A Simple Model

Consider the (political) maximization problem of a central government (or president) that must choose the amount of federal transfers τ_i to allocate to each local government i=1, . . ., N. The president's objective function is made up of two (political) benefits. First, the general goodwill of the citizens in every municipality i toward the central government: U(τ_i), where U′>0, U′′<0. Second, the likelihood that municipality i is run by a mayor aligned with the central government, because either local politicians are important opinion leaders and may be useful allies in the next presidential campaign or they may engage in rent-seeking activities for the president (such as collecting money to finance the party or developing established relationships with organized interests). Thus, if the future margin of victory of the mayoral candidate aligned with the president is positive, the central government increases its political capital by a positive (fixed) amount R.Footnote ⁶ However, the relative weight of the two benefits depends on whether voters give political credit for the increased municipal revenues to the central government or to the mayor. Accordingly, we define θ∈[0, 1] as the political credit spillovers accruing to the incumbent mayor: If θ=0, then the central government can claim full credit for τ_i; if θ=1, then all of the credit goes to the municipal government.

The president's objective function can be expressed as

(1) \begin{eqnarray} \max _{\tau _i} \bigg \lbrace (1-\theta )\sum _i U(\tau _i) + \sum _i R \cdot Pr[MVP_i>0]\nonumber\\ \qquad-\sum _i C(\tau _i) \bigg \rbrace , \end{eqnarray}

The crucial mechanism is that, in Equation (1), federal transfers also affect the margin of victory of the aligned candidate. In particular, we assume that transfers increase the electoral prospects of the incumbent mayor's political party. Thus, if the incumbent is aligned with the president (P_i=1), τ_i will increase the future margin of victory of the aligned candidate, and vice versa if the incumbent is unaligned (P_i=0):

(2) \begin{equation} MVP_i=\rho MVP_i^0+\epsilon _i+\theta (2P_i-1)f(\tau _i), \end{equation}

(3) \begin{eqnarray} Pr[MVP_i>0]=\Phi \bigg [ \frac{\rho MVP_i^0+\theta (2P_i-1)f(\tau _i)}{\sigma } \bigg ], \nonumber\\ \end{eqnarray}

Equilibrium transfers

The first-order condition of the president's maximization is

(4) \begin{eqnarray} &&(1-\theta )U^{\prime }(\tau _i)+\frac{R\theta (2P_i-1)f^{\prime }(\tau _i)}{\sigma } \phi \nonumber \\ &&\quad\times\, \bigg [\frac{\rho MVP_i^0+\theta (2P_i-1)f(\tau _i)}{\sigma }\bigg ]-C^{\prime }(\tau _i)=0, \end{eqnarray}

where φ[.] is the density function of the standard normal. Moreover, note that by definition, P_i=1 if MVP ⁰_i>0 and P_i=0 if MVP ⁰_i<0. From this condition, we derive a set of testable predictions for the direction and size of politically motivated transfers.

If no political credit spillovers exist (θ=0) or if transfers cannot affect elections (because MVP ⁰_i tends to either +∞ or −∞), the central government simply weighs the marginal benefit of citizens’ goodwill against the marginal cost of τ_i, resulting in the (politically) optimal amount of transfers τ* in the absence of tactical redistribution, such that (1−θ)U′(τ*_i)=C′(τ*_i). Things change if θ>0 and MVP ⁰_i∈[−1, 1]. In this case, the central government chooses τ_i by looking at the additional marginal benefit or cost of influencing the next municipal election. The second term on the left-hand side of Equation (4) is either positive if P_i=1 or negative if P_i=0. In other words, at the zero threshold (MVP ⁰_i=0), a sharp jump occurs in the maximization problem of the president, as the marginal cost of sending money to an unaligned incumbent suddenly becomes the marginal benefit of sending money to an aligned incumbent. This phenomenon implies a strong and positive impact of political alignment on transfers in close races.

Proposition 1.The local average treatment effect (ATE) of political alignment on transfers is positive at MVP ⁰_i=0. That is, $\lim _{_{MVP_i^0 \downarrow 0}} \tau _i - \lim _{_{MVP_i^0 \uparrow 0}} \tau _i >0$.Footnote ¹⁰

Proposition 1 highlights the peculiar theoretical content of the causal effect of political alignment on transfers in razor-close elections, which represents an upper bound for the average effect in all races. Furthermore, it is easy to show that politically motivated transfers are a decreasing function of the past margin of victory of the aligned candidate on both sides of the zero threshold.

Proposition 2.The amount of politically motivated transfers is a decreasing function of MVP ⁰_ion either side of the threshold MVP ⁰_i=0. That is, ∂τ_i/∂MVP ⁰_i<0.Footnote ¹¹

The intuition for this result follows from the persistence of electoral outcomes: When MVP ⁰_i<0, the president penalizes unaligned municipalities lost by a narrow margin because, especially there, transfers will make a difference in the next election. Similarly, when MVP ⁰_i>0, the president favors aligned municipalities won by a narrow margin.Footnote ¹²

Identification

The close-race RD design setup (Lee Reference Lee2008) is particularly suited to take the previous predictions to the data and estimate the causal effect of political alignment on the amount of (discretionary) federal transfers. Define τ_it(1) as the potential transfers received by municipality i during term t if the mayor is politically aligned with the president, and τ_it(0) as the potential transfers of the same municipality if the mayor is not aligned with the president. The variable P_it defines political alignment. The observed outcome is thus τ_it=P_it·τ_it(1)+(1−P_it)·τ_it(0). The estimand of interest is the ATE, E[τ_it(1)−τ_it(0)], defined over some subpopulation of interest.

Define W_i as municipality-specific covariates (including state fixed effects), X_it as (time-varying) mayoral characteristics, and δ_t as term fixed effects. In the ordinary least squares (OLS) model,

(5) \begin{equation} \tau _{it}=\alpha +\pi P_{it}+W_{i}^{\prime }\beta +X_{it}^{\prime }\phi +\delta _{t}+\varepsilon _{it}, \end{equation}

$\hat{\pi}$

Instead, a difference-in-differences estimator can control for time-invariant confounding factors through municipality fixed effects γ_i:

(6) \begin{equation} \tau _{it}=\alpha +\pi P_{it}+X_{it}^{\prime }\phi +\gamma _i+\delta _{t}+\varepsilon _{it}. \end{equation}

$\hat{\pi}$

To address the presence of both time-invariant and time-varying confounding factors, we implement an RD strategy in the spirit of Lee (Reference Lee2008) and compare municipalities for which the aligned candidate barely won with municipalities for which the aligned candidate barely lost.Footnote ¹³ Specifically, we calculate the margin of victory of the mayoral candidate aligned with the president in each municipality i and term t (MVP_it): At the threshold MVP_it=0, political alignment P_it sharply changes from 0 to 1.Footnote ¹⁴

MVP_it is viewed as a random variable depending on both observable and unobservable political factors, as well as on random events on election day (the shock ε_i in our model). Lee (Reference Lee2008) shows that identification requires that, for each political candidate, (i) the probability of winning is never equal to 0 or 1 and (ii) the probabilities of winning or losing the election by a narrow margin are identical.Footnote ¹⁵ All factors that are crucial in deciding electoral outcomes (such as campaign financing, ideology, or the unobservable valence of candidates) are not a threat to identification, as long as random events also play a minor role. In other words, electoral outcomes depend on both predictable elements and random chance, but the latter is crucial only when the race is close. Thus the ATE in close elections is defined as

(7) \begin{eqnarray} E[\tau _{it}(1)&-&\tau _{it}(0)|MVP_{it}=0]=\lim _{\epsilon \downarrow 0} E[\tau _{it}|MVP_{it}=\epsilon ]\nonumber\\ &-&\lim _{\epsilon \uparrow 0} E[\tau _{it}|MVP_{it}=\epsilon ]. \end{eqnarray}

Recent empirical literature shows that elections for U.S. Congress do not always meet these conditions. In particular, Caughey and Sekhon (Reference Caughey and Sekhon2011) show that close U.S. House elections during 1944–2008 are prone to manipulation, as bare winners and bare losers are significantly different from each other.Footnote ¹⁷ However, this criticism of the close-race RD design does not directly apply to our setup. Unlike in the original framework, the running variable of our strategy is not the margin of victory per se, but the margin of victory of the aligned candidate. As a result, we do not compare winners and losers but instead municipalities in which the aligned versus unaligned candidate won. For instance, the validity of our identification strategy is not affected by the possibility that incumbents are more successful in close races, as long as incumbents aligned with the president have the same probability of running for reelection as incumbents from the opposition coalition. This phenomenon is indeed the case in Brazil, as we discuss in the section devoted to the empirical results.Footnote ¹⁸

Nonetheless, the U.S. literature points to the fact that every close-race RD design must be validated on a case-by-case basis to show that no political actor is able to sort just above the zero threshold defined by the relevant margin of victory. In our case, we need to show that neither aligned nor unaligned candidates have a higher probability to win in close elections. In the section devoted to the empirical results, we present a vast array of density and balance tests of covariates to support this claim.

Estimation

Various methods are available to estimate the local ATE expressed in Equation (7). We first apply a spline polynomial approximation of order p in MVP_it:

(8) \begin{eqnarray} \tau _{it} = \sum _{k = 0}^{p}\left(\rho _k MVP^{k}_{it}\right) + P_{it} \sum _{k = 0}^{p}\left(\pi _k MVP^{k}_{it}\right)+\delta _t + v_s + \eta _{it}, \!\!\nonumber\\ \end{eqnarray}

$\hat{\pi}$

$\hat{\pi}$

An additional advantage of Equation (8) is that the shape of the polynomial to the left of zero (captured by the estimated $\hat{\rho }_k$) and to the right of zero (captured by the estimated $\hat{\pi}$_k) can indirectly test Proposition 2, shedding light on the relationship between the level of political competition and discretionary transfers to the president's (political) friends or enemies. However, because the assignment to political alignment is endogenous as we move away from zero, this evidence ought to be interpreted with caution (that is, keeping in mind alternative mechanisms that might interact with the relationship between MVP_it and τ_it). The endogeneity of transfers away from zero does not affect the identification of the ATE in close races as long as the conditions previously stated are met.

As an alternative estimation method, we apply local linear regression, which restricts the sample to municipalities in the interval MVP_it∈[−h, +h] and estimates the model:

(9) \begin{eqnarray} \tau _{it}&=&\rho _0+ \rho _1 MVP_{it}+ \pi _0 P_{it} + \pi _1 P_{it} \cdot MVP_{it}\nonumber\\ && +\, \delta _t + v_s + \eta _{it}, \end{eqnarray}

Note that with the above estimators, close races do not need to be defined because we estimate the boundary points of two regression functions on either side of zero and the ATE is identified as the difference between these two points. As a result, close races are those with a margin of victory approaching zero at the limit. For the sake of transparency, we also show results for two alternative estimators, namely OLS in close intervals around zero: [−10; +10] and [−5; +5].Footnote ¹⁹ Here, close races are defined as elections for which the margin of victory is lower than 5% or 10% in absolute terms, respectively.

Brazilian Political System

Brazil is a federal presidential democracy and has a multiparty system. The Brazilian territory is divided into 26 states and one federal district (Brasilia). Municipalities are the lowest layer of administrative division and are ruled by an elected mayor (Prefeito) and an elected city council (Camara dos Vereadores). Municipal governments are in charge of a relevant share of the provision of public goods and services related to education, health, and infrastructure projects. The president, governors, and mayors of municipalities with more than 200,000 eligible voters are directly elected through (runoff) majority rule, whereas mayors of municipalities with under 200,000 eligible voters are directly elected through plurality rule. The elections of the president, governors, and members of Congress all take place at the same time every four years, whereas all municipal elections are staggered by two years and also take place every four years.

The Brazilian party system is highly fragmented, and the composition of coalition governments at the federal level has constantly changed over time. In particular, the fragmentation of the party system increased in the late 1980s because of new legislation that eased the requirements for party organization and because of an exodus from the two largest parties at that time, PMDB (Partido do Movimento Democrático Brazileiro) and PFL (Partido da Frente Liberal). Currently, Brazil has 27 officially registered parties. In addition to PMDB and PFL, PT (Partido dos Trabalhadores) and PSDB (Partido da Social-Democracia Brasileira) are the most important of the parties.

The current president is Dilma Vana Rousseff Linhares, elected in October 2010 and affiliated with PT. After the 2010 parliamentary election, PT became the largest party in the Congress, and it rules some states and major cities. Former presidents were Luiz Inácio Lula da Silva, also affiliated with PT, elected in October 2002 and re-elected in October 2006, and Fernando Henrique Cardoso, from PSDB, elected in October 1994 and re-elected in October 1998. PSDB rules the two largest Brazilian states, Sao Paulo and Minas Gerais. However, PMDB was the largest party measured by number of mayors and by parliamentary seats in the 1990s and 2000s. Because of its relevance and median position, PMDB usually supports the federal government; indeed, it is a member of the current government coalition, also the case in both of Cardoso's mandates and in the second of Lula's mandates.

Brazil has a weakly institutionalized party system, with high electoral volatility, low levels of party identification in the electorate, fragmentation, and lack of strong ideological platforms. However, following the Constitution of 1988, all presidents have been able to build reasonably stable postelectoral government coalitions through several discipline devices, such as veto power on the main issues, special prerogatives on budget allocation (including federal transfers), and the strong power for parties’ leaderships (see Figueiredo and Limongi Reference Figueiredo and Limongi2000, Pereira and Mueller Reference Pereira and Mueller2002). Although partisanship is relatively weak, most Brazilians who claim to identify with a party refer to either PT, which is Lula's party, or to PSDB, which is Cardoso's party (see Samuels Reference Samuels2006).

Role of the President in Allocating Federal Transfers

The resources of Brazilian municipalities come from (i) local revenues, such as service taxes, residential property taxes, and fines, and (ii) transfers from the federal, state, or other municipal governments. The most important source of municipal revenues is represented by federal transfers, which amount to an average of 65% of the municipal budget. Basically, two types of federal transfers exist: (i) constitutional automatic transfers, mostly unrestricted (Fundo de Participação do Municipios, FPM) or tied to education and health programs, and (ii) discretionary transfers (CONVÊNIO), mostly related to infrastructure projects. Municipal governments depend strongly on these federal transfers for their budgets, as tax revenues average only 5.5% of municipal total revenues. The remaining share of municipal revenues is from state transfers.

This study focuses on discretionary federal transfers devoted to infrastructure projects, which amount to approximately 15% of total municipal expenditure in infrastructures. The data source for these transfers is the Finbra data set from the Brazilian National Treasury (Tesouro Nacional), which codes them as transferências de capital. Infrastructure transfers are related to budget items that involve the construction of buildings or bridges, the paving of roads, the improvement of water and sewer systems, the purchase of ambulances, and so on. We focus on these transfers because the bulk of the other revenues are largely nondiscretionary.Footnote ²⁰ Furthermore, these transfers are used to finance highly visible projects, that is, they are an ideal target for politicians willing to woo voters.

Although mayors must make some effort to apply for these discretionary transfers, the legislative and the executive play an important role in defining their precise allocation. The support of a federal deputy and the final consent of the president are crucial ingredients of the allocation mechanism. Effectively, the executive first drafts the annual budget law (Lei Orçamentária Anual), which is then subject to (individual or collective) amendments by legislators. In most cases, the municipalities receiving discretionary transfers are chosen by legislators attempting to bring the pork home, as the bulk of the proposed amendments include targeted benefits to local areas. However, there are limits to the number and value of the amendments that legislators can propose. The budget committee is in charge of authorizing the bill. After a period of discussion, congress votes for the budget law, which is then sent back to the president for the final decision. As the budget law is not mandatory in Brazil, legislators cannot use the threat of not approving it to force the executive to accept any type of request. Thus the president has a major role in deciding the exact allocation of discretionary transfers. Municipal governments must also make some effort because a budgetary amendment can be executed only when an agreement between the municipality and the central government is signed.

At the end of the day, voters in a given municipality will receive discretionary infrastructure transfers depending on three factors: (i) the effort of their municipal administration; (ii) the interest of a federal deputy; and (iii) the interest of the president in executing the budget amendment. The president has the last word in the decision process and has many opportunities to influence the geographic allocation of infrastructure transfers, both in the construction of the budget law and in the implementation process. Therefore we follow the approach of Berry, Burden, and Howell (Reference Berry, Burden and Howell2010) for the U.S. and focus on the role of the president in the game of distributive politics.

Sample Selection and Variables of Interest

Brazil holds elections every two years. Federal elections and municipal elections take place separately; Figure 2 illustrates their exact timing. Our study encompasses federal governments in office during the three four-year terms from 1999 to 2002, from 2003 to 2006, and from 2007 to 2010, as well as municipal governments in office during the three four-year terms from 1997 to 2000, from 2001 to 2004, and from 2005 to 2008. Our baseline results consider federal transfers in the last two years of each mayoral term, that is, the per-capita amount of infrastructure transfers for the periods 1999–2000, 2003–2004, and 2007–2008. For transfers in 1999 and 2000, the president was Fernando Henrique Cardoso, whereas for transfers since 2003, the president was Luis Inácio Lula da Silva. The choice of the timing is motivated by the fact that we want to capture opportunistic transfers in proximity to municipal elections. Indeed, aggregate data show that, in the years of municipal elections, transfers are larger by about 75% with respect to their average value.Footnote ²¹ However, as a second step, we also check whether opportunistic transfers take place in proximity to federal rather than municipal elections. Thus we build an alternative measure of federal transfers, referring to the first two years of each municipal term: 1997–1998, 2001–2002, and 2005–2006. In our data, federal transfers are measured in per-capita real values expressed in 2000 Brazilian reais.

FIGURE 1.

Politically Motivated Transfers

Notes: Politically motivated transfers as a function of the past margin of victory of the candidate aligned with the president.

FIGURE 2.

Timing of Brazilian Elections

Notes: Timing of federal and municipal elections over the sample period.

For the treatment, we define two different variables capturing whether or not the municipality is politically aligned with the president. The first measure indicates whether the mayor belongs to a party in the federal government coalition, and the second measure indicates whether the mayor belongs to the president's party. Supplemental Online Appendix Table A1 (available at http://www.journals.cambridge.org/psr2012016) provides details on the political parties included in the government coalition year by year.

To implement our strategy in the Brazilian multiparty system, we must restrict the sample to municipalities for which only two (or at most three) candidates ran for mayor and one of them was politically aligned with the president. This restriction results from the fact that treatment assignment is considered as good as random only if the mayoral candidate aligned with the president had a fair chance of winning the election. In principle, one could also implement the RD strategy in all races in which one of the first two (or three) candidates was aligned with the president (irrespective of the total number of candidates). However, in this case, the sample selection depended on the electoral outcome of the aligned candidate and, therefore, on the RD running variable.Footnote ²² Instead, in our RD design, sample selection only depends on pre-electoral variables, and races with the same margin of victory share the same total number of candidates.

Our sample selection procedure improves internal validity but comes at the price of lower external validity. Races with only two candidates amount to 51% of the total, whereas races with only three candidates amount to 31%, indicating that races with at most three candidates represent 82% of all municipal elections. If we further restrict the sample to elections in which the aligned mayor belonged to the best candidates, we end up with a sample of two-candidate races that covers 42% (19%) of the total when we look at the president's coalition (party). With three-candidate races, we end up with a sample that covers 55% (28%) of the total when we look at the president's coalition (party). This means that, starting from the entire sample of 22,287 elections, we can implement our RD strategy in four distinct subsamples: (i) two-candidate races with respect to the president's coalition (5,723); (ii) two-candidate races with respect to the president's party (2,612); (iii) three-candidate races with respect to the president's coalition (12,245); and (iv) three-candidate races with respect to the president's party (6,248).Footnote ²³

Focusing on the RD samples of two-candidate and three-candidate races, we can preliminarily check whether significant differences exist between municipalities with aligned versus unaligned mayors. According to the descriptive statistics reported in Table 1 and Table 2, municipalities aligned with the president's coalition (or the president's party) receive larger per-capita transfers when municipal elections are approaching (i.e., in the last two years of the term), whereas no significant difference emerges with respect to the first two-year transfers. The reelection probability of aligned mayors is also higher. Instead, the census characteristics show that no evidence exists of “selection on observables” in our samples; that is, municipalities that are politically aligned with the president do not differ from the others in terms of development indicators. This finding is not particularly relevant for our identification strategy, which accommodates for selection on both observables and unobservables, but represents additional information on Brazilian politics.

TABLE 1.

Descriptive Statistics, Aligned versus Unaligned Municipalities in Two-candidate Races

Notes: Two-candidate races only. Aligned municipalities are those where the winner is affiliated with the president's coalition or party. Unaligned municipalities are those where the winner is not affiliated with the president's coalition or party. All columns except those with p value report the average values in the respective subsamples; p value refers to the statistical significance of the difference between means. First two-year transfers (last two-year transfers) are the average infrastructure transfers from the federal government to municipalities in the first (last) two years of the mayoral term (per-capita real values in 2000 Brazilian reais for both variables). Reelection rate captures whether the incumbent mayor is reelected or not. Population is the number of resident inhabitants in 2000. Per-capita income refers to monthly income in 2000 and is measured in Brazilian reais. The following variables refer to the 2000 census and are expressed in percentage terms: urban population is the fraction of people living in urban areas; water access; sewer, and electricity; literacy rate is the fraction of people above 20 who are literate; are the fraction of houses with access to water supply, sewer, and electricity, respectively; radio captures whether there is at least one local radio station in the municipality. North, northeast, center, south, and southeast are macroregions.

TABLE 2.

Descriptive Statistics, Aligned versus Unaligned Municipalities in Three-candidate Races

Notes: Three-candidate races only. Aligned municipalities are those where the winner is affiliated with the president's coalition or party. Unaligned municipalities are those where the winner is not affiliated with the president's coalition or party. All columns except those with p value report the average values in the respective subsamples; p value refers to the statistical significance of the difference between means. First two-year transfers (last two-year transfers) are the average infrastructure transfers from the federal government to municipalities in the first (last) two years of the mayoral term (per-capita real values in 2000 Brazilian reais for both variables). Reelection rate captures whether the incumbent mayor is reelected or not. Population is the number of resident inhabitants in 2000. Per-capita income refers to monthly income in 2000 and is measured in Brazilian reais. The following variables refer to the 2000 census and are expressed in percentage terms: urban population is the fraction of people living in urban areas; water access; sewer, and electricity; literacy rate is the fraction of people above 20 who are literate; are the fraction of houses with access to water supply, sewer, and electricity, respectively; radio captures whether there is at least one local radio station in the municipality. North, northeast, center, south, and southeast are macroregions.

Effect of Political Alignment on Federal Transfers

In this section, we present results on the impact of political alignment on federal transfers both in all elections and in razor-close races. As a benchmark, Table 3 shows OLS and difference-in-differences (DD) estimates. Instead, Table 4 shows the main results with different RD estimators. For the reasons previously discussed, we look at two outcomes: In the first two columns of both tables, the dependent variable represents the transfer amounts during the last two years of the mayoral term (when the next municipal election is approaching); in the last two columns, the dependent variable represents the transfer amounts during the first two years of the term (when the next federal election is approaching). For both variables, we have two separate columns because we measure political alignment that refers to both the president's coalition and the party: In the first case, the control group includes parties from the opposition coalition; in the second case, the control group includes all parties but the president's party. To ease comparisons, we implement OLS and DD in both the sample of all municipal elections and our RD samples of two-candidate and three-candidate races.

TABLE 3.

The Impact of Political Alignment on Federal Transfers, OLS and DD Estimates

Notes: Dependent variables: last two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the last two years of the mayoral term; first two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the first two years of the mayoral term (per-capita real values in 2000 Brazilian reais). Two-candidate (three-candidate) races are those where only two (three) candidates run for mayor and one of them is affiliated with the president's coalition or party. Estimation methods: OLS and DD (i.e., difference-in-differences) specifications as in Equations (5) and (6), respectively. Robust standard errors clustered at the municipality level are in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***.

TABLE 4.

The Impact of Political Alignment on Federal Transfers, RD Estimates

Notes: Dependent variables: last two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the last two years of the mayoral term; first two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the first two years of the mayoral term (per-capita real values in 2000 Brazilian reais). Two-candidate (three-candidate) races are those where only two (three) candidates run for mayor and one of them is affiliated with the president's coalition or party. Estimation methods: RD specifications with third-order spline polynomial approximation as in Equation (8); local linear regression with optimal bandwidth as in Equation (9); OLS in restricted intervals around the threshold of zero margin of victory, i.e., [−10; +10] and [−5; +5] as specified. Robust standard errors clustered at the municipality level are in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***.

According to the cross-sectional and panel evidence reported in Table 3, politically aligned municipalities receive more infrastructure transfers during the last two years of the term, whereas estimates on the transfers received during the first two years are never statistically significant. The results in the all sample and in two-candidate (or three-candidate) races are similar to each other, which reaffirms the external validity of the RD sample restriction. For all municipalities (panel A), with respect to the average level, transfers increase by 18.4% or 24.1% when the mayor belongs to the coalition or party of the president, respectively. In two-candidate (panel B) and three-candidate races (panel C), the OLS estimates, respectively, point to a 27.3% or 20.5% effect of coalition alignment, and to a 24.9% or 35.8% effect of party alignment. The size of the DD estimates is in the same ballpark, with just one exception, but the DD estimates tend to be slightly larger than OLS. Although the discrepancy might arise from downward bias in the OLS coefficients, the DD results might suffer from (time-varying) omitted bias.

Table 4 reports the main RD results. We implement four specifications: spline polynomial approximation with full bandwidth as in Equation (8); local linear regression with optimal bandwidth as in Equation (9); and two OLS estimations as in Equation (5) but in close intervals around the zero threshold, [−10; +10] and [−5; +5], which test for the difference in means in close elections defined by a margin of victory lower than 10% and 5% in absolute value, respectively.

In the two-candidate sample (panel A), according to the baseline estimation with spline polynomial, being affiliated with the coalition of the president increases per-capita infrastructure transfers by approximately 36.9%.Footnote ²⁴ Instead, the effect of being affiliated with the president's party is not statistically significant. In the three-candidate sample (panel B), being affiliated with the president's coalition increases transfers by approximately 26.3%. The effect of being affiliated with the president's party amounts to 34.3%, although significant only at the 10% level. Based on local linear regression, being affiliated with the president's coalition entails an increase in per-capita infrastructure transfers by 40.7% (two-candidate) or by 32.9% (three-candidate).Footnote ²⁵ The estimates on the president's party are never statistically significant. The OLS specifications in close intervals confirm these results in terms of both size and statistical significance.

On the whole, the RD results on the president's coalition are a direct confirmation of Proposition 1 in our model. The fact that the results on the president's party are less robust could receive a twofold explanation: From a political point of view, the president might be forced to please political allies even more than his or her own party to keep the government coalition united; from a statistical point of view, as the point estimates on the president's party are fairly similar to those on the president's coalition, the lower significance might be a problem of accuracy induced by the smaller sample size.Footnote ²⁶

Table 4 also confirms that no effect exists of political alignment on the infrastructure transfers received in the first two years of the municipal term. Consistent with the aggregate evidence discussed in the section devoted to the data description, no evidence exists of opportunistic transfers in proximity to federal elections. This result also reinforces the notion that mayors can claim the lion's share of the political credit from larger transfers (i.e., that the parameter θ in our model is greater than zero) because tactical redistribution takes place only in proximity to the elections for mayor and not to the elections for the Brazilian president.Footnote ²⁷

Figure 3 shows the estimated spline polynomial in MVP_it to highlight not only the jump in transfers at MVP_it=0 but also the shape of the relationship between transfers and political competition for aligned municipalities (on the right of zero) and unaligned municipalities (on the left of zero). With respect to the president's coalition, Figure 3 clearly shows that the ATE at zero is driven by a sizable cut in transfers to unaligned municipalities close to zero. In other words, the central government penalizes municipalities ruled by mayors of the opposition coalition, especially if they won by a narrow margin, thereby tying the hands of its (political) enemies for the next election. This evidence on the left of zero is consistent with Proposition 2 in our model.

FIGURE 3.

Political Alignment and Federal Transfers

Notes: The central line is a spline third-order polynomial in the margin of victory of the president's coalition or the president's party, fitted over the interval [−0.80, +0.80]; the lateral lines represent the 95% confidence interval. Scatter points are averaged over five-unit intervals. Two-candidate (three-candidate) races are those where only two (three) candidates run for mayor and one of them is affiliated with the president's coalition or party.

However, the evidence on the right of zero shows no increase in transfers to aligned municipalities that won by a narrow margin. In contrast, some evidence exists that shows that the federal government gives more money to its strongholds (where MVP_it is large), although we do not know whether the endogenous characteristics of these municipalities or political motivations (as predicted by Cox and McCubbins Reference Cox and McCubbins1986) drive this phenomenon. Moreover, one possible explanation for the asymmetry in the findings might be incumbency advantage, which is not incorporated in our model. In fact, a slight incumbency advantage might make close losses on the left of zero an even more important target for the president (the “tying hands” effect), and might make close victories on the right of zero less of a target (because the aligned candidate could count on some incumbency advantage at the party level in the next election).

Politics of Opportunistic Transfers

Our model assumes that federal transfers improve the reelection prospects of the incumbent. Others have estimated the effect of transfers on electoral outcomes using Brazilian data (e.g., Brollo et al. Reference Brollo, Nannicini, Perotti and Tabellini2010, Litschig and Morrison Reference Litschig and Morrison2010) and our RD design does not allow us to credibly identify such an effect. Yet, identifying the effect of political alignment per se on the reelection of the incumbent political party is interesting, keeping in mind that many factors besides politically motivated transfers may drive this differential incumbency advantage (if any). For instance, a positive impact of political alignment on reelection might capture a “bandwagon” effect on the side of voters and interest groups prone to support politicians connected with the national leader who won the last presidential election. In the first two columns of Table 5, we repeat our RD specifications with the incumbent party's reelection as a dependent variable.

TABLE 5.

The Impact of Political Alignment on Electoral Outcomes, RD Estimates

Notes: Dependent variables: incumbent party's reelection, i.e., the probability that the mayor's political party wins the next election; president's vote share, i.e., the incumbent president's vote share in a given municipality at the next presidential election. Two-candidate (three-candidate) races are those where only two (three) candidates run for mayor and one of them is affiliated with the president's coalition or party. Estimation methods: RD specifications with third-order spline polynomial approximation as in Equation (8); local linear regression with optimal bandwidth as in Equation (9); OLS in restricted intervals around the threshold of zero margin of victory, i.e., [−10; +10] and [−5; +5] as specified. Robust standard errors clustered at the municipality level are in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***.

Being aligned with the president increases the chances for reelection, although point estimates are precisely estimated only in the (larger) three-candidate sample or with OLS in close intervals. Under the additional hypothesis that the reelection effect of transfers is first order with respect to other motivations and given that, on average, political alignment increases transfers by approximately one-third and reelection by approximately 25%, the elasticity of reelection with respect to transfers should be slightly less than 1.

Our model also assumes that aligned mayors are valuable assets for the president. This assumption may be related to electoral advantages (if mayors seek votes for their preferred presidential candidate), rent seeking (such as collecting funds on behalf of the party), or mere policy congruence. To assess the plausibility of the first channel, we repeat our RD estimations with the municipal vote share of the president (or the candidate of the same party) in the future presidential election as a dependent variable in the last two columns of Table 5. The baseline estimates with spline polynomial approximation point to a gain of about 2 (coalition) or 3 (party) percentage points for presidential candidates who can count on aligned mayors. Yet, the results are not robust for three-candidate races or for the use of alternative estimators. However, some evidence exists that at least part of the political capital associated with aligned mayors might arise from electoral advantages.

The results presented above disclose a strong effect of political alignment on federal transfers in razor-close elections, which our model identifies as an upper bound of tactical redistribution. To shed more light on the politics of opportunistic transfers and to assess the comparative statics predictions of our model, we now evaluate the heterogeneity of this treatment effect. In other words, we interact our baseline RD spline polynomial specification with different (binary) heterogeneity dimensions to assess whether the estimated ATE is statistically different in some respects. In particular, we focus on two dimensions: (i) whether or not the value of political capital for the central government is high (see Table 6); and (ii) whether or not transfers are expected to have a strong impact on the electoral outcome (see Table 7).

TABLE 6.

Last Two-year Transfers and Political Capital, RD Heterogeneity

Notes: Dependent variable: last two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the last two years of the mayoral term (per-capita real values in 2000 Brazilian reais). Panel A compares municipalities without and with a radio station. Panel B compares municipalities below and above the median population. Panel C compares municipalities where the president obtained more than the median vote share in the past presidential election (president's strongholds) versus the others (opposition strongholds). Estimation methods: RD specifications with third-order spline polynomial in different subsamples as in Equation (8). Robust standard errors clustered at the municipality level are in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***.

TABLE 7.

Last Two-year Transfers and Effectiveness, RD Heterogeneity

Notes: Dependent variable: last two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the last two years of the mayoral term (per-capita real values in 2000 Brazilian reais). Panel A compares first-term and second-term mayors, where the latter face a binding term limit. Panel B compares states where the governor is unaligned and aligned with the president. Panel C compares municipalities with turnout below and above the median level at municipal elections. Estimation methods: RD specifications with third-order spline polynomial in different subsamples as in Equation (8). Robust standard errors clustered at the municipality level are in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***.

As empirical proxies for the first dimension, we use two measures: (i) the population size of the municipality; and (ii) whether or not there is a local radio station. The intuition behind these measures is that, in small municipalities or municipalities without radio stations, local politicians play a strong role in influencing the public opinion; therefore they are valuable allies for the president. Table 6 shows that the effect of political alignment on federal transfers is significantly greater in places without a radio station and in small municipalities (i.e., with a population size below the median).

However, we are aware that the previous findings should be interpreted with caution because other unobservable characteristics at the city level might interact with population size. Therefore we look at political factors as additional heterogeneity dimensions. We use three measures as empirical proxies for the political effectiveness of transfers: (i) whether the incumbent mayor is allowed to run for reelection or faces a binding term limit; (ii) whether or not the incumbent mayor is aligned with the state governor; and (iii) whether the turnout in municipal elections is high or low (based on the median turnout of the last municipal election). The intuition is that mayors who run for reelection or who may have difficulty obtaining resources from other sources will put more effort into using transfers for political purposes.Footnote ²⁸ Moreover, irrespective of the mayor's effort, increased municipal revenues are expected to be more effective in terms of electoral outcomes when turnout is high.Footnote ²⁹ The results in Table 7 show that the effect of political alignment on transfers is significantly greater for first-term mayors, for mayors unaligned with the state governor, and in municipalities with high turnout. These results support the notion that transfers are diverted exactly to where they are more politically effective.

Finally, Table 8 provides additional heterogeneity results based on the nature and structure of political parties in Brazil. We estimate our baseline RD specifications using a spline polynomial approximation and a local linear regression; however, we leave in the treatment group only specific subsets of parties, namely, the president's party (i.e., PSDB under Cardoso and PT under Lula); the pivotal party PMDB; medium-size parties; and small parties. The control group only includes parties in the opposition coalition at the federal level. Note that this estimation with the president's party in the treatment group is different from the estimations presented in Table 4 because, for Table 4, the control group includes all remaining parties (both in the government and in the opposition coalition).

TABLE 8.

Last Two-year Transfers and the Role of Political Parties, RD Estimates

Notes: Dependent variable: last two-year transfers, i.e., the average infrastructure transfers from the federal government to municipalities in the last two years of the mayoral term (per-capita real values in 2000 Brazilian reais). Two-candidate (three-candidate) races are those where only two (three) candidates run for mayor and one of them is affiliated with some of the parties of the president's coalition: the president's party, PMDB, medium-sized parties, or small parties, as specified. Estimation methods: RD specifications with third-order spline polynomial approximation as in Equation (8); local linear regression with optimal bandwidth as in Equation (9). Robust standard errors clustered at the municipality level are in parentheses. Significance at the 10% level is represented by *, at the 5% level by **, and at the 1% level by ***.

The estimates in Table 8 show that the effect of being aligned with the president's coalition is primarily driven by the president's party. This result should also be interpreted based on the nature of Cardoso's and Lula's parties. Indeed, PSDB and PT are established parties under the spotlight at all levels of electoral competition, and for them partisan affiliation is fairly stable (as discussed in the section devoted to Brazilian institutions). Therefore the effect of political alignment on opportunistic transfers is more relevant for structured parties, and our findings are more likely to generalize to political environments with strong party structures. Instead, no evidence exists that medium-size and small parties, which in turn are also less structured and prone to party switching by local politicians, drive our results. Furthermore, we find that strategic considerations such as the incentive to woo coalition switching and pivotal parties are second order, as we find no effect for PMDB.

Validity Tests

For our RD strategy to be internally valid, political parties must not be able to sort above the threshold of zero margin of victory. In other words, neither aligned nor unaligned candidates should be able to systematically win close races. To check for the absence of manipulative sorting, Online Appendix Table A4 shows estimates of discontinuities in the lagged margin of victory of the aligned candidate, in the geographic location of the municipality, and in the 2000 census variables at 0. In particular, we run tests of the difference in means between aligned versus unaligned municipalities in close intervals near zero, namely [−10; +10] and [−5; +5]. Because most of these variables are predetermined with respect to the treatment (political alignment), we should observe no discontinuity as long as no manipulative sorting occurs around zero.Footnote ³⁰ This is indeed the case, as all of the variables are balanced around the threshold, excluding a few exceptions for geographical location in three-candidate races with respect to the president's party.Footnote ³¹

The result from balancing the lagged margin of victory is particularly reassuring regarding the validity of the RD setup, as it implies that predetermined electoral outcomes are not deciding close races between aligned and unaligned candidates. This result is also consistent with the descriptive evidence that incumbents (shown to be over-represented in close victories in the U.S.) belonging to the government coalition are not more likely to run for reelection than incumbents belonging to the opposition coalition. In fact, the probability that aligned versus unaligned incumbents run for reelection is 23.3% versus 23.5% (two-candidate with coalition dummy); 21.8% versus 22.0% (two-candidate with party dummy); 23.5% versus 24.8% (three-candidate with coalition dummy); and 21.8% versus 22.5% (three-candidate with party dummy). In addition, these differences are never statistically significant.

To further check for the absence of manipulative sorting, in the Online Appendix we show the results of (i) visual inspection of the histograms of MVP_it in Figure A1; and (ii) formal tests of the continuity of the density at MVP_it=0, in the spirit of McCrary (Reference McCrary2008), in Figure A2. The latter procedure tests the null hypothesis of continuity of the density of the margin of victory at zero and is implemented by running separately kernel local linear regressions of the log of the density on both sides of zero. These procedures do not show any evidence of manipulative sorting around zero. Municipalities in which the aligned candidate barely won are not over- or under-represented with respect to municipalities in which the aligned candidate barely lost. Therefore random shocks rather than choice variables such as federal transfers or campaigning effort and ability decide (ex post) close races between aligned and unaligned politicians.Footnote ³²