*I am grateful to Patricia Anderson, Andriana Bellou, Gordon Dahl ...

Timing is Everything: The Impact of School Lunch Length on Children’s Body Weight
Rachana Bhatt*
Georgia State University
December 2012
The large number of overweight children in the U.S. has prompted school administrators and
policy makers to identify practices in schools which contribute to unhealthy weight outcomes for
children and develop strategies to prevent further increases. Advocates for school nutrition
reform have suggested that it is important for children to have an adequate amount of time to eat
meals in school in order to maintain a healthy weight. This paper examines whether the length of
time children are given to eat lunch in school has an impact on their weight. We find evidence
that an increase in lunch length reduces the probability a child is overweight, and this finding is
robust across various econometric specifications, including a two-sample IV and
difference-in-differences model that account for the potential endogeneity of lunch length.
JEL Classification: I10, I20
Keywords: childhood nutrition, school lunches
*I am grateful to Patricia Anderson, Andriana Bellou, Gordon Dahl, Inas Rashad Kelly, James
Marton, Uta Schönberg, Rusty Tchernis, participants in the 2010 CeMent workshop, seminar
participants at Georgia State University, the University of Rochester, and various conferences for
their suggestions and input. I would like to thank Ted Macaluso from the United States
Department of Agriculture, and Anne Gordon from Mathematica Policy Research for their help
with the School Nutrition Dietary Assessment-III data. I would also like to thank Lisa Whittle
from the Center for Disease Control and various state Youth Risk Behavior Surveillance System
coordinators for granting permission to access micro data from their states. All errors are my
own. Please contact the author with comments and suggestions at rbhatt@gsu.edu.

1. Introduction
The connection between the food children eat in school and their body weight has been the
topic of much discussion among public health agencies (U.S. Department of Health and Human
Services, 2001; Child Nutrition and WIC Reauthorization Act, 2004). 1 Children eat close to 180
lunch meals in school each year, in addition to snacks, and for some students, breakfast.
Consequently, emphasis has been placed on identifying practices in schools that contribute to
children’s excess weight, and to develop strategies to reduce further gains. Past research has
examined the link between children’s weight and the availability of “junk” food in schools,
school provided lunches, and participation in the National School Lunch Program (NSLP) and
School Breakfast Program (SBP) (Anderson and Butcher, 2006; Schanzenbach, 2009; Gundersen
et. al, 2012; and Millimet et al., 2010).
One aspect of the eating environment at school that has received attention among advocates
for school nutrition reform, but less so in research is the length of time children are given to eat
lunch. For instance, both the National Association of State Boards of Education (NASBE), and
the School Nutrition Association (SNA) have recommended that schools provide students with at
least 20 minutes to eat lunch once they are seated (NASBE, 2000; SNA, 2005). 2
Two natural questions arise in light of this recommendation. First, what is the typical amount
of time students are given to eat lunch, and is this lower than what is recommended? Second,
how might lunch length impact weight? Figures 1 & 2 display the distribution of lunch lengths in
schools using two data sets, the School Nutrition Dietary Assessment-III (SNDA-III) and the
1 The U.S. Department of Health and Human Services developed the Healthy People Initiative which highlights
public health goals to be achieved by 2010 and 2020. Targets include increasing access to healthier food in schools,
reducing access to sweetened beverages, and reducing the proportion of overweight and obese children. The Child
Nutrition and WIC Reauthorization Act of 2004 requires all education agencies which participate in the National
School Lunch Program to develop wellness policies outlining goals for nutrition promotion and education, physical
activity, and other school-based activities that promote student wellness by the start of the 2006 school year.
2 These organizations also recommend a minimum length of time be allocated for children to eat breakfast.
2

School Health Policies and Programs Study (SHPPS), respectively. Lunch length in the
SNDA-III reflects the entire length of a scheduled lunch period (i.e. inclusive of travel time
to/from the cafeteria and time spent in service to obtain food). Although the average lunch length
is 32.7 minutes, the amount of time available for eating could be substantially less: A report by
the National Food Service Management Institute indicates that in order for children to have
twenty minutes of eating and socializing, an additional three to eight minutes is necessary for
service and cleaning, and four minutes for travel (Conklin et al., 2002). Figure 2 provides insight
on how much time is actually available for eating using the SHPPS. Here lunch length is
measured as the amount of time children have to eat lunch once they are seated with food. The
average is 22.7 minutes, but there are a non-trivial amount of schools with less than this amount,
suggesting that at least some students receive less than the recommended amount of time to eat.
There are a few ways that lunch length may affect weight. First, short lunch periods may lead
students to skip lunch or eat less than is nutritionally necessary and this can result in overeating
later. Second, short lunches may affect students’ food choices and lead them to substitute away
from healthy food items to less healthy alternatives that are quicker to eat. Finally, there is a
medical literature which suggests that there is a delay between eating and satiation (Rolfes et al.,
2009; Hayes, 2009), and as a result, eating at a slow pace can lead to lower calorie intake relative
to eating at a fast pace (Andrade et al., 2008; Martin et al., 2007). If children that have short
lunches eat quickly, this could translate to greater calorie consumption during lunch.
This paper examines whether the length of time children are assigned to eat lunch in school
has an impact on their Body Mass Index (BMI). OLS estimates indicate a negative relationship
between lunch length and BMI, however these estimates may be biased due to the potential
endogeneity of school lunch length. To overcome this we estimate a two-sample IV model,
3

where the instrument is each state’s law regulating lunch lengths in school. We find a ten minute
increase in predicted lunch length is predicted to reduce BMI by 1.2 percent and the probability
of being overweight by 2.5 percentage points. These findings are upheld in a difference-
in-differences model where we estimate the reduced form effect of the state laws on BMI.
To understand the mechanism(s) behind the BMI and lunch length relationship, we analyze
calorie intake data. We find that students with shorter lunches consume fewer calories during
lunch, but more during dinner, compared to their counterparts with longer lunches. This suggests
the negative relationship between BMI and lunch length may be due to longer lunches shifting
consumption toward times of the day when there are more opportunities to expend calories.
The remainder of this paper is organized as follows: Section 2 describes the current literature
on schools and children's nutrition, and the potential links between meal length and weight.
Sections 3 and 4 discuss the data and OLS estimates. In Section 5, the TSIV analysis and results
are described, and Section 6 provides a discussion. Section 7 concludes.
2. Background & Related Literature
2.1 School Meals & Children's Body Weight
This paper contributes to the literature on eating at school and children’s weight.
Schanzenbach (2009) finds that students in Kindergarten and first grade who eat school lunches
are more likely to be obese than their classmates who bring their lunches from home due to the
extra calories consumed by the former. Anderson and Butcher (2006) examine the sale of junk
foods in schools and find that increased access results in higher BMI. Gundersen et al. (2012)
find that children who receive free/reduced price lunches have higher self-reported health
compared to eligible children that do not participate. Millimet et al. (2010) find that children who
receive free/reduced price breakfasts (lunches) are less (more) likely to be obese compared to all
4

other students. Finally, a series of studies examine the effect of other factors in school such as
physical education, the structure of the school day/year, and recess on children’s weight (Cawely
et al., 2007; Frisvold and Lumeng, 2011; von Hippel et al., 2007; Getlinger et al., 1996).
To the best of our knowledge, only one other study, by Bergman et al. (2003), examines the
relationship between lunch length and consumption. The authors compare food intake among
children that receive 20 minutes for lunch versus 30 minutes and find that plate waste is higher at
schools with the shorter lunch but the nutritional content of food consumed by students in these
schools is higher. This suggests that longer lunches may lead to better food choices, although one
caveat is that this study was based on a sample size of four schools.
2.2 Meal Length & Weight: Potential Mechanisms
Meal length and food consumption can be related in a variety of ways. First, meal length may
affect the food choices children make. Short lunches may encourage students to substitute
towards quicker, less healthy foods, such as choosing apple sauce (194 calories if sweetened)
over a raw apple (81 calories). These foods can be brought from home or bought at school. In the
SNDA-III eighty percent of students purchased a la carte items from the cafeteria, fifteen percent
from vending machines, and three percent from school stores. In addition, short lunches may
affect purchases of school meals since this requires spending time in the service line, and the
nutritional content of these meals could have an impact on weigh (Schanzenbach, 2009). 3
A second way that meal length may affect consumption is if short lunches lead children to
skip lunch or eat less than is nutritionally necessary because they don’t have time. Prior research
3 See www.nutrientfacts.com. The nutritional quality of food in vending machines has been found to be low (Center
for Science in the Public Interest, 2004). Anderson et al. (2003) find that older students have greater access to
vending machines and this is also true in the SNDA-III: Ninety-nine percent of high school students have vending
machines at school, compared to sixty-five (thirty-four) percent of middle (elementary) school students.
5

has found that kids who skip meals are more likely to be obese due to overeating at other meals
or because they engage in unhealthy snacking (Deshmukh-Taskar, 2010; Leidy; 2011; Lafay et
al. 2001). Thus, longer lunches could reduce BMI by shifting consumption towards times of the
day when there are more opportunities to expend calories (i.e. after lunch, in the afternoon),
rather than later in the day when there are fewer opportunities (i.e. after dinner). 4
The third link is based on the physiology of appetite. When an individual eats, signals are
sent from the stomach to the brain to indicate that the individual is getting full and should stop
eating. These signals take ten to twenty minutes to reach the brain once eating starts (Rolfes et
al., 2009; Hayes and Landan, 2009), and it is under this setting that individuals are recommended
to eat slowly, so that they do not overeat. 5 If children with short lunches eat quicker, then these
students may consume more calories than students who eat slower (by virtue of having a long
lunch). No direct evidence exists on whether students eat quicker when they have a short time to
eat, although this seems plausible. For instance, in the SNDA-III, a higher proportion of students
with shorter lunches report being dissatisfied with their lunch length: 32% of students with less
than 30 minutes report dissatisfaction, compared to 25% of students with more than 40 minutes.
Finally, it is important to note that while the mechanisms described above provide motivation
for why longer lunches may lead to lower bodyweight, it is possible that long lunches actually
increase children’s weight. For instance, long lunches may simply give children too much time to
eat, so that for instance, they purchase extra food at school and consume more food than if they
were time-constrained. As will be discussed in Section 4, we find no evidence of this.
2.3 Meal Length & Weight: Potential Biases
4 Skipping meals can lead to delayed insulin response, thus increasing the risk of diabetes (Carlson et al., 2007).
5 This has been tested in lab experiments where some participants are instructed to eat at a slow pace and others at a
fast pace. The former consumed fewer calories (Andrade et al., 2008; Martin et al., 2007).
6

A major issue with estimating the effect of lunch length on weight is that a child’s lunch
length may be correlated with unobserved factors that directly influence weight. For instance,
parents that otherwise encourage healthy nutrition and exercise habits may send their children to
schools with longer lunches, and it is these habits, and not lunch length which influence weight.
Similarly, schools facing accountability pressures may reduce the time allocated for lunch as
well as physical education in an effort to improve achievement, and it could be the reduced
exercise time that ultimately impacts weight. Moreover schools facing financial constraints may
reduce cafeteria staff (plausibly reducing eating time) at the same time that they increase junk
food sales on campus, thus affecting children’s BMI (Anderson and Butcher, 2006).
While some of these issues are more likely to occur than others, we cannot rule out that lunch
length is endogenous. As a result, we estimate the impact of lunch length on children’s weight
using a variety of methods. This includes OLS where we condition on a rich set of controls, a
two-sample IV model, and a difference-in-differences model. Together, the results from these
models offer useful insight on the effect of lunch length on children’s weight.
3. Data & Descriptive Statistics
3.1 School Nutrition Dietary Assessment-III
The SNDA-III is a nationally representative survey of children in grades one through twelve
at schools in the U.S. The survey covers 2,314 students at 285 schools. Children, their parents,
principals and food service managers are surveyed regarding the diet and exercise behaviors of
kids and the nutrition environment at schools. In addition, survey administrators measured each
child's height and weight, providing a measure of BMI. This data is used for the OLS analysis,
and one important limitation of this data is that it does not contain state identifiers.
Children are not asked how long their lunch periods are. We construct lunch length using
7

principals’ reports of the start and stop time of each lunch period and which grades are assigned
to eat in each period. 6 We combine this with each child’s grade to construct lunch length for
each child. Missing information on lunch length, BMI, and key demographic characteristics
reduces the final sample size to 1,287 students from 179 schools. 7 Figure 1 displays the
distribution of lunch lengths among students in this final sample; as discussed above, the average
lunch length is 32.7 minutes, although this reflects all time allocated for lunch including travel
time to/from the cafeteria, time spent waiting for food, cleaning up, and using the restroom.
Appendix Table 1 provides descriptive statistics for all the variables in the SNDA-III that are
included in the OLS analysis. The proportion of obese children (defined as having a BMI above
the 95 th percentile of a historic baseline for a given gender and age) is twenty-one percent, and
overweight (above 85 th percentile) is thirty-nine percent. The average time a student spends in
line waiting to purchase food is around five minutes. Close to seventy-nine percent of students
take physical education, forty-three percent receives a free or reduced priced lunch, and close to
seventy percent purchased a school meal on the day prior to the interview. 8
3.2 School Health Policies & Programs Study
The SHPPS collects data from schools and states and is used for the TSIV and DD analysis.
The SHPPS is conducted every six years starting in 1994. In each survey year, all fifty states and
D.C. are surveyed about state laws governing school nutrition practices. Within states, a sample
6 If students attend schools where the same grade can eat in multiple periods, we use the average.
7 The majority of missing information is due to non-response on lunch length by principals (only 208 schools
provided information). In conversations with survey administrators, it was determined that the questions about lunch
periods were only asked during an initial survey, and there was no follow-up. In an omitted analysis (available upon
request), we compare average characteristics of schools with and without missing information on lunch length and
find no systematic differences. We omit 11 students with lunch lengths of 100 minutes since they are potential
outliers, and 2 students who were in Kindergarten and may attend school half day. Results are similar (and available
upon request) if these observations are included (or if the 14 students with 90 minute lunches are also excluded).
8 Information on school characteristics is taken from principal responses, except for average service time which is
provided by the food service manager. Information on demographic and socio-economic characteristics of each child
and family, and each child’s exercise and eating habits were taken from the parent survey. Information on if a child
ate a school lunch was taken from the child survey.
8

of schools is surveyed regarding their nutrition and health practices. This analysis uses the 2006
SHPPS, and draws on both the school and state questionnaires. Overall, 944 schools from
forty-one states (state identities are known) were selected to participate in the school survey. 9
Lunch length is constructed from the school questionnaire. School administrators are asked
how much time (on average) children in their school have to eat lunch once they are seated with
food. Missing information reduces the sample of schools to 916, and Figure 2 displays the
distribution of lunch lengths for these schools. 10 The state questionnaire provides information on
the instrument. State food service administrators are asked: ``Does your state require or
recommend [to schools] a minimum amount of time students will be given to eat lunch once they
are seated?'' Answers include “require”, “recommend”, or “no policy”. In 2006, six states had a
requirement, 26 had a recommendation, and 18 states plus D.C. had no policy. 11
The SHPPS does not contain information about the required or recommended times. In
conversations between the author and state food service agencies, the requirements were
determined to be: Delaware (N/A), Kansas (20 minutes once students are seated with food), New
Mexico (30 minutes for the entire lunch period), Nevada (20 minutes once seated), South
Carolina (20 minutes once seated for grades K-5, a recommendation for all other grades), and
West Virginia (20 minutes once seated). States with recommendations do not provide a
9 Schools in Delaware, Hawaii, Nevada, Rhode Island, South Dakota, Utah, Wyoming, and D.C. were not surveyed.
10 Seven schools did not report lunch length; average characteristics are similar across schools with and without
missing information (results omitted for brevity but available upon request). Thirteen schools in Alaska are dropped
because the state questionnaire was not completed. Four schools that report providing less than 8 minutes for lunch
are omitted (results are similar if they are included, however, and are available upon request). One school that had
no state identifier was dropped. Nine schools (3 elementary, 6 middle/high) in South Carolina were surveyed. South
Carolina has a requirement for grades K-5 but recommendation for older grades. It is not clear whether it is
appropriate to use all schools from South Carolina in the first stage since for all other observations, variation in the
covariates occurs only across states. Ultimately, since the second stage utilizes data from the YRBSS (described in
Section 3.3), which consists of only high school students, the three elementary schools are omitted. Results are
similar (and available upon request) if these schools are included in the first stage.
11 The states which have a recommendation are: Alabama, Arkansas, California, Colorado, Connecticut, Georgia,
Iowa, Idaho, Louisiana, Missouri, Mississippi, Montana, North Carolina, North Dakota, Nebraska, New Jersey,
Ohio, Pennsylvania, South Dakota, Texas, Utah, Virginia, Vermont, Washington, Wisconsin, and Wyoming.
9

numerical time; rather they suggest that schools provide “enough” time for children to enjoy and
consume their food. We construct the instrument as a set of dummy variables equal to one if the
state requires, recommends, or has no policy, respectively.
Figure 3 illustrates the distribution of lunch lengths in the SHPPS for schools in states with
different lunch laws, and previews the findings from the first stage of the TSIV model. Average
lunch length is higher in states with a requirement compared to a recommendation or no policy:
In states with a requirement the average lunch length is 26.3 minutes, versus 23.1 minutes and
23.3 minutes in states with a recommendation and no policy, respectively. A higher proportion of
schools in states with a requirement offer lunch lengths of 35 or more minutes, whereas, a higher
proportion of schools in states with a recommendation or no policy offer less than 20 minutes. 12
There are three aspects of the state laws which are important to note. First, it is not possible
to determine the exact year in which states adopted their policies, although there is evidence to
suggest that the majority of states implemented their policy between 2000 and 2006. The 2000
SHPPS asked the same question about state minimum lunch laws to a subset of twenty-two
states, and in 2000, no states had a requirement, two had recommendations, and twenty had no
policy. Finally, in conversations with school food service agencies it was determined that the
state lunch laws from 2006 were still in place by 2010. 13
Second, although it is not possible to determine the exact reason why each state adopted their
particular lunch policy, anecdotal evidence suggests it may have been driven by federal pressure
that affected all states at once, rather than in response to a state-specific event or trend. In June of
12 To gain insight on how much lunch length varies across states, we regress school lunch length on a set of state
dummies. The results indicate that variation across states explains 11.9% of the observed variation in lunch length.
13 Substantial effort was made to identify the year in which states adopted their lunch law through direct
conversations between the author and administrators at state school food offices, as well as by examining state
statutes. Consistent information on year of adoption was not able to be determined through either of these methods.
The two territories with a recommendation were Louisiana and D.C. Information from the 2000 SHPPS is used to
assess the validity of the instrument in Section 5.4. No questions about lunch laws were asked in the 1994 SHPPS.
10

2004, the U.S. Congress passed the Child Nutrition and WIC Reauthorization Act, which
required each educational agency that participates in federally funded school meal programs to
develop and implement a wellness policy by the 2006 school year. The policy should address
goals and practices for nutrition, physical/health education, and food provision; the latter could
include minimum lunch length. These wellness policies were constructed in consultation with
states, and as a result states themselves may have identified a set of “best practices” such as a
minimum lunch length to implement at the state level (Council of State Governments, 2007).
A third and related point is that in order for state lunch length laws to be a valid instrument in
the TSIV model and exogenous regressors in the DD model, they must be uncorrelated with
unobserved characteristics of states which can directly influence children’s health outcomes.
Although this cannot be tested directly, a series of falsification tests and empirical evidence
(discussed in Section 5.4) suggests that these state laws are plausibly exogenous.
Information on other state school nutrition policies is drawn from the SHPPS state
questionnaire. These are listed on the left side of Appendix Table 2; the majority of questions in
the state survey are asked in the “require”, “recommend”, and “no policy” format, hence these
variables are constructed as binary indicators. A little less than half of all schools are in states
that ban junk foods in vending machines, and sixty-seven percent are in states where schools
must follow physical education standards. The right hand side of Appendix Table 2 details the
set of characteristics that we append to the SHPPS from various auxiliary data sources. This
includes variables that proxy for socio-economic status such age and ethnicity breakdowns, and
the percent of students receiving free or reduced priced lunches, which are taken from the
National Center for Education Statistics and U.S. Census. State-level school characteristics
include average pupil-teacher ratio (22.7 students per teacher), teacher salary, and eighth grade
11

proficiency on National Assessment of Educational Progress mathematics exam (34% of
students across states achieved above proficiency status). These variables are designed to control
for school environmental factors, such as overcrowding and financial constraints which could
affect student outcomes. Finally, adult obesity rates are obtained from the Center for Disease
Control, and indicate close to 27% of adults across all states are obese. 14
3.3 Youth Risk Behavior Surveillance System
The YRBSS is conducted every two years starting in 1991 and most recently in 2009. The
YRBSS is a cross-sectional survey of ninth through twelfth grade students, and collects
information on students’ eating habits. Students self-report their weight and height, providing a
measure of BMI. The YRBSS does not contain information on lunch length, but students’ state
of residence is known, and consequently all the state level information from the SHPPS and
auxiliary data can be merged into the YRBSS. This data is used for the TSIV and DD analysis.
For the analysis, we pool the 2007 and 2009 YRBSS surveys. Waves prior to 2007 are not
used to ensure that all students whose outcomes we observe were exposed to the laws that
existed as of 2006. 15 The data is pooled across years to increase sample size; the final sample
consists of over 180,000 students in forty-one states. Appendix Table 3 displays summary
statistics for students in the YRBSS. Thirteen and twenty-eight percent of students are obese and
overweight, respectively, a little less than half are female, and thirty-eight percent are minority.
It should be noted that BMI in the YRBSS is constructed from self-reported height and
weight, which can be reported with error. Previous research indicates that the correlation
14 All variables are averaged over the years 2006-2008, with the following exceptions: Adult Obesity (2007-2009),
Facility Overflow (2003-04), Per Pupil Expenditure, Pupil Teacher Ratio (2005-07), Teacher Salary (2005, 2007,
2008), Eighth Grade Math Proficiency (2009). All dollars are in 2007-08 dollars.
15 Some states and territories (California, D.C., Minnesota, Nebraska, Oregon, Virginia, and Washington) do not
conduct a state YRBSS. Among the states that do, the CDC has permission from a subset of these states to directly
provide researchers with the data. For the remaining states, a private data agreement must be reached with each
state's YRBSS representative. Through these two methods, state YRBSS data was obtained for all the states that
conduct a survey except two (Georgia and Louisiana).
12

etween self-reported and actual BMI is high, however the former is generally an underestimate
(Morrissey et al., 2006). If the measurement error is classical, this will not bias the estimates but
does result in lower precision. In the TSIV analysis, where we instrument lunch length with state
laws, measurement error in students’ weight will not result in biased estimates, so long as the
measurement error is uncorrelated with the instrument. While it seems unlikely that misreporting
is correlated with a state’s lunch length law, this cannot be ruled out entirely, and the results
should be considered with this caveat. Other studies faced similar issues with self-reported
measures (Anderson and Butcher, 2006; Cawley et al., 2007).
4. OLS Estimation & Results
We estimate the following model using the SNDA-III data:
health = S
is
0 1lunchlengt
his
2X
is 3
13
s
is
(Eq.1)
Here health is represents one of three health outcomes for child i at school s. The first is the
natural log of BMI, and the other two are binary indicators for being overweight and obese. We
use a linear probability model; results are similar using probit and are available upon request.
The variable lunchlengt his
is the length of a child's assigned lunch period in minutes. X is is a
vector of child and parent characteristics, and Ss is a vector of school characteristics. All the
variables included in X isand
Ss are listed in Appendix Table 1. Standard errors are clustered at
the school level and SNDA-III sample weights are used.
In Columns 1 and 2 of Table 1 we display the OLS results for the three outcomes. For
brevity, only the coefficient on lunch length is shown. Column 1 shows the results when we
condition on only basic characteristics of children (age, sex, race, family income), schools

(attendance rate, grade span) and dummies for region of residence. 16 The estimates indicate that
a ten minute increase in lunch length is predicted to decrease BMI by 1.1%, lower the probability
of being overweight by 3.7 percentage points (9.5%), and there is negative but insignificant
effect on obesity. Column 2 provides the results when we include a richer set of school and
individual level covariates which may be correlated lunch length and weight outcomes. This
includes whether there are vending machines or fast food restaurants at schools, whether children
can leave campus for lunch, time spent in service, whether the child participates in sports, and
whether the child buys lunch at school (for a full description of the controls see the Table 1 notes
and Appendix Table 1). In spite of these additional controls, the estimated effect of lunch length
is similar: A ten minute increase in lunch length is predicted to decrease BMI by 1.3% and
reduce the likelihood a child is overweight by 3.9 percentage points. 17 We continue to find no
significant effect on obesity. One explanation for this is that there may be heterogeneous effects
of lunch length across the BMI distribution; while lunch length may affect children along the
margin of the 85 th percentile, there may be little impact around the 95 th percentile. 18
As discussed above, long lunches could increase opportunities to eat, thus leading children to
gain weight. To test this, we split the continuous lunch length variable into four categories:
10-29, 30, 31-44, and 45 plus minutes. Following Equation 1 we regress the natural log of BMI
on the indicators (10-29 minutes is omitted) and all the covariates in Appendix Table 1. These
results are provided in the last column of Table 1. We find no significant difference between the
16 A control for recess is included in the basic model because there may be a tradeoff between lunch length and
recess time. No information about the time allocated for recess is available in the SNDA-III; a discussion of the
tradeoff between recess and lunch time is provided in Section 5.4.
17 Lunch length can vary within schools, however this only occurs for 20% of students in the SNDA. Due to the
small sample size of the data and somewhat limited variation, we do not include school fixed effects in Equation 1.
That said, in an omitted analysis (upon request) we estimate Equation 1 controlling for all the covariates in
Appendix Table 1 along with school fixed effects and find no significant effects of lunch length on any of the three
outcomes. The standard errors in these regressions are more than double those in Column 2 of Table 1.
18 For studies that find heterogeneous effects of independent variables on BMI, see Beyerlein et al., 2009; Herbst and
Tekin, and Han et al., 2011.
14

omitted group and 30 minutes, but do find negative and large effects for the other groups: The
BMI of students with 31-44 minutes is 2% lower (although the estimate is not statistically
significant), whereas it is 4% lower (significant at the 10% level) for those with 45 or more
minutes. This provides no evidence that longer lunches lead to increases in weight for children.
Overall, the estimates in Table 1 suggest there is a negative relationship between lunch
length and BMI. In an omitted analysis (available upon request), we show these results are robust
to the inclusion of additional controls and corrections for measurement error. 19 That said, there
may unobserved factors which bias the OLS results, and so we turn to the TSIV and DD models.
5. TSIV & DD Estimation
5.1 First Stage
In the first stage we use the SHPPS to estimate the relationship between lunch length and
each state’s laws on lunch length. Formally, we estimate:
lunchlengt h = R (Eq. 2)
sm
0 1Z m
2M
m 3
m
Here, lunchlengt hsm
denotes lunch length in school s and state m, and Z m is a set of dummy
variables for if a state requires, recommends, or has no policy on lunch length. M m is a vector of
state level school nutrition and physical education policies, financial characteristics, and
socio-economic, and Rm is a set of region dummies. All control variables included in the first
and second stage of the TSIV model are listed in Appendix Table 2.
19 The OLS results are robust to the inclusion of a control for whether or not a child eats within 15 minutes of
midday. Both the SNDA and NASBE recommend that children eat close to midday in order to avoid going long
periods of time without food. With the addition of the control, the magnitude of the lunch length estimates remains
similar, and we find negative but insignificant effects of eating at midday on BMI. To assess measurement error, we
omit all students that were assigned an average lunch length (see Footnote 6). These point estimates are slightly
larger (in absolute value), which is consistent with measurement error theory. We also estimate the effect of lunch
length separately for students in younger and older grades. We find a stronger negative effect for children in grades
one through five (a 7 percentage point drop in the probability of being overweight in response to a ten minute
increase in lunch length) compared to grades nine through twelve (1.3 percentage point drop). We discuss this
finding in further detail in Section 6.1.
15
sm

The set of conditioning variables was purposefully chosen to be large and exhaustive in an
attempt to control for state characteristics that may be correlated with weight and lunch length.
The state nutrition and health policies control for whether students can purchase junk foods
through vending machines or school fundraisers, whether schools face serving size limits, or
must meet state physical education standards. Average per pupil spending, teacher salaries, and
pupil-to-teacher ratios should proxy for the financial conditions of schools and overcrowding.
Adult obesity rates capture variation in health and eating behavior across states, and math
proficiency rates are meant to reflect performance accountability pressures. Because no school
covariates can be included in the model, these state variables are intended to control for school
factors; for instance, controlling for whether a state bans the sale of junk food in vending
machines should be indicative of the presence of vending machines in schools. Finally, standard
errors are clustered by state and SHPPS sample weights are used throughout.
The first stage results are displayed in Table 2. Recall we omit the category require. Focusing
first on the instrument, the estimates indicate that schools in states with a recommendation have,
on average 5.7 minute shorter lunches compared to schools in states with a requirement. An even
stronger effect is found at schools where there is no policy: lunch lengths are about 6.6 minutes
shorter. The F-statistic testing the joint significance of the instrument is 30.62 [p-value=0.0000]
and given at the bottom of the table. In addition, we test whether the coefficient estimates on
recommend and no policy are statistically different from one another and find that they are at the
5% level. This indicates that our identifying variation is not only based on a comparison of states
with and without a requirement, but also across states with no policy versus a recommendation.
Finally, the estimates for the other state controls are mostly intuitive. For instance, schools in
16

states with higher pupil-teacher-ratios have longer lunches, presumably to service all students. 20
5.2 Second Stage
For the second stage, we use the YRBSS data with the merged SHPPS and auxiliary data.
Using the estimates from Eq. 2, we construct predicted lunch length (denoted, predlunch m ) for
each student in the YRBSS, and estimate the following regression:
health = B
(Eq. 3)
im
0 1predlunch
m
2M
m
3Rm
4
Here, healthimrefers to any one of three outcomes for student i in state m (natural log of BMI,
overweight, obese). M m and Rm are defined as in Eq. 2 and are listed in Appendix Table 2, and
Bimrepresents a vector of student characteristics from the YRBSS and are listed in Appendix
Table 3. Standard errors are clustered at the state level and YRBSS sample weights are used.
Standard errors are adjusted (using the approach outlined by Murphy and Topel (1985)) to
correct for the fact that lunch length is predicted.
The second stage results are displayed in Table 3. For each dependent variable Eq. 3 is
estimated with and without B im,
and for brevity only the coefficients on these variables and
predicted lunch length are reported. Focusing on Ln(BMI), we find that a ten minute increase in
predicted lunch length is predicted to reduce BMI by 1.3% when no individual covariates are
included and 1.2% when they are. The addition of the individual controls does not substantially
change the estimate on lunch length, suggesting these additional covariates are not correlated
with the exogenous portion of lunch length predicted by the laws. The remaining columns
suggest that a ten minute increase in predicted lunch length is predicted to reduce the probability
20 All the school characteristics in the first stage are measured at the state level, and as a consequence of being the
average for schools within a state, these variables may be mismeasured. If the mismeasured variables are correlated
with a state’s lunch law this can bias the estimates of the laws. In an omitted analysis (available upon request), we
check the robustness of the first stage results to different assumptions about the extent of the measurement error, and
find that the relationship between the instrument and lunch length would not change substantively if there was
measurement error present.
17
im
im

that a child is overweight by 2.5 percentage points (9%) and there is no significant effect obesity.
Overall, the TSIV results indicate that longer lunches lead to healthier weight outcomes.
5.3 Difference-In-Differences
We examine the robustness of the TSIV results by estimating a simple difference-in-
differences model of each state’s lunch law on BMI. This is analogous to estimating the reduced
form regression of the instrument on the outcome. The DD model compares BMI before and
after a state switched from no policy to a recommendation or requirement, relative to the same
before and after for states where the state continued to have no policy. To do this, we utilize data
from the 1999, 2001, 2007, and 2009 YRBSS, and the 2000 and 2006 SHPPS.
To construct the data for the DD analysis, we first combine data from the YRBSS survey
years 1999 and 2001 (this serves as our “pre” period) and data from the 2007 and 2009 surveys
(our “post” period). We limit our sample to states which were surveyed in both the 1991/2001
and 2007/2009 surveys. We then look to the 2000 SHPPS, where states were asked if they had
any policy on lunch service. Among the 22 that said yes, they were then asked if they had a
requirement/recommendation/no policy on lunch length. None of the states that were asked and
are represented in the 99/01/07/09 YRBSS data had a policy in 2000, and we assume that states
that didn’t have a policy on lunch service also had no policy on lunch length. From the 2006
SHPPS, we know which states then adopted requirements, recommendations, or had no change
by 2006. Our final sample includes more than 230,000 students across 32 states. We estimate:
health
imc
Require
4
= Require
m
0
1
* C0709
mc
m
Recommend
Recommend
5
2
m
18
m
* C0709
C0709
3
mc
mc
M
6
mc
B
7
imc
imc
(Eq. 4)
Here, i refers to the individual, m to the state, and c to the YRBSS cohort (i.e. 1999/2001 or
20007/2009). Require m is a binary variable equal to one for states where there was no policy in

2000 but by 2006 there was a requirement, and similarly for Recommend m and NoPolicy m (the
omitted category). C0709mc is an indicator equal to one for observations from the 2007/2009
YRBSS and zero for the 1999/2001 YRBSS. M mc is a vector of the same state level controls that
were included in the TSIV model. Finally, Bimc is a vector of individual characteristics of students
in the YRBSS. The coefficients 4 and 5 provide the difference-in-difference estimates. 21
The results are provided in Table 4. Only the estimates for 1 5
are presented for brevity.
Column 1 (2) shows the results without (with) controls for student demographic characteristics.
In both specifications, the effects of the laws are in the expected direction. That is, we find that
there is a decrease in BMI by approximately 2.3-2.5% for students who live in states that went
from no policy in 2000 to a requirement in 2006 (relative to students in states where there was no
policy in both 2000 and 2006). There is a similar, but smaller drop in BMI of 0.8-0.9% in states
that went from no policy to a recommendation. Overall, these results provide additional support
that stricter state lunch laws reduce BMI vis-à-vis lunch length.
5.4 Validity of Instrument
In order to interpret the estimates from the TSIV and DD analysis as estimating causal
effects, a state’s lunch law should only affect weight outcomes through lunch length. Doubts
about validity can be aggregated into two broad categories. First, states’ lunch length laws could
reflect unobserved differences that exist across states, and it is these differences that affect
outcomes, rather than lunch length. For instance, a state may adopt their lunch law in response to
pre-existing trends in BMI. Second, state laws may induce changes in schools with respect to
other nutrition practices that can directly influence BMI, which will confound the results.
21 Appendix Table 2 displays the state level covariates attached to the 2007/2009 data, and similar data was collected
for the year 2000 and attached to the 1991/2001 observations (descriptive statistics available upon request).
19

Although we cannot conclusively rule out these issues, below we describe a series of falsification
tests and empirical evidence that suggests these issues are unlikely to bias our results.
We test for evidence of whether states adopted their lunch laws in response to prior trends in
children’s weight. If this is the case, we should observe a relationship between the lunch laws in
place by 2006 and BMI in states prior to that year. To test this, we estimate the TSIV model (first
and second stages are defined in Eq. 2 and 3), where in lieu of using data from the 2007/2009
YRBSS, we use student BMI data from the 1999/2001 YRBSS. We condition on all the controls
listed in Appendix Table 2 (in both the first and second stage), and the results are provided in
Table 5. 22 One issue that limits inference from this test is that students in the 1999 and 2001
YRBSS were exposed to state lunch laws that existed prior to 2006, and to the extent that these
laws are correlated with the 2006 laws, we may estimate a non-spurious, non-zero relationship
between the 2006 laws and prior levels of BMI. In spite of this issue which would bias the
estimates in favor of finding an effect, the estimate is not statistically different from zero. 23
We cannot control for school level covariates in the TSIV and DD models since the YRBSS
does not include information on schools. The lack of school controls will be problematic if a
state’s lunch law not only affects lunch length, but also other school practices which impact
weight. We investigate this in the upper-half of Table 6 using data on schools in the 2006
SHPPS. We consider four characteristics of schools which may plausibly change in response to
22 The first year that BMI was collected in the YRBSS was in 1999. Survey coverage was small in 1999; only
fourteen states conducted surveys. To increase sample size and representativeness, we combine the 1999 data with
2001 data, resulting in a sample size of close to 78,000 students in forty states.
23 In an analysis omitted for brevity (available upon request), we estimate the first stage of the TSIV model, where
in lieu of controlling for each state’s school lunch length law we include each state’s workplace lunch length laws.
These laws indicate whether (in 2006) a state required or recommended to firms that they provide their employees
(adults and minors, respectively) with a minimum amount of time for meals (Department of Labor, 2006). If the
school lunch laws reflect unobserved health and behavioral characteristics of a state, the workplace laws likely
reflect similar characteristics, and therefore should be correlated with school lunch lengths. In practice, we find no
evidence of this: The F-statistic testing the joint significance of the dummies for requiring a minimum lunch length
for minors and adults is F(2,40)= 0.0.33 [p-value=0.7203]).
20

lunch length and could affect children’s weight: overcrowding of eating locations, amount of
time allocated for recess and physical education, and bans on junk foods in vending machines. In
the table, we group schools by their state’s law, and calculate the fraction of schools with the
given characteristic. Rates of overcrowding and recess time are similar across groups, but
schools with lunch length requirements tend to ban junk food at a higher rate and offer more time
for physical education. That said, these differences are not statistically significant once we
control for observed state level characteristics: The last column provides the F-statistic testing
the joint significance of the lunch law dummies from a regression of these characteristics on the
dummies and all the covariates in Appendix Table 2, and in all cases we reject the null. 24
A remaining concern is whether schools changed their lunch lengths in response to their
state’s law, or if schools were already setting systematically different lunch lengths prior to the
law. This is tested using data from the 2000 SHPPS. Schools in the 2000 SHPPS were asked to
report the length of time children are given to eat lunch once seated with food. The last row of
Table 6 provides average lunch time in 2000 for schools grouped by the law their state adopted
by 2006. Lunches range from 22-24 minutes, and the differences are not statistically different,
suggesting no evidence of systematic differences prior to adoption. 25
6. Discussion
6.1 Comparison of Results
The results of the TSIV and OLS models are largely similar. The most pronounced difference
24 For brevity, we only display the results from a few school characteristics in Table 6. In an omitted analysis
(available upon request), we also compare the fraction of schools which allow students to leave campus during lunch
and find no statistically significant differences across schools in states with different state lunch laws. Other
differences that may arise due to state lunch laws include changes in the length of the school day. This information
is not included in the SHPPS, therefore we turn to the SNDA-III and compare average school day length across
schools with 10-29, 30, 31-40, and 41plus minutes. We find little difference in school day length across groups.
25 Schools interviewed in the 2000 and 2006 SHPPS are not necessarily the same. We construct a graph analogous to
Figure 3 using 2000 school lunch lengths, and generally find that across lunch lengths, there are a similar proportion
of schools in states that adopted a requirement, recommendation, or no policy by 2006.
21

is for the outcome of being overweight, where the OLS estimate indicates there is a 3.9
percentage point drop in the probability of being overweight, compared to the 2.5 percentage
point drop estimated by TSIV. Taking these results at face value, this suggests that the OLS
estimates are biased downward. This is consistent with a scenario where schools that offer longer
lunches (or children that attend these schools) also follow other “best practices” such as
serving/eating healthy meals or engaging in physical exercise and nutrition education.
Ultimately, the reader should be cautious about comparing the magnitude of the OLS and
TSIV findings and interpreting the difference as the result of unobserved heterogeneity. This is
because there are a few differences across the data sets which limit comparability of the
estimates. First, the SNDA-III spans students from all grades, while the YRBSS only covers
grades nine through twelve. Response to meal length may differ by age because of differences in
metabolism and growth patterns. Moreover, health outcomes of younger children are generally
more malleable to new health and nutrition practices than adolescents (Tanofsky et al., 2003). To
examine this we re-estimate Equation 1 using just a sample of SNDA-III students who are in
grades nine through twelve, and estimate a smaller reduction in being overweight than in the full
sample: a 1.3 percentage point decrease (although it’s not statistically significant), compared to
the 3.9 percentage point decrease estimated using the whole sample. This suggests differences in
the ages of students across samples may explain some of the difference across estimators, rather
than the presence of unobserved heterogeneity. A second difference is that meal length in the
SNDA-III refers to the entire meal period, whereas in the SHPPS it is the time remaining once a
child has his/her food. On one hand, the SHPPS provides a more accurate description of the time
available to eat, however it is also an average for all children, including those who “brownbag”
and those who buy school lunches, and thus does not capture timing differences between the two.
22

6.2 Economic Significance of the Estimates
We can evaluate the magnitude of the findings by estimating the monetary and non-monetary
savings that accrue from providing longer lunches. This is done below, but it is important to note
that these calculations do not factor in any of the potentially adverse consequences of or costs
associated with longer lunches. For instance, increasing lunch length could come at the expense
instructional and non-instructional time, leading to lower achievement and less opportunity for
exercise. Moreover, lengthening lunches could necessitate hiring more cafeteria staff and
incurring other expenses. Any formal assessment of the welfare benefits of longer lunches should
consider the potential costs and benefits, and is out of the scope of this analysis.
Fewer overweight children would translate into lower costs of treating weight-related
illnesses. 26 Transande and Chatterjee (2009) estimate that the two year medical cost of
outpatient services for treating overweight children (relative to normal weight) is close to $79 per
child. Pairing this with the TSIV finding, this indicates that if all students in the YRBSS were
given ten minutes more for lunch, this would reduce the share of overweight children by 2.5
percentage points, from an average of 27.7% to 25.2%, which represents a cost saving of nearly
$33 million dollars (2005 dollars) over two years. Additionally, we can consider the impact on
academic outcomes. Geier et al. (2007) find that overweight children miss 1.7 more days of
school per year than their normal weight classmates. Based on this figure, if all the students in
the YRBSS were given ten minute longer lunches, this would reduce the number of absences by
close to seven hundred thousand days per year. This is a non-trivial amount, given that close to
60% of all tenth-grade students miss anywhere from one to six days per year (Snyder, 2011). 27
26 Cost savings can accrue regardless of if the costs of weight-related medical care are borne by all taxpayers, or
largely internalized (Bhattacharya and Sood, 2011).
27 The weighted YRBSS data represents 16.65 million children, of which 27.7% are overweight. A reduction in the
percent of overweight children to 25.2% yields 333,000 fewer overweight children. Multiplying 333,000*
23

6.3 Calorie Consumption
As discussed above, there are (at least) three potential channels through which lunch length
may affect children’s weight. Shorter lunches may lead students to skip lunch or eat less than is
nutritionally necessary, lead students to substitute towards higher calorie, less healthy food
options, and lead students to eat at a quick pace, thus increasing the likelihood that they overeat.
To gain some insight on these potential mechanisms, we examine data on calorie consumption
from students in the SNDA-III. Students completed a 24 hour dietary recall diary, where they
listed their food consumption over a random school day, and the calorie content of these foods
was computed by survey administrators. We use this data to estimate the relationship between
calorie consumption and the length of time a child is assigned to eat lunch in school using OLS.
The dependent variables is the amount of calories consumed during meals throughout the day,
and to capture any non-linear effects of lunch length, we formulate the dependent variable as a
set of four indicators for lunch lengths of less than 30, 30, 31-44, and 45 plus minutes (less than
30 minutes is omitted). All regressions control for all the covariates in Appendix Table 1.
The results are presented in Table 7. 28 Looking at breakfast and snack calories (first and
fourth column), there is no systematic or significant difference in calories consumed during
breakfast or as snacks for students who have longer lunch lengths relative to students who have
less than 30 minutes. In contrast, for lunch calories, we observe that students who have 30
minutes to eat lunch consume an average of 59 more calories during that time than students who
eat for less than 30 minutes, those with 31 to 44 minutes eat 46 more calories, and those with 45
or more minutes eat an average of 89 more calories. Moving to calories consumed during dinner,
$79=$33,070,005. Similarly, 1.7 days *333,000=711,633 fewer absences.
28 Total calories are restricted to be less than 6,000, and the sample is restricted to students who ate lunch on the day
of the dietary recall. Calorie intake data may be misreported; we check the robustness of the results in Table 7 by
removing implausible reports using the method outlined by Huang et al. (2004) and Ventura et al., (2006). Results
are similar to those reported in Table 7 and are available upon request.
24

we observe a similar, but opposite-signed pattern: Students with 30 minutes to eat lunch
consume 88 fewer calories during dinner than students with less than half an hour to eat lunch,
those with 31 to 44 minutes consume 47 calories less (although this estimate is not statistically
significant), and those with the longest lunch consume 99 calories less.
The estimates in Table 7 suggest that students with longer lunches actually tend to consume
more calories during lunch, but this increase is exceeded by a slightly larger decrease in calories
consumed during dinner. Such eating behavior is consistent with a scenario where children that
have shorter lunches tend to eat less because they do not have time to eat, but subsequently, they
arrive at the next meal and eat more because they are hungry.
These results suggest that there are two possible ways that shorter lunch lengths could lead to
increases in BMI. First, students with longer lunches consume, on net, fewer calories during
lunch and dinner combined compared to students with shorter lunches, and this can affect weight
gain. For instance, students with 30 minute lunches consume about 29 calories less during lunch
and dinner combined, relative to students with less than 30 minutes. The magnitude of this
calorie reduction is not trivial and is consistent with the estimates of the effect of lunch length on
BMI presented in Table 3. For instance, in Schanzenbach’s 2009 study, the author found that an
increase in 40 calories per day is enough to increase the rate of obesity of first grade students by
close to 1.7 percentage points. One caveat to applying Schanzenbach’s (2009) findings to this
study is that the SNDA-III includes adolescents and teens, and calorie intake may impact weight
differently based on age. That said, the overall results from the calorie analysis are in line with
our general finding that longer lunches lead to better weight outcomes for students.
Second, the timing of consumption may have important implications for weight gain.
Students arguably have more time to burn off any calories consumed during the day with after
25

school exercise and play, whereas there are fewer exercise opportunities for students after dinner.
Longer lunches may simply shift consumption towards times of the day where students have
more opportunities to expend calorie intake, thus leading to reductions in BMI.
7. Conclusion
The consequences of childhood obesity are considerable. Overweight children are at risk
for having high blood pressure and elevated levels of cholesterol, are more prone to asthma,
sleep apnea, Type 2 diabetes, and experience psychosocial problems which can lead to risky
behavior and depression (Dietz, 1998; Farhat et al., 2010; Freedman et al., 1999; Mallory et al.,
1989; Serdula et al., 1993). Such illnesses can lead to absenteeism from school, which has
implications for human capital accumulation and skill-formation (Geier et al., 2007). Finally, the
financial costs of treating childhood obesity are non-trivial: Transande and Chatterjee (2009)
estimate the costs of treating weight-related illnesses of children at 14.1 billion annually.
The severity of childhood obesity has prompted interest at the national and local level for
developing strategies to reduce obesity and prevent further increases from occurring (U.S.
Department of Health and Human Services, 2001; NASBE, 2010). At the forefront of this
discussion are schools, and of particular interest is the role of meals and food eaten at school.
Proposed school-based methods to achieve reductions in obesity include, among other things,
increasing the availability of fresh fruits and vegetables, reducing the availability of calorie
sweetened beverages (U.S. Department of Health and Human Services, 2001), and as argued by
some advocates for nutrition reform, scheduling lunch periods of sufficient length so that
children have time to enjoy and consume their food (SNA, 2005; NASBE, 2000).
To date, there has been no evidence about the impact of lunch period length on children’s
weight. This paper is the first to provide evidence on this link. The TSIV results indicate that
26

extending lunch length by ten minutes is associated with a 1.2% reduction in BMI, and reduces
the probability a child is considered overweight for his/her age and gender by 2.5 percentage
points. These magnitudes are economically meaningful when translated into potential cost
savings that could result from a reduction in treating weight-related illness among children.
Although the results indicate that allocating more time for school meals has potential benefits
for lowering children’s weight, there are many practical considerations that limit the extent to
which schools can offer longer lunches. If the school day is held constant, longer lunches would
presumably come at the expense of other non-instructional/non-core activities (physical
education, art) or instructional activities (math, reading), and there are drawbacks to this. While
increasing the length of the school day is one possibility, this is likely to meet resistance unless
sufficient compensation can be given to staff and administrators. Given the already strained
budgets of schools and districts, this seems difficult to manage. The tension between finances
and practices which can affect weight is not unique to this study. For instance, Anderson and
Butcher’s 2006 study highlights the tradeoff between allowing junk food sales on school
property and increased revenue to schools from these sales.
This analysis underscores the need for further research to be conducted not only on the food
students eat in school, but also the environment in which they eat it in. For instance, it may be of
interest to examine whether the spacing of lunch between the beginning and end of the school
day, or eating location (supervised cafeteria space, versus unsupervised locations such as
hallways or commons) has an effect on children’s weight. It would also be interesting to examine
the impact of lunch length on other outcomes of children, such as their behavior and achievement
in school. Given the results of such research, policy makers and educational administrators can
work towards developing effective and creative solutions for bettering the outcomes of children.
27

References
Anderson, Patricia and Kristine Butcher and Philip Levine. 2003. “Maternal Employment and
Overweight Children,” Journal of Health Economics, 22, pp. 477-504.
Anderson, Patricia and Kristin Butcher. 2006. “Reading, Writing and Raisinets: Are School
Finances Contributing to Children's Obesity?,” Jounal of Human Resources, 41, pp. 467-494.
Andrade, Ana and Geoffrey Greene and Kathleen Melanson. 2008. “Eating Slowly Led to
Decreases in Energy Intake within Meals in Healthy Women”. Journal of the American Dietetic
Association. Vol. 108, pp. 1186-1191.
Bergman, Eathan A. and Nancy S. Buergel and Annaka Femrite and Timothy F. Englund and
Michael R. Braunstein. 2003. “Relationships of Meal and Recess Schedules to Plate Waste in
Elementary Schools,” Insight: A Publication of the National Food Service Management Institute.
No. R-71-03. pp. 1-34. Accessed online at: http://www.nfsmi.org
Beyerlein, Andreas and André Toschke and Rüdiger von Kries. 2008. “Breastfeeding and
Childhood Obesity: Shift of the Entire BMI Distribution or Only the Upper Parts?”. Obesity.
Vol.16 No. 12, pp. 2730–2733.
Bhattacharya, J. and N. Sood “Who Pays for Obesity?,” Journal of Economic Perspectives, 25(1),
139–158.
Carlson, Olga and Bronwen Martin and Kim Stote and Erin Golden and Stuart Maudsley and
Samer Najjar and Luigi Ferrucci and Donald Ingram and Dan Longo and William Rumpler and
David Baer and Josephine Egan and Mark Mattson. 2007. “Impact of reduced meal frequency
without caloric restriction on glucose regulation in healthy, normal-weight middle-aged men and
women”. Metabolism Vol. 56 (12), pp. 1729-34.
Cawley, John and Chad Meyerhoefer and David Newhouse. 2007. “The Impact of State Physical
Education Requirements on Youth Physical Activity and Overweight”. Health Economics Vol.
16, No. 12, pp. 1287-1301.
Center for Science in the Public Interest. 2004. “Dispensing Junk: How School Vending
Undermines Efforts to Feed Children Well” Accessed online at: http://www.cspinet.org/
Child Nutrition and WIC Reauthorization Act. 2004. Public Law. 108–265
Conklin, Martha and Laurel Lambert and Janet Anderson. 2002. “How Long Does it Take
Students to Eat Lunch? A Summary of Three Studies”. The Journal of Child Nutrition &
Management Vol. 26, Issue 1, pp. 1-6 of Current Issues.
Council of State Governments. 2007. School Wellness Policies Legislator Policy Brief
Accessed online at: http://www.healthystates.csg.org
Deshmukh-Taskar, Priya R. and Theresa A. Nicklas and Carol E. O'Neil and Debra R. Keast and
28

John D. Radcliffe and Susan Cho. 2010. “The Relationship of Breakfast Skipping and Type of
Breakfast Consumption with Nutrient Intake and Weight Status in Children and Adolescents: The
National Health and Nutrition Examination Survey 1999-2006,” Journal of the American Dietetic
Association. Volume 110, Issue 6, pp. 869-878.
Dietz, W. 1998. “Health consequences of obesity in youth: Childhood predictors of adult
disease”. Pediatrics Vol. 101, pp. 518–525.
Department of Labor. 2006. “Minimum Length of Meal Period Required Under State Law For
Adult Employees in Private Sector 1 January 1, 2006”. Wage And Hour Divsion. Accessed
online at: http://www.dol.gov
Farhat, T. and R.J. Iannotti and B.G. Simmons-Morton. 2010. “Overweight, obesity, youth, and
health-risk behaviors”. American Journal of Preventative Medicine Vol. 38, No. 3, pp. 258-67
Freedman, D.S. and W.H. Dietz and S.R. Srinivasan and G.S. Berenson. 1999. “The relation of
overweight to cardiovascular risk factors among children and adolescents: The Bogalusa Heart
Study”. Pediatrics Vol. 103, pp. 1175-1182.
Frisvold, David and Julie Lumeng. 2011. “Expanding Exposure: Can Increasing the Daily
Duration of Head Start Reduce Childhood Obesity?” Journal of Human Resources Vol. 46, No.
2, pp. 373-402.
Geier, Andrew B. and Gary D. Foster and Leslie G. Womble and Jackie McLaughlin and Kelley
E. Borradaile and Joan Nachamani and Sandy Sherman and Shiriki Kumanyika and Justine
Shults. 2007. “The Relationship Between Relative Weight and School Attendance Among
Elementary Schoolchildren”. Obesity Vol. 15, No. 8, pp. 2157-2161.
Getlinger, Mary Jane and Carol Laughlin and Elizabeth Bell and Christine Akre and Bahram
Arjmandi. 1996. “Food waste is reduced when elementary-school children have recess before
lunch.”. Journal of the American Dietetic Assocation Vol. 96, Issue 9, pp. 906-908.
Gundersen, Craig and Brent Kreider and John Pepper. 2012. “The Impact of the National School
Lunch Program on Child Health: A Nonparametric Bounds Analysis”. Journal of Econometrics.
Vol. 166, Issue 1, pp. 79-91.
Han, Euna and Lisa Powell and Oksana Pugach. 2011. “The heterogenous relationship of food
stamp participation with body mass: Quantile regression model”. Food Policy, Vol. 36, Issue 4,
pp. 499-506.
Hayes, Dayle and Rachel Landan. Editorial Advisers. Food And Nutrition Volume 4 Glycemic
Index to Legislation and Food. 2009. Marshall Cavendish Corporation. pp. 531-540.
Herbst, Chris and Erdal Tekin.“Childcare Subsidies and Childhood Obesity”, Forthcoming in
Journal of Urban Economics
29

Huang, Terry and Nancy Howart and Biing-Hwan Lin and Susan Roberts and Megan McCrory.
2004. “Energy Intake and Meal Portions: Associations with BMI Percentile in U.S. Children”.
Obesity Research. Vol. 12, No. 11. pp. 1875-1885.
Lafay, Lionel and Frederique Thomas and Louise Mennen and Marie Aline Charles and Eveline
Eschwege and Jean-Michel Borys and Arnaud Basdevant. 2001. “Gender Differences in the
Relation Between Food Cravings and Mood in an Adult Community: Results from the Fleurbaix
Laventie Ville Sante´ Study” International Journal of Eating Disorders. Volume 29, Issue 2, pp.
195–204.
Leidy, Heather and Rebecca Lepping and Cary Savage and Corey Harris. 2001. “Neural
Responses to Visual Food Stimuli After a Normal vs. Higher Protein Breakfast in
Breakfast-Skipping Teens: A Pilot fMRI Study”. Obesity. 2011.
Mallory, G.B. and D.H. Fiser and R. Jackson. 1989. “Sleep-associated breathing disorders in
morbidly obese children and adolescents”. Journal of Pediatrics Vol. 115, pp. 892-897.
Martin, Corby and Stephen Anton and Heather Walden and Cheryl Arnett and Frank Greenway
and Donald Williamson. 2007. “Slower eating rate reduces food intake of men, but not women:
Implications for behavioral weight control”. Behaviour Research and Therapy Vol. 45, No. 10,
pp. 2349-2359.
Millimet, Dan and Rusty Tchernis and Muna Husain. 2010. “School Nutrition Programs and the
Incidence of Childhood Obesity’’. Journal of Human Resources Vol. 45, No. 3, pp. 640-654.
Morrissey, SL and LM Whetstone and DM Cummings and LJ Owen. 2006. “Comparison of
self-reported and measured height and weight in eighth-grade students”. Journal of School
Health Vol. 76, No. 10, pp. 512-5.
Murphy, Kevin M. and Robert H. Topel. 1985. “Estimation and Inference in Two-Step
Econometric Models”. Journal of Business and Economic Statistics Vol. 3, No. 4, pp. 370-379.
National Association of State Boards of Education. 2000. “Fit, Healthy, and Ready to Learn, Part
I” Publication of the NASBE pp. 1-236. Accessed online at:
National Association of State Boards of Education. 2010. Obesity Prevention Policies for Middle
and High Schools: Are We Doing Enough? Issues in Brief. Accessed online at: http://nasbe.org/
Rolfes, Sharon Rady and Kathryn Pinna and Ellie Whitney. Understanding Normal and Clinical
Nutrition. Eighth Edition. 2009. Wadsworth, Cengage Learning. Chapters 8 and 9.
Schanzenbach, Diane Whitmore. 2009. “Do School Lunches Contribute to Childhood Obesity?”.
Journal of Human Resources Vol. 44, No. 3, pp. 684-709.
School Nutrition Association. 2005. “School Nutrition Association Local Wellness Policy
Recommendations”. pp. 1-11. Accessed online at: www.schoolnutrition.org
30

Serdula, MK and D Ivery and RJ Coates and DS Freedman and DF Williamson and T Byers.
1993. “Do obese children become obese adults? A review of the literature”. Preventitive
Medicine Vol. 22, pp. 167-177.
Snyder, T.D. and Dillow, S.A. 2011. Digest of Education Statistics 2010 (NCES 2011-015).
National Center for Education Statistics, Institute of Education Sciences, U.S. Department of
Education. Washington, DC.
Tanofsky-Kraff, Marian and Helen Hayden-Wade and Paul Cavazos and Denise Wilfley. 2003.
“Treating And Preventing Pediatric Obesity”. Chapter 11 in Obesity: Etiology, Assessment,
Treatment, and Prevention Ross Andersen (Ed.). Human Kinetics Publisher, 1 ed.
Trasande, Leonardo and Samprit Chatterjee. 2009. “The Impact of Obesity on Health Service
Utilization and Costs in Childhood”. Obesity, Vol. 17, pp. 1749–1754.
Ventura, Alison K. and Eric Loken and Diane Mitchell and Helen Smiciklas-Wright and Leann
Birch. 2006. “Understanding Reporting Bias in the Dietary Recall Data of 11-Year-Old Girls”
Obesity Vol. 14, No. 6. pp. 1073-1084.
von Hippel, Paul T. and Brian Powell and Douglas B. Downey and Nicholas J. Rowland. 2007.
“The Effect of School on Overweight in Childhood: Gain in Body Mass Index During the School
Year and During Summer Vacation”. American Journal of Public Health Vol. 97, pp. 696-702.
U.S. Department of Health and Human Services. 2001. “Healthy People 2010: Volumes I and
II”.
31

(1) (2) (3)
DV: Ln BMI
DV: Ln BMI
Lunch Length -0.0011 -0.0013 30 min 0.0001
(s.e.) (0.0007)* (0.0007)** (s.e.) (0.0172)
R 2
0.3645 0.3887 31-44 min -0.0200
(s.e.) (0.0186)
Lunch Length -0.0037 -0.0039
(s.e.) (0.0018)** (0.0017)** 45 plus min -0.0398
(s.e.) (0.0234)*
R 2
0.0660 0.0993
Lunch Length -0.0009 -0.0013 R 2 DV: Overweight
DV: Obese
0.3893
(s.e.) (0.0013) (0.0013)
R 2
Table 1: OLS Estimates. Effect of Lunch Length on Body Weight (SNDA-III)
0.0870 0.1205
Controls
Controls
School Basic Col (1) + Col (1) +
Child Basic Col (1) + Col (1) +
Parent Basic Col (1) + Col (1) +
N 1,287 1,287 1,287
Standard errors are clustered at the school level and regressions are estimated using SNDA-III sample weights. The results in Column 1 condition on the following
covariates: month of interview, urbanicity, region, grade span, avg daily attendance, school participates in National School Breakfast Program, school has recess
family income, employment status of parents, age, sex, race, grade level, and child receives free or reduced lunch. The results in Column 2 condition on all the covariates
from Column 1 plus additonal controls at the child/parent and school level that may be correlated with lunch length and influence children's outcomes. At the school
level this includes whether students can eat eat on/off campus, whether the school has a wellness program, whether the school allows extra-curricular activities to
be scheduled during lunch, whether the school has vending machines, a school snack store, serves brand name fast food on campus, and average time spent in the
food service line. At the parent and child level this includes parents highest education, whether English is the main language spoken at home, parents marital status,
number of household memebers, whether a child bought lunch from school in the day prior to the survey, whether the child takes physical education, plays sports,
and number of hours spent watching television. The omitted category in Column 3 is less than 30 minutes. * denotes statistically different from zero at the 10%
level, ** at 5%, *** at 1%.
32

Table 2: TSIV-First Stage Estimates. Effect of State Lunch Laws on Lunch Length
(SHPPS+Auxiliary)
Offer Minimum Lunch Length (omit require) Avg Pupil-Teacher Ratio 0.379
Recommend -5.710 (s.e.) (0.190)**
(s.e.) (0.758)***
Ln Avg Teacher Salary 0.938
No Policy -6.603 (s.e.) (1.907)
(s.e.) (0.868)***
Percent Above Proficiency -0.071
Ban Junk Food in Vending Machines (omit require) (s.e.) (0.039)*
Recommend -4.244
(s.e.) (0.675)*** Percent Reduced/Free 0.054
(s.e.) (0.050)
No Policy 0.885
(s.e.) (0.524)* Percent White -0.046
(s.e.) (0.019)**
Ban Junk Food for Fundraising Sales (omit require)
Recommend -3.290 Percent Male 3.744
(s.e.) (0.765)*** (s.e.) (1.658)**
No Policy -0.127 Percent Age 10 to 14 -3.715
(s.e.) (0.774) (s.e.) (0.664)***
Limit Serving Size of Food Items (omit require) Percent Age 15 to 17 0.377
Recommend 2.897 (s.e.) (0.482)
(s.e.) (0.788)***
Percent of Obese Adults -0.612
No Policy -0.308 (s.e.) (0.125)***
(s.e.) (0.718)
State Requires Schools Follow PE Guidelines N 916
Require 1.740 R 2
0.1061
(s.e.) (0.549)***
F-Statistic [p-value]
H0: βrecommend, β no policy = 0 F(2,40) = 30.62
H0: βrecommend = β no policy F(1, 40) = 4.43
Standard errors are clustered at the state level and SHPPS weights are used. The first stage regression controls for all the
covariates listed in Appendix Table 2, as well as region dummies (NE,S,W). The coefficients on the region dummies are
omitted from this table for brevity. * denotes statistically different from zero at the 10% level, ** at 5%, *** at 1%.
33

Table 3: TSIV-Second Stage Estimates. Effects of Predicted Lunch Length on Body Weight
(YRBSS+SHPPS+Auxiliary Data)
Ln(BMI)
Overweight Obese
Predicted Lunch -0.0013 -0.0012 -0.0020 -0.0025 -0.0009 -0.0012
(s.e.) (0.0008) (0.0007)* (0.0014) (0.001)* (0.001) (0.001)
Year 2009 -0.0048 -0.0149 -0.0080
(0.002)*** (0.003)*** (0.004)**
Minority 0.0441 0.0895 0.0474
(0.003)*** (0.007)*** (0.004)***
Female -0.0261 -0.0705 -0.0656
(0.002)*** (0.003)*** (0.002)***
Age 0.0209 -0.0069 -0.0004
(0.0007)*** (0.002)*** (0.001)
Mean of Dependent Variable
Mean 3.1306 0.2769 0.1256
[s.d.] [0.1859] [0.4475] [0.3314]
N 186,831 186,831 186,831
Standard errors are clustered at the state level and regressions are estimated using YRBSS weights. The second stage regressions
control for all the covariates included in the first stage (except the instrument) which are listed in Appendix Table 2. We estimate the
second stage without and with demographic characteristics of the students in the YRBSS and a control for being in the 2007 or 2009
YRBSS survey. * denotes statistically different from zero at the 10% level, ** at 5%, *** at 1%.
34

Table 4: Difference-In-Difference Estimates.
Effect of State Lunch Laws on Body Weight
(YRBSS + SHPPS+ Auxiliary Data)
Ln(BMI)
(1) (2)
Requirement -0.001 -0.003
(s.e.) (0.004) (0.004)
Recommendation 0.012 0.011
(s.e.) (0.008) (0.008)
C0709 -0.009 -0.008
(s.e.) (0.005)* (0.005)*
Requirement*C0709 -0.025 -0.023
(s.e.) (0.008)*** (0.008)***
Recommendation*C0709 -0.009 -0.008
(s.e.) (0.003)*** (0.003)***
Individual Demographic Characteristics N Y
State School and Demographic Characteristics Y Y
N 230,329 230,329
R 2
0.01 0.05
F-statistic [p-value]
H 0: β req*c0709, β rec*c0709 = 0 F(2, 32) = 7.75 F(2, 32) = 6.86
[0.0018] [0.0033]
H 0: β req*c0709 = β rec*c0709 F(1, 32) = 3.95 F(1, 32) = 3.12
[0.0555] [0.0869]
Standard errors are clustered at the state level and regressions are weighted using YRBSS survey weights. All regressions
control for the covariates included in Appendix Table 2. Regression results in Column 2 additionally control for the age, race,
and sex of the YRBSS student. * denotes statistically different from zero at the 10% level, ** at 5%, *** at 1%.
35

DV: Ln(BMI) 1999-2001
Predicted Lunch 0.002
(s.e.) (0.001)
Year 2001 0.008
(s.e.) (0.002)***
Minority 0.045
(s.e.) (0.003)***
Female -0.044
(s.e.) (0.0021)***
Age 0.018
(s.e.) (0.0013)***
Mean of Dependent Variable
Mean 3.1160
[s.d.] [0.1800]
N 78,453
Standard errors are clustered at the state level, and regressions are estimated using YRBSS survey
weights. The regression results condition on all the controls listed in Appendix Table 2 in both
the first and second stage. * denotes statistically different from zero at the 10% level, ** at 5%,
*** at 1%.
Table 5: TSIV-Second Stage Estimates. Falsification Test.
Effect of Predicted Lunch Length on Prior BMI
(YRBSS+SHPPS+Auxiliary Data)
36

Table 6: Characteristics of Schools, by 2006 State Lunch Length Policies
Require Recommend Neither F-stat [p-value] N
2006 SHPPS
Cafeteria/Gym Overcapacity 0.0616 0.0431 0.0564 F(2, 40) = 1.11 904
[s.d.] [0.2449] [0.2033] [0.2311] [0.3382]
Ban Junk Food Vending 0.5783 0.4486 0.4951 F(2, 40) = 1.84 862
[s.d.] [0.5026] [0.4979] [0.5007] [0.1717]
Minutes of Recess Time Per Week (conditional on schools offering recess)
139.1788 149.1318 140.0770 F(2, 37) = 0.04 291
[37.4167] [65.801] [59.5422] [0.9615]
Minutes of Physical Education Per Year
(conditional on schools requiring physical education for grade promotion)
5204.13 4127.88 3750.33 F(2, 40) = 0.31 528
[s.d.] [2571.72] [2487.32] [2105.69] [0.7338]
2000 SHPPS
2000 Lunch Lengths 21.0540 23.7600 24.3210 F(2, 43) = 0.45 822
[s.d.] [4.540] [7.754] [8.018] [0.6409]
Standard deviations are in parentheses and SHPPS weights are used. The sample size differs across outcomes
due to missing information. The F-statistic displays the results of a test of the joint significance
of state lunch law dummies from regressions that condition on all the variables in Appendix Table 2.
37

Breakfast Lunch Dinner Snacks
30 minutes 9.584 58.702 -88.200 -25.656
(s.e.) (21.580) (23.614)*** (37.243)** (38.711)
31 to 44 -22.067 46.261 -47.352 -19.807
(s.e.) (21.999) (26.309)* (36.710) (36.173)
45 plus 43.796 88.690 -98.963 36.236
(s.e.) (27.724) (36.577)*** (56.664)* (44.849)
Mean 336.407 608.461 646.248 494.509
(262.196) (307.673) (430.783) (444.152)
R 2
Table 7: OLS Estimates. Effect of Lunch Length on Calorie Intake (SNDA-III)
DV: Calories Consumed During Meals
0.2007 0.1437 0.1149 0.1105
N 1211 1211 1211 1211
Standard errors are clustered at the school level, and SNDA-III sample weights are used. All regressions control for the covariates
in Appendix Table 1. The omitted category is less than 30 minutes * denotes statistically different from zero at 10% level, ** at 5%, *** at 1%.
38

Density
0 .05 .1 .15
Source: SNDA-III
Density
0 .05 .1
Density
.15 .2
0 .05 .1 .15 .2
Source: SHPPS
Figure 1: Length of School Lunch Period (SNDA-III)
DV: Overweight (>=85th)
DV: Obese (>=95th)
0 20 40 60 80 100
lunchperiodlength
Figure 2: Length of School Lunch, Once Seated (SHPPS)
10 10 20 20 30 30 40 40 50 50 60 60
lunchperiodlength_onceseated
39

Density
0 .1 .2 .3
Source: SHPPS
Figure 3: Length of School Lunch, Once Seated by State Law (SHPPS)
10 20 30 40 50 60
lunchperiodlength_onceseated
Require Recommend
No Policy
40

Appendix Table 1: Summary Statistics (SNDA-III)
Outcomes & Lunch Length
Ln BMI 3.0361 Lunch Length (minutes) 32.6800
[s.d.] [0.2400] [9.0197]
Obese 0.2116 Lunch Length Is Averaged 0.0656
[0.4086] [0.2477]
Overweight 0.3888
[0.4877]
School Characteristics
Grade Span: K to 2 0.4778 Eat in Cafeteria 0.9572
[0.4997] [0.2024]
3 to 5 0.5214 Eat on Campus 0.3677
[0.4997] [0.4823]
6 to 8 0.4268 Eat off Campus 0.1218
[0.4948] [0.3272]
9 to 12 0.2851 School has Wellness Program 0.403
[0.4516] [0.4907]
Average Daily 803.3729 Vending Machine 0.6308
Attendance [686.0951] [0.4828]
School Participates in National 0.8728 School has Store or 0.2051
School Breakfast Program [0.3333] Snack Bar [0.4040]
School offers Recess 0.5173 Average time spent in service 5.4764
[0.4999] [3.8432]
Other Activities Scheduled 0.2474 Fast Food Brand 0.2756
[0.4316] [0.4470]
Child Characteristics
Age 11.5776 Receive Reduced or 0.4311
[3.4249] Free Lunch [0.4954]
Female 0.5049 Ate school lunch day 0.6871
[0.5002] before survey [0.4638]
Minority 0.3256 Regularly eats breakfast 0.7889
[0.4688] [0.4083]
Grade Level* 5.9579 Parent or adult makes 0.7647
[3.3401] breakfast [0.4244]
Takes Physical Education 0.7875 Has Allergies 0.0769
[0.4092] [0.2666]
Plays on Sports Team 0.5854 Hours of TV per Day 1.7690
[0.4928] [1.2908]
41

Appendix Table 1 (cont.): Summary Statistics (SNDA-III)
Parent Characteristics
Income* Parent's Highest Education*
15 to 30 K 0.1907 High School 0.2389
[0.3930] [0.4266]
30 to 40 K 0.1006 Some College 0.3635
[0.3009] [0.4812]
40 to 60 K 0.1528 College 0.1741
[0.3600] [0.3794]
60 to 80 K 0.1508 Professional 0.1108
[0.3580] [0.3141]
80 to 100 K 0.0777 Parents Married 0.7299
[0.2678] [0.4442]
100 Plus K 0.1715 Number Household 4.3632
[0.3771] Members [1.3292]
At least one parent 0.9324 Household speaks English 0.8616
employed [0.2511] [0.3455]
Geographic & Interview Date Characteristics
Region* Urbanicity*
Midwest 0.1747 MSA, Not Central City 0.3744
[0.3799] [0.4842]
Mountain 0.0849 Not MSA 0.2049
[0.2788] [0.4038]
Month of BMI Data Collection*
Northeast 0.0794 February 0.1919
[0.2705] [0.3939]
Southeast 0.2163 March 0.2653
[0.4119] [0.4417]
Southwest 0.1754 April 0.1977
[0.3804] [0.3984]
West 0.1484 May-June 0.2870
[0.3556] [0.4525]
*The omitted categories are less than $15 K, mid-Atlantic, less than high school, MSA Central
January. In the regressions, grade level enters as a series of dummy variables for grades 1-12
with omitted category grade 1. The sample size is N=1,287. Weighted means with
their sample standard deviations are reported using SNDA-III weights.
42

Appendix Table 2: Summary Statistics (SHPPS + Auxiliary Data)
State Policies State Level School Characteristics
Lunch Length (school level) 22.6779
Offer Minimum Lunch Length (omit require) [7.4253]
Recommend 0.6019 Avg Pupil-Teacher Ratio 22.6779
[0.4898] [7.4253]
0.3722 Ln Avg Teacher Salary 15.7755
[0.4837] [2.5099]
Ban Junk Food in Vending Machines (omit require) Overflow 19.4337
Recommend 0.2960 [5.4632]
[0.4567] State Level Student Characteristics
Neither 0.2747 Ln Average Enrollment 14.1297
[0.4466] [0.8867]
Ban Junk Food for Fundraising Sales (omit require)
Recommend 0.3908 Above Proficiency 33.8764
[0.4882] [7.0503]
Neither 0.4204 Percent Reduced/Free 40.7754
[0.4939] [8.1535]
Offer Recess to Elementary Students (omit require)
Recommend 0.1287 Percent Black 15.7768
[0.3351] [9.8450]
Neither 0.6861 Percent White 61.7451
[0.4643] [17.0800]
Limit Serving Size of Food Items (omit require)
Recommend 0.4011 Percent Male 51.2020
[0.4904] [0.11534]
Neither 0.3331 Percent Age 10 to 14 38.3978
[0.4716] [0.4267]
State Requires Schools to Serve Breakfast
0.1037 Percent Age 15 to 17 24.4447
[0.3051] [0.7525]
State Requires E/M/HS to Offer PE State Level Demographic Characteristics
0.8885 Ln Median HH income 10.8528
[0.3150] [0.1188]
State Requires Schools Follow PE Guidelines Percent of Obese Adults 26.9239
0.6719
[0.4698]
[2.6343]
N=916. Weighted means with their sample standard deviations are reported using SHPPS
weights. Regressions include indicators for region (NE,S,W) but are omitted for brevity.
43

Ln BMI 3.1306 Minority 0.3791
[s.d.] [0.1859] [0.4852]
Overweight 0.2769 Female 0.4892
[0.4475] [0.4999]
Obese 0.1256 Age 16.0679
[0.3314] [1.2136]
Year 2009 0.5099
N 186,831
Weighted means with their sample standard deviations are reported using state YRBSS weights
Year 2009 is equal to 1 for YRBSS observations from 2009, and =0 for year
2007 observations.
Appendix Table 3: Summary Statistics (YRBSS)
44