Influence of participation in the national school lunch program and food insecurity on child well-being

This study examines two research questions: the child- and family-specific factors that predict food insecurity and participation in the National School Lunch Program (NSLP) and the effects on school-age children of food insecurity and participating in the NSLP. Results show that factors representing families' economic status are significantly associated with food insecurity and that a broader range of cultural and attitudinal factors predicts participation in the NSLP. Food insecurity is associated with children's health and behavior, but not test scores. After addressing selection, participation in the NSLP does not predict improved child outcomes.

The Influences of Participation in the National School Lunch Program and Food Insecurity on Child Rachel Dunifon Cornell University Lori KowaleskiJones University of Utah This study examines two research questions: the child- and family-specific factors that predict food insecurity and participation in the National School Lunch Program (NSLP) and the effects on school-age children of food insecurity and participating in the NSLP. Results show that factors representing families* economic status are significantly associated with food insecurity and that a broader range of cultural and attitudinal factors predicts participation in the NSLP Food insecurity is associated with children's health and behavior, but not test scores. After addressing selection, participation in the NSLP does not predict improved child outcomes. Every day millions of U.S. children rcccivc free or rcduccd-pricc lunchcs through the National School Lunch Program (NSLP). However, few studies explore the links between participating in this program and broad measures of child well-being, particularly when also considering children's food-sccurity status. Understanding the ways in which both food insecurity and participation in the NSLP may affcct children in­Social Service Review (March 2003). © 2003 bv The University of Chicago. All rights reserved. 0037-7961 / 2003/7701-0004$ 10.00School Lunches and Well-Being 73 forms the public policy debate on how federal food assistance programs influence children's development. This study uses data from the Child Development Supplement of the Panel Study of Income Dynamics (CDS-PSID) to examine two research questions. First, what are the child- and family-specific factors that pre­dict food insecurity and participation in the NSLP? Second, what are the associations between both food insecurity and participating in the NSLP on the social and cognitive development of school-age children? The National School Lunch Program The NSLP is a federally sponsored nutrition program serving approx­imately 26 million children each day, with estimated expenditures of $5.8 billion in 1998 (Oliveira 1999). Recent estimates indicate that the NSLP is available to 92 percent of U.S. students and that 56 percent of these students participate in the program (Burghardt and Devaney 1995). The aim of the program is to provide nutritious foods to school- age children at no, or reduced, cost. Eligibility for free NSLP lunches is limited to families whose incomes are at or below 135 percent of the poverty line. Reduced-price lunches are available to families whose in­comes are between 135 and 185 percent of the poverty line.1 The NSLP also subsidizes full-price lunches in most schools, so virtually all school children may benefit from the NSLP (Devaney, Ellwood, and Love 1997). Approximately 47 percent of all school lunches under this program are served to children whose family incomes are less than 185 percent of the poverty level. Lunches provided by this program are expected to enable students to consume at least one-third of the Recommended Daily Allowance of specific nutrients and include core items: meat or meat alternative, two or more vegetables and fruits, whole-grain or en­riched breads, and milk. Many of the available evaluations of the NSLP document the rela­tionship between participating in the program and increased nutrient intake (Burghardt and Devaney 1995). There is some concern that al­though the NSLP is effective in delivering on the promise of achieving Recommended Daily Allowance goals of increasing nutrients, the lunches are likely to be higher in fat and saturated fat than recom­mended by The Dietary Guidelines for Americans (Gleason and Suitor in press). Few studies examine the links between participation in school-based food assistance programs and behavior and cognitive outcomes for chil­dren. One such work finds that the Massachusetts school breakfast pro­gram (SBP) is associated with higher children's test scores and lower levels of school tardiness and absences (Meyers et al. 1989). This study suggests some potential links between food assistance programs and child performance outcomes.274 Social Service Review If participation in the NSLP leads to improved nutritional intake, participating children may also see improvements in their ability to learn and regulate their behavior, as well as their overall health. However, it is possible that NSLP participation may not translate into improved outcomes among children along these domains. First, children may already be receiving adequate nutrients in their diet. Second, the food eaten as part of the NSLP may not be of adequate nutritional value to improve children's outcomes. Finally, even if children participate in the NSLP and increase their nutrient intake as a result, their overall nutrient intake may not improve if their parents compensate for this by decreas­ing the food provided to children in the home. Overall, then, it is unclear whether NSLP participation will be associated with improved well-being among children, suggesting the need for empirical research. Food Insecurity and Child Development Another nutrition-related factor affecting child development is food security. Food security is defined by the U.S. Department of Agriculture (USDA) as "access ... to enough food for an active, healthy life" (U.S. Department of Agriculture 2001), and is assessed with an 18-item survey measured at the household level and capturing experiences (in the last 12 months), such as running out of food, perceptions that the food in a household is of inadequate quality or quantity, and reduced food intake by adults or children, all because of financial constraints. In 1999, 10 percent of U.S. households were food insecure. That is, they affirmed at least three of the 18 items in the food security scale (U.S. Department of Agriculture 2001). Measures of food insecurity are associated with deficits in nutritional consumption. Using a 24-hour diet recall, as well as a survey of house­hold food supplies, Anne Kendall, Christine Olson, and Edward A. Fron- gillo (1996) find that food-insecure households have lower rates of con­sumption of fruits and vegetables and less food on hand than food secure households. This suggests that the USDA measure of food insecurity is associated with actual food intake. Food insecurity is likely to affect children via two key pathways. First, food insecurity in the home may translate into a source of family stress that could affect both parenting behaviors and children's reactions to parenting styles. This stress may have implications for children's be­havior. This expectation is consistent with research showing that eco­nomic hardship is linked to shortfalls in parent-child interactions, which lead to subsequent increases in children's behavior problems (as in McLoyd 1990). Second, food insecurity may have direct effects on chil­dren's health outcomes. Children who are living in homes where the availability of food is threatened are more likely to have limited access to healthful foods that will enhance their nutritional intake.School Lunches and Well-Being 75 To be sure, compared with children in developing nations, relatively few children in the United States actually experience hunger. Mark Nord and Gary Bickel (2001) find that only .9 percent of all U.S. children lived in families experiencing child hunger in 1998. Yet, researchers still find associations between food insecurity and child outcomes. For ex­ample, Katherine Alaimo and colleagues (2001) find that food shortfalls adversely affect children's health outcomes, including reported health status, iron deficiency, and colds. Alaimo and colleagues (20016) also find that poor children are at greater risk for poor health status, as are food-insufficient children, but that the combination of poverty and food insufficiency puts children at even greater risk. This reflects the fact that poverty and food insecurity are different phenomena, although they are highly correlated.® In 2000, the prevalence of food insecurity was 36.8 percent among poor house­holds, compared with 4.6 percent among households with incomes above 1.85 times the poverty line (Nord et al. 2002). Very little research relates the USDA measure of food insecurity to child well-being. However, a larger body of research examines the in­fluence of food insufficiency on child and family adjustment. Food in­sufficiency is assessed in surveys with a single item indicating that a family often or sometimes does not have enough food to eat (National Health and Nutrition Examination Survey 2003). In a series of studies, Alaimo, Olson, and Frongillo (2001; Alaimo et al. 2001) find food in­sufficiency to be associated with adverse outcomes among children, in­cluding lower math scores, greater problems getting along with others, poor health status, and more frequent illness. However, it is important to highlight the distinction between food insecurity and food sufficiency According to Alaimo, Olson, and Fron­gillo (1999, p. 269), "food insecurity consists of 1) depletion of house­hold food stores, 2) eating unsuitable food, 3) worrying about the food supply, and/or 4) acquiring food in socially unacceptable ways, such as begging or scavenging." Food insufficiency, by contrast, refers simply to an insufficient quantity of food and does not address issues such as food quality, worries about food, or ways of obtaining food.4 Because food insecurity measures both the quality and quantity of food, it could affect children even if it does not occur along with child hunger. Children participating in food-assistance programs, such as the NSLP, are at increased risk for living in food-insecure households. In the cur­rent study, 24 percent of the children who participate in the school lunch program live in food-insecure households, compared with 5 per­cent of nonparticipating children (difference significant at 1 percent level). The aim of this study is, first, to evaluate which child- and family- specific factors predict food insecurity and participation in the NSLP and, then, to investigate the main effects of participation in the NSLP76 Social Service Review and of living in a food-insecure household on child health, behavior, and academic development. Sample We study the associations among NSLP participation, food insecurity, and child outcomes using the 1997 CDS-PSID. Since 1968, the PSID has followed and interviewed annually a nationally representative sample of about 5,000 families. Split-off families are followed when children leave home, when couples divorce, and when more complicated changes break families apart. Except for problems of immigration, this proce­dure produces an unbiased population sample each year (Fitzgerald, Gottschalk, and Moffitt 1998). In 1997, the PSID supplemented its core data collection with data on parents and a maximum of two of their children ages 12 or younger, a project known as the CDS. The CDS includes assessments of the cog­nitive, behavioral, and health status of 3,500 children. One of the strengths of the CDS-PSID is its link to multiple years of data on the children's parents. Our analyses focus on children ages 6-12 in 1997, for whom we have measures of participation in the NSLP in 1997. Measurement NSLP Participation Participation in the NSLP is measured with a dichotomous variable asking the primary caregiver (usually the mother) whether the 6-12- year-old child currently receives a free or reduced-price lunch at school. Food Insecurity Food insecurity is measured using the 18-item USDA scale. We use all 18 items in a continuous scale created with a Rasch coding model de­veloped and implemented by the USDA. This is a linear scale that de­pends on the number of affirmative responses to a series of increasingly severe food security items (U.S. Department of Agriculture 2001). The intervals of this scale are meaningful, and an increase in the scale in­dicates increased severity of food insecurity. The appendix describes in more detail how the continuous Rasch scale was coded. Other analyses use a simple indicator of whether a household is food insecure, which is indicated by an affirmative response to at least three of the 18 items in the overall scale (U.S. Department of Agriculture 2001). Finally, we create a measure of "marginal" food insecurity if a family responded affirmatively to one or two of the food-insecurity scale items but did not affirm three or more items.School Lunches and Well-Being 77 Child Outcomes Child well-being is measured in three areas: health limitations, behav­ioral adjustment, and achievement. We measure health limitations with a variable indicating whether the child has any health limitations (re­ported by the parent) that affect participation in childhood activities, school attendance, or the performance of schoolwork. A child with a limitation in any area is given a score of one. The rest are given scores of zero. This measure taps the propensity for a child to be debilitated with illness to the degree that it interferes with normal childhood ac­tivities, which may be an important precursor to success on achievement outcomes. The measure of health limitations is significantly correlated with global measures of child health and number of hospital stays since birth, suggesting reasonable construct validity.5 Children's behavioral adjustment is measured with the Behavior Prob­lems Index (BPI), adapted from the Thomas M. Achenbach and Craig S. Edelbrock behavioral checklist (1981), which measures both exter­nalizing behavior (16 items; examples include bullying other children or destroying things) and internalizing behavior (13 items; examples include moodiness or fearfulness) using maternal reports. A higher score indicates increased behavior problems. We also use a 10-item index of positive behavior, in which mothers report whether a specific behavior (examples include cooperating with others and sharing) is not at all like the child (1), totally like the child (5), or somewhere in between. A higher score indicates more positive behavior. With two measures from the Woodcockjohnson Psycho-Educational Battery-Revised, which were administered to children in the CDS-PSID (Mather 1991), we assessed children's achievement. Math achievement is represented by a combination of scores on applied problems and calculation tests, and reading achievement is represented by a combi­nation of scores on letter-word and passage comprehension tests. Control Measures In all analyses, we control for an extensive set of background charac­teristics of the child's family Each of these controls is theorized to be associated with child outcomes, as well as the measures of household food insecurity and children's participation in the NSLP. Although only one wave of CDS-PSID data is currently available, it is possible to control for characteristics of children's families spanning the years from birth to 1997 by linking children to information on their families in the PSID main data. The control measures capture sociodemographic characteristics of the children and their families. Additionally, they include psychological and attitude measures of the parents in order to address issues of omit­ 78 Social Service Review ted variables that may bias estimates that rely on parental reports of both independent and dependent variables. Several of the control mea­sures are measured over the period between the child's birth and 1997: average family income in 1996 dollars, the percentage of time that a child's parents owned their own home, the percentage of time that a child's family received food stamps, and the percentage of time that a child lived in a married-couple family. Other measures are taken in 1997: number of siblings; maternal and paternal educational attainment; child gender, age, race, and ethnicity; maternal age; health insurance status of the child; whether the primary caregiver smokes; whether the primary caregiver drinks; the primary caregiver's score on an assessment of self-esteem (10 items); the primary caregiver's score on an assessment of personal control (six items); the primary caregiver's score on a mea­sure of depression (10 items); and the primary caregiver's self-report of aggravation in parenting (four items). (Because of missing data on these last six items, as well as on paternal education, all analyses include a missing data dummy for these items.)6 Method First, we examine the child- and family-specific factors that are associated with participation in the NSLP and with the various measures of food insecurity described above. Next, we examine the main effects of food insecurity and NSLP participation on children's well-being in the areas of health, behavioral adjustment, and test scores. Our first set of analyses uses logistic regression and ordinary least squares (OLS) methods to estimate equations (1) and (2), in which various measures of household food insecurity for family / at time t (INSECy,) and NLSP participation (NSLPI/() for child i in family /at time t are functions of the child- and family-specific (Xp) control mea­sures detailed above. Each equation also contains a child-specific error term (e^) and a family-specific error term (efl). INSECy = + [3:, X";/7 + /3 ; Xp + g(y, + gp, (1) NSLP,. = a. ift + (3,Xi7, + ftXy, + eif, + ef,. (2) We next examine the main effects of food insecurity and NSLP par­ticipation on children's well-being in the areas of health, behavioral adjustment, and test scores, as shown in equation (3). Here, using OLS (or logistic regressions when predicting the indicator of health limita­tions), the outcomes of child i in family/at time t (CHILD ,y,) are func­tions of demographic controls at the child- and family level (Xifl and Xp) and an indicator for participation NSLP (NSl.P,,.), as well as the continuous measure of household food insecurity in 1997 (INSECy,).School Lunches and Well-Being 79 CHILD,/( = aift + j8,INSECyj + 0,NSLP^ + P?,Xifl +$4Xfl+eifl+efl. (3) We also address issues of selection that may bias estimates of the effects of NSLP participation on individual outcomes. Despite the extensive set of control variables available in the PSID, we may not be able to account for all of the ways in which families using the NSLP may differ from those who do not, leading to bias in the estimates of equation (3). Specifically, the family-specific error term (eft) may be correlated with the likelihood that a specific child participates in the NSLP and with the measures of child well-being.7 To deal with issues of selection into NSLP participation on the basis of family-level unobservables, we perform sibling comparison analyses in which we compare the outcomes of a sibling who does participate in the NSLP with those of a sibling who does not participate. In our sample of 1,854 children ages 6-12, 574 children lived in families in which there is nonmissing data on all measures for a group of at least two siblings ages 6-12, referred to as the sibling sample. A total of 266 children in the sibling sample lived in families in which at least one child participated in the NSLP, and of these children 32 did not par­ticipate in NSLP themselves. By comparing the outcomes of siblings within the same family, who differ in participation in NSLP, this method controls for the self-selection of families into NSLP participation on the basis of unobservable family-level characteristics. This method controls for all characteristics that are shared between siblings, including meas­urable characteristics such as parental education and unmeasured fac­tors such as parental mental health or motivation. The sibling fixed-effects estimator subtracts the measures from one sibling from those of another, as shown below: (CHILD 1/( - CHILD,,,) = afl + 0, (NSLP1/( - NSLP,/() + Pi(X1fi-Xm) + (e1fi-eifl). (4) Here subscripts 1 and 2 reference siblings 1 and 2 within family f and equation (4) shows that the outcome for sibling 2 is subtracted from that of sibling 1, as are all control variables, including the indicator of NSLP participation. Thus, we are relating differences in NSLP partici­pation between siblings to differences in outcomes, such as behavior problems, controlling for other factors that differ between siblings, such as age or gender. By subtracting one sibling from another, the fixed- effect models net out the effects of family characteristics that are shared between siblings, both observed and unobserved. Because all measures that are shared between siblings are not directly estimated in this model, family-level food insecurity cannot be directly estimated, but is still con­trolled, in the model. Measures that differ between siblings, such as age,80 Social Service Review gender, and the percentage of a child's lifetime living in a married- couple family, are directly estimated in the sibling model. Although this method controls for all family-level unobservable char­acteristics, it is possible that parents of siblings may choose to have one child participate in the NSLP while the other does not on the basis of child-specific characteristics. To test this, we examined whether the like­lihood of split participation is determined by child gender, age, health insurance status, health status, or body weight and height. None of these factors was associated with the likelihood of split participation between siblings. Thus, we do not find evidence that important child-specific factors differ between children who do and do not participate in the NSLP within the same family. Examining differences between siblings can lead to increases in mea­surement error. Zvi Griliches (1979) notes that bias will result when the right-hand variables are measured with error. In the current study, the right-hand variable of interest, NSLP participation, is an objective mea­sure that is likely to be measured with a great deal of precision. By contrast, some of the dependent variables, such as maternal reports of behavior problems, may be measured less precisely. Although the use of a sibling-based model under these circumstances (measurement error in the dependent, but not independent, variable) can produce higher standard errors, it does not impart increased bias to the coefficients of estimation (see Gujarati 1995). Finally, it is important to address concerns about the reduction in sample size in the sibling model. The use of the sibling model reduces the potential for bias resulting from family-based unobservable char­acteristics; thus, results from the sibling model are likely to be less biased than results from the other analyses. The only way that bias could be increased through the use of the sibling model would be if households that split participation do so on the basis of a factor that is not measured and that also affects child behavior. As noted above, this does not appear to be likely. Thus, the sibling model reduces one source of bias, giving us confidence in the results from this model. IS "LlltS Weighted means for all variables are presented in table 1, separately for the full sample and for the sibling comparison sample. In the full sample, 31 percent of children receive a free or reduced-price lunch through the NSLP.8 The mean Rasch scale is less than zero, indicating that the average family did not affirm any items in the food-insecurity scale; families affirming no items in the scale were given a score of -3.4 (see appendix). Overall, 10 percent of the CDS-PSID sample of children ages 6-12 lived in households that would be classified as food insecure by the USDA (a household affirming at least three of the 18 items is con-School Lunches and Well-Being 81 Table 1 Weighted Means of Sample Characteristics Full Sample Sibling Sample N Mean SD N Mean SD Internalizing behavior problems 1,600 16.73 4.68 567 16.93 5.31 Externalizing behavior problems 1,592 22.96 5.85 561 22.79 5.88 Positive behavior 1,616 41.91 5.65 573 41.86 5.37 Whether any health limitations 1,618 .09 .40 574 .05 .25 Math test score 1,297 49.63 16.82 566 49.32 15.87 Reading test score 1,304 57.91 19.70 574 56.79 19.50 Participates in national school lunch program 1,618 .31 .46 574 .32 .47 Recoded Rasch food-insecurity scale 1,618 -2.19 2.90 574 -2.37 2.83 I lousehold is food insecure 1,618 .10 .30 574 .09 .29 I lousehold is marginally food insecure 1,618 .07 .25 574 .05 .21 Lifetime average family income (1996 dollars) 1,618 50,251.28 45,130.58 574 50,774.87 42,700.60 Number of siblings 1,618 1.64 1.09 574 1.97 1.02 Years parents owned home (%) 1,618 .63 .40 574 .64 .41 Maternal education 1,618 13.26 2.18 574 13.29 2.21 Paternal education 1,074 13.65 2.35 396 13.76 2.24 Child is African American 1,618 .20 .40 574 .19 .39 Child is Hispanic 1,618 .03 .17 574 .02 .12 Child is male 1,618 .51 .50 574 .50 .50 Age of child 1,618 8.97 2.01 574 8.89 1.97 Age of primary caregiver 1,618 37.37 6.79 574 36.86 5.78 Child is insured 1,618 .94 .24 574 .96 .19 Years received food stamps (%) 1,618 .17 .30 574 .19 .33 Years in a married-parent fam­ily (%) 1,618 .74 .37 574 .78 .36 Primary caregiver smokes 1,043 .45 .50 428 .47 .50 Caregiver drinks 1,042 .22 .42 429 .24 .43 Caregiver efficacy 999 34.32 4.38 407 34.51 4.36 Caregiver self-esteem 1,036 19.15 2.83 426 19.27 2.91 Caregiver aggravation in parenting 1,040 9.48 2.58 428 9.51 2.59 Caregiver depression 1,017 16.28 5.71 418 16.01 5.43 Note.-NSLP = National School Lunch Program. sidered food insecure [U.S. Department of Agriculture 2001]). An ad­ditional 7 percent of the households are measured to be marginally food insecure. Table 1 also shows few differences between the full sample and the sibling comparison sample. In table 2, we examine the associations between child- and family- specific factors and three different measures of food insecurity (the continuous Rasch food-insecurity scale, the indicator of marginally food insecure, and the food-insecure dummy), as well as the indicator for participation in the NSLP. These analyses allow us to identify whetherTable 3 T5 w 1=1 3 3 U < Ok Ck J £ C * < H 2 p o c o o pH o * I « |d O cj o s tin 1-1 , Z Oi u o £ a h be ,2 u oi o, o &k 2 M ^ R oi iis H vi z ^ 5 O -i o &k !5i-Ht^iocMi-HinrH^oocMcot'3oao^ co ^!5in3DrHt>.l>^!5rHrHj)JO^OO^lOin ,0 CO H H CM 00 i" i" i" H H CM CM CM " i" i" o •t> cr> O cr> JiOO oo cr> j> CD CM O 00 CM cr> j> p CO p p 00 >o i> cr> p p p >q CM >o p p p p p I-H I-H CO CM ' I-H I-H I-H CO >o 00 J> CO CD -f cr> >o I-H >q --J p p 00 p co p p I-H I-H i" I-H I-H i" " l-H I-H i-H O cd CM -f cd cr> I-H 00 ^H yD cd cd cd TjH CO o O O j> p 00 >q CM p p p p p p co t-- p p p p I-H I-H I-H I-H i-H i-H I-H I-H I-H I-H o 00 hOh I-H TjH CM O 00 cd cr> cr> cr> 00 CO p >q p p oqooppo >q p p O CO p p p p I-H I-H ' I-H ' ^ ^ ^ CM I-H I-H I-H I-H I-H o o o o s? iS 5 .a '-* - -I '&.$■ | S's I s -p c« S ^ ‘5? i ^ '"■h ^ ^ 2 » £ - * .a 2i^i^ oc&- isls-Sall^-S ^ 2 O _, < .S ^-< s-I s-I s-I s-I s-I s-I h S'd'd'd^^'d '5c'5c'5b'5b'5b'5b b'ii^^^^DD^bbs-.s-.s-.s-.s-.s-. Ck J W* V < H ^.. O .. T. I j Z ? 3 £ u u u < < u ; £ ; > 0 0 0 0 0 0 S ; = < School Lunches and Well-Being 83 different factors predict marginal food security versus food security, or NSLP participation versus food security, for example. Because our data consist of some children with the same family of origin, Huber-White robust standard errors, clustered by the family of origin (in 1968), are used. Results from table 2 show associations that operate in the expected direction. Family income is significantly and negatively associated with the continuous food-insecurity measure, as is the percentage of time that a child's parents owned their own home, paternal education, and whether the child is insured. The number of years that a child received food stamps and maternal depression are both positively associated with the contin­uous food-insecurity measure. When predicting the food-insecurity in­dicator, family income again is the most significant predictor, along with number of siblings and food stamps receipt. Considering the indicator of marginal food insecurity, results indicate that paternal education is associated with a reduced likelihood of marginal food insecurity. A broader set of family and child measures is predictive of NSLP participation. Family income, paternal education, and whether the care­giver drinks are all significantly and negatively associated with partici­pation. The number of a child's siblings, whether the child is African American, the percentage of time that the child received food stamps, and the percentage of time that the child lived in a married-parent family are all positively and significantly associated with participation in the NSLP. Noteworthy is the large odds ratio on the indicator that a child is African American; African-American children are almost five times more likely to participate in the NSLP than other children. These results suggest that while family income and parental education are the most important predictors of food insecurity, a wider range of measures, including aspects of cultural and behavioral attributes, is predictive of NSLP participation. In table 3, we use the linear food-insecurity scale and participation in NSLP to predict the outcomes of 6-12-year-old children in the full sample. Analyses using the other measures of food insecurity did not differ substantially from those using the continuous measure. All anal­yses control for the covariates described above. The results show that participating in the NSLP is associated with increased externalizing be­havior, an 82 percent increase in the odds of having a health limitation, and lower math test scores. An increase in food insecurity is associated with decreased levels of positive behavior and an 8 percent increase in the odds of health limitations. The other variables in the model operate in the expected direction: male children have more behavior problems, parental education is highly associated with test scores, and parental aggravation in parenting is associated with increased reports of chil­dren's behavior problems, for example. It is unlikely that participating in the NSLP is actually detrimental forTable 3 3 SJ CJ w T5 T5 S 3 .O H o H w o * z R H a; K P Hr^ Z g o 3 h 'I < t> jjjj §> j* K *a s-< o 10 1 t> GO 10 i 0 Th 0 GM Th 3 t> t> 10 GM Th >q GCi 0 t> GCi GCi GO >q 0 J t-4 T-4 T-4 0 r Th T-4 T-4 1 r GT 25 GO 10 10 10 GM iO 2 CD 1-1 iO GM 2 O I> GO 0 GO O O CT) Th Th 0 GM GO GM GO GO O G] CT) O T-4 T-4 T-4 ' t-4 T-4 t-4 T-4 t-4 GM t-4 i-< EC CJ Z O o Bjj cm cq 5 i-i ['■>■ GM Th GM i-i'-ii c o o tM c z < z o H 2 H ca: < Cm a- J h cq X w H cq z ) t> GM GM i 5 O GO GO iO t> GO Th ■MW^OOOO fi 5b O £ 3 J £ H , 'p ; a - 5 'S .a $ 3 B s •= 3 = s.y a, be •3 81 Z Pi( J ^2s *-< '"O '"O '"O ^ ^ '"O - «■*$ zj ID & i! 5 .S 2 i±3 i±3 be P­" 5 ^ ^ s 2 *3 S o s ^ "5b'5b'5b'5b'5b'5b *2 ^ Z^llouS-?'<f U .2 Z CL, J Z§9 H O* |School Lunches and Well-Being 85 children, as suggested in table 3. The pattern of associations between NSLP participation and children's outcomes suggests that omitted var­iables may be biasing the results presented in table 3. Specifically, chil­dren participating in the NSLP may be different in unmeasured ways from those not participating, and these unmeasured variables may be associated with detrimental outcomes observed among children partic­ipating in NSLP. To address this, table 4 presents the results of analyses using the sibling comparison model shown in equation (4). This model compares the outcomes of siblings within the same family, one of whom participates in the NSLP and one of whom does not. (Because of estimation diffi­culties when using a sibling fixed-effect model when predicting dichot- omous outcomes, we use a linear probability model when predicting health limitations rather than a logistic regression.) Here, the NSLP indicator is not a significant predictor of any of the child outcomes. Additionally, in some cases, the sign on the NSLP coefficient is reversed compared with that in table 3. For example, the NSLP coefficients when predicting math and reading scores are negative in table 3 but are positive in table 4; coefficients predicting health limitations are positive in table 3 but negative in table 4.9 However, when predicting external­izing and positive behavior, the coefficients in table 4 operate in the same direction and are actually larger than those in table 3. For these outcomes, the lack of significant association with NSLP participation found in table 4 likelv results from an increase in the standard error, or precision, of the estimates. The remaining variables in the model operate in the expected direction, and their coefficients are consistent with those in table 3. The full-sample results indicate that NSLP participation is associated with detriments in child well-being. However, the sibling comparison results indicate that these results are due to unmeasured family-specific factors that bias the earlier estimates of associations between NSLP par­ticipation and child outcomes, particularly when predicting children's test scores and health limitations. Neither the restricted-sample nor the full-sample results suggests that participation in the NSLP is associated with improvements in child well-being. However, it is possible that the NSLP may benefit subgroups of especially disadvantaged children. To test this, we examine whether the impact of participation in the NSLP on children differs for children living in a single-parent family and for children whose family income-to-needs ratio in 1996 was below the fed­eral poverty line. The impact of participation in the NSLP does not differ for these subgroups (analyses not shown here). Finally, we examine whether parenting practices mediate associations between food insecurity and child well-being. We examine the measure of aggravation in parenting, described above, as well as the Home Ob­servation for Measurement of the Environment Short Form (HOME-o % y O ti O & 05 _l 1> 30 ^030 30 30 ^ •jf •jf ■5fr •jf ■5fr >n cm m m o 30 ^ o 30 ^ I r" 1 m z o to 2 £ s 0 U o ?; H g s o G u ^ 'f\ & r- CM '-£> 30 >n 30 1> 30 05 r- 05 05 cm ^ r- 30 CM 1> ^ ' >n 05 r- •jf •jf •jf o CM 05 o 30 05 O o 05 >-i <£> O 05 i' CM I 30 V I o u h P O ?; o > h 2 p u w 2h -1 2 tn O W U aa- £ M -O J ? o < s& J2 H £ S pq * 2 66 S i»; h CQ X w ^ a 31 H CQ y=3 O O <£> m ^ >n >n c£5 C£5 <£> <£> >-^ 30 >0 >-^ >-^ <£> >n ^ 'S ' 30 >n 05 05 •jf •jf •& 30 '-£} 1> 30 r- 30 o m 05 30 ^ o >-( 30 O CM l' >n CM i' ^ 05 CM CM CM 30 O CM >n <£> O m o r- ' 30 'S id 30 >n C£5 ^ 30 '-£> 05 in CM !£> r- 30 30 CM ^ q ^_' ^_' _!i CM l> 30 CM 1 ^ 1 m 05 30 ^ >n m 05 i> >n >n r- 05 05 30 >n CM 30 •jf CM 05 30 30 ^ CM CM !£> CM 05 O 30 30 05 O r ' ^ o be o 3 J z H cxrs II U '-*-< S. | J'S m ^ Z J o T5 O o b g ix,_-a -a i*h -a "o .3 : = S£aa°aSS. JH ^ ^ ^ 5 ' s- as­s'^ '.u '‘"■J «j s 2 Si 'OS'-' .8 !s .2 £ Oh zgS V V 1*3 " h -®s. 5 -%• S ■5fr ■5frSchool Lunches and Well-Being 87 SF) scalc (Bradley and Caldwell 1984a, 19846). This scale captures two dimensions of children's home environments: the degree of appropriate cognitive stimulation and the degree of maternal warmth directed to­ward the child. We do not find evidence that these parenting measures mediate associations between food insecurity and child well-being (anal­yses not shown here). Discussion Previous research on the impact of the NSLP and food insecurity on child well-being is extremely limited. The results from this analysis pro­vide much needed insight regarding the ways in which a widely used food assistance program and a newly developed measure of food inse­curity may influence the overall well-being of U.S. children. Results indicate that factors representing families' economic status are significantly associated with both the continuous measure of food insecurity and the dummy indicator of insecurity. When predicting mar­ginal insecurity, however, only paternal education is a significant pre­dictor. This suggests that although food insecurity itself is primarily a product of financial constraints, marginal food insecurity may be more attributable to human capital factors. Families affirming some, but fewer than three, of the food-security items could face constraints unrelated to financial factors when struggling to have an adequate amount of food. Additionally, families experiencing marginal food insecurity may benefit from a different set of interventions than families that are food insecure. While food-insecure families could benefit from financial sup­ports, marginally insecure families may benefit more from nutrition education or other initiatives aimed to address barriers to food security that go beyond financial factors. As expected, measures of families' economic status are highly asso­ciated with children's participation in the NSLP. One unique finding is the dramatically increased likelihood of participation in the NSLP among African-American children, even after controlling for a wide range of socioeconomic measures. This may represent a difference in the school-level prevalence of participation in the program among chil­dren in different racial groups. For example, African-American children may attend schools in which a high percentage of children participate in the NSLP, thus reducing the stigma associated with participating compared with that felt by white children and increasing the odds that a specific child is likely to participate. Future work is needed to examine this association further. It is also surprising that living in a married-parent family is associated with increased odds of participating in the NSLP compared with living in a single-parent family. Perhaps, after adjusting for the wide range of socioeconomic factors in this data, married parents have more time or88 Social Service Review energy to devote to learning about programs such as the NSLP. They may also have more information about the program itself, leading to increased participation among their children. Also unusual is the find­ing that the indicator of parental drinking is associated with a decrease in the likelihood of participating in the NSLP. We are not aware of other research examining parental drinking as a predictor of NSLP partici­pation; however, it is possible that parents who drink more than others are less willing or able to enroll their children in the school lunch program. Overall, results from table 2 suggest that a broader range of factors in children's families predicts participation in NSLP, compared with those measures associated with food insecurity. Turning to the influence of food insecurity on child outcomes, results in table 3 suggest that food insecurity is associated with behavioral and health outcomes among children but not with cognitive outcomes. This may be because very few children in this sample experience hunger caused by food insecurity. The 18-item food security scale contains eight items that refer specifically to hunger or lack of food among children. Consistent with coding developed by the USDA (Nord and Bickel 2001), we coded families affirming at least five of these eight items as expe­riencing child hunger. In our sample, only 0.3 percent of the children are in families classified as experiencing child hunger. Rather than leading to child hunger and reduced food intake, which could translate into cognitive deficiencies, food insecurity may instead represent a level of stress in a household that is translated most reliably into children's behavioral adjustment. Indeed, research from develop­ing countries suggests that undernourished children respond to stressful situations with higher levels of two key indicators of stress, cortisol and heart rate, than adequately nourished children (Fernald and Grantham- McGregor 1998). In addition to leading to increased stress among chil­dren, food insecurity can lead to stress among parents. Parents of chil­dren in families where food supplies are tight maybe be under strain, which could affect parent-child interactions or lead to stress in the chil­dren themselves. Research bv Rebekah L. Colev and P. Lindsav Chase- / j j Lansdale (2000) shows that parents' perceptions of financial strain can lead to increased behavior problems among adolescents. In our analyses, food insecurity may represent a specific type of financial strain that is likely to lead to reductions in positive behavior among children. Additionally, food insecurity is associated with our measure of health limitations, which represents limitations in school attendance or daily activities. Again, it is possible that the stress of living in a family expe­riencing food insecurity leads to negative physical outcomes, even if children are not experiencing actual hunger. Turning to the associations of NLSP participation on child well-being, results in table 3 indicate detrimental associations between NSLP par­ticipation and child outcomes. Most of these associations are eliminatedSchool Lunches and Well-Being 89 after adjusting for family-level factors associated with the selection of children into participation in the program. The differences between the whole-sample and sibling-sample results highlight the importance of addressing issues of selection when examining the effects of NSLP participation on child well-being. We are not aware of another study of the NSLP that addresses the selection issue in this way. It is noteworthy that even the sibling comparison results do not suggest that participation in the NSLP is associated with improvements in child outcomes. There are several potential explanations for these findings. It is possible that additional nutrients in a child's diet that result from participation in the NSLP do not lead to overall improvements in child well-being because most children are already receiving an adequate nu­tritional intake. As some researchers note, "above some level, extra nu­trients [may be] superfluous" (Butler and Raymond 1996, p. 782). In addition, the likelihood exists that families may use the NSLP to replace food the child would have eaten anyway, rather than to add to the child's overall diet (this issue is discussed with respect to the Special Supplemental Nutrition Program for Women, Infants, and Children; see Besharov and Germanis 2001). Thus, participation in the NSLP may not have an impact on children because it does not significantly improve their overall nutritional intake. Finally, it is possible that NSLP partic­ipation does not significantly change children's intake of vitamins and minerals, and therefore, does not lead to benefits in the domains mea­sured here. Our data do not allow us to distinguish between these various interpretations. The lack of longitudinal data in this study makes it difficult to draw strong conclusions about the impact of NSLP participation on children. Thus, although we find no evidence of benefits of NSLP participation on child well-being along a variety of domains, it is too early to draw conclusions about the overall effectiveness of this major federal pro­gram. Research examining the impact of NSLP participation on child development is rare, and more work is needed better to understand the impact of this program. Overall, the NSLP represents a major federal policy effort to alleviate the problems associated with hunger and a lack of adequate nutrition among low-income children. Additionally, many children participating in the NSLP are also at risk for food insecurity. Thus, it is important to evaluate the effectiveness of both the NSLP and food insecurity on child well-being. This study takes an important first step in this direction and opens the door for further research in this understudied area. Results presented here show that food insecurity is associated with children's behavioral adjustment, but not with cognitive test scores. Re­sults also show that factors predicting food insecurity differ from those associated with marginal food insecurity, suggesting that marginally in­secure and insecure families may benefit from different types of inter-90 Social Service Review vcntions. Additionally, one of the main contributions of this study is to emphasize the need to address selection effects. We observe substantial differences in the pattern of results once selection issues are addressed in our models. Future evaluations of the NSLP should seriously consider this issue. Appendix In the Rasch scale, a household that has affirmed no items is given a score of zero. However, this is problematic because it assumes that the difference between affirming zero items and affirming one item is the same as any other interval (e.g., the difference between affirming five items and affirming six). To address this, we follow a procedure sug­gested by researchers at the USDA and used by others working in this area (Mark Nord, personal e-mail communication, November 7, 2001). We regressed the Rasch score (with those affirming no items given a score of zero) on all of our control measures, as well as a dummy indicator for a Rasch score of zero. We then calculated, from the co­efficient on the dummy variable, the "correct" score for those house­holds who affirm no items, which was -3.4. We gave the households affirming no items a score of -3.4 instead of zero and use this contin­uous measure of food insecurity in our analyses. References Achenbach, Thomas M., and Craig S. Edelbrock. 1981. "Behavioral Problems and Com­petencies Reported by Parents of Normal and Disturbed Children Aged Four through Sixteen." Monograph of (he Society for Research in Child Development, Serial 188, vol. 46, no. 1. Ann Arbor, Mich.: Society for Research on Child Development. Alaimo, Katherine, Christine Olson, and Edward Frongillo. 1999. "The Importance of Cognitive Testing for Survey Items: An Example from Food Security Questionnaires." journal of Nutrition Education 31 (5): 269-75. ----------. 2001. "Food Insufficiency and American School-Aged Children's Cognitive, Aca­demic and Psychosocial Development." Pediatrics 108 (1): 44-53. Alaimo, Katherine, Christine Olson, Edward Frongillo, and Ronette Briefel. 2001. "Food Insufficiency, Poverty, and Health in U.S. Pre-school and School-Age Children." Amer­ican journal of Public Health 91 (5): 781-86. Besharov, DouglasJ., and Peter Germanis. 2001. Rethinking W1C: An Evaluation of the Women, Infants, and Children Program. Washington, D.C.: AEI. Bradley, Robert 11., and Bettye M. Caldwell. 1984a. "The HOME Inventory and Family Demographics." Developmental Psychology 20 (2): 315-20. ----------. 19846. "The Relation of Infants' Home Environments to Achievement Test Per­formance in First Grade: A Follow Up Study." Child Development 55 (3): 803-9. Burghardt, John, and Barbara Devaney. 1995. "School Nutrition Dietary Assessment Study." American journal of Clinical Nutrition, suppl., 61 (1): 178-81. Butler, J. S., and Jennifer Raymond. 1996. "The Effect of the Food Stamp Program on Nutrient Intake." Economic Inquiry 34 (4): 781-98. Centers for Disease Control and Prevention. "National Health and Nutrition Examination Survey." 2003. Family Questionnaire Codebook. Centers for Disease Control and Pre­vention, Atlanta. Available at http://www.cdc.gov/nchs/data/nhanes/fq-fs.pdf.School Lunches and Well-Being 91 Coley, Rebekah L., and P. Lindsay Chase-Lansdale. 2000. "Welfare Receipt, Financial Strain, and African-American Adolescent Functioning." Social Service Review 74 (3): 380-404. Devaney, Barbara, Marilyn R. Ellwood, and John M. Love. 1997. "Programs That Mitigate the Effects of Poverty on Children." Future of Children 7 (2): 88-112. Fernald, Lia C., and Sally M. Grantham-McGregor. 1998. "Stress Response in School-Age Children Who Have Been Growth Retarded since Early Childhood." American Journal of Clinical Nutrition 68 (3): 691-98. Fitzgerald, John, Peter Gottschalk, and Robert Moffitt. 1998. "An Analysis of the Impact of Sample Attrition on the Second Generation of Respondents in the Michigan Panel Study of Income Dynamics ."Journal of Human Resources 33 (2): 300-343. Gleason, Phillip M., and Carol W. Suitor. In press. "Eating at School: How the National School Lunch Program Affects Children's Diets." American Journal of Agricultural Economics. Griliches, Zvi. 1979. "Sibling Models and Data in Econometrics: Beginnings of a Survey." Journal of Political Economy, suppl., 87 (5): S37-S64. Gujarati, Damodar. 1995. Basic Econometrics. New York: McGraw-Hill. Kendall, Anne, Christine Olson, and Edward A. Frongillo. 1996. "Relationship of Hunger and Food Insecurity to Food Availability and Consumption." Journal of the American Dietetic Association 96 (10): 1019-24. Mather, Nancy. 1991. An Instructional Guide to the Woodcock-Johnson Psycho-Educational Bat­tery- Revised. New York: Wiley. McLoyd, Vonnie. 1990. "The Impact of Economic Hardship on Black Families and Chil­dren: Psychological Distress, Parenting, and Socioemotional Development." Child De­velopment 61 (2): 311-46. Meyers, Alan F., Amy E. Sampson, Michael Weitzman, Beatrice L. Rogers, and I lerb Kayne. 1989. "School Breakfast Program and School Performance." American Journal of Diseases of Children 143 (10): 1234-39. Nord, Mark, and Gary Bickel. 2001. "Estimating the Prevalence of Children's I lunger from the Current Population Survey Food Security Supplement." Paper presented at the Second Food Security Research Conference, USD A Economic Research Service, Au­gust, on-line. Available at http://www.ers.usda.gov/publications/fanrrll-2/, last ac­cessed January 2003. Nord, Mark, Nader Kabbani, Laura Tiehen, Margaret Andrews, Gary Bickel, and Steven Carlson. 2002. "Household Food Security in the United States, 2000." ERS Food Assistance and Nutrition Research Report no. 21. U.S. Department of Agriculture Economic Research Service, Washington, D.C. Oliveira, Victor. 1999. "Food-Assistance Expenditures Fall for Second Year." Food Review 22 (1): 38-44. U.S. Department of Agriculture. 2001. "Guide to Measuring Household Food Security." Available at http://www.ers.usda.gov/briefmg/foodsecurity/surveytools/index.htm. Last accessed January 2003. U.S. Department of Agriculture. 2003a. "Nutrition Program Facts: The National School Lunch Program." Food and Nutrition Research Service, Washington, D.C. Available at http://www.fns.usda.gov/cnd/Lunch/AboutLunch/NSLPFacts02.pdf. U.S. Department of Agriculture. 2003&. "Nutrition Program Facts: The School Breakfast Program." Food and Nutrition Research Service, Washington, D.C. Available at http: //www.fns.usda.gov/cnd/Breakfast/AboutBFast/BFastFacts.pdf. Notes We gratefully acknowledge support from the USD A Small Grants program, administered through the Institute for Research on Poverty at the University of Wisconsin. We also appreciate helpful comments from Judi Bartfeld; J. S. Butler; John Cawley; Pinka Chatterji; Mary Corcoran; Sheldon Danziger; Edward Frongillo; Craig Gunderson; Nader Kabbani; Mark Nord; John Karl Scholz; Ken Smith; Scott Winship; Cathleen Zick; seminar partic­ipants in the Department of Policy Analysis and Management at Cornell University, Ithaca, N.Y., the USD A Economic Research Service Small Grants Conference, Washington, D.C.,92 Social Service Review and the 2001 Annual Meetings of the Association of Policy Analysis and Management, Washington, D.C.; and the anonymous reviewers at Social Service Review. 1. In 1997, the year of our study, the federal poverty line was $16,276 for a family of two adults and two children. 2. The SBP is operationally different from the NSLP. The NSLP served 27 million as compared with the 8 million children served by SBP in fiscal year 2001 (USDA 2003a, 20036). The SBP is also perceived as more directly targeted to low-income children, po­tentially introducing stigma that may prevent many children from participating and many schools from implementing a breakfast program, thus making comparisons between the two programs difficult. 3. In our data, the correlation between poverty and food insecurity is .19, which is significant at the 1 percent level. 4. In our data, the correlation between the 1997 food-insecurity dummy and the 1996 food-insufficiency dummy is .37. 5. In our data, the correlation between health limitations and children's global health rating is .18, and the correlation between health limitations and number of hospital stays is .24. Both of these correlations are statistically significant. 6. Because of concerns about the potential endogeneity of these measures of parental attitudes and behaviors, we ran all analyses omitting the measures of smoking, drinking, self-esteem, personal control, depression, and aggravation in parenting. The removal of these measures did not change the associations between NSLP participation and child outcomes. 7. It is also likely that food-insecure families differ from other families in unobservable ways; however, because food insecurity is measured at the household level, the data do not allow us to correct for this in the way described below. 8. Based on census reports and USDA program use data, the participation rates for NSLP participation were 31 percent in 1997. This number was derived from dividing the 15.1 million children who received free or reduced price lunches in 1997 (http:// www.fns.usda.gov/pd/slsummar.htm) by the approximately 48 million children who were enrolled in first grade through the end of high school as of October 1997 (http:// www.census.gov/prod/99pubs/p20-516u.pdf). This figure is quite comparable with our reported rate of participation among our sample members of 38 percent. 9. In order to confirm that the sibling analyses themselves cause the NSLP coefficients to become insignificant, rather than simply the use of a different selected sample, we ran OLS analyses using the sibling comparison sample relating NSLP participation to all mea­sures of child outcomes. These OLS analyses, like our full-sample results, indicate detri­mental associations between NSLP participation and child well-being. Thus, we are assured that it is the reduction of selection bias through the sibling fixed-efleets method that produces the results in table 4, rather than simply the use of a more selective sample.