The new corruption crusaders: Security sector ties as an anti-corruption voting heuristic

Abstract

Despite the salience of corruption in elections in Latin America and beyond, it remains unclear what makes certain candidates attractive to voters as solutions to address corruption. Building on studies about the effect of candidates’ professional affiliation on voting behavior, we hypothesize that police and military officers are perceived to be more competent to address corruption. We test our theoretical expectations through an online survey of Brazilian voters with an image-based factorial experiment that presents respondents with three randomly generated handbills, varying candidates’ professional affiliations and potential confounders, such as economic policy, insider versus outsider status, and demographic features. Our results demonstrate that candidates affiliated with the police or the military are perceived to be more effective at reducing corruption, all else equal. The effect of police or military professions on candidates’ perceived effectiveness to fight corruption varies according to respondents’ ideology and is particularly significant among conservative voters.

Over the past decades, several scholars have sought to understand the consequences of corruption on voting behavior and public attitudes toward politics. Some scholars focus on the conditions under which voters punish politicians accused of corruption (Pavão 2018; Weitz-Shapiro and Winters 2017; Anduiza et al. 2013). Other scholars note that corruption also affects public trust in political institutions (Anderson and Tverdova 2003) and citizens’ engagement with the political system (González Ocantos et al. 2023). Still others show that corruption scandals increase the chances that voters support outsider candidates; that is, noncareer politicians (Petersen 2020; González Ocantos et al. 2023).

However, previous studies have left untheorized why certain candidates are perceived to be more competent to address corruption than others. This is an important omission because there is suggestive evidence that candidates from specific professions may benefit more than others in elections. For example, in Brazil, after several corruption scandals uncovered by a large-scale investigation known as Operation Car Wash (Operação Lava Jato), there was a substantial increase in the number of police and military who won elected positions (Gelape et al. 2018). ¹ This phenomenon was not restricted to Brazil: similar candidates emerged in other countries across Latin America (Rosen 2021; Bonner 2019) and beyond (Garrido 2020). Nevertheless, it is not clear if support for these candidates stems from voters’ beliefs that police and military officers are more viable alternatives to fight corruption.

To understand what voters find compelling in anti-corruption candidates, the literature on the effect of professional titles on voting behavior can provide some guidance (Carnes and Sadin 2015; McDonald et al. 2020; Boas 2014). When candidates display their profession in their campaign advertisements, they are signaling specific areas of expertise, which can lead voters to evaluate these candidates positively in these respective areas (Pedersen et al. 2019; Coffé and Theiss-Morse 2016; Teigen 2013). For example, candidates who work in the education sector are perceived to be more competent to serve on the school board (Atkeson and Hamel 2020). Professional affiliations may also provide hints about the candidate’s ideology and which areas the candidate is expected to focus on (Crowder-Meyer et al. 2020; Kirkland and Coppock 2018; McDermott and Panagopoulos 2015).

We build on this literature to test whether and how candidates’ occupation affects their perceived competence to address corruption. Our research draws on an original experiment embedded in a national sample of 1,052 respondents in Brazil recruited by the marketing research firm Netquest. Brazil presents a particularly good case to explore the role of professional heuristics in voting because Brazil’s open-list proportional representation system requires candidates to compete within their own party for votes (Samuels 2003) and, as a result, to develop strong personal brands, giving outsized importance to candidates’ individual attributes compared to their party affiliation. Even the ballot itself is a space to advertise professional affiliations, with professional titles such as Delegado (Sheriff), Pastor (Pastor), Doutor (Doctor), or Professor (Teacher) appearing frequently in the candidate’s self-reported ballot name (Boas 2014; Novaes 2023). While mayoral elections, by definition, have a district magnitude of 1 and are therefore nonproportional, mayoral campaigns frequently follow the same logic of personal branding, whether because of the generally weak party brands or because candidates often have prior experience running in legislative elections.

Our conjoint experiment presented respondents with three randomly generated hypothetical campaign handbills, known as santinhos. The candidates presented in these santinhos varied along several dimensions, including profession, experience in politics, and economic ideology. Respondents were asked to rate each candidate along several different dimensions, including the likelihood of support and effectiveness in managing corruption, public security, and unemployment.

Results show that respondents—in particular respondents who identified as conservative—rated police and military officers as more competent to address corruption relative to otherwise identical profiles. Results also showed that affiliation with the police or the military also increased support for the candidates among conservative respondents. Furthermore, the results showed that, surprisingly, teachers are also perceived to be more competent to address corruption.

We leverage qualitative evidence elicited from post-survey open-ended questions to hypothesize two mechanisms for how candidates’ professional affiliation affects their perceived competence to address corruption. One was being a member of a profession understood to be characterized by high levels of integrity and discipline, and the other, having previous experience working on corruption cases. Respondents who rated military officers and teachers as competent to address corruption typically talked about the moral character of these professions, emphasizing how their integrity could translate into anti-corruption work. In contrast, police officer candidates received comments highlighting their experience in corruption investigations, although police officers are also likely to benefit from the first mechanism, given the militarized nature of policing in Brazil and the relatively high trust in the police compared to political institutions (Corporación Latinobarómetro 2021).

Because of the generally low trust that voters have in the police in Latin America (Malone and Dammert, 2021), and that police officers are frequently the very face of corruption in many contexts, this relationship should not necessarily be expected. However, the Brazilian context suggests that while trust in the police and military might be low, the abysmal state of trust in political institutions and the cultural influence of high-profile anti-corruption investigations, such as Lava Jato, mean that these “iron fist” approaches to combating crime enjoy a comparative advantage in the eyes of voters.

Professional Affiliation and Voting Behavior

Given the complexity of policy issues in the modern era, many voters have neither the time nor the inclination to make choices fully informed by candidate positions on substantive issues (Downs 1957; Palfrey and Poole 1987). Instead, voters frequently rely on heuristics, such as party (Rahn 1993), gender (Krupnikov et al. 2016), race (Weaver 2012), age (Webster and Pierce 2019), religion (Calfano and Djupe 2009), disability (Reher 2021), appearance (Ahler et al. 2017), and even ostensibly nonpolitical cultural images (Hiaeshutter-Rice et al. 2021) to get an idea of which candidates generally align most closely with their worldviews. While generally thought of as imperfect shortcuts for low-information voters, heuristics contain important information for voters and are helpful for organizing electoral competition. Carnes and Sadin (2015), for instance, show that class heuristics can lead voters to make relatively accurate conclusions about candidate economic policy positions.

Recent scholarship has focused on professional affiliations as key heuristics for voters in the United States (Carnes and Sadin 2015; McDonald et al. 2020; McDermott 2005) and beyond (Boas 2014; Carnes and Lupu 2016). Some studies find an effect of occupation on vote preference regardless of voters’ individual characteristics. For example, Campbell and Cowley (2014) find that doctors receive more support than candidates with no visible occupation, whereas others argue that veterans, on average, receive more support from voters than civilians do (Hardy et al. 2019). In another example, Sigelman and Sigelman (1982) find that business owners are more likely to receive votes than attorneys, while others show that lawyers are perceived to be more competent than warehouse assistants (Pedersen et al. 2019).

For some authors, the effect of professional affiliation on voter support is moderated by voters’ party or religious identification. For example, candidates who are business executives (Crowder-Meyer et al. 2020), police officers (Kirkland and Coppock 2018), or who have experience serving in the military (McDermott and Panagopoulos 2015) tend to receive more support among Republicans. In contrast, Democrats prefer teachers (Crowder-Meyer et al. 2020). Other studies find that pastors enjoy more support than other candidates, but only among Evangelical Christians (Boas 2014).

Previous studies have also examined why candidates’ occupation affects voters’ choice, showing, for example, that professional affiliations signal areas of expertise (Pedersen et al. 2019), thereby leading voters to evaluate candidates as more competent to work on issues associated with that expertise. For example, Teigen (2013) shows that professional experience in the army may affect a candidate’s level of support because voters perceive candidates who are veterans to be more competent to handle national defense and security policies. Similarly, Atkeson and Hamel (2020) find that voters prefer candidates for school boards who have worked in education (Yanez Ruiz 2018). In addition, shows that business executives are rated as more competent in the economy and doctors are rated as more competent in healthcare policy (see also Coffé and Theiss-Morse 2016). Although previous studies have examined the effect of candidates’ occupations on their competence to address issues related to national security, the economy, or healthcare, they have left untheorized the question of whether professional affiliations affect candidates’ perceived competence to fight corruption. This is an important omission because corruption is often an important issue for voters, in particular across Latin America (Carlin et al. 2015). We build on the literature on occupations as voting heuristics to develop a framework to explain whether and how candidates’ professional affiliation affects their perceived competence to fight corruption.

Professional Affiliations and Corruption

We hypothesize that candidates from the military and the police will be perceived as more competent to address corruption when compared to candidates with no professional affiliation. We expect that there are two non–mutually exclusive mechanisms that lead voters to rate police and military officers as more competent to fight corruption.

First, occupation can be a strong indicator of worldview and morality. This signal is especially important in a context such as Brazil, where party brands are generally weak and legislative campaigns are highly personalistic, with candidates frequently relying on biographical details to differentiate themselves from their competition. We hypothesize that military officers will be perceived as more competent to address corruption—compared to candidates with no professional affiliation—because the military is a profession typically associated with a morality of integrity, authority, and discipline (Schumm et al. 2003), all of which may be viewed as important characteristics in the fight against corruption. Indeed, in Brazil and across the vast majority of countries in Latin America, people tend to trust the army substantially more than other branches of government, such as the legislative or executive (Corporación Latinobarómetro 2021). Symbols of militarism are often used to evoke ideas of order and respect, and the “quasi-military” organization of police departments was begun at least partly to curb public perceptions of police corruption (Bittner 1970). Police and military candidates can credibly associate themselves with these symbols, and therefore, rightly or wrongly, with an image of honor and respectability.

In addition to a relationship with symbols of order, we expect police officer candidates also to benefit because the police have the institutional mandate to enforce the law against individuals who violate society’s moral codes. However, this effect may vary according to the population’s level of trust in the police. In Brazil, for example, approximately half the population somewhat or strongly trusts the police, but confidence levels in the police are substantially lower across the vast majority of countries in the region (Corporación Latinobarómetro 2021).

Besides the feature of signaling specific moral characteristics that voters perceive to be useful for fighting corruption, we also expect a second mechanism to be at play. Occupations matter to voters because they indicate the areas in which candidates have professional experience (see also Yanez Ruiz 2018; Campbell and Cowley 2014; Coffé and Theiss-Morse 2016). Although the work of police officers generally revolves around public security, we expect that police officers will be perceived as more competent to address corruption compared to candidates with no professional affiliation because police officers—like prosecutors—typically also have the legal mandate to investigate corruption. In other words, police officers may be perceived as more competent to address corruption because voters may believe that these candidates already worked on corruption investigations prior to running for office. This effect may vary, however, according to the extent to which police officers play a prominent role in corruption investigations. ²

Because of these two mechanisms—cultural associations between symbols of security sector professions, such as the uniform, and abstract concepts of order and discipline and perceived professional experience with anti-corruption investigations—we propose the following hypothesis:

H1: Voters rate military and police officers as more competent to combat corruption than candidates with no professional affiliation.

Heterogeneous Effects: How Progressives and Conservatives Respond to Security Sector Candidates

We hypothesize that military and police officers are more likely, on average, to be rated as competent to address corruption, but we expect this effect to vary according to voters’ ideology. This is because progressive and conservative voters tend to have different attitudes toward certain professions. For example, in the United States, conservative voters are generally more favorable to the military and the police (McDermott and Panagopoulos 2015). While this study was specific to the context of the United States, we expect that this effect is also present in other countries where conservative-minded people have higher trust in the police and the military than progressives, such as Brazil and most places across Latin America (Holland 2013).

We also expect the effect of professional affiliation on support for a candidate to vary according to voters’ ideology. This is because even if progressive voters evaluate candidates from the military or from the police as competent to address corruption, these voters may choose to support candidates based on other policy issues in which police or military officer candidates fare worse.

These heterogeneous effects based on ideology might vary according to the context in which voters are embedded, in part because their reactions to security sector candidates may depend on which parties were targeted in corruption scandals that preceded elections. In Brazil, for example, the two largest corruption scandals that preceded our experiment—the Mensalão (monthly bribes) and Lava Jato (car wash) investigations—disproportionately targeted left-wing and center parties (Da Ros and Taylor 2022; González Ocantos et al. 2023; Kerche 2022). As a result, Brazil is a context in which we would expect conservatives to have more positive attitudes toward candidates from the police and the military not only because they tend to trust these organizations more, but also because conservatives recently witnessed police officers arrest and prosecute multiple high-level politicians from center- and left-wing parties. Therefore, while conservatives, on average, tend to trust the police and military more, it is also possible that progressive voters show more support and evaluate more positively candidates from the police and the military in contexts where recent corruption scandals targeted right-wing politicians, a topic we will return to in the conclusion when discussing the boundary conditions of our theory.

H2a; Conservative voters are more likely to rate candidates with occupations in the police or military as more competent to address corruption when compared to progressive voters.

H2b: Conservative voters are more likely to support candidates with occupations in the police or military when compared to progressive voters.

Corruption and Elections in Brazil: The Rise of Law and Order Candidacies

Corruption has been one of the most salient issues in Brazil in past elections (Pavão 2018), in part because of several recent scandals of high-level corruption in the government. For example, in 2012, several politicians were convicted of receiving bribes from the governing party in exchange for legislative support for the president’s agenda in a case that became known as Mensalão (Praça and Taylor 2014). Between 2014 and 2021, prosecutors launched an even bigger investigation called Car Wash (Operação Lava Jato), which uncovered schemes of bribery and kickbacks between public agencies and construction companies and resulted in the conviction of more than two hundred politicians and business executives (Ministério Público Federal 2020).

As a result of successive corruption scandals, several candidates started to run for office leveraging an anti-corruption campaign (Rennó 2020). In the aftermath of Lava Jato, police officers and candidates with military backgrounds were elected in higher numbers when compared to previous years (Gelape et al. 2018). Berlatto et al. (2016) note that, in addition to increasing in number, there seemed to be a qualitative shift in their identity, with police and military candidates running with smaller, ideologically ambiguous parties, as opposed to the large right-wing parties they had traditionally attached themselves to. However, police and military candidates remain ideologically heterogeneous, with many running with left-wing parties.

The growth in presence of police and military candidates is not limited to the federal level. Figure 1 plots the growth in police candidacies in mayoral elections over time. Mayoral races pose an interesting case for law and order candidates because while the Brazilian Constitution endows municipal governments with significant policymaking power in many areas, municipal governments wield little influence over questions of public security (Corralo and Kemmerich 2016). The police, for example, are commanded by state governments, meaning that mayors hold little direct power over their conduct. Given that mayors have comparatively less influence over public security policies, police officer candidates need to leverage other areas of strength from their professional histories. ³

Figure 1. Proportion of Mayoral Candidates Reported as Police Officers to the TSE Since 2004.

One possible source of comparative strength for military and law enforcement agents is their perceived ability to combat corruption. Several military and police officers have launched electoral campaigns specifically focused on fighting corruption. Figure 2, for example, shows a captain in the military police of São Paulo running for Congress who explicitly combines the issues of corruption and public security in his campaign advertisement, inserting the image of former judge Sergio Moro, famous for his role in the Lava Jato corruption probes. Similarly, figures 3a and 3b show a military police officer and a federal police agent running for Congress and mayor, respectively, and drawing on messages related to corruption in their campaign advertisements. In figure 3a, the military police officer from Paraná expresses his support for a punitive solution to government corruption: increasing the sentence for corruption crimes. In figure 3b, a federal police agent poses with handcuffs and an anti-corruption slogan.

Figure 2. Campaign Advertisement of Military Police Captain Running for Congress.

Figure 3. Campaign Advertisements of Military Police Officer and Federal Police Agent Running for Office.

These campaign images suggest that police and military officers at least believe that their professional identity combines well with an anti-corruption message. However, it remains unclear whether and why voters see these candidates as more competent to address corruption. Our experiment clarifies these questions and opens up new avenues for future research on the effect of candidates’ professional affiliation on voting behavior.

Experimental Design

To evaluate our hypotheses about military and police officer affiliations on voter perceptions, we fielded an online survey on the Qualtrics platform to a diverse sample of Brazilian voters, embedding an image-based conjoint experiment. Our sample was collected by the online polling firm Netquest. Netquest maintains large online panels of respondents across the Americas, including a panel of more than 300,000 respondents in Brazil. The firm collected a representative quota sample of 1,052 voters from its Brazil panel from November 24 through December 12, 2021.

The experiment evaluates the effect of candidates’ professional affiliation on voters’ support and evaluation of the candidate in three dimensions: fighting corruption, increasing public security, and creating jobs. The experiment follows the format of the conjoint experiment (Hainmueller et al. 2014)—also known as a factorial experiment (Auspurg 2015)—in which respondents are presented with several randomly generated profiles in sequence and asked to rate each one along several different dimensions, including the likelihood of supporting them in a hypothetical election. By randomly varying certain features of the profiles, this experiment recovers an average marginal component effect (AMCE) for each possible level that a feature can take. For example, we are particularly interested in the average expected difference in rating between a candidate who signals no professional background and one who signals that they are a police officer. The difference between the average rating of these two profiles represents an unbiased estimate of the “police” effect.

The setup for the factorial experiment is similar to the forced-choice experiment but allows respondents to focus on one profile at a time, reducing the mental burden. AMCEs can be estimated via linear regression with random intercept models or fully within-respondent models to account for the clustering of observations within respondents. Another common method is not to directly model the hierarchical nature of the data, but instead to estimate the model via OLS and cluster standard errors at the respondent level to account for the greater uncertainty of modeling clustered data. While the three approaches produce roughly similar estimates, we report the results of all three.

Respondents viewed three candidate profiles in succession. Each candidate was represented by a randomly generated political handbill, known as a santinho, that is commonly distributed by political campaigns in Brazil. Image-based experiments have recently been fielded by Flores-Macías and Zarkin (2022) and Poertner (2021). To make sure the information in the santinho got across to respondents, each profile had a small block of text underneath, reemphasizing the important information: the occupation and the economic policy the candidate supported. Table 1 presents a summary of all features of the hypothetical candidates, as well as the possible levels they could take. ⁴

Table 1. Candidate Features and Their Possible Labels

To produce the images, we hired a photographer based in Brasília to hold a photoshoot with six models representing two gender and three racial categories. Each model wore an identical costume to portray each of the occupations represented in the design. The santinhos were then randomly generated in-browser, layering a randomly selected candidate image with other attributes. Occupation was signaled through two bundled cues: the professional title and the costume. A candidate with the Sheriff (Delegado) title was always presented wearing the police officer costume, for example. (See the appendix for images of all the models used and costume examples.)

See figure 4 for an example of two of these santinhos, demonstrating how the profession might vary between candidates while holding some other features constant. Respondents were asked to evaluate each candidate on several attributes. First they answered the question, “How likely would you be to support this candidate?” with options on a five-point scale ranging from “extremely unlikely” to “extremely likely.” Respondents also evaluated how well they thought each candidate would perform on the dimensions of public security, corruption, and unemployment.

Figure 4. Sample of Candidate Profiles in the Survey.

We expect police and military candidates to be perceived as more effective at addressing public security and corruption. This means that candidates who signal a professional history with the military or police should score higher on effectiveness in handling corruption. Including a police officer profile and a military profile in the experiment helps us to parse out whether voters are rewarding law enforcement experience in particular or military organization in general. To account for how much candidates’ anti-corruption credibility is driven by the security sector connection rather than by the mere presence of a professional cue, we included in the experiment other professions of similar status, such as doctor, teacher, and pastor. All of these professions require some degree of education and specialized training, and the occupations of doctor and teacher are tenured public servants in the Brazilian education and medical systems, just as police officers are in the public security apparatus. Because of this selection of professions, we can be confident that results are not driven by a factor such as prestige.

To account for the possibility that respondents were supporting or evaluating candidates positively because they associated the candidate’s profession with being an outsider, rather than because of characteristics associated with the profession, we included information in the santinhos about the candidate’s insider or outsider status. Political experience was manipulated through a small seal on the santinho indicating whether the candidate had “six years’ experience in the state legislature” or would “bring fresh ideas to politics.” We also explicitly identified the candidate as new to politics or an experienced politician in the accompanying text vignette. Our design expanded on other studies highlighting the potential benefits to outsider candidates (e.g., González-Ocantos et al. 2023) and parsed out exactly what sort of outsider would be likely to benefit most.

One notable exclusion from our list of features was the candidate’s political party. Although parties tend to be comparatively more fluid and less institutionalized in Brazil (Mainwaring et al. 2018), party identification is still likely to overpower the other features and wash out the effects of theoretical interest, especially parties with recent high-profile national corruption scandals. To maintain realism while avoiding unwanted connections to well-known party brands, we assigned each santinho a two-digit number, placing it in the top right corner of the santinho. In Brazil, all parties are represented by two-digit numbers that a voter must type into the electronic voting machine on the day of the election. Our numbers were drawn from a list of unregistered numbers.

The exclusion of known party brands from the experiment may raise the question of whether our results might be driven by antipetismo (Samuels and Zucco2018). We accounted for this possibility in two different ways. First, to ensure that respondents’ evaluation and support for the candidate resulted from the professional titles, we included a slogan that signaled the candidate’s economic ideology. This guaranteed that respondents saw left-wing and right-wing versions of every professional profile. While left-wing and Partido dos Trabalhadores (PT) are not synonymous, respondents should generally consider left-wing candidates to be more closely aligned with the PT, all else equal. This allows us to parse out whether respondents are rating all police and military officers as more competent to address corruption, or just the ones that present a right-wing economic slogan.

Second, several of the professions that we include in the design could conceivably benefit from antipetista attitudes. This is especially true of the pastor. Similarly to the increase in police officer candidates, Brazil has seen massive growth in the number of Evangelical pastors running for office in the past decade, running mostly with right-wing parties (Boas 2023). If results are driven by antipetismo, respondents should react similarly to the security sector candidates and the pastors, as they are both popularly connected with antipetista movements. By the same logic, we should observe similar movement with the right-wing economic policy signal. While we do not believe that the association between antipetismo and security sector candidates taints the validity of our results, we do note that this is a possible limitation of the current design. Future work could address this by replicating this study in other contexts beyond Brazil, or by testing directly whether attitudes toward security sector candidates in Brazil are moderated by the candidate’s political party.

The final consideration for the construction of the santinhos was the candidate’s name. We randomly assigned each santinho a name drawn from a list of six common Brazilian names, three masculine and three feminine. We then cross-checked these names with a recent list of high-level officials to ensure that none of the last names were associated with well-known and controversial figures in Brazilian politics and culture.

After seeing a santinho, respondents were asked about their overall support for the candidate and how they rated the candidate along three dimensions. First they were asked, “How likely are you to support this candidate?” (Qual a chance de você apoiar esse[a] candidato[a]?) and were given five options on a scale that ranged from “extremely unlikely” to “extremely likely.” Next, respondents were asked “How effective do you think this candidate would be in the following areas?” (Quão efetivo você acha que o[a] candidato[a] seria nos seguintes temas?). Respondents were then presented with three areas—corruption, public security, and unemployment—and had to select a response for each one from a five-point scale ranging from “extremely ineffective” to “extremely effective.” Respondents were asked to perform this task three times in succession, evaluating three distinct fictional candidates. While the order of these questions was not randomized, we do not believe that question order had a strong effect, as respondents on average clearly did not respond identically to each posttreatment evaluation.

Because we wanted to examine heterogeneous effects, we asked respondents about their ideology on a seven-point scale common to many public opinion surveys in the region. More specifically, the question on ideology read as follows: “In politics, people often talk about ‘left’ and ‘right.’ Where do you position yourself on this scale, where 1 means left and 7 means right?”

To explore potential mechanisms and to collect additional qualitative data about how respondents perceived the candidate profiles they viewed, we included several open-ended questions at the end of the survey, asking the respondents to explain their choices. Specifically, we presented one randomly selected profile viewed by the respondent, as well as their previous evaluations of this profile for vote choice and anti-corruption effectiveness, and asked them to explain why they made those specific evaluations. Previous research has suggested that asking respondents to explain their choices in the context of a survey experiment can reduce social desirability bias by allowing respondents to frame their choices in a positive light (Krupnikov et al. 2016). While the sequencing of the experiment with the open-ended questions does not allow for respondents to make such adjustments, respondents will probably be driven by the same logic to provide a clear rationale for their choices, especially those who viewed unusual or controversial profiles. (See the appendix for open-ended question wording.)

Sample and Descriptive Results

The sample roughly mirrors the Brazilian population as a whole along gender, racial, class, and geographic lines, as well as how they voted in the first round of the 2018 presidential election. First, the respondents in our sample were relatively distrustful of nearly all the institutions that we asked about. Only 15 percent of respondents reported having at least some trust in the National Congress, while about 25 percent reported having at least some trust in their mayor. The institutions with the most trust, on aggregate, were the police and especially the armed forces, which had 49 percent and 56 percent, respectively, reporting at least some confidence in these institutions.

It is important to note, therefore, that while absolute levels of confidence in the police and armed forces remain somewhat low, with about half of respondents expressing a lack of trust, confidence in traditional political institutions is substantially lower, meaning that the police and armed forces enjoy a comparative advantage in this area. The average Brazilian voter may not love the police, but they trust them a good deal more than “traditional” politicians.

These results are similar to other studies of Brazilian public opinion, such as the 2020 round of the Latinobarómetro survey and the 2019 data from LAPOP. We find that our estimates of trust in the institutions are all roughly the same as the ones depicted by these two other surveys; our estimates have overlapping confidence intervals with at least one of the other major national surveys (figure 5). This suggests that our sample is not more or less trusting of any institution than the Brazilian population at large.

Figures 5. Levels of Trust in Various Institutions in Surveys of Brazilian Citizens.

Results

Main Effects

The image-based conjoint experiment confirmed part of our hypotheses and suggested intriguing new potential lines of research for future experiments. The models show that, as expected, police officers and military officers are perceived to be more competent in addressing corruption when compared to similar candidates with no clear professional affiliation (figure 6). Specifically, marginal effects show that being affiliated with the police increases candidates’ perceived competence to address corruption by 0.25 on a five-point scale. This same effect is 0.16 for colonels. While the magnitude on a five-point scale may seem small, these effects are on par with estimated effects for similar studies. ⁵ They also are a size similar to that of well-established features, such as race and gender. ⁶ Thus, Brazilian voters, on average, rated the police officer and the military officer as more effective in addressing corruption than the baseline category, irrespective of the race or gender of the candidates, insider versus outsider status, or whether the candidates displayed a left-wing or right-wing economic slogan. ⁷

Figure 6. Results from Image-based Factorial Experiment.

The evidence shows that this effect of candidates’ professional affiliations and their perceived effectiveness in addressing corruption is not a product of voters’ generalized affinity for the candidates. As a robustness test, we analyzed how respondents evaluated military and police candidates along other dimensions. When looking at respondent ratings of candidates along other dimensions, we observe a pattern consistent with the idea that respondents are ascribing certain specific competencies based on profession, and not simply generalized affinities. For instance, respondents did not rate police and military officers as more competent to address other issues where we would not, a priori, expect them to have specific strengths or weaknesses, such as healthcare provision and unemployment. Likewise, profession had little impact on the likelihood that a respondent would support the candidate. Other than pastors, who suffered a massive penalty of 0.41 points, no profession was consistently preferred or disliked across model specifications on the vote choice dimension. The professions had the strongest influence on the dimensions of specific issue competencies.

Besides combating corruption, the only other dimension on which police and military officers had a distinct advantage was the issue area of public security. The police officer candidate had an estimated gain of 0.50 points on this dimension, while the military officer enjoyed a gain of 0.38 (figure 6). Public security is the most straightforward area where these professions should have an advantage in the eyes of voters, as their professional experiences are most directly related to this issue area. Because the police and military candidates differed significantly from the other profiles on only this dimension and combating corruption, we can conclude that voters were indeed ascribing specific advantages to different candidates based on their professional experiences. Therefore, the increase in ratings for effectiveness against corruption is not just generalized affinity for the candidate but instead an issue-specific evaluation.

Qualitative responses illuminated two different reasons why respondents rate candidates from some professions—police, military, and teachers—as more competent to address corruption. The first one draws on candidates’ prior experiences working on corruption-related issues, and was particularly common when respondents were assigned to rate santinhos of police officers. ⁸ For example, one respondent tapped into police officers’ backgrounds in corruption investigations to explain why they expected these candidates to be competent in fighting corruption: “Because she is a detective, she knows how to organize and create tactics to work on this topic [corruption].” Another respondent agreed: “Being a police officer, they have some knowledge of how to work on this topic.” Several respondents pointed to the fact that, for them, prior experience in corruption investigations should give police officers a boost when trying to fight corruption from the mayor’s office.

Whereas respondents usually explained their understanding of police officers as being more competent to address corruption because of these candidates’ prior experiences, when it came to colonels, respondents emphasized the integrity of these candidates. For example, one respondent said they rated the military candidate as highly effective in addressing corruption “because in the military career one learns respect, ethics, and morality.” Another respondent added, “because of training on integrity that they [the candidate] received in the military headquarters.” Yet another respondent agreed: “[Because of] the military education…. Military schools are excellent.”

This integrity mechanism may help explain one unexpected result of the experiment, which is that teachers, too, enjoyed greater voter confidence to combat corruption relative to the baseline category. Teachers were estimated to have a rating 0.21 points higher than the reference category, an effect on par with police officers. While teachers overwhelmingly run with left-wing parties, such as the Workers’ Party, this effect does not seem to be driven by left-wing respondents. Instead, there seems to be a general sense across ideological groups that teachers are trustworthy and therefore would be effective at fighting corruption. In the open-response questions, respondents typically drew on the integrity and honesty of teachers to express why they believe teachers would be good candidates to combat corruption. One respondent, for example, justified their rating of teachers to address corruption by saying, “Being a teacher, I believe this candidate would have good conduct when working on this subject [corruption].” According to another respondent, “teachers are usually honest.” Yet another respondent simply stated, “I trust teachers.” While this result was unexpected, we highlight the finding as a potential extension of our theoretical framework and an avenue for future research.

Interestingly, the corruption slogan itself had only a small effect on respondents’ perceptions and was significant only in the OLS with cluster–robust standard errors model. This finding suggests that while voters were reading and comprehending the slogans, they had very low confidence in politicians and did not necessarily trust their campaign promises, especially when the promises concerned corruption. This sentiment was reflected in a number of open-ended responses to the activity. For example, one respondent shared their opinion that “most candidates say [that they will reduce corruption] but never follow through.”

This distrust seems to be especially strong with anti-corruption promises, with one respondent insisting that “generally, those who say that they are going to end corruption are corrupt!” Compared to other slogans, the anti-corruption slogan performed particularly poorly. For example, we detected a larger positive effect of having a campaign slogan mentioning unemployment on perceptions of effectiveness in managing unemployment in the city.

Overall, the professional cues corresponded with the strongest effects on individual perceptions—stronger than race, gender, economic policy, or campaign slogan. ⁹ This could be a design effect, as the professional cues are presented in the most visually appealing and obvious fashion, but we believe that this is probably not the case, for at least two different reasons. First, several dimensions, including race and gender, were also clearly signaled by the images. If the visual nature of the professional cue was driving results, then we should have observed a similar boost for these other visual dimensions. Second, the text-based elements were mentioned frequently in the open-ended response section at the end of the survey instrument. Respondents who chose to provide answers frequently referenced the slogans and economic positions signaled by the santinhos, suggesting that respondents were indeed making multidimensional choices, weighing several factors that were presented with images and text.

This image-based design improves text-based factorial or conjoint designs because it more closely resembles the types of campaign messages and media that Brazilian voters are used to consuming. If the candidate’s profession has an outsized effect in this design, it is also likely that these cues have an outsized effect in real life, meaning that this experiment has come closer to capturing the true decisionmaking process of Brazilian voters in local elections than previous experimental work. Observational studies have also emphasized the importance of nonpolicy heuristics, such as occupation, in low information contexts, such as local Brazilian politics, and have shown the weakness of platform and policy cues (Aguilar et al. 2015; Mechtel 2014). This design therefore also presents a contribution to the study of campaign symbols, further demonstrating the strength of uniforms as a cultural symbol (Flores-Macías and Zarkin 2022). By focusing on policy cues that are comparatively less credible than a symbol like a uniform, which only certain individuals can legally use, researchers may be missing an important dimension of criminal justice policy preferences. This contribution is therefore in line with other recent experimental evidence that suggests that professional affiliation is an important mediator for the relationship between candidate preferences and public security policy signals (Ventura et al. 2023).

Conditional Effects

The analyses also revealed significant heterogeneity among respondent groups. The police officer and the military officer did not, on average, fare better in terms of support, but our models show that the responses to professional security sector connections vary according to respondents’ ideology. ¹⁰ Specifically, our models show that the marginal effect for police and military candidates on vote choice (figure 7) and corruption effectiveness (figure 8) is strongest among right-wing respondents, and that the heterogeneity is strongest for military candidates.

Figure 7. Estimated Effects on Vote Choice for Military and Police Candidates Conditional on Respondent Ideology.

Figure 8. Estimated Effects on Corruption Effectiveness for Military and Police Candidates Conditional on Respondent Ideology.

When we plot these conditional effects, we observe some important details about how different respondent groups reacted to the experiment. For example, as shown in the previous section, the AMCEs for police and military candidates did not significantly differ from zero. This finding suggests that respondents were not more or less likely to support these candidates than the reference category, on average. However, figure 7 demonstrates that this is because of the symmetry between right-wing affinity and left-wing dislike for these profiles and not general ambivalence. While left-wing dislike (conditional effect of –0.748 when ideology = 0) was stronger than right-wing affinity (conditional effect of 0.463 when ideology = 6), the difference was small enough that the behavior of the two groups roughly canceled each other out when estimating the average effect. A similar story can be told for the police candidates, though the heterogeneity is not as strong, mostly because the farthest left respondents did not respond as negatively to the police candidates as they did to the military candidates.

Perceptions of effectiveness in fighting corruption tell a slightly different story (figure 8). While right-wing respondents give police and military candidates a higher rating on this dimension (conditional effect of 0.524 and 0.548 when ideology = 6, respectively), the left-wing penalties are not as large as they were for vote intention (0.212 and –0.300 when ideology = 0, respectively), and are not significantly different from zero at the α = 0.01 level. Voters in the center or just slightly to the right are estimated to give police and military candidates a meaningful advantage in fighting corruption (0.265 for police when ideology = 4), even when those same respondents were unconvinced about supporting the candidate in general (vote choice). These slightly divergent outcomes suggest that the anti-corruption brand of police or military candidates has at least some broad resonance beyond the farthest right-wing voters. While the farthest left voters are not convinced, they do not penalize security sector candidates on this dimension, and the Brazilian center voter seems open to believing that police and military candidates are effective anti-corruption leaders. ¹¹

To evaluate whether respondents answered identically to all posttreatment candidate evaluations, we analyzed ideological heterogeneity among two additional outcomes: effectiveness at providing public security and improving employment. When analyzing these conditional effects, we see that right-wing respondents reacted more positively across the board to police and military candidates. However, there are important differences between the outcomes, suggesting that respondents were making multidimensional choices. For example, figure 9 shows that there was general consensus across ideological groups that the military and police officers would have an advantage in providing security to the city, while figure 10 shows that there was little expectation even among right-wing respondents that police and military candidates would perform well on economic issues, such as generating employment. These differences between the outcomes suggest that each outcome was considered independently, and that respondents did not answer uniformly positively or negatively on all outcomes for a given candidate profile. Our estimate for strength fighting corruption is therefore credibly interpreted as such, and not just a general affinity for that particular candidate.

Figure 9. Estimated Effects on Effectiveness to Address Public Security for Military and Police Candidates Conditional on Respondent Ideology.

Figure 10. Estimated Effects on Effectiveness to Address Unemployment for Military and Police Candidates Conditional on Respondent Ideology.

Discussion and Conclusions

Drawing on an original experiment in Brazil, this study shows that respondents perceive candidates from the police and the military to be more competent to address corruption. This perception that police and military officers are more effective was strongest among conservative voters, but the large number of Brazilian center voters were also amenable to the idea. Our findings contribute to debates on corruption and voting behavior (Pavão 2018). Previous work suggests that voters prefer outsiders to address corruption (Petersen 2020; González Ocantos et al. 2023), but we find that previous political experience has no significant impact on perceived competency in combating corruption. Instead, we find that specific professions—police and military officers, as well as teachers—receive a boost in their perceived ability to fight corruption. This finding can help to understand recent trends in Brazilian elections and the potential unintended consequences of massive corruption investigations in young democracies.

Moreover, these findings expand the explanatory domain of previous work on the effect of professional affiliation on voting behavior (Carnes and Sadin 2015; McDonald et al. 2020; Boas 2014; Carnes and Lupu 2016). We show that candidates’ professions affect not just voters’ support for them or voters’ perceptions of which areas they will prioritize (McDermott and Panagopoulos 2015; McDermott 2009; Yanez Ruiz 2018), but also candidates’ perceived competence to address issues that are both directly and indirectly related to candidates’ profession. Finally, while previous studies have explored the effect of professional affiliation on areas such as national security (McDermott and Panagopoulos 2015; Teigen 2013), business (Yanez Ruiz 2018), or education (Atkeson and Hamel 2020), we focus on candidates’ ability to fight corruption, which is one of the most salient electoral issues in Brazil (Rennó 2020) and beyond.

Future studies could more directly test the mechanisms through which candidates’ professional affiliation affects their perceived competence to address corruption. The qualitative evidence from the survey suggests two possible mechanisms. First, it may be that candidates from certain professions, such as police officers, are understood to have experience working on corruption. Therefore, their prior experience makes them a compelling choice—in the eyes of voters—to fight corruption. Moreover, certain candidates are affiliated with professions that are understood to have more integrity than other professions. This may lead voters to believe that these candidates either are more proactive in bringing integrity to politics or trying to combat corruption. The integrity mechanism could also help explain an unexpected result—that teachers also are perceived to be effective against corruption. By systematically testing these two and other possible mechanisms, future scholarship can deepen our understanding of how candidates’ professional affiliations affect perceived anti-corruption competencies.

Future studies could also test the effect of other professional titles on voting behavior. One limitation of this project is that we only examined the effect of being doctors, pastors, teachers, police, and military officers. However, it is possible that other occupations also affect voters’ support and evaluation of candidates’ competence across different areas. In particular, future studies could examine the effects of candidates who campaign as prosecutors or judges. Typically, it is rare to see prosecutors or judges running for political office in Brazil because both these professionals are tenured civil servants recruited through impersonal exams and are forbidden from having partisan affiliation (Arantes 2015). Therefore, to run for political office, prosecutors and judges must first quit their jobs. However, in the 2022 elections, the lead prosecutor (Deltan Dallagnol) and judge (Sergio Moro) who spearheaded the Lava Jato investigations in Brazil were elected federal representative and senator, respectively (Mendonça 2022). This suggests that these occupations are also perceived as more competent to address corruption, given that this was the primary focus of these two civil servants before they decided to run for office. Future studies could test whether and why prosecutors and judges receive more support and are perceived to be more competent to fight corruption.

The results of our experiment are subject to specific boundary conditions. The first one is the relative public trust of security sector institutions, such as the army or the police, compared to other state institutions, such as Congress or political parties. As previously mentioned, in Brazil, the public tends to trust the police substantially more than political institutions (Corporación Latinobarómetro 2021), such as national-level (federal representatives) and local politicians (mayors). This may be because bureaucratic corruption in Brazil is low when compared to grand corruption; that is, corruption at the highest levels of government (Rose-Ackerman 1999). For example, data from Transparency International show that while only 11 percent of Brazilians reported having to pay bribes to access public services (such as schools or healthcare), 90 percent of the population evaluates corruption as a serious problem (Pring and Vrushi 2019). In contrast, in countries where bureaucratic corruption is as high or higher than grand corruption, we should expect trust in the police to be lower. As a result, candidates from the police or military should be perceived as a part of the problem and less effective at combating corruption. Future research could compare how the public evaluates the performance of security sector candidates across countries with varying levels of trust in law enforcement organizations to examine whether the effect we found here holds in countries where the police are perceived to be equally or more corrupt than politicians.

Another possible boundary condition of this argument is the degree of visibility of law enforcement agents in anti-corruption prosecutions on the public agenda. We fielded this experiment shortly after a large-scale corruption investigation, in which detectives and prosecutors pressed charges against more than one thousand politicians and business executives involved in bribery schemes across multiple public and private organizations (Ministério Público Federal 2020). Therefore, it may be that the public perceives candidates from the police as more effective in addressing corruption partly because people have recently been bombarded with media content on anti-corruption investigations by the Federal Police (Feres Júnior et al. 2018). In contrast, when the issue of corruption is more associated with other government institutions—such as the Comptroller’s Office—or nonprofit organizations, candidates from the police may no longer receive a boost in their perceived ability to address corruption.

One more possible boundary condition of the argument relates to the outcomes of corruption scandals that respondents were recently exposed to. We fielded this experiment in Brazil in 2021, shortly after the Lava Jato investigations, which disproportionately affected left and center parties (González Ocantos et al. 2023). Therefore, it is possible that the heterogeneous effects we uncovered—in which conservative respondents are more likely to support and rate police and military officers as competent to address corruption—were partly explained by these particularities of the Brazilian context. It is possible, for example, that if corruption scandals implicated more heavily right-wing politicians, progressive voters would show more support for police and military officers. However, it is not clear whether security sector candidates would be able to capitalize on corruption scandals targeting right-wing politicians in the same way that they have in scandals targeting center- and left-wing candidates, such as Lava Jato. We hypothesize that candidates from the police and military should be able to benefit from scandals that target right-wing politicians because, before Lava Jato, these candidates were not mostly affiliated with right-wing parties as one might expect; instead, most of them were affiliated with center parties with no clear ideology (Berlatto et al. 2016). Future studies could further enhance our understanding of occupations as voting heuristics by testing the effects of candidates’ occupations in contexts where recent corruption scandals disproportionately targeted right-wing politicians.