A spatio-temporal hierarchical model to account for temporal misalignment in American Community Survey explanatory variables

The American Community Survey (ACS) is one of the most vital public sources for demographic and socioeconomic characteristics of communities in the United States and is administered by the U.S. Census Bureau every year. The ACS publishes 5-year estimates of community characteristics for all geographical areas and 1-year estimates for areas with population of at least 65,000. Many epidemiological and public health studies use 5-year ACS estimates as explanatory variables in models. However, doing so ignores the uncertainty and averages over variability during the time-period which may lead to biased estimates of covariate effects of interest. In this paper, we propose a Bayesian hierarchical model that accounts for the uncertainty and disentangles the temporal misalignment in the ACS multi-year time-period estimates. We show via simulation that our proposed model more accurately recovers covariate effects compared to models that ignore the temporal misalignment. Lastly, we implement our proposed model to quantify the relationship between yearly, county-level characteristics and the prevalence of frequent mental distress for counties in North Carolina from 2014 to 2018.