+1234567890
1, My Address, My Street, New York City, NY, USA
Junk Food junk food banned in schools articles in Schools and Childhood Obesity
Junk Food junk food banned in schools articles in Schools and Childhood Obesity
We find that junk food availability has small positive associations with BMI and obesity in basic OLS models that only control for a limited set of covariates, but those associations become insignificant when controls for BMI at school entry and state fixed effects are added. Our IV models, which address potential bias in the OLS models, generate somewhat larger, albeit less precise, point estimates that are also not statistically significant. Even if the IV point estimates were statistically significant, they would still represent only minor increases in BMI and obesity, generally one-third of one percent. Moreover, reduced form estimates, which are more precisely estimated than IV estimates, provide further support because combined school attendance has no significant effects on 5 th graders’ BMI and obesity. These results are robust to alternative measures of junk food availability and sample restrictions. The models also produce the expected findings on various falsification tests. RAND Corporation, 20 Park Plaza, 7th Floor, Suite 720, Boston, MA 02116, USA, [email protected] , Phone: 1-617-338-2059 x4227 where, BMI iks , denotes fifth grade BMI for child i attending school k located in state s , JF k captures junk food availability in the child’s school, X i and S k are the vectors of individual/family and school characteristics , respectively, and ε iks is the error term. The child’s baseline BMI is included to address potential heterogeneity that can bias OLS estimates such as student demand for junk foods, genetic susceptibility, and sorting. Because junk food availability is collected only in fifth grade, we do not know the length of exposure during prior school years. Therefore, BMI at school entry is the preferred baseline because it is measured prior to any exposure to the school food environment. Finally, since states differ markedly in terms of obesity prevalence in their populations as well as the policy environment geared towards combating obesity, we include state fixed effects to control for state-specific time-invariant unobserved heterogeneity that may be correlated with school food environments and children’s weight. Effects of Junk Food Availability on BMI and Obesity in Fifth Grade with Alternate Sample Restrictions Furthermore, there is very limited evidence on the influence of older peers on food choices. Cullen and Zakeri compared changes in food consumption of 4 th graders who transitioned to middle school in 5 th grade and gained access to school snack bars to changes in food consumption of 5 th graders who were already in middle school. Fourth graders who transitioned to middle school consumed fewer healthy foods compared with the previous school year, but it is not clear whether this was due to the presence of older peers or the change in school food environment. We now turn to our main results, which examine the effects of junk food availability on BMI and other outcomes. We first estimate basic OLS models of BMI and obesity, then augment with state fixed effects and baseline BMI to address omitted variable bias and selection, and finally estimate the IV and reduced form specifications . In Section 5.2, we examine the sensitivity of our results to alternate measures of junk food availability and various sample restrictions. We also report findings from falsification tests. And finally, in Section 5.3, we describe results from ancillary regressions that explore the potential mechanisms underlying our BMI findings. In particular, we examine in-school and total consumption of selected foods and beverages and the availability of and participation in physical activity. Effects of Junk Food Availability on Physical Education, Recess and Physical Activity in Fifth Grade Notes: N=9,380. Robust standard errors clustered at school level are shown in brackets. Other covariates in the model include male, age , male*age, race/ethnicity, kindergarten BMI, mother’s education, income, private school dummy, categories for percent minority in school and school enrollment, and state and urbanicity dummies. First stage results are shown in Table 2 . To provide a sense of the caloric contribution of these purchases, we multiplied the increase in the probability of purchase from attending a combined school by the median number of times that food was purchased among children who purchased at least once, by the number of the calories per unit. 28 Summing across the three junk food groups yields 50 calories per week from in-school junk food purchases. The caloric contribution of in-school purchases is much higher among children who purchase these foods . But even the 62 calories per day represents less than a quarter of the daily discretionary calorie allowance for a moderately active fifth grader. 29 33 “Schools expect budget cuts as economy sours: State problems, decline in property values eat away at district funds”. Available at: snbc.msn.com/id/23116409/ . Junk food availability is a prominent issue for middle and high schools in the U.S. However, there is also widespread legislation and regulation targeting junk foods even in elementary school . Young children’s access to junk foods in school is an important concern due to the strong correlation between childhood overweight and obesity in adolescence and adulthood . In this paper, we examined whether junk food availability increased BMI and obesity among a national sample of 5th graders. Those 5th graders who attend a combined school are much more likely to have junk food availability relative to those in elementary school. While estimates from naïve models that only control for a limited set of covariates suggest a positive association between junk foods in school and BMI and obesity, fully-specified OLS models that control for BMI at school entry and state fixed-effects demonstrate no statistically or economically significant relationships among these young children. Likewise, the IV and reduced form models, which are not subject to the potential bias undermining OLS models, confirm the null findings for BMI and obesity. These results are not sensitive to various robustness checks including alternate measures of junk food availability and sample restrictions. To address these issues, we estimate instrumental variables and reduced form regressions using grade span as the instrument: whether the 5 th grader attends a combined school with older peers. 20 The IV point estimates are relatively larger than the OLS estimates, but less precisely estimated rendering them statistically insignificant . 21 , 22 IV estimates from models that do not control for state and urbanicity dummies and baseline BMI are much larger than those in our preferred specification , although they are not statistically significantly different from each other. Even if the IV point estimates in our preferred specification were significant, they would represent only small increases in BMI and obesity of less than one-third of one percent. Hausman tests that check for the endogeneity of junk food availability by comparing estimates from the fully-specified OLS regression with the IV cannot reject the null hypothesis that both estimates are consistent. Therefore, we also report the reduced form estimates of BMI and obesity regressed directly on our instrument . The coefficients on the instrument are close to zero and very precisely estimated, which further confirm the null findings. Given concerns about unobserved heterogeneity in the OLS specifications and the larger standard errors in the IV specifications, the reduced form estimates are preferred.
Effect of Grade-Span on Academic and Social-Behavioral Outcomes Among Schools Without Junk Food Availability in Fifth Grade Our findings may have implications in the current economic environment. Half of the states are projecting budget shortfalls that threaten staffing, compensation, extracurricular activities, and policy initiatives such as mandated limits on class size. 33 Many schools subsidize their funding with revenue from the sale of junk foods. In total, elementary schools earn approximately $442 million annually from junk food sales . In light of our findings, certain policy measures, such as outright bans on junk food sales , might appear premature given that they remove a key source of discretionary funds. OLS and reduced form estimates for school- and parent-reported physical activity measures are reported in Table 12 . OLS estimates show no relationship between junk food availability and minutes per week of physical education at school, minutes per week of recess at school, and parent-reported participation in physical activity . The reduced form regressions show no significant effects of combined school attendance on minutes per week of physical education instruction. Children attending combined school have fewer minutes of recess , but slightly higher days of parent-reported physical activity though neither finding is statistically significant at.conventional levels. Overall, the regressions do not provide consistent evidence that increased energy expenditure explains the null finding for BMI and obesity. 12 We rely mainly on the first measure of junk food availability because it is the most specific with respect to the quality of foods and because school-level policies regarding junk food availability are frequently set by school principals and staff . We prefer this measure over the simple dichotomy of having any competitive food outlets because the outlet-based measure does not differentiate the type of foods sold . We also prefer it over the child-report because children who do not consume junk foods are less likely to accurately report availability and because children reported only the availability of any sweets, salty snacks, or sweetened beverages, but did not differentiate specific items . 15 However, Clark and Loheac estimate how substance use behavior of students within the same school who are one year older influences adolescent substance use and find a positive relationship. It is possible that children substitute in-school purchases for snacks brought from home or eaten at home either due to satiation or parental monitoring. With our simple dietary recall measures, we cannot explicitly test the nature of potential substitution. We can, however, examine the total intake of soda and fast food consumed in and out of school. Soda is of particular interest because it is the only item for which children were asked about both their in-school and total consumption separately. Fast food, on the other hand, does not correspond exactly to the in-school snack food consumption categories. We find that junk food availability is not associated with significant increases in children’s total consumption of soda or fast foods . 30 The OLS regressions show negative, though generally insignificant, estimates. 31 More importantly, the reduced form estimates confirm that there is no relationship between combined school attendance and total consumption of soda and fast food. The fact that children who consume soda and other junk food in schools show no evidence of an increase in total consumption provides support for the substitution hypothesis. This finding is also consistent with the literature, which indicates that only 27 percent of soda and sweetened drinks consumed in elementary schools are bought at school compared to 67 percent brought from home . Notes: Each estimate represents a separate regression. All models include the full set of covariates. Robust standard errors clustered at school level are shown in brackets. Sales of competitive foods have the potential to generate significant revenues for schools. During 2005–2006, middle and high schools earned an average of $10,850 and $15,233, respectively, from a la carte sales alone . In addition, nearly a third of high schools and middle schools earned between $1,000–$9,999 during that same year from vending machines, another ten percent earned between $10,000–$50,000, and a small number earned in excess of $50,000 per year. These revenues may in turn be supplemented by on-site school stores and pouring contracts with beverage companies. While availability and revenues were less common in elementary schools, nearly half of elementary schools had pouring rights contracts, and competitive food sales from fundraising activities were also common. Several studies have examined peer effects on academic, social-behavioral and substance use outcomes by leveraging variation in school grade span . Most studies compare students in the same grade who attend middle versus combined schools or middle versus elementary schools . 15 These studies generally find that 6 th or 7 th graders who attend middle school fare poorly compared to those who attend elementary or combined schools. 16 However, we are not aware of any studies that compare children in the same grade level who attend elementary versus combined schools . The exception is Rickles , whose findings suggest inconsistent effects of elementary versus combined schools attendance on achievement. 10 The “potatoes” category excluded French fries, fried potatoes, and potato chips. 27 We dichotomize the in-school purchase variables and estimate linear probability models since much of the variation in junk food purchases at school occurs on the extensive margin. Our main results focus on whether the availability of junk foods increases BMI and obesity among 5 th graders . Columns 1 and 4 in Panel A show the results of basic OLS regressions of log BMI and obesity, respectively, on junk food availability controlling for child, household, and school characteristics. 19 These regressions yield a statistically significant increase in both BMI and obesity when junk food is available, although the point estimates are small. The inclusion of state fixed effects and urbanicity dummies and then baseline BMI measured in kindergarten eliminates the significant coefficients. The fully-specified OLS models have very small, precisely estimated, and statistically insignificant point estimates. 32 Given the limitations of the ECLS-K’s consumption variables, we again examined the SNDA-III data and found no evidence that combined school attendance increases total caloric intake. 3 Also, using the ECLS-K, Fernandes found small positive associations between soda availability in schools and both in-school and overall soda consumption of fifth graders. 9 To validate the ECLS-K estimates, we examined the Third School Nutrition and Dietary Assessment Study , which collected 24-hour dietary recall from 2,300 children attending a nationally representative sample of public schools in 2005. Similar to the ECLS-K, eighty percent of elementary school children reported no competitive food purchases. Among children who made a purchase, the median daily caloric intake from these foods was 185 calories. The SNDA estimate is higher than our ECLS-K estimates because it includes healthy foods purchased from competitive food venues: for example, milk was by far the most popular item purchased from competitive food venues and yogurt also ranked highly. 25 Hausman tests cannot reject the consistency of fully-specified OLS estimates in any of our sensitivity checks. Overall, the literature suggests that the presence of older peers may adversely affect academic and social behavioral outcomes, but there is less evidence to support effects on their eating behaviors. Nevertheless, if such an effect exists, the potential bias in our IV estimates due to peer effects is likely to be upward. That is, 5 th graders might emulate older peers who are more likely to consume junk foods in school and would therefore tend to be overweight, independent of the school food environment. In that case, an insignificant finding is unlikely to be undermined. First Stage Regression Estimates of Junk Food Availability in Fifth Grade This research was funded by grants from the Robert Wood Johnson Foundation’s Healthy Eating Research Program, NIH R01 HD057193, the Bing Center for Health Economics at RAND, and the RAND Labor and Population Program. All opinions are those of the authors and do not represent opinions of the funding agencies. The absence of any effects of junk food availability on BMI despite the in-school purchases of junk food also raises questions regarding potential compensatory changes in the availability of and participation in physical activity. For example, revenues from junk food sales may be used to fund playgrounds or pay for physical education instructors. Or it may be that combined schools simply offer more opportunities for physical activity due to their scale and organization relative to elementary schools. Another possibility is that parents or children may increase children’s physical activity to balance junk food intake. If physical activity is greater, then we may find no change in BMI or obesity despite an increase in caloric intake. Note: N=9,380. Robust standard errors clustered at school level are shown in brackets. RAND Corporation, 1776 Main Street, P.O. Box 2138, Santa Monica, CA 90407, USA, [email protected] , Phone: 1-310-393-0411 x7367, Fax: 1-310-260-8161 Detailed information on junk food availability in schools was collected from the school administrators and from children in the fifth grade. School administrators were asked whether students could purchase 17 individual food and beverage items, either from vending machines, school store, canteen, snack bar or a la carte items from the cafeteria during school hours. From these responses, we constructed an indicator variable of junk food availability in school that equals 1 if the administrator reports that students can purchase food and beverage items containing high sodium and/or sugar, including candy, chocolate, baked foods , salty snacks , ice cream or frozen yogurt, or sweetened beverages during school hours, and zero otherwise. 11 Based on these school administrator reports, approximately 61 percent of the children had junk food availability in school. For robustness checks, we also considered two alternative measures of availability. The first is based on whether the modal child at each school reports that foods containing sugar, salty snacks, or sweetened beverages can be purchased at school. Based on this measure, about 75 percent of the children had junk foods available. And the second is based on whether the administrator reports any of the following competitive food outlets operate in the school: vending machines, school stores, canteens, snack bars, and a la carte lines. About 60 percent of the sample had at least one competitive food outlet. 12 Our paper adds to the existing literature in its sample, methodology and scope. First, to our knowledge, ours is the only study that addresses the endogeneity of the school food environment among younger children. The focus on fifth graders is useful because junk food regulations are increasingly targeting elementary and middle schools. 5 And our national sample of children provides a larger and more representative sample with significant variation in school environments. Second, our data contain actual measurements of children’s height and weight, unlike the self-reports from other national datasets that have been used to examine this question previously. Third, our approach improves on the common cross-sectional designs by controlling for children’s BMI at school entry and state fixed-effects, and leveraging variation in schools’ grade spans to estimate IV models. Finally, unlike previous studies, we also provide evidence on the underlying mechanisms by examining effects on food consumption and physical activity. 28 The median number of times an item is purchased in school among children who purchase at least once is 1.5 times . We assume that salty snacks add 140 calories , sweets add 200 calories , and soda adds 150 calories. Given the limitations of the consumption data in the ECLS-K, we caution the reader to treat these caloric intake calculations as approximations. We conducted a number of sensitivity analyses to test the robustness of our findings. In this section, we report results from a few key analyses and then turn to falsification tests. 23 These analyses control for the full set of covariates, including state and urbanicity dummies and baseline BMI. 2 Kubik et al find that a la carte availability in school is negatively associated with overall intake of fruits and vegetables and positively associated with total and saturated fat intake among 7 th graders attending 16 Minneapolis-St Paul schools. Using the same data, Kubik et al show that using competitive foods as rewards and incentives is positively associated with BMI. In-School and Total Food Consumption in Fifth Grade Effect of Junk Food Availability on In-School Junk Food Purchases in Fifth Grade The only effort to address endogeneity is in Anderson and Butcher , who use national data on adolescents aged 14–20 years to examine whether various school food policies influence BMI . In the absence of a single data source containing information on school food policies and BMI among adolescents, the authors use a two-sample IV approach that employs county, state, and regional characteristics as instruments to capture budgetary pressures on schools. They find that a 10 percentage point increase in the proportion of schools in the county that offer junk foods leads to a 1 percent increase in BMI. But this effect is primarily driven by adolescents with an overweight parent, which the authors interpret as a measure of family susceptibility. 4 Their IV approach constitutes an innovation over the literature, but the authors acknowledge that their results may be undermined by a weak first stage. 14 This literature examines peer effects on a wide range of outcomes including substance use , crime , teenage pregnancy , discipline , academic achievement , adolescent food choices and weight . Our sample consists of a single cohort of 5 th graders attending schools with a variety of grade spans. Given that junk food availability is significantly higher in middle and high schools compared to elementary schools, a potentially useful instrument for junk food availability is whether the 5 th grader attends a combined school or whether the 5 th grader is in an elementary school . Our instrument considers only this dichotomy of school type: elementary versus combined. Over 70 percent of our sample attends elementary schools while the remainder attends combined schools usually with grade spans of K-8, K-12 and 5–8 . Effect of Attending a Combined School on Kindergarten Outcomes 16 One exception is Eisenberg who finds that 7 th and 8 th graders who attend schools with older peers are no more likely to use substances relative to those who attend schools with younger peers. 11 The questionnaire separately asked about availability of high- and low-fat options for baked foods, salty snacks, and ice cream/frozen yogurt/sherbert. We include both the low- and high-fat options in our measure, however, in sensitivity analyses, we used only the high-fat versions to construct our school-administrator based measure of junk food availability and found results to be similar. 31 Negative binomial models with a binary treatment variable to account for the count-data distribution of the total consumption variable and the binary nature of junk food availability produced qualitatively similar results. . Notes: Each estimate represents a separate regression. All models control for the full set of covariates. Robust standard errors clustered at school level are shown in brackets. Ashlesha Datar, RAND Corporation, 1776 Main Street, P.O. Box 2138, Santa Monica, CA 90407, USA, [email protected] , Phone: 1-310-393-0411 x7367, Fax: 1-310-260-8161. Another potential concern with this identification strategy is that variation in grade span exposes children to older peers who may influence obesogenic behaviors. Peers, defined broadly, have been shown to influence a wide range of adolescent behaviors and outcomes. 14 However, of particular relevance to our identification strategy is the literature examining a specific type of peer effect, namely, the effect of exposure to older peers due to school grade span. 19 In all models, we estimate robust standard errors clustered at the school level. Effects of Alternate Measures of Junk Food Availability on BMI and Obesity in Fifth Grade Descriptive Statistics in the Fifth Grade The consistent lack of significant findings for BMI and obesity raises questions regarding how the energy balance equation is affected by junk food availability. While we cannot measure children’s energy intake and expenditure explicitly with these data, we can examine whether junk food availability influences general food consumption patterns and physical activity. Unlike BMI and obesity, the consumption and physical activity measures are based on parents’ and children’s reports . As a result, they are subject to measurement error and consequently produce noisier estimates particularly for the IV models. Nevertheless, they represent our best opportunity for understanding important mechanisms underlying our null finding. Therefore, for the in-school junk food purchases, total consumption, and physical activity analyses, we focus mainly on the reduced form results . 26 We also report results from the reduced form, which regresses BMI or obesity directly on the instrument . These results have the advantage of being unbiased and providing evidence of whether a causal relationship exists in the regression of interest. 13 The U.S. Department of Agriculture’s regulations on competitive foods in schools had been comprehensive, but in 1983, a successful lawsuit by the National Soft Drink Association limited the scope of these regulations to food service areas during meal hours . In recent years, several states, districts, and schools have enacted competitive food policies that are more restrictive than federal regulations. And, between 2003 and 2005, approximately 200 pieces of legislation were introduced in US state legislatures to establish nutritional standards in schools or to address the availability or quality of competitive foods . At the federal level, legislation was passed in 2004 requiring local education agencies to develop a “wellness policy” by 2006 that included nutrition guidelines for all of the foods available in schools. More recently, there has been debate in the US Congress over enacting an amendment to the farm bill that would further restrict the sale of unhealthy foods and beverages in schools . At the local level, two of the largest school districts in the nation, New York City Public School District and Los Angeles Unified School District, imposed a ban on soda vending in schools in 2003 and 2004, respectively. 7 Obesity is defined as BMI greater than the 95 th percentile for age and gender on the Center for Disease Control growth charts. Notes: All models include the full set of covariates. Robust standard errors clustered at school level are shown in brackets. Hausman tests for consistency of OLS estimates could not be rejected in any case. The tests are not reported in the table. Unfortunately, we cannot completely separate out these possible explanations because the ECLS-K does not provide us with full information about the daily dietary intake of each child. However, we do have information about in-school purchases of foods with sugar, salty snacks, and sweetened beverages for those children with in-school availability. We also have total consumption of soda, fast food, and a variety of healthy foods for all children in the sample. While not definitive, we can use this information to gain some insight into underlying eating behaviors and lend support for our BMI and obesity findings. Effect of Junk Food Availability on Total Consumption of Selected Unhealthy and Healthy Foods in Fifth Grade Means by Attendance in Elementary Versus Combined School and by Private/Public While we acknowledge their limitations, ancillary analyses of children’s in-school junk food purchases, total consumption of healthy and unhealthy foods, and physical activity are consistent with our null findings for BMI and obesity. Our estimates suggest that the caloric contributions of in-school junk food purchases are likely to be small. Moreover, we find evidence consistent with substitution between in- and out-of-school consumption. Specifically, the total amount of soda and fast food consumed in- and out-of-school, is not significantly higher among those children with greater exposure to junk food in school . And, finally, we find little support for the notion that children substitute calories from healthy foods or increase their physical activity to compensate for increased junk food intake. The child food consumption questionnaire asked about the frequency of overall consumption of specific food items during the past week. Children were asked to include foods they ate at home, at school, at restaurants, or anywhere else. We examine the consumption of two unhealthy items - soda and fast food, and six healthy food items – milk, green salad, potatoes 10 , carrots, other vegetables, junk food banned in schools articles and fruits. The percentage of children not consuming any soda or fast food during the previous week was 16 and 29 percent, respectively, with modal responses at 1 to 3 times per week . Among the healthy foods, green salad, carrots and potatoes were consumed most infrequently with nearly half of children reporting no consumption during the past week. The modal responses for the other healthy foods were 1 to 3 times during the past week. Equation 2.1 represents the first-stage regression where junk food availability is regressed on the combined school attendance instrument , individual and school characteristics, baseline BMI , and state fixed effects . Equation 2.2 represents the second stage where children’s BMI is regressed on the predicted availability of junk foods from the first stage in addition to the common covariates. Nancy Nicosia, RAND Corporation, 20 Park Plaza, 7th Floor, Suite 720, Boston, MA 02116, USA, [email protected] , Phone: 1-617-338-2059 x4227. 23 We also conducted additional sensitivity analyses not reported here. First, given that we do not know the exposure to junk food in previous grades and given concerns that genetic susceptibility may not have a constant proportional effect on BMI at every point in the life cycle, we controlled for 1 st or 3 rd grade BMI instead of BMI in Kindergarten and obtained similar results. Second, inclusion of controls for school meal participation did not change our findings. Third, we used BMI z-scores as the dependent variable to accurately control for age and gender influences on BMI and obtained qualitatively similar results. Fourth, we estimated quantile regressions to test whether the effects of junk food availability varied across the BMI distribution, but found no evidence for heterogeneous effects. Finally, we also re-estimated our BMI and obesity models separately for each gender. The results for junk food availability mirrored those for the full sample. The OLS, IV, and RF models show no significant effects of junk food availability for either boys or girls. Still we may be concerned about differential peer effects, for example, if girls are influenced by older peers’ concerns about body image, which would bias our IV estimates downward. Restricting the sample to those boys and girls attending schools without junk food availability, the coefficients from the reduced form were nearly identical to those based on the full sample of boys and girls, which suggests that peer effects are not an issue even when regressions are gender-specific. While BMI is a widely-used outcome measure, it does not capture nutritional changes. Just because children are not gaining weight does not mean that their diets are not adversely affected by junk food availability. If children are consuming junk food in lieu of healthy foods, there may still be concerns about their nutrition. Columns 3 through 8 of Table 11 examine whether children with in-school availability of junk foods consume less milk, green salad, carrots, potatoes, other vegetables, and fruit. The OLS results show no significant associations with junk food availability. Moreover, reduced form regressions also show that combined school attendance does not significantly impact total consumption of the healthy foods. 32 Effect of Combined School Attendance on BMI, Obesity and Related Behaviors Without Junk Food Availability in Fifth Grade For the sensitivity analyses, we first re-estimate our BMI and obesity regressions with the two alternate measures of junk food availability . Both the child-reported measure of junk food availability and the school-administrator reported measure of competitive food outlet show no effect of junk food availability on BMI or obesity. Next, we re-estimate the models with the exclusion of three particular groups that might confound our instrument . First, because combined schools are much more likely to be private, our instruments may simply capture variation across public versus private schools students, even though the regressions control for private school attendance. We re-estimate the models on a sample that excludes children who attend private schools and find no effects on BMI and obesity. 24 Second, even though Section 4.2.2 suggests there are no peer effects on BMI and related behaviors, we test the sensitivity of our results to exclusion of the oldest peers , but still find no evidence of an effect on BMI and obesity . Finally, children who switch schools for unobservable reasons potentially related to junk food availability may bias our estimates, but estimates from models that exclude children who changed schools between kindergarten and fifth grade confirm no effects . The point estimates from the OLS, IV and reduced form regressions for these sensitivity checks are essentially zero, though less precisely estimated in the IV models. 25 20 In alternate analyses, we used continuous measures of the highest and lowest grades in the school as instruments. In these over-identified models, both instruments had a strong positive association with junk food availability . This approach yielded qualitatively similar results as the exactly-identified models . Identification in our IV models relies on the assumption that, conditional on state and urbanicity dummies, the school’s grade span does not influence BMI except through differences in the availability of junk foods. Districts typically determine the grade span at the time of the schools’ opening based on a number of factors including transportation costs, length of bus ride, desired number of transitions, population size, site availability, preferred school size, and likelihood of parental involvement rather than children’s health outcomes. Changes in grade span over time are possible, but infrequent and similarly-motivated. For example, in our ECLS-K sample, less than 4 percent of the children who remained in the same school between kindergarten and fifth grade experienced a grade-span change from combined to elementary school or vice-versa. While unlikely, it is nevertheless possible that schools may change grade span in response to children’s physical size. Therefore, below we report results from several tests that support the validity of our instruments. These analyses are based on our preferred specification, which controls for the full set of covariates, including state and urbanicity dummies and baseline BMI. 22 A concern with our IV specification estimated via two-stage least-squares is that our first stage models do not account for the dichotomous nature of the treatment variable . Estimates from binary treatment effect IV models confirm that the effects of junk food availability on BMI are neither substantive nor significant . For combined school attendance to be a valid instrument, it must be the case that the school’s grade span has no direct effect on children’s weight except through the junk food environment. One potential concern is that there may be unobserved factors that are correlated with both the likelihood of combined school attendance as well as BMI. For example, it is well known that states differ markedly in the prevalence of childhood obesity. But, states are also likely to differ in terms of factors that contribute to school grade span such as: the size of the school-age population, its distribution within the state, differences in the educational systems and policies, as well as education budgets. Similarly, school grade span can vary across urban versus rural areas , with the latter more likely to have combined schools largely because of a smaller school-age population. The inclusion of state and urbanicity dummies in our regressions controls for unobserved differences across states and across rural/urban areas that may be correlated with combined school attendance and BMI. However, the coefficients from these OLS models may be biased if junk food availability is related to unobserved determinants of children’s BMI. For example, districts with a large population of students at risk for obesity may adopt more stringent nutritional policies that reduce the availability of junk foods in school. In such situations, OLS regressions may show no significant relationship or even a negative relationship between junk food availability and BMI. OLS estimates might also suffer from attenuation bias due to the presence of measurement error in the junk food availability measures. A distinct advantage of the ECLS-K is that it collected height and weight measurements from children at kindergarten entry and in the spring of kindergarten and first, third, and fifth grades. Measurements are superior to self- or parent-reported height and weight data that may introduce non-random measurement error. These measurements are used to compute BMI, defined as weight in kilograms divided by height in meters squared. The average BMI in our sample during the fifth grade is 20.4 . Approximately 20% of the ECLS-K sample is categorized as obese – this is nearly identical to prevalence rates among 6–11 years olds from the 2007–8 National Health and Nutrition Examination Survey . 7 30 The total consumption variables are not dichotomized because there is sufficient variation on the intensive margin. Notes: N=9380. Each estimate represents a separate regression. Dependent variable captures the number of times the food or beverage item was consumed during the last 7 days. All models include the full set of covariates. Robust standard errors clustered at school level are shown in brackets. The prevalence of childhood obesity in the US is at an all-time high with nearly one-third of all children and adolescents now considered overweight or obese . Considerable attention has been focused on schools in an attempt to identify policy levers that will help reverse the obesity epidemic. In particular, the availability of “competitive foods”, defined as foods and beverages available or sold in schools outside of the school lunch and breakfast programs, has been a much debated issue. On the one hand, opponents question the nutritional value of competitive foods and consider them the primary source of “junk foods” in schools. Indeed, the available evidence suggests that these foods are higher in fat compared with foods sold as part of the school meal programs . On the other hand, supporters argue that revenues from these food sales provide much-needed funding for schools, especially in times of budgetary pressures . The ECLS-K is a panel dataset on a nationally representative cohort of kindergarteners in the U.S. who entered school in fall 1998. In the fall and spring of kindergarten and the spring of the first, third, and fifth grades, the study collected information from the children and their parents, teachers, and schools on children′s cognitive, social, emotional, physical development , and their home, classroom, and school environments. One limitation is that the information on the school food environment and children’s food consumption was collected only in the fifth grade. Our analysis sample includes the approximately 9,380 children attending the fifth grade in public and private schools in the 2003–04 school year. 6 In this section, we describe the key variables for our analyses. 26 Although not shown, the IV estimates are easily calculated by dividing the reduced form estimates in Table 10 – Table 12 by 0.2 . The IV coefficients are never significant in part due to the larger standard errors in the regressions of reported eating behaviors and physical activity. 8 Sweets include candy, ice cream, cookies, brownies or other sweets; salty snack foods include potato chips, corn chips, Cheetos, pretzels, popcorn, crackers or other salty snacks, and sweetened beverages include soda pop, sports drinks or fruit drinks that are not 100 percent juice. Effects of Junk Food Availability on BMI and Obesity in Fifth Grade Notes: Figures in brackets are robust standard errors clustered at the school level. Other covariates in the model include male, age , male*age, race/ethnicity, kindergarten BMI, mother’s education, income, private school dummy, categories for percent minority in school and school enrollment, and state and urbanicity dummies. Despite limited empirical evidence, there is growing concern that junk food availability in schools has contributed to the childhood obesity epidemic. In this paper, we estimate the effects of junk food availability on BMI, obesity, and related outcomes among a national sample of fifth-graders. Unlike previous studies, we address the endogeneity of the school food environment by controlling for children’s BMI at school entry and estimating instrumental variables regressions that leverage variation in the school’s grade span. Our main finding is that junk food availability does not significantly increase BMI or obesity among this fifth grade cohort despite the increased likelihood of in-school junk food purchases. The results are robust to alternate measures of junk food availability including school administrator reports of sales during school hours, school administrator reports of competitive food outlets, and children’s reports of junk food availability. Moreover, the absence of any effects on overall food consumption and physical activity further support the null findings for BMI and obesity. 17 We also examined unadjusted differences in children’s individual, family and school characteristics during the 5 th grade . There were slight differences for some of the covariates. However, there was no overall pattern in the socioeconomic factors that would threaten the validity of the IV approach: that is, some differences imply better BMI outcomes for one group and others worse. For example, in our sample, elementary school students are more likely to be Hispanic and Asian while combined school students are more likely to be white. There are no differences in the share that are Black. Similarly, there is no consistent pattern in maternal education. Elementary school students are more likely to have poorly and highly educated mothers . Notes: N=9,380. Means are unweighted. Standard deviation in parentheses. Second, since our instrument leverages across school variation we might be concerned that selection into different schools might undermine the validity of our instrument. To test for differential selection into combined versus elementary schools, we regress BMI, obesity, test scores, social-behavioral outcomes, and parental involvement measured in kindergarten on combined school attendance in 5 th grade . 17 Because these outcomes are determined prior to exposure to school, these comparisons allow us to test for selection. The results suggest that, conditional on observed characteristics, combined school attendance is uncorrelated with pre-exposure BMI, obesity, test scores, social-behavioral outcomes and parental involvement. As falsification tests, we examined whether junk food availability in the fifth grade influenced children’s height in the fifth grade and their pre-exposure BMI. Height should clearly be unrelated. And indeed, the coefficients are essentially zero and insignificant . Because BMI and obesity in kindergarten is measured prior to exposure to junk foods in school, any effects would suggest unobserved heterogeneity. The OLS, IV and reduced form point estimates are close to zero and the reduced form specifications also show no relationship . Results for BMI and obesity measured in first and third grade likewise confirm insignificant effects of junk food availability during fifth grade . However, because our data do not contain information on junk food availability prior to 5 th grade, these results are also consistent with the absence of junk foods in earlier grades. 24 Estimates based only on the sample of private schools yield small and statistically insignificant effects of competitive food availability on BMI in both OLS and IV specifications, although the F-statistics for the instrument in the first stage were smaller . Try out PMC Labs and tell us what you think. Learn More . The food consumption questionnaire collected information on in-school junk food purchase during the fifth grade. These questions asked children about their purchases of sweets, salty snack foods, and sweetened beverages during the previous week. 8 A substantial majority of the children did not purchase junk food in school during the reference week: 77% for sweets, 84% for salty snacks, and 88% for soda . But a large share of these children did not have junk food available in their schools . Conditional on availability, about half the sample purchased any of these unhealthy foods at least once a week in school. Among those who did purchase, the modal response was 1 to 2 purchases per week: 68 percent for sweets, 72 percent for salty snacks, and 70 percent for soda. 9 8600 Rockville Pike , Bethesda MD , 20894 USA Notes: N=9,380. Percentages are unweighted. Figures in the top panel are not conditional on availability in school. Notes: N=9380. Each estimate represents a separate regression. Dependent variables in columns – are dichotomous and capture whether any purchase of that item was made in school during the last week. All regressions include the full set of covariates. Robust standard errors clustered at school level are shown in brackets. 18 To check whether these null findings are merely due to lack of power instead of absence of peer effects, we estimated the same models using social-behavioral outcomes and test scores as dependent variables because the literature finds evidence of peer effects on these outcomes. We were able to identify statistically significant peer effects on social-behavioral outcomes , which suggests that lack of power is an unlikely explanation for the finding of null peer effects on BMI and related outcomes. The debate draws from largely cross-sectional research that rarely addresses the potential endogeneity of the school food environment. Our paper advances the literature by attempting to isolate the causal effect of junk food availability on children’s food consumption and BMI. We use longitudinal data on BMI for a national sample of fifth graders from the Early Childhood Longitudinal Study – Kindergarten Class and an instrumental variables approach that leverages the well-documented fact that junk foods are significantly more prevalent in middle and high schools relative to elementary schools . Plausibly exogenous variation in junk food availability across a cohort of fifth graders is identified using the grade structure in their schools. We argue that a fifth grader attending a combined or middle school is more likely to be exposed to junk foods compared to a fifth grader in an elementary school , but that the school’s grade span has no direct effect on a child’s weight. First-stage regressions confirm that combined school attendance is a strong predictor of junk food availability. Further tests for instrument validity including an examination of sorting and peer effects support our use of the instrument. Not surprisingly, our analysis of in-school consumption of junk foods does confirm that children purchase junk food when it is available. 27 The OLS estimates show a significant relationship for purchases of all types of junk food when junk foods are available in schools . And the reduced form estimates show that children in combined schools are between 5 and 9 percentage points more likely to purchase junk foods compared to those in elementary schools Table 10 , Panel B). The remainder of this paper is organized as follows. We first discuss junk food availability in schools and the findings from the existing literature in Section 2. Section 3 describes our data and relevant analysis variables. In Section 4, we describe our empirical strategy, which leverages longitudinal information on BMI and implements an instrumental variables approach to identify the causal impact of junk food availability. In Section 5, we first discuss our main results for children’s BMI and obesity and then support these findings with robustness checks and falsification tests. We also present supporting evidence from models of in-school purchases of junk food, total consumption of various healthy and unhealthy items, and physical activity. Finally, Section 6 concludes with the policy implications of our findings. Overall, the instrument appears to be strongly predictive of junk food availability and there is no evidence that selection or peer effects threaten its validity. 21 The IV regressions were also estimated without baseline BMI. The point estimates, first-stage F-statistics, and Hausman tests yield similar results . The relationship between junk food availability and children’s BMI in fifth grade can be estimated cross-sectionally using the following linear regression model. One potential explanation for our null findings for BMI and obesity may be that availability does not impact overall food consumption. This may happen for several different reasons. First, young children may not purchase significant amounts of junk food in school either due to limited access to such foods or fewer discretionary resources to purchase them. Second, children may not change their total consumption of junk food because junk food purchased in school simply substitutes for junk food brought from home. Or third, children may not change their overall consumption during the day, but simply substitute between junk food consumed in-school and out-of-school. 13 The value of reduced form regressions has been highlighted by Angrist and Krueger and, more recently, Chernozhukov and Hansen formally show that the test for instrument irrelevance in the reduced form regression can be viewed as a weak-instrument-robust test of the hypothesis that the coefficient on the endogenous variable in the structural equation is zero. Despite the growing support for competitive food regulation, it is hard to deny opponents’ claims that the evidence against competitive foods is limited. Existing research does show that competitive food availability is associated with a decline in nutritional quality of meals consumed at school . 1 However, less is known about the effects on overall diet quality and children’s weight. The literature does provide some evidence of substitution of caloric intake across meals and locations among adults , but the evidence is less clear regarding children for whom parental oversight can also play a role. Only Kubik and colleagues have examined 24 hour dietary recall and BMI among children, however these studies are based on small cross-sectional samples and do not address the potential endogeneity of the school food environment. 2 , 3 29 Discretionary calories are the difference between an individual’s total energy requirement and the energy necessary to meet nutrient requirements. According to Dietary Guidelines for Americans, the discretionary allowance for a 2000 calorie diet is 267 calories. See: s/dga2005/document/html/chapter2.htm#table3 accessed August 22, 2008. 6 All sample sizes have been rounded to the nearest 10 per the ECLS-K’s restricted-use data agreement. We address the potential endogeneity of junk food availability using instrumental variables. Specifically, we estimate the model in Equations and using Two-Stage Least Squares. Finally, we provide further support for the null findings by examining in-school and overall food consumption patterns as well as physical activity. The null effects on BMI and obesity cannot be explained entirely by limited access or limited discretionary resources among young children because 5 th graders do purchase junk food when it is available in schools. However, our results suggest that the caloric contribution of in-school purchases is likely to be small. Moreover, we find no evidence of significant changes in the overall frequency of consumption of soda and fast food, which is consistent with children substituting in-school purchases of junk food for that taken from or eaten at home. Alternative explanations such as compensatory changes children’s consumption of healthy foods and in their opportunities for and participation in physical activity do not appear to play a significant role in explaining our null findings for BMI and obesity. Junk Food junk food banned in schools articles in Schools and Childhood Obesity
Junk Food junk food banned in schools articles in Schools and Childhood Obesity
Notes: Each estimate represents a separate regression. Other covariates in the models include age, male, age*male, race/ethnicity, kindergarten BMI , mother’s education, income, private school dummy, categories for percent minority in school and school enrollment, and state and urbanicity dummies. Robust standard errors clustered at school level are shown in brackets. For reading, math, self control, and interpersonal skills, higher skills indicate better outcomes. For externalizing and internalizing behavior problems, higher scores indicate worse outcomes. Parent involvement is measured as the sum of the number of times/week that the parent engages in 9 activities with the child . First, we report first-stage estimates of the effect of our instrument – combined school attendance – on junk food availability in school. The first-stage estimates show that combined school attendance significantly increases the likelihood of junk food availability with an F-statistic on the instrument that exceeds 22 . Effects of Junk Food Availability on BMI and Obesity in Kindergarten, First, and Third Grade 1 Other studies have examined the effects of price reductions, increases in availability, and promotion of low-fat foods in secondary schools on sales and purchases of these foods as well as their consumption within experimental settings and found positive effects. Third, another concern is that combined school attendance might generate peer effects on BMI, obesity, food consumption and physical activity, independent of junk food availability. We test for the presence of peer effects by regressing these outcomes on combined school attendance using only the sample of schools that do not offer junk foods . The results do not provide any support for peer effects on BMI, obesity, food consumption or physical activity. 18 Notes: Each estimate represents a separate regression. All models include the full set of covariates as well as the baseline measure of the dependent variable. Robust standard errors clustered at school level are shown in brackets. 4 Their results for the other school policies, pouring rights contracts, and food and beverage advertisements are smaller and less precise. The parameter of interest in Equation is β 1 . Obtaining an unbiased estimate of β 1 is challenging because the school food environment is not exogenous to the outcomes of interest. Schools that serve high-fat, energy-dense junk foods may differ on many observable and unobservable factors that are correlated with children’s weight and dietary behavior. In particular, the decision to offer junk foods in schools may be influenced by a variety of factors including budgetary pressures, demands of the student population, parental involvement, and state/district policies. These factors could independently influence children’s weight as well. For example, budgetary pressures may induce schools or districts to scale back or eliminate physical education programs, which might increase children’s weight. As a result, coefficient estimates from the ordinary least squares estimation of Equation 1 would be biased. While our results are robust, we caution that we could not consider the full range of consequences of junk food availability. Not only are the dietary intake measures in the ECLS-K limited, but we are also not able to examine whether related health outcomes such as diet quality or dental caries are influenced by junk food availability. Also, we are unable to examine the generalizability of our findings to older children who may have greater junk food access and intake both in and outside school. And finally, we could not consider whether exclusive contracts between schools and beverage/snack companies influence students’ food choices in the longer run through product or brand recognition. Additional research is necessary to fully understand the potential consequences before costly legislation is implemented. Such research might also consider the consequences of junk food regulations on school finances and the extent to which these financial consequences could be mitigated by the sale of more nutritious alternatives or through alternative financing mechanisms. Competitive foods are sold through a la carte lines, vending machines, school canteens/stores, and fundraisers and, in contrast to the federally-reimbursable school meal programs, are not subject to federal nutritional standards. As a result, competitive foods account for much of the variation in the food environment across schools. Competitive foods are available in a large share of schools, although the availability of these foods varies significantly across elementary, middle, and high schools. For example, as many as 97% of high schools and 82% of middle schools have vending machines compared to only 17% of elementary schools . However, a la carte lines, which are the predominant source of competitive food sales, operate not only in most high and middle schools, but also in a large proportion of elementary schools . Effect of Junk Food Availability in School on Height in Fifth Grade 5 For example, California’s first nutrition policy implemented beverage standards for elementary and middle schools, not high schools. Notes: N=9380. avoid junk food article