Abstract
In long-standing democracies, the partisan attachments of most citizens are stable and not responsive to short-term political events. Recent studies from younger democracies, however, suggest that partisanship may be more malleable in these contexts. In this paper we develop hypotheses about how political corruption might affect voter attachment to the parties of corrupt officials or to the party system as a whole. Using data from an original survey experiment in Brazil, we show that prompts about political corruption shift patterns of partisan attachment for highly educated respondents – specifically, that corruption associated with one political party reduces nonpartisanship and significantly increases identification with other political parties. In contrast, we find that information on corruption has no consistent measurable effect on partisanship for less educated respondents. We conclude by discussing the implications of malleable partisanship for democratic accountability.
Introduction
Can individual political events change partisan attachment? The dominant view of partisanship as an “unmoved mover” suggests not: in most of the literature, partisanship is seen as a stable personal attribute rather than a pliable attitude that varies over time. This scholarly understanding has been shaped predominantly by the study of US politics and public opinion. What is true of citizens in the United States, however, may not be true of citizens elsewhere – especially those in countries with shorter democratic histories. The political parties in these countries, as compared to those in long-standing democracies, are likely to be younger, less institutionalized, and less deeply familiar to citizens. Individual attachment to these parties, consequently, may be shallower and more malleable.
Building on a nascent literature that examines partisan identity as an outcome variable in younger democracies (Brader and Tucker 2008; Baker et al. 2010; Chong et al. forthcoming; Lupu 2013; Klăsnja and Tucker 2013), we examine whether citizens’ partisan identities are responsive to priming about political corruption. Corruption is a highly salient issue for voters in many young democracies. Whereas observers once treated corruption as innocuous or even beneficial (Huntington 1968), it is now well established that corruption is linked to lower economic growth, decreased investment, and greater income inequality (see Lambsdorff 2006). These pernicious effects are commonly highlighted by national media outlets and international organizations like Transparency International, while public opinion surveys reveal widespread citizen discontent with corruption (e.g., Transparency International 2013).
Existing work on how voters respond to corruption falls into two main categories. The first body of literature is mostly case-driven and explores how citizens’ disgust with systemic corruption serves as a catalyst for major political upheaval, from the impeachment or resignation of individual presidents (Pérez-Liñan 2007) to the collapse of entire party systems (Pharr 2000; Della Porta and Vannucci 2007; Chang, Golden, and Hill 2010). The second body of literature uses survey data to examine how corruption prompts changes in individual attitudes like system support (Anderson and Tverdova 2003; Seligson 2006) and individual behavior like protest participation (Gingerich 2009) or vote choice (Chong et al. forthcoming). In the latter body of literature, recent experimental and quasi-experimental work provides strong evidence that voters punish individual politicians involved in malfeasance (Ferraz and Finan 2008; Pereira, Melo, and Figuereido 2008; Chong et al. forthcoming). 1
Some recent work also explores factors that may mitigate dissatisfaction with corruption (e.g., Anduiza, Gallego, and Muñoz 2013; Winters and Weitz-Shapiro 2013).
We build on both approaches to examine whether information about corruption can affect citizens’ partisan attachments, an important individual-level change that can have system-level consequences. 2 Using original survey experimental evidence from Brazil, we show that priming respondents to associate major political parties with corruption leads to greater engagement with the political system among the most highly educated. Compared to those who have not been primed with information about corruption, highly educated respondents are less likely to describe themselves as nonpartisan and more likely to declare a partisan attachment to smaller political parties. Existing findings about the effects of corruption on voting behavior would not predict these results. Thus, our findings point to the importance of treating partisanship as an outcome distinct from voting behavior and suggest a heretofore unidentified pathway through which negative political information may increase political engagement for at least some voters. The effect of this priming, however, is limited to the highly educated group as we find less evidence of changing patterns of partisanship among less well-educated respondents.
In a similar vein, Chong et al. (forthcoming) show that allegations of corruption can have collateral costs for political parties.
Partisanship is commonly understood as a stable individual characteristic. Originating in the Michigan School of the 1960s and based largely on studies of US politics, the dominant view of partisanship depicts it as an “unmoved mover” formed in adolescence that goes on to affect a wide range of individual attitudes and behaviors (Campbell et al. 1960). 3 While not all scholars would go so far as to place partisanship in the same category as “kinship and ethnic ties” (Gerber, Huber, and Washington 2010), even revisionist scholars who treat partisanship as a “running tally” (Fiorina 1981) concur that partisan identity is slow moving (see Fiorina (2002) and Johnston (2006) for summaries of the debate). In long-standing democracies with entrenched parties, identification persists to a great extent even when a party performs poorly in office (e.g., in terms of managing the economy) (Green, Palmquist, and Schickler 2002). 4 The limited experimental work on this topic is broadly consistent with this view. Attempts by Cowden and McDermott (2000) to manipulate students’ partisan attachments through short-term role playing, for example, produce null results. 5
Despite the large body of empirical literature that examines partisanship outside of the United States, most of the theoretical discussion is driven and dominated by the study of the United States.
In fact, rather than updating their beliefs, there is ample evidence in the United States that partisans are likely to seek out information from news sources that tend to confirm their preexisting party evaluations (e.g., Iyengar and Hahn 2009), to evaluate the same policies or economic conditions differently if they are endorsed by or associated with their preferred party (Bartels 2002; Druckman and Bolsen 2011; Lavine, Johnston, and Steenbergen 2012), and even to act in accordance with incorrect, partisan-induced views (Prior, Sood, and Khanna 2013).
Gerber, Huber, and Washington (2010), on the other hand, are able to increase levels of party identification by providing an instrumental reason to become a partisan (i.e., reminding nonaffiliated voters that only registered partisans can vote in party primaries). Running a field experiment in the US state of Connecticut, they find that latent partisans in the treatment condition were more likely to identify with the party to which they were closest.
Given the strong findings on the stability of partisan identity, the vast majority of literature studying partisanship has examined the effects of partisanship on other outcomes. Building on the large body of empirical work from the long-standing democracies of the United States and Western Europe, recent studies outside of these countries have found evidence that strong partisan identities can be formed and that partisanship can predict political behaviors in other contexts, too. Authors studying Brazil (Samuels 2006; Baker et al. 2010) and Mexico (McCann and Lawson 2003) find relatively high levels of partisan attachment and a fair degree of stability in partisan identification. Using survey experiments, Brader and Tucker (2012) and Samuels and Zucco (2014) provide evidence that voter partisanship in developing democracies has meaningful effects on outcomes like policy preferences.
At the same time, many newer democracies have characteristics that should undermine the strength of party attachment. Where parties are young or democratic elections a recent phenomenon, the creation of long-standing attachments from childhood is impossible for many citizens. 6 Party proliferation and turnover – either over time or at any given moment – may weaken citizens’ incentives to learn about and develop psychological attachments to particular parties. Similarly, legislative weakness in some new democracies undermines the importance of parties and party labels (Fish 2006; Carey and Reynolds 2007).
Some countries that experienced prolonged periods of nondemocracy nonetheless have parties that substantially predate the most recent democratic period. The Peronist party in Argentina is a classic example (Lupu and Stokes 2009; Calvo and Murillo 2012).
In a context where some or all of these characteristics are present, we should be cautious in applying the US model of partisan stability. Where partisan attachments are more malleable, the likelihood that individuals will declare themselves partisans of given parties is more likely to be affected by recent political events. In Brazil, Baker et al. (2010) find that declared partisanship within a six-wave panel study experiences short-term shifts in response to the performance of national and, to some extent, subnational officials. In a field experiment in Mexico, Chong et al. (forthcoming) show that distributing information about high levels of municipal corruption decreases identification with the party of the incumbent politician. Relatedly, studies from Eastern Europe and Latin America have shown that rates of expressing partisan identification increase among survey respondents who receive more information about parties’ policy positions (Brader and Tucker 2008; Lupu 2013). Taken together, these studies suggest that changes in the information environment can alter individual partisanship in new democracies in the short and medium term.
How might individuals’ declared partisanship change after they are exposed to negative political information? In this section, we review likely ways in which individuals might change their expressed partisan identities after being primed to think about corruption in conjunction with a particular party. 7 We then describe how individual shifts in partisanship might be reflected in the aggregate.
This project is motivated by the desire to examine how real-world political information about corruption affects expressions of partisanship in the short term. The design of our experiment involves a hypothetical politician. Therefore, respondents do not receive factual information about existing parties, although they are still primed with exposure to an association between a particular political party and the idea of corruption. We use the language of information and priming in this section.
The possible pathways of partisan change depend on how individuals would have expressed their partisan leanings in the absence of priming. Before hearing about a corrupt politician, a given individual might be (a) a sympathizer with the party of that politician, (b) a sympathizer of another party, or (c) a nonpartisan. Table 1 summarizes the range of reactions for these three types of individuals confronted with information linking a partisan politician to corruption. For the purposes of fully enumerating an individual's choice set, we include in Table 1 some outcomes that we do not think are particularly likely.
How Different Types of Partisans Might React to Information about a Corrupt Politician
Source: Authors’ own compilation.
From these reactions, we highlight three likely outcomes of interest. Although it is not impossible, we think it is unlikely that revealed corruption would drive citizens to affiliate themselves with the corrupt party, and so we do not speculate further on that possibility.
Persistent Partisanship/Nonpartisanship
It may be the case that the partisanship of sympathetic voters in situation (a) is sufficiently strong that negative information does not alter their partisan identity. This is what we would expect in settings like the United States, where partisanship is a social identity. Persistent partisanship is even more likely in situation (b), where sympathizers of some other party hear negative information linked to a party with which they do not identify. Negative information about this other party is highly unlikely to prompt individuals to identify with the corrupt party, although it might drive voters to reassess their own party and perhaps thus declare partisanship with a third party. Relatedly, persistent nonpartisanship in situation (c) seems likely and results when a nonpartisan maintains a lack of party preference after being exposed to negative information about a given party.
Switch to Nonpartisanship
In either situation (a) or (b), information about corruption may drive individuals to embrace a new nonpartisan identity. This shift is most likely to occur among people who otherwise would consider themselves copartisans with the party of the corrupt politician (a) but who begin to feel a new distaste for that party. It is also possible that supporters of other parties may respond to information about corruption by minimizing their engagement with the political system as a whole, even if they would have previously identified with a party (b); a dislike of the party system as a whole drives these citizens into nonpartisanship.
Identification with other Parties
A link between corruption and one political party may inspire copartisan or nonpartisan individuals to identify with a new political party toward which they now feel relatively greater sympathy. 8 In situation (a) negative information decreases a voter's support for her or his previously most preferred political party to the extent that the voter identifies with a new party. In situation (c) negative information is enough to inspire a former nonpartisan to newly identify with a party not linked to corruption. 9
Both Turner (1999) and Lupu (2013) argue for the importance of comparative fit. Though they focus on ideology, valence issues like corruption might also contribute to comparative fit.
Gingerich (2009) finds increasing political engagement in the form of protest activity is a common response to exposure to corruption in Bolivia. Similarly, using 2008 AmericasBarometer data from Brazil, we find that personal experience with corruption is positively associated with a variety of measures of political engagement (results not reported).
In our empirical analysis, randomization and a between-subject experimental design allow us to confidently assess the causal effect of corruption priming on the overall distribution of partisanship among respondents. We cannot, however, observe pathways at the individual level, because we do not collect a pre-treatment measure of partisanship. Given the small number of questions in the survey, a pretreatment measure of partisanship might create a bias for subjects to persist in expressing their pretreatment partisan identity in a posttreatment question. Separately, it is possible for some pathways to be effectively “cancelled out” by other pathways. For instance, if nonpartisans respond to information about corruption in Party X by affiliating with Party Y, and if partisans of Party Y respond to that same information by becoming nonpartisans, our empirical approach will not identify any change in the aggregate partisan composition. Some pathways are more likely than others, however, and we use these to develop our three main hypotheses.
At a systemic level, these individual-level pathways map to a set of possible outcomes for the overall distribution of partisanship in the population at large. 10 In our empirical analysis below, we examine shifts in aggregate partisanship in response to a prime about corruption; therefore, we specify three alternative (but not mutually exclusive) hypotheses at the macrolevel:
H1: Decreased Copartisanship: Defining copartisanship as sympathy with the party associated with corruption, we expect that the proportion of people identifying with that party will decrease when they are exposed to information linking that party to corruption. The most robust result from various survey and field experiments on corruption is that politicians lose electoral support (or intended support) when corrupt behavior is revealed (Ferraz and Finan 2008; Chong et al. forthcoming; De Figueiredo, Hidalgo, and Kasahara 2013; Klăsnja and Tucker 2013). To the extent that declarations of partisanship and vote choice result from similar cognitive processes, we expect similar results.
H2: Increased Other Partisanship: After hearing about corruption associated with Party X, citizens may feel relatively greater sympathy with other parties, increasing the overall rate of affiliation with those parties. This movement could come from individuals who formerly sympathized with the party linked to corruption or from individuals who formerly identified themselves as nonpartisans and are sufficiently disenchanted by one party's links to corruption that they now feel greater affinity with a competing political party.
H3: Increased Party Alienation: After hearing that Party X has been linked to corruption, some individuals who would otherwise be copartisans might feel newly alienated from the implicated party but not sufficiently sympathetic with any other party to identify themselves as a partisan. Separately, other individuals who might have identified with a political party not linked to corruption may be turned off from partisan politics in general. Both individual-level processes reflect a form of political disengagement and should result in an increase in the share of the population that describes itself as nonpartisan. This hypothesis receives support from studies of vote choice in the wake of corruption. Both Chong et al. (forthcoming) (in Mexico) and de Figuereido, Hidalgo, and Kasahara (2013) (in Brazil) find higher levels of abstention – another form of political disengagement – among voters randomly exposed to information about actual corruption by a mayor and/or mayoral candidates.
Although exposure to information about corruption might lead to shifts in partisan attachment for anyone in the population, we argue that these effects are likely to be most acute among highly educated citizens. Drawing a link between an individual partisan politician who is described as corrupt and one's own identification with a party at large is a relatively complex cognitive task. Therefore, being primed to think about a corrupt partisan politician is most likely to affect the partisan sentiment of citizens who are attentive to the frame about corruption, are able to keep that framing in mind for a relatively long period, and can then make the association when asked to express their partisan identity. 11 Highly educated individuals are the most likely to maintain access to such associations and to connect particular information about corruption with their attitudes toward the political system writ large. As Druckman and Nelson (2003) observe in the literature on framing effects, experts are the respondents who most often “possess the knowledge and ability to connect the considerations suggested by the frame to their opinions” (Druckman and Nelson 2003: 731).
With regard to recall, in a separate survey from Brazil using a similar vignette, we find that rates of correct recall of information contained in the prompt are highly correlated with educational attainment. Among those with the lowest levels of education, only 50 percent of respondents could correctly answer a question about the prompt, whereas the rate rose to 74 percent among those with the highest level of education.
Our expectation is consistent with other results that examine partisan identity or political attitudes more broadly. Gerber, Huber, and Washington (2010: fn. 24), for instance, find that their field experimental treatment of informing people about the value of registering with a party leads to larger positive shifts in partisanship among the most educated respondents (although the difference from the overall effect is not statistically significant). Results in the literature on framing and political attitudes reveal stronger framing effects among more knowledgeable people after controlling for prior attitudes. 12 Chong and Druckman (2007) argue that this is because knowledge “increases the likelihood that the considerations emphasized in a frame will be available or comprehensible to the individual” (see also Zaller 1992; Druckman 2004). 13
While the more educated may be more responsive to frames, the literature on motivated reasoning (e.g., Taber and Lodge 2006) finds evidence that more politically sophisticated respondents defend their preexisting beliefs on contentious policy issues more strongly than others. The results that we present here might shed light on the types of issues for which new information leads to greater responsiveness among highly educated and sophisticated citizens.
Both Brader and Tucker (2008) and Lupu (2013) find that more sophisticated respondents are somewhat less likely to change their partisan attachment after being exposed to information about parties in a survey environment. In those studies, however, the authors provide respondents with publicly available information about party platforms and/or alliances. Knowledgeable voters are more likely to already be aware of macrolevel party behavior and therefore less likely to update their preferences in response to this information. In our study, we describe a hypothetical local politician about whom the respondents have no prior information. Our results here, which demonstrate more responsiveness among more educated respondents, are consistent with other work that shows greater responsiveness among more educated voters to new information on valence issues like corruption (Weitz-Shapiro and Winters 2014) and economic growth (Alt, Lassen, and Marshall 2014).
Our empirical framework parallels the real world delay between hearing some piece of information at a specific point in time and then thinking about one's partisan identity in the medium or longer term. In our survey we describe a hypothetical local politician as either corrupt or not. We give this politician a partisan identity, but we do not describe the overall performance of the political party in the local arena nor the performance of a party leader or national figure associated with the political party. We thus prime the respondent to think about corruption in conjunction with a particular party only through reference to an individual partisan politician. In the context of our survey, highly educated people should be more likely to recall the partisan affiliation of the politician in the prompt, to recall the association with corruption, and to have these factors in mind when reflecting on their partisan identity later in the survey. 14
An alternative hypothesis that might also predict more movement among more educated respondents would be that these respondents are simply more sensitive to social desirability bias and thus less likely to give responses that indicate support for corruption (a socially stigmatized practice). As we discuss in more detail below, a number of patterns in the data are inconsistent with this hypothesis.
Brazil is a particularly compelling site in which to examine the link between local political corruption and partisan attachment. Although strictly speaking a federal country since the 1891 Constitution, subnational governments gained substantial power in the years immediately following the country's most recent transition to democracy in 1985. 15 Municipal governments are responsible for the provision of a variety of basic services, including in core areas like health, and they spend an increasingly large portion of total government revenue (Falleti 2010). Holding municipal political office can be an important stepping stone in a political career.
The 1988 Constitution invested municipalities with substantial powers in a number of areas.
Although the country experienced a long dictatorship in the second half of the twentieth century (1965–1985), the Brazilian Congress remained open during that period and the military allowed controlled competition between one proregime and one antiregime party. After democratization, these parties lost influence and two new parties came to dominate national politics – the Partido dos Trabalhadores (PT), a leftist worker's party that first emerged in 1980 led by Luiz Inácio Lula da Silva, and the Partido do Social Democracia do Brasileira (PSDB), a centrist party that held the presidency from 1995 to 2002. The dominant narrative about Brazilian politics for much of the current democratic period has depicted parties as quite weak, both legislatively and in terms of their ability to engender individual loyalty among citizens (Weyland 1996; Ames 2002; although see Figueiredo and Limongi 2000 and Desposato 2006, among others, for evidence on the relative strength of parties in the legislative arena).
The recent literature on Brazil questions that portrayal. Baker et al. (2010) show that partisan identity is relatively stable over time, especially for the PT. Similarly, Samuels (2006) and Samuels and Zucco (2014) show the strength of PT partisanship as a social identity and demonstrate that PT (and to some extent PSDB) partisans react to party cues, shifting their opinions in order to conform with those of their party. This new wave of scholarship, however, still acknowledges that partisan attachment and identity in Brazil do not have the strength they enjoy in longstanding democracies. Partisan identification, for example, is far more limited in Brazil than in the United States according to recent surveys: responses to two-stage party identification questions suggest that less than 35 percent of citizens in Brazil identify with a political party compared to 60 percent in the United States. 16
These numbers are based on AmericasBarometer data (Brazil) and the ANES (US).
In 2010 we carried out a nationally representative in-person survey with 2,000 respondents in Brazil. The survey presented respondents with a vignette describing a hypothetical incumbent mayor running for reelection. 17 We employed random assignment to vary across vignettes the hypothetical mayor's party affiliation along with certain aspects of his past performance in office. 18 Some respondents learn that it is well known that the mayor takes bribes while others learn that it is well known that the mayor does not take bribes. 19 We analyze how the survey prime linking major political parties and corruption affects respondents’ declared partisan attachments.
In doing so, we follow a long tradition of survey experimental work that asks respondents to evaluate hypothetical politicians (e.g., Terkildsen 1993; Tomz 2007; Pepinsky, Liddle, and Mujani 2012). As Tomz and Weeks (2013) note, having respondents abstract from actual politicians allows for better testing of general hypotheses.
The vignettes varied information about politician performance simultaneously along two dimensions: one described the mayor's past involvement in corruption, whereas the other varied information provided about a mayor's provision of public works projects. The treatments were orthogonal to one another. We focus here on the information about corruption.
In providing their responses, it is possible that those surveyed react to either the positive or the negative information. In emphasizing the effects of negative information, we follow the convention in the corruption literature (Klăsnja and Tucker 2013; Muñoz, Anduiza, and Gallego 2013). In the appendix, we use data from another survey experiment to show that respondents tend to behave as if a politician is not corrupt unless they are exposed to explicit information about corruption.
The vignette reads as follows:
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 {or PSDB}. 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 experimental vignette was followed immediately by a question that asked the respondent the likelihood (on a four-point scale) that Gabriel (or Gabriela) would vote for the named mayor. 20 We then asked a series of additional questions about mayoral performance and eventually a question about partisan attachment. The time elapsed within the survey between the vignette and the question on partisan identification should attenuate any effect the corruption information has on respondents’ self-reported partisan identification and therefore bias against a significant finding.
The results of this analysis (discussed in Weitz-Shapiro and Winters 2013) are reproduced in the appendix. We use a third person vignette in order to attenuate social desirability bias among respondents exposed to a prompt that mentions corruption. In a separate survey conducted in the United States in 2012, we used the vignette from Brazil but randomly varied the third-person language with second-person language. We did not find any significant differences in response patterns between the two versions.
Partisanship
Respondents were shown a list of 27 parties and asked to select the party to which they felt closest. 21 This one-stage question results in a higher rate of partisanship than that elicited by a two-stage question. About 44 percent of respondents did not express a preference for any party; results from other surveys place the percentage of nonpartisans in the population at somewhere between 50 and 60 percent (Samuels 2006) or 65 and 70 percent (Ames et al. 2013; LAPOP n.d.). Of the total sample, about 32 percent of respondents support the PT; about 7 percent, the PMDB; about 7 percent, the PSDB; and about 16 percent, another party.
Brader and Tucker (2008) use three alternative measures of partisanship: two that are two-stage questions and one that is very similar to our question. They do not find any distinct patterns in treatment effects across alternative measures of partisanship.
Education
At the individual level, we expect the effect of the treatment to be strongest among highly educated respondents. We therefore subset out analyses to compare respondents with some tertiary education to those with completed secondary education or less. About 15 percent of our sample has some tertiary education or more.
To understand how linking corruption information to a political party affects declarations of partisanship, we examine differences in the distribution of expressed partisan identities across the corrupt and not-corrupt conditions in our experiment. 22 The randomized design allows us to attribute changes in this distribution to the corruption prime in the treatment vignettes; the other characteristics of respondents are orthogonal to this information. 23 As described above, we expect to see stronger effects among highly educated respondents and therefore subset our analyses accordingly.
In this discussion, we use the language “partisan attachment” interchangeably with “declared partisan attachment.” A survey allows us to observe the latter, although the former is the phenomenon of theoretical interest. We return to this issue in the discussion of our results.
In the appendix, we demonstrate balance on observable characteristics across the corrupt and not-corrupt treatment groups.
Table 2 presents the results for our three theoretical outcomes of interest: copartisanship, other partisanship, and nonpartisanship. For each panel of the table, we use a χ 2 test to examine whether the distribution of outcomes in the corrupt condition is statistically distinguishable from the distribution of outcomes in the not-corrupt condition.
Distribution of Partisan Declarations under Different Experimental Conditions and across Different Subgroups
Note: Panel A: All Respondents. N = 1,957; table reports proportion of respondents with number in parentheses. χ2 statistic for independence of rows and columns: 1.71 (p < 0.43).
Note: Panel B: Respondents with Completed Secondary Education or Less. N = 1,673; table reports proportion of respondents with number in parentheses. χ2 statistic for independence of rows and columns: 0.19 (p < 0.92).
Note: Panel C: Respondents with Some Tertiary Education or More. N = 284; table reports proportion of respondents with number in parentheses. χ2 statistic for independence of rows and columns: 7.13 (p < 0.03).
Source: Authors’ own calculation and compilation.
In the overall data and among the subset of less educated respondents, we see little evidence of any difference between the two conditions. Among the less educated respondents, in particular, the distribution of declared partisanship is nearly identical in the corrupt and not-corrupt conditions. 24 This suggests that the majority of our respondents do not update their partisan identification after being primed to think about corruption in conjunction with a particular party. Among the highly educated group, however, the corruption treatment has a number of interesting effects on declared partisanship.
For this group, the corrupt treatment has a statistically significant effect on the share of respondents who either fail to answer or say that they do not know if they are close to a party. The share of less educated respondents who decline to answer or say that they do not know goes from 1.5 percent in the clean condition to 3.3 percent in the corrupt condition (p < 0.02). As discussed in section 4.1 below, the corruption prompt depresses vote intention for both groups of respondents.
Panel C focuses on the most educated respondents in the sample. It shows a decrease in copartisanship (consistent with hypothesis 1) and an increase in other partisanship (consistent with hypothesis 2) among this group. Specifically, Table 2 reveals that the share of highly educated respondents sympathizing with the party named in the vignette drops from 25 percent in the not-corrupt condition to 20 percent in the corrupt condition, while the share of respondents declaring a partisan identity with a party other than the one named in the vignette rises by 15 percentage points (from 32 to 47 percent). In contrast to hypothesis 3 (which anticipates that corruption information increases party alienation) our results instead document increased party engagement among this group. When compared with the control condition, in the corruption condition, nonpartisanship drops from 43 percent to 33 percent. Highly educated respondents who might otherwise be copartisans or nonpartisans in the clean condition appear to instead say they identify with parties not named in the vignette in the corrupt condition.
Figure 1 presents our results for the most educated respondents only. It illustrates the substantial increase in other partisanship and the marked decrease in nonpartisanship that result from exposure to the corruption prime. The comparatively limited change in the number of copartisans suggests that the bulk of the increased rate of other partisanship is a result of decreased nonpartisanship among educated respondents in the corruption conditions. Our results are consistent with a world in which few highly educated partisans are swayed by negative information about their preferred political party and where negative information about one party leads many highly educated nonpartisans to feel greater sympathy with one of the other parties in the system. 25 As a result, the association of a particular political party with corruption seems to drive new partisan identification and to increase engagement with the partisan political system among highly educated respondents in Brazil. 26
Lupu (2013) finds that partisanship increases when voters have an increased capacity to differentiate between political parties. To the extent that our respondents are prompted to think about corruption in conjunction with one party, that might assist them in differentiating that party from other parties.
A scenario in which nonpartisans react to the corruption prime by declaring a partisan identity is more plausible than the alternative possibility in which respondents who would otherwise declare themselves partisans of different parties are exposed to the not-corrupt prime and then opt to declare themselves nonpartisans.
Partisan Shifts: Highly Educated Respondents
In the appendix, we show the robustness of these results to several additional tests of the data. First, we run a multinomial probit model in which we use a corruption treatment indicator and an interaction between that indicator and a variable indicating whether or not the respondent belongs to the high education group. This allows us to predict whether or not a respondent expressed copartisanship with the politician in the vignette, other partisanship, or nonpartisanship (the omitted category). Using the parametric model allows us to include basic sociodemographic controls. The results show the same overall pattern as presented above: among highly educated respondents, the corruption information treatment increases the share of other-party identifiers relative to non-partisans. The multinomial probit model allows us to run an explicit test on the difference in the treatment effect between the two groups – for which we find a statistically significant difference (p < 0.03). Second, we run the multinomial probit model only among the respondents who heard about a PT mayor and then again only among the respondents who heard about a PSDB mayor. We find that the differences in treatment effects persist across the two types of vignettes, although the difference is significant only for the PSDB vignettes. Third, we present the cross-tabulated results using a more exhaustive set of categories that underlie our coding of the outcome variable, disaggregating the larger parties that compose the “other party” category. The patterns for our high education respondents are again consistent with the patterns shown here: the respondents in this group who hear about a corrupt politician from either of the two parties used in the treatment are less likely to describe themselves as nonpartisan and more likely to express support for a third party. Regarding the PT vignettes, apart from leading to gains in identification with other parties, the corruption treatment leads the PT to lose support from highly educated respondents. These gains for other parties outweigh the PT losses, which explains the net positive effect on partisan engagement. Among the PSDB vignettes, corruption information does not change identification with this party among the highly educated, but other parties still gain identifiers. Among our low-education respondents, when we disaggregate results by party, the results are somewhat more ambiguous than those presented in Table 2. In the PT treatment, among less educated respondents, we see the same pattern as in the overall data: information about corruption does not shift reported party identification. Within the PSDB vignettes, corruption information does shift the distribution of partisanship, as our respondents move away from both the PT and the PSDB and toward the third largest party in the country, the PMDB. In terms of rates of partisan identification, however, these shifts mostly cancel out – that is, in contrast to the most educated respondents, the corruption prompt does not appear to increase overall partisan engagement among the least educated. We discuss these patterns in greater detail in the appendix.
As argued above, we believe that the existence of a treatment effect conditional on having at least some tertiary education reflects the cognitive demands of linking a prompt about a corrupt partisan politician to one's broader views about political parties. Here, we address two alternative explanations for why the treatment effect might be limited to this group. First, it might be the case that highly educated citizens simply are more likely to condemn corruption than is the population at large. Second, highly educated citizens may be more susceptible to social desirability bias and therefore more likely to give the “right” answer to survey questions on sensitive topics like corruption, regardless of their true opinions. We find a number of patterns in the data that are inconsistent with these two alternatives.
Beginning with the former, if it is the case that our results are driven by different attitudes toward corruption across education levels, we should be able to observe those differences in responses to other questions in the survey. The most obvious candidate is a question that asked respondents how likely they were to vote for the mayor described in the vignette. As this question was asked immediately after the vignette, directly about the mayor described therein, and without explicit reference to the mayor's partisanship, it placed fewer cognitive demands on respondents than the question about partisanship. Table 3 reports the results for this question for the whole sample and then divides them according to education. The results do not provide any evidence that more educated respondents have more negative attitudes toward corruption. On a four-point scale of how likely respondents are to reelect the mayor (where 4 is very likely), noncorrupt mayors score an average of 3.23 among respondents with completed secondary education or less, while corrupt mayors only manage 1.68. Among highly educated respondents, the average likelihood of voting is 3.19 for non-corrupt mayors and 1.79 for corrupt mayors. Thus, the size of the punishment for corruption is slightly higher among the less educated (1.55 versus 1.40), which is the opposite of what an explanation focused on different preferences implies.
Survey Responses on Vote Intention
Survey Responses on Vote Intention
Source: Authors’ own calculation and compilation.
This strongly suggests that the differences in partisan change we observe in our survey are not the result of different attitudes toward political corruption. It also speaks to the importance of thinking about changing partisan affiliation as a response to political corruption that is analytically distinct from electoral punishment of the corrupt actor.
The second alternative explanation is that, in the context of a survey, highly educated people have a stronger desire to communicate their disapproval of corruption than less educated respondents regardless of any underlying difference in views between these two groups – that is, the highly educated might be particularly susceptible to social desirability bias on the sensitive issue of corruption. The results reviewed in the previous paragraph, however, speak against this possibility. If social desirability bias were to explain the different patterns of partisan change across education groups, we would expect the highly educated to be particularly likely to claim that they would not vote for a corrupt politician. As the results displayed in Table 3 illustrate, this is not the case.
In addition, the responses to the question on partisanship itself do not offer support for this alternative explanation. If the responses of the highly educated mostly reflected a desire to give socially conforming answers in the context of a survey, we would expect these respondents to be particularly reluctant to reveal sympathy with a party of a mayor identified as corrupt. However, as described above, the corruption condition has little effect on the share of declared copartisans among the highly educated; instead, it leads to a significant decrease in nonpartisanship and an increase in other partisanship. Neither result is consistent with response patterns motivated by a desire to appeal to an interviewer.
Although we can dismiss these alternative explanations, the question remains as to the extent to which changes in declared partisan attachment in a survey are likely to reflect changes in genuine party attachment in response to real world information. As always, we should exercise caution in extrapolating from the relatively controlled context of a survey experiment to the views and behavior of citizens in the real world (Barabas and Jerrit 2010). Given the limited information surveys provide on electoral contests and the relatively close proximity between receiving information and being asked about partisan leanings, our survey experiment might best replicate cases in which information about political corruption gains the public's attention and receives ongoing attention in the press (e.g., Chang, Golden, and Hill 2010). 27 In Brazil, the use of federal audits aimed at uncovering corruption and the practice of publicizing the names of candidates accused of corruption mean that these high information settings are increasingly common.
In a field experiment in Mexico, Chong et al. (forthcoming) expose treated individuals to information about corruption in a public information flier and then find a follow-up effect on partisan identity a few weeks later. This suggests that in some cases, even a single exposure to a real-world treatment analogous to the one included in our survey might have a demonstrable effect on partisanship.
The external validity of our survey is also buttressed by the fact that our results are consistent with other previously identified characteristics of partisanship and partisan change in Brazil. Baker et al. (2010) use a six-wave panel study in two medium-sized Brazilian cities to document changes in partisanship over time. Their findings emphasize the fluidity of the “nonpartisan” category: across the six waves of their study, the level of nonpartisanship ranges between 50 and 62 percent in one of the cities and 46 and 64 percent in the other. Comparing between two waves of the survey, they show that some 20 percent of declared nonpartisans in one wave of the survey profess a party affiliation in the other wave – though they document far less movement between parties. Based on a nationally representative sample, our results provide further evidence of the importance of shifts in and out of the nonpartisan category (what Baker et al. 2010 call “bounded partisanship”) and the limited shifts between partisan groupings. In addition, using a survey experiment allows us to identify a probable cause of such shifts – that is, the priming of respondents to associate a particular political party with corruption. 28
Our work also builds on Samuels’ (2006) finding that attitudes toward a party's most prominent national leader are correlated with partisanship in Brazil. Our results suggest that the negative portrayal of subnational politicians can also affect partisan attachments.
The notion that partisanship is an ingrained social identity that is relatively unresponsive to short-term political events is well established in the literature on voters in the United States. Recent work presents mixed evidence on the applicability of this view to other countries. Our paper advances our understanding of partisanship in younger democracies by providing experimental evidence from Brazil that priming subjects with short-term political information can cause changes in citizens’ declared partisan identity. In particular, a prompt about corruption tied to a politician from a particular political party can decrease the likelihood that respondents will describe themselves as nonpartisan and increase the likelihood that they will declare themselves sympathizers of other parties. This effect is found only among the highly educated – those people most likely to recall and be able to access the association in the prompt when considering their own partisan sympathies at some later time. Importantly, this does not reflect different preferences over corruption across respondents with different levels of education.
Our finding that some citizens’ declared partisan identities are responsive to short-term political performance contributes to an emerging consensus that the declaration of partisan identity is malleable in less established democracies (Brader and Tucker 2008; Baker et al. 2010; Chong et al. forthcoming; Klăsnja and Tucker 2013; Lupu 2013; Winters and Weitz-Shapiro 2014). 29 Nonetheless, these studies do exhibit variation in the details of who changes their partisanship and in what way. In the survey examined here, we find that changes are concentrated among the highly educated, whereas some other studies find that movement in partisan identification is widespread or even that changes in expressed partisan identification are concentrated among the less educated. Whereas we find that negative performance information brings nonpartisans into new engagement with the party system, other studies find that negative performance information causes people to shift into nonpartisanship. Thus, while the variety of results emerging in the literature demonstrate the malleability of partisanship in younger democracies, they also give us reason to continue to work to understand precisely among which groups and under what conditions partisan sentiment will shift in these democracies.
Baker et al. (2010) explicitly compare their results to those found in Britain and Germany in Zuckerman, Dasović, and Fitzgerald (2007) and note that partisan identities are less stable in Brazil.
What are the implications of our results for our understanding of political accountability in Brazil today? There are two broad schools of thought on the relationship between strong party identity and democratic accountability. 30 On the one hand, there are those scholars who have traditionally portrayed strong, responsible parties as a democratic asset that can organize and channel societal demands. In this light, strong partisan identity is valuable because it both originates in and contributes to strong parties. 31 Academics studying relatively young democracies in Latin America (Mainwaring and Scully 1995; Mainwaring 1998; McCann and Lawson 2003) and the post-Soviet space (Klingemann and Wattenberg 1992; Smyth 2006) have lamented weak partisanship as a deterrent to democratic consolidation and accountability. As McCann and Lawson (2003: 60) argue, “when citizens lack firm, enduring dispositions […] it is difficult for them to hold leaders accountable.”
See Baker et al. (2010) for a similar description of the existing literature and the tensions therein in a comparative context. See also Tilley and Hobolt (2011).
In the US context, this vision was articulated by the 1950 APSA Committee on Political Parties, and it has since appeared elsewhere in the literature (e.g., Burnham 1970: ch. 5; Broder 1971).
On the other hand, there are those scholars that have depicted strong party identification as blind loyalty, which distorts citizens’ ability to objectively evaluate relevant political information. To the extent that strong partisans “have already decided who is right and who is wrong” (Mutz 2006: 128), strong partisanship could actually decrease accountability. In the United States, researchers have shown that partisanship colors voter perceptions of politician and government performance levels across a range of areas (Wilcox and Wlezien 1996; Bartels 2002; Johnston, Hagen, and Hall 2004; Lavine, Johnston, and Steenbergen 2012) and have called attention to the risks of partisan-motivated reasoning (Bolsen, Druckman, and Cook 2012). The cross-national literature that highlights the differences between partisan “winners” and “losers” also offers evidence of the more forgiving nature of copartisans outside of the United States. 32
For example, Anderson and Tverdova (2003) find that being in a corrupt country has less of a negative effect on government evaluations for “winners” as opposed to “losers.” See also Anduiza, Gallego, and Muñoz (2013) and Klăsnja and Tucker (2013).
Where does Brazil fall within this spectrum? This study and several other recent studies (Baker et al. 2010; Winters and Weitz-Shapiro 2014) provide evidence that short-term performance information can change partisan attachment for at least some groups of Brazilian citizens. At the same time, it is clear that for some Brazilians, partisan attachments help them organize and order their understanding of politics (Samuels and Zucco 2014). Together, these results suggest that mass partisanship in Brazil today may occupy a middle ground: parties are strong enough to serve as meaningful cues to citizens, but parties can also expect their fortunes to rise and fall with their accomplishments. The results of the October 2014 presidential election perhaps reflect this dynamic, as Brazil's incumbent PT president Dilma Rousseff was just barely reelected in the face of growing corruption scandals and a slowdown in economic growth.
Finally, our results are particularly encouraging with respect to the spate of new information-provision reforms enacted in Brazil in recent years. Federal audits of municipal accounts, the publication of the names of political candidates accused of wrongdoing in the judiciary, and a new right to information law means that credible information about political performance is more easily available now than ever before. Our findings suggest that the revelation of malfeasance through such venues may serve the double purpose of facilitating punishment of poor-performing politicians while, at least for some citizens, increasing engagement in Brazil's political system.
Footnotes
Portuguese Version of the Vignette
Bolded brackets indicate the randomized components of the vignette.