Data and sample
The sample was drawn from the U.S. National Health Interview Survey (NHIS), an annual cross-sectional survey (see sample flowchart, Supplemental Fig. 1). We used the 1998–2017 waves of NHIS, since surveys prior to 1998 did not contain sufficient information on family relationships, and 2017 was the most recently available when analysis began. While demographic data are collected by the NHIS for all household members (N = 1,989,149 for 1998–2017), this study was restricted to women who reported having children under the age of one in the household to capture the postpartum period (N = 172,903) and to children under five (N = 70,047). Although pregnant women are also eligible for WIC, information on pregnancy was not regularly ascertained by NHIS throughout the study period, and our results are thus only generalizable to postpartum women. Additionally, women are only eligible for WIC during pregnancy and up to six months postpartum (for non-breastfeeding women) or one year postpartum (for breastfeeding women). Thus, restricting our sample to those with children under the age of one captures a likely eligible population. Next, we restricted the samples to those for whom data on WIC receipt was non-missing (N = 145,119 women and 61,133 children). We additionally restricted the samples to those whose household income was less than $75,000 to ensure that the control group of non-recipients was more comparable to the treatment group (N = 81,771 women and 30,798 children). For children, we observed that outcome trends were not parallel in earlier time periods, violating an assumption of difference-in-differences analysis (see additional details below). Thus, we additionally excluded observations which occurred more than 100 months prior to when the revision was implemented (N = 27,780). Not all outcomes were collected for all women or children, resulting in variation in the number of individuals included in each analysis (Supplemental Fig. 1).
The main exposure was whether women or children received the revised WIC food package. The implementation of the WIC revision was staggered across states throughout 2009, with some states implementing the changes as early as January 2009 and others as late as November 2009 . The NHIS assessed whether an individual had received WIC benefits via self-report. Among WIC recipients, we considered women and children to have received the revised food package if they reported WIC receipt after the revision was implemented in their state.
For women, we assessed two measures of physical health. First, we evaluated change in self-reported health over the prior year, which was assessed by the question, “Compared with 12 months ago, would you say your health is better, worse, or about the same?” We dichotomized this variable as better or about the same versus worse. This question has been previously included as a part of the Short-Form Health Survey, which is a well-established and validated instrument [18, 19]. Because improvements to dietary quality (e.g., increased consumption of vegetables) have been found to be positively associated with self-reported health [20,21,22], we hypothesized that women who received the healthier revised WIC food package would have better self-reported health. Second, we assessed body mass index (BMI), which was calculated using a woman’s self-reported height and weight. Prior research has demonstrated the benefits of the revised WIC food package on reduced gestational weight gain , as well as breastfeeding , which is associated with reduced postpartum weight retention. Furthermore, excessive weight gain during pregnancy has been found to be predictive of long-term obesity in women [24, 25]. Therefore, we hypothesized that receiving the healthier WIC food package could manifest in reductions in women’s post-partum BMI.
For children, we evaluated several indicators of physical and mental health. First, we evaluated anemia, which was defined based on an affirmative response to the question, “During the past 12 months, has [child] had anemia?” Anemia is a critical health outcome in early childhood and has been found to impact children’s cognitive outcomes [26, 27] and long-term growth , and prior studies have found that receiving WIC benefits was associated with increased iron intake for children [29, 30]. Second, we evaluated change in parent-reported health similar to the question for adults described above. Finally, results from a systematic review suggest that improvements to overall dietary quality led to improvements in mental health for children , while another study found that the 2009 WIC revision improved child development outcomes among recipient children . Therefore, we additionally evaluated changes to children’s mental health, which was measured for children aged 2–3 years with the Mental Health Indicator (MHI) score. The MHI is a validated tool adapted from the Child Behavior Checklist and includes questions about whether a child had trouble sleeping, was unhappy/depressed, or was nervous/high-strung during the previous two months . Although there was slight variation in the questions asked of girls and boys, the MHI questions were intended to measure the same construct, so scores across sex were pooled into one variable, as has been done in previous studies [33, 34]. Higher scores on the MHI score (range 0–8) indicate increased risk for mental health problems.
All models adjusted for covariates that might confound the relationship between receipt of the revised WIC food package and the outcomes, including age, parent marital status, family size, parental education, race/ethnicity (to capture experiences of structural or interpersonal racism), and inflation-adjusted family income. For race/ethnicity, we used the categories of White, Black, Hispanic, and other (including Asian American, American Indian, and those who did not further self-identify a specific race or ethnicity). The latter category is a heterogeneous group for which effect estimates may be difficult to interpret, although we were not able to create more granular categories due to small cell sizes and unstable estimates. Fixed effects for state were included to account for time-invariant characteristics of states that may have confounded the relationship between the state’s timing of policy implementation and the outcomes of interest, and fixed effects for year accounted for secular trends.
We first calculated descriptive statistics stratified by women’s or children’s receipt of WIC and whether the interview was conducted before or after the revised WIC food package was implemented. Then, we estimated the effect of the revised WIC food package on women’s and children’s health outcomes using difference-in-differences (DID) analysis. DID is a quasi-experimental approach that compares trends in a given outcome in a “treatment” group before and after the implementation of a policy, while “differencing out” the secular trends in the outcome in a “control” group of individuals unaffected by the policy . We leveraged the fact that the WIC food package revisions were unlikely to be associated with the characteristics of individuals in our sample. In brief, DID analysis involves a multivariable linear regression model in which the primary predictor is an interaction term between a binary variable for WIC receipt and an indicator for whether the interview was conducted after the revision. DID estimation requires several assumptions to produce valid estimates. Importantly the baseline characteristics of the treatment and control groups do not have to be the same, but rather the trends in outcomes for the treatment and control groups during the pre-revision period must be similar (i.e., the “parallel trends” assumption). Further details, including the equation, are provided in the Supplementary file 1.
We conducted several subgroup analyses to evaluate differential responses to the revised WIC food package. For both women’s and children’s outcomes, we evaluated heterogeneity in estimates by race/ethnicity (White, Black, Hispanic, other), parental education (high school or less versus some college or more), and women’s/mother’s age (under 35 versus 35 and older).
Additional sensitivity analyses were conducted to test the robustness of the results. First, since self-report of safety net benefit receipt can be unreliable [36, 37], we conducted a sensitivity analysis in which the primary exposure was based on WIC eligibility rather than actual receipt, akin to an intent-to-treat design. We imputed eligibility for WIC on state, year, household size, self-reported income, presence of children under one in the household (for women’s eligibility), and age under five (for children’s eligibility). Because income and household size were self-reported in NHIS and might not correspond to the information provided to WIC to determine eligibility, this approach may result in measurement error and therefore was not considered the primary analysis. Second, we included fixed effects for the interview month to provide a more granular adjustment for secular trends. Third, we evaluated whether results were sensitive to group-specific linear time trends, by including an interaction term for WIC receipt and a continuous variable for time. In effect, including group-specific trends in the main model allows outcome trajectories to differentially change over time, allowing for a relaxation of the parallel trends assumption of DID analysis. Finally, it is possible that women may still benefit from their children’s participation in WIC even if they do not receive benefits themselves, since food might be distributed throughout the household. Thus, for the women’s analysis, we included a sensitivity analysis where we redefined the sample to those with children under the age of five.