Four of the federal government's national household surveys may appeal to users as potential data sources for estimates of the number of children who are eligible for Medicaid and SCHIP plans over the next several years:
- The March supplement to the Current Population Survey (CPS)
- The 1996 and later panels of the Survey of Income and Program Participation (SIPP)
- The National Health Interview Survey (NHIS), which underwent a major redesign in 1997
- The Medical Expenditure Panel Survey (MEPS), which was initiated in 1996, with additional panels scheduled to start in subsequent years
A fifth, privately sponsored survey, the Urban Institute's National Survey of America's Families (NSAF), will also make important contributions to national estimates of Medicaid and SCHIP-eligible children, although the sample design of that survey, with its focus on 13 states, may lead researchers to place greater emphasis on estimates for those 13 states. The first NSAF was conducted in 1997, a second wave was fielded in 1999, and a third wave is planned for 2001.
Another important survey is the Community Tracking Study (CTS), funded by the Robert Wood Johnson Foundation (RWJF) and conducted by the Center for Studying Health System Change. The CTS household survey was fielded in 1996-1997, with a second round in 1999. The household survey sample of more than 60,000 respondents was selected from 60 randomly chosen communities representing about half the U.S. population plus a small, nationally representative supplemental sample designed to improve the precision of the sample for national-level estimates. About half of the 60-site sample were selected from 12 communities, to enable site-level analysis for these localities. Because of the community focus of the sample, the CTS is not an obvious choice for national- or state-level simulation of Medicaid or SCHIP eligibility based on the application of state eligibility rules, we have not included the CTS in our comparison of surveys.
Table IV.1 compares the five surveys with respect to their collection of the data elements needed to simulate children's eligibility for Medicaid and SCHIP in the 10 states we studied. The data elements include state of residence, family income by source, income by family member, the income reference period, resources, program participation, expenditures, family composition, characteristics of children, immigration status, and health insurance coverage.
We note right away that neither the NHIS nor the MEPS public use files identify state of residence, making it impossible for public users to incorporate the state variation in eligibility discussed earlier. Unless provision is made to give researchers access to state identifiers under arrangements that will maintain the security of these data, it will be difficult for NHIS or MEPS to make a major contribution to national (let alone state-level) estimates of SCHIP eligibility. SIPP restricts the identification of a small number of states, due to confidentiality concerns, while the NSAF policy on state identification outside the 13 study states remains unclear. If the state of residence is not released with the NHIS or MEPS files, the major value of those databases for Medicaid and SCHIP simulation may be as sources of ancillary information to support the imputation of variables not collected in CPS or SIPP.
It is evident that, even if reporting the state of residence were not an issue, no single survey captures the full range of information needed for a relatively complete simulation of Medicaid and SCHIP eligibility and that each of the surveys possesses certain comparative strengths, along with weaknesses. For example, data on expenditures and assets are weak generally. While all the surveys identify aliens, none differentiates legal from undocumented aliens--an important distinction for SCHIP eligibility.
The CPS has been, and will almost surely continue to be, the chief source of estimates of Medicaid and SCHIP eligibles, owing to the timely release of March CPS data and the fact that CPS has become the official source for the measurement of family income and poverty in the United States. CPS has nearly achieved that status with respect to the measurement of health insurance coverage as well. Yet CPS has several notable limitations. It contains no data on assets or expenditures, leaving researchers either to impute the missing amounts or to simulate their impact more crudely, or to exclude them from consideration altogether. For instance, it is common for simulations of Medicaid eligibility based on the CPS to provide no estimates of children or adults who are eligible under the medically needy provisions. As an increasingly large proportion of children become eligible under poverty-related expansions and SCHIP extensions, eligibility under the medically needy provisions becomes less and less important for children, although this program remains important for adults and families.
A less obvious limitation of the CPS, but one that is nevertheless important, is its reference period. Except for earnings, for which CPS collects current information, all the income data collected in the survey refer to the previous calendar year. Since Medicaid and SCHIP eligibility are determined on the basis of monthly income, and many of those who qualify have fluctuating incomes, rather than persistently low ones, researchers using CPS must choose among some less-than-perfect options. One option is to calculate eligibility based on one-twelfth of annual income, recognizing that this will understate the number of children eligible in an average month. Another option is to modify the first strategy by adjusting the results to reflect the relationship between monthly and annual poverty that is observed in other data, such as SIPP. A third approach is to simulate monthly income streams for different sources, relying on the number of weeks respondents report having worked in the previous year, to determine the number of months with few or no earnings and to integrate the streams for different sources, based on assumptions or external evidence (such as from SIPP) about their covariation. Eligibility can then be simulated for a selected month. Both the Urban Institute and MPR employ this last approach in their respective CPS-based microsimulation models. Regardless of the approach taken, the reporting of health insurance coverage in annual rather than monthly terms makes it unclear how best to align reported coverage with simulated eligibility for Medicaid and SCHIP.
SIPP provides monthly data on income, family composition, and insurance coverage; also it allows the simulation of waiting periods and collects asset data and expenditure data annually. Finally, SIPP provides the most extensive coverage of the full range of variables reported in Table IV.1. Where SIPP continues to fall short of the ideal most conspicuously is in the timely release of data. While the Census Bureau is currently completing the fourth and final year of data collection, only the first 16 months of data have been released.
To date, NHIS has not been a popular vehicle for estimating Medicaid eligibility. While changes to make the data available sooner are occurring, NHIS lacks many of the items identified in Table IV.1, and, as we reported earlier, state identifiers are not released with the public use files. The principal value of NHIS for simulating Medicaid and SCHIP eligibility may lie in its capture of health insurance coverage--arguably better than CPS, because it captures coverage at the time of the interview, so that reported coverage is not subject to recall error or misinterpretation of the reference period. In addition, NHIS captures detailed information on health status, activity limitations, and access to care.
Because of the aforementioned lack of state identifiers, the MEPS appears to hold greater promise as a source of data for imputing information not captured in SIPP or CPS than as a direct source of estimates of Medicaid and SCHIP eligibility. The comparative strengths of MEPS include the breadth of its information on health care expenditures and employer coverage. In addition, because MEPS uses a different approach than SIPP to capture changes in coverage over time, MEPS data provide a potentially important source for validating some of the health insurance dynamics observed in SIPP.
Finally, NSAF has some of the same limitations as CPS with respect to the reference period for the reporting of income, and it collects fairly limited data on expenditures and resources. These limitations suggest that users of the NSAF would have to employ more simplifications than would be needed with SIPP to simulate eligibility for Medicaid or SCHIP. Its chief value lies in the comparative information it will provide for the 13 states that are represented with very large samples.