represent medical care economic risk. If the cells are big, say 200 cases, then experience within the cell will be a good estimate over the entire distribution. If the cells are small or if one wants to know more extreme tails, one will have to model them, or combine experience on rare events from many cells.
A PROSPECTIVE APPROACH BASED ON 2 YEARS OF DATA
Because this feasible set of calculations based on the annual CPS ASEC is somewhat informative, why should we continue to pursue construction of a prospective measure of MCER? What could be its added value? With its richer data on health conditions, distribution of medical care spending by service type, and 2-year panel, MEPS offers the opportunity to learn much more about the interplay of health status, health insurance, and out-of-pocket medical care spending with respect to family finances as well as to more accurately assess how risk varies with health. Over the next several years, as the landscape of health insurance coverage in the United States undergoes substantial change, understanding the underlying drivers of families’ choices of insurance coverage and their out-of-pocket health care spending and the effects on their resources will be extremely important.
With 2 years of data, as are available from MEPS, one can employ multivariate regression methods to develop predictions about expected outcomes or their distributions. The difference between this and the cell-based approach is that, with cells, one does not share information across different groups. In regression, however, the estimated model has a more limited specification and shares information across observations, under the assumption that the response to individual covariates can be jointly modeled. In reality, these methods are not exclusive alternatives. With limited data or if one combines responses across individuals in the same family, acquiring meaningful detail on risk may require a mixed approach.
Another alternative is to use data on second-period expenses and base-period characteristics together with multivariate regression methods to estimate the probability that a family with given income and resources, family composition, and health will have an expenditure large enough to push the family to the poverty threshold. In the absence of sufficient research on the distribution of out-of-pocket costs relative to SPM thresholds, it will be necessary to do that work empirically.8 For example, one would expect that a working poor family with one or more members in fair or poor health might have a substantial risk even without a hospitalization or high-cost drug regimens. An emergency department visit or a flare-up of a chronic condition might be enough to drop the family below the threshold.
8 See the recommendations for research in this area on the following pages.