Integrating Large-Scale Online Surveys and Aggregate Data at the Constituency Level: The Estimation of Voter Transitions in the 2015 British General Elections

Abstract

What have been the underlying voter shifts that led to the victory of the Con-servative Party in the 2015 British general election – against all predictions bypollsters? Analyses of voter transitions based on (online) surveys and recall ques-tions are plagued by sampling and response biases, whereas aggregate data analysesare suspect of the well-known ecological fallacy. We propose a systematic statisticalcombination of individual and aggregate data at the constituency level to identifyregional electoral shifts between the 2010 to 2015 British general elections, with aparticular focus on England. Large-scale individual data collected by the BritishElection Study Internet Panel (BESIP) allow us to locate more than 28,000 respon-dents in their constituencies. We estimate transitions based on a recently developedBayesian Hierarchical Hybrid Multinomial Dirichlet (HHMD) model. We discovera clear deviance from pure RxC ecological inference and from pure online panel-based estimations of transition matrices. Convergence diagnostics corroborate thesuperiority of the hybrid models.