Census Nonresponse and Electoral Abstention

Abstract

Through censuses and election counts, the modern state regularly requires popular participation in data collection to guide and authorise its policymaking. Both exercises suffer from nonresponse, which generates a missing data problem. This missing data represents both a statistical challenge (because selection biases introduce inferential uncertainty) and a policy problem (because decision-making based on these data can contribute to underrepresentation of non-participating groups). The co-occurrence of local elections and a census in Scotland in 2022 provides a unique opportunity to test hypotheses about this nonresponse, including whether missingness in the two contexts is correlated, and if it is, whether there is sufficient evidence of common causes to posit a theory of generalised nonresponse. Using a geospatial approach, we also investigate the relationship between nonresponse and contextual covariates including age, deprivation, and cohabitation rates. We conclude that understanding these relationships represents an opportunity for improving policy practice to mitigate against decreasing response and turnout rates in future.