Date post: | 18-Jan-2016 |
Category: |
Documents |
Upload: | celio-menezes-rodrigues |
View: | 8 times |
Download: | 0 times |
Lacking Information or Condoning Corruption? Voter Attitudes Toward Corruption in Brazil
Matthew S. Winters1 University of Illinois at Urbana‐Champaign
Rebecca Weitz‐Shapiro Brown University
A well‐known phrase used to describe politicians in Brazil is “rouba, mas faz” – “he robs, but he gets things done.” It has been linked to particular political figures and used to explain President Lula’s persistent popularity in spite of numerous corruption scandals. But is it really the case that voters knowingly overlook corruption because a politician otherwise performs well? Or, instead, do voters fail to turn corrupt politicians out of office because they lack reliable information about corruption? Results from a nationwide survey experiment in Brazil suggest that the information constraint – rather than an informed tradeoff – explains the persistent political corruption there. In the experimental setting, voters express a strong preference for punishing corrupt politicians. Our findings help us to refine existing hypotheses about the role of information in deterring corruption, suggesting that specific, credible, and accessible information facilitates voter rejection of corrupt politicians.
1 This is part of an ongoing collaboration in which the position of first author will alternate between the authors. Both authors contributed equally to this paper. A previous version of the paper was presented at the 2010 American Political Science Association annual meeting. For helpful feedback at different stages of the project, we extend our thanks to Christopher Anderson, Jake Bowers, Damarys Canache, Silvia Cervellini, José Cheibub, Diego Correa, Miguel de Figueiredo, Scott Desposato, Daniel Gingerich, Jim Kuklinski, Eduardo Leoni, Rose McDermott, Jeff Mondak, David Nickerson, Lucio Renno, Mitch Seligson, Dan Treisman, Milan Vaishnav, Cara Wong, and Cesar Zucco, along with participants at a seminar at the University of Illinois, Urbana‐Champaign. Thanks to Amanda Cronkhite and Diego Correa for research assistance. Rebecca Weitz‐Shapiro thanks the Brown University Salomon Research Award for funding. This research was determined to be exempt from human subjects regulations by the Department of Political Science Human Subjects Committee at the University of Illinois at Urbana‐Champaign and by the Research Protections Office at Brown University (Protocol #1005000205).
1
In recent decades, enumerating the ill consequences of corruption has become something of a
cottage industry in both academic and policy circles.2 Although an earlier generation of scholars
suggested that corruption might play a functional role in the modernization process (Leff 1964;
Huntington 1968), the current academic consensus points instead to the negative consequences of
corruption. In cross‐national studies, corruption is repeatedly linked to lower investment and economic
growth (Mauro 1995; Lambsdorff 1996; Bardhan 1997; World Bank 1997b; Treisman 2000; see
Lambsdorff 2006 for a review). Corruption may not only limit overall growth, but also increase poverty
and income inequality (Gupta, Davoodi and Alonso‐Terme 2002; Li, Xu and Zou 2000 and Gyimah‐
Brempong 2002). Furthermore, in a democracy, corruption undermines the quality of representation
when elected politicians make decisions motivated by the desire for personal enrichment rather than
the preferences of the voters who elected them (Rose‐Ackerman 1978; Morris 1991).
Given the negative outcomes of corruption documented by scholars and publicized by NGOs, we
should expect citizens to reject corruption, as well. It is no surprise, then, that citizens in countries with
higher levels of corruption express lower levels of system support than their counterparts in less corrupt
countries (Rose, Mishler and Haerpfer 1998; Pharr 2000; Mishler and Rose 2001; Seligson 2002;
Anderson and Tverdova 2003). We might also expect to observe different levels of corruption across
different regime types. Assuming that citizens disapprove of corruption, and given that, by definition,
democracies provide citizens with the right to choose their leaders, we would expect those regimes
where citizens have the most power to select their leaders to suffer from the least corruption. The
2 By corruption, we mean the abuse of public office for private gain (Nye 1967; World Bank 1997a). Our
focus in this paper is on political corruption – the involvement of elected office holders in corrupt acts—
as distinguished from bureaucratic corruption, which involves unelected public servants. (This usage is
not universal; see Heidenheimer, Johnston and LeVine 1989.) Most empirical measures of corruption do
not distinguish between the two.
2
evidence, however, is at best mixed on this point. According to Keefer and Vlaicu (2008, 372), “in 2004
more than one‐third of all democracies exhibited as much or more corruption than the median non‐
democracy.” Various cross‐national analyses find no conclusive evidence that democracies are less
corrupt than their non‐democratic counterparts (Lambsdorff 2006, Treisman 2007). What could explain
the apparent inability of democracy to more sharply decrease corruption?
The existence of persistent, pervasive political corruption in a democracy means, by definition,
that many voters cast their ballots for corrupt politicians. Although most explanations of corruption
focus on institutional or economic factors, we turn in this paper to the individual level. The literature
points to two main possible explanations for why voters support corrupt politicians. The first, which we
call the “information hypothesis,” suggests that voters support corrupt politicians when they lack
information about a candidate’s involvement in corruption upon which they then could act in the voting
booth. The second, which we call the “tradeoff hypothesis,” suggests that voters knowingly cast ballots
for corrupt politicians because they expect that the overall benefits that accrue from a politician’s term
in office will be greater than the costs associated with corruption. In this paper, we test these
hypotheses using data from a nationally‐representative survey experiment in Brazil.
Brazil has historically suffered from high levels of corruption, a phenomenon that has persisted
in spite of the advent and consolidation of Brazilian democracy in the decades since the mid‐1980’s (e.g.
Geddes and Ribeiro Neto 1992; Fleischer 1997). Indeed, Kurt Weyland (1998) argues that the costs of
corruption reached record highs in Brazil after the transition to democracy. Even the administration of
President Luiz Inácio Lula da Silva, at the helm of a party that made honesty a central theme of its
campaigns (Ward 2006), was racked by a series of influence peddling and kickback scandals that resulted
in dozens of arrests and the resignation of a cabinet minister. In spite of these scandals, Lula was
resoundingly elected to a second term in office and retained unusually high levels of public approval. In
3
the words of three scholars of Brazilian politics, “[t]he conventional understanding of corruption in Brazil
suggests ‘impunity reigns’” (Pereira, Rennó, and Samuels ND).
In Brazil, the tradeoff explanation is succinctly summarized by the well‐known Portuguese
phrase “rouba, mas faz” (“he robs, but he gets things done”). The phrase itself is sufficiently widespread
that, according to popular legend, a well‐known congressman and former mayor of the city of São Paulo,
Paulo Maluf, took [email protected] as his e‐mail address.3 Direct survey questions that ask about
respondents’ attitudes towards “rouba, mas faz” suggest that it enjoys the support of a substantial
minority of the population. In one survey from the year 2000, 47 percent of respondents said they
would prefer a mayor who was not “totally honest” as long as he resolved a municipality’s problems,
whereas only 40 percent claimed they would prefer a totally honest mayor, even if he were “not that
efficient” (Figueiredo 2004).4 Similarly, a 2002 survey finds 40 percent of respondents agree with the
statement that “a politician who carries out a lot of public works, even if he robs a little, is better than a
politician who carries out few public works and does not rob at all” (Almeida 2006, 45).5
On the other hand, there is some aggregate‐ and individual‐level data that suggest Brazilian
voters do punish corruption when they learn about it. The information hypothesis finds support in
3 The origins of the phrase date back at least to the 1950s, where it was the informal campaign slogan of
the politician Adhemar de Barros, alternately governor and mayor of São Paulo between the 1940s and
1960s (Villaméa Cotta 2008; Queiroz and Manhanelli 2009).
4 In Portuguese, respondents were asked whether they would prefer “um prefeito que não seja tão
honesto, mas resolva os problemas do município” or “um prefeito totalmente honesto, mesmo que não
seja tão eficiente.”
5 These data are from the 2002 Brazilian national election study (ESEB). In Portuguese, respondents
were asked whether they agreed that “É melhor un político que faça muitas obras, mesmo que roube
um pouco, do que um polítco que faça poucas obras e nao roube nada.”
4
recent studies at the municipal level showing that, at least under some conditions, voters will turn
mayors out of office when municipal audits undertaken by the national or state government uncover
corruption linked to their administrations (Ferraz and Finan 2008; Pereira, Melo, and Figueiredo 2009).
Similarly, others have shown that national deputies accused of corrupt acts are less likely to run for
reelection and less likely to win reelection if they do run (Pereira, Rennó, and Samuels ND). A field
experiment conducted in São Paulo that provided voters with information about corruption allegations
against two candidates running for mayor produced some evidence that voters may vote against corrupt
politicians or else opt not to turn out to vote (de Figueiredo, Hidalgo and Kashara 2010).
To better understand voter attitudes toward corruption and the extent to which each of these
two explanations accounts for corruption in Brazil, we make use of a nationwide embedded survey
experiment. Our findings do not support the claim that the “rouba, mas faz” tradeoff attitude is
widespread in Brazil but instead are more consistent with the hypothesis that an information problem
explains the continued pervasiveness of political corruption in Brazil. Except for some small subsets of
the population, our results suggest that, when they learn about it, Brazilian voters are not tolerant of
corruption and are unlikely to support corrupt politicians – even if these politicians otherwise perform
well. In general, where voters can clearly attribute corrupt acts to particular politicians, they appear
willing to punish those politicians electorally. Our results also suggest the need to further refine the
information hypothesis to understand the conditions under which information about corruption will in
fact prompt a negative response among voters. Before turning to the results, we first describe in more
depth why voters may cast their votes for corrupt candidates and review the extant literature on voter
responses to corruption.
Why Do Voters Support Corrupt Politicians?
Scholars have proposed a number of explanations for the disparities in the amount of corruption
across countries. Some attribute varying levels of corruption to differences in electoral or political
5
institutions. Federalism and presidentialism have been linked to increased corruption, whereas
decentralization and plurality elections for legislators are associated with lower levels of corruption
(Treisman 2000; Fisman and Gatti 2002; Adsera, Boix and Payne 2003; Persson, Tabellini and Trebbi
2003; Gerring and Thacker 2004; Kunicova and Rose‐Ackerman 2005; Golden and Chang 2006). Others
have explored whether regulatory systems and the availability of rents create incentives for corruption
(Shleifer and Vishny 1993; Djankov, et al. 2002; Ades and Di Tella 1999). Beyond institutional
explanations, inherited or cultural attributes might explain a society’s tolerance of corruption (La Porta,
et al 1999; Treisman 2000; Fisman and Miguel 2006). Other authors have suggested that countries can
become stuck in a trap where the pool of available candidates includes only corrupt politicians (Kurer
2001; Caselli and Morelli 2004; Messner and Polborn 2006).
Here, in contrast, we focus on understanding the individual behavior of voters in a democracy.
Without the support – knowing or unknowing – of voters, corrupt politicians would not survive
electorally. We examine the extent to which an individual’s choice to vote for a corrupt politician can be
explained either by a lack of information or by a rational tradeoff. Although not mutually exclusive,
these hypotheses are addressed in different literatures, and so we address them separately below.
The Information Hypothesis
There are a number of reasons why, even in a democracy, information about corruption may be
limited. For obvious reasons, politicians will try to conceal their illicit activities. Lack of resources and
professionalization among the media – acute problems especially among middle and low income
democracies (Freedom House 2010) – are likely to make efforts at obfuscation more successful. Then,
insofar as voters are not aware of corruption, they cannot vote against corrupt candidates (Rose‐
Ackerman 1978; Geddes 1994).6 Even when allegations of corruption are made public, voters may
discount these reports, especially if they take the form of politically‐motivated partisan accusations
6 Rundquist, Strom and Peters (1977) refer to this as the “ignorant voter” explanation for corruption.
6
(Rundquist, Strom and Peters 1977). If a lack of information explains voter support for corrupt
politicians, we expect voters to withdraw support from corrupt politicians in settings where information
is available.
Although not always stated explicitly, much of the cross‐country research on corruption
assumes that the information hypothesis for the persistence of corruption holds true. For instance,
Treisman argues that democracies have less corruption because “exposure [is] more likely” in
democracies – implying that, after exposure occurs, voters will remove corrupt politicians from office
(1998: 6). Gerring and Thacker (2005) similarly postulate that increased openness and transparency
should be associated with a lower incidence of corruption. Adserá, Boix, and Payne (2003) argue that, if
voters have mechanisms available to hold politicians accountable (such as democratic elections), then as
information provision increases, corruption should decrease. At an extreme, they expect that, “with
perfectly informed voters, politicians’ rent [from corruption] should disappear” (448).
The past few years have seen the emergence of a number of natural and field experiments that
test the information hypothesis.7 These rely on aggregate data at the precinct, polling station, or
municipal level, and have found mixed support for the information hypothesis. A field experiment
carried out in rural India (Banerjee et al. 2010) finds that priming voters about corruption in general has
no effect on either voter turnout or the support politicians receive at polling stations. Looking across
precincts in a survey carried out in Mexico, Chong et al. (2010) find that information about high levels of
corruption depresses turnout and has a marginally significant negative effect on the incumbent’s vote
share. In Brazil, Ferraz and Finan (2008) take advantage of natural experiments generated by
randomized federal auditing of municipalities and find that the information about corruption generated
7 Pande (2011) provides a summary of many of these works, as well as other experimental work dealing
with the effect of voter information on the quality of governance.
7
by these audits, especially in municipalities with local radio stations, decreases support for incumbent
mayors.8
At the individual‐level, several previous studies have used experimental techniques to explore
the ways in which voters react to information about different characteristics of political candidates.
Although they do not examine corruption specifically, these results are generally consistent with the
expectations of the information hypothesis. For example, a series of articles finds that citizens react
negatively to information that politicians are dishonest or lack integrity (Kulisheck and Mondak 1996;
Canache, Mondak and Cabrera 2000; Mondak and Huckfeldt 2006). Scholars have also found that
different types of voters respond differently to information about corruption. Rundquist, Strom and
Peters (1977), for example, conclude that individuals who care about honesty in government are the
most likely to punish corrupt candidates.
Our work examines the information hypothesis with respect to corruption using individual‐level
data. Embedding the experiment into a survey allows us to look at how the effects of treatment vary for
different sectors of the population. Also, importantly, our survey experiment tests the information
hypothesis alongside alternate hypotheses about the relationship between corruption and voting
behavior. Specifically, we examine the tradeoff hypothesis, which has deep resonance in Brazil.
The Tradeoff Hypothesis
The second explanation for voter behavior suggests that voters knowingly elect corrupt
politicians when those politicians provide voters with other valuable benefits. In other words, this
explanation posits that voters make a strategic tradeoff, overlooking corruption when politicians deliver
8 Although they do not use experimental data, Pereira, Melo, and Fiugueiredo (2009) examine state‐
level corruption audits and find that corruption allegations made immediately prior to an election
depress a mayor’s probability of reelection, but that corruption allegations made public at a time farther
from elections have no such effect.
8
other positively‐valued outcomes.9 In the Brazilian context, this logic of a strategic tradeoff is
communicated by the phrase “rouba, mas faz.”
The tradeoff explanation takes a number of different forms. The first says that if the politician
delivers economic growth or other public goods then citizens will accept some losses from public coffers
because overall the government machinery performs well under the helm of the corrupt politician. If
this is the case, then voters should support corrupt politicians where those politicians otherwise act as
competent public servants, whereas a corrupt and poorly performing politician should not be returned
to office. Although this may be the most common interpretation of the tradeoff explanation, it is also
possible that citizens overlook corruption in exchange for either private goods or ideological satisfaction.
Rundquist, Strom and Peters (1977) discuss both possibilities: voters may receive or expect patronage or
other particularistic benefits from corrupt politicians, or voters might support a corrupt politician if he
more closely mirrors their ideological preferences.10
We focus our efforts on examining the primary conceptualization of the tradeoff explanation—
that citizens may overlook corruption when politicians deliver broad public benefits. This is the
dominant scholarly and popular interpretation of the phrase “rouba, mas faz” in Brazil. For example,
President Lula’s electoral success in 2006, which followed upon both corruption scandals and a period of
impressive economic growth, has been explained with reference to this phrase (Desposato 2006; Renno
2007, ND).11 Similarly, Brazilian public opinion surveys that ask directly about the tradeoff between
9 A third possibility is that, all else equal, voters actually prefer corrupt politicians. Although theoretically
possible, we view this as implausible and do not explore it here.
10 See Manzetti and Wilson (2007) and Chang and Kerr (2009) for some evidence for the patronage
hypothesis.
11 Samuels (2004, 2008), however, does not find that PT voters are actually more likely than other voters
to express support for the “rouba, mas faz” principle.
9
corruption and other dimensions of performance also reflect this understanding of the phrase, referring,
for example to “efficiency,” “resolving the problems of the municipality,” or “carrying out public works”
as factors that might excuse corrupt behavior.12
The salience of the tradeoff explanation may vary across groups. Building on previous literature,
we explore the possibility that there are different reactions across social classes and across subsets of
people with different levels of political knowledge. Some existing literature argues that the poor –
because of the pressures of meeting basic needs – are less likely to have the time or inclination to care
about the quality of governance (Kurer 2001; Blake 2009). Empirically, Figueiredo (2004) finds significant
class differences in the patterns of responses to the question about supporting a corrupt yet competent
mayor. Among respondents in the two lowest income brackets, 54 percent claimed to prefer the corrupt
but competent mayor, whereas willingness to support such a mayor drops to 40 percent and 32 percent,
respectively, among the top two income brackets. Pereira, Rennó, and Samuels (ND: 86) similarly report
an inverse relationship between SES and support for corrupt yet competent politicians. According to the
2002 Brazilian National Election Survey (ESEB), the share of voters who said they would support a
corrupt but competent politician dropped steadily as education increased. While 53 percent of voters
with no education said they would offer such support, only 25 percent of respondents with at least
some college education said so.13
12 See, for example, the IBOPE survey cited in Figueiredo 2004 and the ESEB survey cited in Almeida
2006.
13 Though not asking about rouba, mas faz directly, cross‐national data from Latin America also supports
these findings. Blake (2009: 104) finds a negative correlation in Latin American World Values Survey data
between income and tolerance for corruption. On the other hand, using WVS data from a sample of 35
countries, Gatti, Paternostro, and Rigolini (2003) find that the wealthy are actually more tolerant of
corruption.
10
Similarly, in looking at the impact of political scandal on vote intention in a laboratory
experiment, Funk (1996) finds differential responses to scandal in the face of politician competence
among the politically knowledgeable. Specifically, when politicians are described as competent, the
politically knowledgeable respondents in her study were forgiving of marital infidelity but not forgiving
of tax evasion, whereas those respondents who scored low on political knowledge were not willing to
forgive either. Funk argues that this result is consistent with other research that finds “political
knowledge is associated with greater cognitive complexity to process political information” (1996: 16).
In summary, we propose two possible explanations for why a voter might cast her ballot for a
corrupt politician. On the one hand, she may lack information about political corruption and thus
unwittingly support a corrupt politician. On the other hand, she may knowingly vote for a corrupt
politician, calculating that the politician’s anticipated performance in office outweighs the drawbacks of
corrupt behavior. Previous research provides some evidence consistent with each of these mechanisms
but has not tested them simultaneously. In this paper, we explore these two hypotheses together in the
context of a young democracy known for widespread corruption. Research in Brazil has both asserted
the relevance of a “rouba, mas faz” attitude among voters—drawing mostly on direct survey data—and
yet also shown that, in the aggregate, voters withdraw support from politicians who are found to be
corrupt. We use individual level data and an experimental design to examine which hypothesis better
captures the attitudes of present‐day Brazilian voters.
Research Design: An Embedded Survey Experiment
When asked about a sensitive topic in a public opinion poll, respondents may reply to questions
with socially‐conforming answers rather than their genuine opinions. In the case of corruption, we
worry that some voters who will express disapproval of corruption when asked directly in fact have
more complex attitudes. In our survey experiment, rather than directly asking respondents their views
on corruption, we instead randomly assigned each respondent to hear one of 12 different vignettes,
11
varying the information that they received about a hypothetical incumbent politician’s past corruption
and general competence in providing public works. We also varied information about the politician’s
political party. Random assignment to the vignettes ensures that, on average, these groups are
indistinguishable on both observable and unobservable characteristics. By comparing average responses
to standard survey questions across the groups assigned to the different vignettes, we are able to
identify the causal effect of receiving information about corruption on survey responses.14
Our basic vignette read as follows15 (with the phrases in brackets being substituted in
alternative versions of the vignette):
Imagine a person named Gabriel {or Gabriela}, who is a person like you, living in a neighborhood like yours, but in a different city in Brazil. The mayor of Gabriel’s city is running for reelection in October. He is a member of the PT [Partido dos Trabalhadores] {or PSDB [Partido da Social Democracia Brasileira]}. In Gabriel’s city, it is well known that the mayor never takes bribes {or frequently takes bribes} when giving out government contracts. The mayor has completed few {or many; or omit the entire sentence} public works projects during his term in office. In this city, the election for mayor is expected to be very close.
The identity of the hypothetical voter (Gabriel or Gabriela) was chosen by the enumerator to match the
respondent’s gender.16 The partisan identity of the mayor, whether the mayor was known for taking
14 For more on survey experiments, see Sniderman and Grob 1996; Druckman et al. 2006; and Gaines,
Kuklinski and Quirk 2007.
15 This vignette is similar to that described in Weitz‐Shapiro (2008) and also draws on the wording of
vignettes included in the 2006 Mexico Panel Study (Lawson et. al 2007).
16 We used this third‐person prompt for two reasons. First, we wanted respondents to abstract from
their own experience and from the characteristics of elected officials in their own municipalities.
Second, we hoped to reduce social desirability bias (DeMaio 1984; Nederhof 1985) in the responses of
those who were in one of the corruption treatment groups. Given the cultural visibility of “rouba, mas
faz” in Brazil, we assume that respondents who give answers they believe are consistent with the
12
bribes or not, and whether the mayor had completed many or few public works projects (or if this
information was not provided to the respondent) were all randomized across respondents.17
Immediately following the vignette, respondents were asked, “In your opinion, what is the
likelihood that Gabriel(a) will vote for this mayor in the next election: very likely, somewhat likely,
unlikely, not at all likely?” The response to this question serves as our main outcome variable of
interest. For all analyses in this paper, we collapse it to a dichotomous variable indicating positive or
negative vote intention.18 Although the question asks about Gabriela’s voting behavior, recall that we
prompt respondents to consider Gabriela to be “a person like you, living in a neighborhood like yours.”
We therefore treat responses to this question as if they indicated the respondent’s own preference.19
“typical” Brazilian’s viewpoint would claim to be more forgiving of corruption. The bias from the design,
then, would increase the likelihood of finding support for the tradeoff hypothesis. Given the lack of
evidence we find for this hypothesis in the data, any bias generated by this aspect of the vignette design
appears minimal.
17 Hence there were a total of 2 X 2 X 3 = 12 randomly‐assigned vignettes. Assignment took place in
advance of distribution of the surveys and was stratified at the census tract level. Enumerators received
surveys with different vignettes pre‐printed in them. Note that we do not describe a potential
challenger. We made this choice in part to limit the total number of treatment groups to a reasonable
amount. Although the absence of challenger information may add noise to the data, it will not bias the
results as long as there is no systematic correlation between treatment categories and the assumptions
that respondents might make about a challenger.
18 We have reproduced the main results using ordered logistic regression with the original four‐category
variable as the outcome variable. They can be found in the online appendix.
19 To simplify the presentation, we sometimes refer in the text to the respondent’s vote choice, rather
than that of Gabriel(a), but this question was in all cases asked about Gabriel(a). Some follow‐up
13
The survey was administered in summer 2010 by IBOPE, the largest public opinion polling
company in Brazil, as part of their monthly omnibus survey (N=2,002). Our questions were the first
block on the survey, so respondents heard the vignette immediately after answering a small set of
demographic questions that determined their eligibility for the survey. IBOPE sampled 140 cities using a
probability‐proportional‐to‐size method within 25 strata.20 Then census tracts were selected using a
probability‐proportional‐to‐size method with stratification across zones of major metropolitan areas.
Enumerators recruited individual respondents according to a quota sampling scheme designed to
produce a representative sample of the national population in terms of age, gender, education and
occupational category.
Table 1 presents the six treatment conditions – ignoring the variation in the announced party of
the incumbent mayor – and the variable in each cell represents the proportion of respondents in that
treatment condition who say that Gabriel(a) is either very likely or somewhat likely to vote for the
mayor.21
To the extent the information hypothesis is correct, and voters react negatively when they
discover that a politician is corrupt, then we expect to see that respondents are less likely to express
support for the mayor in the corruption treatment conditions as compared to the no corruption
treatment conditions (i.e. B4 < B1; B5 < B2; and B6 < B3). To the extent the tradeoff hypothesis is correct,
questions, especially those not subject to potential problems of social desirability bias, asked the
respondent directly about his or her opinion of the mayor, rather than about Gabriel(a)’s likely behavior.
20 These strata were defined by 25 of Brazil’s 27 states – the monthly survey rotates among three small
states in the northern region of the country.
21 In this paper, we do not explore the role of party ID. We find that partisanship has little or no effect on
responses. In future work, we will explore the interaction of partisanship and attitudes towards
corruption in more detail.
14
and voters are willing to forgive corruption in the face of competence or good performance, then we
expect to see several patterns in the data. First, we expect to see less punishment of corruption as
competence increases (i.e. (B4 – B1) < (B5 – B2)). Second, we expect to see that competent but corrupt
politicians will be weakly preferred to incompetent but clean politicians (i.e. B4 > B2). Note that these
two hypotheses are not exclusive. It is theoretically possible to find evidence for both the information
hypothesis and the tradeoff hypothesis in the data.
Experimental Validity
Before turning to the results, we first discuss the experiment’s validity, both internal and
external. Beginning with the former, we check that the vignettes were appropriately randomized and
that respondents are not systematically different across vignettes. To do this, we run two multinomial
logit models where the 12 categories of treatment assignment define the outcome variable. We
compare a null model with no predictors to a model with predictors for gender, age, education, social
class, indicators for employment status, an indicator for being Catholic, an indicator for being religious,
an indicator for being nonwhite, indicators for being a likely PSDB or PT voter and indicators for being a
PSDB or PT partisan. A likelihood ratio test fails to reject the null hypothesis that the two models are
indistinguishable (p < 0.93), meaning that no meaningful correlations are detected across the treatment
conditions and respondent characteristics. This validates the success of the randomization.
We also check whether respondents understood the prompts in the expected manner. Do
respondents treat information about bribe‐taking as evidence of corruption, and do they treat
information about accomplishing public works as evidence of a politician’s competence or capacity? We
check this by examining correlations between the different stimuli and respondents’ answers to follow‐
up questions in the survey. Asked if the mayor was trustworthy, those in the corruption treatments
were 66 percentage points less likely (p < 0.00) to say that the mayor was very or somewhat trustworthy
than those in the non‐corruption treatments. Similarly, respondents who were told that the mayor had
15
accomplished many public works projects were 23 percentage points (p < 0.00) more likely to say that
the mayor was very competent or somewhat competent at securing federal funds for the municipality as
compared to those respondents who were told that the mayor had accomplished few public works
projects. These results suggest that respondents understood the information about bribes and public
works provision as we had intended.
“Rouba, mas faz” or Not?
Our main test of the information and tradeoff hypotheses examines responses to the first
question those surveyed heard after the vignette—“In your opinion, how likely is it that Gabriel(a) would
vote for this mayor in the next election?” To the extent that the information hypothesis holds true, we
would expect corrupt politicians to receive less support than clean politicians at all levels of politician
performance, and the electoral costs of corruption should be approximately equal across levels of
politician competence. To the extent that the tradeoff hypothesis holds, we expect the costs of
corruption to decrease as competence increases and for corrupt but competent politicians to receive
equal or greater support than clean but incompetent politicians.
Table 2 displays the proportion of people in each treatment category who said Gabriel(a) would
vote for the incumbent mayor. The results provide substantial support for the information hypothesis
and are not at all consistent with the existence of a rouba, mas faz tradeoff dynamic. The first column
shows that, across all respondents, there is a very strong negative reaction to corruption. The average
share of respondents willing to vote for a candidate described as non‐corrupt is 78%, but only 19% are
willing to vote for a candidate described as corrupt, a difference of almost 60 percentage points. The
magnitude of this difference is similar when we look only at the subset of respondents who were told
that the mayor had delivered a large number of public goods during his previous term in office. Within
the subset of respondents who were told that the mayor was incompetent, the effect of the corruption
stimulus is still substantial, although somewhat smaller. Whereas the tradeoff hypothesis would suggest
16
that competence moderates the punishment for corruption, we find no evidence of this in our data.
Instead, the size of the corruption penalty is actually greater when a politician is described as
competent.22
In addition, the rouba, mas faz hypothesis is not supported in the crucial comparison between
Gabriel(a)’s intended vote for corrupt but competent politicians as compared to clean but incompetent
politicians. Among those respondents who were told the mayor was clean but incompetent, fully 62%
of them reported that Gabriela would be somewhat or very likely to support the mayor. In contrast, only
28% of respondents in the corrupt but competent treatment group believed that Gabriela would
support the mayor. This 34 percentage point difference is in the opposite direction of what the tradeoff
hypothesis would predict, and it is significant at the .01 level. The negative reaction to corruption among
our sample is, on average, much stronger than the positive reaction to public goods provision.
Also of interest from table 2 are the results for the subset of respondents who received no
information about competence. When respondents are told that the politician is corrupt but not given
information about competence, their response about Gabriel(a)’s voting intention parallels that of
respondents in the corrupt and incompetent category, whereas the voting behavior of respondents who
are told that the politician is clean but not given information about competence parallels that of
respondents in the clean and competent category. Although far from conclusive, this suggests that
respondents think that corruption and competent public goods provision are negatively correlated,
which also would speak against the existence of a rouba, mas faz mentality among Brazilian voters.
Although we find little evidence that Brazilian voters as a whole are forgiving of corruption when
politicians are otherwise competent, we build on existing results in the literature to explore the
22 This may be due in part to the fact that the 13% support for a corrupt, incompetent politician
approaches the lower bound of possible support.
17
possibility that “rouba, mas faz” does describe the attitudes of some particular subsets of the
population.
Does Socioeconomic Status Matter for Attitudes towards Corruption?
As described above, the results of non‐experimental survey research on rouba, mas faz in Brazil
suggest that there are significant differences in the idea’s resonance across social classes. We examine
the relationship between socioeconomic status and information about corruption by comparing the
effects of the prompts across social classes, using a standard Brazilian measure of social class that draws
on the household’s access to goods and services, as well as the head of household’s education level.23
Table 3 summarizes the results for respondents in the lowest and highest social classes.24 Once again,
we highlight in bold the responses that reflect most directly on the tradeoff hypothesis.
The results suggest the exact opposite conclusion about the relationship between corruption
tolerance and social class than that reached by previous surveys that relied on direct questioning, rather
than an experimental set‐up. As is clear from the table, respondents in the lowest classes have a
powerful negative reaction to information about corruption. For either level of purported competence,
information about corruption results in a sharp drop‐off in support. In addition, when we compare
23 We define upper class as those falling into social classes A1, A2 and B1; middle class as those falling
into social classes B2, C1 and C2; and poor as those falling into social classes D and E. The original set of
social class designations is standard and is determined by a national formula based on the number of
nine different household goods and services (including hired household help) to which respondents
have access and the head of household’s highest level of educational attainment.
24 In the interests of space, we omit results for the middle classes, whose responses fall in between
those reported here, although they are closer to the results reported for the poor. In our table, the
lowest classes make up about 25% of the sample, whereas the upper classes make up about 9% of the
sample. This imbalance reflects Brazil’s highly unequal distribution of income.
18
responses among the poor who were in the corrupt, competent treatment with those in the clean,
incompetent treatment, we find no evidence of widespread support for the tradeoff hypothesis. Fully 65
percent of respondents replied that Gabriela would vote for the clean, although incompetent mayor,
whereas only 32 percent replied that she would vote for the corrupt, yet competent mayor.
Results for upper class respondents show some similarities but also stark differences. The
information hypothesis appears to have significant explanatory power for this group, as well. Once
again, we see that, for a given level of competence, mayors who are revealed to be corrupt receive far
less support than those purported to be clean. Most striking for this group, however, is that, in contrast
to the lower class and pooled population results, the tradeoff hypothesis finds much more support. The
proportion of upper class respondents who say that Gabriel will vote for a corrupt but competent
politician (0.48) is statistically indistinguishable from the proportion who say that Gabriel will vote for a
clean but incompetent politician (0.54). In addition, although these upper class voters clearly punish
corruption, they withdraw support from corrupt politicians to a lesser degree than respondents from
lower social classes.
What might explain the greater tolerance for corruption among the upper class? This result may
reflect the fact that upper class voters tend to be more sophisticated consumers of political information.
Perhaps these voters recognize the complex institutional and political environment Brazilian mayors
face and therefore reward accomplishments, even when accompanied by corruption, because of the
significant barriers to the effective distribution of public goods. We explore this possibility in greater
detail in the next section, where we examine the explanatory power of the information and tradeoff
hypotheses among respondents with varying levels of political interest. It also is possible that wealthy
voters are more likely to personally know or identify with politicians, or even to be the direct
beneficiaries of corrupt transactions with politicians. This personal benefit or personal sympathy with
19
politicians among the Brazilian elite might explain the results we see, although our data do not provide
any direct evidence on either of these points.
With regard to the sharp contrast between our results and those of previous surveys that asked
directly about the tradeoff between corruption and competence, we can only speculate, but we note
that the most privileged respondents to standard survey questions are more likely to be attuned to
international norms about corruption and therefore more subject to social desirability bias in the
context of direct questioning. It may be that the experimental setting is most useful for uncovering the
true opinions of this group. The possibility that social desirability bias is particularly strong among the
upper classes is supported by the distribution of non‐responses to the direct question used in Figueiredo
(2004). The number of “other” or “no responses” to the question steadily increased with income: nine
percent of the lowest income respondents failed to choose one of the two proffered options (clean but
incompetent or corrupt but efficient), as compared to 22 percent of the highest income respondents
who did so.
It is also possible that our results reflect a genuine shift in attitudes among Brazil’s poor over the
past decade. Renno (2007) notes that surveys have detected a shift in the salience of corruption. In
2002, only two percent of respondents in Brazil’s National Election Study (ESEB) cited corruption as the
country’s worst problem, whereas that share rose to 42 percent in 2006. Both continued corruption
scandals and the concerted efforts of advocacy and professional groups, like the Brazilian Magistrate’s
Association, have kept this issue on the public agenda since that time. It may be the case that poor
voters at the time of our survey were simply more aware of corruption and its ills than was the case in
earlier surveys. This remains an area for future study.
Do Politically Interested Citizens Support Corrupt Politicians?
In the above discussion of social class effects, we suggest that the greater tolerance for corrupt
yet competent politicians among respondents in the highest social classes might be due to political
20
sophistication. Although we do not have a direct measure of political sophistication or knowledge for
our respondents, we do have information about political interest, which is likely to be closely related
(Verba, Schlozman and Brady 1995; Delli Carpini and Keeter 1996; Prior 2005). Funk (1996) finds that
politically‐knowledgeable respondents react differently to information about scandals involving
politicians depending on the type of scandal. In our data, we describe as politically interested the subset
of respondents who report discussing politics with their friends and family very often (as compared to
often, rarely or never). The number of respondents who fall into this highest interest category is small, a
little over six percent of our sample. These respondents are more likely to be male, highly educated and
members of the upper class than the average respondent.
As table 4 shows, this subgroup appears to value competence more than those who are not as
interested in politics. Fifty percent of politically‐interested respondents who received a vignette about a
corrupt but competent politician said that Gabriel(a) would vote to reelect the mayor, a rate much
higher than that among the not‐politically‐interested subset (p < 0.06). In addition, those who received
a vignette about a clean and competent politician were almost universal in their belief that Gabriel(a)
would vote to reelect the mayor. Looking at the off‐diagonal cells for the politically interested, we see
that respondents in this group were almost as likely to say that Gabriel(a) would vote for a corrupt but
competent politician as that Gabriel(a) would vote for a clean but incompetent one – the two
proportions are statistically indistinguishable. All three results are very similar to those we find for
respondents in the highest social classes. Once again, this suggests a willingness to make a rouba, mas
faz tradeoff among a particular group of voters.
Although we do not measure political information directly, these politically interested
respondents may be more knowledgeable and sophisticated political observers. These voters may treat
the delivery of public goods as a difficult task and therefore be (1) more tolerant of corruption when a
mayor is competent, (2) particularly supportive of competent and clean mayors and (3) particularly
21
critical of incompetent mayors. As an indication that these voters do, in fact, attach greater importance
to the information about public goods provision, we find that they are more likely than other voters to
attribute skill to the mayor in other areas. Whereas 88 percent of the highly interested respondents in
the competent treatment conditions said that the mayor would be very competent or competent in
acquiring federal resources for the municipality, only 65 percent of other respondents thought this.
These significantly different (p < 0.01) assessments of the skills in another area of a mayor who is
described as good at creating public works projects provide evidence that high‐interest voters are
interpreting information about competence in a more positive light than the average voter.25
Rethinking the Information Hypothesis
Although the phrase “rouba, mas faz” has widespread recognition in Brazil, the results of a
nationwide survey experiment offer little evidence that Brazilian voters actually adopt such a mentality
when evaluating the suitability of candidates for office. In fact, we find just the opposite. Voters in our
survey appear to be very sensitive to information about corruption—indeed so sensitive that the vast
25 An alternative explanation – suggested to us by Bob Pahre – for why there is more support for the
tradeoff hypothesis among highly interested or wealthy voters is that these respondents, asked to
describe what Gabriel(a) would do, answer not for someone like themselves, but rather have in mind
how an “average” respondent would answer. Given the widespread usage of the phrase, these voters
might attribute a rouba, mas faz mentality to Garbiel(a). However, note that the responses among
these groups differ from those of respondents at large for many of the treatment categories, including
those that do not evoke a popular phrase. They are, for example, somewhat more critical of clean but
incompetent mayors and reward clean and competent mayors at exceptionally high rates. In addition,
these upper class or highly interested respondents are more likely to think that a clean politician is
skilled in other areas. All of this then suggests that these respondents simply evaluated mayoral
performance differently than most other respondents.
22
majority are willing to overlook poor public goods provision as long as a politician is clean. Overall, our
results are much more consistent with what we call the information hypothesis—the argument that,
within a democracy, citizens will vote against corrupt politicians as long as they have knowledge of that
corruption. Given the cultural currency of the phrase “rouba, mas faz,” the fact that our results are so
resoundingly at odds with the tradeoff hypothesis is at first glance surprising. However, they are
consistent with other recent empirical work, including that of Ferraz and Finan (2008), Pereira, Melo,
and Figueiredo (2009), Pereira, Rennó, and Samuels (ND) and de Figueiredo, Hidalgo and Kasara (2010),
all of whom who use voting outcomes to examine the effect of information about corruption on
reelection.26 Our research provides individual‐level results consistent with these aggregate‐level findings
and provides evidence of what Pereira, Rennó, and Samuels (ND) call the “mass mechanism” of electoral
accountability.
Both their research and our research provide evidence that information about corruption,
where available, leads Brazilian voters to vote against corrupt politicians. This, in turn, suggests that the
persistent corruption in Brazil is a result of a lack of information. But is this really the case? Although we
discuss the limitations of Brazil’s media below, it has been instrumental in publicizing a number of major
corruption scandals in recent years.27 Data from the 2006 Brazilian National Election Study show that
corruption has become highly salient, with over 40% of respondents naming corruption as the single
26 Pereira, Renno and Samuels explore both “mass” and “elite” mechanisms of accountability, whereas
the other three papers, like ours, focus only on the link between the revelation of information about
corruption and voter behavior.
27 In a forthcoming volume, Power and Taylor (ND) characterize the media this way: “While not all media
organizations are autonomous of political pressures, the pluralism of the national press is a useful
antidote: the scope of media coverage is broad, and the media cooperates closely with both the central
institutions and with civil society organizations” (23).
23
biggest political problem in the country (Renno 2007). Furthermore, data from our survey show that
Brazilians have no illusions about the cleanliness of their politicians: when asked how common it is for
Brazilian politicians to take bribes, 78 percent of respondents said it was very or somewhat common.
Instead of pointing to the importance of broad information about corruption for limiting support
for corrupt politicians, our results instead point to the relevance of specific, credible, and available
information in affecting voting behavior. As always, we need to be cautious in extrapolating from an
experimental setting to the real world. As Barabas and Jerit (2010) point out, survey prompts necessarily
confront respondents with direct information that is uncluttered by the distractions of real life. In this
case, the dissimilarities between the experiment and the real world can serve to refine the information
hypothesis as currently presented in the literature. Our survey presents specific information about a
particular politician (bribe‐taking by a current mayor), presents it from a credible source (the survey was
run by IBOPE, the largest and most well‐known public opinion firm in Brazil), and presents it in close
temporal proximity to a hypothetical voting decision. Similarly, the results reported by Ferraz and Finan
(2008) and Pereira, Melo and Figueiredo (2009) explore the effects of the revelation, by an independent
auditing agency, of information about specific corrupt acts in a mayoral administration. They use data
from Brazil’s national and state municipal audit programs and municipal elections to show that
reelection rates were significantly lower in cities where a preelection audit found corruption violations.
Pereira, Melo and Figueiredo explore the effects of corruption revelations from two waves of audits,
one conducted over a year prior to the election and one conducted within the four‐month official
campaign period leading up to elections. They find that only the latter had a significant effect of mayoral
reelection, suggesting that only when information about corruption is at the top of voters’ minds do
they act on it. Ferraz and Finan find that where there are more local radio stations, the negative effects
of corruption findings are enhanced, again suggesting that the spread of credible information among
24
voters plays a critical role in linking corruption, voter attitudes toward corruption, and politician losses
at the ballot box.28
In contrast, the type of information Brazilian voters typically encounter about corruption may be
quite different; it may be vague, diffuse in who it accuses of wrong‐doing, disseminated by partisan
sources, and not concentrated in the period immediately before elections. Although Brazil has a large,
privately‐owned and active media, its penetration is quite uneven. As Freedom House (2009) reports,
the challenges to information provision are particularly acute away from the main metropolitan areas.
Ferraz and Finan (2008) document that many municipalities do not have even a single radio station.
The importance of distinguishing between diffuse and particular information is reinforced by
patterns in the data from two questions that we asked at the end of the survey. We asked voters how
common they thought it was for politicians in Brazil to take bribes: very common, somewhat common,
not very common or not common. Immediately afterwards, we asked voters whether they personally
had ever voted for a corrupt politician.29
28 In contrast, note that the experimental corruption organized by Banerjee et. al (2010), discussed
earlier, involved a broad discussion about the ills of corruption, rather than a specific description of
corruption on the part of any particular politician. If we are correct, this lack of specific information,
rather than a disregard for corruption on the part of those surveyed, might explain the lack of a
treatment effect on voting outcomes in their study.
29 Both of these questions were asked at the end of the survey and therefore after respondents had
heard the experimental vignette. However, we detect no effect of the stimulus on the ways in which
people answered these two questions – the proportion of respondents that say bribe‐taking is very
common or somewhat common is indistinguishable between the group of respondents that heard a
corruption vignette and the group that heard a vignette that said the politician was not corrupt (p <
25
As noted above and as can be seen in table 5, while 78 percent of the respondents think it is
very common or somewhat common for politicians to take bribes, only 21 percent say that they ever
personally have voted for a corrupt politician. Even among those who say it’s “very common” for
politicians to take bribes, fully 73 percent (601 out of 818 respondents) say they have never voted for a
corrupt politician.30 Much like the familiar observation that the average American “hates Congress but
loves their congressperson,” Brazilian voters appear to think that corruption is everywhere and yet
report at high rates that they have never themselves voted for a corrupt politician. Although Brazilian
voters have a diffuse awareness of corruption, this appears unlikely to affect voting behavior unless – as
in our experimental vignette – voters are given credible, specific, and cognitively available information
that ties corruption to a particular politician.
Conclusions
In summary, we find that Brazilian voters are very sensitive to information about corruption and
are unlikely to support corrupt politicians, even when these politicians deliver public goods. What do our
results tell us about the prospects for limiting corruption in Brazil and other similarly situated countries?
In light of the cultural resonance of the phrase “rouba, mas faz,” we might think of Brazil as a hard case
for the information hypothesis. The fact that we find such strong evidence for it and against the tradeoff
hypothesis suggests that broad cultural explanations of corruption will be limited in their explanatory
power elsewhere, as well. Presumably it is easier to repair an information deficit than to change
preference orderings, and so to the extent that voters believe corruption is costly, even if it is coupled
with performance in other areas, these results should cheer opponents of corruption. A number of
0.63); similarly there is no difference in the proportion of people in each condition who said that they
had ever voted for a corrupt politician (p < 0.86).
30 Note that the question order should, if anything, make consistent replies more likely. This makes the
high degree of incongruence between the general and specific questions even more striking.
26
efforts are currently underway in Brazil to give voters more precise information about candidates’
corrupt activities during the course of the electoral campaign (and to outlaw some corrupt candidates
from running for office). Policies like these, if implemented elsewhere, might have a greater effect on
voter behavior than more general anti‐corruption campaigns. Our results suggest the possibility of
improving accountability and ameliorating or eliminating the paradox of “unpopular corruption and
popular corrupt politicians” (Kurer 2001, 63).
Our results sound some cautionary notes, as well. If, as we argue, the distribution of precise,
credible information is key for generating a voter response, this task is still a substantial one in many
young and lower and middle‐income democracies. Its success will rely on either independent
government agencies and/or active investigative journalism and widespread dissemination of
information. Precisely where corruption is widespread, obtaining and distributing credible information is
particularly difficult. Corrupt politicians may act to prevent the establishment of independent
government auditing agencies. Journalists and other watchdogs may fear repercussions from politicians
and candidates (Stanig 2010). Even where corruption is uncovered, the ability to attract and hold the
public’s attention may vary with the levels of corruption. Where most politicians are clean, identifying
the few corrupt politicians may be relatively easy for journalists and other watchdogs, and then that
information can be readily acted upon by voters. In contrast, if most or many politicians are corrupt, it
may be difficult for voters to identify and recall credible, detailed information about corruption. Our
experiment ‐‐ in the context of Brazil's experience with corruption ‐‐ suggests that although Brazilian
voters may lack credible information, they do not condone corruption.
27
Works Cited
Ades, Alberto and Rafael Di Tella. 1999. “Rents, Competition and Corruption,” American Economic
Review 89:4 (September), pp. 982‐94.
Adsera, Alicia, Carles Boix and Mark Payne. 2003. “Are You Being Served? Political Accountability and
the Quality of Government,” Journal of Law, Economics and Organization 19:2, pp. 445‐90.
Almeida, Alberto Carlos. 2006. Por Que Lula? Rio de Janeiro: Record.
Anderson, Christopher J. and Yuliya V. Tverdova. 2003. “Corruption, Political Allegiances and Attitudes
Toward Government in Contemporary Democracies,” American Journal of Political Science 47:1
(January), pp. 91‐109.
Banerjee, Abhijit, Donald Green, Jennifer Green and Rohini Pande. 2010. “Can Voters be Primed to
Choose Better Legislators? Experimental Evidence from Rural India,” Working Paper.
Barabas, Jason and Jennifer Jerit. 2010. “Are Survey Experiments Externally Valid?” American Political
Science Review 104:2 (May), pp. 226‐42.
Bardhan, Pranab. 1997. “Corruption and Development: A Review of the Issues,” Journal of Economic
Literature 35:3 (September), pp. 1320‐46.
Blake, Charles H. 2009. “Public Attitudes Toward Corruption,” in Corruption and Democracy in Latin
America. Charles H. Blake and Stephen D. Morris, eds. Pittsburgh: University of Pittsburgh
Press, pp. 94‐107.
Canache, Damarys, Jeffery J. Mondak and Ernesto Cabrera. 2000. "Voters and the Personal Vote: A
Counterfactual Simulation," Political Research Quarterly 53, pp. 663‐76.
Caselli, Francesco and Massimo Morelli. 2004. “Bad Politicians,” Journal of Public Economics 88:3‐4, pp.
759‐82.
28
Chang, Eric C.C. and Nicholas N. Kerr. 2009. “Do Voters Have Different Attitudes Toward Corruption?
The Sources and Implications of Popular Perceptions and Tolerance of Political Corruption,”
Afrobarometer Working Paper No. 116, December.
Chong, Alberto, Ana L. De La O, Dean Karlan and Leonard Wantchekon. 2010. “Information
Dissemination and Local Governments’ Electoral Returns: Evidence from a Field Experiment in
Mexico.” Working Paper.
Delli Carpini, Michael X. and Scott Keeter. 1996. What Americans Know About Politics and Why It
Matters. New Haven: Yale University Press.
DeMaio, Theresa J. 1984. “Social Desirability and Survey Measurement: A Review,” in Surveying
Subjective Phenomena. Vol. 2. Charles F. Turner and Elizabeth Martin, eds. New York: Russell
Sage Foundation, pp. 257‐82.
Desposato, Scott W. 2006. “From Revolution to Rouba Mas Faz? Lula's Reelection Campaign in Brazil.”
ReVista: Harvard Review of Latin America 5:1, pp. 29‐32.
Djankov, Simeon, Rafael La Porta, Florencio Lopez‐de‐Silanes and Andrei Shleifer. 2002. “The
Regulation of Entry,” Quarterly Journal of Economics 117:1, pp. 1‐37.
Druckman, James N., Donald P. Green, James H. Kuklinski and Arthur Lupia. 2006. “The Growth and
Development of Experimental Research in Political Science,” American Political Science Review
100:4, pp. 627‐635.
Ferraz, Claudio and Frederico Finan. 2008. “Exposing Corrupt Politicians: The Effects of Brazil's Publicly
Released Audits on Electoral Outcomes,” Quarterly Journal of Economics 123:2, pp. 703‐45.
Figueiredo, Marcus. 2004. “Convite ao ‘Rouba, mas faz,’” Insight Inteligencia, pp. 24‐7, available at
http://www.insightnet.com.br/inteligencia/25/PDF/0125.pdf.
29
de Figueiredo, Miguel, F. Daniel Hidalgo and Yuri Kashara. 2010. “When Do Voters Punish Corrupt
Politicians? Experimental Evidence from Brazil,” available at
http://epstein.law.northwestern.edu/research/LPEMdeFig.pdf.
Fisman, Raymond and Roberta Gatti. 2002. “Decentralization and corruption: evidence across
countries,” Journal of Public Economics 83(3), pp. 325‐345.
Fisman, Raymond J. and Edward Miguel. 2006. “Cultures of Corruption: Evidence from Diplomatic
Parking Tickets,” NBER Working Paper No. 12312, June.
Fleischer, David. 1997. “Political Corruption in Brazil: The Delicate Connection with Campaign Finance,”
Crime, Law & Social Change 25, pp. 297‐321.
Freedom House. 2009. Country Report: Brazil, available at
http://www.freedomhouse.org/template.cfm?page=251&country=7573&year=2009.
Freedom House. 2010. Freedom of the Press 2010, available at
http://www.freedomhouse.org/uploads/pfs/371.pdf.
Funk, Carolyn L. 1996. “The Impact of Scandal on Candidate Evaluations: An Experimental Test of the
Role of Candidate Traits.” Political Behavior 18(1).
Gaines, Brian J., James H. Kuklinski, and Paul J. Quirk. 2007. "Rethinking the Survey Experiment," Political
Analysis 15, pp.1‐21.
Gatti, Roberta, Stefano Paternostro and Janele Rigolini. 2003. “Individual Attitudes toward Corruption:
Do Social Effects Matter?” World Bank Policy Research Working Paper No. 3122, Washington,
D.C.
Geddes, Barbara. 1994. Politician's Dilemma. Berkeley: University of California Press.
Geddes, Barbara and Artur Ribeiro Neto. 1992. “Institutional Sources of Corruption in Brazil,” Third
World Quarterly 13:4, pp. 641‐60.
30
Gerring, John, and Strom Thacker. 2004. “Political Institutions and Corruption: The Role of Unitarism
and Parliamentarism,” British Journal of Political Science 34:2, pp. 295‐330.
‐‐‐‐‐. 2005. “Do Neoliberal Policies Deter Political Corruption?” International Organization 59:1, pp.
233‐54.
Golden, Miriam, and Eric Chang. 2006. “Electoral Systems, District Magnitude and Corruption,” British
Journal of Political Science 37:1, pp.115‐137.
Gupta, Sanjeev, Hamid Davoodi and Rosa Alonso‐Terme. 2002. “Does Corruption Affect Income
Inequality and Poverty?” Economics of Governance 3:1, pp. 23‐45.
Gyimah‐Brempong, Kwabena. 2002. “Corruption, Economic Growth, and Income Inequality in Africa,"
Economics of Governance 3:3, pp. 183‐209.
Heidenheimer, Arnold J., Michael Johnston and Victor LeVine, eds. 1989. Political Corruption: A
Handbook. New Brunswick, NJ: Transaction Publishers.
Huntington, Samuel P. 1968. Political Order in Changing Societies. New Haven: Yale University Press.
Keefer, Philip and Razvan Vlaicu. 2008. “Credibility, Clientelism and Democracy,” Journal of Law,
Economics and Organization 24:2 (Fall), pp. 1‐36.
Kulisheck, Michael R. and Jeffery J. Mondak. 1996. “Candidate Quality and the Congressional Vote: A
Causal Connection?” Electoral Studies 15, pp. 237‐53.
Kunicova, Jana and Susan Rose‐Ackerman. 2005. “Electoral Rules and Constitutional Structures as
Constraints on Corruption,” British Journal of Political Science 35 (4): 573‐606.
Kurer, Oskar. 2001. “Why Do Voters Support Corrupt Politicians?” in The Political Economy of
Corruption. Arvind K. Jain, ed. New York: Routledge, pp. 63‐86.
La Porta, Rafael, Florencio Lopez‐de‐Silanes, Andrei Shleifer and Robert Vishny. 1999. “The Quality of
Government,” Journal of Law, Economics and Organization 15:1, pp. 222‐79.
31
Lambsdorff, Johann Graf. 1996. “Institutions and Economic Performance: Cross‐Country Tests Using
Alternative Institutional Measures,” Economics and Politics VII (November), pp. 207‐27.
‐‐‐‐‐. 2006. “Causes and Consequences of Corruption: What Do We Know from a Cross‐Section of
Countries?” in International Handbook on the Economics of Corruption. Susan Rose‐Ackerman,
ed. Northampton, MA: Edward Elgar.
Lawson, Chappell, et. al. 2007. The Mexico 2006 Panel Study.
http://web.mit.edu/polisci/research/mexico06.
Leff, Nathaniel. 1964. “Economic Development through Bureaucratic Corruption,” American Behavior
Scientist 8, pp. 8‐14.
Li, Hongyi, Lixin Colin Xu, and Heng‐Fu Zou. 2000. “Corruption, Income Distribution and Growth,”
Economics & Politics 12:2, pp. 155‐82
Manzetti, Luigi and Carole J. Wilson. 2007. “Why Do Corrupt Governments Maintain Public Support?”
Comparative Political Studies 40:8, pp. 949‐70.
Mauro, Paulo. 1995. “Corruption and Growth,” Quarterly Journal of Economics 110:3, pp. 681‐712.
Messner, Matthias and Mattias Polborn. 2006. “Paying Politicians,” Journal of Public Economics 88:12,
pp. 2423‐2445.
Mishler, William and Richard Rose. 2001. “What are the Origins of Political Trust? Testing Institutional
and Cultural Theories in Post‐Communist Societies,” Comparative Political Studies 34:1, pp. 30‐
62.
Mondak, Jeffrey and Robert Huckfeldt. 2006. “The Accessibility and Utility of Candidate Character in
Electoral Decision Making,” Electoral Studies 25, pp. 20‐34.
Morris, Stephen D. 1991. Corruption and Politics in Contemporary Mexico. Tuscaloosa: University of
Alabama Press.
32
Nederhof, Anton J. 1985. “Methods of Coping with Social Desirability Bias: A Review,” European Journal
of Social Psychology 15:3, pp. 263‐80.
Nye, Joseph. 1967. “Corruption and Political Development: A Cost‐Benefit Analysis,” American Political
Science Review 61:2, pp. 417‐27.
Pande, Rohini. 2011. “Can Informed Voters Enforce Better Governance? Experiments in Low Income
Democracies,” Annual Review of Economics 3.
Pereira, Carlos, Marcus André Melo and Carlos Mauricio Figueiredo. 2009. “The Corruption‐Enhancing
Role of Re‐Election Incentives? Counterintuitive Evidence from Brazil's Audit Reports,” Political
Research Quarterly 62:4, pp. 731‐44.
Pereira, Carlos, Lucio Rennó and David Samuels. ND. “Corruption, Campaign Finance and Reelection,”
in Corruption and Democracy in Brazil: The Struggle for Accountability. Timothy J. Power and
Matthew M. Taylor, eds. Unpublished manuscript.
Persson, Torsten, Guido Tabellini, and Francesco Trebbi. 2003. “Electoral Rules and Corruption,”
Journal of the European Economic Association 1:4, pp 958‐989.
Pharr, Susan J. 2000. “Officials’ Misconduct and Public Distrust: Japan and the Trilateral Democracies,”
in Disaffected Democracies: What’s Troubling the Trilateral Countries? Susan J. Pharr and
Robert D. Putnam, eds. Princeton: Princeton University Press.
Power, Timothy J. and Matthew M. Taylor, eds. ND. Corruption and Democracy in Brazil: The Struggle
for Accountability. Unpublished book manuscript.
Prior, Markus. 2005. “News vs. Entertainment: How Increasing Media Voice Widens Gaps in Political
Knowledge and Turnout,” American Journal of Political Science 49:3, pp. 577‐92.
Renno, Lucio R. 2007. “Escandalos e voto: as eleicoes presidenciais brasileiras de 2006,” Opiniao
Publica 13:2, pp. 260–82.
33
‐‐‐‐‐. ND. “Corruption and Voting,” in Corruption and Democracy in Brazil: The Struggle for
Accountability. Timothy J. Power and Matthew M. Taylor, eds. Unpublished book manuscript.
Rose, Richard, William Mishler and Christian Haerpfer. 1998. Democracy and Its Alternatives:
Understanding Post‐Communist Societies. Baltimore: Johns Hopkins University Press.
Rose‐Ackerman, Susan. 1978. Corruption: A Study in Political Economy. New York: Academic Press.
Rundquist, Barry S., Gerald S. Strom, and John G. Peters. 1977. “Corrupt Politicians and Their Electoral
Support: Some Experimental Observations,” American Political Science Review 71, pp. 954‐63.
Queiroz, Adolpho and Carlos Manhanelli. 2009. “Beve história dos slogans políticos nas eleições do
Brasil republican,” Signos do Consumo 1:2, pp. 101‐22.
Samuels, David. 2004. “As bases do petismo [The bases of ‘petismo’],” Opinião Pública 10:2, pp. 221‐41,
available at http://www.scielo.br/scielo.php?script=sci_arttext&pid=S0104‐
62762004000200002&lng=en&nrm=iso&tlng=pt.
‐‐‐‐. 2008. “A evolução do petismo (2002‐2008) [The evolution of ‘petismo’],” Opinião Pública 14:2, pp.
302‐18, available at http://www.scielo.br/scielo.php?pid=S0104‐
62762008000200002&script=sci_arttext&tlng=en.
Seligson, Mitchell A. 2002. “The Impact of Corruption on Regime Legitimacy: A Comparative Study of
Four Latin American Countries,” Journal of Politics 64:2, pp. 408‐33.
Shleifer, Andrei and Robert Vishny. 1993. “Corruption,” Quarterly Journal of Economics 108:3, pp. 599‐
617.
Sniderman, Paul M. and Douglas Grob. 1996. “Innovations in Experimental Design in General
Population Attitude Surveys,” Annual Review of Sociology 22, pp. 377‐399.
Stanig, Piero. 2010. “Regulation of Speech and Media Coverage of Corruption: An Empirical Analysis of
the Mexican Press,” London School of Economics.
34
Treisman, Daniel. 1998. “The Causes of Corruption: A Cross‐National Study,” University of California,
Los Angeles, April, available at
http://www.isr.umich.edu/cps/pewpa/archive/archive_98/19980019.pdf.
‐‐‐‐. 2000. “The Causes of Corruption: A Cross‐National Study,” Journal of Public Economics 76:3, pp.
399‐457.
‐‐‐‐‐. 2007. “What Have We Learned About the Causes of Corruption from Ten Years of Cross‐National
Empirical Research?” Annual Review of Political Science 10, pp. 211‐44.
Verba, Sidney, Kay Lehman Schlozman and Henry E. Brady. 1995. Voice and Equality: Civic Voluntarism
in American Politics. Cambridge: Harvard University Press.
Villaméa Cotta, Luiza Cristina. 2008. Adhemar de Barros (1901‐1969): A Origem do ‘Rouba, Mas Faz,’”
masters thesis, University of São Paulo.
Ward, Jake. 2006. “Corruption Makes Brazil Think Twice: Lula Faces Runoff,” The Georgetown Public
Policy Review Online, available at
http://www9.georgetown.edu/grad/gppi/GPPIReview/index.cfm?tpl=subject&subjectID=32.
Weitz‐Shapiro, Rebecca. 2008. “Clientelism and the middle class: results of a survey experiment in
Argentina,” presented at the APSA annual meeting.
Weyland, Kurt. 1998. “The Politics of Corruption in Latin America,” Journal of Democracy 9:2, pp. 108‐
21.
World Bank. 1997a. Helping Countries Combat Corruption: The Role of the World Bank. Washington,
D.C.: The World Bank Group.
‐‐‐‐‐. 1997b. World Development Report 1997: The State in a Changing World. Washington, D.C.: The
World Bank Group.
35
High Competence Low Competence No Information on Competence
No Corruption B1
(N=334) B2
(N=335) B3
(N=333)
Corruption B4
(N=335) B5
(N=334) B6
(N=331)
Table 1. Six Treatment Categories. The variables in the cells note the proportion of respondents in the treatment category who express their belief that Gabriel(a) will vote for the incumbent politician. The treatments also varied in a third dimension not shown here – the announced party of the incumbent mayor – with roughly equal numbers of respondents assigned to hear about the PT (Partido dos Trabalhadores, Workers Party) and the PSDB (Partido da Social Democracia Brasileira, Brazilian Social Democratic Party) in each of the six cells shown.
36
(1) All Vignettes
(2) Competent Vignettes
(3) Incompetent Vignettes
(4) Vignettes without
Competence Information
Not Corrupt
0.78 (0.01) N=989
0.88 (0.02) N=329
0.62 (0.03) N=331
0.83 (0.02) N=329
Corrupt
0.19 (0.01) N=985
0.28 (0.02) N=328
0.13 (0.02) N=328
0.16 (0.02) N=329
Difference
0.58 (0.02)
0.59 (0.03)
0.49 (0.03)
0.66 (0.03)
p‐value on H0: No Difference
0.00 0.00 0.00 0.00
Table 2. Respondents’ Assessments of Whether Gabriel(a) Will Vote for Corrupt and Clean Politicians. Cells report proportion of respondents in each treatment condition who said that Gabriel(a) would be willing to vote for the incumbent mayor.
37
Lowest classes (D,E) Highest classes (A1,A2, B1)
Competent Vignettes
Incompetent Vignettes
Competent Vignettes
Incompetent Vignettes
Not Corrupt
0.88 (0.04) N=81
0.65 (0.05) N=87
1.00 (0.00) N=22
0.54 (0.03) N=35
Corrupt
0.32 (0.05) N=87
0.13 (0.04) N=81
0.48 (0.03) N=23
0.23 (0.08) N=30
“Rouba Mas Faz” Difference
0.33 (0.07)
0.06 (0.14)
p‐value on H0: No Difference
0.00 0.64
Table 3. Variation in Voting Patterns Across Social Class. Cells report proportion of respondents in each treatment condition who said that Gabriel(a) would be willing to vote for the incumbent mayor.
38
Politically Interested Not Politically Interested
Competent Vignettes
Incompetent Vignettes
Competent Vignettes
Incompetent Vignettes
Not Corrupt
0.95 (0.05) N=19
0.54 (0.10) N=24
0.87 (0.02) N=310
0.63 (0.03) N=307
Corrupt
0.50 (0.13) N=16
0.21 (0.08) N=24
0.27 (0.03) N=311
0.13 (0.02) N=304
“Rouba Mas Faz” Difference
0.04 (0.17)
0.36 (0.04)
p‐value on H0: No Difference
0.80 0.00
Table 4. Variation in Voting Patterns Across Levels of Political Interest. Cells report proportion of respondents in each treatment condition who said that Gabriel(a) would be willing to vote for the incumbent mayor.
39
40
How Common Is It For Politicians to Take Bribes?
Have Not Voted for a Corrupt Politician
Have Voted for a Corrupt Politician
Total
Very Common 601 217 818 (45 percent)
Somewhat Common 488 117 605 (33 percent)
Not Very Common 232 41 273 (15 percent)
Not At All 129 13 142 (8 percent)
Total 1,450 (79 percent)
388 (21 percent)
1,838
Table 5. Perceptions of Actual Corruption. The table reports respondents beliefs about how common it is for politicians in Brazil to take bribes and also whether or not they personally have voted for a corrupt politician.