Aggregating Preferences and Information: Estimating a Model of Large Election

Voter turnout has been one of the most fundamental research question in the study of political economy. Most study has focused on turnout in simultaneous election in which pivot probabilities are same for all voters. As pivot probabilities are the same in a simultaneous election, the only way to investigate empirically how pivot probabilities affect turnout decision is to compare outcomes of different elections controlling for other factors. In sequential election, to the contrary, pivot probabilities evolve over time within an election and voters socially learn from preceding votes. We propose a dynamic strategic model of turnout and social learning, and estimate the model using the data of U.S. Presidential Primaries. We find a large effect of social learning and quantify its effects on turnout. Finally, we conduct counterfactual experiment on alternative voting schemes.