Judicial subversion: The effects of political power on court outcomes q

Are politicians in power treated more leniently in court? We show that Brazilian mayoral candidates charged with misconduct are 65 percent less likely to be convicted if they narrowly win the election. Politicians play no direct role in the judges’ careers, suggesting that formal independence does not completely insulate the judiciary from political inﬂuence. The effect is driven by districts with few judges and by judges with higher career instability. (cid:1) 2022 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).


Introduction
Keeping elected politicians accountable to the law is essential for political and economic development.Legal checks on politicians can prevent public funds embezzlement, ensure free and fair elections, reduce political violence, and create a predictable investment climate (North and Weingast, 1989; Voigt et al., 2015;  Mehmood, 2019; Chemin, 2021).Such an outcome, however, might be difficult to achieve in practice.Judges and other public officials often face strong incentives not to enforce the law towards politicians in power.Elected officials might have the power to make a judge's career difficult, to starve the budgets of law enforcement institutions, or to offer easier access to government jobs and services to friendly judges and their families.
In response, most modern societies impose rules to make the judicial system immune to political influence.These rules secure the life-time tenure of judges and ban them from most outside jobs, especially in the executive.There are often similar rules protecting the independence of prosecutors.Are these rules sufficient to remove all influence of political power over judicial decisions?Some studies show that the judiciary favors elected politicians or their party when politicians control the nomination or promotion of the judges (Ramseyer and Rasmusen, 2001; Sanchez-Martinez,  2017; Mehmood, 2019; Mehmood, 2022; Poblete-Cazenave,  Forthcoming).As far as we are aware, however, there is no causal evidence showing whether judges favor politicians in office when politicians have no formal control over the judges' careers.This is an important gap in the literature, since in a majority of countries, trial judges are not directly nominated or promoted by politicians (Garoupa and Ginsburg, 2009).If judges who do not directly depend on politicians for their careers are still swayed by political power, we can conclude that removing formal powers to nominate and promote judges from politicians does not completely protect these judges from political influence.
In this paper, we study court cases against official misconduct called Ações de Improbidade ("improbity cases") involving local politicians in the trial courts of the Brazilian state and federal judiciaries.A range of formal rules ensuring that judges are immune to political influence are in place in Brazilian trial courts: Judges are difficult to remove, cannot be transferred to other positions against their will, and are not allowed to have any other job except teaching.Politicians have no direct influence over the judges' careersthey are appointed by a competitive exam administered by the appeals court, which also determines promotions.Appeals court judges are nominated by the appeals court itself, except for one fifth appointed by the state governor or the president.Finally, judges receive a very high salary, placing them among Brazil's top earners.Similar rules apply to the public prosecutors who are in charge of the prosecution.
Knowing whether these rules are sufficient to prevent politics from influencing judicial decisions is challenging.If elected politicians are more likely to win in court than others are, it is hard to know if this difference is due to political influence or if they just tend to have stronger cases (Priest and Klein, 1984).We solve this empirical challenge by using a close election regression discontinuity design, focusing on official misconduct cases filed before the election. 1If close elections are decided at random, marginal electoral winners and losers will, on average, tend to be involved in similar misconduct cases at the time of the election.We can then conclude that systematic differences in court outcomes between marginal electoral winners and losers are causally due to the election result.In our main regression, we find substantial effects of political power on judicial outcomes-candidates that barely lose an election have a 17 percent conviction rate, while those that barely win and become mayors have a 6 percent conviction rate.We thus document a sizable causal effect of political power on court outcomes in a setting where judges are formally very independent from the executive branch.This result suggests that mayors-while not having any formal power over the judicial branch-still exert significant informal influence over local judges.
In the second part of the paper, we ask why judges are less likely to convict politicians in power.First, we show evidence suggesting that the effect is not driven by mayors influencing judges through legal means-by hiring better lawyers.Marginal electoral winners do not increase the quantity or quality of their lawyers substantially more than electoral losers do after the election.Furthermore, there seems to be an effect of winning the election also for cases in which the lawyers have done all the formal work before the election.
We then show suggestive evidence pointing to two non-legal mechanisms: favor exchanges and career concerns. 2While we cannot document a quid pro quo favor exchange, we argue that a collusive relationship between the mayor and the judge is more likely in districts with few judges.Consistent with such a mechanism, we find that the effect is driven by districts with only one or two judges.We do not, however, detect a larger effect when the court is located in the municipality of the mayor than when the court is located in a neighboring municipality.This result suggests that any favors offered to the judge by the mayor are not tied to the location of the court.For instance, the mayor offering contracts to the judge's relatives is a more likely mechanism than, say, municipal lawyers helping out in the court.
Career concerns seem, at first glance, unlikely to be an important explanation of our results-mayors play no role in appointing or promoting judges, and trial judges rely on the appeals court for their careers.Politicians at the state and federal levels do, however, have influence over the appeals courts-notably by appointing a share of the judges and determining the judiciary's budget.A judge with career concerns might thus be reluctant to convict a mayor aligned with politicians in power at the state or federal levels.Perhaps surprisingly, we find evidence consistent with career concerns being part of the explanation.First, we find that judges with a higher tendency to move between judicial districts and thus relying more on the appeals court for their careers are more influenced by the election.Second, our estimates indicate that the effect is larger if the judge is untenured and if the politician's party has appointed appellate judges.In the Appendix, we present results suggesting that judges who convict mayors are more likely to be promoted by a seniority criterion that depends only on time-ofservice, than by a discretionary criterion, which is a faster career path but that requires a vote by the appeals court.
We end our analysis of the mechanism by discussing five additional explanations we believe are unlikely to be major drivers of our result: Law enforcers not wanting to interfere with the local government due to social costs concerns, judges wrongly attributing winning a close election to be a signal of probity, elected politicians destroying evidence, judges being averse to media attention, and interlinkages between official misconduct cases and criminal cases.
The question of how to create a judicial system immune to the subversion of the politically powerful has captivated researchers at least since Montesquieu (1748) and Smith (1976), spawning a large literature across the social sciences. 3The number of studies seeking to measure the degree to which judges act independently from the executive is, however, limited. 4Existing studies almost all take place in settings where politicians have control over the career paths of law enforcers and focus on higher levels of the executive and the judiciary (e.g. the President and higher courts).Our main contribution is thus to demonstrate that politicians in power might receive a favorable treatment by the judicial system even when they have no formal control over the careers of local judges or prosecutors.In doing this, we corroborate the finding in cross-country studies that de jure judicial independence does not necessarily translate into de facto judicial independence (Hayo and Voigt, 2007; Melton and  Ginsburg, 2014; Hayo and Voigt, 2019).
The rest of the paper is organized as follows.In Section 2, we describe the legal remedies against official misconduct in Brazil, the careers of Brazilian judges and prosecutors, and which tools Brazilian local politicians have at their disposal to benefit or harm law enforcers.In Section 3, we explain how we constructed the data set of misconduct cases involving local politicians, and in Section 4 we introduce the empirical strategy we use to estimate how being elected affects judicial decisions.We present our main results, showing that politicians are less likely to be convicted of misconduct if they win the election, in Section 5, while in Section 6 we seek to explain why.We conclude in Section 7. Sanchez-Martinez (2017), Assumpcao and Trecenti (2020), and Poblete-Cazenave (Forthcoming).Our study differ from Sanchez-Martinez (2017) and Poblete-Cazenave (Forthcoming) by being in a setting where the judges are not appointed by the executive.Assumpcao and Trecenti (2020) study small-claims cases that do not threaten the politicians' careers and are thus fundamentally different from the official misconduct cases we study in our paper.
2 These results should be seen as merely suggestive since we neither directly observe a favor exchange nor have access to random variation in judges' career incentives.Anecdotal evidence, however, suggests that political influence from the appeals court on the judges' career path and favor exchange between judges and politicians are common in Brazil.For example, Eliana Calmon, the first woman to serve on the Superior Justice Court and former top watchdog of the National Justice Council, said it is not uncommon for an appeals judge to call a trial judge to ask for a sentence; those that abide are natural candidates for future promotions (Calmon,  2010).Concerning favor exchanges, there are many stories of judges' relatives being employed in the mayor's or the governor's office and in the courts of audit (Tribunal de Contas), which are run by politicians (see, for example, Junior (2002), Borges (2008), Luchete (2011), Valadares (2018), Fabrini (2019), Marques ( 2021)).Favor exchanges between politicians and judges have recently been documented by Mehmood (2022)  in Pakistan.

Institutional context
In this section, we first describe the legal remedies against misconduct among elected officials in Brazil, with a focus on Ações de Improbidade.Then we describe in detail the judges and the prosecutors who are involved in the cases and the general organization of the judicial system.Finally, we describe the relevant features of the Brazilian government.All rules described in this section are documented in Appendix A.

The legal remedies against official misconduct in Brazil
There are three types of civil suits addressing misconduct and less serious administrative malfeasance: Ação Civil de Improbidade Administrativa ("Ação de Improbidade"), Ação Civil Pública, and Ação Popular.In this paper, we do not consider criminal cases against official misconduct.Criminal cases against a mayoral candidate are sent to the appeals court if the politician wins the electiona rule colloquially known as foro privilegiado-making it difficult to interpret a close election regression discontinuity estimate for these cases.Among the civil suits, we focus on Ações de Improbidade since they are the most serious and they cannot be settled. 5ções de Improbidade can be filed against any act by a public official that either violates administrative principles, causes damage to the treasury, or leads to illicit enrichment of the official.Typical cases involve the hiring of public workers without proper procedure and fraud in government contracting.Only the public prosecutor or the entity harmed by the misconduct can bring the lawsuit.We consider only cases filed by the public prosecutor in this paper. 6The possible penalties are loss of office, loss of political rights for 3-10 years, reimbursing the treasury, fines up to 100 times the monthly wage, and the prohibition of receiving government contracts for 3-10 years.Politicians see the loss of political rights as one of the most severe penalties since it includes not being able to run for elected office.
The cases filed by the public prosecutor are typically initiated by someone filing a complaint to the prosecutor.The prosecutor then investigates (inquérito civil) and chooses whether to file a case depending on the outcome of the investigation.Ações de Improbidade involving local politicians are tried in the federal judiciary if the alleged misconduct or malfeasance involves funds transferred to the municipality by the federal government.
An example of a typical case in our data is an Ação de Improbidade filed by the public prosecutor against the mayor of the municipality Fartura in the state of São Paulo.The mayor had awarded a contract to provide fuel to the municipality to a firm owned by the son of the vice mayor, which is illegal.The judge, however, acquitted the mayor, arguing the misconduct was not done in "bad faith."7

Brazilian trial court judges and prosecutors
Judges and public prosecutors at the trial courts in the state and federal judiciaries are formally independent of politics.State and federal judges are appointed by a competitive public exam administered by the state appeals courts (Tribunal de Justiça) and the federal appeals courts (Tribunal Regional Federal), respectively.Similarly, the public prosecutors are appointed by a competitive public exam administered by the state or federal chief prosecutor.Judges and prosecutors earn a very high wage-just their official wage places them among Brazil's top earners, not counting several perquisites such as housing allowances.
Judges and prosecutors are insulated from political influence by three constitutional provisions (see Appendix A).Politicians at the federal and state levels have, however, some formal powers over the judiciaries and the public prosecution.We explain these powers, the career paths of judges and prosecutors, and the organization of the judicial system in the following subsections.

The judiciary
State and federal trial judges enter service as substitute judges (juiz substituto).After a two-year probation, if the judge did not commit any irregularities, she is granted tenure and becomes a regular judge (juiz vitalício), which happens almost always.Later, the judge can be promoted to a court chief judge (juiz titular de vara), which means she is now tied to a specific court in a lowlevel district, typically in a small municipality, and is the first step to progressing to higher-level districts and, eventually, to the appeals court. 8romotions are dependent on decisions made by the appeals court on criteria that alternate between seniority and a discretionary vote. 9The seniority criterion depends on the time of service and can be ignored only by a two-thirds vote of the appeals court.The discretionary criterion is based on a ballot vote among the appeals court judges.The chief appeals judge chooses among the three judges who received the most votes.The promotion of a judge who appears among the three most voted three times in a row or five times in total is, however, mandatory.In federal courts, the president has the final call on who is promoted by vote to the appeals court among the three judges who received the most votes.
Another formal influence of politics is that one fifth of the state and federal appeals court judges are nominated by the state governor and the president, respectively.These seats are filled alternately by public prosecutors and lawyers.The organizations representing public prosecutors or lawyers prepare a list of six candidates that are reduced to three by the appeals court.
The movement of judges between courts is highly dynamic (see, e.g., Dahis et al., Forthcoming).Judges that are not yet court chief can be freely transferred between judicial districts and regular judges can substitute for another judge or be designated "auxiliary" judge in a different, sometimes higher-level, district or the appeals court.These movements between judicial districts depend on decisions made by the appeals court and thus represent an extra source of influence of the appeals court over lower-level courts.

The public prosecution
The state and federal public prosecution have a parallel structure to the judiciary with sections in each judicial district.The public prosecution (Ministério Público) is formally independent of both the executive and the judiciary and is often called the fourth branch of government.Public prosecutors also receive tenure after two years of service and are promoted by the chief prosecutor.
They are required to live in the judicial district they are stationed, except by authorization of the chief prosecutor.The state and federal chief prosecutors are appointed by the governor and the president, respectively, for a two-year term with the possibility of a one-term renewal.In practice, the governor and the president choose the chief prosecutor from a list of three candidates prepared after a vote among the prosecutors, but they are not obliged to choose from this list.Prosecutors have broad discretion in deciding which cases to work on, but any decision to drop a case is subject to review by the chief prosecutor.

Brazilian mayors
Brazilian mayors are elected via a first-past-the-post electoral system, except for cities with a population greater than 200,000, which have a second-round run-off between the top two candidates if none received more than 50 percent of the votes.
The main responsibilities of the mayor are to administrate the city budget and to collect municipal taxes.In doing so, the mayor has the power to contract firms and hire municipal workers.Many municipal employees are hired via a competitive civil service exam and receive tenure after three years of service, but there are some categories of jobs that give the mayor almost total discretion in deciding whom to hire, such as commissioned posts and temporary jobs.

Data
In this section, we explain how we built a data set of Ações de Improbidade involving local politicians and present summary statistics.

Judicial data
Our main data source is the daily official publication of each appeals court, called Diário de Justiça.The law requires appeals courts to publish all judicial decisions in these outlets, including trial court decisions, among several other minor statements about the case.In Fig. B.1 in the Appendix, we show an extract of a Diário de Justiça publication.To generate a data set from this source, we use regular expressions to select all publications on Ações de Improbidade and then extract the names of litigants, lawyers and judges, the judicial district, and decisions.We use all available issues of the Diário de Justiça across all state and federal appeals courts, except the state appeals courts of Rio Grande do Sul and Distrito Federal. 10 In Fig. B.2, we show the available coverage of the Diário de Justiça over time by appeals court.
Each case is identified by a unique number, allowing us to track cases over time.This number includes the year the case was filed.In the case of a final decision, we extract the penalties applied to each defendant, where this is explicitly stated.A defendant is considered convicted if he or she is found to have received a penalty.A defendant is considered acquitted if there is a final decision containing the expression "julgo improcedente" (petition denied) or explicitly stating that the defendant was acquitted (absolvido).If we are not able to identify any final decision, we code the case as not yet decided. 11Note, however, that being coded as not decided does not necessarily mean that the case has not been decided in real-ity.It could be that we have missed some decisions or that there has been a final decision in the case before our first available date in the Diário de Justiça.The latter is not infrequent-there are publications regarding appeals and sentence execution in a case even after the final decision.We still keep these cases in our sample, since removing them in an automated way is tricky.As the start of the Diário de Justiça is predetermined, keeping these cases should not lead to any bias in our estimates, except that we will consistently overstate the number of undecided cases.
The Diário de Justiça lists the names of the lawyers registered on the case in every publication, in most states together with their unique registration number with the Ordem dos Advogados do Brasil (OAB).We use this information to create a data set with all the lawyers registered on each Ação de Improbidade at each publication date.From this data set, we calculate lawyer experience by the number of previous Ações de Improbidade that the lawyer has worked on and lawyer success rate by the share of these cases that has led to a full acquittal. 12inally, the Diário de Justiça records promotions of judges.We extract this information for the courts that consistently record if the promotion was by the seniority or the discretionary vote criterion. 13We were not able to consistently identify the judge in the state judiciaries of Paraíba, Rio Grande do Norte, and Piauí.These judiciaries are thus excluded whenever we use judge information in our regressions.The Conselho Nacional de Justiça provides the number of chief judge positions (varas) by judicial district and the year each judge was appointed.
We complement the Diários de Justiça with data from the courts' online systems provided by Digesto-a legal intelligence firm.This gives us the exact filing date of the case, the disputed value, and the case subject.See Section C for details.

Electoral data and matching
We use election results and candidate characteristics from the electoral authorities (Tribunal Superior Eleitoral) and construct a data set of all candidates for mayor between 2004 and 2016. 14e keep only candidates who either won the election or received the second-most number of votes-the runner-up. 15We match court cases to politicians on perfect name matching, ignoring accents. 16e match only within states-if a defendant in a case in the state judiciary of Paraná has the same name as a candidate in the state of Goiás, it is not considered a match.
Our main estimation sample consists of all cases that are pending at the time of the election: Cases filed before the election but with no identified final decision prior to the election. 17We consider only cases with the public prosecutor among the plaintiffs. 18In Appendix D, we compare all close election candidates with the 10 The Diário de Justiça of the state judiciary in Rio Grande do Sul does not record the court case type (classe), so we were unable to identify which cases were Ações de Improbidade.Distrito Federal is the territory of Brazil's capital and federal government and does not have any municipalities or local elections. 11A publication in the Diário de Justiça is considered to be a final decision if it contains any of the phrases "julgo procedente," "julgo parcialmente procedente" or "julgo improcedente." 12A decision is considered a full acquittal if it contains the expression "julgo improcedente." 13These courts are TRF2, TRF5, TJAC, TJAL, TJBA, TJCE, TJES, TJGO, TJMA, TJMS, TJMT, TJMG, TJPE, TJSC, TJSP, TJRJ and TJRO.The abbreviations TRF and TJ stand for Tribunal Regional Federal (federal court) and Tribunal de Justiça (state court), respectively. 14Data on municipalities such as population, GDP, and geographic coordinates are all from the Brazilian Institute of Geography and Statistics (IBGE). 15In elections with a second round run-off, we focus on the second round.In Appendix Table E.3, we show that the main result is robust to including only firstround elections. 16In Appendix Table E.3, we show that the main result is robust to excluding common names. 17When we do not know the exact filing date, we keep a case only if it has a publication the Diário de Justiça before the election or the filing year inferred from the case number is before the election year. 18Cases without the public prosecutor among the plaintiffs are typically filed by the municipality.If a mayoral candidate with a case filed by the municipality pending against her wins the election, she essentially becomes both the plaintiff and the defendant in the case and the case is typically dropped.
candidates appearing in our sample.The candidates in our sample differ from the average candidate by having more political experience: Over 75% of the candidates are ex-mayors.This is expected since a politician need to be in office to commit an act of administrative improbity.

Summary statistics
In Table 1, we present summary statistics for our main estimation sample of all identified Ações de Improbidade involving candidates in the 2004-2016 local elections, pending at the time of the election.There are 4,484 observations.Most of the cases-81 percent-involve previous mayors, and 36 percent involve incumbent mayors running for reelection.We observe the politician receive a penalty in 13 percent of the cases, while in 13 percent of the cases the politician is acquitted and in 67 percent we have not been able to identify any final decision. 19Note that due to the discussion in Section 3.1, the true share of undecided cases is lower.The cases have a long duration-for cases that received a final decision, the average time between filing and final decision is almost seven years.In Tables D.1, we show how candidates involved in an Ação de Improbidade compare to other mayoral candidates.In Table D.2, we show the distribution of electoral races across election years and states where we observe mayoral candidates with a pending Ação de Improbidade.

Empirical strategy
We want to estimate the effect of political power on judicial decisions.If politicians in power are shown to be more likely to win in court than opposition politicians, this difference in win rates does not prove that decisions are affected by the political power of the litigant-elected politicians and politicians out of office are likely involved in different types of cases.The ideal experiment would be to randomly allocate elected offices to politicians and look at the effect on judicial decisions on already filed cases.We exploit close elections to simulate this experiment.In particular, we look at misconduct cases filed before the election, and not yet decided at the time of the election, comparing politicians who marginally won the election with politicians who marginally lost the election.The idea is that the winner of a close election is as good as randomly determined.Thus, winning and losing politicians should, on average, be involved in similar cases before the election, and any systematic difference in judicial decisions has to be due to the outcome of the election.
As our main specification, we use the bias-corrected estimator proposed by Calonico et al. (2014) with local linear regression for the estimate and local quadratic regression for the bias correction.The specification for the local linear regression is where i is a politician and c is a misconduct case.The variable M i is the electoral win margin of the politician, and E i indicates whether the politician was elected. 20The outcome y ic varies, but in the baseline model it is a dummy for whether the politician was convicted.To avoid researcher discretion in the choice of control variables, we tie our hands by using no controls.As a placebo check, we run the above regression for cases decided before the election.If close elections are indeed randomly determined, we should not see any effect of the election on these cases.We also report the main result using standard local linear specifications for different bandwidths, including the Imbens and Kalyanaraman (2012) optimal bandwidth.We cluster standard errors at the municipality by election-year level.

Main results: Are winners of close elections less likely to be convicted?
In Appendix Table E.1, we show the results of the estimation of Eq. 1 with a wide range of pre-election covariates as outcome variables.If close elections are indeed randomly determined, there should be no systematic differences between marginal winners and losers in these variables.In the first two rows, we consider all candidates in the 2012-2016 local elections.Marginal winners are not significantly less likely than marginal losers are to be involved in an Ação de Improbidade at the time of the election, or earlier.Thus, politicians that engage in less misconduct do not seem to systematically win in close races.The regression discontinuity histogram in Fig. E.1 in the Appendix also shows no evidence of a discontinuous decrease in politicians charged with misconduct at the threshold for winning the election.In the rest of Table E.1, we show the balance within the estimation sample of Ações de Improbidade pending at the time of the election.The results are consistent with the outcome of close elections being randomly determined.For instance, there is no evidence that electoral winners are less likely to be involved in severe cases as measured by the disputed value or by whether the case is about illicit enrichment.The estimated difference between marginal winners and losers is statistically significant at the five percent level for only one variable-whether the politician has higher education.In the Appendix Table E.2, we show that our main result is robust to controlling for all the covariates where the estimated difference is statistically significant at the 15% level and to controlling for all the remaining covariates in Table E.1.
In Column 1 of Table 2, we present the result of estimating Eq. 1 with outcome variable whether the politician is convicted.We estimate that marginal electoral winners are 11 percentage points less Notes: Ações de Improbidade involving candidates for mayor pending at the time of the election.Politician convicted is an indicator for whether the politician is recorded to have received any penalty, whereas Politician acquitted is an indicator for the final decision containing the expression "julgo improcedente" (petition denied) or explicitly stating that the defendant was acquitted.The experience of a lawyer is defined as the number of other Ações de Improbidade she has worked on prior to the election.Judge experience is the years of judicial experience of the judge on the case at the time of the election.
likely than marginal electoral losers are to receive a penalty-the conviction rate among marginal losers is 17 percent and only 6 percent among marginal winners, a 65 percent decrease in the probability of conviction.The coefficient is statistically significant at the one percent level.In Fig. 1, we show a clear downwards jump in the rate of convictions when the number of votes passes the threshold necessary to win the election.We present local linear regression discontinuity estimates for different bandwidth sizes in Appendix Fig . E.2, including the Imbens and Kalyanaraman (2012) optimal bandwidth. 21The optimal bandwidth selectors choose bandwidths of 12 and 22 percentage points, but the effect is statistically significant at the five percent level for any bandwidth greater than five percentage points.
In Column 2 of Table 2, we show the result of the placebo test.Reassuringly, there is no effect of winning the election on cases decided before the election.The absence of a discontinuity in the conviction rate for cases decided before the election can be visually inspected in Appendix Fig . E.5.In the Appendix, we show in Table E.3 that our main result is robust to excluding cases with no pre-election publication in the Diário de Justiça, in Table E.4 that the main result is robust to various levels of standard error clustering, and in Table E.5 the main result decomposed by election year.
There could be two reasons why a defendant in an Ação de Improbidade has not been convicted: He has been acquitted, or he is still waiting for a decision.In Columns 3 and 4 in Table 2, we consider these two outcomes.The point estimates suggest that the main reason for a lower conviction rate among marginal winners is that they are more likely to be acquitted.We estimate that candidates are 6.7 percentage points more likely to be acquitted and 4.0 percentage points less likely to have their case decided if they win the election.Only the effect on acquittals is statistically significant.Moreover, in Section 5.1, we show that there is no sign of an increase in convictions of the elected mayor after four years when the initial term is finished and many them step out of office, indicating that judges are not just postponing convictions of electoral winners.
In Table E.6, we decompose our main result by office and the various types of penalties.We detect statistically significant effects across all the penalties.

The timing of the effect
When does the gap in conviction rates among marginal winners and losers appear?Does the gap show any sign of closing after the four-year mayor term of the marginal winner is over?To answer these questions, we estimate Eq. 1 with outcome variable whether the case has ended in a conviction within x years after the election where x varies from zero to six years.Fig. 2 presents the results.The dotted and the solid lines show the estimates for marginal winners and losers, respectively, using the local linear fit.The regression discontinuity estimate is the difference between the two lines.The estimates indicate that there are almost no convictions involving marginal winners in the first year after they take office.The slope of the solid line is steeper than the dotted line up until 1.5 years after the election, meaning that there are also more convictions involving marginal losers in the second year after the election.From about 1.5 years and onward, the two lines run in parallel, suggesting that the number of convictions involving marginal winners and losers is roughly equal.Interestingly, judges are not just postponing convictions until many of the mayors are out of office-we do not see more decisions involving marginal winners than involving losers in the two years after the four-year mandate of the marginal winner.Instead, there seems to be a permanent gap in the number of convictions, with no sign of closing even six years after the election.The difference in the number of convictions between marginal winners and losers is statistically significant at the five percent level starting from the first year after the election.

Mechanisms: Why are politicians in power convicted at a lower rate?
There could be many reasons why politicians are less likely to be convicted of misconduct if they get elected.In this section, we first show evidence suggesting that the effect is not driven by mayors influencing judges through legal means-by hiring better lawyers.Then we offer some evidence supporting two non-legal mechanisms: Favor exchange and career concerns.Finally, we discuss other potential mechanisms.

Do electoral winners have superior lawyers?
Electoral winners might be convicted at a lower rate because they are represented by better lawyers.The most direct test of whether our result is driven by lawyers is to measure if marginal electoral winners tend to register more or better lawyers on their cases than losers do after the election.This test is possible since the Diário de Justiça lists the lawyers in each publication regarding the case.As proxies for the quality of each lawyer we use lawyer experience defined as the number of other Ações de Improbidade the lawyer has worked on before the election, and lawyer success rate defined as the share of these cases in which there was a full acquittal, conditional on being decided before the election.For each case, we calculate the average quantity and quality of lawyers in Diário de Justiça publications after and before the election, and define the post-election increase as the difference between these two numbers.
In Table 3, we show the result from estimating Eq. 1 with outcome variable the post-election increase in average lawyer quantity and quality.The point estimates indicate that, on average, marginal winners and losers increase the number of lawyers by, respectively, 0.78 and 0.68 after the election.The difference in effects is not statistically significant.In contrast, the lawyers working for marginal winners have, on average, experience from 0.6 more cases after the election, whereas the same number for marginal losers is À1.9.This difference is statistically significant at the five percent level.Compared to the average lawyer experience of ten cases in our sample, however, this difference is relatively small and unlikely to substantially affect the conviction rate.Finally, there is virtually no difference between marginal winners and losers in average lawyer success.In sum, marginal winners do not increase the quantity or quality of their lawyers substantially more than marginal losers do after the election.
There might be changes to the quality of legal counsel that looking at the lawyers formally registered on the cases does not detect.As an additional test, we consider cases where the lawyers have done all the formal work before the election. 22If electoral winners are convicted at a lower rate due to having better lawyers, we should expect to see no effect of winning the election on such cases.In the Panel A of Table 4, we show the outcome of estimating Eq. 1 for cases where the lawyers made their final allegations before the election. 23hile the relatively small sample size (N ¼ 201) does not allow us to make strong conclusions, we estimate a 28.4 percentage point lower conviction rate for marginal winners than for marginal losers in these cases.In addition to suggesting that lawyers cannot be the only reason marginal winners are convicted at a lower rate, this result indicates that part of the effect comes from politicians influencing the judge, as opposed to from politicians influencing prosecutors and witnesses.

Favor exchange
Mayors might seek to influence the judge through favors or threats.For instance, they can award municipal jobs or contracts to relatives of the judge or use municipal lawyers to help the judge in her work. 24While we do not have direct evidence on such quid pro quo exchanges, we consider two indirect tests.First, it is likely easier for the mayor to cultivate a collusive relationship with the judge if there are few judges in the district. 25Second, some favors-e.g., municipal lawyers helping out in the court-are easier to provide if the distance between the court and the municipality is small.
For the first set of results in Panel B of Table 4, we divide our sample into politicians tried in a district with more than the median (two) number of judges and politicians tried in a district with only one or two judges and estimate Eq. 1 for these two subsamples. 26The point estimate for the politicians tried in districts with few judges is 16 percentage points, compared to only one percentage point for districts with more judges.The difference is statistically significant, with a p-value of 0.02.This result is consistent with the effect being partly driven by a collusive relationship between the judge and the mayor.In the Appendix Table F.1, we show that this result is not driven by the size of the municipalitythe estimated effect of winning the election on convictions is essentially the same for large and small municipalities. 27n the second set of results in Panel B of Table 4, we exploit the fact that judicial districts are typically composed of several municipalities, with the largest municipality hosting the court.We can thus divide our sample into politicians tried in a court located in their municipality and politicians tried in a neighboring municipality.The point estimates are similar for both cases, and the difference is not statistically significant. 28This result suggests that any favors being exchanged are not tied to the location of the court.For instance, the mayor offering of contracts to relatives the judge is a more likely mechanism than, say, municipal lawyers helping out in the court.

Judicial careers
Mayors play no formal role in the promotion of judges. 29Trial judges may, nevertheless, have career incentives to be lenient with mayors.State and federal politicians-connected to local politicians through party networks-determine the salaries of judges, the judiciary's budget, and appoint some appeals court judges.These formal powers mean that the court administration, headed by the appeals court's chief justice, becomes politicized. 30The court administration could use its power over the allocation of substitute judges or promotions by the discretionary vote criterion to exert pressure on trial judges.In this section, we show four pieces of evidence consistent with judges favoring politicians in power due to career concerns.The first three are based on the heterogeneity of the regression discontinuity results.We show that the election has a higher impact on: Judges who tend to switch between judicial districts, untenured judges, and judges sentencing a mayor whose party has appointed appellate judges.The fourth piece of evidence, which we show in the Appendix Section G, suggests that judges who convict elected mayors are more likely to be promoted by seniority than by the appeals court vote.Promotions by seniority usually take longer than Notes: Regression discontinuity estimates where the outcome variable is the difference between the average of the respective variables across all publications made in the Diário de Justiça after the election and the same average before the election.Only cases with publications in the Diário de Justiça both before and after the election.The experience of a lawyer is defined as the number of other Ações de Improbidade she has worked on prior to the election.Her past success is the share of these cases which has lead to a full acquittal, conditional on the case having been decided before the election.Coefficients estimated using the bias-corrected estimator proposed by Calonico, Cattaneo, and Titiunik (2014) with a local linear regression for the estimate and local quadratic regression for the bias-correction.The running variable is the electoral win margin.No control variables.Ações de Improbidade involving candidates for mayor pending at the time of the election.
Standard errors clustered at the municipality by election-year level.Ãp 6 0:1; Ã Ã p 6 0:05; Ã Ã Ãp 6 0:01. 22In our sample-due to overcrowded dockets-it typically takes several years between the lawyers have made their final allegations (alegações finais) and the judge's decision. 23We consider the final allegations to have been made if there has been a publication regarding the case including the term alegações finais in the Diário de Justiça before the election.We show regression discontinuity plots for all heterogeneous effects in Table 4 in the Appendix Fig. F.1. 24Cash bribes is another possible favor.While judicial corruption exists in Brazil, we see cash bribes as unlikely to drive our results since both the marginal loser and the marginal winner arguably would be equally willing to bribe the judge.Also, as shown in Appendix Panel C of F.1, the effect is, if anything, larger for less liquidityconstrained politicians. 25Since cases are randomly assigned to judges, a mayor needs a quid pro quo relationship with all the judges in the district to successfully collude with the judiciary. 26In the Appendix Table F.2, we show heterogeneous effects based on quintiles of the number of judges in the district.The point estimate is largest for the first two quintiles-districts with only one judge. 27Larger municipalities tend to be in districts with more judges.The number of judges in a district, however, also depends on the number and size of the other municipalities in the district and on whether the case is filed in a state court or in a federal court. 28In the Appendix Table F.1, we show the results remain unchanged when we also consider the median distance between the municipality and the court. 29Many of the points in this section also apply to prosecutors.Since we lack data on the careers of the prosecutors, however, we focus on the judges. 30See, e.g., Zaffalon (2018).
promotions by vote and are only based on the number of years in service.
As discussed in Section 2.2.1, judges frequently move between judicial districts, and the appeals court plays an important role in determining these movements.Judges who tend to switch between judicial districts might worry that their decisions in politically sensitive cases could influence their careers.We measure a judge's career instability by the probability that the judge works in different districts in two randomly drawn days in the two years before the election. 31To prevent that the election influences the judge's identity, we focus on the pre-election judge, the judge on the case in the last publication in the Diário de Justiça before the election.Consistent with career concerns, we estimate in the first of results of the Panel C of Table 4 that winning the election reduces convictions by 18 percentage points if the pre-election judge has a higher than median career instability, compared to 1.5 percentage points otherwise.The p-value of this difference in effects is 0.06.
In the second result of Panel C, we use another measure of career stability-whether the judge is tenured.Judges receive tenure after two years on the job and before that are more susceptible to influence by the appeals court.Before a judge's tenure, the appeals court determines their stationing between districts and has the power to fire them from the job.While the limited number of untenured judges in our sample makes it difficult to draw strong conclusions, the point estimates indicate that untenured judges are substantially more influenced by the election: The coefficient for untenured judges is more than twice as large as the coefficient for tenured judges.
Suppose mayors are favored due to the judges' career concerns.Then, we might expect the effect to be larger for politicians from the party in charge of appointing appellate judges-the governor's party for state judges and the president's party for federal judges.In the third result of Panel C, we assess whether the effect is larger if the mayor's party has appointed appellate judges in the past-i.e., current appellate judges are likely to have been appointed by the mayor's party.The point estimate is 57% higher for decisions involving mayors with such party connections than for other decisions, although we cannot reject the null hypothesis of no differences. 32n the Appendix Section G, we select all decisions made during the four years the mayor is still in office and look at the future careers of the judges who made these decisions.Conviction decisions are correlated only with promotions by seniority, which are usually mechanical, in contrast to promotions by a discretionary vote that depend on ballot votes by the appellate judges.
Overall, these results suggest that career concerns might be one reason why judges favor electoral winners.The results are only suggestive, since we do not have access to a source of random variation in career incentives. 336.4.Other mechanisms

Psychological mechanisms
Law enforcers could be more lenient with electoral winners without expecting anything in return.For instance, moving forward with a case involving an elected politician could impede the functioning of local government, and it might be better for society to wait until the politician's mandate is over.Given that we do not see any sign of an increase in the number of decisions appellate judges also having power over the state (federal) judicial budget.In the Appendix Table F.1, Panel D, we separately estimate the effects for politicians aligned with the current governor (president) and for politicians aligned with a past but not the current governor (president).If the effect is driven by the governor's (president's) power over the judicial budget we would expect the effect to be largest for the first cases.The point estimates are, however, almost identical for the two types of politicians.
33 An alternative explanation could be that inexperienced judges-who tend to have higher career instability-are more affected by the election due to other reasons than career concerns.However, in the Appendix Table F.1, Panel E, we show that the point estimate is in fact smaller for judges with below median (eight years) experience than for judges with above median experience.
involving marginal winners after the end of the mandate (Fig. 2), we doubt, however, that this mechanism is the main driver of our result.
Another example of such a mechanism is that judges could wrongly attribute who wins in a close election to be a signal of probity.This mistake could happen if, for instance, electoral winners are less likely to engage in misconduct than electoral losers are, and the judge is not aware that the election was closely contested.We do not have a good way of testing this mechanism.It seems unlikely, however, that the judge, who is required to reside in the judicial district and often would serve as an electoral judge overseeing the local election, does not know that an election was decided with a small win margin.

Destruction of evidence
A large share of the evidence used to convict in an Ação de Improbidade is in the form of documents.If such documents are in the hands of the municipality, an elected politician is in a good position to destroy evidence, making it harder to convict politicians in power.There are two reasons to believe that the destruction of evidence is not the main driver of our result.Most importantly, the prosecution collects most of the evidence during the investigations before they file the case (inquérito civil)-before the election.Also, in Section 6.1 we found that there seems to be an effect for cases ready for decision at the time of the election.There is no production of evidence after the case is ready for decision, except under extraordinary circumstances.34

Media attention
Since judges seem to postpone decisions involving politicians in power, one explanation could be that judges do not want to decide politically sensitive cases to avoid media attention.If this were the mechanism driving our result, we would expect, however, to also see fewer acquittals involving electoral winners, since acquitting an elected politician of misconduct charges have the potential to generate much press attention.Also, in Appendix Table F.1, Panel G, we find no clear relationship between local media presence and the effect on court outcomes of winning the election.

Spillovers from criminal cases
Mayors in Brazil have special privileges in criminal court cases called foro especial por prerrogativa de função.Criminal cases involving mayors are tried in the appeals court, not in the trial courts.The accusation of a politician in an Ação de Improbidade often involve criminal acts, which means that there might be a criminal case running in parallel, investigating some of the same facts.If a mayoral candidate wins the election, any criminal case in the trial court is sent to the appeals court.This change in the jurisdiction over criminal cases can impact Ações de Improbidade for two reasons.First, it becomes harder for the prosecutor in the Ação de Improbidade to collaborate with the prosecutor in the criminal case.Second, the Ação de Improbidade judge must take into account criminal acquittals due to a ruling over a fact.These interlinkages between criminal cases and Ações de Improbidade cannot, however, be the only explanation of our result.We have collected criminal cases involving local politicians in the state of São Paulo, which represent 8% of all cases in our sample.There are four times as many Ações de Improbidade as there are criminal cases.Thus, even if each criminal case is related to an Ação de Improbidade, this channel alone seems unlikely to generate a large effect.

Conclusions
We have documented that having a broad set of formal guarantees of judicial independence is insufficient to prevent politicians in power from receiving a more lenient treatment in court.What can be done to avoid this outcome?While we must leave a careful evaluation of alternative policies to future work, our results nevertheless give us some indications of what could help curtail political influence over judicial decisions.First, the fact that our results are driven by districts with few judges suggests that increasing the size of judicial districts could limit the possibility of collusion between the executive and the judicial branches.Second, we have identified two limitations to the formal independence of Brazilian trial judges that could make them susceptible to political pressure, namely a high propensity to move between judicial districts and a discretionary vote criterion for promotion, both stemming from a discretionary court administration influencing judges' career paths.Reducing this discretion by, for instance, increasing the use of seniority criteria and limiting the number of substitute judges might reduce political influence.

Fig. 1 .
Fig. 1.Regression discontinuity plot.Ações de Improbidade involving candidates for mayor pending at the time of the election.A politician is considered convicted if he or she is recorded to have received a penalty in the Diário de Justiça.The size of the dots indicates the number of observations in each bin.

Fig. 2 .
Fig.2.The timing of the effect.The estimated shares of Ações de Improbidade pending at the time of the election in which the politician has been convicted.Excluding the 2016 election and courts where we do not have judicial data back to 2012.Estimated using the bias-corrected estimator proposed byCalonico et al. (2014) with a local linear regression for the estimate and local quadratic regression for the bias correction.The outcome variable is whether the politician is convicted within x years of the election.The estimates for marginal winners and losers are obtained using the local linear fit.95 percent confidence intervals.The running variable is the electoral win margin.No control variables.Ações de Improbidade involving candidates for mayor pending at the time of the election.Standard errors clustered at the municipality by elec.tion-yearlevel.

Table 1
Summary statistics.

Table 2
Main regression discontinuity results.Regression discontinuity estimates using the bias-corrected estimator proposed byCalonico, Cattaneo, and Titiunik (2014)with a local linear regression for the estimate and local quadratic regression for the bias-correction.The running variable is the electoral win margin.No control variables.Ações de Improbidade involving candidates for mayor pending at the time of the election.Mean Marginal Loser shows the estimated mean of the outcome variable for the marginal loser, using the local linear fit.Standard errors clustered at the municipality by election-year level.Ãp 6 0:1; Ã Ã p 6 0:05; Ã Ã Ãp 6 0:01.

Table 3
The effect of winning the election on the increase in quantity and quality of lawyers.

Table 4
Heterogeneous effects: Testing mechanisms.Regression discontinuity estimates for different sub-samples.p-value of Difference is the p-value of the difference in estimated effects between the two sub-samples, assuming that the two sub-samples are independently drawn.A case is considered ready for decision before the election if there has been a publication regarding the case including the term alegações finais in the Diário de Justiça before the election.The number of judges is calculated as the number of regular judge positions (varas) in the judicial district, excluding substitute judges.The career instability of a judge is the probability that the judge is working in different judicial districts in two randomly chosen days in the two years before the election.A judge is considered not tenured if she has less than two years of experience.To avoid censoring we here exclude cases decided shorter than two years before the start of the Diário de Justiça.A party has appointed appellate judges if it has been the governor's (president's) party for cases in the state (federal) judiciary at any time since 2002.Coefficients estimated using the bias-corrected estimator proposed byCalonico, Cattaneo, and Titiunik (2014)with a local linear regression for the estimate and local quadratic regression for the bias-correction.The running variable is the electoral win margin.No control variables.Ações de Improbidade involving candidates for mayor pending at the time of the election.Mean Marginal Loser shows the estimated mean of the outcome variable for the marginal loser, using the local linear fit.Standard errors clustered at the municipality by election-year level.