The Effects of a Criminal Record on Employment, Welfare Participation, and Health: A Model of Long-Run Behaviors and Outcomes when Lagged Variables are Missing Non-Randomly

The authors study the collateral consequences of women's criminal records on their future employment, welfare participation, and health outcomes. We jointly estimate dynamic structural equations for life-cycle behaviors (employment, school enrollment, and welfare receipt), criminal offenses, and general and mental health outcomes using a cohort of disadvantaged women surveyed at five non-uniform intervals over thirteen years. The detailed survey questions allow us to construct annual behavioral histories so that we can explain contemporaneous behaviors by time-varying policy variables as well as uniformly-lagged past behaviors. However, because the wording of survey questions may differ by responses to preceding questions, individual behaviors may be missing non-randomly in some years. We address the endogeneity of important lagged determinants by modeling observed behaviors over time, conditional on being observed/known, as well as the probability of their missingness. Both the behaviors and the missingness, which is defined partially by the variation in wording at each wave and partially by a woman's chosen behaviors, are functions of her endogenous histories of behaviors and outcomes, exogenous characteristics, permanent and time-varying correlated unobserved heterogeneity, and random shocks. The econometric approach allows us to differentiate between possible direct causal impacts of criminal record on health and indirect effects on health through employment, education, and welfare receipt. We use the estimated dynamic model to simulate behaviors and health trajectories based on different criminal record histories and policy scenarios.