is very high. The cell-based approach can be seen as people getting together so that those who are fortunate in health, with low out-of-pocket spending, subsidize their unlucky neighbors.

Cell-based approaches group similar people or families into cells, and then they use the medical care spending experience of the members of a cell (this year’s experience to create not only a measure of burden, but also one of risk) as a proxy for the range of possible outcomes for each member of the cell (next year’s risk). If all the families in a cell are equivalent ex ante, as reflected in base-period health status, demographic characteristics, insurance coverage, and income and other resources, then the average of their experience on out-of-pocket expenditures is an estimate of burden. For risk, one can use the observed dispersion across the families within the cell or estimate the probability that some family reaches one of the common poverty thresholds (50, 100, or 200 percent of poverty). The observed probability of an out-of-pocket expenditure sufficient to take the family below the poverty threshold is an estimate of the risk for each member of the cell because the cell is homogenous in terms of observable characteristics and risk adjustors. The advantage of this method is that it needs only 1 year of data, which has two benefits—timeliness and allowing the use of nonpanel data like the CPS ASEC.4 A disadvantage is that because nonpanel data sources systematically exclude recent deaths and those who have entered institutions in the immediate past time period—two groups known to have high health expenditures—it will be necessary to use other data sources and the relevant literature to provide an estimate of the missing information for those two transitions and their impact on out-of-pocket medical care spending. Although decedents and institutionalized people are not in poverty, the transitions to death and to institutions will often impose major drains on their families’ resources and could push other members of the household into poverty.

The cells for the retrospective measure must be formed on the basis of characteristics that predict spending. These characteristics and their weights used to build cells typically come from preliminary analysis using a regression approach that calculates an individual’s expected payments based on observable characteristics in a prior year (including diagnoses or other health information) from other data sources, such as MEPS. A problem is that, to actually produce the estimates of retrospective MCER from a data source such as the CPS ASEC, the characteristics that predict


4 As discussed in Chapter 5, the CPS employs a panel sample design in that monthly samples rotate in and out of a sample on a schedule that ensures that 75 percent of the sample addresses in a given month were included in the previous month’s sample, and 50 percent were included in the sample 12 months earlier. The purpose of this design feature is to reduce the sampling variance for estimates of month-to-month or year-to-year change—not to enable longitudinal analysis. The limitations of the CPS ASEC for longitudinal analysis—and why we do not propose such use here—are explained in Chapter 5.

The National Academies of Sciences, Engineering, and Medicine
500 Fifth St. N.W. | Washington, D.C. 20001

Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement