Statistical models to predict mental illness among state and federal prisoners

Abstract

The Substance Abuse and Mental Health Services Administration (SAMHSA) developed a model-based methodology to obtain nationally representative estimates of serious mental illness (SMI) and any mental illness (AMI) among the adult U.S. civilian, non-institutionalized population in the 2008-12 Mental Health Surveillance Study (MHSS) as part of the National Survey on Drug Use and Health (NSDUH). This paper examines the feasibility of adapting and applying this methodology to 2016 Survey of Prison Inmates (SPI) data, collected by the Bureau of Justice Statistics (BJS), to obtain nationally representative estimates of SMI and AMI among state and federal prisoners.
The general methodology maintains a rigorous approach that provides reasonably accurate estimates of SMI and AMI (both overall and at the domain level). However, estimates obtained from the 2016 SPI should be viewed with some caution (particularly at the domain level, e.g., sex or age group). Several potential limitations of the methodology when specifically applied to the SPI arise from the small sample size of the MHSS subsample used to develop the SPI prediction model, limited information in the 2016 SPI available for inclusion in prediction models, and different populations represented in the model development process (parolees, probationers, and arrestees surveyed in NSDUH) and estimation phase (federal and state prisoners surveyed in SPI).
Despite these limitations, this approach to estimating mental illness among prisoners may provide useful indicators that are consistent with estimates produced by other federal agencies and researchers. Estimates of SMI and AMI are higher among the prisoner population than among the general adult population not in prison, large differences exist between federal and state prisoners and between male and female prisoners, and minimal differences exist by age group.