Using Cell-phone Mobility Data to Study Voter Turnout

Scholars of voting behavior are often confronted with poor data availability or unsuitably large units of aggregation for reported turnout. We demonstrate a big-data solution to this challenge, using fine-grained cell-phone mobility data on millions of GPS locations for more than 300,000 eligible voters in Tokyo. Our approach uses the geolocations of polling stations, combined with GPS data points recorded on election day and a reference day, to measure patterns in individual-level (but anonymized) voting behavior. We first test the validity of the measure by comparing it to official aggregated data on turnout. We then demonstrate the measure’s substantive utility with two applications exploring the relationships between turnout decisions and (1) the cost of voting (proxied by distance to the polling station); and (2) the extent of neighborhood-level damages sustained during World War II bombings, building on an emerging literature on the long-run effects of political violence