Design Flaws: Consequences of the Coverage Gap in Food Programs for Children at Kindergarten Entry

Design Flaws: Consequences of the Coverage Gap in Food Programs for Children at Kindergarten Entry Abstract Children age out of the Women, Infants, and Children (WIC) program at 60 months and become eligible for the National School Lunch Program (NSLP) upon kindergarten entry. During this period of time, low-income children experience fewer food support services than at any other time. Using the Early Childhood Longitudinal Study, we examine the effects of the duration of the coverage gap between WIC and NSLP on kindergarteners’ skills. Results show evidence of negative effects on reading, though not on math. Findings also suggest that, for children in full-day kindergarten, effects on reading fade out in the spring term. Food programs, WIC, NSLP, kindergarten, cognitive skills, policy The U.S. federal food and nutrition safety net is a patchwork of programs in which program eligibility depends on age, state of residence, disability, and work status. A significant transition occurs in the food and nutrition programs for which children qualify as they reach age five and enter kindergarten. Before age five, children are eligible for the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) but program eligibility ends at 60 months of age and school-based meal programs, such as the National School Lunch Program (NSLP), are not accessible until children enter kindergarten. This is potentially problematic and creates a growing gap in food and nutrition program coverage. In recognition of the national importance of this issue, U.S. House of Representatives and U.S. Senate bills were introduced in 2016 which included provisions to extend WIC age eligibility until age 6. However, to date there are no studies demonstrating negative consequences of the gap on childhood well-being. This study fills an important gap in the literature by examining the effects of transitions in food and nutrition program coverage at a key point in children’s lives—school entry. We examine the effects of the duration of the gap in program eligibility for WIC and NSLP, which we term the coverage gap length. Our research exploits exogenous age eligibility rules that determine exit from WIC and access to NSLP to identify the consequences of the coverage gap length on child well-being during the fall and spring of the kindergarten year. We use rich data from the Early Childhood Longitudinal Study–Kindergarten Cohort (ECLS-K: 2011). These data contain information about both WIC and NSLP participation, as well as the child’s age in months. We use these data to address two questions about the coverage gap length. First, we examine the impact of the coverage gap length in access to food and nutritional programs on language and mathematical thinking. Second, we examine whether the estimated effects of the coverage gap length found in the fall semester increase, stay the same, or fade out during the spring of kindergarten. We use similar techniques to answer the first question and also run similar sensitivity tests. The average coverage gap between WIC and NSLP is 4.7 months. We find that an additional month of the coverage gap decreases fall reading scores by 0.02 of a standard deviation in the fall for a total effect size of 0.094 of a standard deviation for the average total coverage gap. We observe that this effect fades out by spring of the kindergarten year once children are exposed to the NSLP, and that the difference between fall and spring results is statistically significant. Furthermore, these results are robust to different sample sizes and specifications. In the following section, we provide background on the WIC and NSLP programs and describe previous research evaluating their effects on child outcomes. We then describe the data, sample and empirical strategy. Next, we present our main findings of the effects of the coverage gap on child’s cognitive and socio-emotional skills for fall and spring of the kindergarten year, as well as five sensitivity analyses. We conclude by discussing the implications of our results for research on the coverage gap of safety net programs and for policymakers seeking to extend the duration of WIC up to the start of kindergarten entry. WIC provides supplemental food assistance, nutrition education, and health referrals to low-income pregnant and post-partum women, and to children under age five who are at nutritional risk. In order to be eligible, household gross income must be below 185% of the federal poverty line or households must participate in Medicaid. In 2013, 9.05 million children aged one to four in the United States were eligible for WIC, and 49.8% (4.51 million) participated in the program. Younger children have high participation rates, with about 68.8% of eligible children aged one participating in WIC in 2013, and 50% of children aged two participating in the program. In contrast, the participation rate for children who were three and four years-old for the same year were 47.3% and 32.9%, respectively (USDA 2015). However, despite the lower take-up rate, a sizable number of American children who were aged 4, approximately 858,000, were WIC participants (Thorn et al. 2015). The impact of WIC has been well documented (Bitler et al. 2005; Oh, Jensen, and Rahkovsky 2016); studies show WIC participation has had moderate, positive effects on birth outcomes, that is, decreased preterm birth and increased birth weight (Kowaleski-Jones and Duncan 2002; Bitler and Currie 2005; Figlio, Hamersma, and Roth 2009; Foster, Jiang, and Gibson-Davis 2010; Hoynes, Page, and Stevens 2011); reduced infant mortality rates for African-Americans (Khanani et al. 2010); and increased intake of three of the four key nutrients in early childhood (Yen 2010). Participation in the WIC program has been also associated with improved cognitive and behavioral outcomes at home and in school (Hicks, Langham, and Takenaka 1982; Rush et al. 1988; Jackson 2015). Finally, analyses by Arteaga, Heflin, and Gable (2016) using the Early Childhood Longitudinal Study-Birth Cohort and a regression discontinuity design found that household food insecurity increases when children reach month 61 and age out of eligibility for WIC. However, upon turning old enough to enter school, children can receive school-based food and nutrition programs such as the National School Lunch Program (NSLP). The NSLP is administered at the school level, and upwards of 97% of public schools participate (NCES Schools and Staffing Survey 2013a, 2013b). In fiscal year 2011, over 31 million students received a free or reduced-price lunch daily; according to Dahl and Scholz (2011), participation rates among eligible children are 75% for the NSLP. Empirical studies evaluating the NSLP’s effects on food insecurity, health, and school outcomes are limited (Currie 2003; Gundersen, Kreider, and Pepper 2011). However, estimates of individual impacts of participation in NSLP vary. While Dunifon and Kowaleski-Jones (2003) find no evidence of positive effects on children, Bartfeld and Dunifon (2006), Gundersen et al. (2012), and Frisvold (2015) all find positive effects of school food programs on child outcomes, ranging from obesity and health to math and reading. Closer to the current study, Arteaga and Heflin (2014) used variation in states’ kindergarten age eligibility cutoff date and ECLS-B data and found that NSLP participation protects against household food insecurity during the transition to kindergarten. Present Study This paper answers two research questions that relate to the duration of nutrition program exposure. First, how reasonable is it to assume that the coverage gap length affects children’s cognitive outcomes at school entry? We use multivariate analysis to address this question by regressing the coverage gap length on children’s cognitive test scores and controlling for family background characteristics, childrens’ individual characteristics, demographic characteristics, school characteristics, and participation in other programs. These models include state fixed effects, controlling for all unobserved time-invariant state-level determinants of children’s cognitive skills such as state-specific food programs or policies towards education. A question that might arise is whether parents “select” their children’s coverage gap length. We conduct a thorough analysis and conclude that it is highly unlikely that parents are directly choosing the coverage gap length; given that the coverage gap length is determined by federal rules determining the age when WIC eligibility ends and state and local rules determining the age when children can have access to NSLP. Moreover, our sensitivity tests provide powerful information that indicates that state and school districts’ policy makers are not setting cutoff dates for the beginning of kindergarten based on how prepared they think the children in the state are. In other words, our sensitivity analysis provides evidence that strengthens the case for the exogeneity of the coverage gap-length independent variable. Second, we study whether the effect of the duration in food program access on child’s cognitive development fades out by spring of the kindergarten year, once children are exposed to the NSLP. We expect that an increase in the length of the coverage gap will lead to a reduction in scores for reading and math upon kindergarten entry (hypothesis 1). We also expect that the effects of the length of the coverage gap will fade out by spring of the kindergarten year, once children are exposed to the NSLP through attendance at a full-day program (Hypothesis 2). Data and Sample Analysis of this research question will rely upon data from the Early Childhood Longitudinal Study- Kindergarten (ECLS-K: 2011). The ECLS-K: 2011 is a nationally representative sample of about 18,000 children, selected from both public and private schools, who attended either full-day or half-day kindergarten in 2010–11, and who will be followed through the fifth grade. The panel data for fall and spring of the kindergarten year with non-missing data for child’s cognitive tests consists of 14,600 children. We constrained the sample to those children who attended kindergarten for the first time (n=13,950) because those who attended kindergarten more than once were older and unaffected by the coverage gap. We then constrained the sample to those for whom we had information on the coverage gap (n=12,500). Moreover, we further limited this sample to children who were interviewed during the first two months of the school year when exposure to school meals programs was limited (n=7,850). Additionally, we constrained this sample to those cases for which the ECLS-K collected data on household income level so that we could infer eligibility for food assistance programs (n=6,050). All reported sample sizes are rounded to the nearest 50 in compliance with NCES security standards. In this study, we specifically examine a subsample of children who were income eligible to participate in the WIC and NSLP programs at the time of the survey.1 WIC guidelines are based on household income at or below 185% of the FPL. However, the ECLS-K only reports household income at or below 100% of the FPL and at or below 200% of the FPL. We used the latter for our analysis. The sample size for this group is about 2,350 children and can be thought of as the intent-to-treat sample (ITT). We focus our analysis specifically on children who attended a full-day kindergarten program (n=1,950).2 We focus on this analytic sample because the potential benefit of the NSLP is restricted to children who are present at kindergarten all day in order to access meals provided on-site. In our main analysis, we use an intent-to-treat sample that recodes children who participated in Head Start and who are income eligible to participate in nutritional programs at the time of the survey. For those with coverage gap length equal or greater than three months, we recoded their coverage gap length as three.3 Our main analysis uses this sample to address the concern that children who participated in Head Start had access to the school lunch program during the academic year, regardless of their age, thus contaminating the definition of “coverage gap length.” The ECLS-K: 2011 uses a multi-stage survey design where the first-stage sampling unit is a county or group of counties, the secondary-stage unit is the schools sampled within the counties, and the third-stage sampling unit is the students in the schools. ECLS-K: 2011 uses different sampling weights for each sampling stage and survey weights are used to account for the clustered and multi-staged sampling frame of the ECLS-K: 2011. We used STATA’s svy command for all analyses and we adjusted standard errors to account for the complex survey design. Missing data on covariates, though negligible, were included but were identified with dummy variables. Measures Dependent Variables Our key dependent variables consist of math and reading cognitive variables. The math assessment tested children’s recognition of shapes, colors, sizes, numbers, and number counting, while the reading assessment examined print familiarity, letter recognition, beginning and ending of sounds, rhyming words, word recognition, and vocabulary knowledge. Achievement test scores in reading and mathematics are measured as part of the ECLS-K assessment in the fall and spring of the kindergarten year. The ECLS-K used content domains that were borrowed from the National Assessment of Educational Progress. Some questions were taken directly or adapted from copyrighted instruments such as the Peabody Individual Achievement Test (PIAT-R), the Peabody Picture Vocabulary Test (PPVT-III), the Preschool Language Assessment Scale (preLas 2000), the Test of Early Mathematics Ability (TEMA-3), the Test of Preschool Early Literacy (TOPEL), and the Woodcock-Johnson Psychoeducational Battery (WJ-III), among others. The tests were administered using a computer-assisted method in which the interviewer entered children’s responses. Children were presented images and only had to point to answers or provide verbal responses and were not asked to write or explain their reasoning. The ECLS-K used a two-stage assessment in reading and mathematics. Children were administered basic questions in the first stage and based on their responses, they were then given a set of questions appropriate for their demonstrated skill level, instead of all items in the survey instrument. Both reading and math scores were computed using item response theory (IRT) scores (Rock and Pollack 2002) that are included in the ECLS-K dataset. The IRT uses patterns of correct, incorrect, and omitted answers to obtain ability estimates that allow us to compare scores from different levels of difficulties given that different children respond to different questions in the ECLS-K (Hambleton, Swaminathan, and Rogers (1991). The National Center for Educational Statistics recommends that the age at the time of assessment or date of assessment is used as a control in regression models in which a child’s cognitive score is a dependent variable (Tourangeau et al. 2012). However, because we construct our main independent variable—coverage gap length—using the distance of the child’s age to the cutoff rules, this variable is highly collinear with child’s age. Auxiliary regression models that use raw IRT test scores as a dependent variable, coverage gap length as an independent variable, and child’s age as a control variable show a Variance Inflation Factor (VIF) above 10 for both coverage gap length and child’s age, indicating that these two variables are highly collinear, and thus it is difficult to draw conclusions about the magnitude and significance of estimated coefficients. To address this collinearity problem, we age-normed our measurements of achievement. It is important to mention that there is not a national age-normed version of the instrument because not everybody answered exactly the same questions. Remember that the assessment occurred in two stages and questions were administered according to children’s demonstrated ability in the first stage. Moreover, in the case of ECLS-K, they used different components of different standardized tests for their reading and math components. Thus, after consulting directly with ECLS-K specialists, we decided to use age-group categories that have been previously used in the literature to age-norm other instruments like PPVT-III and WJ-III. Given the ages of children in the Fall of Kindergarten, we created eight different groups: 3.5–4 years of age (category 1); 4–4.5 years (category 2); 4.5–5 years (category 3); 5–5.5 years (category 4); 5.5–6 years (category 5); 6–6.5 years (category 6); 6.5–7 years (category 7); 7–8 years (category 8). It is important to note that we had thousands of observations for categories 2 through 6, but fewer observations for the rest of the categories. When closely looking at our sample for this study (n= 2,350), we found that we had less than 50 observations altogether for categories 1, 7, and 8, so we dropped them from our analysis. We then calculated means and standard deviations for IRT scores for both reading and math for each group using all ECLS-K children because the sample is nationally representative of all kindergarteners. Means and standard deviations are very consistent with expectations, with means increasing for each higher age-group category. Using these means and standard deviations, we age-normed our dependent variables using a typical normalization procedure: IRT score – IRT mean for that age category, divided by the standard deviation for that age category. Thus, we obtained age-normalized scores with a mean of zero and standard deviation of 1, which provides model estimates that can be interpreted as effect sizes (Cohen 1988). Independent Variable Children who participate in WIC can only be enrolled until 60 months of age; according to the USDA, there are no category errors due to over-aged children receiving WIC benefits (USDA 2012). In order to access the NSLP program, children must be enrolled in a formal kindergarten program. Kindergarten enrollment is largely based on turning age 5 by a cut-off date, which varies by state, from July 31 in Nebraska, to January 1 in Connecticut. Moreover, the start of the academic year also varies by state and within state by school district. For example, in the academic year 2010–2011, some school districts in Nevada started the academic year on July 12 while some Ohio school districts started on September 26. As a consequence, our key variable of interest is the coverage gap length, which is equal to the age of the child on the first day of kindergarten, minus 60 months (the age when a child loses WIC eligibility). For example, if the academic year starts on 1 September in state s, and child i is 66 months old at that time, then the coverage gap length for child i is 6 months. Coverage gaps of more than 365 days are dropped from the analysis (n = 1,100 for the full-sample, and n = 350 for the ITT sample) because they indicate that the child was red-shirted, or the parents voluntarily delayed kindergarten entry for a year. We eliminate these children from our analytic sample since we only are interested in the effect of the coverage gap length for children who were affected by school eligibility rules and for whom the coverage gap is exogenous to parental decision-making. We assume that these children would be unlikely to be affected by the coverage gap and would bias our results towards zero. Additionally, not all children have a coverage gap. About 200 children turned five after beginning kindergarten, meaning that they experience no gap in food program coverage; we include these children in our analysis and assigned a coverage gap equal to zero. Thus, our key independent variable provides the basis to estimate intent-to-treat effects of the coverage gap length on child’s well-being.4 States and local education agencies (LEA; also known as school districts) vary on their kindergarten age eligibility criteria. In 2010, the year in which students observed in the ECLS-K entered kindergarten, 44 states and the District of Columbia had a statewide age-eligibility policy, the majority of which fell in September (Education Commission of the States 2011). Six states allowed each LEA to decide its own cutoff date by which children had to turn five to be eligible for kindergarten enrollment: Colorado, Massachusetts, New Hampshire, New Jersey, New York, Ohio, and Pennsylvania. For the approximately 70 school districts observed in ECLS-K from these states, we obtained information about their kindergarten eligibility by contacting officials from the school district. To sum up, to create our independent variable of interest, we combined data from ECLS-K: 2011 and from school districts (direct contact by researchers). Control Variables We control for child and family characteristics. Specifically, we control for race and ethnicity (Black, White, Asian, Hispanic, or Other), gender (male or female), and quarter of the child’s birth (Buckles and Hungerman 2013; Cascio and Schanzenbach 2016). We also control for the following family characteristics: number of children in household, number of adults in household, maternal marital status (married, widowed/divorced or separated, or never married), parental education level (less than high school, high school degree, some college, 4 years or more), 12-month household food security level, prior Head Start participation, TANF participation, SNAP participation, school lunch participation, the percentage of the school participating in free or reduced lunch, and public school attendance. We control for SNAP participation because our variable of interest—coverage gap length—captures the gap in food programs (WIC and NSLP) after controlling for all other programs. Similarly, we control for prior participation in Head Start because children who attend Head Start programs receive NSLP. We also control for the survey’s assessment month, urban residence, and state of residence using state fixed-effects. Table 1 presents descriptive statistics for our control variables. We observe that for our sample of interest (n = 1,950), 37% of children are white, 17% are Black, 35% Hispanic, and the rest are from other race/ethnicity. About half of the children are male (51%). Birth seasonality seems to be very similar among quarters. As expected for our WIC-NSLP eligible sample, 20% of children previously participated in Head Start. Similarly, annual household income was on average about $24,000. Also, a higher percentage of parents have never been married (22%), which is consistent with other studies of low-income families (Cabrera et al. 2004; Pruett et al. 2017). Overall, we observe that the percentage of missing values for each variable category is low and below 8%, except for school percentage in free and reduced lunch program (16%). Table 1 Descriptive Statistics Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Note: Standard deviations appear in parentheses next to the mean value for continuous. As per National Center for Education Statistics requirements, all n values are rounded to the nearest 50. Table 1 Descriptive Statistics Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Note: Standard deviations appear in parentheses next to the mean value for continuous. As per National Center for Education Statistics requirements, all n values are rounded to the nearest 50. Empirical Strategy To examine the relationship between length of the coverage gap and children’s cognitive development, we estimate a series of multivariate regression models. The outcomes of child i living in state s (CHILDis) are functions of the WIC-NSLP’s coverage gap length (DURi), a vector of controls, X, (child and family characteristics), state dummy variables, d, (fixed effects) and an error term, ɛ:   CHILDfallis=β0+β1DURfallis+β2Xis+γs∑ds+ɛis (1) where DURis= AgeBSis – 60, AgeBSis is the age of child i at the beginning of the school year in state s, and 60 is the age in months when a child ages out of WIC. Estimating equation (1) with ordinary least squares (OLS) yields an unbiased estimate of the impact of DUR if the unobserved determinants of child well-being are uncorrelated with the coverage gap length. We argue that this assumption is likely to hold because the construction of DUR is based on eligibility rules that are exogenous to parental decision making. Moreover, while a child’s birthday might be considered endogenous, the distance between the child’s fifth birthdate and the start of school is likely to be exogenous. Most parents do not know five years (and nine months) in advance the age eligibility rule for the school their children will attend. Another potential source of endogeneity might arise from policy makers if they choose the starting date for kindergarten based on how prepared they think the children in the state or LEA are. If this is true, then starting dates might be endogenous even though children and parents are not directly choosing them. We might think of a case where schools with a higher commitment to educational quality have earlier start dates. To address this concern, we regressed school start date on school-level scores, controlling for state fixed effects as an additional sensitivity test. While we acknowledge that exogeneity may be a strong assumption, we provide a series of sensitivity tests to back up this claim. Finally, it is important to note that we control for quarter of child’s birthdate in our regressions. We start our analysis with our intent-to-treat sample (our preferred sample) of children who attend full-day kindergarten and who were WIC eligible at the point of kindergarten entry (n = 2,350), that is, who satisfied the poverty eligibility measure (i.e., intent to treat analyses: ITT). Our hypothesis will be confirmed if we find that β1 is negative and statistically significant in the fall for the cognitive outcomes, meaning that DURis negatively affects children’s cognitive development (or if we find that β1 is positive and statistically significant for internalizing and externalizing problem behaviors). Our hypothesis will be unsupported if we find that β1 is not statistically significant. To examine our second research question regarding the fade out of the duration gap, we also use multivariate regression analysis. We regress children’s outcomes measured in the spring of 2011, controlling for fall 2010 variables, as follows:   CHILDspring is=α0+α1DURis+α2Xfall is+γs∑ds+ɛis. (2) By the spring, children would have been receiving free or reduced-priced lunch for about six months. Thus, the only difference between equations (1) and (2) is that the dependent variable in equation (2) is measured in the Spring instead of the Fall. Our hypothesis will be confirmed if β1 (coefficient of DUR on equation [1]) is statistically greater than α1 (coefficient of DUR on equation [2]); in other words, we expect the effects of DUR to fade out by the spring if children attend a full-day program and can access school-based meal programs. For our sensitivity analysis using our overall sample (full-day and half-day groups), it is possible that our results may be more mixed because of the presence of the half-day kindergarten group which may not have access to school meals. Results Main Results (Full-Day Kindergarten Sample) Fall Outcomes Table 2 reports results for our main sample that includes only children who attended full-day kindergarten, which is our intent-to-treat sample (n = 1,950). We find that an additional month of the coverage gap decreases reading IRT scores by 0.021 of a standard deviation ( p < 0.01). Conversely, an additional month of the coverage gap does not have a significant effect on math IRT scores (coefficient=−0.008 standard deviations; p > 0.10). Table 2 Effects of Coverage Gap Length of Food Programs on Child’s Skills for Children Attending Full-Day Kindergarten (n = 1,950)   Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)      Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Table 2 Effects of Coverage Gap Length of Food Programs on Child’s Skills for Children Attending Full-Day Kindergarten (n = 1,950)   Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)      Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Spring Outcomes Next, we examine our second hypothesis that the effects of the coverage gap length will fade out by spring once children are exposed to the NSLP in school. For our intent-to-treat sample (n = 1,950), we find that an additional month of the coverage gap does not have a statistically significant effect on math or reading IRT scores. Moreover, for reading, the difference in the effects of the coverage gap length for fall and spring is a reduction of IRT scores of 0.016 of a standard deviation, and this difference is statistically significant. Remember that this is equivalent to a total effect size of 0.094 of a standard deviation for the average total coverage gap. These results suggest that the negative effects of the coverage gap length prior to kindergarten entry have faded by spring. Sensitivity Analysis We present four different sensitivity analyses designed to test the OLS estimate’s robustness to a variety of modeling choices, assumptions, and samples. First, we test how sensitive our results are to different kindergarten day arrangements by examining the full sample of kindergarteners and not limiting our analysis to full-day program attendees in order to increase the external validity of our findings. Second, we create a proxy sample to estimate Treatment-on-the-Treated effects by using an “ever WIC” participation variable and constraining the sample to those with WIC eligibility prior to a child’s fifth birth date. Third, in order to test if the coverage gap is correlated with unobservable factors, we replicate our analysis of the coverage gap on a sample of advantaged children who were not eligible for either WIC or school-based nutrition programs based on their kindergarten year. Finally, we test the validity of our DUR measure by analyzing whether it is only identifying the effects of differences in school start date solely, after controlling for child’s age. In all cases, presented below, we find that results are consistent with our expectations and confirm our main finding presented above. Full and Part-Time Kindergarten Program Analysis Table 3 presents results for our first two sensitivity tests. In the first two rows of table 3, we replicate the analysis shown in table 2 for the ITT sample, but we add half-day kindergartners to the analysis sample. Ideally, we would be able to replicate our analysis on the half-day sample alone but sample size (n = 400), prevents us from adopting this approach. For panel A of table 3, results are consistent with those shown in table 2. For reading, we observe that in the fall, the association between DUR and our outcome is negative but there is no association for the spring. Once again, difference between Fall and Spring reading scores is negative and statistically significant. For math, we did not find statistically significant effects for the difference between Fall and Spring. Table 3 Sensitivity Analysis: Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, ITT Analytical Sample   Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)      Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. As defined in the text, panel A includes the full-day, plus half-day kindergarten sample (n = 2,350), and panel B includes those in full day kindergarten who ever received WIC. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Table 3 Sensitivity Analysis: Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, ITT Analytical Sample   Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)      Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. As defined in the text, panel A includes the full-day, plus half-day kindergarten sample (n = 2,350), and panel B includes those in full day kindergarten who ever received WIC. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Treatment-on-the-Treated Effects (ToT) While our initial goal for this study was to examine treatment on the treated effects, our dataset did not allow us to conduct such analysis: the ECLS-K: 2011 did not collect information on WIC participation at age 4 specifically, but only gathered information on whether a child ever participated in the WIC program. Using this information, we constructed a “proxy” of ToT effects that we call ToT Ever WIC, which consists of children who ever participated in WIC and who were eligible for WIC in the fall of Kindergarten (n = 1,550). It is important to keep in mind that this is a proxy variable and likely contains measurement error in terms of identifying children who received WIC at age 4. While typical ToT analyses have larger effect sizes than ITT estimates since the untreated are eliminated from the sample, in this case we lack the precision to correctly identify our ToT sample. Nonetheless, we present results for our proxy ToT measures in panel B of table 3. We find similar results to those found in our main analysis (table 2) for ToT Ever WIC. The difference in the effects between Fall and Spring for reading IRT scores is negative and statistically significant, while we do not find statistically significant effects for math IRT scores. Advantaged Sample Falsification Test Next, we present an additional sensitivity analysis to reduce concerns that the effects that we are identifying are due to other confounding factors and not necessarily related to access to food and nutrition programs. In table 4, we replicate the analysis shown in table 2, but instead of limiting our sample to those who qualify for WIC and NSLP (ITT sample), we limit our sample to those children who live in households with incomes above $50,000 at the time of the survey and who have never participated in the WIC program. If the effect that we are able to identify in our prior models is coming from the coverage gap length and not from other confounding factors, then we would expect to observe no effect for the sample that is not income-eligible for the WIC and NSLP. Results show that the coverage gap length does not have a statistically significant association with either math or reading scores in the Fall or the Spring. Thus, our findings pass this falsification test. Table 4 Falsification Test: Estimated Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, Advantaged (High Income and not WIC Eligible) Sample   Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)      Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)    * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Table 4 Falsification Test: Estimated Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, Advantaged (High Income and not WIC Eligible) Sample   Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)      Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)    * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 School Start Date and School Quality While DUR is the distance between a child’s birthdate and the beginning of the school year, some might be concerned that DUR is picking up the effects of differences in school start date only after controlling for child’s age. This might be the case if schools with a higher commitment to educational quality have earlier start dates. To address this concern, we regressed school start date on school-level scores, controlling for state fixed effects. We used two measures for school start date: date in days, and date in weeks. Table 5 shows no statistical differential effect of school start date on school-level scores for either measure. This finding demonstrates that the main results presented in table 2 are not driven by unobserved differences in school quality related to school start date, but by the coverage gap length in access to food and nutritional programs. Table 5 Sensitivity Analysis: Estimates of the Effects of School Start Date Only on School Level Scores   Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)    Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)  * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses. Each row represents a different regression. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: = p<0.1; ** = p<0.05; *** = p<.01. Table 5 Sensitivity Analysis: Estimates of the Effects of School Start Date Only on School Level Scores   Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)    Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)  * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses. Each row represents a different regression. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: = p<0.1; ** = p<0.05; *** = p<.01. Together, the sensitivity analyses carried out in tables 3 through 5 provide robust evidence of the main results found in table 2: a negative effect of the coverage gap length on child achievement scores for reading in the Fall of kindergarten and a statistically significant difference between Fall and Spring. At the same time, our analysis does not show any consistent effects for math. Furthermore, we consistently find evidence that the negative effects fade out by the spring. Given the analysis shown, we believe that it is reasonable to argue that the negative effects observed in the Fall for reading were likely driven by the coverage gap length and not by other confounding factors. Discussion and Conclusions Our study uses the nationally representative ECLS-K to suggest how having limited access to food and nutritional programs during the time when a child transitions to kindergarten may shape child cognitive achievement over the course of kindergarten. Age eligibility rules for WIC and the timing of kindergarten entry mean that many children experience a gap in access to both WIC and the NSLP that can last from several months up to a year. Our study finds mixed results. We found some evidence of negative effects of the coverage gap length in reading, but not in math. These results were consistent across different models. One consequence of this coverage gap is that children enter kindergarten with lower cognitive scores, on the order of 0.02 for each month of the coverage gap, for reading for each gap month experienced. While at first effect sizes in this range may be considered small, our effects are additive across coverage gap months (0.02 x 4.7 months of coverage gap= 0.094). Thus, when comparing our findings to interventions that affect educational outcomes, effect sizes in this range are considered moderate (Cohen 1988). Results presented in this paper are consistent with previous work that found an increase in rates of food insecurity for children who become age-ineligible for WIC and who have not yet started kindergarten (Arteaga, Heflin, and Gable 2016). Additionally, it has been shown in previous work that when children enter kindergarten and are able to access the NSLP, they become less food insecure (Arteaga and Heflin 2014). This study contributes to the literature by examining the consequences at kindergarten entry for child reading and math scores for the family disruption caused by the withdrawal of food and nutritional support when a child reaches 60 months of age. It should be noted that our study also finds that by the Spring of kindergarten, the consequences of the coverage gap for reading have ameliorated. Moreover, policy makers should be aware of the existence of this nutritional gap during a child’s transition into school as a policy failure that could be avoided by extending WIC eligibility to school-age. Given the heightened scrutiny that all social spending is facing in today’s political climate, our research provides clear empirical support for the positive role that nutritional assistance programs can play in the development of cognitive skills upon kindergarten entry. It is important to note, however, that our study does not speak to which form of food assistance is better (i.e., WIC or NSLP), but to the need to support consistent household food consumption over the childhood period. Limitations Despite the contributions of this study, it is important to recognize several limitations. First, the sample of income-eligible families is small. This is a typical challenge when using nationally representative datasets. Second, the ECLS-K does not contain information on WIC participation at the child’s fifth birthday or whether the household was income-eligible to participate in WIC when the child was four years of age. Thus, there is a possibility of measurement error in determining the sample. We used data collected in the spring of kindergarten, which documents household annual income in the prior year for identifying income-eligibility for our intent-to-treat sample. It is still possible that annual income volatility might have changed income eligibility between the time of data collection and when the child was age-eligible for WIC. Additionally, for our treatment on the treated sample, we used the question that documented if the household “ever participated in WIC” because the ECLS-K did not collect information on WIC participation when the child was four years of age. This might also introduce measurement error into the analysis because national figures show that the percentage of WIC participants is higher for infants and toddlers than for preschoolers. It is then possible that some families who should have not been included because they no longer participated in WIC are included in our ToT sample. However, findings were unchanged when models were re-estimated using a different sensitivity analysis. We consistently observed a reduction in the food coverage gap between Fall and Spring for reading; the difference between these two periods of time was statistically significant. In addition, we argue that our analytical strategy does allow us to make causal inferences about the association between the coverage gap length of food and nutritional programs and a child’s cognitive and socio-emotional outcomes. Both aging out of WIC and NSLP participation are determined exogenously by USDA rules, a state’s age eligibility cutoffs, and school districts’ decisions on when to start the school year. Our sensitivity analysis that estimates the effects of start of school date on test scores provides evidence that state age eligibility rules or school district calendars are unlikely to be endogenously determined with respect to birth date. To the extent that one might wonder if turning five years old is associated with some other unobserved changes related to child outcomes that our analysis is picking up, our sensitivity analysis with the advantaged sample and the school data demonstrates null findings of duration. Therefore, we believe it is unlikely that other positive confounding factors are biasing the negative effects that we find for coverage length gap on children’s outcomes. Policy Implications There are several important implications of these findings. First, our analysis indicates that a coverage gap in access to federal food and nutrition programs is associated with poor child outcomes at kindergarten entry, and these effects may attenuate by spring for children who have access to full day kindergarten. It is important to follow these children over time and see if this effect continues to be zero or if it grows in some children. Our findings suggest that the coverage gap length needs to be reduced because it reduces reading scores at kindergarten entry, a time when children are often placed on learning trajectories and teachers and administrators gather impressions about children’s learning potential. The fact that the gap fades by Spring of the kindergarten year suggests that children are very sensitive to their food intake around kindergarten entry and that it is not negative selection into the WIC program that is behind their low reading scores. Stated differently: our results suggest that given an adequate level of resources, WIC-eligible children can learn. There is no reason for them to begin their educational careers behind their more advantaged peers. An easy way to address this problem would be to extend the age of WIC eligibility to the month of state eligibility for kindergarten entry instead of age 5 as is the current practice. Our analysis suggests that this small change could lead to a substantial improvement at kindergarten entry. Given how key notions of equality of opportunity are to the American Dream, policy makers should address the unintended consequences of compounding disadvantage for millions of children each year who are unlucky enough to be born in the wrong state and in the wrong month, both of which prolong their coverage gap. Funding This research was made possible by the generous support from the Institute for Research on Poverty (IRP) RIDGE Center for Policy Research at the University of Wisconsin and the United States Department of Agriculture. The authors thank participants at the 2016 IRP conference in Madison and Washington D.C., and participants at the 2016 APPAM Fall Research Conference for their helpful comments. The statements and views expressed in this article do not reflect those of the study’s sponsors. Any remaining errors are our own. Footnotes 1 NSLP eligibility is based on household income at or below 130% of the federal poverty line (FPL) for free lunch and at or below 185% for reduced-price meals. The community eligibility provision for the NSLP was not enacted at the time of this study. 2 Once a child starts school, the potential benefit of the NSLP depends on how school districts administer kindergarten, since children must be present at school to access meals provided on-site. In 2013, 76% of kindergarteners in the United States attended full-day programs. There were, however, slight variations in full-day attendance rates by race and the highest level of educational attainment of the parents/ guardians. Indeed, 77% of white, 79% of blacks, and 74% of Hispanic kindergarteners were enrolled in full-day programs. Conversely, the full-day attendance rate for Asian kindergarteners was significantly lower at only 65.5%. Further, 69% of children whose parents or guardians had less than a high school education attended full-day kindergarten, compared to 78% whose parents or guardians had completed high school or had a GED, and 79% whose parents had vocational/technical, or some college education. The full-day attendance rates for children whose parents or guardians had an associate's degree, a bachelor's degree or a graduate or professional degree were 75%, 77% and 72%, respectively (NCES 2014). 3 We do not have information on Head Start participation during the summer. National data shows that there are not many Head Start programs that are open during the summer. Thus, we assumed that children did not have access to Head Start during the summer. 4 Although it would be ideal to have knowledge on whether a child was participating in WIC just before kindergarten entry, ECLS-K data do not provide this information. That is, we only know if a child was income-eligible to participate in WIC at the time of the fall survey of the kindergarten year and whether a child ever participated in WIC. We explored the possibility of using the ECLS-B instead since it contains information on WIC participation at 48 months of age and until a child turned 60 months of age. However, the ECLS-B does not gather administrative information on the start date of the school year. Moreover, the ECLS-B only collects data for kindergarteners once and it can be at any time over the academic year. Given these limitations, we opted to use the ECLS-K. Our analysis uses information for those who were eligible to participate in WIC and calculates the duration of the coverage gap for those who were intended to be treated. References Arteaga I., Heflin C.. 2014. Participation in the National School Lunch Program and Food Security: An Analysis of Transitions into Kindergarten. Children and Youth Services Review  47: 224– 30. Google Scholar CrossRef Search ADS   Arteaga I., Heflin C., Gable S.. 2016. The Impact of Aging out of WIC on Food Security in Households with Children. Children and Youth Services Review  69: 82– 96. Google Scholar CrossRef Search ADS   Bartfeld J., Dunifon R. 2006. State‐level predictors of food insecurity among households with children. Journal of Policy Analysis and Management  25 4: 921– 42. Google Scholar CrossRef Search ADS   Bitler M., Currie J.. 2005. Does WIC Work? The Effects of WIC on Pregnancy and Birth Outcomes. Journal of Policy Analysis and Management  24 1: 73– 91. Google Scholar CrossRef Search ADS PubMed  Bitler M., Gundersen C., Marquis G.S.. 2005. Are WIC Nonrecipients at Less Nutritional Risk than Recipients? An Application of the Food Security Measure. Review of Agricultural Economics  27 3: 433– 8. Google Scholar CrossRef Search ADS   Buckles K.S., Hungerman D.M.. 2013. Season of Birth and Later Outcomes: Old Questions, New Answers. Review of Economics and Statistics  95 3: 711– 24. Google Scholar CrossRef Search ADS PubMed  Cabrera N.J., Ryan R.M., Shannon J.D., Brooks-Gunn J., Vogel C., Raikes H., Cohen R.. 2004. Low-Income Fathers’ Involvement in Their Toddlers’ Lives: Biological Fathers from the Early Head Start Research and Evaluation Study. Fathering  2 1: 5. Google Scholar CrossRef Search ADS   Cascio E.U., Schanzenbach D.W.. 2016. First in the Class? Age and the Education Production Function. Education Finance and Policy  11 3: 225– 50. Google Scholar CrossRef Search ADS   Cohen J. 1988. Statistical power analyses for the social sciences . Hillsdale, NJ: Lawrence Erlbauni Associates. Currie J. 2003. US Food and Nutrition Programs. In Moffitt R. (Ed.), Means-Tested Transfer Programs in the United States , 199– 290. Chicago, IL: University of Chicago Press (Chapter4). Google Scholar CrossRef Search ADS   Dahl M., Scholz J.K.. 2011. The National School Lunch Program and School Breakfast Program: Evidence on Participation and Noncompliance. University of Wisconsin, Institute for Research on Poverty Working Paper. Dunifon R., Kowaleski-Jones L.. 2003. The Influences of Participation in the National School Lunch Program and Food Insecurity on Child Well-Being. The Social Service Review  77 1: 72– 92. Google Scholar CrossRef Search ADS   Education Commission of the States. 2011. State Characteristics: Kindergarten. https://www.ecs.org/clearinghouse/90/71/9071.pdf. Figlio D., Hamersma S., Roth J.. 2009. Does Prenatal WIC Participation Improve Birth Outcomes? New Evidence from Florida. Journal of Public Economics  93 1: 235– 45. Google Scholar CrossRef Search ADS   Foster E.M., Jiang M., Gibson-Davis C.M.. 2010. The Effect of the WIC Program on the Health of Newborns. Health Services Research  45 4: 1083– 104. Google Scholar CrossRef Search ADS PubMed  Frisvold D. E. 2015. Nutrition and cognitive achievement: An evaluation of the School Breakfast Program. Journal of public economics  124: 91– 104. Google Scholar CrossRef Search ADS PubMed  Gundersen C., Kreider B., Pepper J.. 2011. The Economics of Food Insecurity in the United States. Applied Economic Perspectives and Policy  33 3: 281– 303. Google Scholar CrossRef Search ADS   Gundersen C., Kreider B., Pepper J. 2012. The impact of the National School Lunch Program on child health: A nonparametric bounds analysis. Journal of Econometrics  166 1: 79– 91. Google Scholar CrossRef Search ADS   Hambleton R.K., Swaminathan H., Rogers H.J.. 1991. Fundamentals of Item Response Theory . (Measurement methods for the social sciences series, Vol. 2). Newbury Park, CA: Sage. Hicks L.E., Langham R.A., Takenaka J.. 1982. Cognitive and Health Measures following Early Nutritional Supplementation: A Sibling Study. American Journal of Public Health  72 10: 1110– 8. Google Scholar CrossRef Search ADS PubMed  Hoynes H., Page M., Stevens A.H.. 2011. Can Targeted Transfers Improve Birth Outcomes? Evidence from the Introduction of the WIC Program. Journal of Public Economics  95 7: 813– 27. Google Scholar CrossRef Search ADS   Jackson M.I. 2015. Early Childhood WIC Participation, Cognitive Development and Academic Achievement. Social Science & Medicine  126: 145– 53. Google Scholar CrossRef Search ADS   Khanani I., Elam J., Hearn R., Jones C., Maseru N.. 2010. The Impact of Prenatal WIC Participation on Infant Mortality and Racial Disparities. American Journal of Public Health  100 ( S1): S204– 9. Google Scholar CrossRef Search ADS PubMed  Kowaleski-Jones L., Duncan G.J.. 2002. Effects of Participation in the WIC Program on Birthweight: Evidence from the National Longitudinal Survey of Youth. American Journal of Public Health  92 5: 799– 804. Google Scholar CrossRef Search ADS PubMed  Oh M., Jensen H.H., Rahkovsky I.. 2016. Did Revisions to the WIC Program Affect Household Expenditures on Whole Grains? Applied Economic Perspectives and Policy  38 4: 578– 98. Google Scholar CrossRef Search ADS   Pruett M.K., Pruett K., Cowan C.P., Cowan P.A.. 2017. Enhancing Father Involvement in Low‐Income Families: A Couples Group Approach to Preventive Intervention. Child Development  88 2: 398– 407. Google Scholar CrossRef Search ADS PubMed  Rock D. A., Pollack J. M. 2002. Early Childhood Longitudinal Study-Kindergarten Class of 1998-99 (ECLS-K): Psychometric Report for Kindergarten through First Grade. Working Paper Series. Rush D., Leighton J., Sloan N.L., Alvir J.M., Horvitz D.G., Seaver W.B., Devore J.W.. 1988. The National WIC Evaluation: Evaluation of the Special Supplemental Food Program for Women, Infants, and Children. VI. Study of Infants and Children. The American Journal of Clinical Nutrition  48 2: 484– 511. Google Scholar CrossRef Search ADS PubMed  The National Center for Education Statistics. 2014. Digest of Education Statistics: Table 202.20. Percentage of 3-, 4-, and 5-year-old Children Enrolled in Preprimary Programs, by Attendance Status, Level of Program, and Selected Child and Family Characteristics: 2013. Available at: http://nces.ed.gov/programs/digest/d14/tables/dt14_202.10.asp? current=yes (accessed February 6, 2017). The National Center for Education Statistics Schools and Staffing Survey. 2013a. Table 1. Total Number of Private Schools and Students, and Percentage of Schools and Students that Participated in the Title I and Federal Free or Reduced-Price Lunch Programs, by Affiliation: 2011–12. Available at: http://nces.ed.gov/surveys/sass/tables/sass1112_2013312_s2a_001.asp (accessed February 6, 2017). The National Center for Education Statistics Schools and Staffing Survey. 2013b. Table 1. Total Number of Public Schools and Students, and Percentage of Schools and Students that Participated in the Title I and Federal Free or Reduced-Price Lunch Programs, by State: 2011–12. Available at: http://nces.ed.gov/surveys/sass/tables/sass1112_2013312_s2s_001.asp (accessed February 6, 2017). Thorn B., Tadler C., Huret N., Trippe C., Ayo E., Mendelson M., Patlan K., Schwartz G., Tran V. 2015. WIC Participant and Program Characteristics 2014. Prepared by Insight Policy Research under Contract No. AG-3198-C-11–0010. US Dept of Agriculture, Food and Nutrition Service, Alexandria, VA. Tourangeau K., Nord C., Lê T., Sorongon A.G., Hagedorn M.C., Daly P., Najarian M.. 2012. Early Childhood Longitudinal Study, Kindergarten Class of 2010–11 (ECLSK: 2011), User’s Manual for the ECLS-K: 2011 Kindergarten Data File and Electronic Codebook (NCES 2013–061). Washington DC: U.S. Department of Education: National Center for Education Statistics. U.S. Department of Agriculture. 2012. WIC Program Participation and Costs: April 2, 2012. Available at: http://www.fns.usda.gov/pd/wisummary.htm (accessed February 5, 2016). U.S. Department of Agriculture, Food and Nutrition Service, Office of Policy Support. 2015. National and State-Level Estimates of Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) Eligibles and Program Reach (Summary) . Alexandria, VA: Author. Available at: http://www.fns.usda.gov/sites/default/files/ops/WICEligibles2013-Summary.pdf (accessed February 5, 2016). Yen S.T. 2010. The Effects of SNAP and WIC Programs on Nutrient Intakes of Children. Food Policy  35 6: 576– 83. Google Scholar CrossRef Search ADS   Appendix Table A1 Descriptive Statistics for Outcome Variables   Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409    Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409  Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. View Large Table A1 Descriptive Statistics for Outcome Variables   Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409    Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409  Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. View Large © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Economic Perspectives and Policy Oxford University Press

Design Flaws: Consequences of the Coverage Gap in Food Programs for Children at Kindergarten Entry

Loading next page...
 
/lp/ou_press/design-flaws-consequences-of-the-coverage-gap-in-food-programs-for-QfVH7MTIFX
Publisher
Oxford University Press
Copyright
© The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
2040-5790
eISSN
2040-5804
D.O.I.
10.1093/aepp/ppy009
Publisher site
See Article on Publisher Site

Abstract

Abstract Children age out of the Women, Infants, and Children (WIC) program at 60 months and become eligible for the National School Lunch Program (NSLP) upon kindergarten entry. During this period of time, low-income children experience fewer food support services than at any other time. Using the Early Childhood Longitudinal Study, we examine the effects of the duration of the coverage gap between WIC and NSLP on kindergarteners’ skills. Results show evidence of negative effects on reading, though not on math. Findings also suggest that, for children in full-day kindergarten, effects on reading fade out in the spring term. Food programs, WIC, NSLP, kindergarten, cognitive skills, policy The U.S. federal food and nutrition safety net is a patchwork of programs in which program eligibility depends on age, state of residence, disability, and work status. A significant transition occurs in the food and nutrition programs for which children qualify as they reach age five and enter kindergarten. Before age five, children are eligible for the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) but program eligibility ends at 60 months of age and school-based meal programs, such as the National School Lunch Program (NSLP), are not accessible until children enter kindergarten. This is potentially problematic and creates a growing gap in food and nutrition program coverage. In recognition of the national importance of this issue, U.S. House of Representatives and U.S. Senate bills were introduced in 2016 which included provisions to extend WIC age eligibility until age 6. However, to date there are no studies demonstrating negative consequences of the gap on childhood well-being. This study fills an important gap in the literature by examining the effects of transitions in food and nutrition program coverage at a key point in children’s lives—school entry. We examine the effects of the duration of the gap in program eligibility for WIC and NSLP, which we term the coverage gap length. Our research exploits exogenous age eligibility rules that determine exit from WIC and access to NSLP to identify the consequences of the coverage gap length on child well-being during the fall and spring of the kindergarten year. We use rich data from the Early Childhood Longitudinal Study–Kindergarten Cohort (ECLS-K: 2011). These data contain information about both WIC and NSLP participation, as well as the child’s age in months. We use these data to address two questions about the coverage gap length. First, we examine the impact of the coverage gap length in access to food and nutritional programs on language and mathematical thinking. Second, we examine whether the estimated effects of the coverage gap length found in the fall semester increase, stay the same, or fade out during the spring of kindergarten. We use similar techniques to answer the first question and also run similar sensitivity tests. The average coverage gap between WIC and NSLP is 4.7 months. We find that an additional month of the coverage gap decreases fall reading scores by 0.02 of a standard deviation in the fall for a total effect size of 0.094 of a standard deviation for the average total coverage gap. We observe that this effect fades out by spring of the kindergarten year once children are exposed to the NSLP, and that the difference between fall and spring results is statistically significant. Furthermore, these results are robust to different sample sizes and specifications. In the following section, we provide background on the WIC and NSLP programs and describe previous research evaluating their effects on child outcomes. We then describe the data, sample and empirical strategy. Next, we present our main findings of the effects of the coverage gap on child’s cognitive and socio-emotional skills for fall and spring of the kindergarten year, as well as five sensitivity analyses. We conclude by discussing the implications of our results for research on the coverage gap of safety net programs and for policymakers seeking to extend the duration of WIC up to the start of kindergarten entry. WIC provides supplemental food assistance, nutrition education, and health referrals to low-income pregnant and post-partum women, and to children under age five who are at nutritional risk. In order to be eligible, household gross income must be below 185% of the federal poverty line or households must participate in Medicaid. In 2013, 9.05 million children aged one to four in the United States were eligible for WIC, and 49.8% (4.51 million) participated in the program. Younger children have high participation rates, with about 68.8% of eligible children aged one participating in WIC in 2013, and 50% of children aged two participating in the program. In contrast, the participation rate for children who were three and four years-old for the same year were 47.3% and 32.9%, respectively (USDA 2015). However, despite the lower take-up rate, a sizable number of American children who were aged 4, approximately 858,000, were WIC participants (Thorn et al. 2015). The impact of WIC has been well documented (Bitler et al. 2005; Oh, Jensen, and Rahkovsky 2016); studies show WIC participation has had moderate, positive effects on birth outcomes, that is, decreased preterm birth and increased birth weight (Kowaleski-Jones and Duncan 2002; Bitler and Currie 2005; Figlio, Hamersma, and Roth 2009; Foster, Jiang, and Gibson-Davis 2010; Hoynes, Page, and Stevens 2011); reduced infant mortality rates for African-Americans (Khanani et al. 2010); and increased intake of three of the four key nutrients in early childhood (Yen 2010). Participation in the WIC program has been also associated with improved cognitive and behavioral outcomes at home and in school (Hicks, Langham, and Takenaka 1982; Rush et al. 1988; Jackson 2015). Finally, analyses by Arteaga, Heflin, and Gable (2016) using the Early Childhood Longitudinal Study-Birth Cohort and a regression discontinuity design found that household food insecurity increases when children reach month 61 and age out of eligibility for WIC. However, upon turning old enough to enter school, children can receive school-based food and nutrition programs such as the National School Lunch Program (NSLP). The NSLP is administered at the school level, and upwards of 97% of public schools participate (NCES Schools and Staffing Survey 2013a, 2013b). In fiscal year 2011, over 31 million students received a free or reduced-price lunch daily; according to Dahl and Scholz (2011), participation rates among eligible children are 75% for the NSLP. Empirical studies evaluating the NSLP’s effects on food insecurity, health, and school outcomes are limited (Currie 2003; Gundersen, Kreider, and Pepper 2011). However, estimates of individual impacts of participation in NSLP vary. While Dunifon and Kowaleski-Jones (2003) find no evidence of positive effects on children, Bartfeld and Dunifon (2006), Gundersen et al. (2012), and Frisvold (2015) all find positive effects of school food programs on child outcomes, ranging from obesity and health to math and reading. Closer to the current study, Arteaga and Heflin (2014) used variation in states’ kindergarten age eligibility cutoff date and ECLS-B data and found that NSLP participation protects against household food insecurity during the transition to kindergarten. Present Study This paper answers two research questions that relate to the duration of nutrition program exposure. First, how reasonable is it to assume that the coverage gap length affects children’s cognitive outcomes at school entry? We use multivariate analysis to address this question by regressing the coverage gap length on children’s cognitive test scores and controlling for family background characteristics, childrens’ individual characteristics, demographic characteristics, school characteristics, and participation in other programs. These models include state fixed effects, controlling for all unobserved time-invariant state-level determinants of children’s cognitive skills such as state-specific food programs or policies towards education. A question that might arise is whether parents “select” their children’s coverage gap length. We conduct a thorough analysis and conclude that it is highly unlikely that parents are directly choosing the coverage gap length; given that the coverage gap length is determined by federal rules determining the age when WIC eligibility ends and state and local rules determining the age when children can have access to NSLP. Moreover, our sensitivity tests provide powerful information that indicates that state and school districts’ policy makers are not setting cutoff dates for the beginning of kindergarten based on how prepared they think the children in the state are. In other words, our sensitivity analysis provides evidence that strengthens the case for the exogeneity of the coverage gap-length independent variable. Second, we study whether the effect of the duration in food program access on child’s cognitive development fades out by spring of the kindergarten year, once children are exposed to the NSLP. We expect that an increase in the length of the coverage gap will lead to a reduction in scores for reading and math upon kindergarten entry (hypothesis 1). We also expect that the effects of the length of the coverage gap will fade out by spring of the kindergarten year, once children are exposed to the NSLP through attendance at a full-day program (Hypothesis 2). Data and Sample Analysis of this research question will rely upon data from the Early Childhood Longitudinal Study- Kindergarten (ECLS-K: 2011). The ECLS-K: 2011 is a nationally representative sample of about 18,000 children, selected from both public and private schools, who attended either full-day or half-day kindergarten in 2010–11, and who will be followed through the fifth grade. The panel data for fall and spring of the kindergarten year with non-missing data for child’s cognitive tests consists of 14,600 children. We constrained the sample to those children who attended kindergarten for the first time (n=13,950) because those who attended kindergarten more than once were older and unaffected by the coverage gap. We then constrained the sample to those for whom we had information on the coverage gap (n=12,500). Moreover, we further limited this sample to children who were interviewed during the first two months of the school year when exposure to school meals programs was limited (n=7,850). Additionally, we constrained this sample to those cases for which the ECLS-K collected data on household income level so that we could infer eligibility for food assistance programs (n=6,050). All reported sample sizes are rounded to the nearest 50 in compliance with NCES security standards. In this study, we specifically examine a subsample of children who were income eligible to participate in the WIC and NSLP programs at the time of the survey.1 WIC guidelines are based on household income at or below 185% of the FPL. However, the ECLS-K only reports household income at or below 100% of the FPL and at or below 200% of the FPL. We used the latter for our analysis. The sample size for this group is about 2,350 children and can be thought of as the intent-to-treat sample (ITT). We focus our analysis specifically on children who attended a full-day kindergarten program (n=1,950).2 We focus on this analytic sample because the potential benefit of the NSLP is restricted to children who are present at kindergarten all day in order to access meals provided on-site. In our main analysis, we use an intent-to-treat sample that recodes children who participated in Head Start and who are income eligible to participate in nutritional programs at the time of the survey. For those with coverage gap length equal or greater than three months, we recoded their coverage gap length as three.3 Our main analysis uses this sample to address the concern that children who participated in Head Start had access to the school lunch program during the academic year, regardless of their age, thus contaminating the definition of “coverage gap length.” The ECLS-K: 2011 uses a multi-stage survey design where the first-stage sampling unit is a county or group of counties, the secondary-stage unit is the schools sampled within the counties, and the third-stage sampling unit is the students in the schools. ECLS-K: 2011 uses different sampling weights for each sampling stage and survey weights are used to account for the clustered and multi-staged sampling frame of the ECLS-K: 2011. We used STATA’s svy command for all analyses and we adjusted standard errors to account for the complex survey design. Missing data on covariates, though negligible, were included but were identified with dummy variables. Measures Dependent Variables Our key dependent variables consist of math and reading cognitive variables. The math assessment tested children’s recognition of shapes, colors, sizes, numbers, and number counting, while the reading assessment examined print familiarity, letter recognition, beginning and ending of sounds, rhyming words, word recognition, and vocabulary knowledge. Achievement test scores in reading and mathematics are measured as part of the ECLS-K assessment in the fall and spring of the kindergarten year. The ECLS-K used content domains that were borrowed from the National Assessment of Educational Progress. Some questions were taken directly or adapted from copyrighted instruments such as the Peabody Individual Achievement Test (PIAT-R), the Peabody Picture Vocabulary Test (PPVT-III), the Preschool Language Assessment Scale (preLas 2000), the Test of Early Mathematics Ability (TEMA-3), the Test of Preschool Early Literacy (TOPEL), and the Woodcock-Johnson Psychoeducational Battery (WJ-III), among others. The tests were administered using a computer-assisted method in which the interviewer entered children’s responses. Children were presented images and only had to point to answers or provide verbal responses and were not asked to write or explain their reasoning. The ECLS-K used a two-stage assessment in reading and mathematics. Children were administered basic questions in the first stage and based on their responses, they were then given a set of questions appropriate for their demonstrated skill level, instead of all items in the survey instrument. Both reading and math scores were computed using item response theory (IRT) scores (Rock and Pollack 2002) that are included in the ECLS-K dataset. The IRT uses patterns of correct, incorrect, and omitted answers to obtain ability estimates that allow us to compare scores from different levels of difficulties given that different children respond to different questions in the ECLS-K (Hambleton, Swaminathan, and Rogers (1991). The National Center for Educational Statistics recommends that the age at the time of assessment or date of assessment is used as a control in regression models in which a child’s cognitive score is a dependent variable (Tourangeau et al. 2012). However, because we construct our main independent variable—coverage gap length—using the distance of the child’s age to the cutoff rules, this variable is highly collinear with child’s age. Auxiliary regression models that use raw IRT test scores as a dependent variable, coverage gap length as an independent variable, and child’s age as a control variable show a Variance Inflation Factor (VIF) above 10 for both coverage gap length and child’s age, indicating that these two variables are highly collinear, and thus it is difficult to draw conclusions about the magnitude and significance of estimated coefficients. To address this collinearity problem, we age-normed our measurements of achievement. It is important to mention that there is not a national age-normed version of the instrument because not everybody answered exactly the same questions. Remember that the assessment occurred in two stages and questions were administered according to children’s demonstrated ability in the first stage. Moreover, in the case of ECLS-K, they used different components of different standardized tests for their reading and math components. Thus, after consulting directly with ECLS-K specialists, we decided to use age-group categories that have been previously used in the literature to age-norm other instruments like PPVT-III and WJ-III. Given the ages of children in the Fall of Kindergarten, we created eight different groups: 3.5–4 years of age (category 1); 4–4.5 years (category 2); 4.5–5 years (category 3); 5–5.5 years (category 4); 5.5–6 years (category 5); 6–6.5 years (category 6); 6.5–7 years (category 7); 7–8 years (category 8). It is important to note that we had thousands of observations for categories 2 through 6, but fewer observations for the rest of the categories. When closely looking at our sample for this study (n= 2,350), we found that we had less than 50 observations altogether for categories 1, 7, and 8, so we dropped them from our analysis. We then calculated means and standard deviations for IRT scores for both reading and math for each group using all ECLS-K children because the sample is nationally representative of all kindergarteners. Means and standard deviations are very consistent with expectations, with means increasing for each higher age-group category. Using these means and standard deviations, we age-normed our dependent variables using a typical normalization procedure: IRT score – IRT mean for that age category, divided by the standard deviation for that age category. Thus, we obtained age-normalized scores with a mean of zero and standard deviation of 1, which provides model estimates that can be interpreted as effect sizes (Cohen 1988). Independent Variable Children who participate in WIC can only be enrolled until 60 months of age; according to the USDA, there are no category errors due to over-aged children receiving WIC benefits (USDA 2012). In order to access the NSLP program, children must be enrolled in a formal kindergarten program. Kindergarten enrollment is largely based on turning age 5 by a cut-off date, which varies by state, from July 31 in Nebraska, to January 1 in Connecticut. Moreover, the start of the academic year also varies by state and within state by school district. For example, in the academic year 2010–2011, some school districts in Nevada started the academic year on July 12 while some Ohio school districts started on September 26. As a consequence, our key variable of interest is the coverage gap length, which is equal to the age of the child on the first day of kindergarten, minus 60 months (the age when a child loses WIC eligibility). For example, if the academic year starts on 1 September in state s, and child i is 66 months old at that time, then the coverage gap length for child i is 6 months. Coverage gaps of more than 365 days are dropped from the analysis (n = 1,100 for the full-sample, and n = 350 for the ITT sample) because they indicate that the child was red-shirted, or the parents voluntarily delayed kindergarten entry for a year. We eliminate these children from our analytic sample since we only are interested in the effect of the coverage gap length for children who were affected by school eligibility rules and for whom the coverage gap is exogenous to parental decision-making. We assume that these children would be unlikely to be affected by the coverage gap and would bias our results towards zero. Additionally, not all children have a coverage gap. About 200 children turned five after beginning kindergarten, meaning that they experience no gap in food program coverage; we include these children in our analysis and assigned a coverage gap equal to zero. Thus, our key independent variable provides the basis to estimate intent-to-treat effects of the coverage gap length on child’s well-being.4 States and local education agencies (LEA; also known as school districts) vary on their kindergarten age eligibility criteria. In 2010, the year in which students observed in the ECLS-K entered kindergarten, 44 states and the District of Columbia had a statewide age-eligibility policy, the majority of which fell in September (Education Commission of the States 2011). Six states allowed each LEA to decide its own cutoff date by which children had to turn five to be eligible for kindergarten enrollment: Colorado, Massachusetts, New Hampshire, New Jersey, New York, Ohio, and Pennsylvania. For the approximately 70 school districts observed in ECLS-K from these states, we obtained information about their kindergarten eligibility by contacting officials from the school district. To sum up, to create our independent variable of interest, we combined data from ECLS-K: 2011 and from school districts (direct contact by researchers). Control Variables We control for child and family characteristics. Specifically, we control for race and ethnicity (Black, White, Asian, Hispanic, or Other), gender (male or female), and quarter of the child’s birth (Buckles and Hungerman 2013; Cascio and Schanzenbach 2016). We also control for the following family characteristics: number of children in household, number of adults in household, maternal marital status (married, widowed/divorced or separated, or never married), parental education level (less than high school, high school degree, some college, 4 years or more), 12-month household food security level, prior Head Start participation, TANF participation, SNAP participation, school lunch participation, the percentage of the school participating in free or reduced lunch, and public school attendance. We control for SNAP participation because our variable of interest—coverage gap length—captures the gap in food programs (WIC and NSLP) after controlling for all other programs. Similarly, we control for prior participation in Head Start because children who attend Head Start programs receive NSLP. We also control for the survey’s assessment month, urban residence, and state of residence using state fixed-effects. Table 1 presents descriptive statistics for our control variables. We observe that for our sample of interest (n = 1,950), 37% of children are white, 17% are Black, 35% Hispanic, and the rest are from other race/ethnicity. About half of the children are male (51%). Birth seasonality seems to be very similar among quarters. As expected for our WIC-NSLP eligible sample, 20% of children previously participated in Head Start. Similarly, annual household income was on average about $24,000. Also, a higher percentage of parents have never been married (22%), which is consistent with other studies of low-income families (Cabrera et al. 2004; Pruett et al. 2017). Overall, we observe that the percentage of missing values for each variable category is low and below 8%, except for school percentage in free and reduced lunch program (16%). Table 1 Descriptive Statistics Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Note: Standard deviations appear in parentheses next to the mean value for continuous. As per National Center for Education Statistics requirements, all n values are rounded to the nearest 50. Table 1 Descriptive Statistics Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Variable  Full-day sample     Includes Head Start children     (n = 1,950)  Duration of gap, months  4.654  (3.215)  Child characteristics   White  36%     Black  19%     Hispanic  34%     Asian  4%     Other race  6%     Male  51%     Child born in first quarter  27%     Child born in second quarter  26%     Child born in third quarter  21%     Child born in fourth quarter  26%     Head Start participation in preschool  21%    Household characteristics   Household income, thousands  23.729  (13.516)   Household income, missing  0%     No. in household, over 18  2.064  (0.816)   No. in household, under 18  2.691  (1.199)   No. in household missing  12%     Parents married  51%     Parents previously married  15%     Parents never married  23%     Marital status missing  11%     Household is food secure  74%     Food security status missing  5%     Urban  64%     Urban missing  1%     Mother less than high school  23%     Mother high school  32%     Mother some college  34%     Mother college  11%     Family receives TANF  10%     Family receives SNAP  52%    School Characteristics   Private school  7%     School lunch program participation  87%     School lunch program missing  8%     School percent FRL  65%     School percent FRL missing  14%    Note: Standard deviations appear in parentheses next to the mean value for continuous. As per National Center for Education Statistics requirements, all n values are rounded to the nearest 50. Empirical Strategy To examine the relationship between length of the coverage gap and children’s cognitive development, we estimate a series of multivariate regression models. The outcomes of child i living in state s (CHILDis) are functions of the WIC-NSLP’s coverage gap length (DURi), a vector of controls, X, (child and family characteristics), state dummy variables, d, (fixed effects) and an error term, ɛ:   CHILDfallis=β0+β1DURfallis+β2Xis+γs∑ds+ɛis (1) where DURis= AgeBSis – 60, AgeBSis is the age of child i at the beginning of the school year in state s, and 60 is the age in months when a child ages out of WIC. Estimating equation (1) with ordinary least squares (OLS) yields an unbiased estimate of the impact of DUR if the unobserved determinants of child well-being are uncorrelated with the coverage gap length. We argue that this assumption is likely to hold because the construction of DUR is based on eligibility rules that are exogenous to parental decision making. Moreover, while a child’s birthday might be considered endogenous, the distance between the child’s fifth birthdate and the start of school is likely to be exogenous. Most parents do not know five years (and nine months) in advance the age eligibility rule for the school their children will attend. Another potential source of endogeneity might arise from policy makers if they choose the starting date for kindergarten based on how prepared they think the children in the state or LEA are. If this is true, then starting dates might be endogenous even though children and parents are not directly choosing them. We might think of a case where schools with a higher commitment to educational quality have earlier start dates. To address this concern, we regressed school start date on school-level scores, controlling for state fixed effects as an additional sensitivity test. While we acknowledge that exogeneity may be a strong assumption, we provide a series of sensitivity tests to back up this claim. Finally, it is important to note that we control for quarter of child’s birthdate in our regressions. We start our analysis with our intent-to-treat sample (our preferred sample) of children who attend full-day kindergarten and who were WIC eligible at the point of kindergarten entry (n = 2,350), that is, who satisfied the poverty eligibility measure (i.e., intent to treat analyses: ITT). Our hypothesis will be confirmed if we find that β1 is negative and statistically significant in the fall for the cognitive outcomes, meaning that DURis negatively affects children’s cognitive development (or if we find that β1 is positive and statistically significant for internalizing and externalizing problem behaviors). Our hypothesis will be unsupported if we find that β1 is not statistically significant. To examine our second research question regarding the fade out of the duration gap, we also use multivariate regression analysis. We regress children’s outcomes measured in the spring of 2011, controlling for fall 2010 variables, as follows:   CHILDspring is=α0+α1DURis+α2Xfall is+γs∑ds+ɛis. (2) By the spring, children would have been receiving free or reduced-priced lunch for about six months. Thus, the only difference between equations (1) and (2) is that the dependent variable in equation (2) is measured in the Spring instead of the Fall. Our hypothesis will be confirmed if β1 (coefficient of DUR on equation [1]) is statistically greater than α1 (coefficient of DUR on equation [2]); in other words, we expect the effects of DUR to fade out by the spring if children attend a full-day program and can access school-based meal programs. For our sensitivity analysis using our overall sample (full-day and half-day groups), it is possible that our results may be more mixed because of the presence of the half-day kindergarten group which may not have access to school meals. Results Main Results (Full-Day Kindergarten Sample) Fall Outcomes Table 2 reports results for our main sample that includes only children who attended full-day kindergarten, which is our intent-to-treat sample (n = 1,950). We find that an additional month of the coverage gap decreases reading IRT scores by 0.021 of a standard deviation ( p < 0.01). Conversely, an additional month of the coverage gap does not have a significant effect on math IRT scores (coefficient=−0.008 standard deviations; p > 0.10). Table 2 Effects of Coverage Gap Length of Food Programs on Child’s Skills for Children Attending Full-Day Kindergarten (n = 1,950)   Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)      Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Table 2 Effects of Coverage Gap Length of Food Programs on Child’s Skills for Children Attending Full-Day Kindergarten (n = 1,950)   Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)      Fall  Spring  Difference  1) Reading  −0.0207***  −0.0052  −0.0155***    (0.0060)  (0.0095)    2) Mathematics  −0.0083  −0.0052  −0.0031    (0.0075)  (0.0091)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Spring Outcomes Next, we examine our second hypothesis that the effects of the coverage gap length will fade out by spring once children are exposed to the NSLP in school. For our intent-to-treat sample (n = 1,950), we find that an additional month of the coverage gap does not have a statistically significant effect on math or reading IRT scores. Moreover, for reading, the difference in the effects of the coverage gap length for fall and spring is a reduction of IRT scores of 0.016 of a standard deviation, and this difference is statistically significant. Remember that this is equivalent to a total effect size of 0.094 of a standard deviation for the average total coverage gap. These results suggest that the negative effects of the coverage gap length prior to kindergarten entry have faded by spring. Sensitivity Analysis We present four different sensitivity analyses designed to test the OLS estimate’s robustness to a variety of modeling choices, assumptions, and samples. First, we test how sensitive our results are to different kindergarten day arrangements by examining the full sample of kindergarteners and not limiting our analysis to full-day program attendees in order to increase the external validity of our findings. Second, we create a proxy sample to estimate Treatment-on-the-Treated effects by using an “ever WIC” participation variable and constraining the sample to those with WIC eligibility prior to a child’s fifth birth date. Third, in order to test if the coverage gap is correlated with unobservable factors, we replicate our analysis of the coverage gap on a sample of advantaged children who were not eligible for either WIC or school-based nutrition programs based on their kindergarten year. Finally, we test the validity of our DUR measure by analyzing whether it is only identifying the effects of differences in school start date solely, after controlling for child’s age. In all cases, presented below, we find that results are consistent with our expectations and confirm our main finding presented above. Full and Part-Time Kindergarten Program Analysis Table 3 presents results for our first two sensitivity tests. In the first two rows of table 3, we replicate the analysis shown in table 2 for the ITT sample, but we add half-day kindergartners to the analysis sample. Ideally, we would be able to replicate our analysis on the half-day sample alone but sample size (n = 400), prevents us from adopting this approach. For panel A of table 3, results are consistent with those shown in table 2. For reading, we observe that in the fall, the association between DUR and our outcome is negative but there is no association for the spring. Once again, difference between Fall and Spring reading scores is negative and statistically significant. For math, we did not find statistically significant effects for the difference between Fall and Spring. Table 3 Sensitivity Analysis: Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, ITT Analytical Sample   Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)      Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. As defined in the text, panel A includes the full-day, plus half-day kindergarten sample (n = 2,350), and panel B includes those in full day kindergarten who ever received WIC. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Table 3 Sensitivity Analysis: Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, ITT Analytical Sample   Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)      Fall  Spring  Difference  Panel A: Full=Day + Half-Day Sample (n=2,350)        1) Reading  −0.0094  0.0023  −0.0117***    (0.0059)  (0.0081)    2) Mathematics  0.0061  0.0102  −0.0041    (0.0049)  (0.0058)    Panel B: ToT (Ever WIC), Full Day Sample (n=1,550)        1) Reading  −0.014*  −0.005  −0.009***    (0.0073)  (0.011)    2) Mathematics  −0.001  0.004  −0.005    (0.008)  (0.011)    Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. As defined in the text, panel A includes the full-day, plus half-day kindergarten sample (n = 2,350), and panel B includes those in full day kindergarten who ever received WIC. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting, and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Treatment-on-the-Treated Effects (ToT) While our initial goal for this study was to examine treatment on the treated effects, our dataset did not allow us to conduct such analysis: the ECLS-K: 2011 did not collect information on WIC participation at age 4 specifically, but only gathered information on whether a child ever participated in the WIC program. Using this information, we constructed a “proxy” of ToT effects that we call ToT Ever WIC, which consists of children who ever participated in WIC and who were eligible for WIC in the fall of Kindergarten (n = 1,550). It is important to keep in mind that this is a proxy variable and likely contains measurement error in terms of identifying children who received WIC at age 4. While typical ToT analyses have larger effect sizes than ITT estimates since the untreated are eliminated from the sample, in this case we lack the precision to correctly identify our ToT sample. Nonetheless, we present results for our proxy ToT measures in panel B of table 3. We find similar results to those found in our main analysis (table 2) for ToT Ever WIC. The difference in the effects between Fall and Spring for reading IRT scores is negative and statistically significant, while we do not find statistically significant effects for math IRT scores. Advantaged Sample Falsification Test Next, we present an additional sensitivity analysis to reduce concerns that the effects that we are identifying are due to other confounding factors and not necessarily related to access to food and nutrition programs. In table 4, we replicate the analysis shown in table 2, but instead of limiting our sample to those who qualify for WIC and NSLP (ITT sample), we limit our sample to those children who live in households with incomes above $50,000 at the time of the survey and who have never participated in the WIC program. If the effect that we are able to identify in our prior models is coming from the coverage gap length and not from other confounding factors, then we would expect to observe no effect for the sample that is not income-eligible for the WIC and NSLP. Results show that the coverage gap length does not have a statistically significant association with either math or reading scores in the Fall or the Spring. Thus, our findings pass this falsification test. Table 4 Falsification Test: Estimated Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, Advantaged (High Income and not WIC Eligible) Sample   Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)      Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)    * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 Table 4 Falsification Test: Estimated Effects of Coverage Gap Length of Food Programs on Child’s Skills in Kindergarten, Advantaged (High Income and not WIC Eligible) Sample   Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)      Fall  Spring  Difference  Advantaged Sample (n=1,400)        Reading  0.0176  0.0098  0.0078    (0.0140)  (0.0063)    Mathematics  0.0127  0.00923  0.0035    (0.0130)  (0.0071)    * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses and clustered at the school level. Each row represents a different regression. Each regression controls for child’s age, gender, race and ethnicity, season of birth, Head Start participation in preschool, household income, maternal marital status, maternal education, number of children in the household, receiving TANF, food security, urban area, school setting and free/reduced lunch program participation. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: * p<.1; ** p<.05; *** p<.01 School Start Date and School Quality While DUR is the distance between a child’s birthdate and the beginning of the school year, some might be concerned that DUR is picking up the effects of differences in school start date only after controlling for child’s age. This might be the case if schools with a higher commitment to educational quality have earlier start dates. To address this concern, we regressed school start date on school-level scores, controlling for state fixed effects. We used two measures for school start date: date in days, and date in weeks. Table 5 shows no statistical differential effect of school start date on school-level scores for either measure. This finding demonstrates that the main results presented in table 2 are not driven by unobserved differences in school quality related to school start date, but by the coverage gap length in access to food and nutritional programs. Table 5 Sensitivity Analysis: Estimates of the Effects of School Start Date Only on School Level Scores   Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)    Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)  * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses. Each row represents a different regression. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: = p<0.1; ** = p<0.05; *** = p<.01. Table 5 Sensitivity Analysis: Estimates of the Effects of School Start Date Only on School Level Scores   Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)    Starting week (n = 450)  Starting date (n = 450)  Average Reading Achievement  −0.0104  −0.00182  (0.0074)  (0.0012)  Average Math Achievement  −0.009  −0.0015  (0.0100)  (0.0015)  * Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. Standard errors are in parentheses. Each row represents a different regression. We use state fixed effects to control for unobserved characteristics at the state level. Asterisks indicate the following: = p<0.1; ** = p<0.05; *** = p<.01. Together, the sensitivity analyses carried out in tables 3 through 5 provide robust evidence of the main results found in table 2: a negative effect of the coverage gap length on child achievement scores for reading in the Fall of kindergarten and a statistically significant difference between Fall and Spring. At the same time, our analysis does not show any consistent effects for math. Furthermore, we consistently find evidence that the negative effects fade out by the spring. Given the analysis shown, we believe that it is reasonable to argue that the negative effects observed in the Fall for reading were likely driven by the coverage gap length and not by other confounding factors. Discussion and Conclusions Our study uses the nationally representative ECLS-K to suggest how having limited access to food and nutritional programs during the time when a child transitions to kindergarten may shape child cognitive achievement over the course of kindergarten. Age eligibility rules for WIC and the timing of kindergarten entry mean that many children experience a gap in access to both WIC and the NSLP that can last from several months up to a year. Our study finds mixed results. We found some evidence of negative effects of the coverage gap length in reading, but not in math. These results were consistent across different models. One consequence of this coverage gap is that children enter kindergarten with lower cognitive scores, on the order of 0.02 for each month of the coverage gap, for reading for each gap month experienced. While at first effect sizes in this range may be considered small, our effects are additive across coverage gap months (0.02 x 4.7 months of coverage gap= 0.094). Thus, when comparing our findings to interventions that affect educational outcomes, effect sizes in this range are considered moderate (Cohen 1988). Results presented in this paper are consistent with previous work that found an increase in rates of food insecurity for children who become age-ineligible for WIC and who have not yet started kindergarten (Arteaga, Heflin, and Gable 2016). Additionally, it has been shown in previous work that when children enter kindergarten and are able to access the NSLP, they become less food insecure (Arteaga and Heflin 2014). This study contributes to the literature by examining the consequences at kindergarten entry for child reading and math scores for the family disruption caused by the withdrawal of food and nutritional support when a child reaches 60 months of age. It should be noted that our study also finds that by the Spring of kindergarten, the consequences of the coverage gap for reading have ameliorated. Moreover, policy makers should be aware of the existence of this nutritional gap during a child’s transition into school as a policy failure that could be avoided by extending WIC eligibility to school-age. Given the heightened scrutiny that all social spending is facing in today’s political climate, our research provides clear empirical support for the positive role that nutritional assistance programs can play in the development of cognitive skills upon kindergarten entry. It is important to note, however, that our study does not speak to which form of food assistance is better (i.e., WIC or NSLP), but to the need to support consistent household food consumption over the childhood period. Limitations Despite the contributions of this study, it is important to recognize several limitations. First, the sample of income-eligible families is small. This is a typical challenge when using nationally representative datasets. Second, the ECLS-K does not contain information on WIC participation at the child’s fifth birthday or whether the household was income-eligible to participate in WIC when the child was four years of age. Thus, there is a possibility of measurement error in determining the sample. We used data collected in the spring of kindergarten, which documents household annual income in the prior year for identifying income-eligibility for our intent-to-treat sample. It is still possible that annual income volatility might have changed income eligibility between the time of data collection and when the child was age-eligible for WIC. Additionally, for our treatment on the treated sample, we used the question that documented if the household “ever participated in WIC” because the ECLS-K did not collect information on WIC participation when the child was four years of age. This might also introduce measurement error into the analysis because national figures show that the percentage of WIC participants is higher for infants and toddlers than for preschoolers. It is then possible that some families who should have not been included because they no longer participated in WIC are included in our ToT sample. However, findings were unchanged when models were re-estimated using a different sensitivity analysis. We consistently observed a reduction in the food coverage gap between Fall and Spring for reading; the difference between these two periods of time was statistically significant. In addition, we argue that our analytical strategy does allow us to make causal inferences about the association between the coverage gap length of food and nutritional programs and a child’s cognitive and socio-emotional outcomes. Both aging out of WIC and NSLP participation are determined exogenously by USDA rules, a state’s age eligibility cutoffs, and school districts’ decisions on when to start the school year. Our sensitivity analysis that estimates the effects of start of school date on test scores provides evidence that state age eligibility rules or school district calendars are unlikely to be endogenously determined with respect to birth date. To the extent that one might wonder if turning five years old is associated with some other unobserved changes related to child outcomes that our analysis is picking up, our sensitivity analysis with the advantaged sample and the school data demonstrates null findings of duration. Therefore, we believe it is unlikely that other positive confounding factors are biasing the negative effects that we find for coverage length gap on children’s outcomes. Policy Implications There are several important implications of these findings. First, our analysis indicates that a coverage gap in access to federal food and nutrition programs is associated with poor child outcomes at kindergarten entry, and these effects may attenuate by spring for children who have access to full day kindergarten. It is important to follow these children over time and see if this effect continues to be zero or if it grows in some children. Our findings suggest that the coverage gap length needs to be reduced because it reduces reading scores at kindergarten entry, a time when children are often placed on learning trajectories and teachers and administrators gather impressions about children’s learning potential. The fact that the gap fades by Spring of the kindergarten year suggests that children are very sensitive to their food intake around kindergarten entry and that it is not negative selection into the WIC program that is behind their low reading scores. Stated differently: our results suggest that given an adequate level of resources, WIC-eligible children can learn. There is no reason for them to begin their educational careers behind their more advantaged peers. An easy way to address this problem would be to extend the age of WIC eligibility to the month of state eligibility for kindergarten entry instead of age 5 as is the current practice. Our analysis suggests that this small change could lead to a substantial improvement at kindergarten entry. Given how key notions of equality of opportunity are to the American Dream, policy makers should address the unintended consequences of compounding disadvantage for millions of children each year who are unlucky enough to be born in the wrong state and in the wrong month, both of which prolong their coverage gap. Funding This research was made possible by the generous support from the Institute for Research on Poverty (IRP) RIDGE Center for Policy Research at the University of Wisconsin and the United States Department of Agriculture. The authors thank participants at the 2016 IRP conference in Madison and Washington D.C., and participants at the 2016 APPAM Fall Research Conference for their helpful comments. The statements and views expressed in this article do not reflect those of the study’s sponsors. Any remaining errors are our own. Footnotes 1 NSLP eligibility is based on household income at or below 130% of the federal poverty line (FPL) for free lunch and at or below 185% for reduced-price meals. The community eligibility provision for the NSLP was not enacted at the time of this study. 2 Once a child starts school, the potential benefit of the NSLP depends on how school districts administer kindergarten, since children must be present at school to access meals provided on-site. In 2013, 76% of kindergarteners in the United States attended full-day programs. There were, however, slight variations in full-day attendance rates by race and the highest level of educational attainment of the parents/ guardians. Indeed, 77% of white, 79% of blacks, and 74% of Hispanic kindergarteners were enrolled in full-day programs. Conversely, the full-day attendance rate for Asian kindergarteners was significantly lower at only 65.5%. Further, 69% of children whose parents or guardians had less than a high school education attended full-day kindergarten, compared to 78% whose parents or guardians had completed high school or had a GED, and 79% whose parents had vocational/technical, or some college education. The full-day attendance rates for children whose parents or guardians had an associate's degree, a bachelor's degree or a graduate or professional degree were 75%, 77% and 72%, respectively (NCES 2014). 3 We do not have information on Head Start participation during the summer. National data shows that there are not many Head Start programs that are open during the summer. Thus, we assumed that children did not have access to Head Start during the summer. 4 Although it would be ideal to have knowledge on whether a child was participating in WIC just before kindergarten entry, ECLS-K data do not provide this information. That is, we only know if a child was income-eligible to participate in WIC at the time of the fall survey of the kindergarten year and whether a child ever participated in WIC. We explored the possibility of using the ECLS-B instead since it contains information on WIC participation at 48 months of age and until a child turned 60 months of age. However, the ECLS-B does not gather administrative information on the start date of the school year. Moreover, the ECLS-B only collects data for kindergarteners once and it can be at any time over the academic year. Given these limitations, we opted to use the ECLS-K. Our analysis uses information for those who were eligible to participate in WIC and calculates the duration of the coverage gap for those who were intended to be treated. References Arteaga I., Heflin C.. 2014. Participation in the National School Lunch Program and Food Security: An Analysis of Transitions into Kindergarten. Children and Youth Services Review  47: 224– 30. Google Scholar CrossRef Search ADS   Arteaga I., Heflin C., Gable S.. 2016. The Impact of Aging out of WIC on Food Security in Households with Children. Children and Youth Services Review  69: 82– 96. Google Scholar CrossRef Search ADS   Bartfeld J., Dunifon R. 2006. State‐level predictors of food insecurity among households with children. Journal of Policy Analysis and Management  25 4: 921– 42. Google Scholar CrossRef Search ADS   Bitler M., Currie J.. 2005. Does WIC Work? The Effects of WIC on Pregnancy and Birth Outcomes. Journal of Policy Analysis and Management  24 1: 73– 91. Google Scholar CrossRef Search ADS PubMed  Bitler M., Gundersen C., Marquis G.S.. 2005. Are WIC Nonrecipients at Less Nutritional Risk than Recipients? An Application of the Food Security Measure. Review of Agricultural Economics  27 3: 433– 8. Google Scholar CrossRef Search ADS   Buckles K.S., Hungerman D.M.. 2013. Season of Birth and Later Outcomes: Old Questions, New Answers. Review of Economics and Statistics  95 3: 711– 24. Google Scholar CrossRef Search ADS PubMed  Cabrera N.J., Ryan R.M., Shannon J.D., Brooks-Gunn J., Vogel C., Raikes H., Cohen R.. 2004. Low-Income Fathers’ Involvement in Their Toddlers’ Lives: Biological Fathers from the Early Head Start Research and Evaluation Study. Fathering  2 1: 5. Google Scholar CrossRef Search ADS   Cascio E.U., Schanzenbach D.W.. 2016. First in the Class? Age and the Education Production Function. Education Finance and Policy  11 3: 225– 50. Google Scholar CrossRef Search ADS   Cohen J. 1988. Statistical power analyses for the social sciences . Hillsdale, NJ: Lawrence Erlbauni Associates. Currie J. 2003. US Food and Nutrition Programs. In Moffitt R. (Ed.), Means-Tested Transfer Programs in the United States , 199– 290. Chicago, IL: University of Chicago Press (Chapter4). Google Scholar CrossRef Search ADS   Dahl M., Scholz J.K.. 2011. The National School Lunch Program and School Breakfast Program: Evidence on Participation and Noncompliance. University of Wisconsin, Institute for Research on Poverty Working Paper. Dunifon R., Kowaleski-Jones L.. 2003. The Influences of Participation in the National School Lunch Program and Food Insecurity on Child Well-Being. The Social Service Review  77 1: 72– 92. Google Scholar CrossRef Search ADS   Education Commission of the States. 2011. State Characteristics: Kindergarten. https://www.ecs.org/clearinghouse/90/71/9071.pdf. Figlio D., Hamersma S., Roth J.. 2009. Does Prenatal WIC Participation Improve Birth Outcomes? New Evidence from Florida. Journal of Public Economics  93 1: 235– 45. Google Scholar CrossRef Search ADS   Foster E.M., Jiang M., Gibson-Davis C.M.. 2010. The Effect of the WIC Program on the Health of Newborns. Health Services Research  45 4: 1083– 104. Google Scholar CrossRef Search ADS PubMed  Frisvold D. E. 2015. Nutrition and cognitive achievement: An evaluation of the School Breakfast Program. Journal of public economics  124: 91– 104. Google Scholar CrossRef Search ADS PubMed  Gundersen C., Kreider B., Pepper J.. 2011. The Economics of Food Insecurity in the United States. Applied Economic Perspectives and Policy  33 3: 281– 303. Google Scholar CrossRef Search ADS   Gundersen C., Kreider B., Pepper J. 2012. The impact of the National School Lunch Program on child health: A nonparametric bounds analysis. Journal of Econometrics  166 1: 79– 91. Google Scholar CrossRef Search ADS   Hambleton R.K., Swaminathan H., Rogers H.J.. 1991. Fundamentals of Item Response Theory . (Measurement methods for the social sciences series, Vol. 2). Newbury Park, CA: Sage. Hicks L.E., Langham R.A., Takenaka J.. 1982. Cognitive and Health Measures following Early Nutritional Supplementation: A Sibling Study. American Journal of Public Health  72 10: 1110– 8. Google Scholar CrossRef Search ADS PubMed  Hoynes H., Page M., Stevens A.H.. 2011. Can Targeted Transfers Improve Birth Outcomes? Evidence from the Introduction of the WIC Program. Journal of Public Economics  95 7: 813– 27. Google Scholar CrossRef Search ADS   Jackson M.I. 2015. Early Childhood WIC Participation, Cognitive Development and Academic Achievement. Social Science & Medicine  126: 145– 53. Google Scholar CrossRef Search ADS   Khanani I., Elam J., Hearn R., Jones C., Maseru N.. 2010. The Impact of Prenatal WIC Participation on Infant Mortality and Racial Disparities. American Journal of Public Health  100 ( S1): S204– 9. Google Scholar CrossRef Search ADS PubMed  Kowaleski-Jones L., Duncan G.J.. 2002. Effects of Participation in the WIC Program on Birthweight: Evidence from the National Longitudinal Survey of Youth. American Journal of Public Health  92 5: 799– 804. Google Scholar CrossRef Search ADS PubMed  Oh M., Jensen H.H., Rahkovsky I.. 2016. Did Revisions to the WIC Program Affect Household Expenditures on Whole Grains? Applied Economic Perspectives and Policy  38 4: 578– 98. Google Scholar CrossRef Search ADS   Pruett M.K., Pruett K., Cowan C.P., Cowan P.A.. 2017. Enhancing Father Involvement in Low‐Income Families: A Couples Group Approach to Preventive Intervention. Child Development  88 2: 398– 407. Google Scholar CrossRef Search ADS PubMed  Rock D. A., Pollack J. M. 2002. Early Childhood Longitudinal Study-Kindergarten Class of 1998-99 (ECLS-K): Psychometric Report for Kindergarten through First Grade. Working Paper Series. Rush D., Leighton J., Sloan N.L., Alvir J.M., Horvitz D.G., Seaver W.B., Devore J.W.. 1988. The National WIC Evaluation: Evaluation of the Special Supplemental Food Program for Women, Infants, and Children. VI. Study of Infants and Children. The American Journal of Clinical Nutrition  48 2: 484– 511. Google Scholar CrossRef Search ADS PubMed  The National Center for Education Statistics. 2014. Digest of Education Statistics: Table 202.20. Percentage of 3-, 4-, and 5-year-old Children Enrolled in Preprimary Programs, by Attendance Status, Level of Program, and Selected Child and Family Characteristics: 2013. Available at: http://nces.ed.gov/programs/digest/d14/tables/dt14_202.10.asp? current=yes (accessed February 6, 2017). The National Center for Education Statistics Schools and Staffing Survey. 2013a. Table 1. Total Number of Private Schools and Students, and Percentage of Schools and Students that Participated in the Title I and Federal Free or Reduced-Price Lunch Programs, by Affiliation: 2011–12. Available at: http://nces.ed.gov/surveys/sass/tables/sass1112_2013312_s2a_001.asp (accessed February 6, 2017). The National Center for Education Statistics Schools and Staffing Survey. 2013b. Table 1. Total Number of Public Schools and Students, and Percentage of Schools and Students that Participated in the Title I and Federal Free or Reduced-Price Lunch Programs, by State: 2011–12. Available at: http://nces.ed.gov/surveys/sass/tables/sass1112_2013312_s2s_001.asp (accessed February 6, 2017). Thorn B., Tadler C., Huret N., Trippe C., Ayo E., Mendelson M., Patlan K., Schwartz G., Tran V. 2015. WIC Participant and Program Characteristics 2014. Prepared by Insight Policy Research under Contract No. AG-3198-C-11–0010. US Dept of Agriculture, Food and Nutrition Service, Alexandria, VA. Tourangeau K., Nord C., Lê T., Sorongon A.G., Hagedorn M.C., Daly P., Najarian M.. 2012. Early Childhood Longitudinal Study, Kindergarten Class of 2010–11 (ECLSK: 2011), User’s Manual for the ECLS-K: 2011 Kindergarten Data File and Electronic Codebook (NCES 2013–061). Washington DC: U.S. Department of Education: National Center for Education Statistics. U.S. Department of Agriculture. 2012. WIC Program Participation and Costs: April 2, 2012. Available at: http://www.fns.usda.gov/pd/wisummary.htm (accessed February 5, 2016). U.S. Department of Agriculture, Food and Nutrition Service, Office of Policy Support. 2015. National and State-Level Estimates of Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) Eligibles and Program Reach (Summary) . Alexandria, VA: Author. Available at: http://www.fns.usda.gov/sites/default/files/ops/WICEligibles2013-Summary.pdf (accessed February 5, 2016). Yen S.T. 2010. The Effects of SNAP and WIC Programs on Nutrient Intakes of Children. Food Policy  35 6: 576– 83. Google Scholar CrossRef Search ADS   Appendix Table A1 Descriptive Statistics for Outcome Variables   Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409    Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409  Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. View Large Table A1 Descriptive Statistics for Outcome Variables   Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409    Full-Day + Half-Day (n = 2,350)   Variables  Mean  S.D.  Fall outcomes       Math  45.528  8.823   Reading  45.758  8.673  Spring outcomes       Math  46.816  9.599   Reading  47.542  9.409  Note: As per National Center for Education Statistics requirements when using Early Childhood Longitudinal Study–K Cohort data, all n values are rounded to the nearest 50. View Large © The Author(s) 2018. Published by Oxford University Press on behalf of the Agricultural and Applied Economics Association. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

Applied Economic Perspectives and PolicyOxford University Press

Published: May 21, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off