Estimation of Uninsurance Rate: Comparison of Four Estimation Methods

Abstract

Objective: Although high percentage of the uninsured is an important public policy issue, the discrepancies in both state and national estimates of the numbers of uninsured are reported. This study compares four advanced estimation methods for uninsurance, by using Florida Health Insurance Survey data as an example. Design: The four predictive models include decision tree, neural network, general logistic regression, and two-stage logistic regression. Risk factors to uninsurance are identified. Population: The study sample comes from the Florida Health Insurance Study data collected for the Florida Agency of Health Care Administration in 1999, representing the first large-scale study designed exclusively to provide reliable statewide information on health insurance coverage. All 67 counties in Florida were surveyed; 14,000 households and 37,120 persons were interviewed.

Findings: The uninsurance rate in Florida was estimated here to be 16.65%. The two-stage LR model was found to be the best model for the imputation of missing data on health insurance. The results show that parametric methods—LR and network—fit the data better than the nonparametric method—tree model—does. The cost of copayments and having regular or routine care are two major predictors of health insurance status. Conclusions: The two-stage LR model for imputing health insurance serves as a demonstrative method, not a gold standard for substitution for missing data.

Implications: Understanding both the precise number of the uninsured and their characteristics of distribution would improve the efficiency of the Medicare and Medicaid safety net.

Date Created

November 2006

This document is currently not available here.

Share

COinS