Social Interactions in Voting Behavior: Evidence from India

Using the unique staggered nature of the Indian General Elections, where voting takes place in several different phases spanning several weeks, we investigate how spatial variation in electoral dynamics affects subsequent voter turnout. Exploiting quasirandom assignment of constituencies to electoral phases each election, we assess the impact of average voter turnout in a given phase, on turnout in the subsequent phase. Standard endogeneity concerns in the estimation of social interactions are dealt by employing two distinct instrumental variables: 1) constituency specific average historical turnout in elections from the pre-staggered era, 2) voter density as measured by number of voters per polling location in a given constituency. Our estimates from both IVs, show that a 1 percentage point (pp) increase in turnout in a given phase depresses turnout in the subsequent phase by 0.3-0.5 pp. Crucially, falsification tests examining the effect on turnout in the current phase, of constituencies in the same phase or in future phases in the same election, produce no such effect. We find the data broadly support an ethical voter model, in which each agent acts as if setting an example for all and seek to maximize social welfare.