

allow for a trend that depends on a development index for a block ccc23 and another trend that depends on the amount of irrigated land within a block (Irrbd0). Both these variables are measured in 2001 and reflect base values.24 Hence the empirical model we estimate is: Ln fundsbdt ¼ α þ δt þ δbd þ ρdt þ ρ1 Irr bd0  t þ ρ2 Infrabd0  t þ þβ1 INC seatsharebdt þ β2 INC vote m arg inbdt

ð1Þ

0

þβ3 wardsbdt þ μ Z bdt þ εbdt where Z is a vector that includes all the other control variables. To estimate this model, we use a balanced panel of blocks and apply a fixed effects estimator. This eliminates the block level time invariant idiosyncrasies. It also eliminates rainfall shock, as that is measured at the district level, and is therefore collinear with the district trend. The district trend also eliminates the need to include district funds as a variable. We are then interested in examining the sign and statistical significance of β1. It may be contended that the share of seats won by INC may themselves be affected by funds. While this is unlikely for 2005 election (NREGS was not around), it is plausible that funds allocated in 2009–10 affect election outcomes in 2010. This would violate the strict exogeneity restriction. In order to alleviate this problem, we look at close blocks. We define a close block for INC as one where the difference of the vote share of INC and the next highest party was less than or equal to 4 percentage points. The focus on close block implies that we are looking at blocks that are similar to each other in other dimensions. Indeed, a comparison of Tables 1.A. and B. shows that the standard error of most covariates is lower for the sample of close blocks relative to all blocks. However, the choice of the 4 percentage point difference is adhoc. Ideally, in the absence of stipulated thresholds, we would prefer to look at blocks which have very little bias towards any one party, at least at the aggregate level. However, if we move below 4 percentage point difference, the set of blocks shrinks so much that we do not have at least two such blocks in each year for most districts. The loss of at least two blocks per district per year implies that district trends are not useful to take care of time specific district changes in the macro-economy as well as in district funds. We do provide some results in a robustness section with smaller vote share differences, but these are at the cost of omitting district trends. As a baseline, we estimate a slight modification of the empirical specification above. Instead of INC_seatshare, we consider an interaction of INC_seatshare and a dummy for close blocks Dclose. We also introduce the dummy for close blocks to take into account any fund allocation that is driven by the fact that block election was close. The interaction term therefore identifies the relation between funds and INC seat shares only for close blocks. Ln fundsbdt ¼ α þ δt þ δbd þ ρdt þ ρ1 Irr bd0  t þ ρ2 Infrabd0  t þ þβ1 INC seatsharebdt  Dclosebdt þ δc Dclosebdt

ð2Þ

0

þβ2 INC vote m arg inbdt þ β3 wardsbdt þ μ Z bdt þ εbdt While the use of panel data addresses the issue of time invariant block heterogeneity, there still remain issues of strict exogeneity. Besides, most of the relevant variation comes from the temporal variation of INC_seatshare among blocks that are close in both periods, that is 27 blocks. We therefore provide another estimation exercise that uses only the sample of close 23 (Infrabd0) is created using principle component analysis taking into account Average No of Schools per village, Proportion of Villages with power supply, Proportion of villages with a medical facility. 24 The data for these variables are sourced from 2001 census. Similar data are not available currently for the 2011 census at the block level.

22

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

Fig. 5. Temporal dependence in INC seat share: all blocks.

blocks. The sample of close blocks is unbalanced. Over the two elections, there are 147 close blocks with respect to INC. We use data from both 2005 and 2010 elections to gain enough sample size. Using this data we estimate a model akin to that described in Eq. (1) using a random effects model. Since the use of random effects on the sample of close blocks comes at the cost of not being able to eliminate the possible effect of idiosyncratic differences between block councils, it is important to emphasize what problems it addresses. First, as pointed out, a fixed effects estimator would require that the funds in 2009 do not affect the proportion of seats won by INC in the 2010 elections. We contend that the use of close blocks alleviates this problem. To show this, for the sample of close blocks, we regress the proportion of seats won by INC in 2010 (INC_seatsharebd1) on the log of funds in 2009 as well as other baseline covariates and include district fixed effects. Appendix A1.1 shows that the coefficient of the log of funds is always insignificant. Further, it is possible that the probability of being a close block in 2010 is a function of funds which would lead to a sample selection bias. To show that this is not the case, for 2010, we run a linear probability model where the dependent variable takes the value 1 if the block was a close block and 0 otherwise. After controlling for all the confounding factors, we test whether funds in 2009 affect the probability of close blocks. Our exercise shows that the probability of being a close block is not affected by the log of funds in 2009.25 The use of cross sectional variation in the random effects model opens up the possible problem of contemporaneous endogeneity. The question that one needs to address is whether the variation in INC_seatsharebdt is indeed quasi random. We offer two pieces of suggestive evidence to indicate this. The main concern for endogeneity is whether INC_seatsharebdt picks up preferences of voters for the policy. The population threshold based ward formation implies that the correlation between vote share that would measure preference of voters for a certain party and its seat share are not perfectly correlated. For the universe of blocks, correlation between the vote share and seat share for INC is high at 0.74 but not 1. For the sub-sample of close blocks this correlation is lower at 0.46. Fig. 3 plots the relation between INC vote share and INC seat share for close blocks. Among close blocks, the percentage of seats to INC varies from 24 percent to 76%. An explanation for this somewhat lower correlation is that the support for INC is not uniform across wards within the same block. Hence it is possible to have a high vote share overall but not high seat share. To explore this further, we run a regression of INC seat shares on INC vote shares for close blocks. The residual is then the variation in seat shares not explained by the vote share. The number of wards in a block may have something to do with this. Hence we plot the calculated residual against the number of wards in a block in Fig. 4. As can be seen in the figure, even conditional on the number of wards, the residual shows variation. We use this variation to identify the impact of seat share on funds in our regression. Thus, in the base specification, we control for the vote share of INC and the INC vote margin, so as to saturate the effect of aggregate voter preferences. Any significant effect of seat share then measures the impact of a change in proportion of seats. Hence we estimate: Ln fundsbdt ¼ α þ δt þ ρdt þ ρ1 Irr bd0  t þ ρ2 Infrabd0  t þ þβ1 INC seatsharebdt þ β2 INC votem arg inbdt

ð3Þ

0

þβ3 wardsbdt þ þθINC votesharebdt þ μ Z bdt þ εbdt : The other suggestive evidence is in terms of temporal correlation of seat shares for the close blocks. In particular, we contend that close blocks behave a bit like swing constituencies.26 As can be seen in Fig. 5, there is a positive temporal correlation between 25 Since the 2010 election results do not vary post 2010, we consider data from only one time period post 2010. This is in contrast to the overall regression where in there is variation in funds over the years, post 2010, as well as different values for the terms that have trend interactions. 26 We do not call them so, as these are not a specific set of blocks that always show variation in how they vote.

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

23

Fig. 6. Temporal dependence in INC seat share: close blocks.

INC seat shares if the whole sample is considered. However, this relation disappears when we take the subset of close blocks (Fig. 6).27 Analogous to the above specifications with INC, we estimate models where the INC_seatshare is replaced by BJP_seatshare. To maintain comparability, close blocks are defined in terms of vote margins for BJP.28 Next, we test the hypothesis if key political appointees matter for funds sanction within close blocks. We focus on the Member of Parliament. The MP is a member of the district panchayat which, together with the administrative officer, approves block level fund allocations. We construct a variable: INC_MP which takes the value 1 if the district MP is from INC, 0 otherwise.29 Thus, we modify Eq. (3) to include this variable by interacting it with INC_seatshare30. Thus: Ln fundsbdt ¼ α þ δt þ ρdt þ ρ1 Irr bd0  t þ ρ2 Infrabd0  t þ þβ1 INC seatsharebdt þ β2 INC votem arg inbdt þβ3 wardsbdt þ θINC votesharebdt þ

ð4Þ

: þβ4 INC seatshare dt  INC MP dt 0

þμ Z bdt þ εbdt We estimate a similar regression for BJP_MP. The reported standard errors are robust and are clustered at the block level.31 5. Results We present results from a pooled OLS regression without district fixed effects (Table 2; column 1), with district fixed effects (column 2) and with both district fixed effects and district trends (column 3). We turn to the results of column 1 later in order to look at the signs of other variables, that form important determinants of fund allocations (these have less intra district variation). The coefficient of INC_seatshare is insignificant in column 1 but is negative and significant as soon as we include district fixed effects. The coefficient of INC_seatshare retains this negative sign and significant in column 3 and in column 4 when we include block fixed effects in addition to district trends. A negative coefficient implies that larger funds are available where the seat share of INC in the block is low. A block with a 1 percentage point lower INC_seatshare has 0.5 percentage point higher funds.32 The coefficient of INC vote margin is positive and significant, conveying the idea that INC voter preferences do matter for NREGS fund allocation. More funds are sanctioned for blocks which vote more in favor of INC relative to other parties. The total number of wards are correlated with the population size in a block, hence they are positive and significant. Some patterns emerge from these regressions. First, the inclusion of district fixed effects seems to matter since the negative coefficient is significant as soon as we include them. In column (1), it is clear to see that other variables that are natural predictors 27

Regressions are used to test these correlations statistically and find no statistically significant correlation. The number of block councils where BJP has close elections is 136. Anecdotally, it would seem that the pradhan (head of the panchayat samiti) and the head of the district panchayat are also important in getting higher funds for a block. However, the elections for the district and panchayat samiti take place at the same time and the heads are chosen from within the elected members. Hence we look at the district MP, who was elected at the beginning of 2009 via national elections for a period of 5 years. 30 The variable, in its uninteracted form, is collinear with the district trend. 31 Results do not change if we cluster at the district level. 32 Everything else the same, the ratio of funds at two levels of INC_seatshare, one percentage point apart, is equal to exp ( 0.0055) = 1.0055. 28 29

24

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

Table 2 All blocks. Dependent variable: log of funds

(1)

(2)

(3)

(4)

INC-OLS

INC-OLS

INC-OLS

INC-FE

Whole sample

Whole sample

Whole sample

Whole sample

−0.00008 (0.0032) 0.0088⁎⁎

−0.0041⁎ (0.0022) 0.0089⁎⁎⁎

−0.0041⁎ (0.0024) 0.0083⁎⁎

−0.0055⁎⁎ (0.0027) 0.0086⁎⁎

Proportion of scheduled caste individuals

(6.42e−07) 1.961⁎⁎ (0.886) 1.091⁎⁎⁎ (0.341) 9.843⁎⁎ (4.14) 2.15⁎⁎

(0.0033) 0.0427⁎⁎⁎ (0.0075) −0.149 (0.263) −0.0161 (0.119) −0.0239 (0.0355) −0.0559⁎⁎⁎ (0.02) 4.02e−07 (5.24e−07) 0.874 (0.898) 0.123 (0.365) 2.765 (4.524) 1.753⁎⁎

(0.0035) 0.0448⁎⁎⁎ (0.0073) −0.371 (0.279) 0.208 (0.130) 0.032 (0.039) −0.061⁎⁎⁎ (0.019) 2.72e−07 (5.39e−07) 0.874 (0.928) 0.201 (0.374) −1.05 (4.48) 1.66⁎⁎

(0.0043) 0.040⁎⁎⁎ (0.0121)

Total Population

(0.0037) 0.047⁎⁎⁎ (0.0098) −0.662⁎⁎⁎ (0.254) 0.036 (0.117) −0.062 (0.041) −0.041⁎⁎ (0.020) 1.12e−06⁎

(0.838) −0.484⁎⁎⁎ (0.135) −0.676⁎⁎⁎ (0.112) −0.109 (0.132) −0.492 (2.072) No No No 617 0.42

(0.765) −0.242 (0.150) −0.767⁎⁎⁎ (0.116) −0.0189 (0.147) 2.453 (2.197) No Yes No 617 0.64

(0.794)

−0.054⁎⁎ (0.021) −3.55e−06⁎⁎ (1.70e−06) −7.88 (4.84) 0.410 (1.55) −1.115 (7.69) 1.718 (1.422)

−0.1669 (0.126) 1.149⁎⁎⁎

−0.218 (0.226) 1.09⁎⁎⁎

(0.212) 3.686 (2.175) No Yes Yes 617 0.691

(0.244) 6.74⁎ (3.95) Yes

INC seat share INC vote margin Total number of wards Proportion of land irrigated (2001) Proportion of land irrigated (2001) × Trend Infrastructure index (2001) Infrastructure index (2001) × Trend

Proportion of scheduled tribe individuals Proportion of females Proportion of illiterates Rain shock Time dummy (2011–12) Time dummy (2012–13) Constant Block fixed effects District fixed effects District trend Observations R-squared Number of blocks

0.238⁎ (0.129)

Yes 617 0.544 219

Robust Std. Errors ( in parentheses) are clustered at block level. ⁎⁎⁎ p b 0.01. ⁎⁎ p b 0.05. ⁎ p b 0.1.

Table 3 All blocks.

Dependent variable: log of funds BJP seat share BJP vote margin Other controls Block fixed effects District fixed effects District trend Time dummies Observations R-squared Number of blocks

(1)

(2)

(3)

(4)

BJP-OLS Whole sample 0.0038 (0.003) −0.003 (0.005)

BJP-OLS Whole sample 0.0016 (0.002) −0.003 (0.003) Yes (As in Table 2) No Yes No Yes 617 0.63

BJP-OLS Whole sample 0.00058 (0.0027) −0.0006 (0.0036)

BJP-FE Whole sample 0.00327 (0.0028) −0.0031 (0.0035)

No Yes Yes Yes 617 0.688

Yes . Yes Yes 617 0.542 219

No No No Yes 617 0.41

Robust Std. errors ( in parentheses) clustered at block level. *** p b 0.01, ** p b 0.05, * p b 0.1.

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

25

Table 4 Main results.

Log of funds

(1)

(2)

(3)

(4)

(5)

(6)

FE

FE

RE

RE

RE

RE

All blocks

All blocks

Close blocks (INC)

Close blocks (BJP)

Close blocks (INC)

Close blocks (BJP)

INC seat share INC seat share × Dum. close blocks (INC) Dum. close block (INC) INC vote margin

−0.0083* (0.004) 0.457** (0.213) 0.002 (0.004)

−0.0155** (0.0074)

−0.0187** (0.007)

0.035 (0.028)

0.026 (0.027) 0.031* (0.017)

INC vote share BJP seat share BJP seat share × Dum. close block (BJP)

0.004 (0.006) −0.283 (0.252) 0.0005 (0.003)

Dum. close block (BJP) BJP vote margin

0.0073 (0.0067)

0.0036 (0.0068)

−0.009 (0.027)

−0.0140 (0.026) 0.026* (0.015)

BJP vote share (Other controls as in Table 2) Block fixed effects District fixed effects District trend Observations Number of blocks

Yes

Yes

Yes 657 219

Yes 657 219

Yes Yes 211 120

Yes Yes 191 112

Yes Yes 211 120

Yes Yes 191 112

Robust Std. Errors ( in parentheses) are clustered at block level. *** p b 0.01, ** p b 0.05, * p b 0.1.

of funds largely come out to be significant and have the expected sign. The proportion of irrigated land is, as expected, negatively correlated with funds since lesser funds are needed if blocks already have the capacity to deal with droughts. Larger funds go to blocks where the proportion of relatively poorer communities (Scheduled Castes and Tribes) is higher. Similarly higher funds go to where there are higher proportion of illiterates and where the proportion of females is high (as stated earlier, in Rajasthan, the female labour force participation in NREGS is high). Therefore it is clear that NREGS funds do vary by characteristics that they should, in principle, be determined by. However, these characteristics, do not by themselves explain all the funds that are available, especially when we move to intra district analysis. The relatively high R squares in the regressions are not merely reflective of the high number of variables that we consider (the inclusion of trends does increase the number of variables ). Even with just district fixed effects, we are able to explain around 64% of the variation. The results for BJP (Table 3) are different. While the coefficients are positive, they are insignificant. The point estimate suggests that this is mirror image of the result for INC since the two shares are negatively correlated (a correlation of −0.5). We now move to a discussion of the estimation results using the main identification strategy of looking at close blocks. Our results using close blocks are reported in Table 4. In column (1), we report results from a fixed effects estimation exercise for INC. The interaction term of INC_seatshare and Dclose is negative and significant (−0.0083). Hence, larger funds are available to blocks where there was a close election and where the INC seat share was lower. The dummy for close blocks is itself positive and significant. This message is echoed when we look at columns (3) and (5), where we show the result for INC seat share, with and without INC vote shares respectively. Column (3) shows that the coefficient of INC_seatshare is negative and significant (−0.015) Hence the qualitative results from the fixed effects regression go through. However, the much larger value of the coefficient implies that a block with 1 percentage point lower seat share for INC gets 1.5 percentage point higher funds. The coefficient is negative and significant and almost the same, statistically, when we control for the INC vote share in column (5). In line with the results in the fixed effects regression, the coefficient of vote share itself is positive and significant, which implies that higher funds do go to blocks where INC has higher vote share. The positive sign of INC vote share may seem a bit counter to the results for seat share. However, recall that INC vote shares measure the mass of voters whose ideal platform is close to that of INC. It is possible that INC voters have a preference for working on the scheme. However, the partial correlation between funds and seat share, purged of the confounding effect of vote share, points to a more strategic behavior on the part of INC. Next we turn to results for BJP, when we look at the set of close blocks for BJP (Table 4, columns 2,4 and 6). However, we find no significant results for BJP.33

33

There are some significant results for BJP in some specifications if we consider only 2009 and 2012. However, these results are not robust.

26

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

Table 5 Effect of member of parliament. Dependent variable: log of funds

INC seat share INC MP × INC seat share

(1)

(2)

RE

RE

Close blocks (INC)

Close blocks (BJP)

−0.0098 (0.0075) −0.016* (0.009)

BJP seat share

0.0037 (0.008) 0.005 (0.008)

BJP MP × BJP seat share INC vote share INC vote margin

0.029* (0.017) 0.0189 (0.027)

BJP vote share

0.0003 (0.008) −0.008 (0.025)

BJP vote margin Other controls District fixed effects District trends Trend Observations Number of blocks Robust Std. errors (in parentheses) are clustered at block level. *** p b 0.01, ** p b 0.05, * p b 0.1 Marginal effect of INC seat share (INC MP = 1) Marginal effect of BJP seat share (BJP MP = 1)

YES (As in Table 4) Yes Yes Yes 211 120

Yes Yes Yes 191 112

−0.026*** (0.009) 0.0056 (0.0065)

These results have two plausible explanations: First, it may be the case that higher funds are allocated when non INC shares are high (among close blocks). This may be a political strategy of other parties to “channel” funds for patronage. Second, it may be INC that targets funds where its seat share is low, especially among close blocks. To investigate this further, we delve deeper into a mechanism that may drive this result. We use the fact that NREGS funds are not akin to party funds that are available to parties to allocate. These are public funds that are allocated with involvement of the administrative machinery. For any political strategy to work, there must a way to influence public funds. This would require the presence of party members in key positions in the fund allocation process. For this, we look at the results from estimating Eq. (3). We posit that since the funds are approved by a district council comprising of the district MP, it is more likely that strategies can be implemented for any party when the district MP is from the same party. In Table 5 we report results for the regression which includes MP as an interaction. In columns (1) and (2) we estimate equation (3) for INC and BJP MPs respectively. Results in column (1) show that the marginal effect for INC_seatshare is only significant when the MP is from INC. The marginal effect of INC seat share is −0.026 when the MP is from INC This implies that in the districts where the MP is from INC, when the seat share for INC is 1 percentage point lower, the funds transferred are 2.4 percentage points higher. There are no significant results from column (2) which suggests that the BJP MP has no role to play.34 These results are not necessarily driven by the quantum of funds since the average district funds for INC MPs is much lower at Rs. 718 million as compared to BJP MPs whose funds are Rs. 1203 million.35 The two sets of results put together suggest that the position of the district MP being from INC is crucial and that it is her involvement that ensures that the strategy of the INC is played out. 6. Discussion The results indicate an allocation of funds to where INC has lower seat share. The results are better identified using close blocks. However, they are suggested even in the fixed effects estimation. What would explain these results? A convenient tethering point is Bardhan and Mookherjee (2010) which adapts the Gross Helpman model to look at the issue of political influence over policy. In their model, the relationship between seat share and policy implementation is inverted U. There is, however, a key 34 These results are borne out if one looks at the estimation of an equation where Eq. (2) is modified. If we introduce INC_seatshare * Dclose and INC_seatshare * Dclose * INC_MP, then the former is insignificant where as the latter is negative and significant. The total marginal effect of a 1% higher INC_seatshare when there is an INC MP is −0.013 There are no results for when we consider an analogous regression for BJP. 35 This is consistent with a similar strategy by INC at the district level: to target funds to where it loses.

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

27

Table 6 Alternate definitions of close blocks. Dependent variable: log of funds

INC seat share (%)

(1) RE

(3) RE

(INC: 4%)

(INC: 3%)

(INC: 2%)

−0.0187** (0.0078)

−0.0134*** (0.005) 0.00007*** (.00001) YES (as in Table 4) Yes No Yes 161 95

−0.007 (0.009) 0.00008*** (.00002) YES (as in Table 4) Yes No Yes 108 67

District funds Other controls District fixed effects District trends Trend Observations Number of blocks

(2) RE

YES (As in Table 4) Yes Yes Yes 211 120

Robust Std. Errors (in parentheses) are clustered at block level. *** p b 0.01, ** p b 0.05, * p b 0.1.

fundamental difference in the context we look at. Fund allocation is not necessarily an indication of actual policy implementation. While it is correlated with policy implementation (which would be better measured by expenditure and has a correlation of 0.87 with funds), it can be better described as an intent. Given evidence of political economy possibilities at lower level of governance (for example at the level of the Sarpanch, see Himanshu et al., 2015), it is difficult to believe that a party can control expenditure all the way to a village.36 This point is important because in the Bardhan and Mookherjee (2010) model, a big driver of the lower policy implementation, when the seat share of a party is low, is it's low ability to implement it's favored policies. Here as we show, the ability to allocate funds rests with the district council. While bureaucracy has a key role to play in this, our results show that the MP too has an important role to play. This role is not merely implied, but is explicitly designed into the fund allocation process as the MP is a part of the district council. To complete the narrative, one needs to reconcile why we don't observe results similar to INC for BJP. The ability to use funds for political mileage is critically linked to attribution. One explanation that can rationalize our results is that it is not optimal for BJP to use funds from NREGS since it is primarily thought of us as a central government scheme. Hence there may be leakage of goodwill that defeats the purpose of using these funds to influence voters in a close election. Some of the reticence to use NREGS funds for political purposes may be based on history. The state government, governed by BJP, was important in implementing NREGS in it's initial stage between 2006 and 2008. However, INC seemed to have got all the credit from it in the 2009 general election (Zimmerman, 2012). Recall that the political economy results are obtained even after controlling for other characteristics that should determine NREGS fund allocations. Hence, this marginal effect is not the total impact of political economy variables. Party politics can, for example, lead to better resources for marginalized communities. But these effects are already captured by inclusion of the variables like the proportion of marginalized communities in our regression. In this paper we have tried to characterize the effect of party politics over and above its effects through other variables. This, then, is a distortion to the fund allocating process. 7. Robustness In this section we carry out three robustness checks. The choice of 4 percentage point difference, while driven by considerations we have discussed in earlier sections, is still adhoc. Hence first, we check if results change a lot if we define close blocks with narrower vote margins. To accommodate the narrower margins, we need to alter specifications to take into account the fact that district trends cannot be considered.37 We check our results first with a 3 percent vote margin. Since we cannot take into account the district trends (we keep all the controls, the time fixed effects as well as district fixed effects), we need to control for the changing amount of district funds over time. Hence we include district funds as a control variable. Table 6 (column (2)) shows that the results are still similar with 3 percent vote margin. Results are insignificant at 2 percent vote margin but one cannot reject the null that the coefficient of INC seat share in column (3) is the same as the coefficient of the same variable in the base specification in column (1). As a second test, we drop the sample for the year 2009 ( Table 7). As before, results point out to negative and significant coefficient for INC_seatshare. The variation in seat shares, after removing the impact of vote shares is driven, to an extent, by the distribution of votes between wards within a block. As a third robustness check, we therefore include some variables that characterize ward level elections.38 The 36 In many states of India, village leaders have a fluid party affiliation since it does not need to be declared. This is in contrast to, for example, West Bengal, a state studied by Bardhan and Mookherjee, where even village elections are fought on party lines. 37 The inclusion of more years does not solve the problem as the definition of the close block is fixed by the elections. Hence we would still have one block in every time period, irrespective of increasing the time periods. 38 These data were collected from the state election commission office and are poorly maintained. They are only available for a subset of blocks and the election sheets often don't have the parties written against votes. It may be possible to figure out the parties using an algorithm that matches the total votes and the total seats reported in the aggregate data set. However, these fail especially for close margins, which is crucial for this paper. Hence it is not possible to create an identification strategy around these disaggregated data.

28

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30 Table 7 Results excluding 2009–10. Dependent variable: log of funds (2011–12) to 2012–13

RE Close blocks ( INC-2010) −.023** (0.0096) 0.0002*** (0.00006) Similar to Table 4 but for t = 1 Yes Yes 128

INC seat share (2010) District funds Other controlst = 1 District fixed effects Time dummy included Observations Robust Std. Errors (in parentheses). *** p b 0.01, ** p b 0.05, * p b 0.1.

ward level data are only available for 92 of the close blocks (INC), so the sample size is smaller for this section ( Table 8). In column (1), we report results including as a control the ward level difference in vote shares between the parties with the two highest vote share, averaged over all wards in a block. In column (2), we add to this specification the variance of the ward level vote share differences between these two parties. In column (3), we include, instead of the variance, the number of wards where the vote margin between the top two parties was less than 4 percentage points (consistent with our idea of close: these close elections constitute around 20% of ward elections in the block). In column (4), we include the average ward level Herfindahl index using vote shares of the parties. The idea behind all these variables is to control for the distribution of support for INC within blocks. Our coefficient remains robust to the inclusion of all these variables. 8. Conclusion In this paper, we provide evidence on how political competition affects fund allocations for the National Rural Employment Guarantee Scheme. We use the particular context of the National Rural Employment Guarantee Scheme (NREGS) in Rajasthan, a state in India. This scheme was introduced by the Indian National Congress (INC). Using panel data techniques, we show that funds allocated to blocks are negatively correlated with the INC seat share. To alleviate problems of endogeneity, we consider a subset of close blocks, where INC vote share was close to it's rivals. We find that 1.5 percentage point larger funds were sanctioned in areas where INC seat share was 1 percentage point lesser. We offer an explanation for these results in terms of a mechanism. In particular, we point out to the importance of the MP since results seem to be driven by whether a district has an MP from INC. The literature on the National Rural Employment Guarantee Scheme has a plethora of identifying assumptions. Some researchers use timing of the NREGS in districts since the implementation was phased. Others use intensity of NREGS, assuming that it is exogenous. This paper offers a deeper explanation in trying to understand what determines the fund allocation to blocks, albeit in one state of India. The political economy explanation is by no means the sole explanator of fund allocation in any state of India, but this paper suggests that it is a significant one. Table 8 Ward level descriptives: close blocks (INC). Dependent variable: log of funds

INC seat share (2010) Average ward level vote margin difference

(1)

(2)

RE −0.014* (0.0076) 1.25 (1.43)

Variance of ward level vote margin difference

(3)

(4)

RE

RE

RE

−0.014* (0.0076) 0.707 (2.21) 3.26 (9.83)

−0.014* (0.0076) 1.03 (1.70)

−0.014* (0.0078)

−0.008 (0.027)

Number of wards with less than 4% victory margin

0.00007*** (0.00001)

0.00006*** (0.00001)

−1.051 (2.18) .00007 (0.00001)

Yes Yes 153 92

Yes Yes 153 92

Yes Yes 153 92

Average ward level Herfindahl Index District funds Other controls District fixed effects Time dummy included Observations Number of blocks Robust Std. Errors (in parentheses) clustered at block level. *** pb0.01, ** pb0.05, * pb0.1.

0.00006*** (0.00001) Similar to Table 4 Yes Yes 153 92

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

29

Appendix A1.1

Table 1.A Summary statistics: all blocks. Overall

Funds (in Rs 100,000) Log of funds Total No. of wards INC seat share (%) BJP seat share (%) INC vote margin (%) BJP vote margin INC MP BJP MP Rain shock Avg. Prop. of land irrigated (2001) Infrastructure index of block (2001) Total population Proportion of SC Proportion of ST Proportion of females Proportion of illiterates Close blocks (INC) Close blocks (BJP)

Time period (1): 2009

Time periods 2 and 3 (2011–2013)

Count

Mean

S.D.

Count

Mean

S.D.

Count

Mean

S.D.

657 657 657 657 657 657 657 657 657 657 657 657 657 657 657 657 657 657 657

1181.62 6.63 21.78 47.25 37.04 −0.02 −5.21 0.60 0.39 0.03 0.39 0.20 219,228 0.19 0.16 0.48 0.51 0.32 0.29

1148.79 1.07 5.19 17.71 18.52 12.20 11.24 0.49 0.49 0.45 0.22 1.36 88,068 0.07 0.21 0.01 0.08 0.47 0.45

219 219 219 219 219 219 219 219 219 219 219 219 219 219 219 219 219 219 219

1733.61 7.13 22.21 44.11 41.70 −1.77 −2.65 0.15 0.85 −0.35 0.39 0.20 195,207.00 0.18 0.15 0.48 0.56 0.38 0.37

1396.39 0.85 6.28 16.34 16.84 11.33 10.37 0.35 0.35 0.25 0.22 1.36 73,501.86 0.07 0.20 0.01 0.08 0.49 0.48

438 438 438 438 438 438 438 438 438 438 438 438 438 438 438 438 438 438 438

904.93 6.37 21.57 48.82 34.70 0.86 −6.48 0.82 0.16 0.22 0.39 0.20 231,239.68 0.19 0.16 0.48 0.48 0.29 0.25

882.59 1.08 4.54 18.17 18.90 12.54 11.45 0.38 0.36 0.40 0.22 1.36 92,377.71 0.08 0.21 0.01 0.07 0.46 0.43

Table 1.B Summary statistics: close blocks. Close blocks (INC) Count Funds (In Rs. 100,000) Log of funds Total no. of wards INC seat share (%) BJP seat share (%) INC vote margin (%) BJP vote margin (%) INC vote share BJP vote share INC MP BJP MP Rain shock Avg. prop. of land irrigated Infrastructure index of block Total population Proportion of SC Proportion of ST Proportion of females Proportion of illiterates

Close blocks (BJP) Mean

S.D.

211 211 211 211 211 211 211 211

1224.15 6.84 21.12 45.4 42.49 0.05 . 42.84

965.55 0.78 4.96 10.76 14.08 2.28 . 4.91

211 211 211 211 211 211 211 211 211 211

0.49 0.47 −0.07 0.44 0.15 196,510.09 0.20 0.12 0.48 0.51

0.50 0.50 0.41 0.23 1.34 78,806.8 0.07 0.15 0.01 0.08

Count

Mean

S.D.

191 191 191 191 191 191 191

1256.38 6.89 21.00 44.72 45.52 . 0.01

977.57 0.73 4.75 10.31 10.79 . 2.26

191 191 191 191 191 191 191 191 191 191 191

43.32 0.45 0.53 −0.11 0.43 0.11 188,883.98 0.20 0.12 0.48 0.51

4.91 0.50 0.50 0.39 0.23 1.36 72,664 0.07 0.12 0.01 0.08

References Afridi, F., Iversen, V., Sharan, M.R., 2013. Women political leaders, corruption and learning: evidence from a large public program in India. IZA Discussion Paper No. 7212. Arulampalam, W., Dasgupta, S., Dhillon, A., Dutta, B., 2009. Electoral goals and center-state transfers: a theoretical model and empirical evidence from India. J. Dev. Econ. 88 (1), 103–119. Azam, M., 2012. The impact of Indian job guarantee scheme on labor market outcomes: evidence from a natural experiment. IZA Discussion Paper No. 6548. Bardhan, P., Mookherjee, D., 2000. Capture and governance at local and national levels. Am. Econ. Rev. 90 (2), 135–139. Bardhan, P., Mookherjee, D., 2010. Determinants of redistributive politics: an empirical analysis of land reforms in West Bengal, India. Am. Econ. Rev. 1572–1600. Besley, T., Coate, S., 1997. An economic model of representative democracy. Q. J. Econ. 112 (1), 85–114. Besley, T., Pande, R., Rahman, L., Rao, V., 2004. The politics of public goods provision: evidence from Indian local governments. J. Eur. Econ. Assoc. 2 (2–3), 416–426. Besley, T., Personn, T., Strum, D.M., 2010. Political competition, policy and growth: theory and evidence from the US. Rev. Econ. Stud. 77, 1329–1352. Cole, S., 2009. Fixing market failures or fixing elections? Agricultural credit in India. Am. Econ. J. Appl. Econ. 219–250.

30

B. Gupta, A. Mukhopadhyay / European Journal of Political Economy 41 (2016) 14–30

Cox, Gary W., Kousser, J.M., 1981. Turnout and rural corruption: New York as a test case. Am. J. Polit. Sci. 25 (4), 646–663. Dixit, A., Londregan, J., 1996. The determinants of success of special interests in redistributive policies. J. Polit. 58 (4), 1132–1155. Folke, O., 2014. Shades of brown and green: party effects in proportional election systems. J. Eur. Econ. Assoc. 12 (5), 1361–1395. Grossman, G., Helpman, E., 1996. Electoral competition and special interest politics. Rev. Econ. Stud. 63 (2), 265–286. Himanshu, Mukhopadhyay, A., Sharan, M., 2015. The national rural employment guarantee scheme in Rajasthan: rational funds and their allocation across villages (with Himashu and M. Sharan). Econ. Polit. Wkly. 50 (6), 52–60. Imbert, C., Papp, J., 2015. Labor market effects of social programs: evidence from India's employment guarantee. Am. Econ. J. Appl. Econ. 7 (2), 233–263. Klonner, S., Oldiges, C., 2014. Safety net for India's poor or waste of public funds? Poverty and welfare in the wake of the world's largest job guarantee program. AWI Discussion Paper Series No. 564. Mukhopadhyay, A., 2012. The political economy of implementing the national rural employment guarantee scheme in India. ESID Working Paper No. 15 (September). Novosad, P. and Asher, S. (2013). Politics and Local Economic Growth: Evidence from India, Mimeo. Ravi, S., Engler, M., 2015. Workfare as an effective way to fight poverty: the case of India's NREGS. World Dev. 67, 57–71 (March). Zimmerman, L. (2012), “ Jai Ho? The Impact of a Large Public Works Program on the Government's Election Performance in India.” Ann Arbor, MI: University of Michigan, Unpublished.