Dissertation Title: Common Knowledge: Elicited Priors in Political Science
Committee: Alexander Tahk (Chair), Scott Gehlbach, Melanie Manion (Duke University), Rikhil Bhavnani, Ian Coxhead (Agricultural & Applied Economics), Edmund Malesky (Duke University)
Abstract: My dissertation addresses the question: how can divergent prior knowledge be applied to the study of political questions? Determining effective ways to utilize the information experts provide is an important step in bridging quantitative and qualitative analyses—particularly in authoritarian and underdeveloped contexts, where qualitative researchers have significant prior experience but quantitative researchers are beginning to overcome data challenges. In a Bayesian context, elicited priors utilize the substantive knowledge of experts, whether through interviews or published research, to improve the accuracy of posterior estimates. My dissertation draws on machine learning methods to propose a new way to aggregate elicited priors—a way that improves statistical analyses and takes into consideration the much greater variation in "expertise" extant in the social sciences, and especially in authoritarian and underdeveloped contexts. I apply this method directly to a small dataset of the participation of Myanmar members of parliament (MPs) to show that the method allows for better estimation of a complex model even using small-n data, but also uncovers motivations of MPs under authoritarianism that are more complex than previous work has led us to believe. I also apply the method in a study of US elections in order to assess what "expertise" is ideal for use in elicited-priors applications.