Political science research consistently shows that political advertisements have small and uniform effects. This contrasts with claims that microtargeted campaigning can sway electoral outcomes and pose a threat to society. This paper introduces a novel experimental design simulating a targeted campaign to empirically test whether targeting political advertisements can be effective. Participants in an online survey experiment view one of five anti-Biden advertisements.
Respondents assigned to control are allocated advertisements at random. This data is used to train a model predicting Biden favorability from advertisement and pre-treatment traits. Respondents in the treatment group view the advertisement that this model predicts to be the most effective. The difference between targeted and random allocation is an 8.7 percentage point increase in Biden dislike and a 7.1 percentage point decrease in intent to vote Biden among unaligned voters who had not yet cast their vote (N=586). The effect is negligible among partisan voters.