Association for Public Policy Analysis & Management

Medicaid and the state Children’s Health Insurance Program (CHIP) represent the single largest component of federal expenditures on child welfare. These programs’ substantial costs, and the current debate surrounding the reauthorization of CHIP, motivate a full accounting of these programs’ benefits. This paper seeks to explore the relationship between children’s health care receipt and subsequent criminal activity, using a panel of state birth-cohorts’ arrest rates and fractions eligible for Medicaid or CHIP (hereafter jointly referred to simply as Medicaid).

Measures for arrest rates come from the FBI’s Uniform Crime Reporting system, where each of the studied 1981-1984 birth cohorts are observed over ages 18-24, spanning calendar years 1999-2008. Measures of childhood Medicaid eligibility come from applying individual information on state of residence, family structure, and family income from the March Current Population Survey (CPS) to a constructed Medicaid Eligibility Calculator built from state and federal eligibility rules. This produces an estimate of the fraction of households with children of a given age in a given state and time who are eligible for Medicaid.

As characteristics such as family structure and family income effect both Medicaid eligibility and criminal behavior, I adopt the method of simulated instrumental variables, first pioneered in the seminal works of Currie and Gruber (1996) and Cutler and Gruber (1996) and now common in the literature. This method addresses the endogeneity of economic characteristics by constructing a measure of eligibility that varies only with state Medicaid policies. In practice, this is accomplished by applying a national sample from the CPS to the Medicaid Eligibility Calculator to find the fraction of households at a given time with children of a given age who would have been eligible, if they had lived in the given state. By using a national sample that does not vary in demographic or economic characteristics across states, this simulated eligibility measure varies only with state Medicaid policies, offering a “convenient parameterization of legislative differences affecting children in different state, year, and age groups” (Currie and Gruber, 1996). Simulated eligibility is then used as an instrument for actual eligibility in the regression’s first stage, before comparing the variation in arrest rates to the variation in eligibility in the regression’s second stage. With the inclusion of state and temporal controls, this amounts to a difference-in-difference specification that can identify the impact of Medicaid on later criminal outcomes based on variation in the fraction of a national sample eligible due to states’ Medicaid expansions.