More Than Just a Job? Causal Evidence of Employment's Impact on Health Status for SSDI Beneficiaries with a Severe and Persistent Mental Disorder
*Names in bold indicate Presenter
Supporting this change, several studies reported that in addition to increasing earnings, community-based employment positively impacts mental health status. These findings are from vocational rehabilitation RCTs that are not able to randomize employment, as well as from non-experimental studies, that do not control for the possibility of omitted variable bias. Our study aims to control for this possible bias in order to estimate a causal employment effect on health status. This is an important step for assessing the value of vocational rehabilitation programs for persons with SPMD.
Data for this study come from the Mental Health Treatment Study (MHTS), a large SSA-supported randomized trial aimed at increasing employment rates for SSDI recipients with SPMD. The MHTS included 2,238 SSDI beneficiaries whose primary reason for disability is SPMD. It provided two evidence-based services, IPS-SE and systematic medication management (SMM), and other expansions of behavioral-health services for beneficiaries. We assess the impact of employment in the first year of the MHTS on two measures of health status at the two-year final follow-up.
Outcome variables, measured at 24 months, are the mental component score (MCS) and the physical component score (PCS) from the SF-12 instrument. We estimate the effect of employment using a full information maximum likelihood (FIML) model with two parts: 1) a probit regression model for employment in year one and 2) an ordinary least squares model (OLS) for the MCS and PCS scores at 24 months. Employment is included as a covariate in the second part of the FIML estimation. We allow for cross-equation correlation in the errors (which implies omitted variables bias) and cluster standard error estimates on site locations. Covariates include baseline values of: (1) the MCS and PCS, (2) beneficiary demographic and social characteristics; (3) beneficiary recipiency history; (4) beneficiary employment in the past two years; and (5) average local labor-market unemployment rates.
The FIML estimation allows us to test for the correlation in the error terms. The test for the error correlation is positive but is not significant for both outcome variables when the MHTS treatment dummy is included as a covariate. This result holds in more parsimonious models with highly insignificant covariates dropped. Thus, single equation OLS regressions were estimated. The employment effect on both the MCS and the PCS are not significant, however the MCS approaches significance in the parsimonious model (1.25, 0.109). Alternative specifications where the MHTS treatment dummy is excluded are explored. Results suggest that excluding the MHTS treatment dummy leads to positively biased estimations of the effect of employment on the MCS.
Results indicate that omitted variable bias is not an issue in estimating employment effects on the SF-12 measures of health status. This research does not, however, support previous findings of employment benefits to mental health status.