Estimating the Prevalence of Psychiatric Disorders Among National Guard Service Members Using a Bayesian Post-Stratification Model

• Activity Number: 120
• Type: Topic Contributed
• Date/Time: Monday, August 10, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Survey Research Methods Section
• Author(s): Sharifa Barracks* and Qixuan Chen and Sandro Galea
• Companies: and Columbia University and Columbia University Mailman School of Public Health
• Keywords: Calibrated Bayesian models ; penalized spline regression ; post-stratification weighting ; mental health

Abstract

The Ohio Army National Guard Study is a panel survey examining the prevalence and risk factors of psychiatric disorders among National Guard service members. A sample of 2616 soldiers from a population of 10778 completed the first wave of the study between 2008 and 2009. Survey samples often differ from the target population due to nonresponse and under-coverage. To estimate the population prevalence of psychiatric disorders, we propose a logistic post-stratification model, in which administrative data on population race, sex, rank, age, and marital status are used to improve the survey estimates. The stratum specific prevalence rate is modeled using a partially linear model, with the post-stratification weight modeled by a penalized spline to allow nonlinear association with the outcome. Further, post-stratification variables that are related to outcome are included in the model as linear predictors. Simulation study shows that the proposed method is more efficient than the classical post-stratification or raking estimators in population proportion estimation. The proposed method also allows the estimation of prevalence of psychiatric disorders in small sub-populations.