Abstract: Ad campaigns during elections are a common way to influence voter turnout. For example, during the 2020 U.S. presidential election, Aggarwal et al. (2023) conducted a randomized experiment to evaluate a negative, digital ad campaign against President Trump and found the campaign to be ineffective in changing voter turnout. But, their null results may not generalize to elections that are less exceptional than 2020, notably when COVID-19 emerged.

The main question in this paper is as follows: would negative, digital ads against Trump remain ineffective in changing voter turnout for the 2024 U.S. presidential elections? A randomized experiment like that in Aggarwal et al. (2023) is the gold standard to answer this question, but can be expensive. Instead, we propose to use a less-than-ideal, but a significantly cheaper and likely faster alternative based on transfer learning where we transfer knowledge from past experimental data to inform ad effectiveness for 2024. A key component of our analysis is a sensitivity analysis that quantifies the unobservable differences between elections, which can be calibrated in a data-driven manner. Within this framework, we propose two estimators to evaluate the ad effect on the target voter population: a simple regression estimator with bootstrap, which we recommend for practitioners in this field, and an estimator based on the efficient influence function for broader applications. Using our framework, we estimate the effect of running a negative, digital anti-Trump ad campaign on voter turnout for the 2024 U.S. presidential election in Pennsylvania (PA), a key "tipping point" state for the election. Our findings indicate effect heterogeneity across counties of PA and among important subgroups stratified by gender, urbanicity, and education attainment.

This is joint work with Xinran Miao (UW-Madison) and Jiwei Zhao (UW-Madison).


https://pages.cs.wisc.edu/~hyunseung/