![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
We examine how campaigns use advertising in the current digital era – and how spending decisions depend on the stage of the campaign and the particular goal being pursued. Our investigation relies on a database consisting of thousands of ads placed on Facebook by 24 different U.S. Senate campaigns in 2018. In the end, we find that campaigns pursue a large number of goals in their Facebook advertising and that the sequence of the campaign has a major influence on the goals pursued.
Supplementary material
Supplemental data for this article can be accessed on the publisher’s website.
Acknowledgments
We think Riley Dougherty, Miranda Gooler, Julia Ianelli and Redd Whetsel for their work in coding ads.
Notes
1. This is in many ways a second-generation set of research. Initial work in the late 1990s and early 2000s focused on campaign websites and the “webstyle” of those sites (Bystrom et al Citation2004).
2. According to Facebook’s political ad archive, over $1.4 billion was spent by advertisers on local, state, and federal election and policy-related ads on its platform between May 2018 and June 2020. These totals are from a report for “All Dates” at www.facebook.com/ads/library/report. Not all of these ads were explicitly election-related, but Facebook is just one platform for digital ad spending. Google reports Google search ads and YouTube ads, as well as ads purchased directly on specific websites. One might also consider ads purchased on streaming television or radio channels or applications like Pandora or Hulu. A recent report from the Wesleyan Media Project found that through June 27, 2020 (and for the period back to January 2019), ad sponsors in the presidential race alone had spent nearly $370 million on just Facebook and Google ads ($218 million on Facebook and $151 million on Google). See report at: https://mediaproject.wesleyan.edu/releases-070220/(Accessed on 7/9/20).
3. https://www.statista.com/statistics/264810/number-of-monthly-active-facebook-users-worldwide/(accessed July 31, 2019)
4. https://marketingland.com/time-spent-on-facebook-snapchat-remains-flat-but-instagram-sees-growth-261,705 (accessed July 31, 2019)
5. To help establish this point, we conducted a separate analysis of 1,922 television ads aired during 2020 in federal and gubernatorial races. Coders indicated whether each ad was aimed at persuasion, providing information, encouraging a contact, fundraising or selling merchandise and indicated the primary goal of each ad. We found that 88% of television ads had persuasion as the primary goal and 96% had persuasion as a goal.
6. This is enhanced with the comparatively low cost to produce ads on Facebook, especially compared with the cost to produce slick television ads. This allows campaigns to experiment with different ad formats and use the engagement metrics tracked by Facebook to derive what kinds of messages or ad content has more impact on voters. Notably this also allows campaigns to experiment with ad types within a goal. For example, campaigns can test which types of fund-raising messages produce higher click rates.
8. These were the races not rated as “solid” by the Inside Elections April 20, 2018, Senate Ratings. Texas was also initially in this sample, but because the volume of ads in that state was so much greater than anywhere else, we eliminated it from our final sample so as to make coding manageable.
9. This number might, at first glance, seem low given the hundreds of thousands of digital creatives that presidential campaigns place. But we do not believe this number indicates a problem with the data. First, most Senate campaigns do not have millions of dollars to invest in a digital team, like a presidential campaign does, and thus they are unlikely to create thousands of distinct ads. Second, ads with minor variations (a change in the color of a font, e.g.) might not appear as separate creatives in the Pathmatics data. We discuss this also in Appendix A.
10. Pathmatics was unable to capture the creatives for a small number of ads, even though the company was able to estimate spending and impressions for those ads. Thus, we were unable to code these ads. These uncaptured ads account for 5.96% of total impressions and 5.95% of spending.
11. For video ads, coders did not have the ability to click through and watch the entire ad. Instead, they saw the initial image of the video, along with the text surrounding the video. Given that most Facebook users choose not to play videos, this coding procedure mimics the experience of most Facebook users.
12. One of the authors developed an initial list of potential goals (from the perspective of the campaign) based on an examination of 100 Facebook ads sampled from various points in the campaign. This initial list of goals was then given to a coder who did sample coding using that list. The two then met to discuss the categories, which were refined based on the input of the coder, resulting in 14 goals.
13. To assess the reliability of the coding, we asked one of the coders to recode a random sample of 418 ads that were coded by the other three coders. The reliability of the coding varied considerably. More specifically, the Cohen’s kappa for the aggregated categories based on the primary purpose of the ad was 0.75 for persuasion, 0.88 for acquisition, 0.58 for mobilization and 1 for fundraising (the Krippendorf’s alpha was the same as the Kappa to the second decimal point for each category). Although the reliability of the mobilization category is below traditional thresholds, we note that this type of ad is uncommon, a situation that can lead to reliability statistics that fluctuate wildly based on the coding of one or two observations. Because it is an important theoretical category, we include the findings for the mobilization category in spite of the low reliability and urge readers to be cautious in making too much of the findings for that type.
14. An alternative model specification included an indicator of the week before a competitive primary (defined as a primary race in which a candidate won by fewer than 10% of the vote) instead of an indicator of the week before a primary. The substantive findings do not change at all, and the model estimates barely move when we use the alternative model.
15. Television spending data come from the Wesleyan Media Project, which tracks spending on broadcast television stations in all media markets in the United States. These data are commonly used by scholars in the study of television ads in U.S. electoral campaigns (see Fowler et al., Citation2016).
16. All three were rated as “solid Democratic” by Inside Elections in their final pre-election ratings.
17. More specifically, in terms of predicting spending on acquisition, only the November 2017 dummy variable remains a significant predictor. In regard to fundraising, only the first campaign finance deadline remains a significant predictor of spending, and the spike in fundraising appeals in October of 2018 is no longer evident. In terms of mobilization, while the significance of the July 2018 indicator variable remains, the later month indicators are no longer significant. Finally, in speaking of persuasion, the significance of the February 2018 and October 2018 indicators remains, but three other month indicators are no longer significant predictors of spending.
Additional information
Notes on contributors
Travis N. Ridout
Travis N. Ridout is Thomas S. Foley Distinguished Professor and Director of the School of Politics, Philosophy and Public Affairs at Washington State University in Pullman, Washington.
Erika Franklin Fowler
Erika Franklin Fowler is Professor of Government at Wesleyan University in Middletown, Connecticut.
Michael M. Franz
Michael M. Franz is Professor of Government and Legal Studies at Bowdoin College in Brunswick, Maine.