OBJECTIVES: This study examines how capitated managed care contracting affects the production of medical care for the uninsured by federally-supported community health centers. Measuring the effects of managed care contracting is complicated by the endogenous selection of providers into managed care networks--a process subject to selective contracting by plans as well as self-selection by providers. Moreover, measures of managed care contracting are often censored due to providers that do not participate in contracts. METHODS: This study compares three methods for modeling uninsured care production: generalized least squares (GLS) regression; two-stage instrumental-variables (IV) estimation; and discrete-distribution mixture modeling. All models are estimated using a panel data set containing annual information on health centers, managed care plans, and local market and policy characteristics for all U.S. health centers during 1992-96 (3185 center-years). RESULTS: GLS estimates appear positively biased due to contract selection and censoring. The IV and mixture models show no evidence that capitated contracting adversely affects uninsured care production, after controlling for selection and censoring. Mixture model estimates, which are consistent with but more precise than IV estimates, indicate that production increases modestly in response to capitated contracting. As a consequence, centers that are not selected into contracts show diminished production of uninsured care. CONCLUSIONS: Latent-class mixture models, though computationally intensive, may offer advantages over other statistical models for studies faced with censored endogenous variables and selection bias. This study underscores the importance of selecting appropriate statistical models for observational studies of complex health policy issues.
Learning Objectives: At the conclusion of the session, the participant in this session will be able to: (1) Understand the threats to validity posed by selection and censoring in observational studies. (2) examine the relative advantages and disadvantages of three statistical models increasingly used for health policy analysis: generalized least squares regression; two-stage instrumental variables estimation; and latent-class mixture modeling. (3) compare the performance of three statistical models in addressing selection bias and censoring bias in a typical health policy problem
Keywords: Underserved Populations, Managed Care
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.