How Negativity Can Increase and Decrease Voter Turnout: The Effect of Timing

Abstract

Negative ads dominate campaign communication, but scholars continue to disagree over the effects of negativity on voter turnout. While some studies show that negativity leads to a lower likelihood of turnout, others find precisely the opposite. In this article, I leverage the role of timing to unify findings that were heretofore perceived as largely conflicting. I use the same data to show that at a certain time exposure to negativity can be mobilizing, but at other points in time exposure can be demobilizing. A combination of observational data and experimental results highlight these crucial conditions.

• advertising
• voter turnout

Notes

1. Except in cases of mudslinging.

2. Only in 1992.

3. In 2012, there were increases in ads embedded in entertainment programs shown online. Even if individuals can fast forward through televised political ads, most embedded online ads require that the viewer watch all the way through in order to watch the show.

4. This argument assumes that individuals are voting for their preferred candidate rather than against their disliked candidate. It is possible that people do vote based on candidate threat, meaning that they turn out to vote against an opponent. Recent research on voting by threat suggests, however, that this is most likely to happen when people have a very high sense of self-efficacy and a strong attachment to a particular candidate (Schuck & de Vreese, 2012). Thus, it is unlikely that these individuals would have been moved by post-choice negativity no matter their voter motivations.

5. Aggregate turnout patterns reinforce these predictions. Relying on the turnout information Krasno and Green (2008) present, the 2000 turnout rate in the Fresno media market—a market where the bulk of negativity was aired late in the campaign—was 47%. In the Lexington media market, where negativity aired evenly over the course of the campaign, turnout was at 50%. In contrast, however, within the same year, in the Albany-Schenectady media market, the bulk of negativity was aired early; within that media market, turnout was between 58% and 61% (depending on state). Constructing aggregate turnout data for 2004 shows that in the Chattanooga media market, where negativity was early, turnout was 63%. Meanwhile, that same year, in the Dallas media market—where the bulk of negativity came late in the election—turnout was 51%. In sum, although these results are not as consequential as the individual-level analyses in this article, they do hint at an important turnout pattern.

6. I rely on issue ads following previous scholars (Ansolabehere & Iyengar, 1994; Clinton & Lapinski, 2004). Further, the majority of negativity is issue based: In 2000, 93.44% of negative ads were issue ads; in 2002, 88.14% were issue ads; and in 2004, 96.59% were issue ads.

7. To ensure that observed results are not due to the fact that the candidates are simply called “A” and “B,” I replicate this study with an experiment that used real names (online Appendix 3).

8. Party of candidate was randomized.

9. Effect significant at p ≤ .01.

10. Effect significant at p ≤ .1 for the sample.

11. Pretests showed that partisanship is a good approximation of the candidate the individual will select absent any information. I conducted additional checks excluding subjects in the post group who identified as members of one party yet selected a candidate of a different party. The final result holds even when this group is excluded.

12. The NAES is commissioned by the Annenberg School for Communication and the Annenberg Public Policy Center of the University of Pennsylvania (Romer et al., 2004).

13. This is a smaller wave of the campaign than the NAES cross-section. As a result, the N in my analyses will naturally be smaller than the N of the full cross-section. Further, the N is smaller due to the interview dates necessitated by the structure of my analysis. I conduct checks to ensure that there is no distinction in the respondents interviewed on particular dates. All models are robust to controls measured in either pre- or post-election waves. I rely on the 2000 NAES for two key reasons. First, the timing of the pre-election interviews allows for variation in choice state that is highly beneficial for my tests. Second, this particular survey asks choice questions most appropriate for the analysis at hand.

14. Use of the Wisconsin Advertising Project data requires the following acknowledgment: “The data were obtained from a joint project of the Brennan Center for Justice at New York University School of Law and Professor Kenneth Goldstein of the University of Wisconsin-Madison, and includes media tracking data from the Campaign Media Analysis Group in Washington, D.C. The Brennan Center-Wisconsin project was sponsored by a grant from the Pew Charitable Trusts. The opinions expressed in this article are those of the author(s) and do not necessarily reflect the views of the Brennan Center, Professor Goldstein, or the Pew Charitable Trusts.’’

15. Given that these are panel data, tracing respondents’ eventual voting behavior provides additional validation of this measure. Notably, for this measure to be useful, the majority of respondents who are considered “undecided” need not change their minds—but the percentage of preference reversals should be significantly and substantively higher among these respondents than among those who have made choices. Of those individuals who first report that they would vote for Gore and that there is no chance they would vote for Bush, only 1.7% intended to cast ballots for Bush on the day prior to the election. In contrast, of those who first reported that they would vote for Gore, yet said there was still a chance they might vote for Bush, 25% intended to vote for Bush at the end of the campaign. Similarly, of those who initially reported that they would vote for Bush and said there was no chance they would vote for Gore, 1.6% intended to vote for Gore at the end of the campaign; of those who stated they preferred Bush but said there was still a chance they might vote for Gore, 20% reported that they intended to vote for Gore. These results are robust to controls for partisan strength.

16. I consider the level of negativity in two ways—as a percentage and as the raw number of negative ads. Both lead to substantively similar results, but the models in the main text rely on percentages of negativity.

17. The Franz et al. (2007) measure also relies on the natural log of the value obtained when the number of ads is adjusted by television viewership.

18. I also estimate a random effects probit; both models lead to the same conclusions.

19. See online Appendix 4.

20. I do not make predictions about the effects of pre-interview negativity due to data structure, though one can argue that in a media market with a consistent flow of negativity, there could be conditions where pre-interview negativity could set the stage for post-choice demobilization.

21. As Vavreck (2007) notes, there are errors in turnout self-reports. My analyses suggest that while self-reports are by no means a precise measure, there is nothing to suggest that the factors that lead to misreporting are confounded with factors crucial to my theoretical premises. Moreover, Berent, Krosnick, and Lupia (2011) suggest that validated votes do not lead to better data than self-reports.

22. For respondents who have not made choices and watch no television, the marginal effect of negativity is −0.19, p = .431. For respondents who have made choices and watch no television, the marginal effect of negativity is −0.07, p = .656.

23. I also obtain similar patterns of substantive results relying on raw negativity in the triple interaction. The average marginal effect of a choice decreases likelihood of turnout by 0.10 at average levels of negativity and by 0.23 at levels of negativity in the upper quartiles of the negativity distribution (both significant decreases in turnout likelihood).

24. I also leverage the panel structure of the data to consider whether the relationship between advertising and choice can explain changes in turnout intention over the course of the campaign. Using a measure of intention to vote at the start of the campaign and a measure of intention to vote near the end of the campaign, I create a dependent variable that considers whether intention to vote declined. Then I use the relationship between negativity, exposure, and choice to trace patterns in this shift. The results (presented alongside other tests in online Appendix 5) show that when individuals are exposed to negativity after a choice has been made, their intention to vote is significantly likely to decline. In contrast, when individuals are exposed pre-choice, their intention to vote is likely to either remain constant or increase. These results further reinforce that negativity has distinct effects before and after choice.

25. Another robustness check is the aggregate patterns in the same year, by media market. Specifically, using aggregate-level turnout data (Krasno & Green, 2008), I again use October 1 as a proxy for early and late negativity and consider whether we see the expected differential pattern across aggregate turnout results. As controls, I rely on demographics gathered via the U.S. census, previous turnout information, as well as competitiveness information. Using these data, I estimate a series of models relying on different estimation approaches: OLS, a two-stage model that initially uses media market factors to predict negativity patterns within the market and then considers turnout factors, and a model with state-level fixed effects (Krasno & Green, 2008). All results reinforce previous conclusions obtained with individual-level data: Early negativity increases turnout, while late negativity decreases it. The coefficients on the critical negativity variables are included in online Appendix 6.

Additional information

Notes on contributors

Yanna Krupnikov

Yanna Krupnikov is Assistant Professor, Department of Political Science, Stony Brook University.