County and State Health Insurance Coverage Estimation

The Small Area Health Insurance Estimates (SAHIE) program at the U.S Census Bureau is conducting research regarding the model-based estimation of insurance coverage for counties and has produced experimental estimates. There is widespread interest in estimates of health insurance at geographic levels below the state level. While there are several states that produce estimates of insurance coverage for counties within their boundaries, the estimates described here are novel in that they use a uniform method over all of the counties in the U.S. They also make use of several related administrative records, some of which are unavailable to the public. Fisher and Turner(2003) model health insurance coverage, measured by the Annual Social and Economic Supplement to the Current Population Survey (CPS ASEC), as a function of several variables, including administrative records and demographic variables. The primary interest in that paper is to estimate insurance coverage as the CPS ASEC defines it. The Survey of Income and Program Participation (SIPP) was added as a correlated response by Fisher and Turner (2004) with two goals: first, to improve the precision of the estimated CPS ASEC insurance coverage, and, second, to provide estimates of insurance coverage as defined by SIPP. The benefits for the CPS ASEC insurance coverage were small and interest in the SIPP coverage was small, so development of that model has been deferred since the model was complex enough to make fitting a much slower process. One deficiency of the model in Fisher and Turner(2003), labeled FT in this paper, is the variance estimation. There is evidence in that paper that the variances, as modeled, do not fit the data that well. Improvements to the variance model are incorporated here. Since the 2003 and 2004 papers, the models have been presented to several organizations including the State Data Centers (SDC) and the Federal and State Cooperative Program for Population Estimates (FSCPE). These groups had several suggestions, mostly for new covariates. We have also been considering the interpretability of the covariates and their coefficients and have concluded that a different set is indicated.