The authors use data from the 1996, 2001, and 2004 panels of the Survey of Income and Program Participation ( SIPP ). This data is supplemented with state-level information on Medicaid income eligibility rules. The authors classify each person as either: holding no insurance, private insurance, Medicaid, or both Medicaid and private insurance. About 10 percent of Medicaid recipients in the sample also have private insurance.
The authors run an OLS regression of eligibility on insurance status. The regressions model take-up (using any Medicaid insurance as the dependent variable), crowd-out (using private insurance as the dependent variable) and the overall change in insurance (any insurance as the dependent variable). Eligibility, however, is an endogenous variable; people may alter their income in order to receive Medicaid benefits. To address this problem, the authors instrument for Medicaid eligilbity using the fraction of the rest of the population (outside one’s own state) who would be eligilbe for Medicaid under the state’s Medicaid rules.
The problem with this approach, is that eligibility is not a great predictor of Medicaid enrollment. In the authors’ sample, about 40 percent of Medicaid recipients in our sample do not, by our imputation, appear to be eligible. Thus, the authors use an alternative specification that defines eligibility using the Medicaid eligilbity as a share of the federal poverty level for each state-month-year-family size cell.
The authors find the following results:
First, the eligibility expansions result in signiﬁcant increases in Medicaid participation; a “typical” expansion increases Medicaid participation by about four percent of baseline coverage rates. Second, the participation effect is larger for lower initial thresholds and the effect decreases as Medicaid thresholds increase. Third, we ﬁnd no statistically signiﬁcant evidence of crowd out regardless of initial threshold level.