The estimates presented in this series of technical appendices are accompanied by two caveats. First, they are based on a panel sample--that is, a sample of children who were followed over a (three-year) period rather than a sample of children observed at a single point in time. This has implications for what population the sample represents at any given point over the period that we observe it. Second, they indicate too few children covered by Medicaid, based on comparisons with Medicaid program statistics. Both of these caveats have implications for the overall number of children and the proportion of all children that we estimate to be uninsured at any one time or over any period of time. In this brief report we consider how our estimates of the uninsured might be changed if we could expand the representation of the SIPP panel and obtain complete reporting of Medicaid coverage. We conclude that complete reporting of Medicaid coverage would reduce the 1992 SIPP panel estimate of the percentage of children who were uninsured in September 1993 from 13.0 percent to between 10.2 and 10.8 percent and would reduce the estimated percentage of children who were uninsured in September 1994 from 12.6 percent to between 9.5 and 10.1 percent.
Representativeness of the SIPP Sample
At the time of the initial interview, the children in the SIPP sample represented the entire United States population of children except for those residing in institutions or military barracks.1 Over time, however, as the U.S. population of children under 19 grew to 74.0 million, the population of children represented by the SIPP longitudinal sample declined to less than 70 million. As we detailed in Technical Appendix A, except for births the SIPP sample does not represent children who entered the population after the initial interview--that is, children who migrated to the U.S., returned from abroad, or left institutions. Moreover, while the SIPP sample theoretically represents all children born to members of the initial population who remained in the U.S., we demonstrated that these births are underrepresented by as much as a quarter. It is this shortfall of sample births that accounts for the sizeable decline in the population of children that the SIPP sample represents.
The fact that most of the divergence between the SIPP population and the total U.S. population of children is concentrated in the youngest ages carries important implications. As we have seen, the health insurance coverage patterns of very young children are strikingly different from those of other children. Underrepresenting births will very likely affect the overall distribution of coverage, increasing the proportion uninsured and reducing the proportion covered by Medicaid
The fact that the SIPP sample does not represent immigration has implications for estimates of insurance coverage as well. We would expect the rates of insurance coverage to be lower among immigrant children than among the general population of children. Therefore, excluding the immigrants will lower both the absolute number and the proportion of children who are estimated to be without insurance relative to what we would observe with a sample of the total population at the same point in time.
In order to evaluate our SIPP estimates of Medicaid enrollees against those reported by the Health Care Financing Administration(now known as Centers for Medicare and Medicaid Services(CMS)) (HCFA(now known as CMS)), we must expand our estimates of Medicaid coverage to the entire population. This entails first estimating the number of children that we are "missing" due to the SIPP panel's underrepresentation of births and its non-representation of other additions to the population and then estimating the frequency of Medicaid coverage and lack of insurance among these additional children. Our derivation of adjusted population totals, estimates of uninsured children, and estimates of children enrolled in Medicaid is summarized in Table 1.
For September 1993 we added 1.150 million children to correct for the underrepresentation of births and another 1.119 children to account for net growth of the population of children. For September 1994 we added 2.250 million children to correct for the underrepresentation of births and another 1.772 million children to incorporate other new entrants to the population.
To derive estimates of uninsured children and Medicaid enrollees among the children added to correct for underrepresented births, we applied age-specific rates from two tables that were presented in Technical Appendix C. To derive estimates of uninsured children and Medicaid enrollees among the other new entrants to the population, we assumed that these children shared the same probabilities of being uninsured or covered by Medicaid as all children in 1993 and 1994, and we applied rates from a table presented in Technical Appendix A. This is not an entirely satisfactory assumption, given that these new entrants to the population could differ dramatically from the rest of the population. Nevertheless, in the absence of relevant data on this population, the assumption is a reasonable one to apply in this case, and the final results are not very sensitive to it.
The net effect of these operations is to add about 700,000 children to the number of children estimated to have been ever enrolled in Medicaid during FY93 and 1.27 million to the number estimated to have been ever enrolled in Medicaid in FY94. Children added to correct for the underrepresentation of births accounted for about two-thirds of the additional Medicaid enrollees in each year.
Undercount of Medicaid Children
Crude comparisons of CPS estimates and Medicaid program statistics suggest that the CPS underestimates by a substantial margin the number of Medicaid enrollees for the population as a whole. Researchers at the Urban Institute have made a practice of adjusting their estimates of uninsured children and adults to compensate for what they interpret as incomplete reporting of Medicaid coverage by survey households. The effect of these adjustments is sizable. When applied to data for 1994 and 1995 they lowered the estimated number of uninsured children from 10.6 million to 7.6 million and reduced the estimated number of Medicaid-eligible uninsured children from 4.5 million to 1.6 million (Ullman et al. 1998).
Evaluating the Medicaid coverage of the SIPP or the CPS is, in fact, both difficult and problematic. It is difficult because the program administrative statistics are very limited, which complicates the task of creating administrative and survey estimates that apply to the same universe. One major problem, discussed in subsequent sections, is that survey data generally represent points in time while federal administrative data include those covered by Medicaid at any time throughout the year. Beyond that, the age categories for which Medicaid statistics are reported do not lend themselves to identifying children in the way that they are commonly defined in policy analyses--that is, under 19 or under 18. Part of the problem is that the national statistics are assembled from reports submitted by the states, and the state data differ in quality and in some of the conventions that they apply. In 1993 more than half of the state reports were submitted in hard-copy, giving HCFA(now known as CMS) no flexibility to generate additional tabulations--with alternative age categories, for exmple.
Evaluating Medicaid coverage is also problematic. Even if the program data and survey data could be matched with respect to universe, simply comparing them would not tell the whole story. Because the Medicaid statistics come from separate state systems they are not unduplicated across states. If a child moves from one state to another and is enrolled in Medicaid in both states during the same fiscal year, the child will be counted in both states' statistics. In a survey, of course, this same child would be represented as a Medicaid recipient in only one state. Another factor confounding the comparison of survey and administrative estimates is the potential misreporting of Medicaid coverage as something else. Survey respondents may not be fully aware that the source of their medical coverage is Medicaid. States have developed alternative names for their programs and issued health insurance cards that resemble those issued by private insurers, perhaps persuading some Medicaid enrollees that their coverage is from a private source. Finally, a number of states run affiliated programs that do not receive federal reimbursement and whose enrollees, therefore, are not counted in the Medicaid program statistics. If such persons report their coverage in surveys as Medicaid, as seems likely, or if the Census Bureau interprets it and edits it as such, then the survey estimates of Medicaid coverage will include persons who are not counted in the administrative totals.
Impact of the Medicaid Undercount on Estimates of the Uninsured
Determining how the Medicaid undercount affects estimates of the uninsured is even more problematic. Apart from the issue of whether Medicaid coverage may be reported as something else is the matter of other coverage in general. If a child did not report Medicaid and did not misreport it as something else, the child might still have had--and reported--another source of coverage.2 To estimate the impact of even a known Medicaid undercount on the number of children who are estimated to be uninsured requires assumptions about how the children who failed to report their Medicaid coverage may have reported their insurance status with respect to other potential sources of coverage.
Estimates of the Medicaid Undercount
Based on a comparison of the adjusted Medicaid enrollment estimates in Table 1 with estimates derived from program statistics, we determined that the undercount of Medicaid enrollees in the 1992 SIPP panel was 13.2 percent or 2.8 million children in FY93 and 14.6 percent or 3.3 million children in FY94 (Table 2).3
Attributing all of the difference between the survey and administrative estimates to a Medicaid undercount presumes that the administrative count is in fact accurate. While we have reason to believe that there is some double-counting of Medicaid enrollees, HCFA(now known as CMS) has advised users of its program data that participating infants may be undercounted by about 10 percent in as many as half the states (HCFA(now known as CMS) 1994). Furthermore, we have not attempted to estimate the number of children enrolled in fully state-funded programs. These children are not included in the statistics reported to HCFA(now known as CMS) by the states, but they are included, to at least some degree, in the survey estimates of the Medicaid population. On balance, then, we have assumed that these errors and omissions in the administrative statistics net to zero.
Estimates of Uninsured Children
To estimate the impact of the Medicaid undercount on estimates of the number of uninsured children, we must work with estimates of children uninsured at a point in time rather than children ever uninsured during the year. Having Medicaid (or any other form of coverage) during a year does not imply that a child was insured throughout the entire year. Therefore, the impact of a Medicaid undercount on estimates of the number of children who were ever uninsured is too ambiguous to estimate.
Our estimate of the Medicaid undercount represents children who were ever enrolled in Medicaid during a year. While the proportionate undercount need not be the same for the annual-ever and point-in-time statistics, we assumed that it was. In Table 3 we derive estimates of the size of the SIPP Medicaid undercount for September 1993 and 1994.4 We obtained 2.120 million as the estimated number of children who were covered but did not report Medicaid enrollment in September 1993 and 2.445 million as the number who were covered but failed to report their Medicaid enrollment in September 1994.
What proportion of the children who failed to report their Medicaid enrollment also failed to report any other kind of coverage? As we discussed earlier, some children (or their parents, in most cases) may have misreported their Medicaid coverage as another form of public or even private insurance. Others may have even had overlapping coverage and reported the other coverage rather than Medicaid.5 With no evidence to support specific assumptions about these proportions, we applied three alternative assumptions. We assumed that the proportion of unreported Medicaid enrollees who were classified as uninsured was either 95 percent, 85 percent, or 75 percent. These different assumptions may be considered to include our uncertainty about the incidence of duplication of individuals in the national Medicaid statistics.
The alternative assumptions imply that between 1.6 million and 2.0 million children were incorrectly counted as uninsured rather than insured in September 1993, and they lower the estimated number of uninsured children to between 7.5 million and 7.9 million--compared to the SIPP panel estimate of 9.5 million. Taking into account the Medicaid undercount implies that between 10.2 and 10.8 percent of children under 19 were uninsured in September 1993, compared to the SIPP panel estimate of 13.0 percent. For September 1994 the higher estimated Medicaid undercount yields estimates of the number of uninsured children that range between 7.0 million and 7.5 million compared to the SIPP panel estimate of 9.3 million. The alternative estimates of the proportion of children without health insurance range from 9.5 percent to 10.2 percent versus the SIPP panel estimate of 12.6 percent.
1. As a household-based survey, the SIPP also excludes the homeless, although homeless children are included in the population totals to which the SIPP sample weights are controlled. In this sense, then, the homeless are counted in the population estimates that we present, but their characteristics are not represented. If homeless children are more likely to be uninsured or more likely to be enrolled in Medicaid than children in the household population, for example, the frequency of these characteristics in the total population will be understated.
2. While we speak as if the child is the one who gave the responses, the CPS and SIPP collect data on children under 15 from other members of the household.
3. The HCFA(now known as CMS) data are reported for groups of ages, with ages 15 to 20 included in a single category. Since our SIPP estimates apply to children under 19, we had to estimate the proportion of children 15 to 20 who were under 19. We tabulated the March 1994 CPS to obtain the distribution of Medicaid enrollees under age 21 by single year of age, and based on these data we apportioned the HCFA(now known as CMS) count of enrollees 15 to 20 into ages 15 to 18 and ages 19 to 20. States use at least two different conventions for reporting age to HCFA(now known as CMS). States that participate in the Medicaid Statistical Information System (MSIS) and submit electronic case records to HCFA(now known as CMS) assign age as of the end of the fiscal year. States that submit 2082 reports instead of MSIS electronic case records assign age as of the middle of the fiscal year. The latter system classifies about 50 percent more children as infants and reports more persons under any given age than does the first system. (States using the first system may also exclude children born in the final month or two of the fiscal year, according to HCFA(now known as CMS).) Depending on whether a state appeared to have a substantial excess of infants or not, we applied a different algorithm to estimate the number of Medicaid enrollees who were under 19 in each state and then aggregated these to obtain the national numbers shown in Table 2.
4. To obtain the estimated undercount for 1993, we divided the adjusted SIPP estimate of children enrolled in Medicaid in September 1993 by one minus the percentage undercount to obtain the implied total Medicaid enrollment in that month, from which we then subtracted the adjusted SIPP estimate.
5. In assigning insurance coverage each month, we classified a child as enrolled in Medicaid if Medicaid was reported in combination with any other type of coverage.