For every step in the simulation process, our findings indicate that there is at least some variation across the states. Often, there is variation across programs within a state as well. Replicating this variation in simulation algorithms presents an enormous challenge to researchers and one that must invariably involve many simplifications. The data requirements alone are sizable. Other aspects of the simulation of Medicaid and SCHIP eligibility may be difficult to implement, either because a detailed state-by-state and program-by-program description is unavailable, or because the task of coding each nuance demands more resources than most research teams can afford to allocate to the effort.
Determining the family unit continues to be one of the most complex steps, both because of its data requirements and because the rules vary both across and within states; these rules remain among the most difficult to document. Furthermore, the counting or not counting of individual family members (and their income) in calculating the poverty level of a child's family can be the determining factor in whether or not a child is deemed eligible for a particular program. A further complication is introduced by program rules that count unborn children in determining family size. Since information on pregnancy is not collected in two of the major surveys, taking account of this aspect of eligibility in a simulation may require data that simply are not available.
The anti-crowd-out features of the programs developed under S-SCHIP impose important new demands on simulation models--ones that will be difficult to satisfy. Only panel surveys provide the data required to simulate waiting periods, and panel data generally lack the timeliness that is desirable for estimates of eligibles.
The fact that many states seem to be using gross rather than net income to establish eligibility under their S-SCHIP programs implies that simulating eligibility under the full array of programs within a state may be easier than simulating eligibility under Medicaid alone. At the same time, however, the S-SCHIP programs appear to be introducing other complications into the simulation process, if only through their variability across states.
Clearly, to be credible, eligibility simulations will have to be detailed. At the same time, however, our review of the data collected by five major surveys indicates that the simulations will have to incorporate major simplifications to be feasible. No single data source contains all the elements necessary to mimic Medicaid and SCHIP eligibility determination with a high degree of precision. How simulations of Medicaid and SCHIP eligibility can be adapted to the data limitations, while, at the same time ensuring their value as a policy tool, remains a considerable challenge.