We attempt to demonstrate the necessity for communicating HIV/STD prevalence estimates to the public with high spatial and demographic resolution, using a proposed revision the CDC-NCHHSTP HIV/STD Atlas. The Atlas as it currently stands presents either raw count data on HIV/STD prevalence at the county-level|with no covariate information, such as race/ethnicity, sexual orientation, or income| or state-level data which includes some information on covariates. We show that county-level data is insuciently resolved (especially without covariate data) to provide local-level users with the information they need to make informed health choices. To rectify this shortcoming, we present a fully formed, multi-level, Bayesian statistical model with which the CDC data can be analyzed, and from which local- level HIV/STD prevalence predictions conditional on covariate information can be simulated. We argue that model predictions could be displayed on the Atlas in place of the raw data. Since the multi-level model balances and integrates information at the local and global levels, release of the local-level model predictions maintains the anonymity of local-level residents, even if the population size of a county is small and model predictions include covariate data. It does so while providing more meaningfully accurate predictions of HIV/STD prevalence in sub-communities, while simultaneously characterizing uncertainty in prevalence rates conditional on spatial and demographic data.
Paper coming soon.