![Sample our Politics & International Relations journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5947/TOC-Subject-Sample-2021-Politics-International-Relations.jpg"})
Abstract
This article uses panel survey data from 2010 to 2012 to examine how scandals affect perceptions of House incumbents’ character. Scandals negatively impact impressions of the incumbent’s integrity, indicating that voters incorporate information about their representatives’ behavior. Scandals’ effects on attitudes are shown to vary depending on the type of misbehavior that is alleged, with sex scandals having far deeper and wider impact than other types of scandals. Scandals significantly influence the vote choice, although impressions of the incumbent’s integrity mediate more than one-third of their effect.
Notes
1 Despite finding a slightly higher victory rate (81%) among members who stood for re-election, Basinger (Citation2013) found that a smaller fraction of scandal-tainted incumbents (73%) were candidates in the general election itself, compared to 84% in the Peters–Welch–Hibbing analyses. Research on the House Bank scandal found that incumbents’ exits from the electoral arena were predictable. Incumbents who were more deeply involved in the scandal, measured by the number of checks an incumbent bounced, were more likely to retire or lose in their primary; see especially Alford et al. (Citation1994); Banducci and Karp (Citation1994); Jacobson and Dimock (Citation1994).
2 Using just the 2006 general election, which was particularly rough for Republicans, Grose and Oppenheimer (Citation2007) reached the opposite conclusion on the relative sizes of scandals’ effects, with Democrats paying a slightly higher price than Republicans (7% vs. 4%). But, both they and Hendry et al. (Citation2008) confirm that a generic scandal costs a generic incumbent around 5% of the general election vote.
3 Brown (Citation2006) also found that Republicans pay a slightly greater price in the general election for a generic scandal than Democrats do (9% vs. 6%), and Republicans pay a particularly steep price for a sex scandal, at nearly 21%.
4 Although “quality” or “character” might refer to many traits, Kinder (Citation1986) reduced the trait space to four dimensions labeled competence, leadership, integrity, and empathy. Funk (Citation1999) surveys the important scholarship on these dimensions and further reduces the trait space to three dimensions by combining competence and leadership.
5 Granted, incumbents identified as “low quality” did include many politicians tainted by sex scandals. For example, Dan Crane was exposed for having an affair with a 17-year-old House page, Jon Hinson was arrested three times for sodomy, including once in a restroom in the Capitol, and Gus Savage was accused of sexually assaulting a Peace Corps worker in Zaire.
6 More recently, Hendry et al. (Citation2008) asked survey respondents to evaluate how well the words “honest” and “competent” described sitting incumbents in the 2006 election. They found that scandals significantly impacted honesty ratings. They do not report that those ratings influenced vote choices, and they do not carry out a mediation analysis such as in presented in section 6.
7 See too Hendry et al. (Citation2008: 98–101).
8 Values of these two variables correlate at r = .89, although much of this correlation is due to the fact that both evaluations are influenced by partisan factors, as I demonstrate in section 6.
9 To ensure consistency, I exclude districts in which the incumbent faced no major-party challenger, districts in which two incumbents faced each other following a redistricting, etc.
10 Before taking the logarithm I set the minimum expenditure to $5000 for contested races, to accommodate FEC regulations on filing campaign fundraising and spending reports, and I set the minimum value to $1 for uncontested races.
11 An examination of respondents’ ratings in the first wave of the panel, i.e., in 2010, reveals that members who would become involved in scandals during the 112th Congress had average integrity ratings of 0.195 and average competence ratings of 0.304. By comparison, members who were not involved in scandals in the 112th Congress had average integrity ratings of 0.429 and average competence ratings of 0.394.
12 District-specific random effects were included to control for unmeasured heterogeneity across districts, and standard errors were clustered by congressional district.
13 To check the robustness of the regression models, I also calculated the average attitude on each character trait across all respondents in the member’s district in 2010. This measure identifies which members had reputations for high or low integrity and high or low competence in 2010. When average lagged integrity and average lagged competence ratings are used in place of individuals’ lagged ratings, the R2 values in Table 3 fall by nearly half (from 0.33 to 0.18) and the coefficients on the party variables grow larger in magnitude (by roughly 60%), but the coefficients on scandal variables are barely affected. Results are available on request.
14 One might reasonably ask whether this effect is due largely to Anthony Weiner. Only 17 respondents resided in Weiner’s district, compared to 177 for the other four members involved in sex scandals. Re-analysis shows that all results in tables 2 and 3 are robust to inclusion or exclusion of Weiner’s constituents. Attitudes about Weiner’s integrity was especially harmed by the scandal, but there is no difference in voter attitudes about Weiner’s competence after one accounts for the fact that he was involved in a sex scandal and not a financial or political scandal.
15 Sample sizes are 14,589 respondents with a scandal-free incumbent and 917 respondents with a scandal-tainted incumbent, for a total of 15,506 respondents. This particular analysis is limited to districts in which one incumbent faced one challenger from the other major party.
16 Nearly ninety percent of respondents reported voting for a major-party candidate in the election. If the voter chose not to participate in the general election, his or her value is coded as missing. Four additional sets of respondents are omitted from the vote choice analysis: respondents whose incumbent resigned or retired, leaving an open seat; respondents whose incumbent faced no major-party challenger; respondents who had two incumbents facing each other in the general election due to redistricting; and respondents who had two candidates from the same party facing each other in the general election, due to “jungle primary” rules in states such as California).
17 Dimock and Jacobson (Citation1995) estimated models of the House vote in 1992, and found that higher numbers of overdrafts by the House member decreased the likelihood of a respondent voting for the incumbent. Hendry et al. (Citation2008) reached a similar conclusion about involvement in scandals in the 2006 election.
18 Lest one quibble with such claims based on a one-tailed test, note that if one re-runs the model without challenger expenditures, the sample increases by more than 4000, to 15,506; and simultaneously the coefficient on scandal changes to –0.110, with a standard error of 0.049, which is statistically significant at p = 0.026 (two-tailed).
19 The discussion of mediation draws heavily on MacKinnon (Citation2008), particularly ch. 3–5 and 11.
20 The null hypothesis is rejected at the 95% confidence level if |z| > 1.96.
21 Using a linear probability model for the vote choice equation has the practical advantage of providing unconditional partial effects for the mediator variable and covariates, at the cost of modest heteroscedasticity and possible out-of-bounds predictions (Wooldridge Citation2002, ch. 15). According to Wooldridge, “The case for the LPM is stronger if most of the xj are discrete and take on only a few values” (2002: 456), which is precisely the situation for the scandal, party and awareness variables.
MacKinnon and Dwyer (Citation1993) demonstrated that when the outcome variable is binary and a logit or probit regression is used, the difference in coefficients (C–C') will not equal the product of coefficients (A⋅B). However, MacKinnon et al. (Citation2007) show that the product of coefficients is an asymptotically unbiased estimate of the mediation effect. They provide a correction for estimating the mediation effect from the difference in coefficients when employing either a logit or probit regression to model a binary outcome variable; the correction relies on standardizing the C′ coefficient prior to calculating C–C' to approximate the mediation effect; see also MacKinnon (Citation2008, ch. 11). I utilize this approach for the second row of Table 5.
22 All models contain three covariates in both the mediation model and the outcome model. Awareness is coded 1 when the respondent correctly answered three tests of political sophistication and knowledge, and is coded 0 otherwise. Party is coded –1 when the respondent’s party identification is opposed to the incumbent’s party affiliation, 0 when the respondent identifies as an independent, and +1 when the respondent’s party identification is the same as the incumbent’s party affiliation. The natural logarithm of challenger expenditures also is included as a control. All models use random effects to control for unmeasured heterogeneity across districts and cluster standard errors by district.
23 I utilize the sgmediation package in Stata, written by Philip Ender, and then bootstrapped the standard errors using 1000 repetitions.
24 I utilize the binary_mediation package in Stata, written by Philip Ender. As before, I bootstrapped standard errors, using 1000 repetitions.
25 I utilize the medeff command in the mediation package in Stata, written by Raymond Hicks and Dustin Tingley; see Hicks and Tingley (Citation2011).
26 The closest experimental study is Funk (Citation1996). The article’s focus is on whether traits such as warmth and competence insulate incumbents from a scandal’s effects, so its experimental design manipulates rather than measures respondents’ attitudes about the fictional incumbent’s character traits. In a section on manipulation checks, however, the article notes “a significant main effect of scandal on a four-item index of trustworthiness” (15).
27 Numerous studies have examined perceptions of presidential candidates’ character. Prominent among them is Stoker’s (Citation1993) demonstration that Gary Hart’s rendezvous with model Donna Rice led to lower summary evaluations of Hart, particularly among voters ranking high on moral conservatism. Greene (Citation2001) uses a panel study to show that assessments of competence and integrity have a causal relationship with presidential approval. Hayes (Citation2005) shows that trait impressions vary systematically depending on the party of the candidate: Republicans score higher on leadership and morality, while Democrats score higher on compassion and empathy.
28 Interestingly, experimental studies reach the opposite conclusion, suggesting that fictional financial scandals will have more detrimental effects on fictional incumbents’ evaluations than fictional sex scandals (Funk Citation1996; Woessner Citation2005; Doherty et al. Citation2011, Citation2014). The typical experimental study manipulates the type of misbehavior, assigning a subject to read about tax evasion or an extramarital affair.