Bringing registration into models of vote overreporting

Abstract

Voting is a socially desirable act and a basic form of political participation in the United States. This social desirability sometimes leads respondents in surveys, such as the National Election Study (NES), to claim to have voted when they did not. The methodology of previous studies assumes that people only overreport voting and that the sample of potential overreporters (i.e., nonvalidated voters) is not systematically different from the sample of potential voters. In this research note, we explore several different ways of examining the determinants of overreporting at two different stages (registering and voting) and with a consideration for selection bias. Comparing the traditional probit model used in previous research with sequential and heckit probit models, we find that the determinants of overreporting registering and voting differ substantially. In addition, there is a significant selection effect at the registration stage of overreporting. We conclude with a discussion of contemporary implications for pre-election polling and the postelection analysis of survey data.