Computer Simulations of Elections, With Applications to Understanding Electoral System Reform

Abstract

This dissertation addresses problems in the computer modeling of elections, and then uses these models to simulate electoral system reforms. Computer simulations of how people vote often do not produce just one fixed election result, and instead get stuck in unending cycles. To motivate why these models are nevertheless worth studying, I introduce an extremely simple voting heuristic that correctly matches the winner of 95% of the districts in Canada's 2019 federal election. I show that the vote cycles in a broad variety election models obey simple constraints, and that, for some substantively important rulesets, where the model is stopped within these cycles does not strongly affect the aggregate results. I then apply these models to the question of election reforms. Politicians agree to change election rules when they think a change would benefit them, but there is no way to know how a specific party would perform under a specific alternative system. Combining game theory, past election results, and public opinion data, I estimate those effects falsifiably and accurately. The model suggests that an election reform attempt in Canada did not need to fail, and would have helped a supermajority of parliament. In Britain, it instead shows smaller effects from election reform than expected. The model also supports the idea that a convoluted new electoral system in Alaska was designed to protect a specific incumbent senator, but was probably not the best electoral system for the job. Election reform efforts frequently become mired in uncertainty and fail, and this method offers a way to transparently set expectations for different systems' effects.