Designing multi-level studies: Sampling voters and electoral contexts

Every two years in the US, 435 congressional elections take place that scholars study using data from the National Election Studies (NES) survey of the American electorate. With a focus on sampling, this article explores two issues: (1) How best to design a national election study if the aim is to understand voting behavior within and across subnational contexts; and (2) How, by comparison, the existing NES surveys have been designed. Although our arguments specifically address how one should sample individuals and congressional districts in the US, our conclusions apply to any situation where one is sampling micro-level units nested within diverse and influential macro-level contexts.