This week’s issue of The Economist includes an article that analyses the effectiveness of political advertising during the 2016 Republican presidential primaries. For curious readers, we summarise our method here.
The ultimate test of a political campaign comes at the ballot box. However, there are only a handful of competitive contests per primary cycle. Moreover, their outcomes are determined mainly by factors outside campaign managers’ control, such as the quality of the candidates and the mood of the electorate. As a result, trying to discern the impact of advertising using actual vote totals would be futile. In contrast, polls are updated nearly every day. That means they can be used to measure the influence of specific ad campaigns, even if they occur long before election day.
In order to study the effect of ads on polls, we needed comprehensive data sets about each of them. For advertising information, we downloaded the records published by thePolitical TV Ad Archive, a recently launched project that strives to sweep up every television spot aired in the 2016 presidential primaries. The site codes every advert with information on its sponsor, broadcast location and message—whether positive, negative, or contrasting two candidates. For polling data, we used the polling averages published by RealClearPolitics, both nationally and in the early-voting states of Iowa and New Hampshire. Because there are only two Democratic presidential candidates remaining, we decided to focus our study on the much richer dataset presented by the deep Republican field.
Elections are a zero-sum game: holding turnout constant, every vote gained by one candidate is lost by another. As a result, the effect of advertising cannot be measured by looking at a single candidate’s poll numbers. For example, Bernie Sanders and Hillary Clinton have both avoided negative ads almost entirely. As a result, viewers’ opinions of both of them may have improved in recent months. Nonetheless, if they both become more popular by the same amount, then their standing in the polls relative to each other would not budge. In order to capture this zero-sum dynamic, we divided the Republican field not into individual contenders but rather into candidate pairs, such as Ben Carson versus Chris Christie or John Kasich versus Carly Fiorina. Even if both halves of a matchup are gaining or losing ground on the field as a whole, the gap between them might grow or shrink based on their relative positioning in the ad wars.
Just as with interest rates (at least until recently), there is a zero lower bound on a candidate’s poll numbers: no one can ever receive less than 0%, no matter how much they are pummeled on the airwaves. To prevent this asymmetry from distorting our results, we excluded candidate pairs in which at least one member failed to reach the 10% threshold in at least one of the state or national RCP averages for the day and race in question. Starting on November 20th with every possible candidate pair, this filter left us with 858 matchup-state-days—e.g, Jeb Bush compared with Ted Cruz in New Hampshire on January 10th.