Mind the Gap? Political Advertisements and Congressional Election Results

Abstract

Do political television advertisements influence congressional election results? We test a hypothesis that candidates increase their vote share by increasing their advertisement airings relative to their opponent's airings (i.e., “mind the gap”). Using aggregate advertising data over three election cycles, we employ a two-stage least squares estimation with a rank-order instrumental variable, finding that the advertisement gap explains shifts in vote share after controlling for variables standard to congressional election modeling and advertisement volume. This methodological approach mitigates the endogenous relationship between advertisement airings and election results.

KEYWORDS:

• campaigns
• Congress
• elections
• endogeneity
• political advertisements

Notes

^a Significant difference of means with No Advertisement races (p < .05).

^b Significant difference of means with Single Party Advertisement races (p < .05).

^c Significant difference of means with Two-Party Advertisement races (p < .05).

*Significant difference of means (p < .05).

Value in parentheses represents row percentages.

^a Data compiled by authors from CMAG data set.

^b From all sponsors airing advertisements favoring a candidate.

^a Dependent variable is the democratic two-party vote.

*p < .05, **p < .01, ***p < .001.

^a Data obtained by subtracting the Republican two-party vote from the Democratic two-party, with negative values indicating a Republican victory.

^b Data obtained by multiplying the Ad Gap fitted and the beta-coefficient from the Two-Party Model (.029), with negative values indicating a Republican advantage.

*Congressional elections with the percentage of two-party vote gained from the Ad Gap fitted is more than the two-party vote margin.

This statement occurred on MSNBC's The Race for the White House with David Gregory, on 23 June 2008. We obtained the transcript for the show on http://www.msnbc.com.

These figures were compiled by the authors. The percentages we report are means. The data for these figures were obtained from the CMAG data provided by Kenneth Goldstein, Michael Franz, and Travis Ridout, whose work is supported by the Pew Charitable Trust under a grant to the University of Wisconsin-Madison and the Brennan Center for Justice at New York University. This data set tracks the top 75 media markets in the United States for the 2000 election and the top 100 media markets for the 2002 and 2004 elections (Franz et al. 2007), which is 86% of television households according to the Nielson Media Research website, http://www.nielsenmedia.com/DMAs.html. The CMAG data set includes the location and frequency of advertisement airings, which candidate the advertisement favored, who aired the advertisement, in which race the advertisement aired, and the estimated cost of each advertisement airing.

At the request of an anonymous reviewer, we operationalized the advertisement gap variable as the raw number differences between Republican and Democratic advertisement airings. We conducted this analysis because, as the anonymous reviewer stated, the magnitude of the advertisement gap as a percentage does not always reflect the magnitude of the advertisement gap as a raw number. For example, in a race where the total advertisement airing volume was 5,000, and the Democratic candidate's advertisement gap was .2, the Democratic candidate held a 1,000 raw number advantage; however, in race where the total advertisement airing volume was 1,000 and the Democratic candidate's advertisement gap was .3, the Democratic candidate held a 300 raw number advantage. In this example, the magnitude of the advertisement airing gap as a percentage does not consistently reflect the magnitude of the raw number gap. Supporting the raw differences operationalization, Franz and Ridout (2010) find that raw number differences influence presidential vote share in 2004 and 2008 (without controlling for the total number of advertisements aired in a media market, and without this control, the raw number of advertisement airings suffers from the same critique); however, the advertisement-airing gap measured as a percentage reflects the magnitude of the raw number of airing differences in both the 2004 and 2008 presidential elections. When we operationalized our advertisement gap using the raw number of differences, we do not find any statistically significant influence of advertisement airings on the two-party vote. We interpret this non-finding as additional support for our theory about advertising effectiveness: the most important aspect of political advertising is the domination of the established advertisement space, not merely the raw number difference in advertisement airings. Using the raw number differences in advertisement airings makes a questionable assumption about advertisement effectiveness: a 1,000 advertisement-airing difference is the same for a race with 3,000 total advertisement airings as it is in a race with 10,000 total advertisement airings. For the data in our study, first, we control for total advertisement volume (Republican plus Democratic advertisement airings), and second, the correlation between the raw number and the percentage advertisement airing differences is .639 (p < .000), which means the situation described above does occur, but not at a rate that may skew the findings. However, more research must be conducted to discover how advertisement airings influence election results.

While our measure of the ad gap variable provides no variation in the ad gap in the one-party model, variation can occur in exposure to the advertisement airings in the one-party model. Certainly, individuals who do not watch television or who fail to watch certain programming in which advertisements are aired would not be exposed to those advertisements, even in races where only one party aired advertisements. Future research linking GRPs and individual-level data assessing exposure more directly may provide valuable insight into how exposure shapes the relationship between the ad gap and the two-party vote. We thank an anonymous reviewer for illuminating these points.

We thank an anonymous reviewer for stating that although our analysis finds that advertisement airings are similar across times of day, it is very likely that these same airings have important variation across programming. At the presidential level, descriptive analysis has confirmed variation across programming (Ridout et al. 2010); however, these findings have not been definitively linked to election results. Related to both time of day and programming, an important research question emerges: does the advertising gap during high levels of exposure increase vote shares more than during low levels of exposure?

This variable construction and research design is not conducive to making an argument about the mechanisms important to shifting the two-party vote; instead, our research design seeks to analyze the statistical correlation between advertisement airings and election results. The mechanisms that allow advertisements to shift vote share could be the content (e.g., issues, emotions) or tone of the advertisement, its partisan cues, or merely the act of airing the advertisement confers a sense of legitimacy on the candidate favored in the advertisement. Given the lack of substantive findings with content (Sides 2007) or tone (Lau, Sigelman, and Rovner 2007) affecting the two-party vote, we suggest future studies should examine the legitimacy mechanism. The legitimacy hypothesis argues that voters cannot assess the quality of a candidate by the information in the advertisement; thus, it is only the act of advertising that informs voters of candidate A being more legitimate than candidate B (see Nelson 1970, 1974).

We thank Gary Jacobson for generously supplying much of the data for the control variables used in this analysis.

For the Full Model, the low group included values between −1 and −.4976, the medium group between −.4977 and .1902, and the high group between .1903 and 1. For the Two-Party Model, these values were −.9788 to −.1487, −.1488 to .1218, and .1219 to .9547, respectively.

An ordinary least squares for both the Full Model and the Two-Party Model, using the observed values for Ad Gap, produced similar results. The coefficient was larger for the Full Model, but the statistical significance across both models was similar, for all variables and model fit statistics.

When we measure remaining expenditures with the natural log (both Democrat and Republican), they are statistically significant (p < .001).

Also see Franz and Ridout (2010, 310). Studying presidential advertisement effects, the authors show similar advertising relationships between Republican and Democratic presidential candidates in 2004 and 2008.

Author ordering is alphabetical and does not indicate the level of contribution by each author.

Additional information

Notes on contributors

Michael D. Jones

Michael D. Jones is currently an assistant professor at Virginia Tech's Center for Public Administration and Policy and is also a research fellow at the Edmond J. Safra Center for Ethics at Harvard University. His research interests include political communication and public policy.

Paul D. Jorgensen

Paul D. Jorgensen is an assistant professor of political science at the University of Texas-Pan American. His research interests include campaign finance and political economy.