Estimating Insurance Spell Dynamics Using Longitudinal Survey Data
*John A. Graves, Harvard University & Vanderbilt University 

Keywords: Insurance, Survival Analysis, Health Reform

Most health policy research on insurance coverage utilizes point-in-time measures from large household surveys. When considering the impact of major health care reform on insurance coverage, it is also important to think about coverage changes from a dynamic perspective. Obtaining accurate estimates of insurance spell durations is difficult due to the methodological challenges researchers face when using available longitudinal survey data sources. These issues, including the influence of sample attrition, seam bias, censoring, truncation and misclassified spells, are in addition to the well-documented challenges of using survey data to obtain accurate measures of insurance coverage. In this study I provide methodological guidance on each of these issues using recent data from the Survey of Income and Program Participation (SIPP) and the Medical Expenditure Panel Survey (MEPS). Using data on uninsured non-elderly adults, I non-parametrically estimate the distribution of uninsurance spells in the United States between 2001 and 2007. Using the MEPS I find median spell durations of 13 months, compared to 8 months in the SIPP. I also show that efforts to reduce seam bias in the SIPP result in estimated spell durations that are 4 months longer during a time period where estimated MEPS spell durations are constant. Researchers combining multiple waves of the SIPP to study event durations should therefore take caution in interpreting estimates, since estimates could reflect changes in survey design and not changes in the population.