Estimating Neighborhood Effects on Turnout from Geocoded Voter Registration Records∗

Do voters turn out more or less frequently when surrounded by those like them? While decades of research examined the determinants of turnout, little is known about how the turnout of one voter is influenced by the characteristics of other voters around them. We geocode over 50 million voter registration records in California, Florida, and Georgia and estimate the effects of racial and partisan composition of neighborhoods at the census block level. Through cross-section and panel difference-in-differences estimation, we address the general identification problem of neighborhood research: voters in different neighborhoods cannot be directly compared because voters’ individual characteristics as well as those of their neighborhoods differ. We find that a 10 percentage point increase in the out-group neighborhood proportion leads to an approximately 0.5 to 2 percentage point decrease in the turnout probability. These neighborhood effects persist in non-competitive districts, suggesting that mobilization alone cannot explain their existence. ∗We thank Bruce Willsie, the president of Labels & Lists, Inc., for generously providing unlimited access to their database and answering numerous questions. We also thank Bill Guthe and Jonathan Olmsted for geocoding and computational assistance. Eitan Hersh, Marc Meredith, and seminar participants at Princeton University (Center for the Study of Democratic Politics and Center for the Study of Social Organization), Seoul National University, and Vanderbilt University (Center for the Study of Democratic Institutions) provided useful suggestions. †Ph.D. candidate, Department of Politics, Princeton University, Princeton NJ 08544. ‡Professor, Department of Politics, Princeton University, Princeton NJ 08544. Phone: 609–258–6601, Email: [email protected], URL: http://imai.princeton.edu