The Varying Political Toll of Concerns About Corruption in Good Versus Bad Economic Times

ABSTRACT: Under what conditions do citizens connect concerns about corruption to their evaluations of sitting executives? In contrast to conventional scholarship positing a direct, negative relationship between corruption and political support, we build on a small but suggestive body of research to argue that this relationship is conditional on economic context. We test this claim with national survey data collected in 19 presidential systems as part of the AmericasBarometer 2010 study. Using both fixed effects ordinary least squares and hierarchical linear regression analyses, we show that individuals facing bad (good) collective economic conditions apply a higher (lower) penalty to presidential approval for perceived political corruption. This result holds across both an individual-level indicator of national economic assessment and a regional economic measure; we further test, and find less substantial results for, the moderating influence of personal economic conditions on the political toll of corruption perceptions.

CITATION: Zechmeister, E. J., & Zizumbo-Colunga, D. (2013). The Varying Political Toll of Concerns About Corruption in Good Versus Bad Economic Times. Comparative Political Studies, 46(10), 1190-1218. https://doi.org/10.1177/0010414012472468 

Note: What follows is the authors' pre-print version. The final published version can be accessed [HERE]

Political corruption can undermine confidence in various dimensions of governance and contribute to underdevelopment (Mauro 1995). Classic theories posit that voters hold politicians accountable for performance (e.g., Downs 1957; Fiorina 1981). When perceptions of corruption run high, norms of fairness and beliefs concerning corruption’s potential to yield deleterious outcomes should move individuals to express disapproval with the incumbent administration. And, in fact, research focused on both developing and developed countries finds evidence linking corruption to indicators of political support and trust (e.g., Anderson and Tverdova 2003; Bailey and Paras 2006; Canache and Allison 2005; Fackler and Lin 1995; Morris and Klesner 2010; Seligson 2002). At the same time, citizens do not always turn against incumbents when corruption is made salient (Chang, Golden, and Hill 2010; de Figueiredo, Hidalgo, and Kasahara 2012; Krause and Méndez 2009; Rundquist, Strom, and Peters 1977). This raises an important question: Under what conditions do citizens connect concerns about corruption to their evaluations of sitting executives?

Our answer is a simple, but under-examined, proposition: individuals have a propensity to trade-off political corruption for economic well-being. When economic conditions are good, we argue that citizens will tend not to withdraw support from the executive when they see a prevalence of corruption. Conversely, individuals will be relatively more punitive in their evaluations of the executive when politicians are perceived as being corrupt and incapable of delivering good economic times. In short, drawing on and extending extant scholarship in ways we will detail below, we propose that the way in which individuals apply their perceptions of corruption to presidential approval is conditioned by their perception, and the actual nature, of economic times. 

We assess this proposition using responses from national surveys of the 19 presidential countries included in the AmericasBarometer 2010 study, from which we also derive economic measures for 107 sub-national regions.[1] Our dependent variable, presidential approval, is widely considered an important facet of democratic politics, with higher approval ratings translating into political power and influence (see Gronke and Newman 2003). It is well-established that economic factors, both objective conditions and subjective perceptions, matter for presidential approval (see Nannestad and Paldam 1994; Gronke and Newman 2003). In Latin American presidential systems, the relative strength of the executive office makes the study of presidential approval even more important, as this office is frequently a focus point of politics. In fact, public grievances over government performance have led to occasional pre-term departures of executives in this region (Valenzuela 2004; see also Pérez-Liñan 2007). Thus, in this project we consider the interaction between corruption assessments and economic performance as these relate to presidential approval. 

The theoretical perspective we offer here extends a limited set of scholarship that suggests a conditional relationship between economic conditions and corruption with respect to political support (Manzetti and Wilson 2007; see also Manzetti and Rosas 2012; Rundquist, Strom, and Peters 1977; Stokes 2001). Yet, the notion that the public trades off clean government for good performance is not uncontested (e.g., Winters and Weitz-Shapiro 2012). A key contribution we make, then, is to offer a test of this debate, with a particular focus on presidential approval. In doing so, we build on but also go beyond extant research in a number of ways. First, unlike prior studies, we assess the moderating effects of general economic conditions versus the moderating effects of individual economic situations; we demonstrate that, in moderating the link between corruption and political support, one’s general economic context matters more than do evaluations related to one’s pocketbook. Second, we demonstrate the robustness of the argument across both perceptions of the economic situation and a more objective indicator of the sub-national economic context in which the individual resides. Third, we assess and find support for the argument in a broad comparative context consisting of all presidential systems in the Americas. Our findings provide strong evidence that, with respect to their evaluations of the executive, individuals across the Americas tend to tolerate corruption when times are good; and, on the other hand, when times are bad, perceptions of corruption exact a hefty toll on approval of the sitting president. 

The Conventional View on the Performance and Politics

Conventional wisdom posits a direct connection between, respectively, economic and corruption performance on the one hand and political support on the other hand. That is, while a handful of scholars have proposed that citizens tend not to put much effort into monitoring and reacting to government performance, either because they select “good types” into power (Fearon 1999) or otherwise delegate the process of governing to elected officials (O’Donnell 1994), most scholarship suggests that citizens react to government output and update their evaluations of sitting politicians in accord with the nature of that output. In the realm of voting, the process of ensuring quality representation via retrospective monitoring has been termed “accountability representation” by Przeworski, Stokes, and Manin (1999), and stems from classic work by Downs (1957), Fiorina (1981), and others. According to this perspective, citizens maximize the quality of political representation if they hold elected officials accountable when politicians violate their mandate or otherwise deliver output that is not in accord with citizens’ best interests.[2]

Presidential job approval is one of the most important indicators of the support citizens have for the incumbent executive. While the opportunity to vote for the president occurs only every 4-6 years in most systems, public opinion polls constantly measure and report on presidential approval. Presidents and their teams of advisors monitor these polls closely because approval carries political purchase for executives who seek to exert influence over congress, their party, and others (Edwards 1997; Gronke and Newman 2003). One of the most influential determinants of presidential approval is, of course, the economy. At least since Downs (1957), scholars focused on both the developed and developing world have proposed that citizens look at the performance of the economy to form their opinions of government and ultimately to determine whether or not to support the incumbent (Kinder and Kiewiet 1981; Lewis-Beck and Stegmaier 2008; Przeworski 1999). While arguably still an open question, much research asserts that perceptions of the national economy have stronger political effects than pocketbook evaluations (Kinder and Kiewiet 1981; Lewis-Beck 1988; Nannestad and Paldam 1994; see also Singer and Carlin 2013).[3]

At the same time, economic factors may not be the only performance indicators on citizens’ minds when they evaluate elected officials (Fiorina 1981). From a number of available factors one might introduce, we elect to focus on corruption.[4] Corruption, which can be defined as the “misuse of public power for private gain” (Rose-Ackerman 1999, p. 91), is a frequently salient issue in politics around the globe and the Americas are no exception (e.g., Weyland 1998). It is in keeping with classic perspectives on accountability and retrospective voting to posit that citizens will withdraw support from the government for a visible dearth of “integrity” (see Fearon 1999; Seligson 2002). A fairly broad consensus exists that, if individuals in democratic settings are presented with information about corruption, they will downgrade their evaluations of, or preferences over, incumbent politicians (e.g., Adserà, Boix, and Payne 2003; Winters and Shapiro 2012). In fact, in a time series analysis of the United States, Fackler and Lin (1995) find a negative correlation between corruption and presidential vote-share. More generally, corruption has been shown to have a negative relationship with political trust and support (Della Porta 2000). Similar effects have been, in turn, documented in numerous studies conducted across different contexts. 

For example, in their study of sixteen established democracies, Anderson and Tverdova (2003) show that citizens of countries with a higher level of corruption express lower evaluations of their political system and authorities. In the African context, other authors have found corruption to be a major obstacle to a process by which governments build popular trust for their institutions and the electoral process (Armah-Attoh, Gyimah-Boadi, and Chikwanha 2007). In their stance against Asian exceptionalism, Chang and Chu (2006) show political corruption is corrosive to political trust in six Asian democracies. Further, and notable because it runs counter to our argument, they argue for an invariant effect across contexts. In the case of Latin America, and in line with Seligson’s research (2002), Canache and Allison (2005) and Morris and Klesner (2010) are among a number of scholars who have found a negative effect of perceptions of corruption on support for incumbent officials and political institutions. Outside of the realm of survey data, two studies in Brazil add further support to the contention that if corruption is salient, the public will turn against tainted incumbents. First, Ferraz and Finan (2008) show that pre-election audit reports that indicated corruption affected support for mayors, in particular where the media was capable of a broad dissemination of the audit results. Second, Winters and Weitz-Shapiro (2012) report on an experiment in which information about bribe-taking by the mayor decreased political support.[5] 

While a fair amount of evidence indicates that corruption can take a political toll on incumbent politicians and institutions, the following presents itself as an important puzzle: why do corrupt governments sometimes survive and only sometimes expire? One possibility is that the public (or at least important parts of it) is not actually as sensitive to information about corruption as one might expect. In an innovative study, de Figueiredo, Hidalgo, and Kasahara (2012) find that making corruption accusations salient in a mayoral election in Brazil had no discernible effect on the incumbent’s electoral popularity; they find, instead, that the challenger was affected and argue that the public’s sensitivity to corruption can vary across party lines. In addition, Anderson and Tverdova (2003) find that the negative effect of corruption on evaluations of the political system is significantly attenuated among supporters of incumbent political authorities. We concur that sensitivity to corruption accusations may vary by partisanship, among other related factors; however, we nonetheless seek to determine when, in general, the public connects concerns about corruption to presidential approval. Another possibility alluded to above is that information matters. We also concur that information is important, but by focusing on perceptions of corruption we remove from consideration the process by which individuals become aware of political corruption. Thus, we ask: when individuals come to believe that politicians are corrupt, under what circumstances will they downgrade their evaluation of the executive?

Proposing an Interaction between Economic Output and Perceptions of Corruption

Our answer to the aforementioned question is that, when it comes to presidential approval, there exists a general tendency in the public to trade-off perceived political honesty for economic good times and, conversely, to exhibit a strong intolerance of corruption in economic down times. This proposition has been suggested by previous scholars, but subjected to little empirical testing; moreover, the proposition has been critiqued, as we will note below. But, first, we offer our theoretical perspective on why individuals will tend to connect concerns about corruption to presidential approval in bad, but not in good, times. In so doing, we highlight the question of what factors exert a more powerful conditioning effect: household or community level conditions?

Past research on the subject provides reason to expect connections among corruption, economic performance, and political support. One perspective that has been offered focuses on economic benefits at the individual, or household, level. For example, Rundquist, Storm and Peters (1977; see also Peters and Welch 1980) argue that an individual may be willing to trade her vote to improve her material well-being via a direct exchange between the candidate (or his associates) and the voter. Some support for this notion is found in an analysis of 1995 data from Argentina by Manzetti and Wilson (2006), whose focus on personal economic factors extends beyond the simple machine politics referenced by Rundquist and colleagues. They find a null effect of corruption on confidence in government when Argentineans are more satisfied with their household finances.

Another possibility is that general economic conditions exert a more important conditioning effect on the relationship between perceptions of corruption and presidential approval. Returning to Rundquist et al. (1977, p. 956), the authors offer evidence that, on average, individuals whose policy stances align with a candidate are less likely to be dissuaded by issues of corruption from expressing support for that candidate. If we consider the performance of the economy to be an issue, then when output is favorable we have theoretical reason to believe good economic performance could similarly dissuade individuals from linking corruption to support. Such a focus on general, or sociotropic, economic conditions is supported by a number of economic voting studies, which tend to find that the direct effect of these is more influential in comparison to pocketbook evaluations (Kinder and Kiewiet 1981; Lewis-Beck 1988; Nannestad and Paldam 1994).[6] Further, the tendency for citizens to couple corruption and the economy in bad economic times has previously been suggested, but not directly assessed, in research by other scholars (e.g., Stokes 2001).

Anecdotally, examples of incumbents and incumbent parties surviving despite high levels of corruption are not scarce and, further, economic output appears to play a role in these examples. However, it is difficult to discern from anecdotes whether individuals are reacting to household or macro-economic conditions, let alone whether these reflect general tendencies. As one example, we can reflect on the case of Mexican President Carlos Salinas and the PRI. Despite general accusations of corruption against his party and a very specific fraud scandal surrounding his 1988 election to the presidency, Salinas embarked on economic reforms so good that he seemed justified when he declared that Mexico was entering the “first world”. With no evidence that corrupt practices by the then-dominant PRI had changed, Salinas managed to achieve robust approval ratings and the PRI leveraged enough legitimacy to win the presidential seat for the subsequent term. Although good economic performance can give new life to corrupt governments, bad economic performance often sounds their death knell. The case of pre-Chávez Venezuelan politics provides an example. On the eve of Chávez’s ascent to power in Venezuela, corruption had become one the most salient issues facing President Carlos Andrés Pérez’s government. Simultaneously, the country faced an economic crisis. Stokes (2001, p. 135) suggests these bad economic times made corruption particularly politically relevant: “[H]ad political management and economic performance been better, accusations of corruption against Pérez would not have stuck.”

In keeping with the above examples, our argument speaks not only to how good economic times might dissuade individuals from punishing the executive for corruption, but also to the fact that bad economic times might induce individuals to make a particularly strong connection between the two. Returning to the case of Argentina studied by Manzetti and Wilson (2006), this framework implies that economic conditions in general, as opposed to just individual financial situations, conditioned public reaction to corruption, both during Menem’s “golden years” and with respect to his political fall. Despite evidence of widespread corruption in his government, Menem was reelected in the 1995 election after having overseen a period of strong economic performance. However, in 1999, when the country faced severe economic downturns, corruption scandals came to haunt Menem and his Peronist Party. Menem was unable to marshal through the congress legislation allowing a third presidential run and his party’s candidate was severely defeated by an opposition candidate who ran on an anti-corruption platform. While Manzetti and Wilson (2006) convincingly argue that personal economic evaluations matter to explain political support under Menem, it is also quite possible that macro-economic conditions mattered, and potentially more. 

Therefore, in addition to testing whether individuals evaluate the president in a way that trades off honest government for good economic output, we also test whether general economic conditions or personal economic evaluations exert a greater conditioning effect on the relationship between corruption and presidential approval in the Americas. Neither of these possibilities is an uncontroversial proposition. Rather, there exist reasonable counter-arguments and competing hypotheses to the general framework we have asserted here. First, it is possible that under conditions of high corruption, citizens concentrate their energy on applying this directly to presidential approval, paying less attention and giving less weight to other, in particular, economic considerations. This alternative scenario is suggested in discussion by Singer (2011, p. 284), who writes that “Governance crises related to corruption and human rights reduce attention to the economy.” Singer’s focus is on issue salience, but his discussion suggests a plausible alternative to the theoretical perspective that we assert here.[7] That is, it is possible that, when corruption is salient, other factors such as the economy simply become less relevant. It is also possible that there is simply no conditioning relationship between corruption and performance. Winters and Shapiro (2012), for example, find that subjects responding to a survey experiment did not appear to trade off clean government for performance. However, their conception of performance focused on mere output (completed public works projects) while, in contrast, we focus on quality (better or worse economic outcomes). In short, the question of the general relationships among economic conditions, corruption, and political support is an open and debated one, and one that we intend to shed light on with the empirical analyses that follow.

Data 

The data used for our analyses are from national surveys collected as part of the 2010 AmericasBarometer, a regular regional study carried out by the Latin America Public Opinion Project (LAPOP). We use data from all 107 sub-national regions across the 19 presidential countries included in the project. With the exception of the United States, where a web-based survey was administered by Polimetrix, the data are from face-to-face household interviews selected using a complex sample design.[9] Because we are testing a general theory, we include the U.S. case in our analyses; however, our results are robust to its exclusion. 

Our key dependent variable captures evaluations of the president’s job performance.[10] We reversed and rescaled the original five-point measure to run from 0 to 100, where 100 is maximum approval. Mean presidential approval in the 2010 AmericasBarometer dataset is 58.43, with significant variation across countries. Thus, for example, Uruguay, Brazil and Colombia had the highest mean levels of approval while Argentina, Peru and the United States had the lowest mean levels. 

Our principal objective is to test the heterogeneous nature of the relationship between individual level perceptions of corruption and presidential approval. Consistent with Manzetti and Wilson (2006), we measure concerns about Corruption with a variable that gauges the degree to which respondents think corruption is widespread.[11] We rescaled the original 1-4 measure to run from 0 to 1, where 1 indicates the highest perceptions of corruption. The average perception of corruption across presidential systems in 2010 is 0.73, which indicates that on average individuals in the Americas think corruption is fairly common. 

We include three measures of individual level economic assessments. The first two capture, respectively, current national and current personal evaluations. Both were initially asked with a 5-point response option, which we have rescaled to run from 0 to 1 where 1 indicates a very good evaluation. Perceptions of the National Economy and of the Personal Economy across the sample are on average moderately low to neutral (.43 units and .51 units, respectively).[12] The third individual level economic assessment is Household Income Deterioration.[13] This is a dummy variable that captures whether individuals reported their household income to have decreased over the two years prior to the survey (coded as 1) as opposed to having increased or stayed the same (coded as 0). In 2010, 27.8% of citizens in the presidential countries of the Americas reported an income decrease. We include this third individual economic measure across all models because it becomes an important individual level control when we move to our multi-level model; we note that the effects of the other two individual level economic assessment variables are only marginally different (such that our statistical and substantive conclusions are unaffected) when this variable is excluded.

Having being a political supporter of the president in the past is an obvious determinant of presidential approval in the present (whether this is due to shared ideology, party identification or other factors). To control for this influence, we use respondents’ self-report of whether they Voted for the Incumbent in the last election, which accounts for electoral supporters of the current executive (Blais 2007).[14] In addition, we include controls for respondents’ Age, Education, Income, Gender (Female), whether they reside in an Urban (versus rural) area, and region dummy variables to account for differences across sub-national regions.[15]

In addition to individual level data, we assess the effect of general economic conditions with a region-based indicator that takes advantage of intra-national variation and provides a measure of the economic context in which individuals are embedded (see Linz and de Miguel 1966; Snyder 2001; Wood 2009). To identify regions, we use the sub-national strata variable in the LAPOP AmericasBarometer dataset; the variable indicates geographic units within the country that are based on political or other relevant boundaries and the survey data collected within these regions, by the sample design, are representative of those strata. In all, there are 107 regions across the 19 countries. In order to estimate regional economic conditions, we calculated a regional measure based on reported household income deterioration. This indicator, Regional Income Deterioration, is the proportion of people within the subnational region who reported that their household income decreased over the two prior years.[16]

A bivariate analysis of the relationship between the Regional Income Deterioration measure and average regional income (measured by the regional average of self-reported income in the survey data) shows that regions with lower average incomes have a higher proportion of people reporting that their household income decreased over the last two years. Thus, to isolate the effect of the contextual variable we are most interested in, we include Average Regional Income in the analysis.[17]   

Analyses and Results

To refresh, our expectation is that the relationship between perceptions of corruption and presidential approval varies according to economic conditions. To assess this, first, we employ an individual level design based analysis in which we model presidential approval as a function of perceptions of corruption and individual national and personal economic evaluations. This approach, assessed using fixed effects OLS regression,[18] controls for any between region differences among the participants and adjusts standard errors to account for the stratification and clustering inherent in the AmericasBarometer survey data (see endnote 9); we present both a simple model and, to test our expectations, one that contains interaction terms between corruption and economic evaluations (we focus first on national conditions, and then we turn to personal evaluations). Second, we take advantage of the subnational representativeness of the AmericasBarometer (at the level of the 107 regions) and assess a hierarchical model in which the effect of corruption on presidential approval is proposed to be a function of the proportion of households in a region that report that their income has decreased in the last two years. Across both models, as we will show, we find strong support for the notion that economic context moderates the relationship between individual level assessments of corruption and presidential approval.

Individual Level Models

We first assess the direct relationship between, on the one hand, perceptions of corruption and of the national economy and, on the other hand, presidential approval with a basic model for the pooled data of the 19 presidential countries in the AmericasBarometer; the model predicts presidential approval with measures of the individual’s perception of Corruption, reported Household Income Deterioration, perception of the National Economy, and perception of the respondent’s Personal Economy. The model includes a control for whether the individual Voted for the Incumbent, standard controls for socio-economic and demographic factors, and regional dummy variables in order to produce region fixed effects that account for any differences across the 107 regions.[19] All independent variables are scaled to run from 0 to 1 and the dependent variable, presidential approval, is kept on a 0 to 100 scale for ease of interpretation.

Column 1 in Table 1 shows the OLS regression estimates for a region fixed effects model in which no interactions are included. Perceptions of corruption, of the national economy, and of one’s personal economy are all significant predictors of presidential approval. As the table shows, holding other variables constant, a maximum increase in perceptions of corruption is associated with a decrease of 5.31 units in presidential approval; a maximum increase in perceptions of the national economy is associated with a 26.02 unit increase in presidential approval; a maximum increase in assessments of one’s personal economic situation is associated with a 6.79 unit increase; and, finally, respondents who reported that their household income decreased in the last two years approve of the president 2.51 units less than those whose household income did not decrease. These basic findings are in accord with the perspective associated with classic “accountability representation” models, in which citizens punish sitting politicians for poor output and where economic factors (and, in particular, sociotropic indicators) exert strong effects. In addition, those who voted for the president approve more of the president (12.06 units more) compared to those who did not support the executive in the last election. Finally, wealthier, more educated and younger people and those how live in an urban (as opposed to rural) setting are less approving. 

But, do economic perceptions moderate the relationship between corruption and presidential approval, as we anticipated with our earlier discussion? As a first test of this argument, we add to the previous model an interaction term between perceptions of corruption and of the national economy. Column 2 of Table 1 shows that interaction is significant, which supports the argument that the economy conditions the toll of corruption. 

To see this relationship more clearly, Figure 1 presents the marginal effect of perceptions of corruption at different levels of perceptions of the national economy. As the figure shows, the estimated effect of perceptions of corruption on presidential approval (the toll of corruption) when individual evaluations of the national economy are at their lowest is -10.36 (this can also be assessed directly from Table 1 because all independent variables run from 0 to 1). However, as perceptions of the national economy improve, the toll of corruption becomes lower until, at last, it becomes insignificant when people perceive that the country’s economy is very good.[20] 

Personal economic conditions continue to matter, but still at relatively lower levels. That said, do citizens trade off corruption for that individual economic well-being, or is a more general (sociotropic) perception of the economy the more relevant moderating variable? To address this question, we first tested the conditioning influence of personal economic evaluations. The results are found in Table 2, which shows that when we substitute in an interaction between Personal Economy and the corruption variable, we do find a conditional effect, but one that is comparatively less substantial. Specifically, the maximum difference of the effect of corruption as conditioned by pocketbook economic evaluations is 7.89 compared to a maximum difference of 11.98 of sociotropic economic evaluations (the latter value is seen in Table 1 and Figure 1, in the difference between the endpoints of the line in that figure). However, because of the tendency for these two variables to be strongly predictive of one another, it is possible that this interaction is capturing the moderating effect of more general perceptions of the national economy. In order to disentangle their distinct moderating effects, we specified a model in which the two interactions are included at the same time. In that fully interacted model (see column 2 in Table 2), the moderating effect of pocketbook economic evaluations is not statistically significant, while the maximum effect of sociotropic evaluations reaches levels of 18.06. This indicates that the moderation effect observed from perceptions of the household economy is likely driven by perceptions of the national economy. This finding is consistent with the notion that it is the general nature of the economic times, and not personal benefits, that are at the forefront of political evaluations by citizens in the Americas. [21] 

In sum, we find clear support for our expectation that perceptions of the general economic context condition the way in which individuals link perceptions of corruption to presidential job approval. When individuals perceive good economic times, they are reluctant to pin perceptions of corruption on the executive; but, when they perceive bad economic times, they punish the executive for perceived corruption, and to a significant degree. Moreover, we find evidence that perceptions of the more general macro-economic conditions are more consequential to this process than individual economic evaluations. This finding begs the question, do actual differences in the economic context moderate the corruption-approval link? To address this question, , in the following section we assess the moderating influence of the economy by using our region-level economic measure.

Hierarchical Model 

Individuals exist embedded in contexts, and these contexts matter. We focus on regions as contextual units with three considerations in mind. First, the general logic of sociotropic performance voting theoretically ought to apply to the sub-national level (see, e.g., Tucker 2006) and this may be particularly the case where countries tend to be characterized by high levels of intra-regional differences in economic output, such as is the case for much of Latin America. Second, and related, the regions we examine are selected into the sample design for each country because they refer to relevant socio-political boundaries.[22] Third, a regions-based analysis maximizes the number of observations given that the data cover only 19 countries but 107 distinct regions. Thus, this approach offers us sufficient leverage with which to conduct a robustness check of the extent to which our findings for individual-level economic evaluations travel to a context-oriented focus. 

In an initial set of analyses, we determined that the relationship between perceptions of corruption and presidential approval varies across sub-national regions[23] and, thus, we can ask are regional economic conditions a source of this variation? To answer this question, we develop a model that includes perceptions of Corruption and all other individual measures from the previous models and random effects for both the intercept and the slope of corruption on presidential approval. In addition, the model includes Regional Income Deterioration (a measure that represents the jth region proportion of people who reported that their income decreased in the last two years in the AmericasBarometer survey); Average Regional Income; Regional Income Deterioration*Corruption (a term interacting regional income deterioration and individual perceptions of corruption); Average Regional Income*Corruption (a term interacting the average regional income and individual perceptions of corruption); and, Country*Corruption (a vector of interactions between country dummy variables and the individual perception of corruption measure, to account for differences in the regions belonging to different countries).[24] The specification allows us to estimate the moderating effect of Regional Income Deterioration on the corruption-presidential approval link while controlling for regional income as well as any country level factors that might moderate the link between corruption and presidential approval.[25] The model was estimated using HLM software.[26]

Estimates from this model are presented in Column 3 of Table 1. We note that the inclusion of Regional Income Deterioration in the model explains roughly 62 percent of the between-region variance of our estimate of the effect of corruption on presidential approval.[27] As Column 3 of Table 1 shows, regional income deterioration significantly (p<.001) moderates the effect of corruption on presidential approval. That is, the results show that there is a substantively more negative effect of corruption perceptions in regions that have worse economic conditions than in regions with better economic conditions. To illustrate, Figure 2 presents estimates of the effect of perceptions of corruption on presidential approval in regions with different levels of income deterioration.

Endnotes

[1] We limit our analysis to presidential countries given that distinct way the executive functions in these countries. The AmericasBarometer only asks about approval of the “governing party” (not an individual executive) in parliamentary countries. Our results, however, are robust to a pooled analysis of all presidential and parliamentary systems included in the AmericasBarometer.

[2] While citizens can use retrospective evaluations to enhance representation, it is overly optimistic to conclude that retrospective voting always results in the best outcomes (Achen and Bartels 2004; Bartels 2008; Przeworski 1999).

[3] Singer and Carlin (forthcoming) find this holds on average in Latin America, but also that pocketbook voting tendencies appear greater in the poorer countries of the region.

[4] Our specific focus is on perceptions of corruption; we note that mean levels of these are highly predictive of elite-produced country corruption scores (see Tverdova 2011).

[5] The complete effects of corruption extend further, for example to trust in civil servants and interpersonal trust (see Anderson and Tverdova 2003; Seligson 2002, 2006).

[6] Individual and contextual characteristics may condition the weight of personal versus general economic considerations (e.g., Gomez and Wilson 2001; Singer and Carlin forthcoming).

[7] We note that our focus on the strength of the association between issues and presidential approval is different from a focus on the salience of an issue. In our framework there can be times in which corruption is considered a salient (important) issue, but not one for which the executive is held responsible.

[8] We thank LAPOP and its major supporters (USAID, the United Nations Development Program, the Inter-American Development Bank, and Vanderbilt University) for making the data available.

[9] In the individual level models we account for the complex sample design using STATA’s svy command, which yields weighted estimates and standard errors robust to stratification and clustering. The HLM analysis allows us to: separate the within and between region variance in order to include variables at this level; weight the individual level observations using the standard AmericasBarometer probability weights; yet, it does not allow us to account for the within cluster correlations. However, we note that the standard errors of the individual level covariates across the svy models and the HLM models do not reflect substantive differences in size.

[10] M1: Speaking in general of the current administration, how would you rate the job performance of President [name]?

[11] EXC7: Taking into account your own experience or what you have heard, corruption among public officials is: very common, common, uncommon or very uncommon?

[12] SOCT1: How would you describe the country’s economic situation? and IDIO1: How would you describe your overall economic situation? [Very good, good, neither good nor bad, bad or very bad?]

[13] Q10E. Over the past two years, has the income of your household: increased; stayed the same; decreased?

[14] VB3: Who did you vote for in the last presidential elections of [year]]? This measure captures those that both weakly and more strongly identify with the party of the incumbent. Because the AmericasBarometer asks a filter question (do you identify with a political party?) before asking citizens’ party identification, an indicator built on that question would likely capture only strong partisans (especially given that in many countries being a “party identifier” is not as socially accepted as in the United States). In alternative analyses we separated out abstainers from opposition voters; the results for our key measures were unchanged and thus we present the more parsimonious model.

[15] These controls increase the precision of the model as individual characteristics and political loyalties affect not only approval but also perceptions of corruption (see Tverdova 2011).

[16] As one validation check on this measure, we took the case of Bolivia, in which the survey’s region measure maps directly onto departments, and we examined the relationship between our measure and a measure of average change in per capita GDP for each department. While the number of observations, 9 departments, is too small to yield a highly reliable test of this relationship, the two are negatively correlated (Pearson’s R = -.47) as we would expect.

[17] Regional Income Deterioration and Average Regional Income (the average income of the jth region) are only modestly correlated (Pearson’s R = -.23, p<0.015). Our conclusions are robust to excluding the latter.

[18] We use OLS for clarity; the results are robust to ordered logit and probit analyses.

[19] As socioeconomic and demographic controls we include the individual’s household income; the respondent’s age; an indicator variable for the sex of the respondent (coded 1 if female, 0 if male); the individual’s years of education; and, whether the interview took place in an urban or rural setting. Region dummy variables are not shown for space constraints but available upon request. All models are robust to analyses with missing data imputed via MICE in Stata (results available upon request).

[20] Conversely, the results we find can be interpreted as indicating a moderating effect of perceptions of corruption on the effect of the national economy (see Kam and Franzese 2007). In this case, the interaction results show that the effect of the national economy is always significantly positive but stronger among those perceiving higher levels of corruption.

[21] We also considered an interaction between perceptions of corruption and the respondents’ reported experience with vote buying. In that alternative model, the moderating effect of national perceptions of the economy is robust to this alternative model. Interestingly, in that model, being a target of vote buying is associated with an increase in the negative effect of corruption.

[22] Seewww.lapopsurveys.org for more information on the samples; in brief, the samples are stratified by conventional boundaries demarcating major socio-political regions in the country. For example, in Bolivia, the 9 departments constitute the 9 regions in the sample and in Mexico the sample is stratified by the north, south, center, and center-west regions which correspond to the types of regional divisions that scholars of voting behavior (for example) apply in analyses of voter choice in that country (e.g., Domínguez, Lawson, & Moreno 2009).

[23] We estimated an individual level model of presidential approval just like in Table 1, column 1, but allowed the intercept and the effect of corruption on presidential approval to vary randomly across subnational regions. To do this we specified the following model:

PA_ij = γ₀₀ + γ₁₀ × Income_ij + γ₂₀ × Age_ij + γ₃₀ × C_ij + γ₄₀ × NE_ij + γ₅₀ × PE_ij + γ₆₀ × Female_ij + γ₇₀ × Education_ij + γ₈₀ × Urban_ij + γ₉₀ × VotedPres_ij + u₀ + u₃ × C_ij + e_ij

Here, i identifies an individual and j identifies a region. PA_ij is the presidential approval score of individual i in region j; γ₀₀ is the overall intercept; and u₀ is the between-region random effect on the intercept. γ₃₀ is the overall estimated effect of perceptions of corruption (C) on presidential approval; u₃ is the between-region random deviation in that effect; and e_ij is the individual-level random error. NE and PE denote national and personal economic evaluations, respectively. The symbol × indicates multiplication, and _ij identifies an individual within a region.

From this specification the variance component associated with both the intercept and the slope of interest can be estimated. The total variance associated with the effect of corruption on presidential approval across sub-regions is 20.24 (p < .001), which indicates that although the effect of corruption on presidential approval is negative on average (as shown in Table 1), it varies significantly across sub-national regions.

[24] These latter results (Country*Corruption) are not shown for the sake of parsimony, but available.

[25] The full model specification presented in Table 1 is as follows:

Presidential Approval_ij = γ₀₀ + γ₁₀ × Corruption_ij + γ₁₁ × Corruption_ij × Regional Income_j + γ₁₂ × Corruption_ij × Regional Income Deterioration_j + γ₁₃ × Corruption_ij × Φ_jk + Corruption_ij × u₁ + Γ_k0 × Θ_ijk + u₀ + e_ij

Here, Φ_jk is the vector of country dummy variables covering the presidential countries of the Americas, with Mexico as the omitted category. Θ_ijk is the vector of individual-level covariates: age, sex, income, education, household income decrease, perceptions of national income, perceptions of household income, urban/rural residence, and whether the respondent voted for the president in the last election. Γ_k0 represents the corresponding covariate coefficients. The index k labels vector components; vector products represent the combined contributions of those components. The symbol × indicates multiplication, and an underscore (_) introduces an index. u₀ and u₁ are the region-specific random intercept and corruption-slope deviations, respectively; e_ij is the individual-level random error.

The results are robust to the inclusion of direct (non-interacted) country and region fixed effects for the intercept, but these are omitted for the sake of parsimony. Our results remain robust to including a vector of regional dummy variables, a strategy that accounts for the direct effects of these or any other regional or country-level variable (results available from the authors).

[26] We weight the individual level observations using the standard AmericasBarometer probability weights. As is standard with hierarchical models (Raudenbush and Bryk 2001), individual level variables were centered at the regional mean; thus, each individual score represents the individual’s deviation from her regional mean.

[27] When Regional Income Deterioration is included, the variance component associated with the random effect of corruption decreases from 18.89 to 13.08.

[28] By average individual, we mean that the individual would have the average value on all individual covariates for his/her region. By average region, we mean that regional income level and the country level dummies are set at their mean. 32 (29.91%) of the regions have 20% of income deterioration or less, 64 (59.81%) of the regions have between 20 and 40% of income deterioration and 11 (10.28%) of the regions have 40% of income deterioration or more. Figure 2 does not show predicted for regions with income deterioration lower than 20% because we find no evidence of a significant effect of corruption at these levels (results upon request).

[29] As a robustness check, we re-estimated models 1 and 3 in Table 1 by including: 1) an additional interaction between perceptions of corruption and an individual indicator of having voted for the president (in the individual model) and 2) an additional interaction between corruption and the proportion of people who voted for the incumbent in the previous election in the jth region (in the hierarchical model). Our conclusions remain unchanged; results available upon request.

[30] If anything, previous research suggests that individuals are less likely to engage in heuristic processing under threatening conditions (see for example Marcus, Neuman, and MacKuen 2000).