In a Fox News Poll from October 23 to 25, Herman Cain’s 24% led all candidates for the GOP nomination.Footnote 1,21,2Footnote 1,2 On October 30, 2011, Politico reported that two women accused Cain of sexual harassment and misconduct.Footnote 3 Two additional women came forward to accuse Cain of sexual harassment.Footnote 4 In late November, a fifth woman alleged that she had a 13-year affair with Cain.Footnote 5 Although Cain denied the allegations and the affair, he suspended his campaign on December 3 as a result of these “character assassinations.”Footnote 6 This rapid deterioration of Cain’s presidential trajectory illustrates that the public seems to care about the scandalous behavior of candidates. Although several studies identify a negative effect of scandal on the public’s attitudes, individual-level predispositions often moderate this reaction. Specifically, motivated reasoning encourages biased processing of scandalous information such that a candidate’s fellow partisans are least affected by the scandal.
However, primary campaigns pose a unique challenge for the motivated reasoning perspective as partisanship should not bias processing of scandalous information. Instead, we argue that one’s choice of media plays a role in how Republicans responded to the allegations about Cain. Given the range of choices voters face, the decision of which news source to rely on is illustrative of who the person is politically. We use a two-wave survey of likely Republican Iowa caucus attendees to explore how support for Cain conforms to the expectations of a motivated reasoning perspective. The results suggest that the accusations against Cain did not affect respondents equally. Respondents who obtained their information from cable news sources reacted differently than respondents who obtained their information from network news or from other sources.
MOTIVATED REASONING AND SCANDAL
A growing body of political research addresses the effect of scandal involvement on the assessments of political actors (e.g., Brown Reference Brown2006; Carlson, Ganiel, and Hyde Reference Carlson, Ganiel and Hyde2000; Cobb and Taylor Reference Cobb and Taylor2014; Doherty, Dowling, and Miller 2011, 2014; Fischle Reference Fischle2000; Funk Reference Funk1996; Goren Reference Goren2002; Newman Reference Newman2002) and the success of their campaigns (Rottinghaus Reference Rottinghaus2014). This research, despite different methodological approaches, documents a negative effect of scandals. Many studies use surveys to explore the effect of a scandal on the fortunes of politicians. Jacobson and Dimock (Reference Jacobson and Dimock1994) uncover evidence that representatives implicated in the House Banking scandal were more likely to lose in primary and general elections. Using experiments to manipulate scandals, Miller (Reference Miller2010) finds that participants reacted negatively to a politician’s involvement in scandalous activities. Scandals signal to voters that a candidate engaged in unacceptable (and sometimes illegal) behavior. Given that candidates spend time and money creating positive public images, such negative information should have a deleterious effect on candidate evaluations. Moreover, scandalous behavior (despite the media’s focus on scandals) is uncommon and can help differentiate candidates. Hypothesis 1 follows from this discussion:
Hypothesis 1. Herman Cain’s scandal involvement will reduce his support.
In addition, political predispositions may moderate reactions to scandals. Partisanship, political sophistication, and policy congruity have been found to moderate the effect of scandal (Dimock and Jacobson Reference Dimock and Jacobson1995; Funk Reference Funk1996; Stoker Reference Stoker1993). This uneven reaction has been described as an example of motivated reasoning (Fischle Reference Fischle2000; Goren Reference Goren2002). From this perspective, individuals maintain tallies representing their affect for politicians. These tallies are updated as the individual encounters new information; however, prior affect influences the processing of incoming information (Klein and Kunda Reference Klein and Kunda1992; Kruglanski and Freund Reference Kruglanski and Freund1983; Pyszczynski and Greenberg Reference Pyszczynski and Greenberg1987; Sorrentino and Higgins Reference Sorrentino, Higgins, Sorrentino and Higgins1986). In particular, individuals are motivated toward a biased assessment of incoming information (i.e., dismiss the scandalous information) to ensure that “updated” tallies correspond to predispositions. If individuals process information in a biased manner, then some will respond differently to scandal than others. For instance, Goren (2000) finds that partisans weighed scandalous information about the out-party candidate more heavily than information about in-partisans. Hypothesis 2 outlines this expectation:
Hypothesis 2: Predispositions will moderate the negative effect of scandal on support for Herman Cain.
In a primary, partisanship is unlikely to be the source of this biased processing. All of the candidates are in-partisans. Instead, we argue that the respondent’s choice of news media outlet captures the respondent’s motivation toward biased processing of the scandal. The choice of news source by the respondent is indicative of the respondent’s willingness to seek out information that is not always consistent with his or her preferences. Viewers of cable news are less likely to be open to persuasion whereas voters who pay attention to network news are more likely to accept information that is inconsistent with their predispositions. Research also suggests that network news may expose individuals to conflicting attitudes to a greater extent than cable news, especially Republicans (Mutz and Martin Reference Mutz and Martin2001; Stroud Reference Stroud2008). Our expectation, therefore, is that cable news viewers will respond to the scandal differently than network news viewers and those who get their information from other sources.
Viewers of cable news are less likely to be open to persuasion whereas voters who pay attention to network news are more likely to accept information that is inconsistent with their predispositions.
METHOD
We use the Iowa State University Caucus Polls from 2011 collected by the Survey and Behavioral Research Services at Iowa State University. This study included two waves: a November wave with a sample selected from the Iowa voter registration list and a December wave that re-interviewed these respondents. The November sample contained 1,256 respondents in Iowa and was collected from November 1 to November 15, 2011. The second wave contained 940 respondents and was collected between December 8 and December 18, 2011.
Sample
Table 1 provides an overview of the sample. The November wave included 979 registered Republicans and 277 registered independents. The December wave re-interviewed 740 registered Republicans and 200 registered independents. The mean age was 60.8 and 98% were white. The sample was evenly divided between men (49%) and women (51%).
Table 1 Characteristics of the Sample
Measures
The dependent variables in the analyses that follow are based on the candidate trait questions used by the American National Election Studies since 1980. Respondents were asked: “Think about [Candidate name]. In your opinion does the phrase “[Trait]” describe him or her extremely well, quite well, not too well, or not well at all?” Respondents rated four candidates (Herman Cain, Mitt Romney, Michele Bachmann, and Rick Perry)Footnote 7 on three traits (honesty, intelligence, and strong leadership). The dependent variable is an additive scale of the three questions about Herman Cain, scored such that higher values represent a more positive attitude toward Cain. The analyses of these 10-point scales are conducted using OLS.Footnote 8
Our first hypothesis anticipates that the allegations of sexual harassment will reduce Cain’s support. Our operationalization of this is time. We are fortunate in the timing of the survey. The allegations against Cain broke on October 30, right as the survey went into the field. Although we do not have pre- and postscandal measures of attitudes, we can use the timespan that the survey was in the field to measure the effect the allegations have on voters’ attitudes toward Cain. In particular, we dichotomize the measure of time into before or after November 7; the day Sharon Bialek outlined her allegations of sexual harassment against Cain. This was the first time that the specific details of Cain’s behavior were described by one of the victims. At that moment, voters were forced to think about the allegations directly.Footnote 9
The second hypothesis focuses on the role that predispositions, in particular preferences for media sources, play in moderating the effect of scandal on support for Cain. The measure of media choice is based on a survey question asking respondents “which of these news sources would you consider your main information source about the caucus?” Respondents were given the option of national television newscasts, cable television newscasts, local television newscasts, late-night comedy shows, newspapers, radio, Internet, advertising, or friends and family. We create dummy variables for the main news source being network news and cable news. We have tested various other specifications, and the key distinction is between those people who get their news from network TV and those who do not.
The survey also contains several questions we use as controls. Respondents were asked the standard seven-point, two-question party identification question, coded such that higher values represent stronger Republicans.Footnote 10 We also included the standard seven-point ideology question, where higher values represent identification as a liberal. We include a measure of Tea Party identification in the model as well. In addition, we include a measure of age in years, an indicator of being male, identification as born-again Christian, and a five-point scale measuring education, with higher values indicating the respondent is more educated. Finally, we include a measure of how closely the respondent is following the campaign coded such that higher values represent following the campaign more closely.
RESULTS
Before addressing the hypotheses, we provide descriptive statistics of the trait scale for the four candidates in table 2. Most respondents scored the candidates slightly above the midpoint. While Romney was the most popular, Cain came in second.Footnote 11 This suggests that Cain was a legitimate contender at this point in the campaign, despite the growing scandal. These allegations were not made about a second-tier candidate or one that voters felt universally negative toward.
Table 2 Mean Trait Ratings of All Candidates
All of the scores are positively correlated. Evaluations of Cain, Perry, and Bachmann correlate with each other between 0.42 and 0.49. The correlation with Romney is weaker (between 0.22 and 0.29), but still positive. These voters are not rating the candidates highly on all traits, nor are they evaluating their choice highly and the others negatively. Instead, these seem to be real impressions of candidates the respondents generally liked. Next, we developed a baseline model predicting attitudes about Cain. In the model, we included the controls discussed earlier and the variable indicating if the survey took place before or after November 7. The results appear in the first column of table 3.
Table 3 Support for Herman Cain, Wave 1 (OLS results)
Note: Standard errors in parentheses; * p<0.05.
Stronger Republicans viewed Cain more positively, as did Tea Party members, conservatives, born-again Christians, those following the campaign more closely, older voters, and more-educated voters. Those who chose to get their news from cable news were more positive, and respondents who got their news from network news were more negative than those who sought out other news sources. Finally, the timing of the interview mattered. If a respondent was interviewed before the November 7 press conference, he or she viewed Cain more positively than if he or she were interviewed after it.
Next we added the interaction between the timing variable and the news variables and present the results in the second column in table 3. The inclusions of the interactions change the interpretation of the effects of time and news choice. First, the indicator of the effect of time on evaluations of Cain (the effect of time for those who chose something other than cable news or network news) is no longer significant. In other words, the sexual harassment scandal, as measured by time, had no effect on those who received their information from other sources.Footnote 12
The effects of the allegations on those who get their news from network or cable news present a different story. The difference between network news viewers and the rest of the sample in column 1 is entirely because of a shift in opinion after November 7. Prior to that date, network news viewers are no different in their attitudes about Cain than the baseline group. After November 7, however, they become significantly more negative about Cain. They are the respondents for whom the scandal has the largest effects.
The cable viewers’ results are more complicated. Prior to November 7, cable viewers were significantly more positive about Cain than either the baseline group or network news viewers. The interaction between being a cable viewer and the time variable is insignificant, but this does not capture the effect of the scandal on cable viewers. Instead, we test the sum of the interaction term and the indicator of being after November 7.Footnote 13 The resulting F-test (F(2,893) = 4.22) is statistically significant, indicating that cable viewers were more negative about Cain after the press conference.
Finally, we use the data from the December wave to see if the differences across news choices change after Cain’s withdrawal. We use the same set of independent variables included in table 3, but shift the dependent variable to the respondent’s assessment of Cain in the second wave. The results appear in table 4. The difference across media sources continued into December, after Cain withdrew. Cable news viewers remained the most positive, while network news viewers remained the most negative. The effects of several of the control variables change. Age and education, for instance, are no longer related to attitudes about Cain. Most interestingly, born-again Christians are no longer more positive about Cain. Despite being more supportive of him in wave 1 of the survey, they are no different from non-born–again Christians.
The difference across media sources continued into December, after Cain withdrew. Cable news viewers remained the most positive, while network news viewers remained the most negative.
Table 4 Support for Herman Cain, Wave 2(OLS results)
Note: Standard errors in parentheses; * p<0.05.
To summarize these result we plot the mean evaluation of Herman Cain for each of the three media use groups (cable, network, and other) for the three time points (wave 1 before November 7, wave 1 after November 7, and wave 2) and present these means in figure 1. The most striking results are for those who voters get their news from neither cable nor network news. These respondents barely reacted during the first wave of the survey. Their mean evaluation of Cain declined by only 0.14 points—much less than the other two groups. Network news viewers, in contrast, seemed to react quickly and decisively with their mean dropping more than a full point. The differences between the post-November 7 portion of wave 1 and the mean in wave 2 for these groups shows the opposite pattern. The baseline group dropped the most between the two points while the network viewers dropped the least. Overall, the declines in the average rating of Cain from the pre-November 7 to December across the three groups were indistinguishable. Network viewers, on average, declined 1.54 points between the first portion of wave 1 and wave 2. The baseline group dropped 1.40 points while cable news viewers declined 1.48 points. The overall effect of the allegations and other campaign effects were the same for all three groups, but the timing was dramatically different.
Figure 1 Mean Evaluations of Cain by News Source and Time
The obvious question is: were there differences in the nature of the coverage across these mediums? The key to the content analysis we conducted was the connection between language and personality descriptions. As part of the effort to validate the Big Five dimensions of personality, Goldberg (Reference Goldberg, Spielberger and Butcher1982), asked subjects to rate themselves on all personality-relevant adjectives in the English language (1,710 words). We then factor analyzed the responses to uncover what is now known as the Big Five (openness to experience, conscientiousness, extraversion, agreeableness, and emotional stability). For this work, we use the set of factor loadings from Goldberg’s analysis as a dictionary in the content analysis, giving us the connection between each adjective and each of the Big Five dimensions of personality.
To conduct the analysis, we collected the full set of news transcripts from stories that discussed Herman Cain from two sources, Fox News and NBC news from November 1 until November 15. These files were processed to exclude all sentences that did not refer to Cain. A python program scanned the remaining text. The software scanned the text for each of the adjectives in the dictionary and then multiplied the counts by the factor loadings in Goldberg’s work. We then weighted this score by the number of words in the transcript to account for the differential lengths of the stories. This provides us with a score for the image of Cain’s personality on each of the Big Five for each transcript.
It is well established that the ANES-based trait questions tap some of the Big Five dimensions of personality (McGraw et al. 1996). In particular, the openness to experience dimension is connected to what the candidate evaluation literature refers to as competence. Similarly, the conscientiousness dimension is akin to integrity, the extraversion is linked to the extraversion dimension, and the agreeableness dimension is connected to empathy. Therefore, we believe that these measures of the image of Cain’s personality should correspond with the media coverage on the same aspects of Cain’s character as the survey questions asked.
Figure 2a Empathy Scores for Cain. Dashed Line is Fox, Solid Line is NBC
Figure 2c Leadership Scores for Cain. Dashed line is Fox, Solid Line is NBC
Figure 2b Competency Scores for Cain. Dashed Line is Fox, Solid Line is NBC
Not surprisingly, Fox presented a more positive picture of Cain than NBC did. It is also reassuring that our measure finds that the image of Cain’s character declined after November 7, the time that the harassment allegations picked up steam. What is surprising is that across the three dimensions the survey asked about, the effect of time is either the same for the two networks or the decline is stronger for Fox than it is for NBC. Figures 2a through 2c reports these changes and it is clear that the lines are either essentially parallel or the decline for Fox News is steeper than for NBC news.
The implication is that if it were the content of the coverage that was driving the differences in the responses between cable and network news viewers, we would expect that the decline would be stronger among cable news viewers—the exact opposite of what we find. If the content is not responsible for the changes in voter impressions of Cain, it is likely the voters’ motivations to process the information is responsible.
DISCUSSION
This study contributes to a growing body of research addressing the public’s reaction to scandalous information. Previous work on scandal involvement documents a consistently negative effect on public attitudes toward the scandal-ridden political actor. Our study of Herman Cain during the Republican primary confirms this finding. Moreover, research suggests that the reaction to scandal tends to be uneven with some individuals motivated to accept the scandalous information and others motivated to argue against the information. Primary elections provide a unique test for motivated reasoning as they eliminate the key predisposition—partisanship—that motivates biased processing of information. We examine an alternative measure of one’s predilection toward biased reception: choice of media sources and find that likely caucus attendees relying on network news sources reacted more negatively to the allegations of sexual misconduct than other respondents. We anticipate that one’s choice of news source is indicative of one’s willingness to seek out information that is not always consistent with one’s preferences. At this point, we cannot test this as a possible mechanism but our future work aims to explore this mechanism experimentally.
