Liar, Liar, Pants on Fire: How Fact-Checking Influences Citizens’ Reactions to Negative Advertising

AbstractElectoral campaigns are dynamic and an important change in recent elections is the growth of fact-checking; the assessment of the truthfulness of political advertisements by news media organizations and watchdog groups. In this article, we examine the role that fact-checks play in shaping citizens' views of negative commercials and political candidates. We rely on an Internet survey experiment where we vary people's exposure to negative advertisements and a follow-up fact-check article (i. e. , no fact-check, accurate fact-check, inaccurate fact-check). The results of our experiment show that fact-checks influence people's assessments of the accuracy, usefulness, and tone of negative political ads. Furthermore, sophisticated citizens and citizens with low tolerance for negative campaigning are most responsive to fact-checks. The fact-checks also sway citizens' likelihood of accepting the claims made in the advertisements. Finally, negative fact-checks (e. g. , fact-checks challenging the truthfulness of the claims of the negative commercial) are more powerful than positive fact-checks. Keywords: fact-checkingnegative campaigningcampaignselections Notes1. A number of scholars have examined the impact of corrective information on policy attitudes (e. g. , Kuklinski, Quirk, Jerit, Schwieder, Sides & Citrin, Citation2007). 2. See also Lewandowky, Ecker, Seifert, Schwarz, and Cook (2012) and Nyhan, Reifler, and Ubel (Citation2013). 3. Even though trust in the news media has declined in recent years (Ladd, Citation2011), we contend that people will be more likely to view a nonpartisan fact-checking organization as a trustworthy source, especially compared to the source of a political advertisement. 4. http: //www. opensecrets. org/overview/topraces. php? cycle=2012&display=currcandsout5. The Web sites for the two groups are http: //www. 60plus. org and http: //www. majority2012. com. 6. We conducted an original content analysis of fact checks from 2003 to 2012. We located 7, 008 fact-checks produced by 116 news and fact-checking organizations. Fact-checks were used to assess political advertisements, speeches, as well as debates. Of the fact-checks assessing political advertisements, 86. 5% of the advertisements scrutinized were negative commercials. 7. The following is the brief description of the Ohio Senate Race: "Ohio, a microcosm of the country politically, is a crucial battleground state in the upcoming election. In Ohio, U. S. Senator Sherrod Brown is running for re-election to a second term. Brown is being challenged by Ohio State Treasurer Josh Mandel. Mandel won his party's primary with 63% of the vote. Brown and Mandel have raised more than 25 million in their bid for the U. S. Senate. That amount makes the contest for the U. S. Senate the most expensive Senate race in Ohio history. And that's not counting outside spending, which is thought to be about 12 million. Polling in the state is close, with Brown enjoying a slight lead in the most recent polls. "8. See supplemental Appendix B (available on the publisher's website) for a copy of the questionnaire. 9. See the original http: //www. factcheck. org/2012/06/at-it-again/ for a fact-check of the advertisement attacking Sherrod Brown and see http: //www. majority2012. com/2012/05/news/releases/running-man-facts/ for facts that substantiated the claims made in the attack on Josh Mandel. The actual fact-check articles used in the experiment are presented in supplemental Appendix C (available on the publisher's website). 10. We secured approval from the institutional review board (IRB) at our institution and followed IRB protocol. Subjects were not made aware of the purpose of the study and were debriefed after the conclusion of the survey. In addition, we excluded Ohio residents from our sample because of the potential to influence citizens' voting decisions in the midst of a highly competitive campaign. We were required to balance the ethical considerations of deceiving potential voters against the potential gains of improving the external validity of the project. 11. The survey was conducted by SSI (Survey Sampling International), using a sampling platform called SSI Dynamix. ™ Please see http: //www. surveysampling. com/ssi-media/Corporate/Fact-Sheets-2013/ESOMAR-28 for more information about SSI's sampling procedure. 12. See supplemental Appendix B (available on the publisher's website). 13. This difference, according to the results of the one-way ANOVA, is statistically significant at p <. 01 (F = 34. 93). 14. Twenty-seven percent of the respondents rate the advertisements as "not useful at all, " 48% describe the advertisements as "somewhat useful, " and 25% say the advertisements are "very useful. " People's assessments of the ads' usefulness does not depend on whether they watch the advertisement attacking Josh Mandel or the advertisement attacking Sherrod Brown (F =. 746, ns). 15. Overall, 17% of the respondents view the advertisements as "overly hostile, " 51% classify the advertisements as "somewhat hostile, " and 28% rate the advertisements as "not hostile at all. " People's assessments of the ads' tone does not vary depending on whether they watch the advertisement attacking Josh Mandel or the advertisement attacking Sherrod Brown (F =. 127, ns). 16. The ANOVA examining the impact of experimental condition on ratings of the usefulness of the advertisement is significant (F = 7. 12, p <. 01) and the ANOVA examining the impact of the experimental condition on the rating of the tone of the advertisement is significant (F = 7. 86, p <. 01). 17. See supplemental Appendix B (available on the publisher's website) for exact question wording for the "factual questions about the candidates. "In creating the index, we recode people's answers so the index ranges from 3 (disagree strongly with each of the assertions) to 12 (agree strongly with each of the assertions). The Brown index has a mean of 7. 8 with a standard deviation of 2. 1. The Mandel index has a mean of 8. 6 with a standard deviation of 2. 1. 18. The mean difference among Conditions 1–3, according to the results of the one-way ANOVA, is statistically significant at p <. 01 (F = 17. 98, N = 222). Similarly, the mean difference among Conditions 4–6, according to the one-way ANOVA results, is also statistically significant at p <. 01 (F = 7. 69, N = 223). 19. This difference (7. 69 versus 7. 15), however, is not statistically significant (at p <. 05). 20. We sum people's scores on the accuracy, tone, and usefulness measures (i. e. , the measures examined in Figures 1, 2, and 3) into a single index. This global assessment of the advertisement has a range of 3 to 9, with a mean of 6. 2 and a standard deviation of 1. 7. The Cronbach's alpha for the composite index is. 69. 21. We also look at how demographic and political predispositions predict levels of tolerance toward negativity and find age, gender, ideology, and interest are significantly related to level of tolerance toward negativity. Older respondents, women, people with low levels of political interest, and liberals are less tolerant toward negative advertising, compared to younger respondents, men, and politically interested and conservative respondents. Strength of partisanship and political sophistication are unrelated to levels of tolerance. 22. We examined the additive impact of partisanship and strength of partisanship on people's views of the negative commercials. However, both factors failed to reach statistical significance in the models examined in Table 3. 23. For information on interpreting multiplicative interactions, see Brambor, Clark, and Golder (2006) and Friedrich (1982). 24. In calculating these estimates, we set all control variables at their mean and look at people exposed to the inaccurate fact-check and vary people's level of political sophistication. 25. Given the small number of cases in this analysis, we drop political interest and ideology from the models in Table 4 and look to see whether excluding these variables changes the results. However, the interaction coefficients in the models remain far from statistically significant in the reduced models. 26. In calculating these estimates, we set all control variables at their mean and look at people exposed to the inaccurate fact-check and vary the level of intolerance to negativity. Additional informationNotes on contributorsKim FridkinKim Fridkin is Professor of Political Science at Arizona State University. Patrick J. KenneyPatrick J. Kenney is Dean of the College of Liberal Arts and Sciences at Arizona State University. Amanda WintersieckAmanda Wintersieck is a PhD candidate in political science at Arizona State University.