Purpose: This research aimed to identify significantly elevated areas of risk for suicide in Virginia adjusting for risk factors and risk factor uncertainty.
Methods: We fit three Bayesian hierarchical spatial models for relative risk of suicide adjusting for risk factors and considering different random effects. We compared models with and without incorporating parameter estimates' margin of error (MOE) from the American Community Survey and identified counties with significantly elevated risk and highly significantly elevated risk for suicide.
Results: Incorporating MOEs and using a mixing parameter between unstructured and spatially structured random effects achieved the best model fit. Fifty-two counties had significantly elevated risk and 18 had highly significantly elevated risk of suicide. Models without MOEs underestimated relative risk and over-identified counties with elevated risk.
Conclusions: Accounting for uncertainty in parameter estimates achieved better model fit. Efficient allocation of resources for suicide prevention can be attained by targeting clusters of counties with elevated risk.
Keywords: American Community Survey; Margin of error; Measurement error; Suicide; Uncertainty; Virginia.
Copyright © 2018 Elsevier Ltd. All rights reserved.