Bypassing Your Enemy: Distributive Politics, Credit. Claiming, and Non-State Organizations in Brazil. Natália S. Bueno

Transcription

1 Bypassing Your Enemy: Distributive Politics, Credit Claiming, and Non-State Organizations in Brazil Natália S. Bueno 1

2 Abstract How do incumbents prevent the opposition from hijacking credit for spending? Existing explanations argue that central authorities target co-partisans in lower tiers of government to reward allies and punish opponents. Yet this is an incomplete depiction of incumbents strategies because it ignores the range of available instruments. I argue that allocating money to non-state welfare providers allows officials to bypass the opposition and reduce credit hijacking. Central incumbents can thereby provide benefits to opposition-ruled municipalities while reducing political costs. Using regression-discontinuity and difference-indifferences designs with data from Brazil, I build on previous findings that co-partisanship boosts transfers to mayors. Nevertheless, I provide evidence that the election of an opposition mayor causes the central government to shift resources to non-state organizations. Drawing on fieldwork and web-scraped news reports to depict officials credit-claiming behavior, I support my claim that opposition mayors do not hijack credit from spending through non-state organizations. Keywords: Credit Claiming, Intergovernmental Transfers, Non-State Welfare Provision, Regression- Discontinuity Design, Difference-in-Differences, Brazil Word Count: 8,477 (in text) + 1,587 (in captions and notes) 2

3 Political authorities allocate resources to build and cultivate political support. However, in multi-level systems, central officials face a challenge: voters often have difficulty distinguishing between levels of government and correctly assigning responsibility for spending. This makes credit hijacking potentially harmful to the incumbent who spends resources in jurisdictions locally ruled by an opposition incumbent. Previous research has therefore argued that partisan alignment between lower and higher level officials is a key determinant of the tactical allocation of resources, such as transfers between federal governments and municipalities (Arulampalam et al., 2009; Brollo and Nannicini, 2012; Stokes et al., 2013). Yet shifting resources away from opposition-governed municipalities has clear costs for incumbents, who may wish to provide benefits to voters in those localities. Incumbent politicians may care about the welfare of voters in opposition municipalities and they may wish to extract political advantage from resource allocations. How then can they solve the problem of credit hijacking by political opponents at lower levels of government? In this paper, I study an alternate distributive strategy that allows incumbents to minimize the costs and maximize the benefits of spending in opposition-controlled jurisdictions: funding non-state welfare providers. I argue that incumbents can strategically allocate resources to non-state organizations to minimize the risk of credit hijacking by political foes and reduce the political costs of allotting goods to voters under opposition control. In this way, non-state organizations provide a path for the central incumbent to provide goods to voters in districts ruled by the opposition while bypassing the opposition incumbents. Several attributes make non-state welfare providers attractive candidates for tactical distribution of public monies. In many countries, private non-state organizations receive public funding to provide a wide array of welfare services to citizens, such as day cares, health clin- 3

4 ics, and worker training centers. Like subnational incumbents, these organizations receive government funds to provide services, but, in clear contrast to local officials, they do not need to amass voter support. Additionally, these organizations are not organs of the state and, while they may have partisan affinities and affiliations, the entities I study in this paper are not political parties. These features make it more difficult for any political incumbent to claim credit for the activities of non-state welfare providers. Thus, whereas transfers to municipalities allow mayors to claim credit for spending, allocations to non-state organizations minimize the appropriation of credit by local officials. Transfers to non-state organizations and to incumbents are thus strategic substitutes, and the partisan alignment of local officials determines which type of transfer is most politically attractive to central incumbents. In this way, these transfers are not mutually exclusive; officials use both strategically allocating through mayors in some municipalities and through non-state providers in others. This paper tests these claims using novel data and a distinctive use of extensive quantitative and qualitative data. I present evidence from Brazil, which contains fine-grained information about the allocation of transfer payments to non-state organizations and mayors. I use a closerace regression-discontinuity design to identify the effect of partisan mayoral alignment on transfers. This design yields two sets of municipalities that plausibly differ only in the presence or absence of an aligned mayor; that is, whether the mayor and the president are co-partisans or not. My argument implies that the effect of partisan alignment is positive for transfers to mayors and negative for transfers to non-state organizations that is, election of a nonaligned mayor causes central incumbents to expend more funding on non-state providers in that municipality, thereby bypassing the mayor. I compare the effects of alignment on each funding outcome in order to examine the argument that central incumbents use both types of 4

5 transfers in tandem to maximize their parties electoral prospects. I find results consistent with the bypassing hypothesis: among municipalities that elected mayors by a narrow margin, the central government sends more money to non-state welfare projects in municipalities ruled by unaligned mayors those who do not belong to the president s party than to non-state providers in municipalities governed by aligned, copartisan mayors. Substantively, the effect of alignment is large for transfers to non-state service providers and the redistributive consequences are important: the average effect of alignment on transfers to non-state organizations is similar in size to the mean per capita expenditures with the flagship social program Bolsa Família. I also find that partisan alignment causes an increase in intergovernmental transfers to mayors, which is consistent with previous studies (Brollo and Nannicini, 2012; Ravanilla, 2014; Berry, Burden and Howell, 2010; Timmons and Broid, 2013; Barone, 2014). Importantly, my findings also support the claim that central incumbents use these transfers as strategic substitutes: the sum of the absolute values of the effects of alignment on the two types of transfers is larger than each effect separately. Finally, I present detailed evidence to support my contention about the political usefulness of non-state organizations namely, that they make it difficult for opposition mayors to hijack credit. First, using qualitative evidence from over sixty interviews with managers of non-state organizations, bureaucrats, and legislators, I argue that mayors cannot interfere with the resources sent to these organizations; that is, mayors do not control the flow of money between the federal government and non-state providers. Second, I analyze qualitative evidence from news outlets to examine how mayors claim credit. I look at events, such as ribbon-cutting ceremonies in non-state organizations, where the mayor would have a chance to claim credit. My fieldwork motivated additional data collection conducted by web-scraping newspaper articles from local media outlets in municipalities in the regression-discontinuity study group 5

6 to further document credit-claiming activity. Third, I discuss a possible objection to my argument; that these non-state organizations could be clientelistic or corrupt organizations that actually channel resources to the president s party where it lost the mayoral election. I present evidence from a dataset on contract irregularities and government reports, about the type of service provided by these organizations, and novel data on partisan affiliation of organizations board members to make the case that neither clientelism nor corruption serve as alternative explanations to the bypassing strategy. This article makes several contributions to the literature on distributive politics. First, it presents a novel strategy by central incumbents to avoid credit hijacking and disputes over credit in scenarios of low clarity of responsibility an issue discussed by many studies on distributive politics and federalism (Arceneaux, 2005; Ebeid and Rodden, 2006; León, 2010). Second, it uncovers political motives behind the distribution of public funds to non-state organizations, which are crucial service providers in both developing and developed countries. Third, it provides an explanation based the several advantages offered by these organizations to reduce the cost to the central government of investing in voters governed by the opposition. Finally, my findings counter explanations of non-state welfare provision as a function of government failure (Stiglitz, 1991; Bardhan and Mookherjee, 2006). These explanations often assume that the government is constrained in its capacity to provide services to citizens and non-state organizations to compensate for these failures. I show that partisan alignment independent of other constraints causes the central government to sidestep opposition with resources to non-state providers. I return to the broader implications for the politics of nonstate welfare provision in the last section (Thachil, 2014; Cammett and Maclean, 2014). 6

7 1 Non-State Organizations and Transfers in Multi-Level Systems Political elites who wish to tactically allocate resources to voters face a challenge: they typically rely on agents, either brokers or subnational incumbents, to distribute benefits or provide services. However, these agents may have different objectives from leaders (Stokes et al., 2013). This potential conflict may be exacerbated when these agents are not aligned with the central incumbent. Central authorities thus target aligned subnational incumbents in preference to opposition officials to reduce dissent and maximize both their own chances of re-election and their party s interests in subnational elections. Central officials face an additional hurdle when providing goods to voters: the latter may not accurately identify the central government as the source of transfers. Hence, they may give credit to both upper and lower tiers of government (Arulampalam et al., 2009; Albert Sole-Olle, 2008; Pop-Eleches and Pop-Eleches, 2012). Brollo and Nannicini (2012) contend that central incumbents wish to minimize the costs of credit-claiming by opposition officials and predict that central authorities will penalize unaligned rulers a strategy they call tying your enemy s hands. The incumbent, however, may incur costs by rewarding allies and punishing opposition. Cutting opposition municipalities off from federal resources altogether or to a large extent may backfire: voter resentment may hurt both the local and central incumbents popularity. Citizens attribute changes in their welfare improvements as well as deteriorations to both central authorities and local ones (Malhotra and Kuo, 2008; Renno, 2010). Thus, targeting aligned subnational incumbents also creates potentially harmful consequences for the central official. In summary, while most studies on transfers focus on the consequences of credit 7

8 leakages to the opposition for the incumbent and her party s electoral prospects, they neglect to account for the costs to central incumbents of punishing local opposition. The use of non-state organizations to administer the distribution of benefits reduces voters distress over the loss of valued service provision and thus reduces the potential damage that punishing opposition mayors could cause to both incumbents. However, these transfers to nonstate providers still diminish local officials ability to steal credit, and central governments may also claim credit for programs dispensed through these organizations. While transfers to non-state providers reduce the cost of sending money to unaligned municipalities, it may not be easy for central incumbents to claim credit even though, as I discuss in section 3, central incumbents do attempt to claim credit. 1 Transfers to aligned mayors may yield more electoral benefits than resources given to NSPs in municipalities run by aligned mayors. Hence, it is more profitable to send money to aligned mayors than to NSPs. This prediction is consistent with previous findings on intergovernmental transfers, in which central officials prioritize resources to aligned incumbents. In summary, the central incumbent wants to maximize her political benefits (by claiming credit for her and her political allies) and lower political costs (from credit hijacking by the opposition or blame for underprovision of goods). The type of organizations I investigate possess two distinctive qualities: they are not organs of the state 2 and they are not elected. Cammett and Maclean (2014, p. 6) define them as all providers outside of the public sector, including charitable and for-profit institutions, 1 It would be difficult to argue that central officials are simply wasting their time by taking part in credit claiming events. It is plausible, however, that these events with non-state organizations are less effective relatively to similar events with mayors in aligned places. 2 This argument does not rule out that these organizations may have partisan preferences and affiliations, as I discuss in section 3. 8

9 as well as domestic and international actors. I focus here on non-state, usually not-for-profit welfare providers that receive public funding to deliver social welfare to citizens; these include church-run health clinics, community schools, daycares and after-school programs run by neighborhood associations, artisan cooperatives, and other service providers. Contrary to notions of autonomous, self-financed, and self-governed entities, these organizations often rely on public funds and subsidies (Rose-Ackerman, 1996). A comparative study by Salamon, Sokolowski and List (2003) of 35 countries that found that about 35% of all non-state organizations revenue comes from domestic public sources, even though this share ranges widely, from a meager 5% in Kenya and the Philippines to bounteous 77% in Belgium and Ireland, although in countries such as Kenya international donors could play an important role. The proportion that comes from the government s purse is even higher and accounts for the plurality of the revenue if only health and social service organizations are considered. 3 Researchers have also investigated the role of these organizations in providing critical welfare functions in Chile, India, the United States, among others (Salamon, Sokolowski and List, 2003; Clemens and Guthrie, 2011; Hungerman, 2007). Allard (2009), for instance, shows that the expansion of the nonprofit section is a function of public funding. Table 1 presents data on the share of non-state welfare provision in developing countries: more than half of physicians work in NSPs and about 28% of all hospital beds are in non-state organizations. The provision of health in developing countries relies heavily on these type of providers, even if the funding 3 Moreover, these estimates do typically not include other types of indirect benefits that states often award non-state providers, such as tax breaks to donors and the organizations themselves. States also allow non-state providers to use government property, such as buildings and land, to carry out public-oriented activities. 9

10 is public or from foreign donors. 4 Table 1: Non-State Welfare Provision in Developing Countries Proportion of physicians Proportion of hospital who work for non-state providers beds that are in non-state organizations All Middle East Asia Latin America and Caribbean Africa Source: Data for Decision Making Project (Hanson and Berman, 1995). Notes: Means are weighted by country population. Providers who operate officially and on a full-time basis in the non-state organizations. Number of beds in hospitals (not including clinics, nursing homes, etc.) More importantly, non-state organizations differ critically from elected local officials such as mayors in that they lack subnational incumbents electoral incentives to expropriate credit and especially to turn credit to their electoral advantage. It also is especially difficult for local incumbents in particular the opposition to hijack political payoff from these organizations activities. These providers are insulated from interference by the local opposition, as they typically do not rely on local incumbents mediation or approval to receive public funds from the central government. The combination of these key distinctions between local officials and private providers of social services has important implications for distributive politics. While credit claiming, avoiding credit hijacking and blame for underprovision of goods are 4 Data on health provision by non-state organizations is more plentiful in cross-country datasets than on other services, but my argument also applies to other types of services. Unfortunately, I am not aware of any cross-country dataset on public goods provision by non-state organizations with more recent information. 10

11 important motivations for the central incumbent, what other factors influence the incumbent s decision of allocations? After all, if these were the only determinants, we would never observe central incumbents funding unaligned mayors, supporting NSPs in aligned municipalities, failing to reward allied incumbents or bypassing opposition. Yet we often observe such behavior (Golden and Min, 2013). Incumbents have many contending objectives and scarce resources. In multi-tier systems, these transfers play different roles in fulling these goals. Studies suggest a host of additional strategic motivations: incumbents may wish to use these transfers to gain legislative support from opposition (Pereira and Mueller, 2002), to engage in party building (Bawn and Nunes, 2013), mobilize swing and core voters (Cox and McCubbins, 1986), or trade transfers for campaign contributions (Samuels, 2002; Boas, Hidalgo and Richardson, 2014). Central incumbents may wish to target opposition municipalities for benevolent reasons (to provide goods for the most needed areas), and for ideological reasons (to implement a preferred policy even at the risk of losing credit). If there were not a mix of motivations, we would not observe such a variation in transfers. Thus, mayoral transfers to opposition mayors or NSP transfers in aligned municipalities are not necessarily an strategic error, but they motivated by other factors. However, it is not an easy task to pin down the extent to which the central government is driven by each particular motivation. More importantly, these motivations and a mix of them do not exclude the effect of partisan alignment between levels of government on transfers, the underlying argument that NSP transfers reduce opposition mayors ability to hijack credit, as an additional cause of central government s spending decisions. In the next section, I discuss evidence from Brazil. 11

12 1.1 Non-State Organizations and Transfers: Evidence from Brazil Brazil provides an excellent setting in which to test the bypassing hypothesis. Intergovernmental transfers are an important part of Brazil s federal system subnational incumbents, while crucial to service provision, are heavily dependent on revenue from transfers. Additionally, I was able to collect new, unique data to test my predictions about transfers to non-state organizations 5 and to replicate findings on transfers to Brazilian mayors with data different 6 from that used in earlier studies. The executive sends the budget bill to congress. And, while legislators have several opportunities to influence it, after it is signed into law, the Executive has decision-making power over it (Boas, Hidalgo and Richardson, 2014). The central government faces a choice: it may transfer to state governments, municipal governments, non-state providers or spend resources through their own federal agencies. I focus on discretionary transfers (convênios) to mayors and non-state providers 7 Both transfers to mayors and to NSPs fund collective goods locally nonexcludable and nonrival, and not targeted to individuals. These transfers have no spending 5 I collected data from SigaBrasil and SICONV. Earlier versions of this paper used data of discretionary transfers that were effected through legislative budgetary amendments. This version uses data from all discretionary transfers, which were not available at the time I began this project. All values were converted to December This motivated the additional supporting data and a couple of changes in coding decisions. All details can be found the next sections and in the Online Appendix. 6 I use data on all discretionary transfers to mayors, while Brollo and Nannicini use data from discretionary transfers devoted to infrastructure projects. 7 I exclude governor s from this analysis because I am looking at the effect of municipal partisan alignment on transfers and many transfers to governors cover more than one municipality. 12

13 limits and they are disbursed at the executive s discretion during the fiscal year (Limongi and Figueiredo, 2008). Non-state organizations in Brazil usually receive funds to build and expand organizations facilities, organize events, provide social welfare (such as homes for the elderly), or train workers. The plurality of resources (22.7%) is sent to health non-state organizations and the second largest share (19.7%) are sent to organizations that provide social welfare; e.g., to build or renovate hospitals and clinics, or to create a program to provide work training for homeless people. 8. Finally, it is important to highlight that non-state welfare organizations are important providers of services in Brazil: these non-state clinics and hospitals are responsible for 41% of all hospital admissions in Brazil s public health system. 9 Mayors are in charge of providing many public goods basic health, education, and local infra-structure and mayoral transfers are often used to invest in infra-structure (Brollo and Nannicini, 2012). Although mayors and non-state providers clearly have different roles in provision of goods, the type of good they provide often overlap. Mayors themselves contract non-state providers to provide public goods - as Appendix Figure S.A.1 shows, mayors are using non-state providers at an increasing rate. 10 Moreover, many of these federal transfers refer to programs that can be administered either through non-state providers or mayors. Furthermore, these discretionary transfers to non-state organizations are detailed: they 8 Many of these transfers do not allow these resources to be spent on personnel and management. These are considered to be the organizations responsibility (contra-partida) in the agreement with the federal government. 9 See SI D However, under most circumstances, mayor cannot use these discretionary funds to subcontract non-state providers. 13

14 specify in which district the non-state organization will carry out a project and they contain details about the project. Thus non-state providers do not typically have discretion to decide where the project will be executed. An example, based on interviews 11 illustrates how the federal government sends money to NSPs: a non-state hospital run by an evangelical church received federal funds to create a cancer care unit. The transfer named the hospital that would receive the funds, and specified that the money would be used to build a specialized health care facility. 12 Table 2 presents suggestive evidence that mayoral alignment plays a role in how the president decides to target her transfers. Municipalities governed by an opposition party receive about 1.5 times more NSP transfers, on average, than municipalities whose mayor belongs to the Workers Party. Moreover, aligned mayors receive about 1.4 times more mayoral transfers, on average, than unaligned mayors. The pattern is similar for transfers to municipalities governed by coalition parties and parties outside of the governing federal coalition, but the differences are smaller. 13 Overall, these are relevant differences, and they suggest a contrast between the tactical logic of intergovernmental transfers and allocation to non-state organizations. Yet we also observe in Table 2 that central incumbents sending money to unaligned mayors, transferring public monies to providers in aligned municipalities as well as not rewarding allied 11 Interviews HE-2013a and HE-2013b. 12 In this particular case, the amount of money needed for the reform was complemented by a legislative amendment, but most transfers are not coupled with amendments. 13 I follow Brollo et al. s defintion of coalition parties for period between 2003 and 2008 in 2009, 2010, and 2011, I define that coalition parties have a cabinet seat in the beggining of Lula s second term (Boas, Hidalgo and Richardson, 2014). 14

15 Table 2: Partisan Alignment and Transfers, in reais per capita Mean NSP Transfers Mean Mayoral Transfers Number of Municipalities Workers Party ,080 Non-Workers Party ,288 Federal Coalition Party ,864 Non Federal Coalition Party ,069 mayors or not sending resources to NSPs in unaligned municipalities. 14 Moreover, mayoral transfers (either to aligned or unaligned) are much larger than NSP transfers on average. Why does it happen at all? Earlier discussion has already point out a few possibilities the central incumbent also use these transfers with many objectives in mind, and other factors, not only partisan alignment, influence her choice. Moreover, non-state transfers occur at a lower rate than mayoral transfers for two main reasons. While central incumbents can always target mayors, non-state providers are not available in every municipality 15 Additionally, not all government programs can be administered through a non-state entity, albeit many, in particular social welfare, culture, and health, are managed by these organizations. More importantly, these reasons do not challenge my argument, since I am interested in the effect of partisan alignment on NSP and mayoral 14 About 38% of aligned municipalities did not receive any mayoral transfers and about 55% of unaligned municipalities did not receive any NSP transfers. Moreover, incumbents send more money to unaligned mayors than to NSPs in either aligned and unaligned. Even though these discretionary transfers are a small slice of the federal budget, the central government cannot meet all demand without increasing expenditures, taxation, or shifting resources from other parts of the budget (Brollo and Nannicini, 2012, p.745). 15 The central incumbent may always contract a provider from a different municipality to provide goods, but this is not always cost-efficient or even possible for every locality. 15

16 transfers as an additional cause of transfers. Of course, the evidence in Table 2 clearly does not identify a causal effect of partisan alignment. Municipalities governed by an aligned mayor may differ in many ways that influence the mix of transfers to non-state organizations and mayors. As I discuss in the next section, confounding factors are likely. Thus, I present a research design to identify the effect of mayoral alignment. 2 Testing the Bypassing Strategy I expect partisan alignment between the mayor and the president to cause a decrease in NSP transfers and an increase in mayoral transfers. However, pinning down the effect of mayoral alignment on transfers is challenging. Chief among these challenges is that of parsing out the effect of mayoral partisanship from observable and unobservable factors associated with both transfers and mayoral alignment. For instance, municipalities with dense organizational civil society networks might be more likely to be targeted by the Workers Party, and might also be more likely to receive transfers to non-state organizations since the Workers Party has held the presidency during most of the period I study (Zucco and Samuels, forthcoming). Also, unaligned municipalities might lack the state capacity to provide some types of services, and thus be more likely to provide these services through non-state organizations. The list of plausible confounders and suggestions on how to adjust for them is at least as large as the number of alternative theories about the influence of mayoral alignment on transfers. My strategy for identifying the effect of mayoral alignment is to compare winners and losers of close mayoral elections, conditional on whether the winner or loser is a member of the 16

17 president s party. 16 Even though many researchers have studied transfers before, this is the first systematic data set with information on federal transfers to non-state organizations I collected data from 2003 to The period I analyze coincides with the period in which the Workers Party has held the presidency, which does not allow for variation in partisan alignment. This is an important concern which I address in section 3, where I discuss state level evidence. The president s party in Brazil governs in a coalition and other parties have the opportunity to influence the targeting of resources to municipalities and NSPs because they control many ministries. However, I expect the president to prioritize her own party s electoral chances over other parties in the coalition (Boas, Hidalgo and Richardson, 2014). In Brazil s multi-party system, federal allies often are local foes, which makes measures of alignment that account for both federal and local allegiances far from perfect. While many studies have shown that pork, particularly through legislative amendments, is an important glue that keeps the governing coalition together (Pereira and Mueller, 2002, p ) and rewarding mayors who belong to a party that is a member of the federal coalition may help to sustain the federal coalition, it also hurts the president s party local ambitions. This does not mean that coalition does not influence transfers, but alignment with the president s party arguably has a stronger effect on these transfers. Moreover, there are other spoils of office that can be used to keep the coalition together (e.g., cabinet positions, appointments for high-paying federal positions and at state- 16 I follow a strategy similar to that of Brollo and Nannicini (2012). 17 Currently, there are no reliable data available on transfers (convênios) to non-state organizations before Also, because of changes in SICONV in 2011, I have not been able to collect more recent data. 17

18 owned companies) as well as compromises in policy positions. 18 Equations (1) and (2) represent the average effect of exposure to treatment; that is, mayoral alignment. Municipalities in which the mayor belongs to the president s party are aligned and municipalities where the mayor belongs to any other party are unaligned. Equation (1) represents the average treatment effect, τ RD,NSP, of the aligned candidate s victory on NSP transfers. Let Y i,t+1,nsp (1) and Y i,t+1,nsp (0) denote NSP transfers allocated to municipality i when the mayor belongs to the president s party and when the mayor belongs to any other party respectively; these are potential outcomes, since only one is observed for each municipality. The study group (N) is composed of municipalities in which there were close mayoral electoral races at time t. Similarly, let Y i,t+1,m (1) and Y i,t+1,m (0) represent mayoral transfers distributed to municipality i when the mayor and the president are co-partisans and when the mayor belongs to any other party respectively. Thus τ RD,M in equation (2) represents the average treatment effect of the aligned candidate s victory on mayoral transfers. τ RD,NSP = 1 N N Yi,t+1,NSP (1) Y i,t+1,nsp (0) (1) i=0 τ RD,M = 1 N N Yi,t+1,M (1) Y i,t+1,m (0) (2) i=0 I estimate these effects by comparing the averages of transfers between aligned and unaligned municipalities in the study group (in which the president s party barely won or barely 18 Studies suggest that the Workers Party favored their co-partisans over as a party strengthening strategy a type of strategy that did not take place during the PT predecessors terms (Boas, Hidalgo and Richardson, 2014). 18

19 lost in the previous election). 19 Thus a difference of means suffices to estimate τ RD,NSP and τ RD,M for different margins of victory for the aligned party under the assumption that treated and control units in the study group are exchangeable (Dunning, 2012; Bueno, 2014, pp ). I also follow Imbens and Kalyanaraman (2012) and Calonico, Cattaneo and Titiunik (2014) and identify the causal effect of alignment at the cutoff using local linear and local polynomial regressions A Bundle of Transfers Both types of transfers are part of the central incumbents set of tools for cultivating support. While transfers to mayors reward allegiance, NSP transfers avoid credit leakages and reduce the political costs caused by blame for leaving voters governed by the opposition without resources. Incumbents use them as a bundle and they allot each type strategically, considering the relative costs and the expected benefits from the combination of each sort of transfer. I use a difference-in-differences design to capture this substitution effect. The first term in equation (3) is τ RD,M, the effect of alignment on mayoral transfers, which I expect to be positive, and the second term is τ RD,NSP, which I posit to be negative per the bypassing hypothesis. Since the second term is subtracted, I expect τ diff to be positive and larger in magnitude than the absolute value of either term alone. 19 Vote margin is measured as the difference of votes between the winner and the runner-up in the decisive round. 20 See Imbens and Kalyanaraman (2012) and Calonico, Cattaneo and Titiunik (2014) for the definition of the causal parameter at the cutoff. See Online Appendix section 3 for the regression models. 19

20 τ di f f = 1 N Yi,t+1,M (1) Y i,t+1,m (0) 1 N N i=0 N Yi,t+1,NSP (1) Y i,t+1,nsp (0) (3) i=0 Finally, I estimate τ diff using τ RD,M and τ RD,NSP, from equations (2) and (1). Thus the assumptions used in estimating τ diff are the same assumptions necessary for validity of the regression-discontinuity design, even though any measure of uncertainty for τ diff has to account for the non-independence of the outcomes. 2.2 A Natural Experiment on Partisan Alignment Close-race regression-discontinuity designs (RDDs) assume that fair and close races have an element of unpredictability: near-winners and near-losers are determined by chance and should thus be very similar on average. This is an assumption that should not be taken for granted. Caughey and Sekhon (2011) compellingly argue that the unpredictability assumption is not warranted for post-war US House of Representative close races because near-winners and nearlosers are systematically different. In contrast, Eggers, Fowler, Hainmueller, Hall and Snyder Jr (2014) present evidence from many elections in different countries, including mayoral elections for Brazil, in which near-winners and near-losers are, on average, indistinguishable. I also evaluate the claim of as-if randomness of alignment for near-winners and near-losers in Brazilian mayoral races. Evidence suggests that municipalities where the president s party barely won or lost do not differ on observed pre-treatment characteristics, just as they would (in expectation) if the treatment were truly randomly assigned. Table 3 shows the difference of means tests of balance between treatment and control groups for twenty pre-treatment covariates, for different margins of victory. As shown in Figure 1, in the smallest window of 0.5% of the vote margin, the difference between treatment 20

21 and control groups reaches levels of statistical significance smaller for only one of the covariates (Mayor s Age). Reassuringly, I do not observe statistically significant differences for this covariate for other windows or estimators. 21 I also evaluate balance using rank-sum tests, local linear, local polynomial, and a F-test using the entire set of covariates (Eggers et al., 2014). This set of tests shows no systematic imbalance (across different bandwidths and specifications) for any covariate. 22 Moreover, I also test if there is a relationship between lagged NSP and mayoral transfers (transfers from 2005 to 2008) and treatment assignment (elected in 2008). I find no systematic differences between treated and control groups. 23 These tests suggest that the imbalances are not larger than expected by chance. 21 Moreover, results do not change if I control for mayor s age in the RDD analyses. 22 See Online Appendix section See Figures S.B.6 to S.B.9. 21

22 Table 3: Balance Tests Vote Margin (%) Covariates Income per capita Std. error Doctors per thousand inhabitants Std. error Education (IDH) Std. error Income (IDH) Std. error Longevity (IDH) Std. error Illiteracy rate Std. error Infant mortality Std. error Population Std. error Poverty rate Std. error Vote for Lula Std. error Vote for FHC Std. error Vote for PT (fed. dep.) Std. error Votes for PSDB (fed. dep.) Std. error Vote for PT (governor) Std. error Vote for PSDB (governor) Std. error Votes for PT (state dep.) Std. error Votes for PSBD (state dep.) Std. error Turnout Std. error Mayor s Gender Std. error Mayor s Age Std. error Notes: Robust standard errors. Socio-economic covariates and electoral covariates were measured in 2000 and 1998 respectively. All covariates were standardized to have mean 0 and standard deviation 1.

23 Figure 1: P-value plot Covariate balance between aligned and unaligned municipalities for twenty covariates across different windows Income per capita Doctors per thousand inhabitants Education (IDH) Income (IDH) Longevity (IDH) Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Illiteracy rate Infant mortality Population Poverty rate Vote for Lula Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote for FHC Vote for PT (fed. dep.) Votes for PSDB (fed. dep.) Vote for PT (governor) Vote for PSDB (governor) Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Votes for PT (state dep.) Votes for PSBD (state dep.) Turnout Mayor's Gender Mayor's Age Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Vote margin (%) Distance from Cutoff Notes: 100,000 simulated randomizations generated the sampling distribution of the estimated average effect under the null hypothesis of no effect. All p-values are two-sided. 23

24 Sorting of units near the threshold that distinguishes winners and losers is also a potential threat to the use of regression-discontinuity designs. The McCrary (2008) test does not reject the null hypothesis of no sorting (p-value of 0.419). 24 Tests for jumps of transfers at nondiscontinuity points show no evidence for concern about the design The Effect of Alignment on Transfers The results are consistent with the bypassing hypothesis. Figure 2 visually represents the effect of alignment on transfers at the discontinuity. Figure 3a shows, for instance, that municipalities where the president s party won by a 0.5% vote margin or less receive reais per capita less, on average, than municipalities where the president s party lost by a 0.5% vote margin or less the smallest window with enough data available in the RDD. At the optimal bandwidths, the causal effect of partisan align is approximately 9 reais per capita. In other words, municipalities in which the mayor belongs to the opposition receive, on average, more transfers to non-state organizations than do municipalities in which the mayor and the president are co-partisans. 26 The opposite pattern holds when I analyze mayoral transfers: co-partisan mayors receive more resources than opposition mayors. These results corroborate previous findings in the 24 See Figure S.B See Table S.B Most transfers are targeted to municipalities with close races, which partly explains the difference in estimates sizes for specifications using data very close to the threshold and those farther away from it. The literature on RDD extensively discusses the trade-offs involving in using smaller or larger bandwidths. Thus, I present results and evaluate RDD assumptions using different specifications. 24

25 literature on Brazil (Brollo and Nannicini, 2012). 27 The graph in Figure 3b presents the estimated effects on mayoral transfers of having an aligned mayor. The plot shows that aligned mayors who won by a 0.5% vote margin or less receive reais per capita more, on average, than unaligned mayors who won by a 0.5% vote margin or less. At the optimal bandwidths for local linear and local polynomial specifications, the causal effects of align are approximately 12 and 19 reais per capita. How large is the effect of mayoral partisanship on transfers to non-state organizations and mayors in substantive terms? In the optimal bandwidths, election of an opposition mayor causes an increase in transfers to non-state providers of approximately a 12% of the standard deviation of outcome for the control group, which is similar to the effect on mayoral transfers of electing an aligned mayor (12% for the optimal bandwidth for local linear regression and 18% for the optimal bandwidth for the local polynomial regression). 28 What do these estimates mean in context? Are funds given to a non-state provider an important resource in these municipalities? As scholars of Brazilian politics know, these discretionary transfers represent a small part of the federal budget (Arretche, 2009). Still, the effect size of alignment is as large as the per capita expenditures of the conditional cash transfers program Bolsa Família, known to have substantial political relevance (Zucco, 2013). In municipalities where the president s party was a winner or runner up, the average expenditures with Bolsa Família between 2008 and 2012 was 7.67 reais per capita. This is very close, albeit slightly smaller, to the amount of additional money given to non-state organizations in municipalities 27 The estimates are not directly comparable because I use data, time period, and reference year for deflation different from that of Brollo and Nannicini (2012). 28 In the smallest windows, standardized effect sizes are larger: 25% and about 61% for mayoral transfers. 25

26 Figure 2: NSP and Mayoral Transfers, Transfers to Non State Organizations (reais per capita) Transfers to Mayors (reais per capita) Vote Margin (%) Vote Margin (%) (a) NSP Transfers (reais per capita) (b) Mayoral Transfers (reais per capita) Notes: Negative vote margins represent unaligned mayors and positive margins represent aligned mayors. Points represent evenly-spaced binned means, selected to mimick the underlying variability of the data. Solid line represents predicted values from a global fourth-degree polynomial regression. These plots were implemented using algorithm proposed by Calonico, Cattaneo and Titiunik (forthcoming). 26

27 Figure 3: NSP and Mayoral Transfers: RDD Estimates, Number of observations Number of observations Average Difference in NSP Transfers Between Aligned and Unaligned Average Difference in Mayoral Transfers Between Aligned and Unaligned Vote margin (%) Absolute Distance from Cutoff Vote margin (%) Absolute Distance from Cutoff Difference of Means 90% Confidence Intervals (Difference of Means) Difference of Means 90% Confidence Intervals (Difference of Means) (a) NSP Transfers (reais per capita) (b) Mayoral Transfers (reais per capita) Note: Difference of means is the average in transfers for an aligned municipality minus the average for an unaligned municipality. 27

28 run by the opposition (10.03 reais per capita). Effect sizes are also relevant when compared to sources of municipal revenue that do not rely at all on federal transfers. In Brazil, property tax is one of the few sources (and often the main source that is not dependent on intergovernmental grants) of tax revenue for municipalities available to the mayor. The median municipal revenue, for places in which the president s party had a winner or runner-up mayoral candidate, from property taxes (IPTU) is approximately reais per capita. 29 Both property taxes and NSP transfers are revenues to be invested in public goods for the population; the former goes to the municipal treasury while the latter goes to a non-state provider. The redistributive consequences of bypassing an opposition mayor are relevant; non-state providers in municipality run by the opposition mayor receives, on average, the same amount of money spent on a flagship social program and 40% extra additional public resources allocated through a NSPs than funds available to the mayor from the main local property tax. 29 The amount of IPTU revenue in the whole country is much smaller: 6.14 reais per capita. 28

29 Table 4: RDD Estimates, Dependent Variables: NSP and Mayoral Transfers, (reais per capita) 29 Estimator Diff. Diff. Diff. Diff. Diff. Diff. Local Linear Local Polynomial Means Means of Means Means Means Means Regression Regression Vote Margin (%) Opt. Bd. Bd. Non-State Transfers Estimates Std. Error p-value Mean Std. dev non zero n Mayoral Transfers Estimates Std. error p-value Mean Std. dev non zero n Notes: Bootstrapped standard errors for difference of means and local linear regression. Randomization inference one-tailed p-value: probability of observing a treatment effect as large as or larger than the observed effect under the null hypothesis of no effect for all observations. 100,000 simulated randomizations generated the sampling distribution of the estimated average effect under the sharp null hypothesis of no effect. The p-values for the local linear and local polynomial regressions were calculated using a normal approximation and a one-tailed test statistic. Full results for these regressions can be found in Tables S.C.1 to S.C.4. Optimal bandwidth for local linear regression uses Imbens and Kalyanaraman s (2012) method (11% for NSP Transfers and 22% for Mayoral Transfers). Local polynomial regression with triangular kernel use the robust bias-corrected standard error and implementation of mean-squared-error optimal bandwidth selector developed by Calonico, Cattaneo and Titiunik (2014) (main bandwidths of 7% for NSP Transfers and 14% for Mayoral Transfers); bias estimated with quadratic polynomial. I report the standard deviation of the outcome variable in the control group as a reference to calculate standardized effect estimates (Gerber and Green, 2012, p.70).