Does Type of Health Insurance Affect Health Care Use and Assessments of Care Among the Privately Insured?

Health Services Research, April, 2000 by James D. Reschovsky, Peter Kemper, Ha Tu

Among policies mentioned by Household Survey respondents, 72 percent were linked to a specific insurer. Among these, 74 percent were directly matched by the insurer with a specific product. For the remaining cases that were linked to an insurer, but not a specific product, a statistical matching technique based on Household Survey information about health insurance attributes and household characteristics was used to complete the match to a specific product. [5] The resulting matched sample consists of 15,969 policies that cover 28,585 people. Weights were developed to account for the unmatched cases and to make estimates representative of the privately insured U.S. population (i.e., those not uninsured or covered by Medicare, Medicaid, or other public insurance). [6] This analysis excludes 1,328 persons covered by insurance whose policyholder lives outside of the household or is covered by both public and private plans, reducing the analysis sample to 27,257 individuals.

Methods

We estimate how the four major types of health insurance products affect individuals' health care use and their assessments of their care, controlling for an extensive set of individual characteristics and location. Product types are based on insurers' classffication. Logit (for dichotomous variables) and ordinary least squares (OLS) regressions (for continuous variables) were used to estimate these models. For continuous service use measures, we estimated two equations. The likelihood of any use was estimated using logit, and the log of the quantity of use (given any use) was estimated using OLS. All results are reported in natural units (e.g., probabilities, number of visits) as regression-adjusted means. These represent expectations for the standardized population of privately insured persons, that is, if all of these persons were in a given type of insurance. Significance tests are based on the underlying coefficients in the multivariate models, with standard errors adjusting for complex survey design usi ng SUDAAN.

Estimates of the effect of plan type control for numerous factors, including health status (using individual SF-12 variables), income, gender, race, education, marital status, age, family size, preferences for risk, willingness to trade off cost for provider choice, non-group market status, choice of HMO and non-HMO products, whether the current plan is new to the individual in the past year, and location (by using dummy variables for the 60 CTS markets). [7] Table 1 shows that selected socioeconomic characteristics and insurance options differ across the four types of insurance. But the key health status measures differed little, indicating little adverse selection among product types on health status.

If unmeasured factors are related to both individuals' choice of plan type and the dependent variables, coefficients on plan type variables can suffer from endogeneity bias. In this context, this has been termed "selection bias." Although differences in populations enrolled in the four types of plans shown in Table 1 suggest the possibility that results can be affected by selection bias, our extensive set of control variables, particularly health status, income, and consumer preferences, reduces this possibility. Moreover, tests in our earlier work indicated that selection bias did not present a serious problem (Reschovsky 2000a). Because unmeasured factors associated with the choice among the four plan types are not likely to differ from those associated with the choice between HMOs and non-HMOs, the extension from two to four product types is not likely to affect this conclusion.