Can court decisions on high-profile corruption cases influence how ordinary citizens relate to corruption? Some anti-corruption practitioners expect ‘frying big fish’ to motivate citizens to fight corruption (see Huther and Shah Reference Huther and Shah2000; Klitgaard Reference Klitgaard2017), whereas others warn that drawing attention to corruption can backlash and foster cynicism (Rothstein Reference Rothstein2011: 240). Judicial action against well-known politicians can serve as a deterrent for lower-ranking officials, affecting citizens' direct experience with petty corruption (Kang and Zhu Reference Kang and Zhu2021). But even exposure to grand corruption court cases – through the media, for example – could inspire citizens (Barbabela et al. Reference Barbabela, Pellicer and Wegner2021; Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023; Yair et al. Reference Yair, Sulitzeanu-Kenan and Dotan2020), or discourage them from politically engaging against corruption (Magalhães Reference Magalhães2022; Poertner and Zhang Reference Poertner and Zhang2023).
Previous literature has provided three explanations for the discouraging effects of high-profile corruption court cases on public opinion. The explanations emphasize how citizens' cognitive processes influence their reactions to information about corruption. The first is fatigue, when citizens are unresponsive to new information about corruption (e.g. Barbabela et al. Reference Barbabela, Pellicer and Wegner2021). The second is motivated reasoning (e.g. Anduiza et al. Reference Anduiza, Gallego and Muñoz2013), when citizens ignore information about corruption that clashes with their political preferences. The third is priming, when citizens paradoxically focus on how widespread corruption is instead of appreciating the efforts to control it (e.g. Bauhr and Grimes Reference Bauhr and Grimes2014).
Another perspective argues that information about anti-corruption actions, and not only information about corruption, can breed cynicism. When judicial or prosecutors' efforts against corruption fail, citizens can also be discouraged from acting politically against corruption (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023). The proposition is theoretically reasonable, since we know expectations of formal institutions shape attitudes to corruption (Peiffer and Alvarez Reference Peiffer and Alvarez2016; Rothstein Reference Rothstein2011), but convincing empirical evidence of this specific channel is lacking. Experiments testing citizens' reactions to judicial failures have either yielded contradictory results (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023) or may have served to re-expose citizens to scandalous information about corruption, making it more salient than assessments of anti-corruption performance (Magalhães Reference Magalhães2022; Poertner and Zhang Reference Poertner and Zhang2023).
In this article, I argue that citizens can be discouraged from engaging in anti-corruption if they believe institutions are not genuinely committed to fighting corruption. Citizens resent corrupt practices and exhibit a distinct appreciation for institutions they perceive to be combatting such transgressions (Peiffer and Alvarez Reference Peiffer and Alvarez2016). However, uncovering the true motivations behind anti-corruption efforts is often elusive, not only to scholars but also to ordinary citizens (Barbabela et al. Reference Barbabela, Pellicer and Wegner2021). Taking insights from the literature on the politicization of courts (Gibson and Caldeira Reference Gibson and Caldeira2011; Woodson Reference Woodson2015; Zilis Reference Zilis2021), I argue that citizens use heuristics to form assessments about an institution's commitment to fighting corruption. For courts, disagreement between judges on a high-profile corruption case is potentially a sign of insincere institutional commitment to anti-corruption, thus discouraging citizens from standing up to corruption when given the opportunity.
The article explores an unexpected event during a survey design (UESD) (Muñoz et al. Reference Muñoz, Falcó-Gimeno and Hernández2020), an empirical strategy adopted by a growing number of studies (Ares and Hernández Reference Ares and Hernández2017; Magalhães Reference Magalhães2022; Merler Reference Merler2021; Poertner and Zhang Reference Poertner and Zhang2023; Solaz et al. Reference Solaz, De Vries and De Geus2019). The unexpected event is a 2018 case involving Luís Inácio Lula da Silva – henceforth referred to as Lula – the president of Brazil between 2003 and 2011. The event involved different judges issuing contradictory decisions, all on the same day. Respondents interviewed after the event display lower levels of trust in courts and less willingness to report corruption to authorities. The empirical consequences are not in line with other mechanisms. Contrary to fatigue expectations, public opinion is altered as a consequence of the rulings. Contradicting motivated reasoning, negative reactions to the decisions are not driven by supporters of the defendant. Against priming expectations, corruption perceptions and trust in institutions related to politicians are not affected. The article thus contributes to studies about the public opinion consequences of anti-corruption and the politicization of courts.
Why and how would judicial decisions in a grand corruption case reported in the media affect citizens’ willingness to stand up to corruption?
Some studies focus on the role of courts in providing voters with information about corruption, as in a principal–agent problem (e.g. Costas-Pérez et al. Reference Costas-Pérez, Solé-Ollé and Sorribas-Navarro2012; Peters and Welch Reference Peters and Welch1980). Others approach corruption as a collective-action problem, where citizens' willingness to combat corruption depends on their expectations of others' behaviour and evaluations of public institutions (e.g. Peiffer and Alvarez Reference Peiffer and Alvarez2016; Persson et al. Reference Persson, Rothstein and Teorell2013). In this article, I follow the collective-action perspective and argue that citizens' willingness to stand up to corruption is also driven by their belief that institutions are committed to controlling corruption. Thus, citizens simultaneously hold assessments about how widespread corruption is and about how well institutions attempt to address it. The latter channel, focusing on anti-corruption performance, is a distinct if yet complementary mechanism shaping citizens' attitudes towards fighting corruption, also recently highlighted by Ezequiel Gonzalez-Ocantos and colleagues (Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023).
There are various ways in which different actors and institutions may seek to display their commitment to controlling corruption. For instance, political leaders may allocate more personnel and resources to fighting corruption (Quah Reference Quah2010). Courts may impose harsh sentences on politicians charged with corruption (Yair et al. Reference Yair, Sulitzeanu-Kenan and Dotan2020) and may advance cases involving high-ranking rather than low-ranking officials (Barbabela et al. Reference Barbabela, Pellicer and Wegner2021). Importantly, courts may seek to showcase their impartiality by demonstrating defiance under political pressure (Barbabela et al. Reference Barbabela, Pellicer and Wegner2021), relying on legal rituals to convince the public of their fairness and effectiveness (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023: 169).
Additionally, the public opinion consequences of anti-corruption efforts can be ‘intended’ or ‘unintended’. They are intended when anti-corruption efforts shape citizens' attitudes in a congruent manner, such as by making citizens more likely to support anti-corruption spending (Barbabela et al. Reference Barbabela, Pellicer and Wegner2021), displaying positive feelings towards politics (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023) or voting corrupt politicians out of office (Yair et al. Reference Yair, Sulitzeanu-Kenan and Dotan2020). Conversely, they are unintended when they decrease institutional trust and discourage citizens from participating in politics (Bauhr and Grimes Reference Bauhr and Grimes2014).
Building upon the insights from other studies (Barbabela et al. Reference Barbabela, Pellicer and Wegner2021; Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023), my argument considers citizens' perceptions. Citizens resent corrupt practices and exhibit a distinct appreciation for institutions perceived to be actively combatting such transgressions (Peiffer and Alvarez Reference Peiffer and Alvarez2016). The conceptual approach advanced in this article explicitly acknowledges that citizens cannot directly observe the real motivation of actors charged with fighting corruption. In light of such complexity, citizens rely on heuristics (mental shortcuts) to form assessments about how genuine is the anti-corruption commitment displayed by institutions. If they perceive that displays of commitment are not genuine, I argue, it will lead to unintended effects. The argument about ‘genuine commitment’ is not a refutation of previous arguments, but rather a simplification and an attempt to extend congruent propositions. Citizens can potentially use various shortcuts to assess genuine commitment, but not all shortcuts will actually involve evaluations of anti-corruption performance.
Empirical evidence showing that negative evaluations of anti-corruption performance discourage citizens from standing up to corruption is lacking. Observational studies demonstrate an association between the two elements (Peiffer and Alvarez Reference Peiffer and Alvarez2016), whereas causal evidence is less clear, as in the important book by Gonzalez-Ocantos and colleagues (Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023). Their survey experiment priming respondents to think about problematic aspects of anti-corruption crusades in Brazil does not induce cynicism (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023: 234).Footnote 1 Also, the correlation between citizens' negative feelings about the anti-corruption crusade in Peru and their willingness to stand up against corruption is not statistically significant (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023: 243).
The evidence from natural experiments examining the impact of high-profile corruption cases on public opinion is also unclear. Different types of judicial decisions yield similar unintended effects. Convictions of former heads of state in Argentina and Costa Rica for corruption-related offenses resulted in decreased trust in several institutions, discouraged political participation and heightened perceptions of corruption (Poertner and Zhang Reference Poertner and Zhang2023). Similarly, a court decision in Portugal acquitting a former prime minister of corruption charges led to decreased trust in courts and in politicians, as well as increased perceptions of corruption and a negative spillover effect to other institutions (Magalhães Reference Magalhães2022). As I develop in more detail below, although objectively different, both judicial decisions resulted in substantive changes to the status quo of the defendant, thus they virtually re-exposed citizens to corruption scandals. So it is unclear what the most salient cues affecting public opinion are: information about corruption or anti-corruption performance.
Previous studies confirm that ordinary citizens use heuristics to form assessments of courts' policy preferences (Zilis Reference Zilis2021). Particularly, the literature on politicization of courts, seeking to understand the acceptance of Supreme Court rulings in the US, suggests citizens may use different cues. Many cues reflect the characteristics of judges, such as the fact that a Republican or a Democrat politician appointed them (Rogowski and Stone Reference Rogowski and Stone2021). Other cues are more related to the behaviour of judges but too idiosyncratic of the US, concerning the characteristics of judicial campaigns for office (Gibson and Caldeira Reference Gibson and Caldeira2011). Some studies investigate the importance of the attributes of decisions, such as decision-making modes and judges' level of disagreement (Woodson Reference Woodson2015). These studies suggest citizens value judicial decisions they perceive to be fair, such as when judges seem to be oriented by principled reasons rather than strategically following non-legal reasons.
The specific type of cue this article focuses on refers to judicial performance and is common enough to be relevant in cases other than the US Supreme Court: judicial consistency. High-profile corruption court cases draw a lot of media attention to courts, and judges come under intense public scrutiny. When judges consistently reach similar conclusions, it indicates – at a minimum – a shared understanding of principles of justice and integrity. Conversely, when different judges within the same jurisdiction deliver conflicting verdicts on the same or on similar cases, disagreeing with each other, it potentially raises concerns about the true motives and commitment of different judges or the overall institution. Thus, disagreement creates uncertainty, leading citizens to question whether judgments are influenced by factors other than genuine commitment to fight corruption, such as elite capture, ideological preferences, incompetence and so on.Footnote 2 Therefore, as a cue of insincere institutional commitment to anti-corruption:
Hypothesis 1A: Exposure to judicial inconsistency discourages citizens from standing up against corruption.
In addition, as a specific assessment about the performance of one anti-corruption institution and the channel through which the previous effect operates:
Hypothesis 2A: Exposure to judicial inconsistency decreases citizens' trust in courts.
Apart from the explanation of insincere institutional commitment, there are other potential factors driving the unintended public opinion consequences of anti-corruption efforts. Crucially, these alternative explanations differ in their underlying mechanisms and observable implications.
A first alternative explanation involves corruption fatigue. I follow Catherine De Vries and Hector Solaz's (Reference De Vries and Solaz2017) definition of corruption fatigue: ‘That is, corruption may be so widespread that information about one additional case of corruption may make little difference to people's overall corruption evaluations’ (De Vries and Solaz Reference De Vries and Solaz2017: 398). Extending the notion to be applied to reactions to anti-corruption efforts, Letícia Barbabela et al. (Reference Barbabela, Pellicer and Wegner2021) argue that only ‘costly’ signals of anti-corruption commitment are relevant to shape citizens attitudes. When signals are ‘cheap’ – or easy to emulate – citizens ignore them. That is essentially a null hypothesis:
Hypothesis 1B: Exposure to judicial inconsistency does not affect citizens' willingness to stand up against corruption.
A second explanation involves priming. Priming occurs when exposure to specific stimuli activates related thoughts, affecting subsequent attitudes and behaviours, such as when anti-corruption makes citizens more sensitive to corruption. Priming is problematic because exposure to information about corruption stimulates people to act corruptly (Corbacho et al. Reference Corbacho, Gingerich, Oliveros and Ruiz-Vega2016). Also, information about the involvement of individual politicians in corruption schemes negatively affects institutions or groups incidentally associated with individuals charged with corruption, such as parties and politicians in general (Ares and Hernández Reference Ares and Hernández2017; Bowler and Karp Reference Bowler and Karp2004; Solé-Ollé and Sorribas-Navarro Reference Solé-Ollé and Sorribas-Navarro2018). Previous studies have reported priming in anti-corruption efforts, such as transparency reforms (Bauhr and Grimes Reference Bauhr and Grimes2014) and awareness-raising messages (Chong et al. Reference Chong, De La O, Karlan and Wantchekon2015), showing anti-corruption efforts can heighten perceptions of corruption, lead to citizen disengagement from politics and have widespread negative consequences for institutional trust.
Recent studies show that citizens react to high-profile corruption rulings in a way that resembles priming (Magalhães Reference Magalhães2022; Poertner and Zhang Reference Poertner and Zhang2023). The cases involve former heads of government charged with corruption. After the rulings, not only do corruption perceptions increase, but there is a negative impact on trust in other institutions involving elected politicians, such as the government, parliament and parties. In both studies, the judicial decisions have direct and substantive consequences for the defendant's status, determining whether they are considered innocent or found guilty. As a result, the decisions attract renewed media attention and public discourse to corruption – by revisiting evidence related to the case, for example – rather than to anti-corruption performance. In other words, the effects of priming have a broader scope and impact on various aspects of public opinion than the consequences outlined in H1A and H2A:
Hypothesis 2B: Exposure to judicial inconsistency decreases citizens' trust in the government, in Congress, and in parties.
Hypothesis 3: Exposure to judicial inconsistency increases citizens' perceptions of corruption.
The third explanation involves motivated reasoning. Motivated reasoning is the tendency to interpret information and evidence in a way aligning with one's pre-existing beliefs, values or interests. Corruption studies commonly consider motivated reasoning in the context of partisanship, showing that partisans of politicians involved in corruption schemes are more willing to dismiss the allegations as fake or not serious enough (Anduiza et al. Reference Anduiza, Gallego and Muñoz2013; Solaz et al. Reference Solaz, De Vries and De Geus2019). Extending the same mechanism to reactions to anti-corruption is intuitive. The biased reasoning process can lead to a perception of unfairness in judicial decisions, as individuals are inclined to prioritize their partisan preferences over objective assessments of fairness. That is, assessments about judicial performance may be driven by an affective attachment to the defendant rather than by evaluations of the actual behaviour of judges (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023; Klašnja and Pop-Eleches Reference Klašnja and Pop-Eleches2022). This explanation differs from the earlier main explanation as it predicts that supporters, compared to non-supporters, will be more likely to believe rulings involving the defendant are unfair. Conversely, non-supporters would lack the incentive of negatively evaluating judicial performance when the outcome of the ruling displeases them. Thus:
Hypothesis 1C: Exposure to judicial inconsistency discourages citizens from standing up against corruption, but the effect is stronger amongst supporters of the defendant.
Hypothesis 2C: Exposure to judicial inconsistency decreases citizens' trust in courts, but the effect is stronger amongst supporters of the defendant.
Table 1 summarizes the hypotheses advanced in this study, along with the implications related to the three alternative explanations.
Table 1. Summary of Hypotheses: Citizens' Reactions to Judicial Summary of Hypotheses: Citizens' Reactions to Judicial Inconsistency in the Realm of High-Profile Corruption Court Cases in the Realm of High-Profile Corruption Court Cases
Empirical strategy
A case of judicial inconsistency: decisions involving Lula's habeas corpus in Brazil on 8 July 2018
I test the public opinion consequences of a case of judicial inconsistency using the case of Lula's habeas corpus.Footnote 3 Lula was imprisoned for corruption on 7 April 2018, and on 6 July 2018 lawyers filed a habeas corpus on his behalf (see Figure B.1 in the Supplementary Material for a timeline of events and background information). On 8 July 2018, the case received significant public attention as four judges reached different decisions regarding the possibility of Lula's provisional release from prison. The sequence of inconsistent decisions played out in the public eye over the course of a single day. Importantly, by the evening, the matter was resolved without any objective change to Lula's legal status, as he remained in prison.
Unlike other cases where charges against high-profile politicians were dropped or resulted in convictions (Magalhães Reference Magalhães2022; Poertner and Zhang Reference Poertner and Zhang2023), this particular case did not involve discussions about the politician's involvement in corruption. Instead, the most notable aspect of the event was the inconsistency between judges' decisions, which was easily observed by the public as the case unfolded. As such, the event presents an opportunity to examine how citizens react to judicial performance in a high-profile corruption case. Figure 1 outlines how different judges decided in an inconsistent way on the matter.
Figure 1. Habeas Corpus Decisions on 8 July 2018
This is how the event unfolded. In July 2018 the first-degree judge presiding over Lula's case, now former Judge Moro, was on vacation and so could not formally issue any decisions. On the night of 6 July Lula's lawyers filed a habeas corpus request. Other habeas corpus had been filed on Lula's case, and it is normally not a salient matter. In this particular case, however, the request triggered a sequence of inconsistent decisions, issued by different judges, on the same case. At each point when a new decision was issued, it was reported on the news. The inconsistent decisions unravelled roughly within 10 hours, from morning until evening, all on 8 July. The first decision was issued at 9 a.m. Appellate judge Rogério Favreto, who was on-call on Sunday, analysed the request and determined Lula's provisional release from jail. Moro published a formal statement claiming he would not comply with such a decision, alleging Favreto could not rule on the case. A few hours later, Favreto issued a second order, restating the release. Then another appellate judge, Gebran Neto, determined Lula should not be released as Lula's lawyers had led Favreto to a mistaken decision. Favreto then determined Lula's release a third time. The matter was only resolved after the president of the federal court, Thompson Flores, determined Lula would not be released as the habeas corpus conflicted with a collective decision from the federal court and, as such, would be under the competence of the Superior Court of Justice.
Although the outcome of the decisions was objectively inconsequential, the inconsistency between judges was central to the narrative of the event on the media. Newspapers reporting the event the next day employed war-related terms (see Table C.1 in the Supplementary Material), giving prominence to the behaviour of different judges in the case. Some outlets also framed the event as a scheme to get Lula out of jail, while also praising the role of the judges who avoided it, but even in those cases, the headlines made the disagreement between judges the most salient aspect of the reporting. Even if the ultimate outcome was that Lula remained in jail it was obvious the decision was not unanimous among the judges.
Exploring an unexpected event during survey fieldwork
The estimation strategy relies on the timing of the Latinobarómetro survey, fielded between 27 June and 14 July 2018, and uses the day of the habeas corpus decisions (8 July) to account for citizens' exposure to judicial inconsistency. Political scientists increasingly use this strategy to estimate causal effects of political events on citizens' attitudes (Merler Reference Merler2021), particularly in the study of reactions to corruption scandals (Ares and Hernández Reference Ares and Hernández2017; Solaz et al. Reference Solaz, De Vries and De Geus2019), and more recently, to judicial decisions involving former heads of state implicated in corruption (Magalhães Reference Magalhães2022; Poertner and Zhang Reference Poertner and Zhang2023). The treatment (D i) is a binary variable coded in the following way:
The most reliable way of eliminating systematic differences across treatment and control groups is random assignment. Conversely, in studies using unexpected event during survey (UEDS), the argument is that certain events can be considered ‘as-good-as-random’. The full specification for the regions fixed effect model to estimate the difference in means across treated and control groups (ρ) is stated as follows:
where Y ir is the outcome variable (willingness to stand up to corruption or trust in courts), α r is a set of region dummies, D ir is the treatment indicator, X ir is a vector of individual-level covariates, and η ir is the individual-level error term.
The UEDS relies on two assumptions: (1) the only channel through which the timing of the survey affects the outcome must be exposure to the event (excludability assumption); (2) the probability of a respondent being interviewed must be independent from the potential outcomes (ignorability assumption). Striving to observe the considerations by Jordi Muñoz et al. (Reference Muñoz, Falcó-Gimeno and Hernández2020), I evaluate the most pressing risks to both assumptions, explaining the strategies I employed to address them.
Firstly, related to the ignorability assumption, one of the potential risks of using political events as an instrument is noncompliance. Still, assuming that respondents interviewed after 8 July 2018 were aware of the habeas corpus event is plausible as it was a highly salient episode. Figure 2 shows the relative frequency of searches on Google for the term ‘Luís Inácio Lula da Silva’ between 15 June and 30 July 2018. The shaded area represents the fieldwork of the Latinobarómetro survey, and the dotted red line the day of the habeas corpus event.
Figure 2. Google for the Term ‘Luís Inácio Lula da Silva’ between 15 June and 30 July 2018
The ignorability assumption also leads to some important considerations about the survey fieldwork structure. Latinobarómetro is a face-to-face survey conducted using a regional roll-out. To dismiss the risk of attrition, I show that the share of unsuccessful surveys before and after the treatment is similar (see Table F.1 in the Supplementary Material). To address regional imbalance, which might correlate with unobserved characteristics, I employ nearest-neighbour matching, which was the most appropriate method to achieve covariate balance in this case (see Figure G.2 in the Supplementary Material). Following advice by Daniel Ho et al. (Reference Ho, Imai, King and Stuart2007), both in matching and in the analysis, I use a set of pre-treatment relevant covariates, including sociodemographic indicators chosen based on previous studies investigating attitudes towards institutions and corruption (Ares and Hernández Reference Ares and Hernández2017; Peiffer and Alvarez Reference Peiffer and Alvarez2016; Solaz et al. Reference Solaz, De Vries and De Geus2019). I also include geographical indicators, as support for Lula around the time was stronger in the north and north-east, given the expansion of government benefits during his government (Hunter and Power Reference Hunter and Power2007) (see Table E.1 in the Supplementary Material for information about variable coding).
Table 2 shows the balance between treatment and control groups regarding sociodemographic and regional indicators before and after matching. As shown in Table 2, the matching reduces the mean differences between the treatment and control groups, clearly improving the similarity across the sample. In the original sample, the treated group is on average older, less educated in formal terms, but displays slightly higher levels of political knowledge and is more likely to be employed. There are no relevant differences in terms of Catholicism, gender, socioeconomic status or access to social benefits. There are, however, some regional imbalances: respondents from the south-east and the north are more likely to be in the control than in the treatment group, whereas the opposite is the case with the centre-west and the south. The only region represented in a balanced way across treatment and control in the original sample is the north-east. In the matched sample the difference between treated and control groups across all sociodemographic indicators is not statistically significant, and the means are also close across the two groups. The only type of regional imbalance in the matched sample concerns the centre-west region, whose respondents are still more likely to belong to the treated group. This is not a problem as the centre-west was not a distinctive region in terms of Lula support at the time (Hunter and Power Reference Hunter and Power2007), and there is balance in terms of individual socioeconomic indicators. To account for the imbalance, however, I include region fixed effects when running the regressions to test the hypotheses, as well as indicators for all the other variables listed on Table 2.
Table 2. Means across Treatment and Control in Samples before and after Matching
Also importantly, related to the excludability assumption, it could be that respondents anticipate the event, and as such adjust their attitudes accordingly. The habeas corpus decisions on 8 July were unexpected because they were triggered by a request made by Lula's lawyers on 6 July, but it was not on the public radar until the judicial decisions were made.Footnote 4 Unlike other types of judicial decisions on Lula's case, there is no evidence of anticipation in the habeas corpus event (see Figure D.1 in the Supplementary Material). Additionally, I reproduce the main analysis using a more conservative approach to code the treatment, excluding observations from 6 July to 8 July (see Table J.2 in the Supplementary Material). I also test a placebo version of the treatment to rule out the presence of temporal trends prior to the habeas corpus event (see Table J.1 in the Supplementary Material).
Testing the insincere institutional commitment versus alternative explanations
Ideally, testing the hypotheses would involve asking citizens about their perceptions of courts' commitment to control corruption, whether genuine or insincere. A limitation of this study is the impossibility of measuring such perceptions with the data at hand. Rather, citizens' interpretation of the event is a theoretical part of the argument connecting a judicial inconsistency event to a change in a particular set of citizens' attitudes, operationalizable using Latinobarómetro questions. The fact that the treatment comprises judicial inconsistency and the pattern of attitudes subjected to change differ with respect to other explanations (see Table 1) is the evidence presented here. Table 3 indicates all outcome variables used to test different hypotheses. I report the specific survey questions used to measure them in the Supplementary Material (Table E.1).
Table 3. Descriptive Statistics of Outcome Variables
As indicated in Table 3, all these are ordinal variables (see Table M.1 in the Supplementary Material for descriptives of categorical data), although their range varies. As the main estimation strategy, I use them as standardized continuous variables to make interpretation more intuitive, but the result is similar when using non-linear models (see Tables M.2 and M.3 in the Supplementary Material). For robustness, I also test the corruption perceptions hypotheses using a salience indicator, with similar results (see Table M.3, model 6, in the Supplementary Material).
To dismiss the motivated reasoning explanation it is important to account for the possibility that citizens attached to Lula may react to the event in a significantly different way from others. This would weaken the claim that it is the evaluation of judges' behaviour – not the identity of the defendant – that matters to citizens reacting to the habeas corpus event.
I account for the existence of heterogeneous effects affecting the results using two different proxies of support for Lula. The first is a dummy variable indicating whether a respondent receives a social benefit, given the expansion of such policies during Lula's government – which contributes to Lula's popularity (Hunter and Power Reference Hunter and Power2007). The second is a dummy indicating support for the Workers' Party (Partido dos Trabalhadores – PT), of which Lula is one of the founders and main figures, following a more established practice in the literature (Anduiza et al. Reference Anduiza, Gallego and Muñoz2013; Magalhães Reference Magalhães2022; Solaz et al. Reference Solaz, De Vries and De Geus2019). In Section H in the Supplementary Material, I elaborate on the appropriateness of the choice of the partisanship dummy and I conduct tests showing that PT support does not predict treatment assignment (see Tables H.2 and H.3) – following the strategy from Solaz et al. (Reference Solaz, De Vries and De Geus2019). I also show that exposure to the treatment does not affect respondents' partisan attachment or ideology (Table H.4). All main results show tests using both strategies.
Results
Table 4 shows the impact of judicial inconsistency on citizens' willingness to report corruption to authorities. The first column shows the treatment coefficient using only the individual-level covariates included in the matching – that is, PT support is not included. The results of that column can indicate support for the fatigue hypothesis (in case of null results) or, in case of negative results, for other hypotheses (insincere institutional commitment and priming). Columns 2 and 3 include tests for the motivated reasoning hypothesis (H1C). The second column shows the results including PT support as a covariate in the analysis and an interaction with the treatment. The third column shows results using only the individual-level covariates included in the matching and an interaction between receiving social benefit and the treatment.
Table 4. H1A–C – Effects on Willingness to Report Corruption
Note: Table shows ordinary least squares regression coefficients with standard errors in parentheses using standardized dependent variables and matched samples. Individual-level covariates omitted from output: age, age squared, gender, education, socioeconomic status, unemployment, Catholicism and political knowledge. Regional fixed effects are included for the five regions of Brazil (centre-west, north, north-east, south, south-east). + p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
The results in Table 4, column 1, indicate that individuals interviewed after the habeas corpus event are less willing to report corruption than individuals interviewed before it. The results are negative and significant at the 99% level. In terms of magnitude, individuals in the treatment group display a 0.2 standard deviation decrease in willingness to report corruption. The effect is small but as it is significant, it allows us to reject the fatigue hypothesis (H1B), according to which citizens would be unaffected by the event. Still, considering the different levels comprising the treatment and the outcome variable – one referring to a high-profile corruption case, the other to citizens' attitudes about how to react to corruption in their daily lives (an indicator for ‘standing up to corruption’) – the effects indicate worrying repercussions of corruption court cases, in line with H1A, related to the insincere institutional commitment explanation.
I proceed to test the evidence related to the motivated reasoning explanation (H1C). As previously mentioned concerning the outcome of rulings, Lula supporters would have more reasons to be displeased, since Lula remained in prison. Nevertheless, if the disagreement between judges was perceived as problematic in itself, non-supporters would also have reasons to be displeased. I investigate the matter using two different indicators. It is reassuring that the results in columns 2 and 3 are similar in terms of direction and magnitude. In both cases, the interaction terms are not statistically significant, even at the 90% level, and they go in the opposite direction to that predicted by the hypothesis, suggesting, if anything, that people with no particular attachment to Lula drive the result. It is amongst citizens who have fewer incentives to believe in a witch-hunt against Lula that the negative effect of judicial inconsistency is most pronounced.
The results allow us to reject the motivated reasoning explanation (H1C). What the positive interaction – not significant at the 90% levelFootnote 5 – potentially indicates is non-supporters' tendency to become more suspicious of anti-corruption efforts. Prior to the event it was expected that Lula supporters would be more suspicious and likely to identify it as selective enforcement (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023: 214). The habeas corpus event made the non-supporters suspicious as well.
So far, without investigating how other variables are affected, it is not possible to distinguish whether the consequences of the habeas corpus event could also be explained by the priming mechanism, as it also implicates a negative effect on citizens' attitudes towards anti-corruption. As an attempt to distinguish both mechanisms, Table 5 shows the effects on trust in different institutions and corruption perceptions. Whereas the insincere institutional commitment explanation expects only a negative effect on trust in courts, the priming explanation suggests a more general effect, especially affecting institutions closely related to politicians and increasing corruption perceptions.
Table 5. H2A–C and H3: Effects on Institutional Trust and Corruption Perceptions
Note: Table shows ordinary least squares regression coefficients with standard errors in parentheses using standardized dependent variables and matched samples. Individual-level covariates omitted from output: age, age squared, gender, education, socio-economic status, unemployment, Catholicism, social benefit and political knowledge. Regional-fixed effects are included for the five regions of Brazil (centre-west, north, north-east, south, south-east). + p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001.
Out of the five different regressions displayed in each of the columns in Table 5, the only substantial effect is on trust in courts. The effect is negative and significant at the 99% level. The other coefficients are small and not significant. The institutionally specific effect is in line with the insincere institutional commitment explanation (H2A), with no support for the hypotheses in favour of the priming explanation (H2B and H3). Regarding the coefficient magnitude, the effects in column 1 are slightly bigger than the effects on willingness to report corruption. I also include the interactions to test the existence of motivated reasoning with respect to these variables (H2C) and it is clear that Lula supporters are not the ones driving the results.
I repeat the analysis using the original unmatched samples, and overall the results go in line with the main analysis, although the results for willingness to report corruption are slightly less robust than the ones involving trust in courts (see Tables N.1 and N.2 in the Supplementary Material). I have already addressed some concerns involving the empirical strategy, using a range of robustness tests in the Supplementary Material. I also perform three more sets of robustness tests.
The first is to re-analyse the data using a regression discontinuity design. In such a design, the probability of treatment assignment changes at the cut-off (Cattaneo et al. Reference Cattaneo, Idrobo and Titiunik2020), which in this case is being interviewed after 8 July 2018. Figure 3 shows two panels: on the left is willingness to report corruption and on the right trust in courts. The standardized outcome variables are on the y-axis and the days to treatment (the running variable) on the x-axis. The vertical dotted line represents 8 July (Day 0). Two linear regression lines are plotted in each panel, one regressing the outcome variable on the running variable for citizens interviewed before 8 July, and the other for those interviewed after. The jittered scatterplot on the background indicates the raw observations.
Figure 3. Change in Willingness to Report Corruption and in Trust in Courts on the Day of Habeas Corpus Decisions
In both cases, the change during the days following Day 0 suggests a shift to a negative trend. It does not seem as if the negative shift in either case is due to a mere disturbance on the day after the habeas corpus event. The number of observations around the cut-off does not seem to vary much, which is reassuring as a drastic change could be a source of bias. The results relating to trust in courts seem to be more robust, as indicated by the shaded 95% confidence intervals around the regression lines. In the Supplementary Material I show the coefficients for each day, using 7 July 2018 as a reference, and a trend of negative and statistically significant results takes shape after Day 1 (Table I.1). I also reproduce the main results using different windows around the cut-off: five-day (Table I.2) and three-day (Table I.3), and the treatment coefficient increases as windows get narrower, where individuals are more likely to be ‘exchangeable’.
The second robustness tests aim to rule out the possibility of collateral events driving the results. I reproduce the analysis using data from Uruguay, where the fieldwork almost entirely overlaps with the one in Brazil (see Table L.2 in the Supplementary Material). The absence of effects in the neighbouring South American country corroborates the absence of regional trends. The third and final test examines potential alternative explanations related to access to media (see Table K.1 in the Supplementary Material), and shows that the event in itself did not affect institutional trust by way of making individuals change their media information consumption patterns and, in line with the mechanism here proposed, the effect is stronger for individuals with higher levels of political knowledge.
Conclusion
High-profile corruption court cases can have repercussions for how ordinary citizens relate to corruption. Worryingly, such cases are not always ignored by citizens, but seem to foster pessimism and disillusionment with politics. The discouraging effect can stem from citizens' affective attachment to politicians charged with corruption – which clouds their judgement – or from the emphasis on corruption resulting from judicial action. In this study, I consider an additional possibility: that discouragement also comes from citizens' negative appraisals of anti-corruption performance. The evidence suggests that exposure to judicial inconsistency makes citizens less willing to report corruption to authorities. This effect seems to operate through a decrease in trust in courts. Trust in other institutions – the government, congress and parties – is not affected and neither are corruption perceptions. The results are also not stronger amongst supporters of the defendant, which for affective reasons could overlook actual judicial performance to claim anti-corruption drives are biased. Three aspects corroborate the soundness of the empirical strategy. First, newspaper headlines indicate that what was most salient about the event was the disagreement between judges. Second, the treatment comprises an exogenous event, reducing the risk of reverse causality. Third, the treatment consists of a real event, strengthening external validity.
I argue that citizens value institutions committed to anti-corruption and use heuristics to form assessments about whether such commitment is genuine or insincere. When displays of commitment are perceived as insincere, it discourages citizens from standing up to corruption. The case tested in this article explores judicial inconsistency as a cue of insincere anti-corruption commitment. Importantly, the inconsistency between judges did not affect the outcome of the case, as it did not change the status of the defendant imprisoned.
The article contributes to the study of public opinion consequences of anti-corruption. By examining a case where there was no change in the politician's status, this study provides clearer results regarding the consequences of negative perceptions of anti-corruption performance. The results contrast with previous studies, which may have captured citizens' reactions to scandalous information about corruption instead (Magalhães Reference Magalhães2022; Poertner and Zhang Reference Poertner and Zhang2023). Also contrasting with previous evidence, the negative consequences of judicial rulings do not seem to be driven by citizens with a particular attachment to the defendant (Gonzalez-Ocantos et al. Reference Gonzalez-Ocantos, Chirinos, Pavão and Hidalgo2023; Klašnja and Pop-Eleches Reference Klašnja and Pop-Eleches2022), possibly because assessments about anti-corruption commitment matter more when there is uncertainty regarding the outcome of the corruption case. The article also contributes to the literature on the politicization of courts, showing judges' disagreements affect not only acceptance of decisions (Woodson Reference Woodson2015; Zilis Reference Zilis2021) but also attitudes towards corruption, and that the attributes of decisions matter beyond the context of the US Supreme Court.
There are, however, limitations. Due to data constraints, the study does not directly test whether judicial inconsistency shapes perceptions of insincere commitment to anti-corruption. Rather, the claim is part of the theoretical argument connecting judicial inconsistency to citizens' attitudes. Future studies could test different steps of the mechanism advanced here, particularly incorporating strategies more advanced in the study of the politicization of courts, such as the use of survey experiments. For instance, conjoint experiments can help understand what cues are more important to citizens and how they vary.
To conclude, I do not argue against the prosecution and punishment of powerful individuals involved in corruption. In a democratic society, it is important to prevent powerful individuals from exploiting their positions for personal benefit; if they do, they should face appropriate sanctions. However, judges and courts also bear responsibility for maintaining practices and behaviours demonstrating consistency and genuine commitment to combating corruption. Citizens are paying attention.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/gov.2023.36.
Acknowledgements
I want to express my gratitude to Marcia Grimes, Honorata Mazepus, Mariana Borges, Miquel Pellicer, Sergiu Gherghina, Wil Hout, Emma Murphy, Jesper Lindqvist, Martijn Schoonvelde and Eva Wegner for their insightful feedback on earlier drafts of the manuscript. I would also like to thank all the participants of the Citizen Perspectives on Accountability in Developing Democracies workshop at the ECPR joint sessions 2021. Additionally, I appreciate the helpful comments provided by three anonymous reviewers and the editors of Government and Opposition, which greatly improved the quality of this article. This research benefited from funding from the Iseult Honohan Doctoral Scholarship.
Open access
