Model-based and design-based inference goals frame how to account for neighborhood clustering in studies of health in overlapping context types

Accounting for non-independence in health research often warrants attention. Particularly, the availability of geographic information systems data has increased the ease with which studies can add measures of the local “neighborhood” even if participant recruitment was through other contexts, such as schools or clinics. We highlight a tension between two perspectives that is often present, but particularly salient when more than one type of potentially health-relevant context is indexed (e.g., both neighborhood and school). On the one hand, a model-based perspective emphasizes the processes producing outcome variation, and observed data are used to make inference about that process. On the other hand, a design-based perspective emphasizes inference to a well-defined finite population, and is commonly invoked by those using complex survey samples or those with responsibility for the health of local residents. These two perspectives have divergent implications when deciding whether clustering must be accounted for analytically and how to select among candidate cluster definitions, though the perspectives are by no means monolithic. There are tensions within each perspective as well as between perspectives. We aim to provide insight into these perspectives and their implications for population health researchers. We focus on the crucial step of deciding which cluster definition or definitions to use at the analysis stage, as this has consequences for all subsequent analytic and interpretational challenges with potentially clustered data.