Administrative Data

The official USDA numbers regarding WIC caseloads come from counts of the number of people who actually received WIC services in a given month. People who have been certified as eligible and thus who are “on the books” but are not receiving services for some reason are not counted. A shortcoming of the official administrative caseload data is that they are not broken out by demographic subgroups, such as age, race, and education level. To remedy this deficiency, USDA conducts a biannual survey of state program directors called the Survey of Program and Participant Characteristics. In addition to information about such participant characteristics as race and age, this survey asks detailed information about state program characteristics, which we discuss below.

Periodically, the Food and Nutrition Service (FNS) at USDA surveys a nationally representative sample of persons certified for WIC. These surveys allow FNS to assess the degree of need of WIC recipients and also to verify actual eligibility of persons certified for WIC. Most recently, the Survey of WIC Participants and their Local Agencies sampled persons certified for WIC in spring 1998.

1995–1999 CPS Food Security Supplements²

The Food Security Supplement (FSS) is one of two different supplements to the regular monthly CPS that collects data on WIC participation (the annual March demographic survey is the second). The FSS contains questions about WIC participation, but it does not have enough information on income to assess WIC eligibility. The FSS provides information about whether anyone in the household received WIC benefits in the 30 days prior to the interview. A limitation of these data is that the program participation questions are asked about the household rather than about the individual, making it impossible to determine which members of the household receive benefits. A second significant problem is that households are screened before they are asked about participation in food programs, so that only households with incomes below a certain level are asked the questions.³

Since the income screen depends on the number of persons in the household, the size of the household is critical to determining whether or not the questions are asked. In practice, the FSS uses the full number of persons in the household, whether or not these persons are related. This definition of a household may not correspond to the one that would be used by a local WIC office to determine eligibility.

This screening procedure is likely to result in the undercounting of persons receiving WIC for several reasons. First, in states with Medicaid thresholds above the income screen, some people eligible for WIC (and who receive it) are not even asked the WIC questions. Second, other eligible WIC recipients will have incomes above the screen in the first month in which a household is surveyed, but below that level in subsequent months.⁴ Working in the other direction, the FSS's use of the broadest possible measure of the household may help to mitigate the undercounting caused by the income screen because the WIC household may not include all the unrelated members of a household, but only those deemed by a WIC eligibility worker to be “sharing resources.”

A second screen was added prior to the program participation questions in 1998 and 1999. In addition to asking WIC questions to all households passing the income screen (and all those responding “don't know” or “refuse”), households who answer “yes,” “don't know,” or “refuse” to a further screening question about food insecurity are asked about participation in food assistance programs.⁵ This additional question will mitigate the undercounting induced by the income screen only if those who are missed by the income screen experience this type of problem. In order to assess the effect of this change in the screen, Bitler et al. (2002) constructed a WIC participation measure that uses a consistent screen by discarding those persons who were asked about WIC only because of the new screening question. The less restrictive screening procedure yielded additional participants (between 0.89 and 0.97 million). This provides evidence that the income screen causes some participants to be missed.

Annual Demographic File (March CPS)⁶

Starting in 1998, experimental questions on program participation were added to the March CPS. Two specific questions ask whether any females age 15 or older in the household participated in WIC in the last calendar year and the number of WIC participants in the family. In 2001, these variables were included in the publicly released data file for the first time.⁷ As in the FSS, the March CPS questions are asked only if the household's income is less than an income screen, but the income screen is generally much higher than in the FSS, and so would be expected to result in less undercounting.⁸

The March CPS offers a significant advantage over the FSS in that it measures household income and it also asks questions about participation in other programs, such as welfare and Medicaid. The latter is particularly important, since those who participate in Medicaid are adjunctively eligible for WIC, and Medicaid often has income cutoffs above 185 percent of federal poverty guidelines.

Survey of Income and Program Participation

SIPP asks about WIC each period for all households with a woman age 15–45. There are no other screens. Unlike the CPS surveys, SIPP asks which individuals in the household receive WIC. Hence, it is straightforward to estimate WIC eligibility and participation among eligible people using SIPP.

WIC and Other Transfers in the CPS and SIPP

Estimates of Participation Among Eligible People

Table 8-2 presents information on average monthly WIC eligibility and participation in 1998. In the first panel, for example, we classify all infants in SIPP in each month of 1998 into estimated eligibles and ineligibles and into those who do and do not receive WIC. For this portion of the analysis, an adjustment that increases the number of WIC recipients by the amount that the SIPP data undercounts recipients in a particular group is made, using the administrative data as the benchmark. These allocated individuals are placed in the eligible and ineligible groups in the same proportion as individuals whose status is observed in the data. A corresponding adjustment to the number of nonrecipients is made, reducing the number of eligible and ineligible nonrecipients by the increase in the number of eligible and ineligible recipients.

TABLE 8-2

WIC Eligibility and Participation Average Monthly Totals for 1998.

The first panel of the table shows that 58 percent of all infants were eligible for WIC in a given month in 1998. The WIC participation rate among eligible infants was 73 percent. We also find that, of the infants receiving WIC, 6 percent were estimated to be ineligible for the benefits. This error rate is consistent with the error rate for infants reported in the National Survey of WIC Participants (U.S. Department of Agriculture, 2001).

The second panel of the table shows a similar analysis for children. Of the 16 million children in this age group, 57 percent are estimated to be eligible for WIC. Of the 9 million eligible children, 38 percent receive WIC benefits. Of the 3.5 million children receiving benefits, we estimate that 5.4 percent do not meet the income or adjunctive eligibility criteria (and have not done so in the past six months). Thus, our evidence is consistent with that of Burstein et al. (2000), who show, using data from the 1993 SIPP, that infants are much more likely than older children to participate in the program. Indeed, Burstein et al. show that many children exit on their first birthdays, when the dollar value of the WIC package decreases (since it no longer includes infant formula).

The third panel of the table presents information on WIC eligibility and participation by pregnant and postpartum women for the first 6 months postpartum.⁹ We are not able to consider the second 6 months postpartum when only women who are breastfeeding are eligible, since we cannot observe their infant feeding practices and did not want to assume a distribution of women allocated into breastfeeding status by eligibility status. Of the 3.9 million pregnant women and women less than 6 months postpartum, 2.1 million or 54 percent are eligible for WIC. Of those who are eligible, 66.5 percent actually receive benefits.¹⁰ We estimate that 6.2 percent of the 1.5 million women receiving WIC are not eligible for benefits. We have the least amount of confidence in our estimates for women, because, as shown in the table, the WIC undercounting problem in SIPP is more severe for women than it is for other groups. Hence, our assumption that unobserved WIC recipients should be allocated to eligible and ineligible status in the same proportion as observed WIC recipients (among the two groups of women) is a bolder assumption.

The USDA has recently conducted another WIC income verification study where individuals receiving WIC were surveyed and information about their income and program participation status was collected. This study found that over all categories of eligibility, 4.5 percent of WIC participants appeared to be ineligible to receive WIC (USDA, 2001). The figures provided by Bitler et al. (2002) are quite close to the USDA estimates.

The results in Table 8-2 are striking, since they suggest that a program that served all eligible people would be considerably larger than the current one. Only 73 percent of eligible infants, 67 percent of eligible pregnant and postpartum women, and 38 percent of eligible children ages 1 to 4 receive benefits.

These participation estimates differ sharply from the implied WIC participation rates used to prepare budget estimates. Recall that the USDA methodology assumed that 80 percent of eligible persons would participate in WIC. Estimates shown here indicate that participation rates among eligible people in each eligibility category are lower than the rates obtained using the current USDA methodology. Participation rates for children, particularly, are much lower than 80 percent.

CONCLUSION: WIC participation rates among eligible persons are substantially lower than the 80 percent rate assumed in the process of estimating the number of eligible people likely to participate in WIC. WIC participation rates also vary substantially across eligibility category.

These participation rate estimates are based on eligibility estimates that differ from the estimates used to produce the current USDA eligibility estimates. Our eligibility estimates are based on the SIPP data rather than the CPS. Furthermore, our eligibility estimates use monthly rather than annual income, account for certification periods, and account for adjunctive eligibility—factors that the panel demonstrated should be taken into account so that the estimates reflect program rules regarding eligibility as closely as possible. The denominators used to estimate these participation rates (which are estimates of the number of eligible people) are thus larger than those used by the current estimation methodology.

If those who report WIC participation even though they are not eligible are included as participants, these estimated participation rates (using the same estimate of eligibility—that is, the same denominator) would increase to 78 percent for infants, 40 percent for children, and 71 percent for pregnant and postpartum women. Given these estimates, the USDA's 80 percent participation assumption is very close to the estimated participation rate for infants and not far off for pregnant and postpartum women. However, participation rates for children are much below the 80 percent assumption.

Coverage Rates Recalculated

USDA's estimated coverage rates (the ratio of WIC participants from administrative data to the estimated number of eligible persons) were reported in Table 2-1. For the past several years, those coverage rates were estimated to exceed 100 percent for infants and for postpartum women, and to range from 60 to 70 percent for children and pregnant women. The estimates of eligibility presented in this chapter imply that actual coverage rates are much lower than reported in Chapter 2. Using the 1998 estimates of eligibility based on SIPP data (Table 8-2) and the 1998 administrative total number of WIC participants from administrative data, we estimate a coverage rate of 79.6 percent for infants and 41.5 percent for children. For the same year, coverages rates based on USDA estimates of eligibility were 127.7 percent for infants and 74.4 percent for children. We did not estimate participation rates for pregnant women separately from postpartum women and so cannot estimate new coverage rates for these groups. However, coverage rates for pregnant and postpartum women based on these estimates of eligibility should also fall.

CONCLUSION: Coverage rate estimates based on currently used methods of estimating eligibility are overstated. Coverage rates for children, infants, and women based on eligibility estimates that account for monthly income, adjunctive eligibility, and WIC certification practices are substantially lower than those based on current estimation methods.

RECOMMENDATION: If participation rate estimates are used to make budgetary forecasts or to understand responses in changes to program rules and policies, separate estimates should be made for each eligibility category.

State-Level Models of WIC Participation

Results from Bitler et al. (2002) show that variations in WIC participation over the period 1992–2000 are not strongly related to changes in need, at least as measured by the unemployment rate or the poverty rate. However, demographic characteristics are important. The share of the state population that is Hispanic has a consistently large and positive effect on WIC participation rates. The share of blacks has the opposite effect. The share of births to unmarried mothers has a significantly negative effect on the probability that children participate. The programmatic variables indicate that there is no strong relationship between WIC participation and AFDC/TANF participation. However, higher AFDC/TANF benefits are associated with lower WIC participation rates, perhaps because the larger TANF benefits offset the need for the relatively smaller WIC benefits. The Medicaid enrollment rate is positively associated with WIC participation among children but negatively associated with participation for infants.

Features of the way that WIC programs are administered across states are also correlated with participation levels in states. The cost of the women's food package is positively correlated with participation for children, meaning that the higher costs of women's food packages are associated with higher levels of participation for children, and negatively correlated with participation for infants, meaning that the higher costs of women's food packages are associated with lower levels of participation for infants. This coefficient is difficult to interpret, however, given that if a woman does not breastfeed, the value of her food package is smaller than if she does, but the value of her infant's food package increases because of the addition of formula given to her infant. Three other characteristics that relate to the strin-gency with which the programs are operated are negatively correlated with participation: dispersing WIC benefits monthly (as opposed to less frequently, which means fewer visits into the WIC office), requiring proof of income, and having a higher nutritional risk cutoff for pregnant women.

These models were estimated with state fixed effects—that is, a dummy variable for each state was included in the model to control for unmeasured differences between states. The results indicate that there is considerable variation in total WIC participation rates across states, even after controlling for all the variables included in these models. Differences may reflect important unobserved differences in the way that the program operates across states and suggest that further information about how the program is operated might be useful in explaining WIC participation.

Individual-Level Models

Individual-level data from the March CPS and SIPP were used to examine individual characteristics associated with WIC participation among eligibles. Several findings are consistent across both data sets. First, Medicaid enrollment is strongly linked to WIC participation. Results based on the SIPP data show that Medicaid participants are 50 percent more likely to participate in WIC than those not enrolled in Medicaid. Food stamp participation is also positively associated with WIC participation, but the association is not as strong. Results from both data sets also show that Hispanics are more likely to participate in WIC than whites, and Asians are less likely to participate than whites. The mother's education level is negatively associated with WIC participation (i.e., more educated mothers are less likely to participate). This finding may reflect a lack of awareness among some more educated women of their eligibility (i.e., because income eligibility levels for low-income assistance programs are not typically as high as 185 percent of poverty), or a higher opportunity cost of participating in the program among the more educated.

These analyses of WIC participation suggest several tentative conclusions. First, WIC participation does not seem to be strongly correlated with state-level indicators of economic need, such as poverty and unemployment rates. Given the WIC income eligibility cutoff of 185 percent of federal poverty guidelines, it is possible that many families who fall into poverty as a result of an economic recession were already eligible for WIC. Second, WIC participation is strongly associated with individual demographic characteristics, such as education, race, and marital status, even after conditioning on income. For example, eligible Hispanics are more likely to participate in WIC, while eligible Asians are less likely. These findings could, for example, be used as an indicator that outreach targeted toward Asian women might be beneficial. Third, WIC program characteristics may play an important role in WIC participation. In general, factors that increase the transaction costs of applying for WIC (e.g., monthly distribution of benefits compared with quarterly distribution) are associated with reduced participation. This type of analysis could be conducted in a state using local-level data. Such an analysis could give state-level officials a better idea of how administrative changes in the programs could be made in order to achieve program goals.

These results have important implications for the process of forecasting future WIC participation levels. On one hand, if the economy does not have a huge impact on levels of participation, then it may not be crucial to account for changes in the economy when forecasting participation levels into future years, although changes in the economy will affect the number of people who are eligible for WIC. On the other hand, participation rates differ significantly across demographic groups. Demographic changes among groups with high propensities to participate could affect the overall participation rate. Although demographic changes can be slow to occur, such changes could introduce uncertainty in the forecasted counts. Results from these analyses also suggest that changes in program rules or administrative practices are associated with changes in participation rates. Thus, another source of uncertainty in forecasted estimates of participation is introduced if states change the way they run their WIC programs or if changes in program rules are implemented. The next chapter discusses the implications of these results while comparing different methods for forecasting WIC participation.