of this volume). There are zeroes for those who do not use health care services or have very generous policies. For those with any out-of-pocket spending, their spending is skewed, but not as skewed as total expenditures. The skewness is probably large enough to require attention. Whether the statistical methods used for total expenditures would be the same as for out-of-pocket expenditures either in degree or kind is not known.
We do not know which variables matter for out-of-pocket medical care spending. We suspect that chronic disease and age will matter because prescription drugs may not be fully covered by insurance. Because the fraction of the population with mental health or dental coverage is less than that with medical care coverage, one would expect that mental health status and oral health would affect out-of-pocket spending for medical services.
To get at the variability in the burden, one can look at the retrospective burden variance within cells, or one could employ multivariate methods for the variance of expenditures conditional on characteristics. Although models for conditional means of expenditures given characteristics of the family and its members are common at the person level, they are not at the family level. This may require some work for the variance function. But given the skewed characteristics of out-of-pocket medical care spending, it will probably not be sufficient to look at the mean and variance of such spending and total health expenditures at the family level. One would need to observe responses to deductibles and stop-losses to assess the impact on out-of-pocket spending. This has been done by Keeler and colleagues (1988) for the RAND Health Insurance Experiment. But that study required much more extensive modeling of behavior than one would expect the U.S. Census Bureau to do, or it would require more detail on insurance plan details than is commonly available.
As noted, for the mean family burden, one has only to add the conditional means for the individuals as long as the individual means condition on family composition and income. But for the distribution of out-of-pocket medical care spending or its variance, it is more complicated than keeping track of means, variance, and covariances among the family members.
Because there is much to learn about the drivers of out-of-pocket medical care spending for families of varying size, composition in terms of ages, health status, insurance coverage, and resources, we recommend a series of analyses based on MEPS to test out various alternatives and to answer such questions as what factors (e.g., chronic conditions) add to the predictive value of previous spending for future out-of-pocket spending. Such analyses are also needed to answer a series of questions about which approach to use in modeling a family’s out-of-pocket medical care expenses and risk as a function of individual characteristics (age, gender, health status) and family characteristics (income, insurance status—which may vary by family subunit). Past research on mean or adjusted health expenditures indicates