Show Summary Details

Page of

Printed from Oxford Research Encyclopedias, Economics and Finance. Under the terms of the licence agreement, an individual user may print out a single article for personal use (for details see Privacy Policy and Legal Notice).

date: 05 December 2022

# The Academic Effects of United States Child Food Assistance Programs—At Home, School and In-Between

• Michael D. Kurtz, Michael D. KurtzEconomics, Lycoming College
• Karen Smith ConwayKaren Smith ConwayPeter T. Paul College of Business and Economics, University of New Hampshire
•  and Robert D. MohrRobert D. MohrPeter T. Paul College of Business and Economics, University of New Hampshire

### Subjects

• Health, Education, and Welfare Economics
• Labor and Demographic Economics
• Public Economics and Policy

### Introduction

In the decades since the U.S. Department of Agriculture (USDA) began tracking measures of food insecurity, the demographic group with the highest rate has been households with children; 14.8% of households with children—meaning 11.7 million children—are food insecure (Coleman-Jensen et al., 2021, p. 11). The U.S. government as well as many private charitable organizations provide a wide and varied range of food assistance programs designed to ameliorate food insecurity and hunger among children. These programs target both the school (e.g., the National School Lunch Program, NSLP) and the home (e.g., the Supplemental Nutritional Assistance Program, SNAP). The line between home- and school-based programs has become increasingly blurred with the onset of weekend and summer feeding programs and food pantries on school grounds, as well as recent policy initiatives that increase SNAP benefits during times of reduced access to school-based programs (e.g., summer months and periods of remote learning caused by the COVID-19 pandemic).

Past research has established that poverty and food insecurity have many adverse and potentially lasting consequences on children’s health and cognitive development (e.g., Fiese et al., 2011; Gundersen & Ziliak, 2015; Jackson, 2015; Shankar et al., 2017). Likewise, many studies investigate the effectiveness of and challenges facing these feeding programs in alleviating food insecurity (e.g., Fiese et al., 2011; Gundersen & Ziliak, 2014, 2018). Far less is known about the effects of these programs on child academic outcomes. While not the primary goal of food assistance programs, any resulting improvements in educational outcomes could have long-lasting effects on the social and economic attainment of low-income children and are additional benefits to be included in evaluating these programs.

The purpose of this article is to summarize what is known about the effects of U.S. food and nutritional assistance programs on child academic outcomes and to identify remaining gaps and directions for future research. This review begins with a discussion of the universal aspects of this line of research, such as the presumed mechanisms and the empirical challenges in isolating causal impacts. It then summarizes the findings for the primary food assistance programs in the United States. The discussion is guided by the organizational construct presented in Figure 1, which shows the two main settings for food assistance programs—at home and in school—as well as the emerging middle ground in-between. It also shows the two main sources of provision: government agencies and private, charitable organizations. Once again, the line is blurry, as many private programs receive government funding and oversight and some programs actively collaborate with public schools and local governments to reach food insecure children. Figure 1 places each program in context, both in terms of these two features as well as its size and longevity.

The discussion of individual food assistance programs starts with the public programs from the left side of Figure 1, beginning with programs where food assistance is provided directly to the household outside of the school environment (e.g., SNAP) before turning to public school-based programs (e.g., School Breakfast Programs [SBP]). Programs that fall in-between home and school and/or to the right side of the diagram include the two USDA summer food programs, the Child and Adult Care Food Program (CACFP) and weekend feeding programs, like Feeding America’s BackPack program. Each section provides a brief history of the relevant programs, verifies whether a prior literature has linked the programs to reductions in food insecurity, and then summarizes research that links the programs to educational outcomes. Each section is accompanied by a table that summarizes the data, empirical strategy, and results of key studies that examine the link between food assistance and academic outcomes.

### Universal Issues and Challenges—And the Limits of This Review

This review focuses on the effects of home and school food assistance programs (X) on child academic outcomes (Y). However, the path from X to Y consists of many links, complications, and possible mechanisms for effect, as shown in Figure 2 and as revealed by a standard household production model in which food assistance presumably increase the “inputs” (e.g., food) used to produce the “output” (child outcomes). Making a program available (X) does not ensure that a food insecure household participates. Take-up rates of SNAP and school meal programs tend to be low (Keith-Jennings et al., 2019; Vaudrin et al., 2018), which has led to program changes designed to reduce stigma and improve participation such as the replacement of food stamps with the SNAP electronic benefit transfer (EBT) program or expansion to universal free meals (UFM) in schools. Similarly, just because a child is offered a meal at school does not mean that they ultimately consume it. For example, children are more likely to take advantage of free school lunches near the end of the household’s SNAP cycle, when SNAP benefits are running low (Laurito & Schwartz, 2019). And, even if the child consumes the free meal, the benefits may spillover to other members or behaviors and thus be diluted. For instance, evidence exists that mothers compensate for the loss of eligibility of the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC) when their child turns age 5 by reducing their own nutritional intake (Bitler et al., 2022). Increased access to free school meals also results in decreased use of both food banks and retail food purchases (Marcus & Yewell, 2021; Ozturk et al., 2021). These studies further reveal the potentially overlapping and interacting effects of different programs.

Conditional on program participation, the presumed mechanism for food assistance programs to affect child academic outcomes includes, but is not limited to, reduced food insecurity and improved nutrition (i.e., improved “inputs”). It is intuitive that improved nutrition would lead to better child health, including cognitive development and the ability to learn, and past research establishes this link (e.g., Schanzenbach & Thorn, 2020). However, food insecurity has broader implications beyond hunger and poor nutrition. Food insecurity includes limited access to food and uncertainty about the ability to attain enough food (Gundersen & Ziliak, 2018, p. 119). As such, it may capture other deleterious effects of poverty and low income beyond poor nutrition, like parental stress and depression, the reallocation of scarce household resources such as parental time and housing, and effects to the home environment more generally. All these factors, and more, can affect children’s cognitive development and educational outcomes.

Figure 2 helps clarify the empirical approaches and challenges in studying the effects of food assistance programs on child academic outcomes. The intent-to-treat (ITT) approach, which estimates the effect of the availability of a program or policy on educational outcomes (X ➔ Y in Figure 2) and the focus here, sidesteps many of these challenges. It avoids the selection bias caused by the low take-up of these programs, that is, that households who choose to participate may be either negatively (the most disadvantaged) or positively (the best organized, most knowledgeable) selected. Estimating the effects of these programs on the ultimate educational outcome presumes but does not identify the specific mechanisms (e.g., improved nutrition, reduced household stress, etc.). While many studies investigate and help verify the middle links of Figure 2 separately (determinants of take-up, the effects of feeding programs on food insecurity, and the effects of food insecurity on a wide range of child outcomes)—see Bitler and Seifoddini (2019), Fiese et al. (2011), Gundersen and Ziliak (2014, 2015, 2018) for excellent surveys—the focus here is on research that explicitly ties feeding programs and policies to academic outcomes.

An ITT approach helps avoid the selection bias caused by endogenous program participation on the part of families themselves, but several challenges remain. First and fundamentally, finding a weak or null effect could be due to any broken link in Figure 2, which limits the conclusions and policy implications one can draw. Briefly summarizing the available evidence for these intermediate links helps address this limitation. Second, even if the study uses an ITT approach, the potential for selection bias at the program adoption level still exists. For example, early adopters of school breakfast programs or states that enact more generous policies may differ in unobservable ways that could also influence child outcomes. This review, therefore, emphasizes those studies with a credible identification strategy, an empirical approach that plausibly isolates the causal effects of the program from other influences.

The gold standard for identifying causal effects and one that can avoid both types of selection bias is the randomized control trial (RCT), in which the program is randomly assigned to certain individuals and the resulting changed outcomes are compared to those of a control group who did not receive the program. Few RCTs exist in this space, and instead most studies use an alternative source of variation and/or empirical strategy. These strategies include exploiting mandates that are exogenous to the school, staggered program rollout (in which programs are “rolled out” at different times and places) combined with difference-in-difference (DD) or difference-in-difference-in-difference (DDD) estimation, regression discontinuity (in which a program has a sharp cut-off for eligibility that can be exploited) and observable matching (e.g., in which the treated school is “matched” with a suitable “control”).

A third challenge is the cumulative nature of both the food assistance programs and child outcomes. The effects of food assistance programs can begin well before birth, such as the prenatal period when the child’s mother first becomes eligible for WIC. Such an early intervention may have an impact that lasts well beyond the program’s end (Jackson, 2015). Evidence from other settings indicates that the duration of exposure to a program matters in determining the magnitude of effects (Kho & Hunter, 2022; Turner & Chaloupka, 2015). Because educational attainment is itself cumulative, any observed outcome is both an incomplete lifetime measure and potentially the product of many past interventions. Studies frequently attempt to control for the cumulative nature of both past interventions and academic achievement by comparing child outcomes before and after the program change. The richest of these studies have longitudinal data that allows one to perform these comparisons for each individual child. However, such studies may suffer from small sample sizes, attrition bias, and a lack of statistical power, problems common to any longitudinal study. A related challenge is to differentiate the short-term benefits from the cumulative benefits of an intervention. Estimated effects may capture short-term benefits that may soon dissipate, as revealed, for example, by school officials’ attention to school meal contents on testing days (Figlio & Winicki, 2005).

The focus here is on understanding the effect of U.S. food assistance programs on relatively contemporaneous measures of educational outcomes. The interventions considered are only those that provide food or nutritional assistance (e.g., broader antipoverty measures are excluded). Educational outcomes are typically measured by test scores, behavioral or disciplinary issues at school, and attendance. While this limited focus is necessary to keep this review tractable, it may overlook several potentially important strands of literature. For example, recent and plausibly causal findings from other countries linking food assistance to academic outcomes (e.g., Altindag et al., 2020; Lundborg et al., 2022) are excluded. Also, except for a brief discussion of WIC, the long-term or cumulative impact of food assistance programs are not discussed. The effects of antipoverty programs on long-term outcomes is receiving increasing attention (e.g., Aizer et al., 2022; Hoynes et al., 2016). A longer-term focus serves as a cautionary reminder that the outcomes reviewed here may be influenced by previous interventions such as WIC and, at the same time, are only intermediate inputs into educational attainment, learning, and lifetime well-being. Even so, as this review makes clear, there is still much that is unknown in the narrower, more manageable terrain set out here.

### Supplemental Nutritional Assistance Program and Home Food Assistance Programs

SNAP, formerly known as the Food Stamp Program, is the country’s largest public program for combating food insecurity. SNAP provides targeted benefits to needy families in the form of EBT cards that can be used to purchase food and nonalcoholic beverages. In 2020, benefit disbursements to approximately 40 million beneficiaries exceeded $74 billion; about half of all SNAP participants are children (Bitler, 2014; USDA Food and Nutrition Service, n.d.). Several surveys provide excellent reviews of the program’s history, size and scope (e.g., Bitler, 2014; Gundersen & Ziliak, 2015, 2018; Hoynes & Schanzenbach, 2016; Keith-Jennings et al., 2019; Nestle, 2019). Numerous studies link the timing of SNAP benefits to changes in food consumption patterns and document decreased caloric intake near the end of the SNAP disbursement cycle (e.g., Hamrick & Andrews, 2016; Kuhn, 2018; Shapiro, 2005). Stress in families and even increased rates of domestic violence are also linked to the timing of disbursements (e.g., Carr & Packham, 2019). The natural inference from these findings, that SNAP increases food security at least for the early part of the disbursement cycle, is more difficult to determine rigorously, because selection into the program is nonrandom. Among studies that addressed this selection issue (e.g., Gundersen et al., 2017; Mykerezi & Mills, 2010; Zhang & Yen, 2017), more recent work typically indicated that SNAP deceases food insecurity and, for families with children, also alleviates other measures of material hardship that may be linked to educational outcomes (Shaefer & Gutierrez, 2013). Point estimates can be substantial; Katare and Kim (2017) found that benefit cuts that took effect in 2013 resulted in a 3.7 percentage point (7.6%) increase in the number of low food secure households. The available evidence therefore supports that SNAP does indeed reduce food insecurity, a key intermediate link in Figure 2. Given the difficulty in addressing nonrandom selection and the fact that SNAP eligibility requirements do not vary geographically, the literature that links SNAP to learning and academic outcomes generally uses the time elapsed between SNAP disbursement and an observed event as a source of variation. This strategy assumes that the “calorie crunch” or increased household stress at the end of the month generates conditions that can affect academic outcomes. Select studies are summarized in Table 1. This table, and all others in this review, are presented so that a finding of increased treatment (e.g., an event that is closer to the date of SNAP disbursement) resulting in a beneficial effect is indicated by a (+), a detrimental effect with a (−), and a null effect with a (Ø). Null effects include a precisely estimated zero effect or an estimated effect that is statistically insignificant. With no clear standard for distinguishing the two, this discussion makes that distinction only when it is made by the original authors themselves. #### Table 1. Studies of the Effects of SNAP and WIC on Academic Outcomes Authors Test Scores Behavioral Outcomes Data, Empirical Strategy, and Key Findings A. Intervention: SNAP Bond et al. (forthcoming) (+) Uses a sample of SAT test takers from states that determine the day of SNAP disbursements by the recipient's last name. ITT estimates indicate that low-income high school juniors and seniors perform approximately 6 points worse (0.06 SD) on the SAT if the test date is more than 2 weeks after the SNAP disbursement date. Cotti et al. (2018) (+) Using math test scores from South Carolina standardized tests (grades 3–8), the study shows that receiving SNAP benefits more than 26 days before the day of the test lowers test scores by an estimated 0.014 to 0.045 SDs. Effects are largest for African American students. Gassman‐Pines and Bellows (2018) (+) Uses children’s names and birthdates to match North Carolina administrative data on SNAP distributions to educational data, including end of grade testing dates and test scores (grades 3–8). Authors estimate a 0.17 SD (0.12 SD) trough to peak swing in reading (math) scores, which peak 17–19 days after the date of benefit issuance, suggesting a beneficial effect of a sustained period of relative food security. Gennetian et al. (2016) (+) Merges administrative data on SNAP distributions to enrollment and disciplinary records for Chicago Public Schools for students in grades 5–8. Student disciplinary infractions increase at the end of the SNAP cycle. The within‐month difference in disciplinary infractions for students in SNAP recipient families is 7 percentage points larger than for nonrecipients. Differences are particularly pronounced for males. Participation in an after-school snack program or extended experience with the SNAP program may mediate this effect. B. Intervention: WIC Arteaga et al. (2019) (+) Uses longitudinal data on kindergartners eligible for WIC and NSLP to estimate the ITT effects of the gap between when WIC ends at age 5 and the child begins receiving school lunch upon school entry on math and reading scores. An additional month of this gap decreases fall reading scores by 0.021 SD. Effect becomes statistically insignificant by the spring, and no effect for math scores is found. Jackson (2015) (+) Using longitudinal data sets along with coarsened exact matching and maternal fixed effects, the study estimates the effect of prenatal and early childhood WIC participation on short‐ and long‐term outcomes on test scores at age 11. WIC increases reading test scores by 0.19–0.26 SD with no significant effect on math. Note: Summary represents assessment of the predominant findings for academic outcomes and participation in each paper. Symbols represent a beneficial (+) or detrimental (−) effect on the outcome in the heading. The Ø symbol indicates the outcome was examined but no significant effect was identified. Estimated effect magnitude reported for test scores when examined in standard deviations (SD). Abbreviations: SNAP, Supplemental Nutrition Assistance Program; WIC, Special Supplemental Nutrition Program for Women, Infants, and Children; ITT, intent to treat. Cotti et al. (2018) and Gassman-Pines and Bellows (2018) relied on variation between the day that benefits are disbursed and the day that children take standardized tests to identify a connection between SNAP benefits and math and reading test scores for elementary and middle school students. Cotti et al. (2018) estimated that receiving SNAP benefits more than 26 days before the day of the test lowers scores by 0.014 to 0.045 standard deviations (SD). Gassman-Pines and Bellows (2018, p. 912) estimated a 0.17 SD (0.12 SD) peak to trough swing in reading (math) scores that varies with the number of days elapsed since SNAP benefits receipt. Test scores peak 17–19 days after benefit issuance, which is consistent with a sustained period of relative food security at the start of the disbursement cycle. Bond et al. (forthcoming) linked the time elapsed from SNAP disbursement to the SAT testing date, exploiting the variation caused by recent policy changes in seven states and Washington, D.C. These states now determine the day of SNAP disbursement by the recipient’s last name.1 ITT estimates indicate that high school juniors and seniors perform approximately 6 points worse (0.06 SD) on the SAT if the test date is more than 2 weeks after the SNAP disbursement date. Gennetian et al. (2016) studied disciplinary records of Chicago public school students in grades 5–8. The authors compared the difference in disciplinary infractions from the beginning (the time of disbursements) to the end of the month by SNAP beneficiary status. SNAP beneficiaries have a relatively larger increase in infractions at the end of the month. Table 1 shows (+) symbols for both papers because these two studies found a relative benefit to test scores or behavioral outcomes shortly after SNAP disbursement. Studies from nutrition and public health that focus on younger children found similar positive associations between SNAP participation and educational outcomes. These papers, which are not individually listed in Table 1, come with the caveat that estimates that rely on comparing SNAP recipients to non-recipients generally cannot fully control for selection. Frongillo et al. (2006) found that early elementary school children in households that started Food Stamp Program participation showed improvements in reading and mathematics scores relative to children in households that stopped program participation. Hong and Henly (2020) linked SNAP participation to improved test scores in early grades and found that the association is strongest for students from high-poverty households. Beharie et al. (2017) found that SNAP participation moderates the positive association between difficulty in affording basic necessities and the likelihood of repeating a grade. Recent developments have blurred the lines between EBT programs like SNAP and programs that provide food assistance through school. Summer and Pandemic EBT (P-EBT) programs are used to increase SNAP benefits during periods when access to school meals is removed. Collins et al. (2018) reported the results from an RCT during the summers of 2011–2013 when some households were randomly assigned an increase in SNAP benefits per child during the summer months; both food security and dietary quality of the treated children improved. A similar policy intervention occurred in response to the shift to remote learning and reduced hours resulting from the COVID-19 pandemic. The Families First Coronavirus Response Act of 2020 authorized increases in SNAP EBT benefits to those children in affected areas (Families First Coronavirus Response Act, 2020; Toossi, 2021). Preliminary findings by Bauer et al. (2021) suggested that this P-EBT program, which ended in summer 2022, substantially reduced food insufficiency.2 It appears, however, that the effects of these extended EBT programs on academic outcomes has not yet been investigated—the first of several critical gaps identified here. SNAP and associated EBT programs are sizable and fit the scope of this review because households receive benefits contemporaneously to educational outcomes. However, other programs that provide food assistance to households might also affect students. WIC can begin when the child is conceived and ends with the child’s fifth birthday. While most WIC studies therefore focus on early child development and other outcomes, two articles extend that research into outcomes relevant to this review. Jackson (2015), in addition to providing an excellent review of WIC outcomes research, showed that prenatal and early childhood WIC participation is associated with improved learning outcomes at school.3 Arteaga et al. (2019) investigated the effects of the gap between when WIC ends (on the child’s fifth birthday) and when the child becomes eligible for school meals upon entry to kindergarten. The negative effect they found for reading scores lends support to the hypothesis that food assistance improves academic outcomes, and Table 1 shows a (+) symbol to reflect this benefit. This study also highlighted the interactions between programs and the blurry line between at-home and in-school food assistance programs. ### The National School Lunch Program, School Breakfast Program, and Other Public Food Assistance at School Federally funded school-based meal programs started during the Great Depression and were initially targeted at select schools with high levels of malnourishment. The National School Lunch Program (NSLP) was adopted in 1946 with the specific goal of alleviating childhood malnutrition. The program operates in both public and nonprofit private schools, and a school lunch is available to nearly all school-aged children in the United States. Prior to the COVID-19 pandemic, nearly 30 million students would consume a school lunch on any given day (USDA Food and Nutrition Service, 2022). Children from low-income families are eligible for either reduced price (household income less than 185% of the Federal Poverty Level, FPL) or free (household income less than 130% of the FPL). The most fundamental expansion to school meal offerings occurred with the introduction of the School Breakfast Program (SBP). The SBP began as a pilot program in 1966 that focused on nutritionally needy students living in very poor areas or far from school. In 1975, the program was permanently authorized, and in fiscal year 2018, SBP served 14.7 million students at a total federal cost of$4.4 billion (USDA Food and Nutrition Service, 2022). The SBP operates similarly to the NSLP in that schools can offer, and be reimbursed for, free breakfast to students with family income below 130% of the FPL and a reduced-price breakfast to those below 185% FPL. Students can also qualify for both programs based on participation in other programs such as SNAP or Temporary Assistance to Needy Families (TANF) (Bernstein et al., 2004). However, participation in SBP is less prevalent than school lunch—only about two-thirds of schools participate in the program, and, at schools that do participate, fewer students receive breakfast than lunch. Nonetheless, the breakfast program reaches about 15 million children, 85% of whom benefit from free or reduced-price breakfasts (USDA Food and Nutrition Service, 2022).4 The primary factors of the low participation rates are thought to be the stigma associated with participating in a program for low-income students and the difficulty of arriving at school early enough to get breakfast.

In order to make breakfast more accessible, there have been two main changes to how some schools administer the program. First, schools began offering universal free breakfast (UFB) that is not conditional on family income; UFB may mitigate the potentially negative effect of stigma on participation (Schwartz & Rothbart, 2020). Second, schools have explored adjusting the timing or location of breakfast to alleviate the need to arrive at school early, creating programs often called Breakfast After the Bell (BAB) or Breakfast in the Classroom (BIC). These terms can overlap. BAB may be in the classroom, may take the form of “grab and go” breakfast where food is distributed in a high-traffic area, or may be a “second-chance breakfast” where the traditional before-school breakfast is supplemented with an additional opportunity after the bell. The discussion here begins with the (limited) studies that examined the SBP alone and then turns to papers that examined the move to UFB and BAB/BIC programs.

The Community Eligibility Provision (CEP), enacted as part of The Healthy, Hunger-free Kids Act of 2010, substantially increased the availability of free school breakfast and lunch. The CEP, which underwent a staggered implementation starting in the 2011–2012 school year, incentivizes schools with a high level of need to offer free meals to all students (by reimbursing schools for some portion of the cost of meals offered to students for free regardless of household income). For a school to qualify for CEP, at least 40% of students must be from households that are “categorically eligible,” meaning that the household receives some other form of assistance like SNAP or TANF. Because school districts contribute to the cost of providing free meals to additional students, not all schools choose to participate in the CEP. The Families First Coronavirus Response Act (2020) further increased access to school meals. The law increased reimbursement rates, relaxed requirements for applying for subsidized meals, and allowed the USDA to grant waivers that permit states to expand their school meal offerings. Like the expanded summer EBT and P-EBT programs, these provisions expired in the summer of 2022.

Establishing causal links between school meal programs and food insecurity is a difficult empirical challenge due to endogenous selection into the program at both the student and school levels. Some families choose to provide their own food to students rather than participate in a school meal, even if that meal is free. Some schools choose not to offer breakfast or CEP (see Rogus et al., 2018 for a description of CEP school characteristics). Nonetheless, the existing literature indicated that school meal programs have a substantial impact on food insecurity (Arteaga & Heflin, 2014; Gundersen & Ziliak, 2018; Gundersen et al., 2012). Other work compared food insecurity during the school year to summer, when school meals are not available, and found that, during the summer food insecurity increases and quality of nutrition decreases (Bhattacharya & Currie, 2001; Huang et al., 2015; Nord & Romig, 2006).5 Thus, evidence suggests that the intermediate link in Figure 2 (feeding programs reduce food insecurity) holds here as well.

#### National School Lunch Program (NSLP) and Academic Outcomes

Table 2 summarizes the literature linking the NSLP to academic outcomes. Because the NSLP has been in place for many decades and the eligibility criteria for free or reduced-price lunch (FRL) does not vary across states, it is difficult to devise an empirical strategy that estimates the impact of exposure to the NSLP or FRL offerings on academic outcomes. Hinrichs (2010) used an instrumental variable strategy relying on a change in the funding formula that allocated money to states to identify differences in birth cohorts by state. The author found a strong positive effect of NSLP exposure and educational attainment, as measured by years of schooling. On the other hand, Dunifon and Kowaleski‐Jones (2003), which compared participating and non-participating siblings, found no statistically significant effect on either evaluations of math or reading achievement or behavioral outcomes, which they suggested is due to a lack of precision.

#### Table 2. Studies of the Effects of the National School Lunch Program on Academic Outcomes

Authors

Educational Attainment

Test Scores

Behavioral Outcomes

Data, Empirical Strategy, and Key Findings

A. Intervention: Exposure to NSLP

Dunifon and Kowaleski‐Jones (2003)

Ø

Ø

Uses the 1997 Child Development Supplement of the Panel Study of Income Dynamics to identify sibling pairs (ages 6–12) where siblings differed in FRL participation. No statistically significant differences in measured outcomes for sibling pairs.

Hinrichs (2010)

(+)

Uses IV to exploit a change in the formula to allocate federal cash assistance across states using 1980 census data about individuals born between 1941 and 1956. Findings reveal a statistically significant positive relationship between NSLP exposure and educational attainment (years of schooling). A 10 percentage point increase in NSLP increases education attainment by 0.365 years in women and nearly a year in men.

B. Intervention: Change to lunch content

Anderson et. al. (2018)

(+)

Uses a DD approach to study the impact of contracting (or canceling contracts) with vendors that provide healthier school meals and school‐by‐grade-level data from California public elementary, middle, and high schools. A healthy food vendor is found to increase test scores by 0.03–0.04 SDs relative to in‐school meal provision.

Figlio and Winicki (2005)

(+)

Uses school menus to calculate the caloric content of lunches on testing and non‐testing days with data on fifth-grade test scores from middle schools in Virginia. Schools threatened with accountability sanctions increase the caloric content on testing dates. A regression of fifth-grade pass rates on previous pass rates and calorie differentials during the testing period provides evidence that this intervention increases test scores.

Note: Summary represents assessment of the predominant findings for academic outcomes and participation in each paper. Symbols represent a beneficial (+) or detrimental (−) effect on the outcome in the heading. The Ø symbol indicates the outcome was examined but no significant effect was identified. Estimated effect magnitude reported for test scores when examined in standard deviations (SD).

Abbreviations: NSLP, National School Lunch Program; FRL, free/reduced‐price lunch; IV, instrumental variables; DD, difference in differences.

Instead of attempting to estimate the full impact of program exposure on academic outcomes, two papers have instead focused on the content of school meals. Figlio and Winicki (2005) calculated the caloric content of school lunches. They showed that schools threatened with accountability sanctions increase school lunch calories on testing days and that this strategy is associated with increased test scores for fifth graders. Similarly, Anderson et al. (2018) studied the school-level decision to use vendors that supply healthier school meals, both breakfast and lunch, and found a 0.03–0.04 SD increase in school by grade level standardized test scores.

#### The School Breakfast Program and Academic Outcomes

As with the NSLP, only a few studies attempt to estimate the impact of exposure to the SBP program, with more research focused on the timing, location, or availability of breakfast. Among the studies identifying exposure to the program overall, Meyers et al. (1989) and Bartfeld et al. (2019) studied the initiation of the SBP at, respectively, Massachusetts and Wisconsin schools. Both studies found positive effects on both test scores and absences. Frisvold (2015) confirmed that similar results—the program increased test scores by 0.05–0.12 SDs—hold nationally. Frisvold (2015) accounted for selection by using a regression discontinuity design based on state-level mandates requiring economically disadvantaged schools to offer a SBP. The top panel of Table 3 summarizes these papers.6

#### Table 3. Studies of the Effects of Various School Breakfast Programs on Academic Outcomes

Author

Policy

Participation

Attendance

Test Scores/GPA

Behav. Outcomes

Data, Empirical Strategy, and Key Findings

A. Intervention: Adoption of SBP only

Frisvold (2015)

SBP

(+)

Uses RD econometric strategy on national student‐level data to study effect of mandate to implement/expand SBP. Reading (math) scores increase by 0.05–0.12 (0.08–0.09) SD.

Meyers et al. (1989)

SBP

(+)

(+)

Examines implementation of SBP at six elementary schools in Lawrence, Massachusetts. Study focuses on children eligible for free meals. Study finds beneficial effects on test scores and attendance.

B. Intervention: Change in cost to students (from SBP to UFB)

Leos‐Urbel et al. (2013)

UFB

(+)

Ø

Ø

Student‐level data from NYC public schools during a time period when the school system expanded UFB to all schools. Finds no effect of UFB on test scores or attendance.

Murphy et al. (1998)

UFB

(+)

(+)

Uses comparison of means on student‐level data and interviews with parents from three schools in Pennsylvania and Maryland. Findings suggest beneficial effect on test scores.

Norwood (2020)

UFB

(+)

(+)

Uses DD and fuzzy RD with school‐level data from Texas. Findings suggest beneficial effects on behavioral outcomes and test scores.

Ribar and Haldeman (2013)

UFB

(+)

(+)

Ø

Uses data from Guilford County, North Carolina, during a time period when the revised eligibility standards caused some schools to stop offering UFB. Finds no effect of UFB on test scores. UFB associated with improvements in Participation and small improvements in attendance.

C. Intervention: Change in location or timing (from SBP to BAB or BIC)

Anzman‐Frasca et al. (2015)

BIC

(+)

(+)

Ø

Uses 2012–2013 data from elementary schools in an urban school district; 257 schools (57.6%) implemented a BIC program. Another 189 schools serve as controls. Uses propensity score weights to match treatment and control schools and then estimates the average treatment effect among the treated. Finds substantial improvements in participation and attendance. No effect on test scores.

BAB

(+)

Uses DD and data from Arkansas schools for fifth to seventh grade. Treatment group is students who first receive BAB in fifth grade and continue to get it through seventh. Finds a decrease in disciplinary infractions.

BAB

(+)

Ø

(+)

Ø

(−)

Uses DD and data from Arkansas on schools up to third grade. Effect on attendance and test scores largely insignificant, but slightly positive for test scores for some economically disadvantaged students and slightly negative effect for some non‐disadvantaged students.

Dotter (2013)

BIC

(+)

Ø

(+)

(+)

Uses longitudinal student data from San Diego elementary schools (2001–2011). Exploits staggered implementation of BIC in eligible schools (>70% FRL). Study finds that reading and math scores increase by 0.10–0.15 SD, classroom behavior improves, but there are no significant effects for attendance.

Hearst et al. (2019)

BAB

(+)

Ø

Uses data from 16 rural high schools in Minnesota that were randomly assigned into either a treatment or delayed treatment group for Project BREAKfast, which offered grab‐and‐go and second‐chance breakfast. Study finds no statistically significant effect on GPA.

Kirksey and GottFried (2021)

BAB

(+)

(+)

Ø

Uses data from Nevada and Colorado schools (2014–2015). These states mandated implementation of BAB if 70% or more of students qualified for FRL. RD strategy compares schools just above and just below the 70% cutoff. No statistically significant effect on test scores, but substantial improvements to attendance.

D. Intervention: Multiple Interventions

UFB w/ BAB

Ø

Uses a DD empirical strategy with student‐level data from the ECLS‐K 2011 cohort matched to state and district mandates to adopt BAB and UFB. Effect on test scores is generally positive but imprecisely estimated.

Bartfeld et al. (2019)

SBP & UFB

(+)

(+)

Uses data from 1,000 Wisconsin elementary schools from 2009–2014. Study observes different changes including adoption of SBP, shifting to a BIC/BAB format or implementing UFB. Implementation of SBP and UFB both associated with improved attendance and scores. BAB/BIC are associated with no beneficial effect on attendance and a slight decrease in math scores.

BIC or BAB

Ø

(−)

Bernstein et al. (2004)

UFB or UFB w/ BIC

(+)

Ø

Ø

(−)

Studies the impacts of the School Breakfast Program Pilot Project (SBPP) whereby randomly selected schools received federal funding to implement UFB or UFB with BIC (see also Schanzenbach & Zaki, 2014 for re‐analysis). No significant effect on test scores or attendance. Substantial increases in participation (larger for schools that chose the BIC option). UFB associated with greater behavioral incidents (especially for BIC implementations).

Corcoran et al. (2016)

UFB w/ BIC

(+)

(+)

Ø

Studies the staggered implementation of the NYC public schools’ (elementary and middle) adoption of BIC. Study finds insignificant effects on math and reading test scores, modest improvements to attendance, and large increases in participation.

Imberman and Kugler (2014)

UFB w/ BIC

Ø

(+)

Uses student‐level data from a large urban school district in the Southwest with a staggered implementation of a BIC program. Reading (math) scores increased by 0.06 (0.09) SD—most pronounced for low‐performing students; no effect on attendance.

Schanzenbach and Zaki (2014)

UFB or UFB w/ BIC

(+)

Ø

Ø

Ø

(+)

Builds on Bernstein et al., 2004) in studying the SBPP, where the opportunity for treatment is randomly assigned; school can decide between UFB or UFB with BIC. No significant effect on test scores, attendance, or behavioral incidents (BIC may improve behavioral outcomes for some economically disadvantaged students). Substantial increases in participation.

Note: Summary represents assessment of the predominant findings for academic outcomes and participation in each paper. Symbols represent a beneficial (+) or detrimental (−) effect on the outcome in the heading. The Ø symbol indicates the outcome was examined but no significant effect was identified. Estimated effect magnitude reported for test scores when examined in standard deviations (SD).

Abbreviations: SBP, School Breakfast Program; BAB, breakfast after the bell; BIC, breakfast in the classroom; UFB, universal free breakfast; RD, regression discontinuity; DD, difference in differences; FRL, free/reduced‐price lunch.

The remaining panels of Table 3 review studies that estimated the academic effects of changes to the price, location, or timing of breakfast. Changes to the price of breakfast typically come from a switch to or, in the case of Ribar and Haldeman (2013), away from Universal Free Breakfast (UFB), where all students at a school are offered free breakfast regardless of household income. Changes in the location include offering breakfast in the classroom or at a “grab and go” station in the hallway. Changes in timing may mean that breakfast is offered after the school day has begun (“after the bell”) or offered both before and after the bell in the form of a “second chance” breakfast. Reviewing the papers in panels B–D of Table 3 poses the challenge that the characteristics of programs differ and, furthermore, that many reforms to location or timing are accompanied by a simultaneous change to UFB. It is possible that such a simultaneous change may have a unique and distinct effect from either policy alone. Nonetheless, some patterns emerge.

Making breakfast more accessible in terms of cost, location, or timing is likely to increase participation substantially (Anzman-Frasca et al., 2015; Bernstein et al., 2004; Ribar & Haldeman, 2013; Schanzenbach & Zaki, 2014). Likewise, there is emerging consistent evidence of beneficial effects of expanded breakfast availability on behavioral or disciplinary measures (Cuadros-Meñaca et al., forthcoming; Dotter, 2013; Norwood, 2020; Schanzenbach & Zaki, 2014). The evidence on attendance, grades, and test scores is more nuanced. Expanded breakfast offerings appear to have a modest beneficial effect on attendance, but the precision is not sufficient for the effect to rise to the level of statistical significance in all studies (Anzman-Frasca et al., 2015; Bartfeld et al., 2019; Bernstein et al., 2004; Corcoran et al., 2016; Cuadros-Meñaca et al., 2022; Dotter, 2013; Imberman & Kugler, 2014; Kirksey & Gottfried, 2021; Leos-Urbel et al., 2013; Ribar & Haldeman, 2013; Schanzenbach & Zaki, 2014).

Cuadros-Meñaca et al. (2022) pointed out that the effect of changing breakfast on grades or test scores could be either positive or negative. Improved nutrition is likely to improve learning (as in Figure 2) but programmatic changes like BAB or BIC may also involve shifting some instructional time, and perhaps other school resources, toward the provision of breakfast. Consistent with this insight, the literature finds mixed results. The majority of studies reviewed in Table 3 found either a positive, but insignificant, effect (Abouk & Adams, 2022; Anzman-Frasca et al., 2015; Bernstein et al., 2004; Corcoran et al., 2016; Kirksey & Gottfried, 2021; Leos-Urbel et al., 2013; Ribar & Haldeman, 2013; Schanzenbach & Zaki, 2014) or statistically significant improvements to test scores (Bartfeld et al., 2019; Dotter, 2013; Murphy et al., 1998; Norwood, 2020). That conclusion is not universal. Hearst et al. (2019) found no effect of expanding breakfast options on grade point average, and Bartfeld et al. (2019) found a small negative effect from BIC on math test scores of lower income boys. Cuadros-Meñaca et al. (2022) used rich student data and a DD empirical strategy to isolate a negative effect on test scores for relatively affluent students, who most likely do not suffer from food insecurity. Even accounting for these small negative effects, the totality of the literature supports the assessment by Cuadros-Meñaca et al. (2022) that changes to breakfast will not adversely impact academic achievement in any meaningful way and may lead to improvements for some subsets of students.

#### Universal Free Meals, the Community Eligibility Provision, and Academic Outcomes

The USDA has long enabled high-poverty schools to offer universal free meals under Provisions 1–3 of the National School Lunch Act, but this option was relatively rarely implemented (Cohen et al., 2021; Schwartz & Rothbart, 2020). The Community Eligibility Provision (CEP), which was first piloted in 2011 with national adoption occurring during the 2014–2015 school year, incentivized many more schools to offer UFM. The staggered rollout of the program, combined with the fact that only certain schools are eligible for federal assistance, makes some version of a DD ITT approach the most typical econometric strategy for studying the academic impacts of this policy. For prior reviews of the literature linking UFM or CEP to academic outcomes see Cohen et al. (2021) and Hecht et al. (2020).

#### Table 4. Studies of the Effects of the Community Eligibility Provision and Universal Free Meals Programs on Academic Outcomes

Authors

Participation

Attendance

Test Scores

Behav. Outcomes

Data, Empirical Strategy, and Key Findings

Bartfeld et al. (2020)

(+)

Uses a DD empirical strategy with student-level attendance data from Wisconsin elementary schools. CEP did not affect attendance in the initial year but resulted in a 3.5-percentage-point decrease in students with low attendance in the year after implementation.

Davis et al. (2020)

Ø

Uses a DD empirical strategy with student‐level data from an Atlanta school district. CEP showed no statistically significant effect on attendance.

Gordanier et al. (2020)

(+)

(+)

Uses a DD empirical strategy and student‐level data to study the adoption of CEP in South Carolina. Study finds a 0.06 SD increase in math scores along with improved attendance, for elementary school students. Effects for test scores are insignificant for middle school students.

Gordon and Ruffini (2021)

(+)

Uses school‐level data from the Civil Rights Data Collection (CRDC) to examine suspensions during the national CEP rollout. Findings show an insignificant effect for high school students but statistically significant reductions in elementary and middle schools. Benefits are concentrated on areas with high levels of food insecurity.

Hecht (2020)

(+)

(+)

Ø

(+)

(+)

Uses a comparative interrupted time series (DD) design with school‐level K–12 data from Maryland to compare schools that adopt CEP to eligible/near‐eligible that do not adopt. Finds an increase in lunch participation for elementary, middle, and high schools and increased breakfast participation for elementary and middle schools. Elementary attendance and middle school disciplinary referrals improve. Middle school science proficiency improves; math and reading change not significant for any grades.

Kho and Hunter (2022)

Ø

(+)

Uses a DD empirical strategy and student‐level data from Tennessee (grades 3–12) during the implementation of CEP. No significant effects on attendance; no contemporaneous effects on on‐time promotion, but magnitudes grow with program duration; disciplinary incidents decrease by 1 percentage point.

Ruffini (2021)

(+)

(+)

Uses a DD empirical strategy on a staggered rollout of CEP and data from all states (all grades) with district- grade‐ year-level data. Lunch (breakfast) participation increases by 12% (38%). In districts with relatively low baseline FRL eligibility rates, CEP improved math scores by 0.02 SD; Improvements are driven by elementary and Hispanic students.

Schwartz and Rothbart (2020)

(+)

(+)

Uses a DD and IV empirical strategy and a sample of New York City students in grades 3–8, including transaction-level data on meal participation during implementation of UFM. UFM increases math (language arts) scores by 0.08 (0.06) SD for non‐poor students with smaller benefits for poor students. Lunch participation increases by 11.0 (5.4) percentage points for non‐poor (poor).

Note: Summary represents assessment of the predominant findings for academic outcomes and participation in each paper. Symbols represent a beneficial (+) or detrimental (−) effect on the outcome in the heading. The Ø symbol indicates the outcome was examined but no significant effect was identified. Estimated effect magnitude reported for test scores when examined in standard deviations (SD). Kho (2018) contained additional outcome measures and specifications not included in Kho and Hunter (2022).

Abbreviations: CEP, community eligibility provision; UFM, universal free meals; DD, difference in differences; IV, instrumental variables.

Table 4 summarizes recent research on the academic effects of adopting UFM in U.S. schools. The move to UFM results in increased participation in both lunch and breakfast programs (Hecht et al., 2020; Pokorney et al., 2019; Ruffini, 2021; Schwartz & Rothbart, 2020). Effects on test scores are generally, but not universally, positive. Math (Gordanier et al., 2020; Ruffini, 2021; Schwartz & Rothbart, 2020) and science test scores (Hecht, 2020) show the most consistent positive effects; only one study in Table 4 finds a statistically significant improvement in reading/language arts scores (Schwartz & Rothbart, 2020). Furthermore, both economically disadvantaged students, who were already eligible for free meals, and non-disadvantaged students show improvements in test scores (Gordanier et al., 2020; Schwartz & Rothbart, 2020). Most studies have found that beneficial effects of UFM are found in, or driven by, younger (elementary) students. Ruffini (2021) and Kho (2018) both found that the effects on test scores are stronger in schools where CEP had a stronger “dosage” by adding more students into a free meals program.

On absences, results are mixed. On the one hand, Bartfeld et al. (2020), Hecht (2020), and Gordanier et al. (2020) found that implementation of UFM decreases absences, specifically for elementary and economically disadvantaged students. On the other hand, Davis et al. (2020) and Kho and Hunter (2022) found that the CEP has statistically insignificant effects of a small magnitude on absences.

On other outcome measures, Gordon and Ruffini (2021) found that a move to UFM through the CEP decreases suspensions by as much as 17% for elementary students, coinciding with Norwood (2020), which uncovered similar findings for universal free breakfast (Table 3). Kho and Hunter (2022) found increases in on-time promotion to the next grade but only after two years of the program. The effect not only persists, but grows, in the third year. This lag could explain the null findings for some of the other studies that are unable to observe a long post-period or separately identify effects by implementation year. Kho and Hunter (2022) noted that this has important implications for policy where many schools/districts may be tempted to cut a program that doesn’t create immediate results.

### Programs In-Between—Emerging Research and Future Directions

Thus far, this review has focused on government-provided food assistance programs at home (SNAP and, to a lesser extent, WIC) and in school (NSLP, SBP, and the CEP), which constitute the top and bottom cells of the first column of Figure 1. This section discusses the remaining programs listed in Figure 1, beginning with the government-provided programs that fall into the area between home and school (column 1). These include summer feeding programs and feeding programs that occur in Head Start programs, childcare centers, and after-school programs (through the Child and Adult Care Food Program, CACFP). The section also discusses programs, such as weekend feeding programs and food pantries, that are initiated and provided by nongovernment entities, such as charitable and religious organizations. While some studies have established the effectiveness of these government and nongovernment programs in reducing food insecurity, very little is known about the effect of the programs on academic outcomes.

#### Summer Food Assistance Programs

Turner and Calvert (2019) and Fleischhacker et al. (2020) provided excellent reviews of summer food assistance programs: their history, how they operate, the meals they provide, the challenges they face, and their effects on food insecurity and other outcomes. Vericker et al. (2021) reported on the USDA Summer Meals Study (in 2018) and provided the most up-to-date report on and detailed information for these programs.

The Summer Food Service Program (SFSP) began in 1968 and became permanent and nationwide in 1975. In 2001, the USDA introduced the Seamless Summer Option (SSO), a program designed to “streamline the transition from school meals to summer meals, with the goal of increasing the number of districts that provide meals in the summer” (Turner & Calvert, 2019, p. 972). While the SFSP can be operated by a variety of sponsors, including government or community agencies, schools, summer camps, and charitable/religious organizations, the SSO is available only to school food authorities. Schools can therefore choose to participate in either SSO or SFSP to provide summer meals. SSO is set up as an extension of the school-year meals program, following the same requirements and reimbursement schedule and therefore requiring less paperwork. In contrast, the SFSP has higher reimbursement rates and more flexibility in meal provision but is a separate program.7 If considered individually, SSO is squarely in column 1 of Figure 1 whereas SFSP spans both columns. Despite this difference, most research, including this review, considers the two programs collectively, and the programs are depicted jointly in Figure 1. The other significant development in summer feeding programs is the 2011 pilot Summer EBT program, which was the basis for the ongoing Pandemic EBT program.

Despite its longevity and efforts to increase participation, summer feeding programs continue to reach only a small fraction of eligible children. Any child eligible for a free or reduced-price school lunch is eligible for these summer feeding programs. However, while approximately 22 million children receive a free or reduced-price lunch at school each day, only 3 million lunches were served on average daily in July 2017 (Turner & Calvert, 2019). This number represents an uptake rate of only 15% nationwide among students participating in the free and reduced-price lunch program, which itself has less than 100% take-up (Turner & Calvert, 2019). Fleischhacker et al. (2020) reported similar figures for 2018. Both reviews noted that the availability of summer programs varies widely across the country and tends to be lower in rural counties than urban ones (Fleischhacker et al., 2020).

Little is known about the effects of summer feeding programs on child outcomes, although limited evidence suggests that it is associated with reduced food insecurity. Neither review lists any research on cognitive or academic outcomes; Turner and Calvert (2019, p. 980) stated:

Thus far, no work has examined the relationships between food insecurity and dietary changes in the summer, and subsequent academic, behavioral, and cognitive outcomes.

A search turned up one additional case study of a summer nutrition program in Maryland (Orovecz et al., 2015). Linear regression models that control for the proportion of FRL students indicated that schools with a nutrition program have a higher percentage of students achieving proficiency in reading and math. In addition, controlling for FRL percentage, participating schools have higher graduation rates. While these results should not be interpreted as causal, they emphasize that investigating the effects of summer feeding programs is a promising future direction.

There are several reasons to expect that summer programs might affect academic outcomes. The first is that the decline in academic performance over the summer, “the Summer Slide” or “Summer Achievement Gap,” is well established. This decline disproportionately affects low-income families, precisely the group targeted by such programs; the cumulative impact of this gap over multiple school years explains a substantial portion of the diminished academic achievement for low-income students (Alexander et al., 2007; Turner & Calvert, 2019). Evaluating the effects of summer feeding programs on subsequent academic outcomes also avoids the short-term—and thus perhaps less meaningful—boost in performance that comes from recent meal consumption. As such, its effects could provide new insight into the effects of feeding programs on outcomes. Finally, understanding the benefits of summer programs are likely critical to the political future of the summer and Pandemic EBT programs and may help determine the need for expanding the reach of summer meals programs (e.g., Litt et al., 2020). Vericker et al. (2021) reported that 80% of children in low-income households live within 1 mile (10 miles) of an urban (rural) summer meals site, satisfaction among participant caregivers is high (90%), and lack of awareness is the most common reason given for a lack of participation, all suggesting ample opportunity for expansion. Again, while improving academic outcomes is not the goal of these programs, finding they have lasting academic effects would be an important part of the calculus.

#### The Child and Adult Care Food Program

The Child and Adult Care Food Program (CACFP) is also a little studied program (Fiese et al., 2011; Gordon et al., 2011; Heflin et al., 2015). The USDA administers and funds the CACFP via grants to states, and the programs are then administered in a decentralized way by state agencies like educational agencies or health/social services departments (Fiese et al., 2011). CACFP through these state agencies then reimburses participating childcare centers, homeless programs, and after-school programs for the meals and snacks provided, which places this program closer to the second column in Figure 1.8 Children are eligible for the program if their household has an income below 185% of the FPL (Heflin et al., 2015); age eligibility depends on the setting (e.g., is under 12 in childcare facilities versus 18 or under in emergency shelters; Fiese et al., 2011). CACFP served approximately 4.2 million children daily in 2015 (Ralston et al., 2017), so in terms of numbers of children served is similar to although slightly larger than the summer feeding programs.

Gordon et al. (2010) investigated the characteristics of CACFP participants as well as the association between participating and a child’s dietary intake and nutritional outcomes. They found “that program eligibility rules leave many poor children outside the CACFP program” (Gordon et al., 2010, abstract) and that program participation is associated with positive child outcomes. Heflin et al. (2015) similarly found that accessing childcare through CACFP-participating providers “results in a small reduction in the risk of food insecurity” (abstract). Both Gordon et al. (2010) and Heflin et al. (2015) summarized the scant research on CACFP. A search that included searching papers that cited these works turned up no studies that investigated the developmental, behavioral, or educational outcomes of CACFP. Given the vulnerable populations served, the possibility of a strong effect on a broader set of child outcomes is promising and worthy of research.

#### Weekend Feeding Programs and Other Private Food Assistance Programs

Food banks and other private organizations also provide food assistance programs specifically targeted toward households with children. These programs sit in the second column of Figure 1 and often blur the lines between at home or in school because charitable organizations often partner with schools to reach students. In general, very little research has been published about either the reach or effects of these programs, except for a recent literature on weekend feeding programs.

Weekend feeding programs provide food assistance, in the form of a food pack that is distributed on Fridays, at the end of the school day, to food-insecure children. The first such program was started in 1995, at an Arkansas elementary school where a school nurse acted after observing that children were noticeably hungry when they returned to school on Mondays. That single program rapidly expanded across the state. In 2006, Feeding America used this model to develop a nationwide program, the “BackPack” program, which it rolled out through its member food banks to now reach more than 500,000 students. Independent organizations, not affiliated with Feeding America, also sponsor weekend feeding programs, meaning that the total number of children receiving supplemental food packs likely exceeds 800,000 (Fram & Frongillo, 2018).

The disaggregated nature of weekend feeding programs means that they are challenging to study and require merging information about students and schools with program data. The fact that programs are initiated and administered by community groups suggests that selection may occur at several levels, affecting both a school’s ability to initiate a program and which students might be invited to participate. For this reason, much of the early literature on weekend feeding programs was qualitative or case study based. Given that caveat, survey and structured interview responses indicate that the program increases food security (Ecker & Sifers, 2013; Shanks & Harden, 2016; Wright & Epps, 2016) and similarly subjective measures indicate improvements to academic performance, grades, and attendance (Berry et al., 2018; Laquatra et al., 2019).

Four studies, reported in Table 5, evaluated the relationship between weekend feeding program participation and academic outcomes more rigorously. Rodgers and Milewska (2007) used school-level data to show that program participation is associated with a higher proportion of eighth graders, but notably not elementary students, achieving proficient scores in math and literacy. Because it lacks individual data, the authors could not identify differential effects on the targeted population (i.e., food insecure children). Mangrum (2019) also had school-level data only but used younger children as a control group in a DDD analysis. Preliminary findings showed statistically significant improvements to language arts and math standardized tests and attendance at the beginning and end of the week.

#### Table 5. Studies of the Effects of Weekend Feeding (“Backpack”) Programs on Academic Outcomes

Authors

Attendance

Test Scores

Data, Empirical Strategy, and Key Findings

Fiese et al. (2020)

(+)

Use a DD empirical strategy and student‐level data from Illinois to compare program participants to a control group of similar children on the program's wait list. Study finds that participants have relatively higher attendance on Fridays, the day that food is distributed.

Kurtz et al. (2020)

Ø

(+)

Uses a DDD strategy (before and after adoption, economically disadvantaged vs. advantaged students), merging student‐level data from North Carolina elementary schools with food bank participation data. The weekend feeding program did not affect attendance but was associated with a 0.09 SD improvement to reading scores, and a weaker but similar effect on math scores. Strongest effects found for the youngest and academically weakest students.

Mangrum (2019)

(+)

(+)

Uses DD and DDD empirical strategies with school‐level data from low-income elementary schools in the Mississippi Delta. Finds statistically significant increases to standardized test performance in language arts and math at two treated schools. Attendance improves on Fridays (day of food distribution) and also on Mondays and Tuesdays.

Rodgers and Milewska (2007)

Ø

(+)

Uses a school fixed effects regression and school‐level data for fourth, sixth, and eighth grade test scores in Arkansas schools. Findings indicate a positive and statistically significant association between the program and eighth‐grade standardized test performance in math and literacy; there is no statistically significant effect for fourth and sixth graders.

Note: Summary represents assessment of the predominant findings for academic outcomes and participation in each paper. Symbols represent a beneficial (+) or detrimental (−) effect on the outcome in the heading. The Ø symbol indicates the outcome was examined but no significant effect was identified. Estimated effect magnitude reported for test scores when examined in standard deviations (SD).

Abbreviations: DD, difference in differences; DDD, difference in difference in differences

Fiese et al. (2020) had rich student-level data and compared the attendance of students receiving weekend food packs to a control group of food insecure (i.e., program eligible) students at the same schools. Absenteeism is relatively lower for the treated students on Fridays, the day that the food packs are distributed, than on Monday–Thursday. However, because the data come from a single school year, they preclude a before-and-after analysis that could identify whether Friday absenteeism is reduced for these students relative to the pre-program period.

Kurtz et al. (2020) addressed these limitations by using elementary school student-level data and a DDD empirical analysis to identify the impact of the program on test scores and attendance for potentially treated students (i.e., economically disadvantaged students at participating schools) relative to untreated students at the same or other schools. This study found that a weekend feeding program improves reading test scores by 0.09 SD with a similar but weaker effect on math scores. Effects are strongest for the youngest students and those with the weakest past academic performance. The estimated effect on attendance is generally centered on zero and statistically insignificant, but the data preclude studying attendance by day of the week (as in Fiese et al., 2020 or Mangrum, 2019).

While the studies on weekend feeding programs are a promising line of recent research and the emerging empirical literature indicates important beneficial effects, other aspects of weekend feeding programs and charitable interventions more generally are unstudied. For example, little is known about the factors that drive patterns of program adoption across schools; that is, unlike government-provided programs, little is known about which children may or may not have access to these programs. Preliminary results from Kurtz et al. (2022) indicate that many factors, beyond the level of need at a school, influence the program adoption decision. Other research points out that weekend feeding programs, which put some of the responsibility of transporting food on students themselves, might exacerbate the stress, anxiety, and stigma associated with food insecurity (Fram & Frongillo, 2018), a potential detrimental effect that merits further exploration. Perhaps in response to these concerns, food banks have devised new interventions, including on-site school pantries, mobile food pantries that regularly visit school grounds, and Kids Café programs for after-school meals. None of these programs have been studied with empirical strategies that can identify causal impacts to scholastic outcomes.

### Concluding Remarks

The primary goal of a food assistance program is to mitigate the detrimental impacts of food insecurity. For the major programs (e.g., SNAP, WIC, NSLP, SBP), there is convincing evidence that each program helps to achieve this goal. Modest investments, as little as an additional $30 per month per child, can have a significant impact on food security and diet quality (Collins et al., 2018). Even marginal food insecurity is linked to significant detrimental outcomes like anemia, asthma, birth defects, obesity, tooth decay, poor diet quality, lower levels of physical activity, and worse categorizations in a generalized measure of overall health (Frongillo et al., 2019; Gundersen & Ziliak, 2015; Shankar et al., 2017). It is reasonable to expect that food assistance programs might therefore also yield academic benefits. That said, it is also possible that diverting school financial resources or instructional time to food assistance (e.g., BIC), or inadvertently making children feel more responsible for managing household food resources (e.g., weekend feeding programs) is detrimental to learning (Cuadros-Meñaca et al., 2022; Fram & Frongillo, 2018). For all the interventions studied, an important objective is participation—that is, to reach more food-insecure children. Perhaps not surprisingly, participation is the outcome with the largest and most consistent impacts. Changes to the timing or location of breakfast and the offering of universal free breakfast and/or lunch produce sizable increases in participation across multiple studies. For other outcomes, the literature generally finds beneficial effects on test scores, attendance, and measures of disciplinary acts, but magnitudes are often modest and don’t always rise to the level of statistical significance. For example, many of the studies in Tables 15 found that various programmatic changes to food assistance programs cause test scores to improve by around 0.1 SD or less. While such an impact is small in absolute terms, the benefit is still economically significant. First, recall that studies typically produce ITT estimates for cases where only a small portion of the eligible population participates; the effect on the treated is likely much larger. Furthermore, Kurtz et al. (2021) showed that even small improvements to scores (0.07–0.09 SD) can account for a substantial proportion of the performance gap between economically disadvantaged and non-disadvantaged children. Figlio et al. (2018) argued that the present value of the benefits of a similarly sized (0.05 SD) improvement to reading test scores might reasonably exceed$3,000 per student. Food assistance that produces a small improvement to test scores is likely to easily pass a cost-benefit test, even before considering the program’s primary goal—to alleviate the effects of food insecurity.

Given the potential for economically significant benefits of food assistance programs, there are numerous opportunities for new research. First, the myriad of policy changes in response to the COVID-19 pandemic, including expanded access to the SSO and UFM as well as the summer- and Pandemic-EBT programs, are still unstudied. Second, a clear pattern in this review is that programmatic details—the dietary content of meals and the mechanism by which meals are distributed—can have substantial effects. Existing results on meal quality (e.g., Anderson et al., 2018) suggest that this topic merits further study. Third, the interactions between programs must be better understood. For example, the SNAP cycle is associated with changes in school lunch consumption and access to school meals and is, in turn, associated with food bank usage, suggesting that academic outcomes can be affected by the interaction of two public and one charitable form of food assistance (Laurito & Schwartz, 2019; Marcus & Yewell, 2021). The interaction between different programmatic changes is evident in the literature on school breakfast (Table 3), where the estimated effects of offering universal free breakfast are sensitive to whether schools simultaneously change the location of breakfast. Fourth, the necessary duration and cumulative nature of food assistance are not fully understood. Clearly food assistance can have the immediate impact of alleviating hunger, which may result in increased test scores that day (Figlio & Winicki, 2005). On the other hand, research is finding that food assistance has a cumulative effect that may not be evident for a period of months or years (Kho & Hunter, 2022; Turner & Chaloupka, 2015) but result in benefits that persist long after the food assistance ends (Jackson, 2015).

Finally, there are many unstudied programs. For example, the academic impacts of the USDA’s two summer food programs and CACFP are unknown. Besides the blurring of the lines between food assistance at home and school, which has been exacerbated by the pandemic, there is also the expanded range of programs offered by charitable organizations, the right side of Figure 1. In addition to studying their academic impacts, investigating which students get access to these programs and how these programs interact with publicly provided ones are areas worthy of future study.

### Acknowledgments

We thank Laura Earle, Charles Hunt, and Catherine Taylor for excellent and dedicated research assistance and thank Heather Oliver for creating Figure 1. We are grateful to James Ziliak for advice and insight.

### Notes

• 1. This type of policy mitigates the surge in grocery demand at some food markets (Fone, 2022) and allows recipients to increase expenditure smoothing (Cotti et al., 2021).

• 2. Food insufficiency indicates that survey respondents “sometimes or often did not have enough to eat in the last week.”

• 3. Jackson (2015) addressed the selection bias of WIC participation through “coarsened exact matching” and by including maternal fixed effects.

• 4. Figures are from 2019. There was a steep drop-off in school meal participation in subsequent years, presumably due to school closures and remote learning.

• 5. Comparing food insecurity during the summer to food insecurity during the school year is complicated by the fact that some students receive food over the summer from the USDA’s summer food programs, which we discuss in the next section. These programs are much smaller than the NSLP or SBP (see Figure 1).

• 6. Bartfeld et al. (2019), which also analyzed several other programmatic changes, is in the bottom panel of Table 3.

• 7. The increased funding of school meals during the COVID-19 pandemic reduced this disparity but like the other provisions is set to expire in the summer of 2022.

• 8. The program also serves eligible adults enrolled in adult day-care centers, but it is not an entitlement program, and it is small (Fiese et al., 2011).