Abstract
When do scandals hurt candidates? In this study, we employ a candidate conjoint experiment to understand the conditions under which candidates are most penalized for scandals. Survey respondents evaluate candidates in election scenarios that vary candidate partisanship, the presence and type of negative news about the candidate, and the amount of other information available to voters about the candidates, such as issue positions and demographics. Averaging across respondents and the distribution of candidate attributes, the results show scandal decreases the probability of voting for a candidate, but the size of this negative effect varies by context. The negative effects of scandal on voting decisions are mitigated by the amount of other information available to voters. Our findings also reveal that it is not always the case that voters are blind to the moral failings of scandal-plagued candidates in the presence of other information; voters may continue to rate scandal-plagued candidates negatively in terms of morality but prefer these candidates in terms of partisanship and shared political views.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
Scandal is nothing new in the American political landscape. From a new country’s early infidelity scandal, the Hamilton–Reynolds affair, to accusations about Hillary Clinton’s email usage and Donald Trump’s treatment of women, scandals abound, though the repercussions of each greatly vary. Politicians like Ted Kennedy and John McCain not only survived their respective scandals, they became some of the Senate’s most prominent figures. Yet, Anthony Weiner fell from grace, and it took two centuries and a musical for Alexander Hamilton’s reputation to improve. We examine under what conditions scandal influences candidate evaluations and vote choice, and particularly how the information environment may augment or curtail the effects of scandal. We conduct an online survey experiment in which respondents evaluate and choose between candidate pairs in hypothetical Congressional election scenarios that vary candidate partisanship, the presence and type of candidate scandal, and the amount of other information available to voters, such as candidate issue positions and demographics.
We find support for the hypothesis that as the information environment expands, scandals have less of an impact on voting decisions. In low information environments where voters only know about a candidate’s partisanship, scandal decreases the likelihood a person votes for the scandal-plagued candidate by 12–26 percentage points, on average, relative to if the voter had no additional news about the candidate. However, the introduction of other information about a candidate significantly reduces the effect of scandal on vote choice. We suggest that this is not because voters are less likely to receive information about scandal in a high information environment, as scandal remains prominent as an attribute in the experiment’s visual presentation, but rather, that voters often prioritize other information in saturated information environments.
Our findings have several implications. First, a great deal of research has focused on examining which types of scandals are most damaging for candidates (e.g., Carlson et al. 2000; Funk 1996; Doherty et al. 2011; Basinger 2013, 2018; Von Sikorski 2018). Our findings emphasize the importance of context in shaping the degree to which the scandal may have an influence on voting decisions. We also find that while partisanship matters for voting decisions across all information environments, partisans punish both copartisan and opposing party candidates for scandals. Overall, partisans show a significant preference—but not blind loyalty—to candidates of their party. Finally, our results help illuminate the reasons why people vote for scandal-plagued candidates. We argue that it is not that voters do not care about scandal, so much as voters often consider other criteria—when available—in arriving at their voting decisions.
When Do Scandals Matter?
Scandals undoubtedly impact voter perceptions of candidates, almost always negatively (Rundquist et al. 1977; Peters and Welch 1980; Abramowitz 1991; Funk 1996; Welch and Hibbing 1997; Brown 2006; Cobb and Taylor 2014; Von Sikorski 2018). In a study of post-Watergate Congressional scandals, Basinger (2013) finds that scandal-tainted incumbents lose five percent of their vote share, on average, in the next election, and Dimock and Jacobson (1995) similarly conclude that the House bank scandal reduced the 1992 incumbent vote by 5 percentage points. Our first hypothesis is, thus, a restatement of findings in existing workFootnote 1:
Scandal Hypothesis: On average, voters will be less likely to vote for a candidate when they learn about a scandal involving the candidate.
However, existing work also suggests not all scandals are created equal, and not all scandal-plagued candidates suffer equal consequences. Often, scandals are far from “dealbreakers.” Basinger (2013) finds 59% of scandal-tainted incumbents returned to Congress the following term, and, in subsequent elections, scandal-plagued Members of Congress tend to recover what vote share they lost (Pereira and Waterbury 2018; Praino et al. 2013).
Our study builds on a set of literature that argues that the effects of scandal depend on context. For example, previous scholars have shown time can mitigate a scandal’s negative consequences (Alford et al. 1994; Praino et al. 2013), chiefly when it comes to sex scandals (Doherty et al. 2014; Basinger 2018).Footnote 2 More generally, type of scandal also matters,Footnote 3 with scandals that demonstrate hypocrisy (McDermott et al. 2015; Bhatti et al. 2013) or violate a salient societal norm (Barnes et al. 2020) punished more severely.
Variation in the Information Environment
In this study, we focus on variation in the information environment as an important contextual variable, which aligns with previous studies that point to the roles of media and information as moderating the effects of scandal. In particular, the congestion of the existing news agenda shapes the likelihood that a scandal emerges in media coverage and how much coverage it receives (Nyhan 2015, 2017), which influences the scandal’s importance in voters’ minds (Iyengar and Kinder 1987). Scandals with greater media visibility are more detrimental to politicians’ electoral chances than lower-profile scandals (Hamel and Miller 2019).
Likewise, whether voters have pre-existing information and beliefs about political figures may also play a role. At the respondent level, Dimock and Jacobson (1995) note that subjects with preconceptions about the elected officials involved in the House Bank scandal reacted differently than respondents with weaker priors. Additionally, Klašnja et al. (2020) points to political awareness in shaping how much voters will penalize corruption. Green et al. (2018) show that the negative impact of scandal on favorability of political figures is most pronounced for politicians who are relatively less known.
In our study, we build on this work to understand how contextual factors—and specifically, the information environment—may exacerbate or mitigate the effects of scandal on candidate evaluations and voting decisions. We test how the quantity of information a person has about candidates shapes the influence of scandal.Footnote 4
Zaller (1992) suggests that individuals reach evaluations by sampling and averaging over considerations available to them.Footnote 5 In our experiment, we exogenously vary the quantity of considerations that respondents receive for two candidates in an election, therefore generating variation between respondents in the information flows they receive, rather than relying on pre-existing differences in voters’ chronic awareness. Some elections in the United States, such as presidential elections, are very high information environments, where voters are bombarded with information about the candidates, of which, a scandal may be one consideration. In contrast, lower-level elections have much more sparse information about the candidates, allowing scandal to play a larger role. For example, Green et al. (2018) find that subjects were more likely to be familiar with a governor and speaker of the state house in their studies relative to state-level representatives and local businessmen.
We work from a starting place that in most state and federal elections in the United States, voters know at least the partisan identification of the candidates. However, what other information voters know is highly variable across elections, and in some cases, partisanship may be among the only candidate attributes salient to voters. These low information cases, where partisanship is the primary attribute, share two features: First, voters are only provided with one basis for decision making. If partisanship is the only consideration available, we would expect this to guide voters’ decisions.
Second, voters may still have considerable uncertainty about whether their party’s candidate has their preferred attributes because they have very few “considerations” upon which to make their evaluations (Zaller 1992). In line with Green et al. (2018), we argue that when voters have little information about a political figure, any new information could sway opinions. In a low information environment, the introduction of negative news about a candidate’s scandal ought to uniformly negatively influence voters’ evaluations because voters likely not only receive the information about scandal but also sample and weigh it in their decisions. In real world elections, this type of “low information” environment could exist either because of sparse information available about a lower-level political figure (such as in a state legislative race) or because media coverage on a scandal has dominated and potentially crowded out other information that could otherwise be made available.Footnote 6
As the information environment expands to include additional contextual information about the candidates in the election, information about scandal still provides a uniformly negative signal about the candidate. However, scandal may be less likely to influence decision making in a high information environment. As only one consideration among many in a saturated information environment, voters may no longer sample and weigh the consideration in decision making, substituting in other information. In complex information environments, voters do not—and often, cannot—weigh all information available (Lau and Redlawsk 2001). Even if voters do weigh scandal as a consideration, other more positive considerations may still offset the negative effects of scandal. In summary, the information hypothesis is:
Information Hypothesis: In high information environments, where voters know a great amount of information about the candidates, we expect that negative information about a scandal will matter less in voters’ decisions than in low information environments where they typically only know partisanship of the candidate.
In real world election scenarios, the high information environment could resemble a context where voters are already very familiar with the candidates or cases where a scandal receives relatively less media coverage compared to other information, allowing voters the possibility to receive and attend to other information about candidates.
This information hypothesis may fail to be supported in two cases. If scandal proves to be a perpetually trivial consideration, voters may ignore information about scandal even in relatively low information environments. On the other hand, if scandal is always a salient consideration in voters’ minds, scandal may matter similarly across information environments. For example, in an August 2015 Gallup poll that asked voters what they had recently heard about Hillary Clinton, more than 40 percent of U.S. adult respondents mentioned something related to “emails” or “email scandal” involving Clinton’s use of private email while serving in government (Saad and Newport 2015). Clinton’s policy stances or other substantive positions were rarely mentioned. If voters become fixated on scandal, it may influence decisions even in a high information environment.
Heterogeneity in Candidate Evaluations
We also anticipate that the effects of scandal may vary by whether voters share partisanship with the scandal-plagued candidate. Even though our experiment will exogenously fix the amount of information provided to voters, as Zaller (1992) suggests, what voters do with that information might vary based on their predispositions. There are multiple ways partisanship may condition the effects of scandal. On one end, voters might process new information without partisan bias. If the way partisans incorporate information about scandal into their evaluations is not influenced by their pre-existing partisan preferences, we would expect information about scandal may have a negative effect for both partisans evaluating a copartisan and an out-partisan, but potentially an even bigger negative effect for copartisans. This is because for partisans evaluating a political opponent, information about scandal serves to only reinforce their baseline partisan preferences. For partisans evaluating a copartisan, information about scandal runs against their baseline preferences and may alter their voting decisions.
Alternatively, partisanship may shape whether voters resist or accept information about scandal as a consideration (Zaller 1992). Following the theory of motivated reasoning (Kunda 1990), Taber and Lodge (2006) suggest that people will be more skeptical of information that goes against their prior attitudes. Partisans may be less likely to believe—and subsequently accept—information about scandal related to their copartisan candidates (Dimock and Jacobson 1995). If this occurs, we could expect that scandal would generally not matter for partisans evaluating copartisan candidates.Footnote 7
Even if voters do accept information about scandal as a consideration when evaluating copartisans, they may still weigh these considerations differentially, according to their predispositions. For example, Klašnja et al. (2020) finds high-awareness voters are less likely to vote for a corrupt politician than low-awareness voters, but partisanship weakens this effect. Partisans could conceivably downweight information about a scandal related to their copartisans and upweight information about scandal about out-partisans. In this case, partisans become more negative in their evaluations of copartisan (and opposing party) candidates, but do not punish copartisans to the degree expected from fully unbiased learning (Bartels 2002).
The existing literature on scandal provides some evidence consistent with partisan motivated reasoning. For example, Walter and Redlawsk (2019) find that partisans are more likely to have negative emotional reactions to scandals where an out-partisan commits moral transgressions. Fischle (2000) shows that Clinton detractors were more likely to believe that the charges against Clinton were important and credible, and Vonnahme (2014) finds that the effects of scandal decay to a greater degree among pre-existing supporters. However, Green et al. (2018) find that both Democrats and Republicans update their beliefs about the same partisan political figures to be more negative in response to negative coverage on scandal. Following the literature that points to a stronger role for partisan biases, we anticipate a degree of partisan resistance:
Partisanship Hypothesis: On average, copartisans will penalize candidates from the opposing party more severely for scandals than partisans of their own party.
It is possible that the information environment might also influence how much partisanship influences voting decisions about candidates. While we do not offer a hypothesis on the interaction, we can examine the role of partisanship in both low and high information environments in our experimental study.
Effect of Scandal on Candidate Evaluations
A final choice we make is to consider both vote choice and other dimensions of candidate evaluation. Most scandal-tainted Congressional candidates returned to Congress in Basinger’s (2013) study of post-Watergate scandals, meaning that a majority of scandal-tainted candidates won their elections. The question still remains: Do voters that vote for scandal-plagued candidates ignore scandals or do they “hold their nose” and vote for scandal-plagued candidates despite their scandals? When we only observe binary voting decisions, it can be unclear how heavily scandals weigh in voters’ minds—and to what extent scandals permeate all dimensions of candidate evaluation.
In our study, we measure both voters’ evaluations of a candidate’s morality, as well as whether voters think the candidate shares their views. Because scandal most directly provides information about a candidate’s integrity/morality, we anticipate that offending candidates could receive a greater penalty in terms of morality than other dimensions. This expectation is consistent with Basinger (2018) who finds scandals especially harm impressions of House incumbents’ integrity.
However, beyond morality, as more considerations relevant to other dimensions of candidate evaluation (e.g., policy information) enter the information environment in high-information settings, voters may come to weigh this new information more heavily in their vote choice. The type of information at voters’ disposal may shape the likelihood that a voter is willing to cast a ballot for a scandal-plagued candidate. Rundquist, Strom, and Peters argue that even a perfectly rational voter may choose a “corrupt” candidate if the voter “perceives both candidates as corrupt, or if he decides that a corrupt candidate who is closer to his own preferences on other issues is preferable” (1977, p. 956). In an experiment, the authors find that knowledge of a candidate’s opinion on President Nixon’s Vietnam policy made it especially more likely that a voter would stick with a corrupt fictional Congressional candidate. Likewise, in an analysis of opinion toward President Bill Clinton after the Monica Lewinsky scandal, Zaller (1998) also argued that the public weighed political substance more heavily than scandal.
Still, it is not a given that voters can separate how they evaluate a candidate across different dimensions, sampling and weighing different sets of considerations.Footnote 8 Instead, voters may feel pressure to evaluate their preferred candidate consistently on all dimensions (Kunda 1990) or subconsciously allow information about one dimension of candidate evaluation to influence other dimensions (Rapoport et al. 1989). That said, Doherty et al. (2011) show that voters can separate evaluations of a political figure as a person from the figure’s performance as a representative. Similarly, McCurley and Mondak (1995) show that evaluations of candidate competence and integrity are not always correlated. We test which dimensions of candidate evaluation are most influenced by scandal to understand the full range of consequences for candidates across information environments.
Experimental Design
We conducted an online survey experiment in August 2018 through Lucid. Lucid partners with survey recruitment firms to provide near-representative non-probability samples of the national adult population of the United States. Studies using Lucid have replicated several of the effects of major social science experiments conducted on other survey platforms (Coppock and McClellan 2019). The sample includes 2135 respondents—with 967 Democrats/leaners, 857 Republicans/leaners, and 311 pure independents or unaffiliated voters. Online Appendix Table A2 provides descriptive statistics on the sample.
The experiment was a conjoint design in which respondents choose between hypothetical candidate pairs that vary along a number of different attributes, the levels of which are randomly assigned. Conjoint designs allow the researcher to examine how multiple factors influence voters’ decisions by allowing those factors to experimentally vary simultaneously (Hainmueller et al. 2014). Scholars have previously used conjoint designs to study voting preferences (e.g., Peterson 2017; Franchino and Zucchini 2015). Our design builds on Peterson (2017), in which the author assessed how the effect of partisanship varies across information environments. In this study, we focus on how the effect of scandal changes across information environments.
The conjoint experimental module was part of a larger survey. Prior to entering the treatment conditions, respondents answered demographic and attitudinal questions, including their age, sex, income, and party identification. In the conjoint experiment, respondents were presented with tables that compare two Congressional candidates (Candidate A and Candidate B) and were asked about their vote choice and perceptions of the candidates. Respondents completed three of these tasks, each of which compare two candidate profiles, for a total of six candidate profiles per respondent, and an overall number of 12,810 candidate profiles.Footnote 9
Scandal Manipulation
Each election task contained a “news” attribute with a piece of information about each candidate. Previous research has found that the effects of scandal can fluctuate by type, so we included multiple variations in the design to make sure our effects are not specific to any single situation. This attribute included six values, or “levels”, three of which describe a negative scandal involving the candidate: Recently accused of sexual harassment, Recently accused of cheating on spouse, or Recently accused of leaking confidential information. Alternatively, candidates could have the more neutral or positive news attribute levels: No recent news, Recently honored for public service, or Recently celebrated wedding anniversary.Footnote 10 These levels allow us to compare the effect of specific scandals to each other and to alternative types of candidate news. Each level of the news attribute had an equal probability of appearing in every candidate profile. Levels were independently randomly assigned between candidate profiles. This means that in some tasks, Candidate A and B may have the same news attribute levels just by chance. The randomization was independent across tasks and respondents.
Information Manipulation
For each task, respondents were also independently randomly assigned to a low, medium, or high information environment. The low information environment only included two candidate attributes: party and news. Candidates were constrained to be of different political parties in each task—one candidate was a Democrat, one a Republican. The party attribute always appeared first, and the news attribute was always second in the table.Footnote 11
In the medium information environment, respondents saw five attributes. The first two attributes remained candidate party, followed by candidate news. The next three were a randomly chosen subset of additional candidate attributes (out of eight possible additional attributes).Footnote 12 In the high information environment, respondents were exposed to all ten possible candidate attributes. Candidate party and news were again the first two attributes. The order of the remaining eight attributes was randomized across respondents. The eight non-party and non-news attributes included candidate gender, race, profession, religion, age, abortion stance, government spending stance, and immigration stance. Within each of these attributes, the specific levels were assigned at random for each candidate profile, according to pre-determined probabilities.Footnote 13 (See Online Appendix Table A1 for all attribute levels and probabilities of assignment. No attribute levels were restricted from appearing together.)
We deliberately kept the candidate party and news attributes in the same location across information environments to help isolate what is varying. This decision helps reduce the possibility that the reason the effect of scandal varies across information environments is because respondents are less likely to see the news attribute. We keep the news attribute toward the top of the table to mimic the possibility that in a real-world election, a scandal would often be among the more prominent pieces of information available to voters. Table 1 depicts what a conjoint table looked like for respondents in the experiment.Footnote 14 For each task, respondents were asked to read the descriptions of two candidates, who are “similar to those who are running for election” in the Congressional midterms and then exposed to a low, medium, or high information table.
Dependent Variables
After each task, respondents were asked, “If this election were being held in your district, would you vote for [Options: Candidate A or Candidate B].” For each candidate profile, we code this primary outcome so that 1 = voted for the candidate, and 0 = did not vote for the candidate. In addition, respondents were asked, “Please indicate how much you agree with the statements about Candidate [A/B]: Candidate [A/B] shares my views; Candidate [A/B] has good morals; and Candidate [A/B] cares about people like me.” Respondents rated each candidate on a five-point scale: “strongly disagree,” “somewhat disagree,” “neither agree nor disagree,” “somewhat agree,” “strongly agree,” coded 0–1 for analysis.
Results
Our primary interest is how scandal influences voting behavior across different information environments. We regress whether the candidate was chosen (Vote choice) in the election on the interaction between the news attribute and an indicator for the information environment (Low, Medium, or High). The reference levels are “No recent news” for the news attribute and “Low Information” for the information variable. We follow the guidance of Hainmueller et al. (2014) and use a linear regression with standard errors clustered at the respondent level, using the estimatr R package (Blair et al. 2018).Footnote 15
Figure 1 displays the average marginal component effects (AMCE) separately by information condition with 95% confidence intervals.Footnote 16 This represents the average change in vote share expected when changing the news attribute from No recent news to a particular type of news, marginalizing over the distribution of candidate and opponent attributes. Our results support the Scandal Hypothesis. Scandalous news leads voters to be less likely to vote for a candidate. In the low information environment, all three scandals have a significantly negative effect on vote choice relative to the No recent news condition. When voters only know a candidate’s party and a piece of negative news about the candidate, they tend to punish candidates for scandals.
However, the results also suggest that the magnitude of the effects of scandal change as the information environment changes. Relative to the low information environment, on average, scandal has less of a negative impact as the information environment expands. Table 2 column 1 displays the full coefficient results from the regression model, providing a formal statistical test of how information moderates the effects of scandal. The significant interaction terms in Table 2 column 1 between the scandal conditions and the medium and high information environments reveal that the effect of negative news on voting (relative to No recent news) grows weaker with information. The positive sign of the interaction terms can be interpreted as a reduction in the baseline negative impact of scandal from the low information condition. The average marginal effect of a candidate having been accused of sexual harassment relative to a candidate with no recent news declines about 17 percentage points from the low to high information environment.
Overall, our results find support for the Information Hypothesis, that scandals affect voting decisions to a lesser degree in more saturated information environments. Table 2 column 2 presents the contrast in the effects of scandal between the medium and high information conditions only. There is a smaller and non-significant change (see the null interaction effects) in the effects of scandal moving between the medium and high conditions. The 10-attribute condition is not sufficient to fully eradicate the negative average marginal effects, though we discuss below how specific estimates may be sensitive to the distribution of election scenarios.Footnote 17
It is worth recalling that the news attribute was always the second attribute in the candidate profiles. By fixing the news attribute as second in the conjoint table, we can be more confident that, even in high information environments, respondents were likely to see the information about scandals (news). Due to this design choice, the diminished effects of scandal in higher information environments likely have more to do with how much weight respondents give to positive and negative news, and not the likelihood the respondents received the information.Footnote 18 Had we varied the position of the news attribute, therefore making it less salient, we suspect the effects of scandal in a high information environment might be even lower, making it easier to detect declines in scandal’s effect across information environments.
Evaluating Alternative Election Distributions
One weakness of the conjoint experiment is that due to the random assignment of candidate attribute levels, a respondent may see an election scenario that is uncommon in real world settings. In particular, a respondent may see a scenario where partisan candidates have one or more counterstereotypical positions on political issues. For example, in the medium and high information environments, a Democratic candidate may show a pro-life abortion stance. While this certainly occurs to a degree in real elections, the distribution of candidate attribute combinations within the experiment is likely not always representative of the distribution voters will see in the real world.
To focus on cases where candidates take positions more programmatically aligned with their party, we conduct an analysis that eliminates election scenarios where either one of the candidates holds a counterstereotypical position. We focus on the low and medium information environments, where we still have a substantial amount of candidate cases to detect significant effects—4250 candidate profiles in the low and 1856 cases in the medium information environment. Figure A2 (top) describes this analysis and displays the results. Consistent with the full sample, the AMCEs of scandal in the medium information environment remain negative (though no longer significant except in the case of sexual harassment) and smaller than in the low information environment. This gives confidence that the presence of counterstereotypical candidates is not driving the pattern of results and provides an additional useful estimate of the effects in a potentially theoretically interesting subset of election scenarios.
In addition, in approximately 25% of election scenarios in the study, both candidates have some type of negative news. We view this as realistic. One need only look back as far as the 2016 presidential election to find a case where one candidate had been accused of sexual harassment and infidelity, while the other had been accused of sharing confidential information. However, there are, of course, many cases in the real world where only one candidate is plagued by a scandal. In Online Appendix Figure A2 (bottom), we analyze the subset of tasks in which one candidate was accused of scandal, and one was not. The effects of scandal are larger in this subset, but the decline in the effects of scandal when moving from the low to medium or high information environments remains.
Partisanship and Scandals
The average effects reported in Fig. 1 may mask heterogeneity. In particular, we hypothesized shared partisanship between respondents and candidates may influence how much a candidate is penalized for scandal. Figure 2 displays the AMCE of voting for a candidate that is a copartisan of the respondent versus a candidate that is from the opposing party for the low, medium, and high information environments, separately for Republican and Democratic respondents (including leaners). Each figure displays the results of a separate regression analysis. Full regression results are in Online Appendix Table A4.
Across all partisan voter-candidate combinations, in low information environments, partisans are on average less likely to vote for both the copartisan and the opposing party’s candidate who is accused of wrongdoing relative to candidates with no recent news. Thus, partisanship does not entirely blind voters to scandal. Going against the Partisanship Hypothesis, partisans do not punish the opposing party for negative news (relative to no recent news) significantly more so than they punish copartisans in the low information environments.Footnote 19
As the information environment goes from low to medium and high, similar to the overall results, we see the effects of scandal somewhat dissipate similarly across figures. This is especially true for Democratic respondents. Whether evaluating a copartisan or opposing party candidate, moving from the low to high information conditions generally significantly reduces the negative impact of scandals. The evidence for information effects for Republican respondents is somewhat weaker, though the effect of scandal relative to the No recent news level does still decline with increasing information, though not significantly so (Online Appendix Table A4).Footnote 20
In the Appendix, we further examine the role of partisanship by taking strength of respondent partisanship into account. Across all seven partisanship levels (from strong Democrat to strong Republican), voters prefer in-party and opposing candidates with neutral or positive news to candidates with negative news (Online Appendix Figure A4). Even strong partisans are not fully “resistant” to information that goes against their partisan predispositions. Stronger partisans appear to incorporate information about scandal, at least to some extent. In addition, comparing the low and high information environments, the gaps between the probability of voting for candidates with positive and negative news become somewhat smaller, on average.
The results still do not diminish the role of partisanship in voting. Pooling across scenarios in the low information environments, more than 75% of the time, Democrats and Republicans voted for the copartisan candidate. In addition, scandals do not appear to consistently matter significantly more in evaluations of copartisans, going against one interpretation of unbiased information processing discussed in the theoretical section. Overall, we emphasize more limitedly, that we also find evidence against the possibility that partisans are entirely resistant to information about scandals that involve their own party. Partisans are less likely to vote for copartisans or opposing party candidates with negative news about scandal, on average, relative to candidates that have no recent news or positive news.Footnote 21 As the information environment expands, the AMCE of scandal also dissipates in both cases.
Effects of Scandal on Candidate Evaluation
In the last section of the analysis, we seek to better contextualize the findings to understand what may lead voters to continue to vote for scandal-plagued candidates. We conduct three supplemental analyses: We initially examine the effect of scandal on candidate morality, a dimension of candidate evaluation closely related to scandal. Second, we compare the effects of scandal on candidate morality to the effects on a dimension that is less related to scandal. Lastly, we show how the presence of shared policy views influences voting decisions.
We first repeat central analysis of the interaction of scandal and information on the outcome of how strongly the respondent agrees or disagrees that the candidate has good morals (Online Appendix Figure A10 displays the distribution responses). Figure 3 displays the AMCE of each type of news on the perceived morality of the candidate. Like vote choice, in low information environments, scandals have a significant negative effect on morality ratings. In addition, like vote choice, more information reduces the negative impact of scandal on morality ratings. The gap in morality ratings between candidates with news about scandal vs. no recent news becomes slightly smaller in medium and environments and grows smaller, again, in the high information environment (Online Appendix Table A6). Still, for this evaluative dimension, the gap remains large—more than 12 percentage points for each scandal in high information environments. We find similar results by partisanship (Online Appendix Figure A7).
At the same time, the effects of scandal do not influence all dimensions of candidate evaluation equally. Figure 4 shows that for each type of scandal and information environment, the average effect of a scandal on morality ratings is significantly larger (more negative) than the effect of the scandal on the average belief the candidate “shares my views.”Footnote 22 For each type of scandal, the figure compares the AMCE of scandal on average ratings for morality to the AMCE of a scandal on average ratings for the outcome “shares my views.” All estimates are negative, which means that candidates suffer a greater penalty from a scandal in their morality ratings than on perceptions of their views. This is similar to Basinger (2018), which distinguishes perceived candidate integrity from competence.
These results show that respondents have some ability to separate their perceptions of the candidate's views—an alternative basis of support for the candidate—from candidate morality. We, lastly, directly examine the role of shared policy information with candidates on voting.Footnote 23 We evaluate the AMCEs of each candidate attribute on vote choice in the high information environment, where all attributes are always shown to respondents (Figure A8).Footnote 24 In Figure A9, we similarly analyze the effects of each attribute, but coding candidate party and issue positions based on whether or not they are shared with the candidate in order to quantify the effects of sharing partisanship, as well as sharing particular issue positions with the candidate.Footnote 25 These figures provide a way to benchmark the size of the effects of scandal against other information in this more saturated information environment, where the effects of scandal are relatively smaller. In particular, here, the negative effects of scandal are smaller than the positive effect of sharing partisanship or an abortion position with a candidate and similar to sharing an immigration position (Figure A9). When provided, policy information matters.
Conclusion
Are scandals ever “dealbreakers” for voters? Or, instead, do voters appear to ignore scandals or “hold their nose” and vote for scandal-plagued candidates despite their scandals? Our answer is it depends. Scandals do matter, but so does the context in which they occur. We find support for our hypothesis that scandals negatively influence a candidate’s electoral chances, but this effect declines in more saturated information environments. This does not mean, however, that people turn a blind eye to impropriety or misconduct in crowded information environments. Even in high information environments, scandals have a small average negative effect on voting decisions and influence perceptions of candidate morality.
Our effects, based on a controlled experimental setting, are somewhat larger than those in existing work that employ alternative methods to study actual candidates, but show similar patterns. For example, using observational data with a matching analysis, Hamel and Miller (2019) study the effect of variation in the visibility of scandals on the extent to which voters penalize candidates in U.S. House elections between 1980 and 2010. The low information environment in our study may reflect a high-visibility situation where, we argue, coverage of a scandal may crowd out other information about the candidates. In contrast, in our experiment’s high information environment, scandal is less dominant, more closely resembling Hamel and Miller’s low-visibility scenario. Hamel and Miller find that nationally visible scandals result in a significant average decrease in vote share of about 8 percentage points (2019, p. 123) relative to a < 5 percentage point decline for low-visibility scandals. In the experiment, we find a slightly larger expected decline in the predicted probability of voting for a candidate of between 12 and 26 percentage points in low information settings. Our pattern of results is similar. Moving from the low to high information condition, like moving from a more to less visible scandal, we find a significant decline in the average effect of scandal.
Our pattern of results also resembles those in Green et al. (2018), who use a set of field experiments to study the effect of negative newspaper coverage on the favorability of political figures. The authors study the effect of negative news coverage on both lesser-known and more well-known political figures, an alternative way to resemble the low information and higher information conditions in real elections—situations where voters are likely to have either little or a lot of pre-existing information about candidates (beyond knowledge of their scandals). In Study 4 in Green et al. (2018), regarding a school reform scandal, negative coverage resulted in an 8.7 percentage point increase in the likelihood of viewing a relatively more well-known figure (state speaker) unfavorably compared to a 13.5 and 16.5 percentage point increase for lesser-known figures, and Fig. 3 further shows a more general association between prior uncertainty and size of the treatment effect across scandals and political figures (2018, p. 255). Our experimental results provide a more controlled test relative to observational and field experiment studies to help isolate the effect of information on voting decisions. Despite this difference, our similar patterns of results give confidence in the external validity of the findings.
The experimental results show mixed evidence for the role of partisanship. Partisanship does not make respondents completely insensitive to scandal. Partisan voters, on average, punish both copartisans and out-party candidates for scandals. On the other hand, shared partisanship is a significant predictor of vote choice across all information environments and types of news, and in the real world, there may be even more room for partisan biases to enter because the framing of information may be partisan. Likewise, unlike the real world, our study cannot directly speak to the roles of other contextual or candidate information not included in the attributes, such as name recognition, incumbency status, or time between the infraction and election.
Overall, if our design included only dense information environments, we might wrongly conclude that scandals never have a large influence on voting decisions. Instead, this study’s design shows that the effect of scandal is context dependent. Scandals matter, but to a lesser degree in high information settings. Does this mean candidates should respond to accusations of impropriety by listing their attributes and achievements rather than addressing the scandal? Not necessarily, but the sheer volume of information—especially substantive policy information—in certain elections may help explain the limited impacts of scandal on voting decisions. Second, even in high information environments, scandal-plagued candidates receive significantly more negative morality ratings than candidates without a scandal. This has potentially important long-run implications given existing research (Paschall et al. 2019). With lower approval on certain dimensions, elected officials may have more difficulty achieving their policy and professional goals.
Notes
We did not formally pre-register our hypotheses. As a result, we keep our analyses to a limited set of tests focused on those most central to the study. Replication data available at https://doi.org/10.7910/DVN/0IN4PX.
Doherty et al. (2014) find that the passage of time does less to mitigate the negative consequences of scandals when it comes to scandals involving tax evasion.
Our study differs from Green et al. (2018) in two ways that provide advantages in internal validity that complement the field experiment: In Green et al. (2018), the evaluators or political figure being evaluated varied as uncertainty varied, while our experiment holds constant the figure being evaluated (Congressional candidate). Second, instead of relying on respondents’ self-reports to quantify uncertainty, we exogenously manipulate the quantity of information.
Zaller defines a consideration as, “any reason that might induce an individual to decide a political issue one way or the other” (1992, p. 40).
This could occur when a scandal generates high levels of national attention, similar to Hamel and Miller (2019). For example, when the Associated Press reported 2020 Congressional candidate Cal Cunningham’s extramarital affair, his campaign tried to pivot media attention to his policy positions, but soon retreated from interviews and public events except for defensive statements regarding the scandal (Robertson 2020).
One thing that could curtail partisan motivated reasoning is a divide within a partisan’s own party (Rundquist et al. 1977).
This may be further influenced by the nature of the scandal and reporting of it. For example, when the Pittsburgh Gazette broke the story of Republican Tim Murphy urging his “mistress” to have an abortion, the scandal was directly tied to his abortion policy positions (Ward 2017). In contrast, scandals are often reported with little substantive information about the candidate’s record, potentially helping to keep separate evaluations of candidate integrity and morality. For example, when Democratic Congressman Anthony Weiner admitted to tweeting a photo of his genitalia, The Washington Post report noted his partisanship “Weiner (D),” and focused on the potential ethical and legal issues of the scandal (Horowitz 2011).
Bansak et al. study how the number of tasks (up to 30) influences quality of responses and find that, “satisficing is not a serious concern that should dictate the number of tasks” (2018, p. 118).
We chose scandals that resemble the types of wrongdoing for which high profile political figures, such as Hillary Clinton and Donald Trump, had recently been accused. The effects could vary if the candidate was proven to have committed the wrongdoing. Likewise, we might also anticipate effects of scandal to vary if the vignette indicated if the wrongdoing involved abuse of power (Doherty et al. 2011).
The low-information setting has at least two parallels to real elections. It may reflect situations where scandal has dominated the popular discourse, making it harder to learn other information about a candidate. It may also approximate situations where scandal is the only salient information about a lower-profile candidate (beyond party).
In low information cases in the real world, voters may still be able to infer other information about candidates just from their name, such as gender. However, gender does not have a significant marginal effect when present in the experiment.
Bansak et al. (2019) find that response quality does not suffer from “excessive satisficing” even in an application with 18 attributes.
We present information in a tabular format to make it easier for respondents to process information. This format performed well in a validation study on immigration preferences (Hainmueller et al. 2015).
The full regression equation is in pg. 4 of the Appendix. We use linear regression with cluster robust standard errors. Hainmueller et al. show the consistency of linear regression and note linear regression represents a “convenient procedure for applied researchers” (2014, p. 15). Results are similar with logistic regression (Online Appendix Table A3).
One concern about having respondents represented in the data multiple times could be that the first task may influence respondents’ answers on subsequent tasks (Hainmueller et al. 2014). Results are similar for only the first task respondents completed (Online Appendix Figure A1).
We chose ten total attributes for the high information environment to match the amount of information presented in the highest information environment from Peterson (2017).
An alternative design could have randomized where “News” appeared in the table. However, in that design, because the position of the news attribute and the amount of information would be varying, we might see the effect of scandal dissipate either because voters weigh scandal less heavily, or instead, because they do not see the information about scandal.
Supplemental analyses that collapse Republicans and Democrats or types of scandals in the analysis are in Online Figure A3 and Table A5. We do not detect significant differences in effects of scandal by copartisanship.
Republicans evaluating copartisans are somewhat less likely to vote for copartisans with any news (including positive news) relative to the No recent news condition in the high information environment, a finding that we would not have theoretically expected.
Similar to scandal, the effect of sharing partisanship with a candidate also declines as the information environment expands, consistent with Peterson (2017). Most attributes do not show significant differences as the information environment expands from medium to high, though effects vary (See Online Appendix Figures A5, A6).
Confidence intervals are 95% cluster bootstrap percentile intervals.
This measure of shared policy positions is based on comparing respondents’ pre-treatment self-reported attitudes with candidate positions. The more shared policy stances, the greater likelihood of voting for the candidate (Online Appendix Table A7).
In this model specification, we include all attributes and levels instead of solely the news and information variables. Given that issue positions and candidate demographics are not present in the low information environment, we did not include these attributes in the main specification. Due to the independent randomization of attributes and levels, including these variables in the specification does not substantially change the effects of scandal.
This analysis is limited to partisans in order to evaluate the effect of shared partisanship.
References
Abramowitz, A. I. (1991). Incumbency, campaign spending, and the decline of competition in U.S. House elections. Journal of Politics, 53(1), 34–56.
Alford, J., Teeters, H., Ward, D., & Wilson, R. (1994). Overdraft: The political cost of congressional malfeasance. Journal of Politics, 56(3), 788–801.
Bansak, K., Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2018). The number of choice tasks and survey satisficing in conjoint experiments. Political Analysis, 26, 112–119.
Bansak, K., Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2019). Beyond the breaking point? Survey satisficing in conjoint experiments. Political Science Research and Methods. https://doi.org/10.1017/psrm.2019.13.
Barnes, T., Beaulieu, E., & Saxton, G. (2020). Sex and corruption: How sexism shapes voters’ responses to scandal. Politics, Groups, and Identities, 8(1), 103–121.
Bartels, L. (2002). Beyond the running tally: Partisan bias in political perceptions. Political Behavior, 24(2), 117–150.
Basinger, S. J. (2013). Scandals and congressional elections in the post-Watergate era. Political Research Quarterly, 66(2), 385–398.
Basinger, S. J. (2018). Judging incumbents’ character: The impact of scandal. Journal of Political Marketing, 18(3), 216–239.
Bhatti, Y., Hansen, K. M., & Olsen, A. L. (2013). Political hypocrisy: The effect of political scandals on candidate evaluations. Acta Politica, 48(4), 408–428.
Blair, G., Cooper, J., Coppock, A., Humphreys, M., & Sonnet, L. (2018). estimatr: Fast estimators for design-based inference. R package.
Brown, L. M. (2006). Revisiting the character of congress. Journal of Political Marketing, 5(1–2), 149–172.
Carlson, J., Ganiel, G., & Hyde, M. S. (2000). Scandal and political candidate image. Southeastern Political Review, 28(4), 747–757.
Cobb, M., & Taylor, A. (2014). Paging congressional democrats: It was the immorality, stupid. PS: Political Science & Politics, 47(2), 351–356.
Coppock, A., & McClellan, O. A. (2019). Validating the demographic, political, psychological, and experimental results obtained from a new source of online survey respondents. Research & Politics. https://doi.org/10.1177/2053168018822174.
Dimock, M., & Jacobson, G. (1995). Checks and choices: The House bank scandal’s impact on voters in 1992. Journal of Politics, 57(4), 1143–1159.
Doherty, D., Dowling, C. M., & Miller, M. G. (2011). Are financial or moral scandals worse? It depends. PS: Political Science & Politics, 44(4), 749–757.
Doherty, D., Dowling, C., & Miller, M. (2014). Does time heal all wounds? Sex scandals, tax evasion, and the passage of time. PS: Political Science & Politics, 47(2), 357–366.
Fischle, M. (2000). Mass response to the Lewinsky Scandal: Motivated reasoning or Bayesian updating? Political Psychology, 21(1), 135–159.
Franchino, F., & Zucchini, F. (2015). Voting in a multi-dimensional space: A conjoint analysis employing valence and ideology attributes of candidates. Political Science Research and Methods, 3(2), 221–241.
Funk, C. (1996). The impact of scandal on candidate evaluations: An experimental test of the role of candidate traits. Political Behavior, 18(1), 1–24.
Green, D. P., Zelizer, A., & Kirby, D. (2018). Publicizing scandal: Results from five field experiments. Quarterly Journal of Political Science, 13(3), 237–261.
Hainmueller, J., Hopkins, D. J., & Yamamoto, T. (2014). Causal inference in conjoint analysis: Understanding multidimensional choices via stated preference experiments. Political Analysis, 22(1), 1–30.
Hainmueller, J., Hangartner, D., & Yamamoto, T. (2015). Validating vignette and conjoint survey experiments against real-world behavior. Proceedings of the National Academy of Sciences of the United States of America, 112(8), 2395–2400.
Hamel, B., & Miller, M. (2019). How voters punish and donors protect legislators embroiled in scandal. Political Research Quarterly, 72(1), 117–131.
Horowitz, J. (2011). Rep. Weiner admits tweeting lewd photo of himself. The Washington Post.
Iyengar, S., & Kinder, D. R. (1987). News that matters: Television and American opinion. Chicago: University of Chicago Press.
Klašnja, M., Lupu, N., & Tucker, J. A. (2020). When do voters sanction corrupt politicians? Journal of Experimental Political Science. https://doi.org/10.1017/XPS.2020.13.
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480–498.
Lau, R. R., & Redlawsk, D. P. (2001). Advantages and disadvantages of cognitive heuristics in political decision making. American Journal of Political Science, 45(4), 951–971.
Long, N. (2019). The impact of incumbent scandals on senate elections, 1972–2016. Social Sciences, 8(4), 1–19.
McCurley, C., & Mondak, J. J. (1995). Inspected by #18406313: The influence of incumbents’ competence and integrity in U.S. House elections. American Journal of Political Science, 39, 864–885.
McDermott, M., Schwartz, D., & Vallejo, S. (2015). Talking the talk but not walking the walk: Public reactions to hypocrisy in political scandal. American Politics Research, 43(6), 952–974.
Nyhan, B. (2015). Scandal potential: How political context and news congestion affect the president’s vulnerability to media scandal. British Journal of Political Science, 45(2), 435–466.
Nyhan, B. (2017). Media scandals are political events: How contextual factors affect public controversies over alleged misconduct by U.S. governors. Political Research Quarterly, 70(1), 223–236.
Paschall, C., Sulkin, T., & Bernhard, W. (2019). The legislative consequences of congressional scandals. Political Research Quarterly. https://doi.org/10.1177/1065912919826224.
Pereira, M. M., & Waterbury, N. W. (2018). Do voters discount political scandals over time? Political Research Quarterly. https://doi.org/10.1177/1065912918795059.
Peters, J. G., & Welch, S. (1980). The effects of charges of corruption on voting behavior incongressional elections. American Political Science Review, 74(3), 697–708.
Peterson, E. (2017). The role of the information environment in partisan voting. Journal of Politics, 79(4), 1191–1204.
Praino, R., Stockemer, D., & Moscardelli, V. G. (2013). The lingering effect of scandals in congressional elections: Incumbents, challengers, and voters. Social Science Quarterly, 94(4), 1045–1061.
Rapoport, R., Metcalf, K., & Hartman, J. (1989). Candidate traits and voter inferences: An experimental study. Journal of Politics, 51(4), 917–932.
Robertson, G. D. (2020). Cunningham keeps low in NC Senate race marked by his affair. Associated Press. https://apnews.com/article/election-2020-virus-outbreak-senate-elections-sex-scandals-north-carolina-5b2812777ae5e5582858360aaf22b76d.
Rundquist, B. S., Strom, G. S., & Peters, J. G. (1977). Corrupt politicians and their electoral support: Some experimental observations. American Political Science Review, 71, 954–963.
Saad, L., & Newport, F. (2015). “Email” defines Clinton; “immigration” defines Trump. Gallup. https://news.gallup.com/poll/185486/email-defines-clinton-immigration-defines-trump.aspx.
Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50(3), 755–769.
Von Sikorski, C. (2018). The aftermath of political scandals: A meta-analysis. International Journal of Communication, 12, 3109–3133.
Vonnahme, B. M. (2014). Surviving scandal: An exploration of the immediate and lasting effects of scandal on candidate evaluation. Social Science Quarterly, 95(5), 1308–1321.
Ward, P. S. (2017). Rep. Tim Murphy popular with pro-life movement, urged abortion in affair, texts suggest. The Pittsburgh Post-Gazette.
Walter, A. S., & Redlawsk, D. P. (2019). Explaining voters’ heterogeneous responses to politicians’ immoral behavior. Political Psychology, 40(5), 1075–1097.
Welch, S., & Hibbing, J. R. (1997). The effects of charges of corruption on voting behavior in congressional elections, 1982–1990. Journal of Politics, 59(1), 226–239.
Zaller, J. R. (1992). The nature and origins of mass opinion. New York: Cambridge University Press.
Zaller, J. R. (1998). Monica Lewinsky’s contribution to political science. PS: Political Science & Politics, 31(2), 182–189.
Acknowledgements
We thank four anonymous reviewers and the editors for their thoughtful feedback during the review process.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary Information
Below is the link to the electronic supplementary material.
Rights and permissions
About this article
Cite this article
Funck, A.S., McCabe, K.T. Partisanship, Information, and the Conditional Effects of Scandal on Voting Decisions. Polit Behav 44, 1389–1409 (2022). https://doi.org/10.1007/s11109-020-09670-x
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11109-020-09670-x