miss proffessor

https://doi.org/10.1177/1532673X16684574

American Politics Research 2017, Vol. 45(2) 256 –279

© The Author(s) 2017 Reprints and permissions:

sagepub.com/journalsPermissions.nav DOI: 10.1177/1532673X16684574

journals.sagepub.com/home/apr

Article

Uninformed Voters and Corrupt Politicians

Marko Klašnja1

Abstract Repeated studies have shown that voters in mature democracies often do not punish corrupt politicians. The majority of existing explanations focus on institutions and their effect on voters’ ability to monitor politicians. In this study, I examine the differences between voters with varying levels of political awareness. The previous literature provides contradicting expectations, suggesting that both the low- and high-awareness voters may be less tolerant of corruption. Drawing on several decades of data on corruption and voting in the U.S. congressional elections, I find that the high-awareness voters are on average less likely than the less politically aware to vote for corrupt incumbents. This association appears to stem from the high-awareness voters’ greater knowledge and better understanding of incumbents’ involvement in corruption. However, partisanship mitigates the differences between the low- and high-awareness voters, as the highly aware are more partisan, and strong co-partisans are more willing to forgive corrupt incumbents.

Keywords political awareness, corruption, U.S. Congress, accountability

Introduction

Representative Charles C. Diggs Jr. was convicted to 3 years in prison on October 7, 1978, for illegally diverting more than US$60,000—around

1Georgetown University, Washington, DC, USA

Corresponding Author: Marko Klašnja, School of Foreign Service and Government Department, Georgetown University, Washington, DC 20057, USA. Email: [email protected]

684574APRXXX10.1177/1532673X16684574American Politics ResearchKlašnja research-article2017

Klašnja 257

US$410,000 in present value—of his congressional employees’ salaries to his personal use.1 Exactly 30 days later, he won reelection to the U.S. House of Representatives with 79% of the vote. Less than a year later, he was cen- sured by Congress, and agreed to repay more than US$45,000, eventually resigning on June 3, 1980. Representative Diggs’s case is far from unique, because repeated studies have shown that voters in mature democracies fre- quently elect or reelect corrupt politicians.2 For example, more than 60% of incumbent legislators in the U.S. House of Representatives alleged of or charged with corruption between 1968 and 1990 were reelected (Peters & Welch, 1980; Welch & Hibbing, 1997); similar reelection rates are found among corrupt legislators in Italy (Chang, Golden, & Hill, 2010) and Japan (Reed, 1999).

Why do voters often reelect corrupt politicians? A number of studies have focused on how institutions may promote or constrain voters’ ability to moni- tor and oust politicians (e.g., Chang & Golden, 2006; Kunicova & Rose- Ackerman, 2005; Persson, Tabellini, & Trebbi, 2003), while assuming that the electorate is generally homogeneous. Others have explored heterogene- ities within the electorate that may mitigate the effect of corruption, such as voters’ partisan allegiances (e.g., Anderson & Tverdova, 2003; Anduiza, Gallego, & Muñoz, 2013). In this article, I systematically examine another potentially consequential heterogeneity in the electorate: voters’ political awareness. Based on a rich existing literature on political awareness, we have reasons to believe that the large variation in political awareness in the elector- ate could be associated with the uneven propensity to punish incumbents’ corruption. Intriguingly, the current body of theory and evidence provides conflicting expectations as to whether the low- or the high-awareness voters may be more likely to punish corrupt politicians. Some studies suggest that the low-awareness voters may be more likely to respond to corruption scan- dals, as they are more sensitive to campaign-specific conditions, less parti- san, and thus less biased than the high-awareness voters in evaluations of incumbent performance (Dimock & Jacobson, 1995; Zaller, 2004). Other studies imply that the high-awareness voters may be more likely to punish corrupt politicians, as they are more knowledgeable of the candidates, are better able to link retrospective judgments to vote choice, and are less suscep- tible to incumbent campaign appeals (Delli Carpini & Keeter, 1996). Finally, some studies suggest that there should be no differences between the low- and the high-awareness voters, as less attentive voters may compensate for the lack of information by using various heuristics and cues (Lupia, 1994; Popkin, 1994).

Because it is difficult to credibly experimentally manipulate corruption scandals and habitual political awareness, this study is limited to describing

258 American Politics Research 45(2)

associations in the observational data, constructed by merging the cases of corruption involving members of Congress with the individual-level survey data from the American National Elections Studies over several decades. Causal relationships are hard to establish because for example, politicians may strategically choose to engage in corruption when facing more or less attentive constituencies, or because voters may strategically choose their level of attentiveness to politics in response to corruption scandals. As is the case in most observational studies, it is also challenging to control for all the potential unobserved confounding variables.

I find that, on average, the high-awareness voters are consistently less likely to support corrupt incumbents compared with clean incumbents than the low-awareness voters in both the House and Senate elections. Consistent with the existing literature supporting this finding, this positive association between political awareness and sensitivity to corruption appears to stem from the high-awareness voters’ greater knowledge as well as better under- standing of the details of incumbents’ involvement in corruption. When incumbent corruption is less complex, the low-awareness voters are equally sensitive to corruption; however, when the corruption cases are more compli- cated, only the high-awareness voters exhibit negative predispositions toward corruption politicians.

However, in line with expectations that the more highly aware may be more biased in incumbent evaluations because of their stronger partisanship, I also find evidence that the difference between the low and high-awareness voters is less pronounced among incumbent co-partisans than the indepen- dents and out-partisans. Thus, the correlation between awareness and partisan- ship somewhat mitigates—but does not eliminate—the positive association between political awareness and the electoral sanctioning of corruption.

Overall, this positive association between sensitivity to corruption and political awareness is quite stable across a number of different specifications, alternative measures of incumbent support, and modeling choices. Moreover, my simple formal analysis (discussed in the Online Appendix) suggests that if incumbents are more likely to strategically engage in corruption in con- stituencies with lower political awareness, it would be harder to observe the correlations I do observe, thus leading to attenuation bias.

Literature

Corrupt incumbents in the U.S. Congress running for reelection on average lose votes compared with their “clean” counterparts, but have nonetheless been reelected more than 60% of the time in the past several decades (Abramowitz & Segal, 1992; Peters & Welch, 1980; Welch & Hibbing,

Klašnja 259

1997). Even the representatives implicated in the 1992 House Bank scandal, which coincided with the greatest turnover in the House in 40 years, were reelected at the rate of more than 80% (Alford, Teeters, Ward, & Wilson, 1994; Dimock & Jacobson, 1995).3 Similar reelection rates are found among legislators in Italy (Chang et al., 2010) and Japan (Reed, 1999), as well as mayors in Spain (Barbera, Fernandez-Vazquez, & Rivero, 2016; Jimenez & Cainzos, 2006) and France (Lafay & Servais, 2000).

Many existing explanations of why corrupt politicians can maintain public support focus on institutions and their effect on voters’ ability to monitor and sanction politicians. Scholars have examined the effects of electoral rules (Kunicova & Rose-Ackerman, 2005; Persson et al., 2003), institutions that affect the “clarity of responsibility” of incumbent governments, such as party cohesion (Tavits, 2007), or freedom of the press (Besley & Burgess, 2002; Brunetti & Weder, 2003); the effects of decentralization (Fisman & Gatti, 2002; Treisman, 2000) as well as seniority rules in legislatures and other institutional sources of the incumbency advantage that may help insulate cor- rupt incumbents (Herrick, 2000; Nyblade & Reed, 2008; Peters & Welch, 1980; Reed, 1999).4

These explanations usually assume that all voters are equally motivated and capable of assessing the nature, significance, and consequences of cor- ruption. However, such assumptions obscure potentially consequential het- erogeneities in the electorate. Scholars have examined some important sources of variation within the electorate itself, most notably the extent to which individuals’ partisan allegiance affects the weight they place on cor- ruption in their political evaluations. These studies find strong evidence that co-partisans are more forgiving of corrupt politicians than independents or out-partisans (Anderson & Tverdova, 2003; Anduiza et al., 2013; Blais, Gidengil, & Kilibarda, 2015).5

In this article, I contribute to this line of inquiry by focusing on another potentially important variation in the electorate—in voters’ political aware- ness. Dimock and Jacobson (1995) showed that while more than 90% of respondents in the 1992 National Election Study claimed to be aware of the House Bank scandal, only 43% got it right as to whether their representative wrote any bad checks, and that even “some of them were clearly guessing” (p. 1152). In this vain, I propose to examine more systematically the associa- tion between voters’ political awareness and support for corrupt incumbents relative to incumbents not involved in corruption cases.

My focus on voters’ political awareness is related to but different from two recent strands of research that examine the role of information about corrup- tion in voter behavior. First, several recent studies have focused on the con- sequences of revealing information about corruption in low-transparency

260 American Politics Research 45(2)

environments in developing democracies (e.g., Chong, De La O, Karlan, & Wantchekon, 2015; Ferraz & Finan, 2008). Unlike these studies, the corrup- tion cases I study (described in Tables A1 and A2 in the Online Appendix) are instances of publicized corruption charges that are observable by voters—to the extent that they are attentive to them.6

Second, several studies have focused on the effects of randomly exposing survey respondents to information about hypothetical corruption scenarios on their attitudes and behavior in the context of survey experiments (Anduiza et al., 2013; Banerjee, Green, McManus, & Pande, 2014; Doherty, Dowling, & Miller, 2011; Klašnja & Tucker, 2013). Although these studies may have greater internal validity than the present article, which relies on observational data, they are of limited external validity as they focus on a limited set of (hypothetical) corruption scenarios. My study draws on a large number of real-world cases over several decades. Moreover, unlike those studies, my study utilizes the variation in the types of real-world corruption cases (such as their complexity) to directly engage with the conflicting expectations based on the existing literature on political awareness.

Theoretical Expectations

Scholars have shown repeatedly that the level of knowledge of politics in the American public is generally low and highly variable (e.g., Campbell, Converse, Miller, & Stokes, 1960; Converse, 1964; Delli Carpini & Keeter, 1996). Could the low level and high variance of political awareness contribute to the frequent electoral impunity of corrupt politicians? The existing litera- ture on political awareness has not addressed this question systematically, but its insights provide contradicting expectations. Some scholars argue that vot- ers with low attentiveness to politics may compensate for the lack of informa- tion by using various heuristics, such as cues from reference groups (e.g., Lupia, 1994; Lupia & McCubbins, 1998; Sniderman, Brody, & Tetlock, 1991), “gut-level” reasoning (Popkin, 1994), or simple retrospective judgments of incumbent’s performance (e.g., Fiorina, 1981; Sniderman, Glaser, & Griffin, 1990). The publicized corruption charges are likely to be strong signals during an election campaign and may represent a ready heuristic device. Also, cam- paigns can help close the gap between the inattentive and attentive by the time of the election (Sekhon, 2006). These arguments imply that the low-awareness voters may be as likely to withdraw support from corrupt politicians as the more politically attentive voters. In other words, the null hypothesis is:

Null Hypothesis (H0): Variation in political awareness is not associated with support for corrupt incumbents relative to clean incumbents.

Klašnja 261

Some studies suggest a negative association between political awareness and support for corrupt incumbents. The less attentive voters tend to be less knowledgeable of incumbents and challengers (Zaller, 1992), which may extend to the details of any involvement in corruption. The high-awareness voters may be better able to understand the consequences of corruption; for example, that representative dogged by corruption investigations may be less effective, as in the aforementioned example of Representative Diggs.7 The low-awareness voters may also be less able to differentiate between law- breaking corruption and negative campaigning, perhaps because they are less able than the high-awareness voters to assess the relevance and quality of reference group cues (Kuklinski & Quirk, 2000; Luskin, 2002). Furthermore, low-awareness voters have been shown to be more susceptible to dominant incumbent campaigns (Kuklinski & Quirk, 2000; Zaller, 1992), and incum- bents facing corruption charges are likely to campaign particularly hard. These arguments imply the following alternative to H0:

Hypothesis 1 (H1): Support for corrupt incumbents relative to clean incumbents decreases as political awareness increases.

Other studies suggest that the low-awareness voters may in fact be more responsive to incumbent corruption than the high-awareness voters. The high-awareness voters tend to be more ideological, and strong partisans have in turn been shown to form biased evaluations of incumbent performance (e.g., Duch, Palmer, & Anderson, 2000; Krause, 1997), and to more heavily discount corruption allegations against their preferred candidate (e.g., Dimock & Jacobson, 1995, and citations mentioned above). Moreover, the low-awareness voters have been shown to be more sensitive to campaign- specific events (in presidential elections) like economic conditions, foreign policy crises, or candidates’ ideological shifts (Zaller, 2004). They may there- fore be more responsive to corruption scandals. This implies another alterna- tive hypothesis:

Hypothesis 2 (H2): Support for corrupt incumbents relative to clean incumbents increases as political awareness increases.

Measures of Corruption and Awareness

I now describe how I define and measure incumbent corruption and voters’ political awareness. I then discuss the main empirical specification and the quantities of interest, followed by the presentation of the results. Due to space

262 American Politics Research 45(2)

constraints, much of the methodological discussion is relegated to the Online Appendix.

Incumbent Corruption

I rely on the definition of corruption commonly used in the applied literature, whereby corruption is understood as a potential misuse of office for personal or political gain (e.g., Rose-Ackerman, 1978). This definition leads me to exclude such high-publicity scandals as Idaho Senator Steven Symms being visibly drunk on the Senate floor in 1986, or Arkansas Senator Tim Hutchinson’s messy divorce and subsequent marriage to a former staffer in 2002, neither of which involve actual misuse of office. Other scholars, as well as my analysis below, show that voters are considerably less sensitive to scan- dals not involving some form of abuse of power (Doherty et al., 2011; Zaller, 1998).8 The focus on misuse of office also implies that the majority, though not all, of the corruption cases I study involve an action from the authorities, such as the federal prosecutors, the congressional ethics committees, the Federal Election Commission, the Internal Revenue Service, and so on. In those cases where there is a subsequent investigation, the cases are included irrespective of their final outcome (i.e., indictment, conviction, acquittal or dropped charges), so as not to presuppose the results of the analysis.9 As with any data on corruption, its covert nature means that the dataset likely underes- timates the actual level of corruption.

The dataset of corruption cases covers the period 1968-2002, to corre- spond with the biannual National American Election Studies surveys I use to construct the measure of political awareness and other relevant variables, which I describe below and in Section A5 in the Online Appendix.10 Drawing on my own data collection and a number of secondary sources (Brown, 2006; Congressional Quarterly, 1992; Congressional Quarterly Almanac, n.d.; Hirano & Snyder, 2012; Noyer, 1995; Roberds, 1997), I identify a total of 470 corruption cases (406 in the House and 64 cases in the Senate). Of these, 136 corruption cases (119 in the House and 17 in the Senate) are represented with at least one survey respondent in the American National Election Study (ANES) data.11 On the House side, this gives 1,670 respondents from district- years with a scandal and 27,504 respondents from district-years without a corruption case.12 For the Senate, there are 1,276 respondents in state-years with a scandal and 15,436 respondents in state-years without one. The num- ber of corruption cases represented in the analysis relative to the full dataset of corruption charges is lower because of the limited coverage of congres- sional districts and states in the ANES data, and the fact that some politicians

Klašnja 263

implicated in corruption resign or retire. Short descriptions for those cases retained in the analysis are given in Tables A1 and A2 in the Online Appendix.

Political Awareness

I construct an individual-level measure of political awareness, following the established practice of using knowledge and policy issue questions from sur- veys (Althaus, 2003; Delli Carpini & Keeter, 1996; Zaller, 1992).13 Specifically, I rely on a number of items from the biennial ANES time-series surveys conducted between 1968 and 2002. The items probe respondents’ factual knowledge of politics and institutions (e.g., the party holding the majority in the House), their ability to recognize important political figures (e.g., the name of the vice president), and the proper placement of parties and candidates on a left–right scale on policy issues. These items have been proven superior in measuring the differences in political awareness to other proxies such as education, self-reported media use, or political participation (Luskin, 1987; Zaller, 1992).14 The political awareness score for each respon- dent from each survey is calculated using factor analysis on all the relevant survey items; the primary factor is large, and the construct validity of the scores is high. I rank the score for each respondent in a particular survey with respect to that survey’s maximum; this procedure removes time trends and makes the scores comparable across surveys. Section A2 in the Online Appendix provides further details on the construction of the political aware- ness score.

Empirical Model

To evaluate the competing expectations laid out above, I specify the follow- ing simple ordinary least squares (OLS) model:

Y I C I C Ei j t t i j t j t i j t j t i j t i j t, , , , , , , , , , , , ,= + + + × + +α β β β β1 2 3 4PID (1)

where i is an individual, located in a district or state j, surveyed in an election year t. In the main analysis, Y is the binary incumbent vote choice, coded as 1 if the respondent reported voting for the incumbent. I discuss further below and show in Figure A2 in the Online Appendix that the results are very simi- lar with incumbent feeling thermometer rating and incumbent approval as alternative dependent variables. I is the respondent’s political awareness score, C is the corruption indicator variable equal to one if the incumbent was involved in a corruption case, I × C is the interaction between political aware- ness and the corruption variable; PID is the respondent’s self-reported

264 American Politics Research 45(2)

partisanship, taking values {−3, 3}, where 3 represents a self-reported strong identifier with the incumbent’s party, 0 represents an independent, and −3 represents a strong out-partisan. αt are the election year fixed effects, control- ling for all the election-specific factors that lead some incumbents to perform better than others. Tables A7, A8, A9, and A10 in the Online Appendix show that the results are substantively very similar when this specification is enhanced with a number of other potentially important individual, candidate, and district- or state-level control variables, and when these variables are interacted with political awareness, partisanship, and corruption.15

Because the ANES suffers from considerable item nonresponse, I use multiple imputation to handle the missing data, because discarding the miss- ing data induces inefficiency and possibly bias (Rubin, 1987).16 All the esti- mates presented below are combined across the imputed datasets using the standard rules developed in Rubin (1987). The imputation process is dis- cussed in more detail in Section A4 in the Online Appendix. All the estima- tions are weighted by the sampling weights, and the standard errors are clustered by district-year or state-year.

Given the focus on observational data, the analysis in this article is limited to describing associations rather than causal relationships, because both cor- ruption and political awareness are potentially endogenous to voters’ incum- bent support. There may also be unobserved factors that make it implausible to assume that conditional on observed covariates, the high-awareness voters are identical to the low-awareness voters in all respects except for political awareness.

That said, using simple formal reasoning, Section A7 in the Online Appendix shows that an important, potentially nonrandom, aspect of the data-generating process—incumbents choosing to strategically engage in corruption based on their constituents’ political awareness—likely induces attenuation bias in rejecting H0 in favor of either H1 or H2. For example, assuming that H1 is true leads to the conclusion that we should observe fewer corruption cases among incumbents from districts with higher-awareness voters, making it harder to find a negative association between higher politi- cal awareness and support for corrupt incumbents (i.e., making it harder to find evidence consistent with H1). The same kind of logic applies if H2 is assumed to be true (see Section A7 in the Online Appendix).

Based on the estimates from Equation 1, the main quantity of interest pre- sented below is the difference in the predicted probability of voting for a corrupt incumbent relative to a clean incumbent, estimated for different lev- els of political awareness, while holding other variables constant.17 For ease of exposition, the results are presented graphically in the main text, while the coefficient estimates are given in the Online Appendix. To examine the full

Klašnja 265

range of values of political awareness, I present differences in predicted vote probabilities from the first percentile to the 99th percentile of political aware- ness, in five-percentile steps. If the differences in predicted vote probabilities for a corrupt incumbent relative to a clean incumbent change little across the different values of political awareness, then naturally H0 cannot be rejected. If the predicted support for a corrupt incumbent decreases (increases) relative to that for a clean incumbent as political awareness increases, then H0 is rejected in favor of H1 (H2).

Results

The lines in Figure 1 represent the difference in the predicted probability of voting for a corrupt incumbent relative to a clean incumbent across different levels of political awareness for the House (left panel) and the Senate (right panel). The shaded area represents the 90% confidence interval.18

The figure shows that the H0 is rejected in favor of H1 for both the House and Senate elections—the predicted support for a corrupt incumbent relative to a clean incumbent declines as political awareness increases. In the House, a voter with the highest level of political awareness is about 11 percentage points less likely to vote for a corrupt incumbent compared with a clean incumbent. This is a nontrivial difference of about 16% of the average

P re

di ct

ed vo

te pr

ob ab

ili ty

fo

r co

rr up

tv s.

cl ea

n in

cu m

be nt

-.

2 -.

1 0

.1

etaneSesuoH

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

P re

di ct

ed vo

te pr

ob ab

ili ty

fo

r co

rr up

tv s.

cl ea

n in

cu m

be nt

-.

2 -.

1 0

.1

Figure 1. Corruption and incumbent vote across political awareness. Note. The lines represent the difference in the predicted probability of voting for a corrupt incumbent relative to a clean incumbent for different levels of political awareness. Results are based on the model in Equation 1. Estimates are calculated for values of political awareness from the first percentile to the 100th percentile in five-percentile steps. The shaded area represents the 90% confidence interval. All quantities are averaged over 15 imputed datasets.

266 American Politics Research 45(2)

reported unconditional propensity of voting for the House incumbent (71%). On the contrary, individuals with the lowest level of political awareness, ceteris paribus, are about a percentage point more likely to vote for corrupt incumbents compared with clean incumbents, but this estimate is statistically indistinguishable from zero. In between the two extremes, support for a cor- rupt incumbent compared with a clean incumbent steadily declines with the increase in political awareness, and it is statistically different from zero for roughly the upper half of the distribution of political awareness.19

Estimates for the Senate are generally quite similar, albeit somewhat less precise. The most attentive voters are about 10 percentage points less likely to support a corrupt incumbent than a clean incumbent (about 17% of the average unconditional probability of voting for the incumbent in a Senate race). The estimates are statistically different from zero for the top quintile in terms of political awareness, noticeably less than in the House.20

Although the predicted support for corrupt incumbents decreases relative to clean incumbents as political awareness increases, Figure 1 does not fully reveal the extent to which the high-awareness voters are different from the low-awareness voters. Figure 2 examines the estimated difference in pre- dicted vote probabilities for each level of political awareness relative to the

-. 25

etaneSesuoH

10 1

C ha

ng e

in co

rr up

tv s.

cl ea

n in

cu m

be nt

vo te

pr

ob ab

ili ty

re la

tiv e

to lo

w es

ta w

ar en

es s

-. 2

-. 15

-. 1

-. 05

0

C ha

ng e

in co

rr up

tv s.

cl ea

n in

cu m

be nt

vo te

pr

ob ab

ili ty

re la

tiv e

to lo

w es

ta w

ar en

es s

-. 2

-. 15

-. 1

-. 05

0

-. 25

.2 .4 .6 Political awareness

.8 0 .2 .4 .6 Political awareness

.8

Figure 2. Corruption and incumbent vote for the high- versus the low-awareness voters. Note. The dots show the difference in the predicted vote probability for a corrupt incumbent and a clean incumbent, for respondents at a given level of political awareness compared with respondents with the lowest political awareness (at first percentile). Results are based on the model in Equation 1. Estimates are calculated for values of political awareness from the first percentile to the 100th percentile in five-percentile steps. The shaded area represents the 90% confidence interval. All quantities are averaged over 15 imputed datasets.

Klašnja 267

lowest-awareness voters. It shows that the slope of the line is statistically significantly steep for both the House and the Senate. For example, voters in the House elections with the average level of political awareness are about six percentage points less likely to vote for a corrupt incumbent compared with a clean incumbent than the least attentive respondents, and are about six percentage points less likely to do so than the most politically aware voters.

Figures 1 and 2 strongly suggest that the high-awareness voters are less likely to support corrupt incumbents than the low-awareness voters. However, the dependent variable—the self-reported incumbent vote—is potentially problematic because voting for the election winner is typically overreported (Wright, 1993), and incumbents are overwhelmingly the win- ners. To guard against this concern, Figure A2 in the Online Appendix shows that the results are qualitatively very similar with two alternative measures of incumbent support: the incumbent feeling thermometer score and incumbent approval.

Mechanisms

As discussed above, the high-awareness voters may react more strongly to incumbent corruption because they are more familiar with the details of the corrupt act than the less attentive voters. Alternatively, the low-awareness voters may respond differently than the high-awareness voters to corruption even when they are equally informed about it, perhaps because they under- stand corruption differently, or are less able to tell apart lawbreaking from negative campaigning. Of course, the two explanations need not be mutually exclusive. Teasing out these expectations is constrained by the lack of sys- tematic survey questions about voters’ knowledge and understanding of the specific corruption cases covered in the analysis. However, I offer two addi- tional analyses to evaluate these mechanisms. First, the 1992 National Election Study included a battery of questions directly related to the 1992 House Bank scandal, which is included among the corruption cases in the House sample. There are both advantages and disadvantages to focusing on the House Bank scandal. On one hand, it was a big scandal involving many representatives (275 House members wrote at least one bad check). Given such magnitude, this scandal is likely a relatively hard case for examining the differences in the familiarity with corruption among voters with different lev- els of political awareness. On the other hand, the representatives varied widely in the extent of the check overdrafts, from 920 bad checks written by the office of Rep. Robert Mrazek to 33 representatives writing only one bad check. Also, except for about two dozen cases, the writing of bad checks did not involve breaking of any laws per se.21 These nuances make the House

268 American Politics Research 45(2)

Bank scandal a good case to examine the variation in understanding of the details by respondents with different levels of political awareness.

The 1992 ANES asked the respondents whether they had heard of the House Bank scandal and if so, whether they knew if their representatives wrote any bad checks, approximately how many, and whether it involved breaking of any laws. The survey also elicited respondents’ attitudes toward the proper electoral punishment of the overdrafts. Cross-checking with the data on the actual involvement of respondents’ own representatives, Figure 3 evaluates how the knowledge and understanding of the details of the case differ among respondents with low and high political awareness. Each esti- mate shown in the graph represents the percentage point difference in the predicted probability of a response between respondents at the 75th and the

Heard of House Bank scandal?

Aware of own repr. involvement?

Correct on extent of check writing?

Correct on lawfulness?

Lose office if wrote any bad checks?

Lose office only if wrote a lot of bad checks?

-.2 -.1 0 .1 .2 .3 Change in predicted response between

high and low awareness (75th vs. 25th percentile)

Figure 3. Knowledge and understanding of the House Bank scandal for the low- versus the high-awareness voters. Note. The dots show the difference in the predicted probability of a response between the respondents at the 75th and the 25th percentiles in the distribution of political awareness, controlling for a number of important respondent characteristics such as partisanship, education, and income. More details on the regression specifications are given in the Online Appendix. The caps represent the 90% confidence intervals. All quantities are averaged over 15 imputed datasets.

Klašnja 269

25th percentiles of the distribution of political awareness, controlling for a number of important respondent characteristics such as partisanship, educa- tion, and income.22

The results in Figure 3 appear to support both notions as to why the low- and high-awareness voters respond differently to corruption. The highly politically aware are more likely than the less attentive voters to have heard of the House Bank scandal (the top estimate), and to have correctly identified the involvement and the number of check overdrafts by their own representa- tive (the second and the third estimate from the top). Moreover, the high- awareness voters have a more nuanced understanding of the nature of the scandal. They are more likely to have understood that most representative- ness did not break any laws (the third estimate from the bottom of Figure 3). Also, they were less indiscriminate in their attitudes about electoral punish- ment: they were less inclined to believe that all representatives who wrote any bad checks should be voted out of office (the penultimate estimate), but more inclined to support the electoral punishment of the representatives with a lot of overdrafts (the last estimate in Figure 3).

These results are further corroborated by examining the variation in incumbent support among the low- and high-awareness voters when the cor- ruption cases are broken down along two dimensions: (a) whether charges are more or less complex (e.g., multiple charges for insider trading vs. failure to report a vacation home on tax returns); and (b) whether the actions are clearly illegal vs. those which may represent abuse but are not illegal per se.23

The results are shown in Table A11 in the Online Appendix. The high- awareness voters are more sensitive than the low-awareness voters to scandals that involve more complex charges (column 1), whereas the low-awareness voters are as reactive as the high-awareness voters to the less complex scan- dals (column 2), suggesting that it is perhaps the lack of familiarity with the details of complex cases that makes the low-awareness voters less sensitive to corruption overall. As for legality of the charges, the results suggest that the electoral punishment is primarily directed at illegal acts, and mainly by the high-awareness voters (column 3), whereas the results for “shady but not ille- gal” acts are both substantively and statistically weaker (column 4). These findings seem broadly consistent with the notion that the high-awareness vot- ers are generally better able to understand the nuances of incumbents’ corrupt actions.

The Mitigating Role of Partisanship

An important a priori argument against H1 (that support for corrupt incum- bents is on average decreasing in political awareness), and in favor of H2

270 American Politics Research 45(2)

(that the opposite of H1 is true) discussed above is that the high-awareness voters are more likely to be strongly partisan (Zaller, 1992), and that strong partisans are more forgiving of corrupt officials from their own party (e.g., Anduiza et al., 2013). Although the results presented so far are in line with H1, here I examine to what extent there may be evidence consistent with H2 for certain partisan subgroups, which could mitigate the overall differences in sensitivity to corruption between the high- and low-awareness voters.

Figure 4 evaluates this possibility. Going from left to right, it plots the same quantity of interest as in Figure 1—the difference in predicted vote probabilities for a corrupt relative to a clean incumbent across political awareness—but splits the sample by voter partisanship.24 The two panels on the left show the results for the strong and weak out-partisans (self-reported identifiers with the opposition party), the panel in the middle shows the results for independents and leaners, and the two panels on the right show the results for the weak and strong co-partisans (identifiers with the incumbent party), respectively. The top panel shows the results for the House, and the bottom panel for the Senate.

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

House

Strong Out-Partisans Weak Out-Partisans Independents/Leaning Weak Co-Partisans Strong Co-Partisans

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

Senate

Strong Out-Partisans Weak Out-Partisans Independents/Leaning Weak Co-Partisans Strong Co-Partisans

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

0 .2 .4 .6 .8 1 Political awareness

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

E st

im at

ed ef

fe ct

of sc

an da

lo n

in cu

m be

nt vo

te pr

ob ab

ili ty

-.

2 -.

1 0

.1 .2

Figure 4. Corruption and incumbent vote across political awareness for different partisan groups. Note. The lines represent the difference in the predicted probability of voting for a corrupt incumbent relative to a clean incumbent for different levels of political awareness. Results are based on the model in Equation 1, augmented by a triple interaction between political awareness, partisanship, and corruption. The estimates are calculated for values of political awareness from the 1st percentile to the 100th percentile in five-percentile steps. The shaded area represents the 90% confidence interval. All quantities are averaged over 15 imputed datasets.

Klašnja 271

Figure 4 does support the notion that greater sensitivity to corruption by the high-awareness voters is mitigated by the strength of co-partisanship. The negative association between political awareness and support for corrupt incumbents (relative to clean incumbents) is stronger for out-partisans and independents than for the weak and particularly the strong co-partisans (i.e., the downward slope is less steep among the latter).25 At the same time, there is no evidence that the high-awareness voters are more likely to support cor- rupt incumbents than the low-awareness voters for any partisan group (i.e., there is no evidence strictly consistent with H2).

Conclusion

A number of previous studies have shown that corrupt incumbents can main- tain support when institutions force trade-offs between corruption and some other important dimension of vote choice, or when they make it harder for voters to clearly ascribe responsibility to the incumbent. Although institu- tions affect the ability of voters to monitor politicians, the results presented in this study suggest that improving the incentives of voters and their motiva- tion to engage in monitoring may also be important. I find considerable dif- ferences in the sensitivity to corruption between the low- and high-awareness voters. These differences appear to stem from the high-awareness voters both having more information and a better understanding of the details of incum- bent corruption. Although these differences are descriptive and need not rep- resent the causal effect of political awareness, they potentially point to the importance of a highly politically aware electorate for electoral accountabil- ity. Although the effects of increasing political awareness are somewhat countervailed by the fact that the high-awareness voters are more partisan and thus more polarized in their reactions to incumbent corruption, it may nonetheless be the case that incumbents in the U.S. Congress are able to maintain public support partly because large sections of the electorate ignore their incumbents’ involvement in corruption, and/or underappreciate the con- sequences of such incumbent behavior.

Several questions are potentially worth exploring in the future. Most directly, it may be useful to examine the differences in sensitivity to corrup- tion between the low- and high-awareness voters in other developed democ- racies with meager punishment of corrupt politicians, such as Italy, Japan, Spain, and France.26 Moreover, while I have presented evidence suggesting that political awareness is correlated with both knowledge and understand- ing of corruption, future work could examine further whether the low- awareness voters are more likely to support corrupt incumbents because they are unaware of corruption, or whether they vote differently than the

272 American Politics Research 45(2)

high-awareness voters even conditional on knowing about and understand- ing the consequences of corruption.

Acknowledgments

The previous version of this article was titled “Why Do Malfeasant Politicians Maintain Political Support? Testing the ‘Uninformed Voter’ Argument” Special thanks are due to Patrick Egan for many useful discussions and suggestions. For valu- able comments, I would also like to thank Michaël Aklin, Pablo Barberá, Miriam Golden, Sandy Gordon, S. P. Harish, James Hollyer, Michael Laver, Christopher Lawrence, Rebecca Morton, Jonathan Nagler, Natalija Novta, John Zaller, partici- pants at the 2010 APSA Annual Meetings, as well as the editor and several anony- mous referees. I am grateful to Lara Brown, Stephen Roberds, Jim Snyder, and Roc’ıo Titiunik for sharing their data. All errors are my own.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publi- cation of this article.

Notes

1. “Ethics and Crimes.” Congressional Quarterly (CQ) Press Electronic Library, CQ Almanac Online Edition, http://library.cqpress.com/cqalmanac/document. php?id=cqal78-1237462.

2. In this article, I rely on a view of corruption consistent with the common definition used in applied research—the misuse of office for private or political gain (e.g., Rose-Ackerman, 1978). See “Incumbent Corruption” section for more details.

3. Although the scandal was associated with a higher rate of retirement, it is not con- sidered the main cause (Groseclose & Krehbiel, 1994; Hall & van Houweling, 1995).

4. Others have focused on how noninstitutional contextual factors may moderate the electoral cost of corruption, such as the state of the economy (Klašnja & Tucker, 2013; Zechmeister & Zizumbo-Colunga, 2013) or electoral partisan stakes (Eggers, 2014).

5. Scholars have also explored the role of personal economic experiences (Tverdova, 2011) and education (Xezonakis, Kosmidis, & Dahlberg, 2015).

6. For example, Representative Diggs’s indictment, mentioned above, which took place almost 8 months before the election, was widely reported in the contem- porary press. See, for example, “U.S. Indicts Rep. Diggs for Fraud,” Chicago Tribune, p. 2, March 24, 1978.

Klašnja 273

7. Prior studies have shown that more educated Americans are more likely to con- sider things like bribery and illegal influence-peddling as corrupt (Gardiner, 2002; Johnston, 1986; Redlawsk & McCann, 2005), perhaps because they are more likely to be socialized into politics in a way that their views conform to the more common negative view of corruption.

8. Several cases included in the analysis, most notably the House Bank scandal, arguably do not involve real misuse of office. Although some House Bank scan- dal cases did involve actual legal abuses, such as bank fraud or using the check overdrafts to lend to one’s campaign, the majority of the cases rather involved an abuse (actual or perceived) of privilege conferred by holding office. I believe these cases are still in the spirit of the definition I use in this study.

9. Some cases involve congresspersons being acquitted or charges against them being dropped even though there were strong indications of impropriety. For example, in 1968, Life magazine twice accused Senator Edward Long of using his position to aid the imprisoned Teamster Union President James Hoffa, but the Senate ethics committee and the grand jury dropped the charges each time. In some cases, the formal charges were made but then dropped during the same term in office of the politician involved, allowing voters to observe that the investigation was closed. For example, in 1977, Rep. William Clay was under a grand jury investigation for alleged tax evasion and fraud. The Internal Revenue Service terminated the investigation in July 1977, 4 months before the election. I exclude such cases from the main analysis, but the results are very similar with them included, given that there only a few of them. Similarly, the results are insensitive to the inclusion of the cases where investigations were initiated only after a politician’s departure from Congress. For example, in the Koreagate scan- dal that broke out in 1976, representative Otto Passman was indicted on March 31, 1978, on charges of receiving illegal payments from a Korean lobbyist and tax evasion. However, he lost the party seat primary in 1976.

10. The more recent corruption cases are not included because the recent American National Election Study (ANES) data do not cover the majority of the constitu- encies involved.

11. Some corruption cases match onto only one or a few respondents in the ANES. Table A13 in the Online Appendix shows that the results are very similar when excluding the constituency-years with a low number of observations.

12. As shown in Table A6, the results are very similar when excluding the House Bank scandal, which accounts for approximately two thirds of the scandal obser- vations in the House sample. Table A6 also shows that the results based only on the House Bank scandal are similar to the main results, but somewhat statisti- cally weaker.

13. I will therefore perform all the analyses at the individual level. Another approach is to focus on districts or states as the unit of analysis. However, this is dif- ficult due to the lack of political awareness data at the aggregate level over a longer period of time. One potential solution is to estimate the aggregate-level political awareness from the individual-level survey data through multilevel

274 American Politics Research 45(2)

modeling and poststratification (Lax & Phillips, 2009; Warshaw & Rodden, 2012). However, the ANES—the only long-running surveys containing political knowledge items—are not suitable for such an approach because of clustered sampling and small samples (Stollwerk, 2013). Other aggregate proxies such as education do not correlate strongly enough with political awareness.

14. Moreover, general knowledge of politics is less susceptible to short-term changes in response to corruption scandals than proxies such as self-reported media use, thus alleviating one potential problem of endogeneity of the measure of political awareness to corruption.

15. Section A5 in the Online Appendix contains more details about the coding of each variable included in the various specifications. Despite the dependent vari- able in Equation 1 being binary, the model is estimated with OLS because a pro- bit or logit model with fixed effects would drop districts in elections for which there is no variation in the dependent variable, creating biases away from zero. Nonetheless, the results are very similar when using a probit model, as shown in Table A5 in the Online Appendix. The results are also very similar if the data are preprocessed with propensity score matching, or when using a multilevel model with individuals nested in districts/states and election years.

16. The principal reason for using multiple imputation is the challenge of construct- ing the political awareness scores in the presence of missing data. Because the awareness measure is comprised of a relatively large set of variables, even a small fraction of missing values in each variable, when combined across all vari- ables, can lead to a loss of a considerable number of observations. See Section A4 in the Online Appendix for more details.

17. Although the main analysis compares the support for corrupt incumbents with that for clean incumbents, the actual voter choice involves evaluating incum- bents against challengers. Table A9 in the Online Appendix shows that the results are substantively very similar when several variables capturing challenger char- acteristics are controlled for.

18. I report the 90% rather than the 95% confidence intervals for two reasons. First, the inclusion of election year fixed effects reduces the statistical power because the analysis only draws on within-election variation in political awareness. Second, corruption cases are rare events (comprising only 5.7% of observations in the House and 7.5% in the Senate), making it harder to precisely estimate the support for corrupt incumbents.

19. There is some indication in the existing literature that political awareness may have nonmonotonic effects on political behavior (Zaller, 1992). Adding higher order terms for political awareness does not change the substance of the results.

20. The Senate results may be weaker than the House results for several related reasons. First, there is usually higher name recognition and greater media attention devoted to the Senate candidates, which likely reduces the discrepan- cies between the high- and low-awareness voters. Second, the rate of retire- ment following a scandal is higher in the Senate (about 20%) than in the House (about 8%), suggesting that a scandal is costlier in the Senate. Related, the

Klašnja 275

Senate races are generally more competitive than the House races following a scandal. Finally, fewer Senators (2.5%) than Representatives (about 8%) resign after a scandal, which may suggest that the scandals in the House are on average more egregious.

21. Twenty-two House members were reprimanded by the House Ethics Committee for having abused their privileges at the House Bank, and several representatives subsequently faced felony charges and convictions. For example, Rep. Carl C. Perkins was sentenced to 21 months in prison on three felony charges, including a charge for bank fraud connected to the check overdrafts at the House Bank and several other financial institutions (see http://library.cqpress.com/cqalmanac/ document.php?id=cqal94-1102400#H2_10).

22. More details about the regression specifications and the coding of the depen- dent variables underlying Figure 3 are given in Section A3 in the Online Appendix.

23. This analysis includes only the House scandals. There are simply too few scan- dals in the Senate to make this kind of analysis feasible.

24. The estimates are based on a specification that adds a triple interaction between political awareness, partisanship, and corruption and the constituent two-way interactions to the model in Equation 1. The coefficient estimates are shown in Table A12 in the Online Appendix.

25. These results are consistent with high-awareness partisans engaging in motivated reasoning, in addition to or in place of their greater attention to and understand- ing of corruption. However, partisanship does not strongly moderate the relation- ship between scandal and incumbent vote, as the two-way interaction between partisanship and corruption shown in A12 in the Online Appendix (as well as in column 4 of Tables A7 and A8) is neither substantively large nor statistically significant. I thank an anonymous reviewer for pointing this out.

26. De Vries and Giger (2014) find differences in economic voting between the highly politically sophisticated and the less sophisticated voters across a number of countries.

References

Abramowitz, A. I., & Segal, J. A. (1992). Senate elections. Ann Arbor: University of Michigan Press.

Alford, J., Teeters, H., Ward, D. S., & Wilson, R. K. (1994). Overdraft: The political cost of congressional malfeasance. The Journal of Politics, 56, 788-801.

Althaus, S. L. (2003). Collective preferences in democratic politics: Opinion surveys and the will of the people. New York, NY: Cambridge University Press.

Anderson, C. J., & Tverdova, Y. V. (2003). Corruption, political allegiances, and attitudes toward government in contemporary democracies. American Journal of Political Science, 47, 91-109.

Anduiza, E., Gallego, A., & Muñoz, J. (2013). Turning a blind eye: Experimental evidence of partisan bias in attitudes towards corruption. Comparative Political Studies, 46, 1664-1692.

276 American Politics Research 45(2)

Banerjee, A., Green, D. P., McManus, J., & Pande, R. (2014). Are poor voters indif- ferent to whether elected leaders are criminal or corrupt? A vignette experiment in rural India. Political Communication, 31, 391-407.

Barbera, P., Fernandez-Vazquez, P., & Rivero, G. (2016). Rooting out corruption or rooting for corruption? the electoral consequences of corruption scandals in Spain, 2003-2011. Political Science Research and Methods, 4, 379-397.

Besley, T., & Burgess, R. (2002). The political economy of government responsive- ness: Theory and evidence from India. Quarterly Journal of Economics, 117, 1415-1451.

Blais, A., Gidengil, E., & Kilibarda, A. (2015). Partisanship, information, and per- ceptions of government corruption. International Journal of Public Opinion Research, doi: 10.1093/ijpor/edv037

Brown, L. M. (2006). Revisiting the character of congress: Scandals in the U.S. house of representatives, 1966-2002. Journal of Political Marketing, 5, 149-172.

Brunetti, A., & Weder, B. (2003). A free press is bad news for corruption. Journal of Public Economics, 87, 1801-1824.

Campbell, A., Converse, P. E., Miller, W. E., & Stokes, D. E. (1960). The American voter. New York, NY: John Wiley.

Chang, E. C. C., & Golden, M. A. (2006). Electoral systems, district magnitude and corruption. British Journal of Political Science, 37, 115-137.

Chang, E. C. C, Golden, M. A., & Hill, S. J. (2010). Legislative malfeasance and political accountability. World Politics, 62, 177-220.

Chong, A., De La O., A. L., Karlan, D., & Wantchekon, L. (2015). Does corruption information inspire the fight or quash the hope? A field experiment in Mexico on voter turnout, choice and party identification. Journal of Politics, 77, 55-71.

Congressional Quarterly. (1992). Congressional ethics: History, facts, and contro- versy. Washington, DC: Congressional Quarterly Press.

Congressional Quarterly Almanac. (n.d.). Retrieved from http://library.cqpress.com/ cqalmanac/

Converse, P. E. (1964). The nature of belief systems in mass publics. In D. E. Apter (Ed.), Ideology and discontent (pp. 205-278). New York, NY: Free Press.

De Vries, C. E., & Giger, N. (2014). Holding governments accountable? Individual heterogeneity in performance voting. European Journal of Political Research, 53, 345-362.

Delli Carpini, M. X., & Keeter, S. (1996). What Americans know about politics and why it matters? New Haven, CT: Yale University Press.

Dimock, M., & Jacobson, G. C. (1995). Checks and choices: The house bank scan- dal’s impact of voters in 1992. The Journal of Politics, 57, 1143-1159.

Doherty, D., Dowling, C. M., & Miller, M. G. (2011). Are financial or moral scandals worse? it depends. Political Science & Politics, 44, 749-757.

Duch, R. M., Palmer, H. D., & Anderson, C. J. (2000). Heterogeneity in perceptions of national economic conditions. American Journal of Political Science, 44, 635-652.

Eggers, A. C. (2014). Partisanship and electoral accountability: Evidence from the UK expenses scandal. Quarterly Journal of Political Science, 9, 441-472.

Klašnja 277

Ferraz, C., & Finan, F. S. (2008). Exposing corrupt politicians: The effects of brazil’s publicly released audits on electoral outcomes. Quarterly Journal of Economics, 123, 703-745.

Fiorina, M. P. (1981). Retrospective voting in American national elections. New Haven, CT: Yale University Press.

Fisman, R., & Gatti, R. (2002). Decentralization and corruption: Evidence across countries. Journal of Public Economics, 83, 325-345.

Gardiner, J. A. (2002). Defining corruption. In A. J. Heidenheimer & M. Johnston (Eds.), Political corruption: Concepts and contexts (pp. 25-39). New Brunswick, NJ: Transaction Publishers.

Groseclose, T., & Krehbiel, K. (1994). Golden parachutes, rubber checks, and stra- tegic retirements from the 102nd house. American Journal of Political Science, 38, 75-99.

Hall, R. L., & van Houweling, R. P. (1995). Avarice and ambition in congress: Representatives’ decisions to run or retire from the U.S. house. American Political Science Review, 89, 121-136.

Herrick, R. L. (2000). Who will survive? An exploration of factors contributing to the removal of unethical house members. American Politics Research, 28, 96-109.

Hirano, S., & Snyder, J. M. (2012). What happens to incumbents in scandals? Quarterly Journal of Political Science, 7, 447-456.

Jimenez, F., & Cainzos, M. (2006). How far and why do corruption scandals cost votes? In J. Garrard & J. L. Newell (Eds.), Scandals in past and contemporary politics (pp. 194-212). Manchester, UK: Manchester University Press.

Johnston, M. (1986). Right & wrong in American politics: Popular conceptions of corruption. Polity, 18, 367-391.

Klašnja, M., & Tucker, J. A. (2013). The economy, corruption, and the vote: Evidence from experiments in Sweden and Moldova. Electoral Studies, 32, 536-543.

Krause, G. A. (1997). Voters, information heterogeneity, and the dynamics of aggre- gate economic expectations. American Journal of Political Science, 65, 131-143.

Kuklinski, J. H., & Quirk, P. J. (2000). Reconsidering the rational public: Heuristics, cognition, and mass opinion. In A. Lupia, M. D. McCubbins, & S. L. Popkin (Eds.), Elements of reason: Understanding and expanding the limits of political rationality. New York, NY: Cambridge University Press.

Kunicova, J., & Rose-Ackerman, S. (2005). Electoral rules and constitutional struc- tures as constraints on corruption. British Journal of Political Science, 35, 573-606.

Lafay, J.-D., & Servais, M. (2000). The influence of political scandals on popularity and votes. In M. S. Lewis-Beck (Ed.), How France votes (pp. 189-205). New York, NY: Chatham House.

Lax, J. R., & Phillips, J. H. (2009). How should we estimate public opinion in the states? American Journal of Political Science, 53, 107-121.

Lupia, A. (1994). Shortcuts versus encyclopedias: Information and voting behavior in California insurance reform elections. American Political Science Review, 88, 63-76.

278 American Politics Research 45(2)

Lupia, A., & McCubbins, M. D. (1998). The democratic dilemma: Can citizens learn what they need to know? New York, NY: Cambridge University Press.

Luskin, R. C. (1987). Measuring political sophistication. American Journal of Political Science, 31, 856-899.

Luskin, R. C. (2002). From denial to extenuation (and finally beyond): Political sophistication and citizen performance. In J. H. Kuklinski (Ed.), Thinking about political psychology (pp. 281-305). New York, NY: Cambridge University Press.

Noyer, M. A. (1995). Catalog of congressional ethics cases, 1976-1992. Washington, DC: Brookings Institution.

Nyblade, B., & Reed, S. A. (2008). Who cheats? Who loots? Political competition and corruption in Japan. American Journal of Political Science, 52, 926-941.

Persson, T., Tabellini, G., & Trebbi, F. (2003). Electoral rules and corruption. Journal of the European Economic Association, 1, 958-989.

Peters, J. G., & Welch, S. (1980). The effects of charges of corruption on voting behavior in congressional elections. American Political Science Review, 74, 697-708.

Popkin, S. L. (1994). The reasoning voter: Communication and persuasion in presi- dential campaigns. Chicago, IL: University of Chicago Press.

Redlawsk, D. P., & McCann, J. A. (2005). Popular interpretations of corruption and their partisan consequences. Political Behavior, 27, 261-283.

Reed, S. A. (1999). Punishing corruption: The response of the Japanese electorate to scandals. In O. Feldman (Ed.), Political psychology in Japan: Behind the nails which sometimes stick out (and get hammered down) (pp. 131-148). Commack, NY: Nova Science.

Roberds, S. C. (1997). Sex, money, and deceit: Incumbent scandals in U.S. house and senate election, 1974-1990 (Doctoral thesis). Graduate School of the University of Missouri–St. Louis.

Rose-Ackerman, S. (1978). Corruption: A study in political economy. New York, NY: Academic Press.

Rubin, D. B. (1987). Multiple imputation for non-response in surveys. New York, NY: John Wiley.

Sekhon, J. S. (2006). The varying role of voter information across democratic societies. Working paper. Retrieved from http://sekhon.berkeley.edu/papers/ SekhonInformation.pdf

Sniderman, P. M., Brody, R. A., & Tetlock, P. E. (1991). Reasoning and choice: Explorations in political psychology. Cambridge, UK: Cambridge University Press.

Sniderman, P. M., Glaser, J. M., & Griffin, R. (1990). Information and electoral choice. In J. A. Ferejohn & J. H. Kuklinski (Eds.), Information and democratic processes (pp. 117-135). Urbana: University of Illinois Press.

Stollwerk, A. F. (2013, September). The application of multilevel regression and poststratification to cluster sampled polls: Challenges and suggestions. Paper Presented at the 2013 Annual Conference of the Midwest Political Science Association, Chicago, IL.

Klašnja 279

Tavits, M. (2007). Clarity of responsibility and corruption. American Journal of Political Science, 51, 218-229.

Treisman, D. (2000). The causes of corruption: A cross-national study. Journal of Public Economics, 76, 399-457.

Tverdova, Y. V. (2011). See no evil: Heterogeneity in public perceptions of corrup- tion. Canadian Journal of Political Science, 44, 1-25.

Warshaw, C., & Rodden, J. (2012). How should we measure district-level public opinion on individual issues? Journal of Politics, 74, 203-219.

Welch, S., & Hibbing, J. R. (1997). The effects of charges of corruption on vot- ing behavior in congressional elections, 1982-1990. The Journal of Politics, 59, 226-239.

Wright, G. C. (1993). Measuring vote choice in the national election studies, 1952-88. American Journal of Political Science, 37, 291-316.

Xezonakis, G., Kosmidis, S., & Dahlberg, S. (2015). Can elections combat corrup- tion? Accountability and partisanship. In C. Dahlström & L. Wängnerud (Eds.), Elites, institutions and the quality of government (pp. 283-304). New York, NY: Palgrave McMillan.

Zaller, J. R. (1992). The nature and origins of mass opinion. New York, NY: Cambridge University Press.

Zaller, J. R. (1998). Monica Lewinsky’s contribution to political science. Political Science & Politics, 31, 182-189.

Zaller, J. R. (2004). Floating voter in U.S. presidential elections, 1948-2000. In W. E. Saris & P. M. Sniderman (Eds.), Studies in public opinion: Attitudes, nonat- titudes, measurement error, and change (pp. 166-212). Princeton, NJ: Princeton University Press.

Zechmeister, E. J., & Zizumbo-Colunga, D. (2013). The varying political toll of con- cerns about corruption in good versus bad economic times. Comparative Political Studies, 46, 1190-1218.

Author Biography

Marko Klašnja is an assistant professor of political science at Georgetown University, with a joint appointment in the Edmund A. Walsh School of Foreign Service and the Government Department. He speacilizes in comparative politics, political behavior, and political economy of democratic accountability.