Food Instability and Academic Achievement: A Quasi-Experiment Using SNAP Benefit Timing

Abstract

Although social policies aimed at low-income families are thought to promote children’s educational success, little research has examined how these policies are related to children’s academic achievement. This article focuses on the Supplemental Nutrition Assistance Program (SNAP), the United States’ largest food assistance program. Using administrative data on over 148,000 SNAP-receiving public school children, we analyze the recency of SNAP benefit transfer and children’s end-of-grade math and reading achievement test scores. Results indicate differences in students’ math and reading performance based on the recency of SNAP benefit transfer. Although the relationship is stronger for reading than for math, the relationship between students’ test scores and SNAP transfer is roughly curvilinear. Test scores peak in the third week following benefit transfer.

Keywords

food stamps SNAP academic achievement poverty low-income children

One persistent gap in educational achievement is between low-income and high-income children: Children living in poverty lag behind their higher-income peers when it comes to school achievement (Michelmore & Dynarski, 2016; Reardon, 2011). Although many factors lead to these gaps in school achievement between low-income and high-income children, for poor children, life outside of school has a particularly large influence on performance within school. Poor children’s families experience much more economic instability than higher-income families (Morris, Hill, Gennetian, Rodrigues, & Tubbs, 2014) and economic instability has been linked to worse educational outcomes (Gennetian, Wolf, Hill, & Morris, 2015). Additionally, poor children are exposed to other risk factors for poor school performance, including exposure to more environmental toxins, family instability, and overcrowded or unsafe housing arrangements (for a review, see Evans, 2004), each of which can lead to worse performance in school.

Social safety net programs, such as the Supplemental Nutrition Assistance Program (SNAP; formerly known as Food Stamps), are meant to buffer economic instability for low-income families. SNAP provides cash-like benefits to low-income individuals and families once a month to use only for purchasing food. Benefits are given to households as a unit and are transferred to households once a month. In Fiscal Year 2012 (FY2012), an average of 42.1 million individuals received $5.7 billion in benefits each month (Eslami, 2015). Of those individuals receiving SNAP in FY2012, an estimated 12.6 million (30%) were school-aged children. Understanding how being in a household that receives SNAP affects children’s school achievement is crucial to both research and policy efforts aimed at closing the achievement gap between low-income and high-income children.

SNAP constitutes a large percentage of many households’ budgets, and conservative estimates suggest that SNAP lifts about 2.1 million children out of poverty annually (Hoynes & Schanzenbach, 2014). Although SNAP provides an important support for low-income families, there is growing evidence that its benefit levels are insufficient for many, leading to different food and nutrition-related outcomes depending on the amount of time within a month that has passed since SNAP benefit transfer. SNAP recipients’ food shopping and food and caloric intake vary throughout the month, with recipients spending more money and consuming more and healthier food right after benefit transfer, compared to at the end of the benefit month (Castner & Henke, 2011; Hastings & Washington, 2010; Wilde & Ranney, 2000). In this article, we build on this literature by focusing on within-month variation in SNAP-recipient children’s academic achievement. The variation in food spending and consumption throughout the SNAP benefit month suggests that SNAP-recipient families also likely experience variability in other important outcomes, such as stress and family interactions. Given literature that has linked nutrition as well as family stress with cognitive functioning and achievement test scores (Frongillo, Jyoti, & Jones, 2006; Gershoff, Aber, Raver, & Lennon, 2007; Gómez-Pinilla, 2008), variability in either or both of these is likely to lead to variability in children’s academic achievement throughout the SNAP benefit month.

Using a matched administrative dataset from North Carolina with over 148,000 SNAP-receiving public school students, we examine the impacts of the SNAP benefit cycle, or the amount of time elapsed since SNAP transfer, on children’s achievement test scores. We focus on academic achievement test scores not only because of the importance of cognitive development but also because of the widening gap in test score performance between low- and high-income students, much of which appears attributable to out-of-school factors (Reardon, 2011).

Effects of SNAP Receipt and the SNAP Benefit Cycle on Children and Families

Most prior research on SNAP has focused on the effects of benefit receipt for families’ economic well-being, as well as individuals’ physical health. Although some early studies found that SNAP receipt was associated with negative dietary and health outcomes, these studies relied on comparisons between participants and nonparticipants and were therefore limited by fundamental differences between these two groups, as SNAP recipients tend to be more disadvantaged than eligible nonrecipients (Bitler, 2014). More recent research, taking advantage of variation in timing of implementation at the county level as a natural experiment, has found that the introduction of the Food Stamp program improved families’ economic well-being as well as individuals’ physical health (Hoynes & Schanzenbach, 2009). Specifically, the initiation of SNAP in the 1960s decreased low birth weight infants, infant mortality, and serious physical health problems (Almond, Hoynes, & Schanzenbach, 2011). Other recent research on SNAP, relying on the variation in eligibility for immigrants from 1996 to 2003, found similar positive effects of the program: Losing one year of parent eligibility in early childhood was associated with an increase of $90 per child in later health costs (East, 2015).

Despite this recent increase in the number of well-designed studies examining outcomes associated with SNAP participation, little research has examined how SNAP impacts other aspects of family well-being or examined SNAP’s impacts on children’s outcomes in domains other than physical health (Gassman-Pines & Hill, 2013). One prior study, using individual fixed effects, found that children beginning participation in the Food Stamp Program (FSP) between kindergarten and third grade improved their math and reading scores, as compared with children ending FSP participation during the same time period (Frongillo et al., 2006). This study provides important evidence that SNAP participation may affect academic achievement.

In addition to the general effect of SNAP benefit receipt on families, specific elements of SNAP program design, such as amount of benefit and mode of distribution, may also impact family outcomes. Throughout the U.S., nearly all SNAP recipients receive benefits once a month through electronic benefits transfer (Hoynes & Schanzenbach, 2014). SNAP recipients often exhaust their benefits before the month has ended; in 2009, 95% of households with children used at least half of their benefit within the first 2 weeks after transfer, and 46% of households with children used over 90% of their benefits within the first 2 weeks after transfer (Castner & Henke, 2011). Therefore, towards the end of the month, recipients spent less on food (Hastings & Washington, 2010). These patterns can be explained partially by families’ shopping habits: Many SNAP-recipient families make one large shopping trip at the beginning of their SNAP month (Damon, King, & Leibtag, 2013; Wiig & Smith, 2009; Wilde & Ranney, 2000). Such shopping trips can be costly in terms of both time and transportation; therefore, spending the majority of benefits quickly after transfer in a large shopping trip is one strategy to maximize benefits and reduce shopping-related costs.

However, for many SNAP recipients, benefits appear to be insufficient to sustain families’ food budgets. Recently, researchers have focused on the implications of the SNAP benefit cycle, seeking to understand how the recency of SNAP benefit transfer is related not only to families’ spending of SNAP benefits but also families’ nutrition and caloric intake. SNAP recipients take in fewer calories at the end of the month, compared to the beginning of the month (Shapiro, 2005; Todd, 2015). Although one study found little evidence that SNAP recipients’ diet varies over the month (Hastings & Washington, 2010), others have found that SNAP recipients consume fewer meat products towards the end of the month (Todd, 2015). SNAP recipients themselves report buying and eating differently throughout the month. At the beginning of the month, they describe buying and consuming a larger variety of foods, particularly meat and produce; at the end of the month, they report eating more prepackaged foods and cheap carbohydrates (Darko, Eggett, & Richards, 2013; Seefeldt & Castelli, 2009). Importantly, these relationships may not be linear: In both the early and late stages of the SNAP benefit cycle, recipients appear to consume higher levels of calories, fat, and protein as opposed to fruit, vegetables, and whole grains (Kharmats et al., 2014).

Because of the significant instability in SNAP recipients’ nutrition and dietary choices throughout the month, researchers have begun asking whether families also experience within-month variability in other important outcomes. New research suggests that this variability may have important health effects: In the last week of the SNAP benefit month, risk for hospital admission due to hypoglycemia increases by 27% amongst the low-income population, with no similar increase among the high-income population (Seligman, Bolger, Guzman, Lopez, & Bibbins-Domingo, 2014). Few studies, however, have examined how this variability may impact the outcomes of children in SNAP-recipient households, and no studies have examined how this variability may impact children’s cognitive functioning. Monthly instability due to the SNAP benefit cycle may impact children’s cognitive functioning through two primary pathways: (a) nutrition or (b) stress and family functioning.

Potential Mechanisms Linking the SNAP Benefit Cycle to Student Test Performance

Nutritional variability due to SNAP timing might impact student performance either through hunger or access to key nutrients. A growing body of literature suggests that specific nutrients can improve cognitive functioning (Bryan, Calvaresi, & Hughes, 2002; Gómez-Pinilla, 2008) as well as the executive functioning skills needed to successfully take standardized achievement tests (Cohen, Gorski, Gruber, Kurdziel, & Rimm, 2016). Although students who receive SNAP are likely also to receive free school breakfast and lunch, access to those meals is consistent throughout the month for all students. Free school breakfast and lunch also do not preclude children from being affected by nutritional variability at home due to SNAP timing. Even if students are unlikely to be hungry, there is evidence that changes in nutrition directly affect students’ test scores (Figlio & Winicki, 2005). Very little research has examined the functional form of change in nutrients within the SNAP benefit month, but emerging evidence suggests that the relationship between time since SNAP benefit transfer and macronutrient intake may be curvilinear, improving after transfer and then declining again at the end of the benefit month (Kharmats et al., 2014).

The SNAP benefit cycle may also impact children’s achievement test performance indirectly through varying levels of family stress, especially as it relates to worry over food acquisition or sufficiency. Prior research has shown that families’ economic well-being and material hardship are related to children’s behavioral and cognitive outcomes through family stress and functioning (Gershoff et al., 2007). Even if children in SNAP households are not directly affected by the SNAP cycle in terms of their food intake or the nutritional quality of their food, others in the family—particularly parents—may not be buffered from food insecurity. Parents’ related stress may spill over and affect children.

Preliminary evidence suggests that SNAP-recipient parents report varying levels of daily food–related hardship over the SNAP benefit cycle, with worry about food the lowest in the middle of the month and substantially higher at the end of the month, compared to immediately after SNAP transfer (Gassman-Pines & Schenck-Fontaine, 2017). Parental stress over limited resources may be transmitted to children through poorer interactions. Family interactions may also be worsened if parents are unable to provide children with foods they enjoy at different points in the SNAP month. Alternatively, many children are highly aware of food cost and may associate cheaper foods at the very beginning and at the end of the SNAP month with decreased resources (Ludvigsen & Scott, 2009).

Variation in levels of family stress may also impact student achievement indirectly through children’s behavioral well-being, as research has linked family stress to problems with children’s externalizing behaviors (Conger & Donnellan, 2007). Preliminary research does suggest that the SNAP benefit cycle may impact children’s behavioral well-being: Grade 5 through 8 students receiving SNAP are more likely to receive a disciplinary infraction towards the end of the SNAP month, as compared to non-SNAP receiving students (Gennetian, Seshadri, Hess, Winn, & George, 2016). Stress and family interactions have also been directly linked to children’s cognitive functioning and test performance (Eysenck, Derakshan, Santos, & Calvo, 2007; Pianta & Egeland, 1994).

Variation by Family and Child Characteristics

The relationship between SNAP recency and children’s test performance may also vary by child characteristics, including children’s gender and race and ethnicity. Girls experience more negative consequences of declines in family economic status than boys and may also be more strongly affected by family economic supports (Elder & Caspi, 1988). Girls may also be more affected by SNAP, in particular, because they are more likely than boys to be engaged in shopping and food preparation activities within the household: On average, girls complete more housework than boys, and children are also typically assigned gender-stereotyped housework (Raley & Bianchi, 2006). Additionally, low-income parents rely on children, particularly daughters, to assist with housework to a greater extent than higher-income parents (Dodson & Dickert, 2004).

In addition, some prior research suggests SNAP participation has greater effects for female, as opposed to male, children. In one prior study examining the impacts of SNAP participation on child test scores, researchers found significant improvements in test scores from kindergarten to 3rd grade associated with SNAP participation (Frongillo et al., 2006). These improvements were driven entirely by female students in reading and mostly by female students in math. Recent research also suggests that the initiation of SNAP in the 1960s improved later-life economic self-sufficiency amongst women exposed to the program at an early age; a similar effect was not found for men able to access SNAP at an early age (Hoynes, Schanzenbach, & Almond, in press).

Additionally, although students receiving SNAP are all disadvantaged, levels of disadvantage vary within this population and could lead to differential effects on students from different racial groups. Prior research indicates that Black children are likely to live in households with lower incomes (Darity & Nicholson, 2005). Additionally, Black families have significantly lower levels of assets than White families, even at the lowest income levels (Darity & Nicholson, 2005; Oliver & Shapiro, 2006). Thus, Black children may live in families that rely more heavily on SNAP than White children and may be more strongly affected by the timing of SNAP transfer. Families who run out of money to purchase food during the month often turn to savings, either in terms of previously purchased food or other assets; and the ability to rely on assets at the end of the benefit month likely varies for different racial groups.

Some children may have access to additional nutrition-related resources based on family characteristics; in particular, SNAP families with very young children are additionally eligible for the Special Supplemental Nutrition Program for Women, Infants, and Children (WIC). The cross-incidence of SNAP and WIC receipt appears to be high among families eligible for both programs. In Illinois, low-income families (as indicated by their receipt of Medicaid) with eligible children overwhelmingly participated in both programs: About 70% of families received both SNAP and WIC (Lee & Mackey-Bilaver, 2007). Therefore, we hypothesize that children with siblings under 2 years of age (when WIC participation is highest; Hoynes & Schanzenbach, 2014) will be less impacted by the SNAP benefit cycle than children without siblings under 2, as WIC provides additional support to those families and likely serves as a buffer.

Current Study

The current study addresses a number of gaps in the literature by examining the impacts of the SNAP benefit cycle on outcomes other than families’ nutrition and caloric intake, specifically SNAP-recipient children’s academic achievement test scores. Building on the literature that has compared student achievement for those entering SNAP compared to those exiting SNAP (Frongillo et al., 2006), the current study focuses on understanding within-month variability among students in SNAP-recipient households. We are able to determine the causal relationship between timing of SNAP transfer and student performance on end-of-grade (EOG) achievement tests because of a unique attribute of the North Carolina SNAP distribution schedule: Instead of all households receiving SNAP on the same day each month, distributions are staggered throughout each calendar month based on the last digit of the recipient’s Social Security Number (SSN). Therefore, timing of benefit transfer within the month varies randomly at the household level. We hypothesize that children’s test scores will decrease over the SNAP benefit month, with students who take achievement tests at the end of their families’ benefit months scoring lower than students who take tests at the beginning of their families’ benefit months.

Method

Data

Our study drew on a unique dataset constructed from multiple administrative data sources in North Carolina during the 2011–2012 academic year, prior to a major SNAP data system change that prevented us from utilizing post-2012 data. First, test score and student demographic data were from the North Carolina Education Research Data Center (NCERDC). NCERDC maintains all of the administrative records on North Carolina public school students that are collected by the state Department of Public Instruction and makes them available to researchers. We utilized EOG test scores in reading and math from the 2011–2012 academic year. All third- through eighth-grade students in North Carolina are required to take EOG achievement tests in both reading and math. In October 2010, the State Board of Education stopped requiring schools’ use of EOG scores in student promotion decisions in Grades 3, 5, and 8. However, EOG scores continued to be used to compute school growth and performance as required by North Carolina’s ABCs Accountability Program and to determine adequate yearly progress (AYP). EOG tests in reading comprehension measure the ability to demonstrate understanding of a written passage and knowledge of vocabulary. EOG tests in math measure proficiency in five areas: numbers and operations; measurement; geometry; data analysis and probability; and algebra. EOG test score files include raw test scores, as well as students’ race/ethnicity, sex, grade level, and school.

EOG test score files also included test dates, which are generally scheduled within the last 15 days of the school year. However, the test dates were approximate and may not have represented the exact date on which students took a test (for example, they might have indicated the date tests were scanned or entered into district recordkeeping systems). Therefore, in order to obtain actual test dates, we contacted testing administrators in all North Carolina school districts. Out of 115 school districts contacted, we were able to obtain updated testing information for 79 districts. The quality of information varied: Some testing administrators could provide exact dates for each test, some provided the week of testing, and some provided a wider range. Based on conversations with testing administrators about typical test administrations, when ranges were provided, we assumed the first Tuesday as the default reading test day and the first Wednesday as the default math test day. Using this information, we updated test dates for all students in these 79 districts. For charter schools and districts without updated information on test dates, we used the original administrative data. Excluding these students does not substantially change our results (results available from authors). Through our conversations with test administrators, we also learned that EOG math tests for Grades 3 through 7 occurred across two consecutive days in the 2011–2012 academic year. Unfortunately, we did not have access to individual test items and so could not separate math EOG test scores by day of test. Instead, we used the first math test day in our analyses and examined total math scores. Therefore, we expect our results for math to be less precise than our results for reading.

Data on children receiving SNAP in April, May, and/or June of 2012 were obtained from the Division of Social Services in the North Carolina Department of Health and Human Services. The SNAP data included recipient children’s names, birth dates, and addresses, as well as benefit amounts and date of benefit transfer.

Data Matching

To match children across the two data sources, we used DataMatch Enterprise, a commercial matching software that uses a fuzzy logic algorithm to match names from different sources and that can account for typos and misspellings (e.g., it can match “Karen” to “Caren” and “Smithh” to “Smith”). In our matching process, we matched on children’s first and last names, birth dates and county of residence, requiring exact matches for date of birth and county and using a less stringent cutoff for first and last names. Using these procedures, we were able to match 87% of children ages 9 to 14 with NCERDC data, for a total of 198,582 students with both SNAP receipt and an EOG math or reading score. Nationally, about 4% of families earning under $20,000 annually send K–12 students to private school (U.S. Census Bureau, 2013). If the same share applies to North Carolina, we would only expect to match a maximum of 96% of SNAP-recipient children with a public school test record, suggesting that our match rate of children who attend public school is even higher. Although there are not common standards on match rates, we note that a match rate of 85% is commensurate with or higher than other research that has matched individual children across administrative data sources in North Carolina (Gibson-Davis, Ananat, & Gassman-Pines, 2016; Ladd, Muschkin, & Dodge, 2014; Muschkin, Ladd, & Dodge, 2015).

To test the quality of the match, we compared matched and nonmatched children on demographic and other characteristics. Most importantly, match rates did not vary by day of SNAP distribution. We also found no significant differences between matched and nonmatched children in terms of age or family size (results available from authors). We did find a significant difference between matched and nonmatched children in terms of benefit size; nonmatched children lived in families that received $8 more per month, on average, than the families of matched children. This difference is not substantial and may reflect higher levels of mobility between schools for children in the most disadvantaged families, who are typically receiving more in SNAP. Finally, although there was some variation in match rate by county, match rates were consistently high across nearly all counties. Only eight counties had match rates below 80%, and no county had a match rate below 70%.

Measures

For our primary dependent variable of interest, EOG math and reading test scores, we used any student with a valid first (regular administration) test score in either math or reading. We excluded retest scores, as well as any students who had only retest scores, because students were scheduled to take retests nonrandomly (students who took retests were mostly students who scored at not proficient levels on the regular administration). We standardized scores using the entire population of students within the same grade and subject. Because students receiving SNAP perform, on average, below students not receiving SNAP, the average standardized test scores for students in the SNAP population were negative.

To construct our primary independent variable of interest, number of days elapsed since SNAP benefit transfer, we subtracted the date of most recent SNAP transfer from test date.

Sample

We restricted our sample in two ways. First, we restricted our sample to students who received SNAP within one month prior to their first test date and excluded cases where SNAP transfer in April, May, and June occurred on days not indicated in the North Carolina Monthly Benefit Issuance Schedule. According to the schedule, families should receive SNAP on the 3rd, 5th, 7th, 9th, 11th, 13th, 15th, 17th, 19th, and 21st of each month, according to the last digit of the household head’s SSN. In the SNAP administrative data, however, a substantial portion of families received SNAP transfers on dates not indicated in the schedule in at least one month, although very few families only received benefits on dates not indicated in the schedule over the 3-month period. Based on our conversations with SNAP policy administrators, we believe that, when families received benefits on dates not indicated in the schedule, these benefits may have been past due. Therefore, families receiving benefits on days not indicated in the schedule may be different from other families because they were new to the SNAP program or had missed transfers in the recent past. Although we report results with these families, provided that they also received benefits on a day indicated in the schedule, excluding them does not change the pattern of findings (see Table 4). Second, we excluded some students whose families received SNAP on the 3rd of the month. For recipients without a SSN, benefits are transferred on the 3rd of the month. Households in which the recipient lacks an SSN are likely to be systematically different from other SNAP-receiving households. Unfortunately, individuals whose SSN ends in “1” also receive benefits on the 3rd of the month. Hypothesizing that most recipients without SSNs are likely new Hispanic immigrants, we also excluded children who received SNAP on the 3rd of the month and are Hispanic but include children who otherwise received SNAP on the 3rd of the month. Our final sample, therefore, consists of 147,418 students with reading scores and 148,316 students with math scores. Compared to the full sample of SNAP-recipient students, our analysis sample is similar along demographic lines, but appears to have more over-time variability in SNAP receipt (results available from authors).

Analytic Plan

As described above, in North Carolina, SNAP benefit disbursement is determined by the last digit of the household head’s SSN, as indicated in the North Carolina Monthly Benefit Issuance Schedule. When EOG tests are administered within a school, some children in SNAP-receiving families are in households that have just received benefit payments, whereas other children are in households at the end of their monthly cycle. This structure is depicted in Figure 1, showing students in a hypothetical school with tests given on the 24th of the month. Within that school, all students take the test on the same day but the day on which they receive SNAP varies and thus the amount of time that has elapsed between SNAP transfer and the test date also varies. This difference is depicted in Figure 1 showing SNAP transfer dates for four hypothetical students: Student A receives SNAP on the 21st of the month and takes the test 3 days after SNAP transfer; Student B receives SNAP on the 17th of the month and takes the test 7 days after SNAP transfer; Student C receives SNAP on the 5th of the month and takes the test 19 days after SNAP transfer; and Student D receives SNAP on the 9th of the month and takes the test 15 days after SNAP transfer. Comparing test scores for these children provides information about the effect of time elapsed since SNAP benefit transfer on test performance, holding constant the share of children in a school who receive SNAP. Thus, our study is a quasi-experiment where the treatment is recency of transfer of monthly SNAP payment. The distribution of treatment is essentially random, as it is determined only by the last digit of participants’ SSNs. Differences in test scores that are found are therefore unlikely to be due to systematic differences between children, because those who are at the beginning of their families’ monthly cycle are likely the same, or highly similar, to those at the end of the cycle.

Figure 1.

Construction of days since SNAP variable for hypothetical students in the same school.

To check our randomization, we tested the relationship between recency of SNAP transfer and various characteristics, including race/ethnicity, gender, presence of a sibling under age 2, geography (urban vs. rural county of residence), and benefit amount. We find no significant or substantial relationships for any of the characteristics, except geography (results available from authors). Given that test dates are often determined at the district level and districts typically overlap with counties, SNAP recency may not be adequately randomized between rural and urban counties. We also checked the randomization of SNAP recency by examining the relationship between student test performance and SNAP transfer day. Figure 2 shows the unadjusted relationship between day of SNAP transfer (3, 5, 7, 9, 11, 13, 15, 17, 19, 21) and average student reading and math test scores. In this graph, there appears to be no relationship between average test scores and day of SNAP transfer. Taken together, these analyses suggest that day of SNAP transfer was adequately randomized.

Figure 2.

Average test scores by day of SNAP receipt.

Although we initially hypothesized a linear relationship between days since SNAP transfer and test score, our preliminary analyses indicated that the relationship was nonlinear (see Results section below). Therefore, we include both a linear and quadratic term for number of days elapsed since SNAP benefit transfer in our models. We use ordinary least squares (OLS) regression to model the effect of recency of transfer of monthly SNAP payment on student test score, using the following equation:

S c o r e_{i s} = β_{1} D a y s_{i s} + β_{2} D a y {s_{i s}}^{2} + β_{3} G r a d e_{i s} + β_{n} X_{i s} + α_{s} + ε,

where Score is the EOG math or reading test score of a given student i in school s in a given subject, Days is a continuous variable of the number of days between SNAP transfer and student test date (1–30), Days² is the squared term of the Days variable, Grade is a set of indicators for student’s grade, X is a vector of student demographic characteristics, and α_s are school fixed effects.

School fixed effects improve our estimates’ precision by removing variance in students’ performance due to stable school characteristics. Importantly, school fixed effects also account for variation in school test dates: although most North Carolina schools administer EOG tests within the last 15 days of the school year, test dates differ across schools, and school years end at different times. When we regressed student performance on test dates, we found that test date had a negative linear relationship with student performance. We hypothesize that schools and school districts schedule tests later in the school year if students in that school, generally, do not perform well on exams; schools and school districts in which most students score well on exams may be less concerned about maximizing learning time and likely to schedule tests earlier. School fixed effects prevent any temporal differences in test scheduling that are also associated with student performance from affecting our results; indeed, once we include school fixed effects in regressions of school performance on test dates, results are no longer significant (all results are available from authors). We also include heteroskedasticity-robust standard errors clustered at the school level to adjust for the nonindependence of observations within schools.

Results

Descriptive Results

We find that approximately 29.7% of North Carolina students in Grades 3 through 8 received SNAP in March, April, or May of 2012; considering that we were not able to match all SNAP-receiving children with NCERDC records, a slightly larger percentage of North Carolina public school students likely receive SNAP. The proportion of students receiving SNAP varied substantially by county, ranging from 14.2% to 57.6%. Relative to nonrecipient students, SNAP-recipient students scored lower on EOG exams: Unadjusted, SNAP-recipient students score 0.64 standard deviations lower in math and 0.63 standard deviations lower in reading. Controlling for race/ethnicity, gender, and grade, as well as school fixed effects, the difference between SNAP-recipient and nonrecipient students shrank but remained substantial: 0.36 standard deviations for math and 0.35 standard deviations for reading.

Table 1 displays descriptive statistics for our population of SNAP-recipient students, including sex, race/ethnicity, residency in a county that contained a Metropolitan Statistical Area, whether or not the student has a sibling under 2 years, SNAP benefit amount, and grade. As shown, the largest group in our population consisted of Black students, followed closely by White students. The number of students receiving SNAP was fairly similar within the early grades, with a slight drop-off in middle school, especially 8th grade. For our main independent variable of interest, number of days elapsed since SNAP benefit transfer, the distribution of students associated with each number of days (1–30) is fairly uniform (results available from authors). However, there is some variation, as test administration dates were not randomly assigned and no individuals receive SNAP during the last third of the month (from the 22nd on).

Table 1

Sample Characteristics

Female	.50
Race and ethnicity
Asian/Pacific Islander	.01
Black	.47
Hispanic	.10
Native American	.03
Multiracial	.05
White	.35
Grade
3	.18
4	.18
5	.17
6	.17
7	.16
8	.14
Lives in urban area	.63
Has sibling under age 2 years	.12
Average monthly SNAP amount—Mean (SD)	$464.70
	($241.40)
Days since SNAP receipt
1–5 days	.23
6–10 days	.22
11–15 days	.18
16–20 days	.11
21–25 days	.10
26–30 days	.17
N	148,316

Note. Statistics refer to proportion of respondents unless otherwise indicated. N presented is for the number of students with a valid math test score. A total of 147,418 have a valid reading test score.

Main Effects of SNAP Timing on Student Achievement

To descriptively demonstrate the relationship between recency of SNAP transfer and average student test scores, Figures 3 and 4 depict the average test score for students who took the test that number of days since SNAP transfer. We also present a Lowess curve (locally weighted scatterplot smoothing), to descriptively demonstrate the overall pattern. Although the relationship is stronger for reading (Figure 4) than for math (Figure 3), the relationship between students’ test scores and SNAP transfer is fairly consistent across the two test types of tests and appears to be curvilinear. As noted above, the small difference between reading and math scores is likely due, in part, to the 2-day nature of the math tests, which reduces precision. Both student reading and math test scores appear to peak in the period from the 21st to the 25th day post-SNAP transfer. In both cases, the effect size is modest: The average score by days since SNAP transfer ranges from the peak to the trough by 17% of a standard deviation for reading and 12% of a standard deviation for math.

Figure 3.

Average math test scores by days since SNAP receipt.

Figure 4.

Average reading test scores by days since SNAP receipt.

These initial results are confirmed in our OLS models, which are presented in Table 2.

Table 2

OLS Regression Results Predicting End-of-Grade Math and Reading Achievement Test Scores From Days Since SNAP Receipt

	No Fixed Effects		School Fixed Effects and School-Level SE Clustering
	Math	Reading	Math	Reading
	b	b	b	b
	(SE)	(SE)	(SE)	(SE)
Days since SNAP receipt	.0034***	.0059***	.0025**	.0027**
	(.0011)	(.0011)	(.0012)	(.0012)
Days since SNAP receipt squared	−.0001**	−.0002***	−.0001*	−.0001**
	(.0000)	(.0000)	(.0000)	(.0000)
Observations	148,316	147,418	148,316	147,418
Number of schools			1,904	1,903

Note. All models control for student covariates: gender, race and ethnicity, grade, presence of a young sibling, and SNAP benefit amount. Robust standard errors are in parentheses.

p < .1. **p < .05. ***p < .01.

In the first column, we show results without school fixed effects, but controlling for student demographic characteristics. Consistent with the pattern of findings shown in Figures 3 and 4, student performance on EOG math and reading tests has an inverse U-shaped relationship with days since SNAP transfer. In both models, there is a statistically significant quadratic relationship between days since SNAP transfer and student test scores. In the second column, we show results including school fixed effects, which control for all stable differences between schools and is our preferred specification, as described above. The pattern of results is unchanged, with statistically significant quadratic relationships between days since SNAP transfer and both math and reading test scores. When school fixed effects were included, however, student performance appeared to peak slightly earlier than in the unadjusted analyses, at Day 19 for math and Day 17 for reading. Additionally, after adjusting for school fixed effects, impacts were smaller than in the unadjusted results: Average test scores by days since SNAP transfer ranged from 2.1% of a standard deviation for reading to 2.2% of a standard deviation for math, from peak to trough.

In addition to examining the effect of the number of days since SNAP transfer on the mean achievement test score, we also considered a related outcome: whether the student’s score met the standard for grade level proficiency (GLP) in that subject. These results appear in Table 3, which shows that the number of days since SNAP transfer was related to the probability of meeting the GLP standard. Consistent with the mean test score findings, these results are curvilinear, with students most likely to meet GLP standards in reading when they took the test 16 days following SNAP transfer and most likely to meet the GLP standards in math when they took the test 20 days following SNAP transfer. At the peak, compared to the beginning of the SNAP month, students’ probability of meeting GLP standards in reading were 0.9 percentage points higher and students’ probability of meeting GLP standards in math were 1.1 percentages points higher.

Table 3

OLS Regression Results Predicting Probability of Meeting Grade Level Proficiency Standards in Math and Reading From Days Since SNAP Receipt

	Proficient in Math	Proficient in Reading
	b	b
	(SE)	(SE)
Days since SNAP receipt	.0011*	.0011*
	(.0006)	(.0006)
Days since SNAP receipt squared	−.0000	−.0000*
	(.0000)	(.0000)
Observations	148,316	147,418
Number of Schools	1,904	1,903

Note. All models include school fixed effects and control for student covariates: gender; race and ethnicity; grade; presence of a young sibling; and SNAP benefit amount. Robust standard errors, clustered at the school level, are in parentheses.

p < .1. **p < .05. ***p < .01.

Robustness Checks

We ran a set of alternate model specifications to test the robustness of our findings; results are robust to these different model specifications (all robustness check results are shown in Table 4). First, we estimated models with school fixed effects but only including schools with at least 30 SNAP-receiving students; results did not change. Second, we also estimated models with district fixed effects, with heteroskedasticity-robust standard errors clustered at the district level. Although test dates are typically set at the district level, we prefer the school fixed effect model to the district fixed effect model for several reasons. School policies often differ dramatically, even within districts. Also, districts in North Carolina typically encompass the entire county, meaning that variation in student populations across schools but within districts is larger than it might be in a state with more districts. Nevertheless, the results with the district fixed effects were highly consistent with the school fixed effects results. Third, because SNAP is transferred to households and not individuals, we clustered the standard errors at the family level. These results, which adjust standard errors for the presence of siblings in the same household, were also consistent with results clustering standard errors at the school level. Fourth, results were robust to the inclusion of an additional covariate indicating whether the student ever had a SNAP distribution on a date not included in the benefit issuance schedule. Similarly, when we used all available SNAP transfer dates, and not only those that conformed to the benefit issuance schedule, the results are highly consistent for reading scores and somewhat smaller but in the same direction for math scores. Fifth, results were consistent and remained significant when we excluded students from charter schools, when we excluded all students who receive SNAP distributions on the 3rd of the month, and when we excluded students with a nonstandard SNAP transfer date at any point. Sixth, although estimated with less precision due to the smaller sample size, the results were the same when we limited our analysis to only those schools with exact test dates or a specific week. Seventh, we altered our assumptions about test days when the school district was not able to provide specific dates. Instead of assuming that reading tests were on Tuesday and math tests on Wednesday, we ran two separate models, one assuming reading tests were on Monday and math tests were on Tuesday, and one assuming reading tests were on Wednesday and math tests were on Thursday. Estimates in both alternate specifications were substantially similar to those in the main model. All of these analyses increased our confidence in the main findings.

Table 4

Summary of Robustness Checks

	District Fixed Effects and District-Level SE Clustering		Family-Level SE Clustering		Including Covariate Flagging Nonstandard Day of SNAP Receipt		Excluding Schools With Fewer Than 30 SNAP- Recipient Students		Exclude Students at Charter Schools		Excluding All Students Receiving SNAP on 3rd of the Month		Excluding Students With Nonstandard Day of SNAP Receipt
	Math	Reading	Math	Reading	Math	Reading	Math	Reading	Math	Reading	Math	Reading	Math	Reading
	b	b	b	b	b	b	b	b	b	b	b	b	b	b
	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)
Days since SNAP receipt	.0025**	.0027**	.0021*	.0023*	.0025**	.0027**	.0022*	.0024**	.0024**	.0027**	.0030**	.0030**	.0027**	.0028**
	(.0012)	(.0014)	(.0012)	(.0012)	(.0012)	(.0012)	(.0012)	(.0012)	(.0012)	(.0012)	(.0013)	(.0013)	(.0012)	(.0012)
Days since SNAP receipt squared	−.0001	−.0001*	−.0001	−.0001	−.0001*	−.0001**	−.0001	−.0001*	−.0001	−.0001**	−.0001**	−.0001**	−.0001*	−.0001**
	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)
Observations	148,316	147,418	148,316	147,418	148,316	147,418	141,584	140,633	145,283	144,397	133,321	132,601	133,504	132,722
Number of schools			1,904	1,903	1,904	1,903	1,484	1,480	1,812	1,811	1,898	1,897	1,900	1,899
Number of districts	208	208
	Including All SNAP Distribution Dates		Assume Reading Monday and Math Tuesday		Assume Reading Wednesday and Math Thursday		Only Including Exact and Within Week Test Dates		Including All Distribution Dates
	Math	Reading	Math	Reading	Math	Reading	Math	Reading	Math	Reading
	b	b	b	b	b	b	b	b	b	b
	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)	(SE)
Days since SNAP receipt	.0013	.0022*	.0017	.0021	.0016	.0022*	.0024	.0026*	.0014	.0005
	(.0011)	(.0012)	(.0012)	(.0013)	(.0012)	(.0012)	(.0015)	(.0015)	(.0011)	(.0011)
Days since SNAP receipt squared	−.0000	−.0001*	−.0000	−.0001	−.0000	−.0001	−.0001	−.0001*	.0000	.0000
	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)	(.0000)
Observations	157,134	155,380	147,882	148,715	147,303	149,288	94,951	93,975	168,658	166,734
Number of schools	1,922	1,925	1,910	1,910	1,906	1,908	1,126	1,125	1,922	1,925

Note. All models include school or district fixed effects and control for student covariates: gender, race and ethnicity, grade, presence of a young sibling, and SNAP benefit amount. Robust standard errors, clustered at the school or district level, are in parentheses.

p < .1. **p < .05. ***p < .01.

To consider nonlinearity in other ways, we also ran alternate models with a set of indicator variables representing 5-day periods between SNAP benefit transfer and test date (Table 5). When the relation between days since benefit transfer and test performance is modeled as a set of indicator variables representing 5-day periods, we also see significant differences between the test performance of students receiving SNAP 11 to 15 days and 21 to 25 days prior to their test day and students receiving SNAP 1 to 5 days prior to their test day. Models examining indicators for 2-day periods were also consistent with the main results (results available from authors).

Table 5

OLS Regression Results Predicting End-of-Grade Math and Reading Achievement Test Scores From Categorical Groups of Days Since SNAP Receipt

	Math	Reading
	b	b
Days Since SNAP Transfer	(SE)	(SE)
6–10	0.0081	0.0127
	(0.0071)	(0.0078)
11–15	0.0128*	0.0165**
	(0.0074)	(0.0081)
16–20	0.0147	0.0136
	(0.0095)	(0.0101)
21–25	0.0223**	0.0211**
	(0.0094)	(0.0099)
26–30	0.0129*	0.0082
	(0.0078)	(0.0083)
Observations	148,316	147,418
Number of schools	1,904	1,903

Note. Omitted category is 1–5 days since SNAP transfer. All models include school fixed effects and control for student covariates: gender, race and ethnicity, grade, presence of a young sibling, and SNAP benefit amount. Robust standard errors, clustered at the school level, are in parentheses.

p < .1. **p < .05. ***p < .01.

We also examined the robustness of our findings by examining the relation between the 1st of the month and students’ test performance, to determine whether our effects were actually due to SNAP or due to other social program benefits that may be transferred on the 1st of the month. As shown in Table 6, we found no relation between the 1st of the month and test performance for students in our sample, increasing our confidence that the effects reported here were due to SNAP and not other benefits that may be distributed on the 1st of the month.

Table 6

OLS Regression Results Predicting End-of-Grade Math and Reading Achievement Test Scores From Days Since 1st of the Month

	Math	Reading
	b	b
	(SE)	(SE)
Days since 1st of month	.0009	−.0084
	(.0127)	(.0192)
Days since 1st of month squared	−.0001	.0002
	(.0004)	(.0007)
Observations	148,316	147,418
Number of schools	1,904	1,903

Finally, an additional robustness check aimed to further explore whether unmeasured factors could be driving the pattern of results. To do so, rather than using actual SNAP transfer dates, SNAP-recipient students were randomly assigned SNAP distribution dates and the regressions were rerun. We repeated this simulation analysis 1,000 times and found that the estimated coefficients on both days and days squared were normally distributed around 0, with the coefficients from the main analyses in the far tails of the distribution and statistically significantly different from the mean, suggesting that our results were highly unlikely to be occurring randomly.

Subgroup Analyses

We further disaggregated results by students’ race/ethnicity, gender, and sibling age (comparing students with a sibling under age two to students without siblings under 2 years) to investigate how these effects differ by child characteristics. We discovered that certain groups of students appeared to be driving results for the overall population. Table 7 summarizes results by student subgroups. When we disaggregated by race, results did not reach statistical significance for either group but were fairly consistent for both Black and White students. However, for both math and reading, White students appear to peak in performance more days post-SNAP transfer than Black students.

Table 7

OLS Regressions Predicting Math and Reading Achievement Test Scores From Days Since SNAP Transfer, by Student Subgroups

	Math	Reading
	b	b
	(SE)	(SE)
Student race
White (N = 52,498)
Days since SNAP receipt	.0035*	.0037*
	(.0021)	(.0021)
Days since SNAP receipt squared	−.0001	−.0001
	(.0001)	(.0001)
Black (N = 69,204)
Days since SNAP receipt	.0027	.0020
	(.0016)	(.0017)
Days since SNAP receipt squared	−.0001	−.0001
	(.0001)	(.0001)
Student gender
Male (N = 73,661)
Days since SNAP receipt	.0022	.0010
	(.0017)	(.0018)
Days since SNAP receipt squared	−.0001	−.0000
	(.0001)	(.0001)
Female (N = 74,655)
Days since SNAP receipt	.0030*	.0044***
	(.0016)	(.0017)
Days since SNAP receipt squared	−.0001	−.0001***
	(.0001)	(.0001)
Presence of a young sibling (under age 2)
Without young sibling (N = 130,422)
Days since SNAP receipt	.0031**	.0030**
	(.0013)	(.0013)
Days since SNAP receipt squared	−.0001**	−.0001**
	(.0000)	(.0000)
With young sibling (N = 17,894)
Days since SNAP receipt	−.0003	.0004
	(.0036)	(.0038)
Days since SNAP receipt squared	.0000	−.0000
	(.0001)	(.0001)

Note. All models include school fixed effects and control for student covariates. Robust standard errors, clustered at the school level, are in parentheses. Ns presented are for the number of students with a valid math test score.

p < .1. **p < .05. ***p < .01.

Disaggregating by student gender, we showed that effects on reading scores were stronger for female, rather than male, students, with female students appearing to be driving the overall results. In the reading model for female students, while coefficients on both the linear and quadratic terms were statistically significant, indicating the female students peak in performance around Day 16, corresponding coefficients on the linear and quadratic terms in the reading model for male students did not reach statistical significance. Female students’ scores in reading appear to vary about 3.1% of a standard deviation over the month. Effects on math scores did not appear to vary by student gender.

Finally, we disaggregated by whether or not the student has a sibling 2 years old or under, as a proxy for WIC eligibility. Our hypothesis was that the additional nutritional assistance provided by the WIC program would buffer the effects of the SNAP cycle. Results were consistent with our hypothesis, as effects remained statistically significant for students without a sibling under 2, whose families would be unlikely to be also accessing the WIC program, but were not statistically significant for students with a sibling under 2, whose families would be more likely to be accessing WIC. Students without siblings under 2 were predicted to score the highest in reading around 17 days post-SNAP transfer and in math around 18 days post-SNAP transfer. Test scores ranged by 2.5% of a standard deviation for both reading and math, for those students without a young sibling.

Discussion

Low-income students’ school achievement is thought to be greatly affected by outside of school factors. This article examined the effect of one important out-of-school factor—the timing of SNAP benefit receipt—on low-income students’ academic achievement. We provide the first evidence that the SNAP benefit cycle not only affects recipients’ nutritional outcomes (e.g., Darko et al., 2013; Hastings & Washington, 2010; Kharmats et al., 2014), but has impacts on children’s test scores as well. This study is the first to examine how the SNAP benefit cycle impacts children’s cognitive functioning. Our results show that students’ test scores have a curvilinear relationship with the time elapsed since SNAP transfer, peaking at approximately 17 days since SNAP transfer for reading and 19 days since SNAP transfer for math. This evidence is particularly compelling because the schedule of SNAP distribution in North Carolina is essentially randomly assigned. Within a school on test day, SNAP-recipient students vary in the number of days since their household has received SNAP benefits and the number of days since transfer is not related to household or student characteristics.

These findings add to the growing body of research demonstrating that, although SNAP is an important support for low-income families, its benefit levels are insufficient for many, leading to within-month variations for members of SNAP-recipient households in important outcomes. The majority of prior work has examined SNAP recipients’ food- and nutrition-related outcomes, showing that SNAP recipients spend more money and consume more and healthier food right after benefit transfer, compared to at the end of the benefit month (Castner & Henke, 2011; Hastings & Washington, 2010; Wilde & Ranney, 2000). The current results show that it is not only those outcomes that are directly targeted by the SNAP program that demonstrate within-month variability, as children’s test performance is also affected. Other research has also demonstrated that the SNAP benefit cycle affects children’s behavior in school (Gennetian et al., 2016). These findings further underscore the importance of out-of-school factors in influencing the school performance of low-income students.

Examining the mechanisms linking the timing of SNAP receipt to children’s achievement test performance was outside the scope of this study; however, there are two plausible ways of considering the timing of SNAP transfer, mediating mechanisms and student test scores, and the limited literature on potential mechanisms is potentially consistent with both pathways. The first way of considering the potential pathway is that test performance is affected immediately by important mediators, such as nutrition and stress, but that those mediators do not change immediately after receiving SNAP. Consistent with that potential causal chain, some prior research suggests a curvilinear functional form to the relationship between recency of SNAP transfer and macronutrient intake, with macronutrient intake highest in the middle of the SNAP month (Kharmats et al., 2014). And emerging evidence also suggests that parents’ perceived food hardship, a marker of stress related to food acquisition, is also lowest in the middle of the SNAP month (Gassman-Pines & Schenck-Fontaine, 2017), in part because SNAP-recipient families may be most likely to borrow money from members of their social networks about three weeks after SNAP transfer (Schenck-Fontaine, Gassman-Pines, & Hill, 2017). Other mediating links could follow this pattern as well, such as changes in students’ propensity to acquire or consume school lunch.

What is less well understood, however, is the timing of how a change in nutrition and/or stress is related to student test performance. If students’ test performance is affected immediately by these mediators, then the pattern of findings is consistent with our results, in which test performance peaks in the middle of the month. For example, others have shown that increases in the caloric content of school lunches on test day were associated with higher scores on EOG achievement tests (Figlio & Winicki, 2005), providing some evidence that changes in nutrition can have same-day effects on test scores. It is also important to note, however, that total calories is only one measure of nutrition and other factors, such as macronutrient intake, could also be related to children’s test performance. Further, performance on high-stakes EOG academic achievement tests likely reflects both actual learning that has occurred throughout the school year, as well as the ability of the student to focus and concentrate during the test itself.

Alternatively, SNAP transfer may improve nutrition and stress relatively quickly, but those changes may take time to accumulate and ultimately affect test performance. This is consistent with prior research showing that SNAP recipients buy and consume a larger variety of foods, particularly meat and produce, at the beginning of the SNAP month; at the end of the month, they report eating more prepackaged foods and cheap carbohydrates (Darko et al., 2013; Seefeldt & Castelli, 2009). Students with peak test performance (who received SNAP around 2 weeks prior to their test date) may have benefited from access to sufficient food resources and lowered stress not only on the day of the test but for the previous 2 weeks. These mediators may not lead to immediate changes in test performance but may take time to lead to improvements.

It is also worth noting that the date on which SNAP households receive their benefits does not vary throughout the year, so the effects found here could occur through any mechanisms related to the timing of SNAP all year, in ways that are either systematic or sporadic. For example, if key pieces of content were taught in classrooms on the same day of the month as the test was ultimately given, that could be driving the results.

Beyond specific content being taught on particular days of the month, it is important to note that, in general, our results suggest students receiving SNAP may experience school achievement cycles throughout the year that are reflected in their EOG test performance. During zenith periods, they may be less able to learn because of temporarily lowered cognitive functioning or less ability to pay attention. Even if these periods represent only a few days each SNAP benefit cycle, these consequences of these “lowered learning” days over the course of the school year may accumulate over the course of the academic year. If, for example, SNAP-receiving children are only affected for 3 days per month, over the course of a school year, these experiences will accumulate as a 10% decline in the share of schooling for which SNAP-receiving children are fully attentive relative to their higher-income peers. This accumulation over the school year may, in part, explain test score gaps between low-income and higher-income children. Indeed, the difference in performance between SNAP-recipient students on their lowest-scoring day and their highest-scoring day—2.2% of a standard deviation for math and 2.1% of a standard deviation for reading—account for about 6% of the overall test score gap (adjusted by covariates and school fixed effects) between SNAP-recipient and nonrecipient students.

It is clear that additional research will be needed to fully understand the causal chain linking SNAP timing to student end-of-grade test performance. Future research should seek to examine the mediators linking the SNAP benefit cycle to low-income students’ test performance and to understand how mediators are related to test performance over time. To do so will require detailed daily-level data collection on food intake, macronutrients, family stress, and other potential mechanisms, over a number of months or a school year. In particular, research using such data collection strategies will be needed in order to uncover how long it takes for changes in mechanisms to affect test performance, and whether different mechanisms are linked over time to test performance differently.

Importantly, our results are consistent across various model specifications and robustness checks. These additional models increase our confidence that the timing of SNAP receipt affects students’ academic achievement. Our results cannot be explained by student, school or district characteristics, or by other social program benefits that may transfer to low-income families on the 1st of the month. Our results additionally cannot be explained by timing of test dates alone; they rely on the interaction between the day of SNAP receipt and test date. Although it would be interesting to compare our pattern of findings with the pattern of findings for non-SNAP students, such an analysis is not possible because it is not possible to construct the key predictor, the number of days between SNAP transfer and test date, for non-SNAP students who do not have a SNAP transfer date. In a state in which SNAP is transferred on the first of the month, the effect of SNAP timing on school disciplinary incidents was evident over and above the relationship between time of month and school discipline for non-SNAP students (Gennetian et al., 2016). We also note that the SNAP distribution schedule in NC is unrelated to the distribution schedule of any other social safety net or social insurance programs, including Temporary Assistance to Needy Families (TANF), Unemployment Insurance, and Supplemental Security Income (SSI). Thus, our results are not being driven by the transfer of other government-provided supports to low-income families.

When we examined the effects of the SNAP benefit cycle by important student subgroups, we found difference that, in general, confirmed our hypotheses. Effects are stronger for female rather than male students in reading, as initially hypothesized. Female students may be more aware of their families’ economic circumstances than male students; they therefore are possibly more affected by the SNAP benefit cycle. Additionally, the SNAP benefit cycle impacts a traditionally gendered household activity: food preparation. Since female students are more likely than male students to assist with food preparation within households, we might suspect that female students will be more aware of cyclical food availability.

As hypothesized, students with siblings under 2 years are much less impacted by the SNAP benefit cycle than students without a young sibling. Since families with young children who are receiving SNAP typically also have access to WIC (Hoynes & Schanzenbach, 2014), they may rely less exclusively on SNAP than families with older children and may be better able to smooth food consumption throughout the month. WIC appears to serve as an important buffer of the effects of the SNAP cycle on student test performance.

Although we had hypothesized differences by student race, impacts of the SNAP cycle do not appear to differ by race/ethnicity. However, although results are not statistically significant, Black students appear to peak in test performance sooner after SNAP transfer, relative to White students. Part of the differences between effects for Black and White students may be related to socioeconomic differences among the SNAP participant population. Overall, the average Black student in our sample is in a family receiving significantly more in SNAP ($473) than the average White student in our sample ($451). Therefore, our results suggest that Black children’s families may rely more centrally on SNAP for support and therefore exhaust that support more quickly. Nevertheless, the consistency of findings across racial groups suggests that the SNAP benefit cycle is experienced across racial groups with different average levels of resources.

Although we were able to examine differences by student characteristics, further research is needed to understand whether results generalize in other contexts. For example, families may have different food-buying patterns and stressors at the end of the school year than, for example, in December during the holiday season. Similarly, the study was conducted in just one state and may not generalize to other states. We do note, however, that North Carolina is the ninth most populous state, with a demographically diverse population, substantial numbers of people living in urban and rural areas, and a sociodemographic profile that closely mirrors that of the U.S. in terms of age, education, marital status, and employment.

Schools and school personnel are acutely aware that many students they serve are experiencing food insecurity (Fram, Frongillo, Fishbein, & Burke, 2014). In response, they employ many formal and informal strategies for addressing their students’ food and nutritional needs (Edwards & Cheeley, 2016; Fram et al., 2014). If these efforts to support students’ nutritional needs, such as giving food to students outside of school meals, are provided more completely during EOG testing windows, then the results reported here may be conservative estimates of the effect of the SNAP cycle on student academic performance, as day of test nutrition also affects test performance (Figlio & Winicki, 2005).

Our results have implications for policy and practice, as understanding how to address the influence of out-of-school factors on low-income students’ school success has been deemed important. Indeed, our findings provide additional support for the argument that addressing the school achievement of low-income students likely requires a whole-child approach. Improving educational outcomes for low-income children may require looking beyond the school door to examine the effectiveness of other types of policies. By understanding the monthly variability low-income students experience through their participation in social programs, educators and other in-school personnel may be better able to support low-income students’ learning, in addition to the various efforts they currently undertake to address food insecurity and support student learning (Fram et al., 2014). For example, it may make sense for schools to target their supports to students may require different levels of support at different times of the month. Or schools or community agencies may be able to target additional food resources to students at different times of the month. Although learning from families when they receive SNAP may be challenging, it could prove useful to educators who could use that information to tailor the supports they are often already providing to individual students at particular times. Educators and other stakeholders across the U.S. could also learn and understand their state’s SNAP benefit issuance schedule.

Second, this study has implications for policies beyond those implemented within schools, adding to the growing body of evidence suggesting that, although SNAP is an important support, benefits are also insufficient for many families. Recent research suggests that, after the temporary boost in SNAP benefit amounts provided by the American Recovery and Reinvestment Act, families receiving SNAP were better able to smooth their food consumption throughout the month (Todd, 2015). Our findings suggest that increasing benefit amounts would also have the added benefit of improving school achievement outcomes for low-income children.

Footnotes

ORCID iD

Anna Gassman-Pines

Notes

Anna Gassman-Pines is an associate professor at the Sanford School of Public Policy, Duke University, Box 90312, Durham, NC 27708; e-mail: agassman.pines@duke.edu . Dr. Gassman-Pines’s work focuses on the development of low-income children in the United States and the ways that contexts external to the family, such as social policies and the labor market, affect family functioning and child well-being.

Laura Bellows is a doctoral candidate in the Sanford School of Public Policy, Duke University. Her work focuses on the links between social policies and children’s academic achievement.

References

Almond

Hoynes

H. W.

Schanzenbach

D. W.

(2011). Inside the war on poverty: The impact of food stamps on birth outcomes. Review of Economics and Statistics, 93(2), 387–403.

Bitler

(2014). The health and nutrition effects of SNAP: Selection into the program and a review of the literature on its effects (Discussion Paper No. DP2014-02). Lexington: University of Kentucky Center for Poverty Research.

Bryan

Calvaresi

Hughes

(2002). Short-term folate, Vitamin B-12 or Vitamin B-6 supplementation slightly affects memory performance but not mood in women of various ages. Journal of Nutrition, 132(6), 1345–1356.

Castner

Henke

(2011). Benefit redemption patterns in the Supplemental Nutrition Assistance Program. Washington, DC: U.S. Department of Agriculture.

Cohen

J. F.

Gorski

M. T.

Gruber

S. A.

Kurdziel

L. B.

Rimm

E. B.

(2016). The effect of healthy dietary consumption on executive cognitive functioning in children and adolescents: A systematic review. British Journal of Nutrition, 116, 1–12.

Conger

R. D.

Donnellan

M. B.

(2007). An interactionist perspective on the socioeconomic context of human development. Annual Review of Psychology, 58, 175–199.

Damon

A. L.

King

R. P.

Leibtag

(2013). First of the month effect: Does it apply across food retail channels? Food Policy, 41, 18–27.

Darity

W. A. J.

Nicholson

M. J.

(2005). Racial wealth inequality and the Black family. In McLoyd

V. C.

Hill

N. E.

Dodge

K. A.

(Eds.), African American family life: Ecological and cultural diversity. New York, NY: Guilford.

Darko

Eggett

D. L.

Richards

(2013). Shopping behaviors of low-income families during a 1-month period of time. Journal of Nutrition Education and Behavior, 45, 20–29.

10.

Dodson

Dickert

(2004). Girls’ family labor in low-income households: A decade of qualitative research. Journal of Marriage and Family, 66, 318–332.

11.

East

(2015, November). The effect of food stamps on children’s health: Evidence from immigrants’ changing eligibility. Paper presented at APPAM 37th Annual Fall Research Conference, Miami, FL.

12.

Edwards

O. W.

Cheeley

(2016). Positive youth development and nutrition: Interdisciplinary strategies to enhance student outcomes. Children & Schools, 38(3), 170–177.

13.

Elder

G. H.

Caspi

(1988). Economic stress in lives: Developmental perspectives. Journal of Social Issues, 44, 25–45.

14.

Eslami

(2015, August). Trends in Supplemental Nutrition Assistance Program participation rates: Fiscal Year 2010 to Fiscal Year 2013. Washington, DC: U.S. Department of Agriculture, Food and Nutrition Service, Office of Policy Support.

15.

Evans

G. W.

(2004). The environment of childhood poverty. American Psychologist, 59, 77–92.

16.

Eysenck

M. W.

Derakshan

Santos

Calvo

M. G.

(2007). Anxiety and cognitive performance: Attentional control theory. Emotion, 7, 336–353.

17.

Figlio

D. N.

Winicki

(2005). Food for thought: The effects of school accountability plans on school nutrition. Journal of Public Economics, 89, 381–394.

18.

Fram

M. S.

Frongillo

E. A.

Fishbein

E. M.

Burke

M. P.

(2014). Roles for schools and school social workers in improving child food security. Children & Schools, 36(4), 231–239.

19.

Frongillo

E. A.

Jyoti

D. F.

Jones

S. J.

(2006). Food stamp program participation is associated with better academic learning among school children. Journal of Nutrition, 136, 1077–1080.

20.

Gassman-Pines

Hill

(2013). How social safety net programs affect family economic well-being, family functioning, and children’s development. Child Development Perspectives, 7(3), 172–181.

21.

Gassman-Pines

Schenck-Fontaine

(2017). Food hardship varies within the SNAP benefit month. Paper presented at the Society for Research in Child Development Biennial Meeting, Austin, TX.

22.

Gennetian

L. A.

Seshadri

Hess

Winn

George

(2016). Supplemental Nutrition Assistance Program (SNAP) benefit cycles and student disciplinary infractions. Social Service Review, 90, 403–433.

23.

Gennetian

L. A.

Wolf

Hill

H. D.

Morris

P. A.

(2015). Intrayear household income dynamics and adolescent school behavior. Demography, 52, 455–483.

24.

Gershoff

E. T.

Aber

J. L.

Raver

C. C.

Lennon

M. C.

(2007). Income is not enough: Incorporating material hardship into models of income associations with parenting and child development. Child Development, 78, 70–95.

25.

Gibson-Davis

C. M.

Ananat

E. O.

Gassman-Pines

(2016). Midpregnancy marriage and divorce: Why the death of shotgun marriage has been greatly exaggerated. Demography, 53(6), 1693–1715.

26.

Gómez-Pinilla

(2008). Brain foods: The effects of nutrients on brain function. Nature Reviews Neuroscience, 9, 568–578.

27.

Hastings

Washington

(2010). The first of the month effect: Consumer behavior and store responses. American Economic Journal: Economic Policy, 2, 142–162.

28.

Hoynes

H. W.

Schanzenbach

D. W.

(2009). Consumption responses to in-kind transfers: Evidence from the introduction of the Food Stamp Program. American Economic Journal: Applied Economics, 1, 109–139.

29.

Hoynes

H. W.

Schanzenbach

D. W.

(2014). U.S. food and nutrition programs. In Means-tested transfer programs in the United States. Cambridge, MA: National Bureau of Economic Research.

30.

Hoynes

Schanzenbach

D. W.

Almond

(in press). Childhood exposure to the Food Stamp Program: Long-run health and economic outcomes. American Economic Review.

31.

Kharmats

A. Y.

Jones-Smith

J. C.

Cheah

Y. S.

Budd

Flamm

Cuccia

… Gittelsohn

(2014). Relation between the Supplemental Nutritional Assistance Program cycle and dietary quality in low-income African Americans in Baltimore, Maryland. American Journal of Clinical Nutrition, 99, 1006–1014.

32.

Ladd

H. F.

Muschkin

C. G.

Dodge

K. A.

(2014). From birth to school: Early childhood initiatives and third-grade outcomes in North Carolina. Journal of Policy Analysis and Management, 33(1), 162–187.

33.

Lee

B. J.

Mackey-Bilaver

(2007). Effects of WIC and Food Stamp Program participation on child outcomes. Children and Youth Services Review, 29, 501–517.

34.

Ludvigsen

Scott

(2009). Real kids don’t eat quiche: What food means to children. Food, Culture & Society, 12, 417–436.

35.

Michelmore

Dynarski

(2016). The gap within the gap: Using longitudinal data to understand income differences in student achievement (NBER Working Paper No. w22474). Cambridge, MA: National Bureau of Economic Research.

36.

Morris

P.A.

Hill

H. D.

Gennetian

L. A.

Rodrigues

Tubbs

(2014). Income volatility in U.S. households with children: Another growing disparity between the rich and the poor (Discussion Paper 1429-15). Madison, WI: Institute for Research on Poverty.

37.

Muschkin

C. G.

Ladd

H. F.

Dodge

K. A.

(2015). Impact of North Carolina’s early childhood initiatives on special education placements in third grade. Educational Evaluation and Policy Analysis, 37, 478–500.

38.

Oliver

M. L.

Shapiro

T. M.

(2006). Black wealth, White wealth: A new perspective on racial inequality (2nd ed.). New York: Taylor & Francis.

39.

Pianta

R. C.

Egeland

(1994). Predictors of instability in children’s mental test performance at 24, 48, and 96 months. Intelligence, 18, 145–163.

40.

Raley

Bianchi

(2006). Sons, daughters, and family processes: Does gender of children matter? Annual Review of Sociology, 32, 401–421.

41.

Reardon

S. F.

(2011). The widening academic achievement gap between the rich and the poor: New evidence and possible explanations. In Duncan

G. J.

Murnane

R. J.

(Eds.), Whither opportunity? Rising inequality, schools, and children’s life chances (pp. 91–116). New York: Russell Sage Foundation.

42.

Schenck-Fontaine

Gassman-Pines

Hill

(2017). Use of informal safety nets during the Supplemental Nutrition Assistance Program benefit cycle: How poor families cope with within-month economic instability. Social Service Review, 91, 456–487.

43.

Seefeldt

K. S.

Castelli

(2009). Low-income women’s experience with food programs, food spending, and food-related hardship: Evidence from qualitative data (Contractor and Cooperator Report). Washington, DC: U.S. Department of Agriculture.

44.

Seligman

H. K.

Bolger

A. F.

Guzman

Lopez

Bibbins-Domingo

(2014). Exhaustion of food budgets at month’s end and hospital admissions for hypoglycemia. Health Affairs, 33, 116–123.

45.

Shapiro

J. M.

(2005). Is there a daily discount rate? Evidence from the food stamp nutrition cycle. Journal of Public Economics, 89, 303–325.

46.

Todd

J. E.

(2015). Revisiting the Supplemental Nutrition Assistance Program cycle of food intake: Investigating heterogeneity, diet quality, and a large boost in benefit amounts. Applied Economic Perspectives and Policy, 37, 437–458.

47.

U.S. Census Bureau. (2013). Table 8: Enrollment status for families with children 5 to 24 years old, by control of school, race, type of family, and family income. School Current Population Survey October 2013. Retrieved from http://www.census.gov/hhes/school/data/cps/2013/tables.html

48.

Wiig

Smith

(2009). The art of grocery shopping on a food stamp budget: Factors influencing the food choices of low-income women as they try to make ends meet. Public Health Nutrition, 12, 1726–1739.

49.

Wilde

P. E.

Ranney

C. K.

(2000). The monthly food stamp cycle: Shopping frequency and food intake decisions in an endogenous switching regression framework. American Journal of Agricultural Economics, 82, 200–213.