Zusammenfassung:
Aggarwal et al. (2023) analyze the effects of an 8-month-long advertising program on voter turnout in the 2020 US presidential election. Therein, 2 million voters were exposed to pro-Biden and anti-Trump advertisements on social media in five battleground states. The study finds no average treatment effect on voter turnout but differential effects when modeling by Trump support: Biden supporters are 0.4 percentage points more likely to vote while Trump supporters are 0.3 percentage points less likely to vote (t = −2.09 with p-value < 0.05). We conduct a direct reproduction of the paper by using their data and code. In addition, we check that their claims are robust to new analyses for understanding heterogeneity through the use of the causal forest methodology. We confirm the sign, magnitude, and statistical significance of the point estimates for the new analyses for understanding heterogeneity. The only significant discrepancy in results is that we find greater and statistically significant effects for the ATE, nearly all CATEs (age 18-39, gender, race, vote margin, partisanship (except Democrats), and Trump support score), and the differential effects of the Trump support score using a causal forest. These differences are likely due to the use of the causal forest and do not question the validity of the findings of the original paper.