Keywords: electoral reform, synthetic controls, regression trees, political science
Oregon implemented a statewide policy of automatic voter registration on January 1st, 2016. This law, called Oregon Motor Voter (OMV), made it so that unregistered Oregon residents who interacted with the DMV would be registered automatically – unless they explicitly opted out. Although it is almost certain that this change increased overall turnout, it is much more difficult to assess the magnitude of the effect. After all, there is no obvious way to distinguish between voters who would not have headed to the polls had they not been registered through OMV and those who would have registered themselves in time without the policy being in place. The fact that only one presidential election and one midterm have occurred since the policy was implemented also makes it harder to study turnout changes as turnouts are higher in those presidential elections regardless. In consideration of these difficulties, we are using two methods to get at the impact of OMV on Oregon elections. First, we are conducting a synthetic control analysis, essentially using data from other states and previous elections to construct a ‘fake’ Oregon without automatic registration and comparing the simulated and real results. This allows us to isolate the effects of OMV. We are also using a machine learning algorithm, boosted regression trees, to create Census Block Group level predictions of voter turnout and assess whether including OMV registration data affects our predictive accuracy. Over the past few years, automatic voter registration policies have become a topic of interest and some controversy among both lawmakers and political scientists. We expect our results will make positive contributions to this debate, and we hope to see greater academic attention given to this subject in the coming years.