Political Organizations and Political Scope Research

1 Political Organizations and Political Scope 1 Introduction In politics, as in production, there are economies of scope. They arise when control over public funds is dispersed across elected officials at different levels of government. Using or misusing public funds therefore requires coordination between these officials. This is difficult by design. Power is dispersed to prevent any one politician from abusing her office for personal or political gain. But much as firms coordinate supply chains to reduce contracting frictions, a political organization may coordinate veto-holders towards circumventing these checks and balances. If it succeeds it can subvert government resources to finance its own operations in ways unavailable to opponents who control fewer offices. It may exploit its competitive advantage to ultimately deepen and broaden its scope of control. This paper studies how a political organization can exploit economies of scope to expand its influence, both vertically into new levels of government and horizontally into new political constituencies. The context of our study is West Bengal, an Indian state of 90 million residents. We study how its ruling party exploited state and local control over a massive make-work scheme, the National Rural Employment Guarantee Act (NREGA), to consolidate and grow its power. While the state government ultimately controls which local governments receive NREGA projects, the local councils control the allocation of NREGA jobs to villagers within their jurisdiction (called the gram panchayat, or panchayat for short). This dispersion of power creates complementarities between control of the state and local governments. The ruling party, which first took control of the state government in 2011, subsequently won absolute majorities on many local councils in 2013. We exploit quasi-random variation created by close contests in this local election to study the impact of complementary control. We show how the party used the program to expand “vertically” during the subsequent national elections in 2014, and “horizontally” in the following years by recruiting opposition councilors in local governments where it originally failed to win a majority. Our research design is based on a local institutional setup that is common worldwide, although a quirk relative to other Indian states. Local candidates in West Bengal campaign with explicit party affiliations to represent a single constituency in a local council. These elected councilors then choose a president who, with the consent of the majority, administers antipoverty 2 programs—essentially a Westminster model. Whichever party holds an absolute majority can appoint a president who is unconstrained by coalition partners, giving the party sole discretion over which individuals get NREGA payments. The majority party is determined by the collective outcomes of the individual council member races. The margins by which these seats are won or lost determine how close the state’s ruling party came to winning or losing the absolute majority. Using a multi-dimensional regression discontinuity design, we construct the univariate distance to the threshold of having an absolute majority. We causally identify how the ruling party shifts resources when it narrowly takes control of the council in closely contested elections. Crucially, we can distinguish how the party acts when it holds an absolute majority, and thus complete control, from the case where it is merely the largest party and holds the presidency subject to the veto of a coalition. We measure its actions by scraping and compiling administrative records for 300 million NREGA payments made to named individuals spread across an eligible pool of 11 million households. We combine these records with data on thousands of candidates contesting the 2013 local council elections, and polling station-level data on vote returns for the 2014 national election. We leverage this unique dataset to study program allocations across and within gram panchayats. We can observe whether it targets politically loyal constituencies, and whether it siphons funds to its own candidates and the opposition candidates it recruits. These data, together with our design, is what makes it possible to identify whether the mechanism driving our results is complementary political control. We find that panchayat councils barely controlled by the state ruling party receive 12 percent higher aggregate program allocations than areas where it barely misses the absolute majority. During the 2014 national election, panchayats narrowly controlled by the ruling party return an additional 2 percentage points for its parliamentary candidate. We rule out that the overallocation of funds is driven by co-partisan efficiencies, or that the voting effects are driven by the endorsement of the local president, by exploiting the difference between holding an absolute majority versus being the largest party. We construct a different running variable that measures how close the ruling party came to being the largest party in the local council. We show that the AITC is far more likely to hold the presidency when it becomes the largest party even when it does not hold an absolute majority. But we find no similar surge in NREGA funds, suggesting 3 the state government will not allocate excess funds to local councils where a coalition creates additional checks and balances. These areas also show no increased electoral returns, suggesting it is the funds and not a presidential endorsement that drives the electoral returns. But is the mechanism for this effect necessarily direct action by local officials on behalf of the national candidate? And if so, how does the party organization sustain their cooperation? We answer the first question by testing for evidence of politically-motivated targeting of the program in the lead-up to the national election. We show that within panchayats controlled by the ruling party, the areas with higher ruling party vote share in prior elections—and thus more ruling party voters—receive consistently higher payouts both in and after the national election. This result holds regardless of whether there is an elected ruling party councilor in the area. Since all areas within the panchayat are governed by the same council, this result is hard to rationalize except as a reward for supporters of the ruling party. Meanwhile, we find that demographics correlated with poverty are no more likely to get jobs when the ruling party takes control. These results suggest that local officials are distorting payments to muster support for national co-partisans. This result is more surprising than it seems because the national candidate has no direct means of rewarding local officials. How does the party organization buy their cooperation? We answer this question by studying payments made through the jobs scheme to the local politicians themselves. We find that panchayat councils controlled by the ruling party make excess payments to job cards registered to its local candidates. These payments are more than twice as large as the already magnified payments to typical households in these panchayats. Party candidates receive the same excess payments regardless whether they won or lost their own races. Since losing candidates hold no official position, they could not have authorized these payments personally. Instead, they could only have been made with the complicity of the party. That suggests the corruption is a feature of the party organization rather than an individual abuse of power. In addition to fixed payments, we also find suggestive evidence of incentive pay. In panchayats controlled by the ruling party its candidates receive a bonus payment in the 4 weeks immediately after polls close in the national election. This payment is systematically larger for candidates from areas that returned more votes for the ruling party. This effect cannot be explained by the results of earlier elections or the overall size of payments received by non-candidate house- 4 holds. And there is no similar pattern in a non-election year, suggesting the payment is directly linked to the election. It is consistent with a scheme that rewards local candidates for turning out voters for the national candidate. Taken together these results suggest it is the opportunity for self-dealing that aligns incentives and undermines checks and balances. Aside from expanding its vertical scope into new levels of government, a party organization can expand its horizontal scope to take control of new areas. We test for whether and how the party takes control of additional local councils by inducing officials aligned with the opposition to switch allegiance. We measure party-switching using the party affiliations of candidates in the subsequent 2018 local election. We infer that an individual registered under a different party in 2018 than in 2013 has switched parties. We show that the ruling party is far more likely to retain candidates than the opposition, and opposition candidates who switch to a new party are overwhelmingly switching to the ruling party. We find that the ruling party candidates who switch also typically receive less money through NREGA, whereas opposition candidates who switch—especially those who switch to the ruling party—are paid more. Among candidates from the two main opposition parties, this effect only occurs for those who won their election in 2013 and thus can help the ruling party gain majorities on councils where it did not win enough seats in 2013 to control the council. The ruling party on average sees no net recruitment in panchayats where it already holds a majority. In short, the party gains recruits only in places where it needs them, suggesting the effect is not merely a unilateral decision by opposition candidates to join the winning side. The net impact is to further consolidate the power of the ruling party. Although the 2014 national election came too soon for the consequences of this effort to be visible, by the 2016 state election these panchayats are returning votes for the ruling party at the same rate as panchayats where it did win the majority. The effort is so effective that the discontinuity in vote returns visible in 2014 has vanished by 2016, and it is largely because of improved returns in “control” panchayats. Taken together our results imply that the party organization plays a key coordinating role in the pursuit and abuse of power. We show that state officials channel funds to local officials to assist with the elections of national politicians. It is hard to imagine such trilateral coordination could arise from decentralized clientelism or a standard model of incumbency-driven targeted 5 spending. These models would likewise fail to account for the organization’s role in exploiting the complementarities created by fiscal checks and balances. The arrangement is unavailable to other parties even if they do hold absolute majorities on the local council because the state ruling party can effectively starve them of projects. Demonstrating the role of political complementarities is our key contribution. It has the biggest implications for the empirical literature on politically targeted spending.1 Much of this work has focused on the incentives of incumbent candidates and parties to win their own election by targeting government resources to either swing or core constituencies (Bardhan and Mookherjee, 2006; Golden and Picci, 2008; Cole, 2009; Baskaran et al., 2015). Though the actors in these studies are often parties, the importance of the party organization itself is not central to their hypothesis. The predictions would be similar for a single elected official contesting her own election, and thus the targeting is based on the political lean of a constituency rather than the allegiances of local officials. The exceptions are a few studies (e.g. Brollo and Nannicini, 2012; Curto-Grau et al., 2018) that study whether state or national resources are targeted to co-aligned local officials. The specific resources are generally infrastructure or earmarked funds that are discretionary at the higher level but not locally, making the mechanism entirely different from the one we study. The mechanism proposed in these prior studies is that higher level officials help local officials with their own election, who may return the favor in future elections (as in Persico et al., 2011).2 Our results cannot be driven by this mechanism because both state and local officials are working on behalf of a national candidate. Our work is unique in studying resources that are jointly controlled by two levels of government, making complementary control the key mechanism. Our work is also unique in having an institutional context and a massive individuallevel dataset that lets us explore multiple aspects of the mechanism, from multi-level targeting to incentive-based payments. The mechanism we propose is potentially more pernicious than those of earlier models because it is inherently dynamic. An incumbent seeking reelection can at worst tilt the odds to keep herself in power. An organization that gains control of a comple1 2 For an overview of the literature on distributive politics see Golden and Min (2013). Asher and Novosad (2017) show evidence of a different kind of favoritism where ruling party state legislators are able to influence bureaucrats to fast-track permits and cut red tape. This is a substantially different mechanism from any of these other papers (and our own). 6 mentary office can unlock additional government resources it can use to not only hold power but to expand its power. Our work is also related to the theoretical literature on political parties. Most prior work has modeled parties as either bundles of policies (Downs, 1957; Aldrich, 1995; Levy, 2004; Anesi et al., 2009) or vehicles that arise to mobilize and advocate for segments of voters united by attributes, ideology or economic interests (Lipset and Rokkan, 1967; Stokes, 2003; Chandra and Boulet, 2012; Muirhead and Rosenblum, 2020). Our approach instead draws on organizational economics, which posits that firms exist to resolve contracting problems that arise when there are complementarities in production (Coase, 1937; Gibbons and Roberts, 2012). To our knowledge this is the first paper to apply the framework to understand how party organizations exploit complementarities that arise from holding different political offices. Finally, our work also draws on the literature on clientelism. Most of this literature focuses on the bilateral relationship between a patron and his supporters (see Hicken, 2011, for a review). A traditional model of clientelism envisions little role for an organization. A relatively new subset of the literature studies how several puzzling strategies followed by patrons can be explained by their need to hire intermediary “brokers,” which comes closer to our work by expanding the set of agents (Stokes et al., 2013). Our micro-level targeting results could be consistent with clientelism and brokers, but the precise mechanism of the targeting is not the focus of our study. We focus on the role of the organization in circumventing checks and balances that would otherwise prevent accessing government resources to finance clientelism (or any form of targeted spending). The coordination between state and local officials on behalf of a national candidate provides clear evidence that a political organization can be more than an association of patrons using similar methods of brokerage and patronage. 2 2.1 Model Setup There is a state politician who controls the aggregate benefits (“projects”) allocated to each locale k. She allocates projects to maximize both social welfare and the vote return from targeting 7 funds: max " X {Bk } # Wk (Bk ) + α X Vk (Bk ) (1) k k subject to X (2) Bk = B̄ k The term Wk (Bk ) is the (indirect) welfare function for the allocation of projects to locale k, and Vk (Bk ) the vote return. The parameter α ≥ 0 captures the relative importance of generating votes in the election. Note that Vk (Bk ) is not necessarily votes in the state politician’s own election, merely votes for a particular candidate in some election. It is possible that α = 0 implying the state politician does not care about this election. Locale k is governed by a local politician who decides which individuals receive benefits (“jobs”) from the aggregate allocation Bk . He allocates benefits to maximize both the welfare of households under his jurisdiction and an electoral objective that may or may not coincide with that of the state politician: " max {bi } # X ui (bi ) + βk X i vi (bi ) (3) i subject to X bi = Bk (4) i The term ui (bi ) gives the indirect utility to household i from receiving bi jobs. The term vi (bi ) gives the expected vote return from targeting i. The term βk , as we shall explain, captures the extent that the electoral objectives of the two politicians coincide. Let {b∗i (Bk ; βk )} be the set of benefits that maximizes (3) subject to (4). Impose the following conditions to relate Equation 1 to Equation 3: Wk (Bk ) = X Vk (Bk ) = X ui (b∗i (Bk ; βk )) (5) vi (b∗i (Bk ; βk )) (6) i i These conditions imply that the state politician’s payoff from giving additional projects to k de- 8 pend on how the local politician allocates benefits. The key question is whether α and βk have the same sign, which implies the political objectives of the politicians are aligned. 2.2 Distinguishing Political Complementarities from Existing Models This general model nests several of the most common models in the literature. Targeted Spending/Pork-Barrel Politics: Assume βk = 0, meaning the local politician simply allocates benefits to maximize local welfare, and that vi (bi ) = λk ui (bi ) for some constant λk that captures the electoral significance of k in the election. Under this assumption the electoral benefit depends solely on total benefits to the locale. If this is the state politician’s own re-election, meaning α is large and positive, then the state politician allocates disproportionate projects to locales where λk is relatively high. If this is not the state politician’s own re-election, meaning α = 0, then there is no political misallocation. Crucially, the misallocation does not depend on who runs the local government. The model of Political Fiscal Cycles is a special case where α > 0 in an election year and α = 0 in a non-election year. Decentralized Clientelism, Vote- or Turnout-Buying: Assume Bk is fixed (say, by law) and only the local politician’s decision matters. In the standard model of clientelism βk is large and positive. The local politician allocates benefits to individual voters in return for their allegiance, which includes voting for the politician during the election. The model of vote-buying is similar except that βk is large and positive only during an election year, and zero otherwise. Partisan Favoritism: Define an indicator ωk that equals 1 if the top-ranking official in k is a member of the state politician’s party (or faction, or ethnicity). One version of this theory assumes pure in-group preference, which can be represented by modifying Equation 5 to P Wk (Bk ) = ωk i ui (b∗i (Bk ; βk )). Another version assumes the state politician wants to elevate the standing of her factional allies (Persico et al., 2011), which can be represented by modfying P Equation 6 to Vk (Bk ) = ωk βk i vi (b∗i (Bk ; βk )). In both cases it is assumed that 1) α = βk = 0 if neither state nor local politician is up for election and 2) ωk is defined based on the identity of the top-ranking official rather than on control of local resources. These are two of the crucial distinctions between partisan favoritism and complementary control, and are among the tests we describe in the next two sections. 9 2.3 Complementary Control Misaligned Incentives—Ambition Checks Ambition: Political complementarities arise in this model because the state and local politician act as checks on one another. To see why, consider the Madisonian ideal where βk = 0 whenever α > 0, and vice-versa. If βk = 0 then the targeting of benefits {b∗i (Bk ; βk )} chosen by the local politician will maximize local welfare Wk . To the extent that vi differs from ui , the political impact will be blunted because b∗i (Bk ; 0) will not be targeted to maximize Vk . Knowing that jobs will not be targeted to maximize political returns, the state politician has less incentive to allocate projects for political gain (see Appendix A.1.1). Conversely if βk > 0 and α = 0, the state politician will allocate projects to maximize welfare conditional on the local politician’s targeting. To the extent that the local politician mis-targets benefits for political reasons, it will decrease the aggregate welfare of households in the locale Wk (Bk ). The state politician will respond by allocating fewer benefits. Her response effectively disciplines the local politician. If he mis-targets too egregiously his budget will be cut (see Appendix A.1.2). Aligned Incentives—The Role of the Organization: The party organization exists to align the incentives of these politicians. It first ensures that α > 0 in any election contested by a member of the organization—in particular, a national election contested by members of the ruling party. Suppose the national politician can promise to deliver federal transfers to the state politician (by increasing the size of B̄). The role of the organization is to enforce this unofficial contract (say, by threatening to expel the national politician if he reneges). Given this credible promise, the state politician is willing to over-allocate projects to locales important to the national politician’s election. Given that α > 0, the organization will likewise ensure that βk > 0 in any local government controlled by members of its organization. To be precise, there are three cases: • Ruling party organization has complete control of local government: βk = α > 0 • Opposition party organization has complete control of local government: βk = −α < 0 • Ruling party organization governs local government in coalition: βk = 0 In Case 1, objectives are fully aligned and the ruling party can exploit complementary con- 10 trol. Jobs that come with any project allocated to k will be targeted to generate electoral returns as well as household welfare. In Case 2, objectives are opposed. Jobs that come with any project will be targeted to reduce electoral returns for the ruling party candidate. In Case 3, the checks and balances hold because the coalition can bring down the ruling party if it engages in political targeting. The model then makes several predictions: Prediction 1 (Complementary Control) 1. The state politician allocates more projects (ag- gregate jobs) to local governments controlled by the ruling party 2. The ruling party’s national candidate reaps more votes in locales controlled by the ruling party 3. These effects are either absent or muted in areas where the ruling party governs but does not have complete control It is crucial to note what makes these predictions distinct from the usual models of porkbarrel politics or co-partisan favoritism. Misallocating funds to co-partisan governments is not the same as geographically targeting funds. The prediction is not driven by the electoral significance of the locale but by control of the local council. Co-partisan favoritism, on the other hand, predicts more projects will be allocated to a locale where the leader is a member of the ruling party, regardless of whether there is an outside check on his control. Finally, neither model predicts that state or local politicians will distort allocations for the sake of a third national politician who has no direct authority over either. 2.4 Mechanisms: Targeting and Financial Incentives The key mechanism behind Prediction 1 is that locales controlled by the party get extra projects with the expectation that the local politician targets politically valuable households. This mechanism is another key distinction from standard models of pork-barrel politics and co-partisan favoritism. To illustrate this mechanism, impose the following assumption: vi (bi ) = λk γi ui (bi ) (7) 11 The difference between this assumption and the one imposed for simple pork-barrel politics (in Section 2.2) is the term γi , which captures the electoral significance of the household as distinct from the locale. This assumption implies there are specific households within the locale who are more likely to vote for the ruling party if they receive a job (those with high values of γi ). It yields the following prediction: Prediction 2 (Within-Locale Targeting) Within a locale controlled by the ruling party, benefits will be targeted to politically significant households or areas. Our model does not make a specific prediction about whether households with high values of γi are core supporters, as would be true for a model of brokers or turnout-buying, or swing voters, as in a model of vote-buying. It also does not require or rule out conditionality, as in a model of clientelism. It only predicts some within-locale targeting based on political preferences. Hence our model is not mutually exclusive with any theory of micro-level targeting. It embeds them within a larger system to explain how aggregate targeting arises from the anticipation of subsequent micro-targeting. The other key mechanism in our model is that the organization actively motivates the local politician to engage in targeting. Section 2.3 simply imposed that βk = α > 0, which would hold if the organization punishes the local politician for not targeting benefits correctly. But this is only possible if the organization observes {γi }, which may not hold if the local politician has private information about the electoral significance of each household. Then the accuracy of the local politician’s targeting is essentially a hidden action. The theory of organizational economics predicts that a firm can alleviate moral hazard by offering incentive pay: a wage that is a function of the outcome of the worker’s effort. If the local politician is allowed to divert funds to himself— that is, engage in self-dealing—the organization can use diverted funds to sustain cooperation. Amend Equation 3 to " max {bi } # X ui (bi ) + uL (wk ) (8) i where uL (wk ) is the local politician’s utility from diverted funds wk . If u0L > 0, u00L < 0 then the optimal contract will include both a constant payment—a wage—and a payment increasing in the electoral return Vk —a performance bonus (Hölmstrom, 1979). 12 Prediction 3 (Compensation) The local politician will receive fixed and incentive payments through diverted funds. There is one small but crucial nuance in taking this prediction to the data. The formal authority to target benefits lies with the local council (as long as the ruling party holds a majority). But individual local council candidates, by virtue of their presence in a district, are likely the ones who observe and report {γi }, and the ones who make any complementary electioneering (e.g. telling voters that they got the job solely through the party’s intercession). Since the individual candidates face the moral hazard problem, we would expect them to receive the fixed payments, and the incentive payments based on returns from their own area within the locale.3 2.5 Long-Run: Horizontal Expansion Section 2.3 predicts the party organization will align state and local incentives to assist a national politician: vertical expansion. But it also has an incentive to expand horizontally to take control of more locales. It has this power because the state politician can both starve oppositioncontrolled locales of funds and offer payments to local politicians who switch. To be precise, suppose that each local government contains several local councilors. A party organization (either that of the ruling party or the opposition) gains control of the local council if it has a majority. A councilor will want to switch parties if Equation 8 is made higher if they switch than it would be otherwise. The ruling party (opposition) will accept the new recruit if Equation 1 is made strictly higher (lower) by the switch. Since the ruling party controls the allocation of projects, all local councilors will want to either remain with or switch to the ruling party. But the ruling party will only accept new recruits in areas where it does not already control the local government. Prediction 4 (Recruitment) The party recruits local councilors in locales that it currently does not control (βk ≤ 0). It will not recruit in locales it already controls (βk > 0). 3 This prediction holds regardless of whether the candidate holds a seat on the council, which distinguishes it from the prediction of the next section.