out-of-pocket medical care spending must logically be defined at the start of the year in that data set. So, cells cannot be defined by current spending because that would produce overly small observed variation in spending. Similarly, health characteristics and the risk adjustors based on them that predict spending may be the result of health shocks throughout the year and not defined at the start.5 In most data sets, some covariates are measured before and some measured afterward. For example, one typically knows income at the end of the year, not before.
If in the past year a given percentage of families had out-of-pocket spending for both premiums and care received, then one could use data on the expenses incurred to say that families with certain characteristics were more or less likely to fall below the poverty threshold last year. If the world were in a steady state—that is, there were no changes in the general cost of care, insurance plans, mandates, or the business cycle—then that retrospective analysis would provide a consistent prediction as long as the covariates were measured at the start of the year. Two-year panels solve this problem by using first-year information to predict second-year behavior.
In the CPS ASEC, one could use also logistic regression of an indicator defined as out-of-pocket medical care spending greater than or equal to the difference between SPM-adjusted income (without the subtraction of out-of-pocket spending) and the SPM family characteristics. The same caveats on when predictor variables are measured would apply.
An Initial Retrospective Measure of MCER
In the short term, with the data now being collected, the CPS ASEC could be used to report the burden of out-of-pocket medical care spending retrospectively, roughly 10 months after the end of the calendar year for which income and spending are reported. Furthermore, with additional assumptions, the retrospective measure of burden could serve as a proxy for the prospective MCER: for example, if x percent of families and individuals were moved into poverty this year, then the same x percent is the best estimate of those who will be in poverty next year, assuming no other major policy initiatives or differences in the business cycle.6
5 If one develops risk adjusters for health conditions based on 1 year of health experience and uses that experience to explain expenditures for that year, one would arrive at a biased assessment of the variance because the covariates are not independent of the out-of-pocket spending (see Manning, Newhouse, and Ware, 1982).
6 The preliminary analysis of MEPS, discussed below, could help to identify which family characteristics were most important in predicting out-of-pocket medical care expenditures. Instead of relying on parametric models, the probability of a family being at or near poverty could be determined empirically if risk cells were based on particular family or individual characteristics.