Download
Abstract
Voter turnout is a cornerstone of democratic stability, yet participation continues to decline across developed nations. This study investigates the mechanisms of information diffusion within geographical neighborhoods, examining how four social pressure treatments: Civic Duty, Hawthorne, Self, and Neighbors—influence both treated and untreated individuals. Using a combination of spatial autoregressive models and causal machine learning, including Double ML and Causal Forests, I re-analyze data from a large-scale field experiment involving 180,002 households. While direct treatment effects remain robust, with the “Neighbors” treatment generating the highest impact (~8%), I find no evidence of simple positive spillovers to untreated neighbors. Crucially, the machine learning models reveal a non-linear, inverted-U relationship between neighborhood treatment intensity and turnout for public pressure treatments. These results suggest that social pressure acts as a contagion only up to a saturation point, beyond which community fatigue or free-riding diminishes the effect. These findings provide a strategic framework for mobilization, suggesting that campaigns should target moderate treatment densities to optimize turnout.