The effect of insurance on self -rated health: Assessing the role of selection bia

Critical reviews of the literature on the effects of health insurance on health demonstrate that selection bias has not been adequately addressed and remains a threat to the validity of these estimates. Furthermore, previous studies have primarily measured insurance at one point in time, leaving unaddressed the role coverage stability plays in these estimates.;The primary goal of this dissertation is to produce an unbiased estimate of insurance effects on health for an adult population. To achieve this estimate, a two-step estimation method with instrumental variables is used. This analysis relies on a sample of 4,975 single adults in the 1996 Medical Expenditure Panel Survey (MEPS). This data includes a 24-month insurance calendar. The dependent variable is a change score derived from the global single-item self-rated health measure at Round 5 subtracted from the same measure at Round 1. Three measures of insurance are tested in three independent models: (1) insurance at baseline; (2) proportion of months uninsured; and (3) type of transition. Insurance and change in self-rated health are separately modeled in two-step estimation, in which the predicted values of insured states from the first-stage equation is substituted for actual insured state in the second equation to achieve a consistent estimate of insurance effects. First-stage equations are reduced form and identified through a set of exogenous instrumental variables extracted from state and county level datasets and representing economic demand for insurance, state insurance reforms in the individual and group markets, and state public insurance generosity and eligibility.;No insurance effects were evident among the full population. Among an older subsample, coefficients for insurance shifted in magnitude but were not significant before or after correction. Among a chronically ill subsample, estimates unadjusted for selection were insignificant, but after correction for selection, group insurance significantly increased self-rated health by .6 units and public insurance by .5 units in the model testing baseline insurance. Models using proportion of months uninsured demonstrated similar trends but remained insignificant.