SAHIE Age Model Methodology 2000: Estimation Details
Several features of the county estimates should be noted:
We estimate the number of people with health insurance coverage. The method is a mixed effects linear regression, where the log proportion insured is modeled as a linear combination of several predictors, mostly administrative records. The proportion insured in the dependent variable is a 3-year average of county-level observations from the Annual Social and Economic Supplement (ASEC) of the Current Population Survey (CPS). Although we use only the approximately 1,200 counties with CPS ASEC sample cases to estimate the equation, we estimate insurance coverage for all 3,140 counties in 2000.
Since the regression is in the log scale, one can think of the model as multiplicative; that is, we model the proportion of people with insurance as the product of a series of predictors. While we may omit reference to logs in the description, all variables in the county regression models for proportions of people with insurance are logarithmic, except for the region indicator variables, which have the value of 1 for counties in the region and 0 otherwise.
The CPS ASEC estimates for different counties are of different reliability because of the size of the sample and the proportions insured in them. Our estimates take this factor into account.
To use the information contained in the direct estimates for the approximately 1,200 counties in the CPS ASEC, we combine the regression predictions with these direct estimates implicitly using Bayesian techniques. The effect of these techniques is to weight the counties' contributions to the parameters' estimates according to the precision of their direct estimates.
We control the estimates for the counties so the following conditions are met:
state insured and uninsured sum to the CPS ASEC population for that state;
numbers of insured sum to the CPS ASEC national direct estimate of insured;
numbers of uninsured sum to the CPS ASEC national direct estimate of uninsured;
county numbers of insured sum to state numbers of insured; and
county numbers of uninsured sum to state numbers of uninsured.
We form the state-level estimates by aggregating the county-level estimates.
We provide a confidence interval, which represents uncertainty from both sampling and modeling, for each estimate.