Content uploaded by Carlisle Runge
Author content
All content in this area was uploaded by Carlisle Runge on Apr 06, 2020
Content may be subject to copyright.
CORBETT GRAINGER, BENJAMIN SENAUER, AND
C. FORD RUNGE
Nutritional Improvements and Student Food Choices
in a School Lunch Program
This study analyzed data on students’ food purchases linked to their
school records to examine factors affecting the healthiness of their food
choices and the impacts of reforms to promote healthier eating in a high
school lunch program. U.S. Department of Agriculture’s Healthy Eat-
ing Index was used to evaluate the nutritional quality of the foods pur-
chased, as well as an alterative ranking developed by the school
dietitian. The new lunch program was associated with an improvement
in the nutritional quality of students’ food choices. Girls tended to pur-
chase relatively healthier food than boys, but male students had
a greater improvement in the healthiness of their food choices.
Ninety-nine percent of U.S. public schools and 83% of public and pri-
vate schools combined participate in the National School Lunch Program
(NSLP). These schools receive cash subsidies and commodities from the
U.S. Department of Agriculture (USDA) for each meal served. In return,
they are supposed to satisfy USDA’s nutritional requirements and provide
free or reduced-price lunch to eligible children. During 2004–2005, the
reimbursement rates for free, reduced-price, and paid lunches were
$2.24, $1.84, and $0.21, respectively (USDA, Food and Nutrition Service
[FNS] 2005b, 2005c). In 2004, U.S. schools served 4.8 billion lunch meals,
reaching over 25 million children (Guthrie 2003; USDA, FNS 2005c). For
many, especially students from lower-income families enrolled in federally
subsidized free and reduced-price lunch programs, these are the main meals
of the day. In addition to these federally subsidized meals, millions of
Corbett Grainger was an MS student in the Department of Applied Economics when this work was
completed (grainger@econ.ucsb.edu). He is currently a PhD student in economics at the University of
California, Santa Barbara, CA. Benjamin Senauer is a professor in the Department of Applied Econom-
ics and co-director of The Food Industry Center at the University of Minnesota, St. Paul, MN
(bsenauer@umn.edu). C. Ford Runge is Distinguished McKnight University Professor of applied
economics and law, and director of the Center for International Food and Agricultural Policy at the
University of Minnesota, St. Paul, MN (frunge@umn.edu).
This research was supported by a grant from the McKnight Foundation. The authors thank Bertrand
Weber and Michele Wignal for their assistance. They were the director and dietician for the Hopkins
School Food Service at the time of this study.
The Journal of Consumer Affairs, Vol. 41, No. 2, 2007
ISSN 0022-0078
Copyright 2007 by The American Council on Consumer Interests
WINTER 2007 VOLUME 41, NUMBER 2 265
students choose foods from a` la carte offerings provided by most school
food services. These are not part of the NSLP and are referred to as ‘‘com-
petitive foods’’ by USDA.
The American Obesity Association (2005) reported that 30.3% of chil-
dren ages 6–19 were overweight or obese in 2000. This figure tripled from
the early 1970s to 2000. Many factors are behind this trend, but there is
evidence that meals served in many schools contributed to poor nutrition
and obesity (Cooper and Holmes 2006). In a nationwide assessment,
USDA found that the average lunch served in 1998–99 met its dietary
guidelines of fewer than 30% of total calories from fat and less than
10% from saturated fat in only one out of seven secondary schools (USDA,
FNS 2001). Many secondary students eat high-fat cheeseburgers, French
fries, and pizzas on a daily basis. In some cases, they eat only a candy bar
and a super-sized soft drink. Two reporters visiting six schools in New
York City and Montgomery County, Maryland (an affluent Washington,
DC, suburb), observed hundreds of students eating lunch and saw only five
who took the green vegetable offered with the full meal (Becker and Burros
2003).
School food service directors often must focus on ease of preparation
rather than healthy options because they lack both the skilled staff and facil-
ities necessary to do more. Many schools depend on major food service
vendors to supply highly processed foods that require little more than heat-
ing to prepare. Faced with tight budgets, food service operations are driven
by cost considerations and a need to serve what students will eat. Major fast
food chains’ products are being served in a substantial number of schools,
because that is what students want. School districts have also signed vend-
ing contracts with snack and soda companies for the commission generated
to fund their programs. Schools have begun to be heavily criticized for not
providing healthier food choices and guiding students toward healthy life-
long eating habits (Cooper and Holmes 2006; USDA, FNS 2001).
Recently, some school districts have taken steps to change their food
programs. A 2005 USDA study, Making It Happen! School Nutrition Suc-
cess Stories, reports on 32 schools and school districts that have made inno-
vative changes to improve the nutritional quality of their food programs.
The encouraging message from these case studies is that ‘‘students will buy
and consume healthful foods and beverages, and schools can make money
from healthful options’’ (USDA, FNS 2005a, executive summary). Hop-
kins School District 270 in Minnesota has been one of the innovators.
Its food service program has improved the quality of food served, the vari-
ety of healthful choices available, and the school food service environment.
Moreover, the Hopkins food service program demonstrated its financial
266 THE JOURNAL OF CONSUMER AFFAIRS
viability. The program is able to operate on a revenue-neutral basis, not
requiring a subsidy from the school district (Royal Cuisine 2006). Because
of these changes, Hopkins High School provided an opportunity to study
the lunch choices of adolescents and how they might be changed.
In previous research using cross-sectional data, Akin, Guilkey, and
Popkin (1983) found that students who participated in the NSLP had higher
intakes of vitamins and minerals over a 24-hour period compared to non-
participants and that the impact was stronger for low-income children. In
a similar study, Gleason and Suitor (2003) examined the effect of NSLP
participation on students’ dietary intakes, also in comparison to nonparti-
cipants. Using a fixed-effects model, the researchers found that NSLP par-
ticipation had an improvement in the 24-hour intake of six vitamins and
minerals. Participants consumed higher levels of fat but fewer added sugars
than nonparticipants.
A relationship between childhood obesity and school feeding programs
was established by Anderson and Butcher (2005), who linked the availabil-
ity of snack foods and beverages in schools to adolescent obesity. They
found that an association between schools with vending machine contracts
and increased body mass index levels. The nutrition literature suggests that
interventions to promote healthy eating can work. Studies typically focus
on evaluating the impact of controlled interventions designed by the
researchers that are implemented on a trial or temporary basis. For example,
Perry et al. (2004) and French et al. (1997) assigned schools randomly to
a control or an intervention group, in which interventions were successfully
initiated to encourage fruit and vegetable consumption.
Simply offering nutritious foods does not necessarily improve students’
diets, if the foods are not appealing. Students will avoid these choices or
leave them uneaten. Ralston, Buzby, and Guthrie (2003) outlined strategies
to increase the appeal of school lunches and breakfasts, suggesting that
schools expand the offerings available, allow student input in food service
decisions, improve the selection of USDA commodities, increase the use of
fresh produce and local foods, and improve methods of preparation.
Beyond the food itself, the school lunchroom is frequently chaotic and stu-
dents must rush to eat in the very short time allotted.
With access to a unique data set, we examined the school lunch choices
over three years (2002–2005) in Hopkins High School, located in a western
suburb of Minneapolis, and the impact of the food service changes on meas-
ures of nutritional quality. Students at Hopkins use debit accounts and per-
sonal identification numbers to purchase food. This point-of-sale (POS)
data allowed students to be linked with their purchases on the cash registrar
records and with their demographic and other information through their
WINTER 2007 VOLUME 41, NUMBER 2 267
school records. USDA’s Healthy Eating Index (HEI) was used to evaluate
the nutritional quality of the specific food items offered. An alternative
nutritional ranking developed by the school dietitian was also utilized,
for reasons explained below. This approach provided a measure of the
overall healthiness of foods purchased for lunch by a student over a
10-day period. Econometric methods were then used to examine factors
affecting a student’s food choices and whether they changed after the
program innovations.
To our knowledge, this is the first study to use POS data linked to stu-
dents’ records to examine the factors affecting the healthiness of students’
food choices and the impacts of comprehensive reforms to promote health-
ier eating in a school lunch program.
Although it is a case study of changes in one school’s food service oper-
ations, Hopkins is in many respects a typical suburban school district. Les-
sons for many other schools nationwide might be drawn from its food
service reforms.
THE HOPKINS FOOD PROGRAM
The food service program in the Hopkins School District underwent dra-
matic changes after 2003. Hopkins is typical of many older, inner-ring sub-
urbs in major American metropolitan areas and is not particularly wealthy.
Most of its students come from middle rather than upper-income house-
holds, and there are an increasing number of minority students and
lower-income families. Hopkins food service was also typical in providing
a` la carte foods in addition to the NSLP lunches, as do 80% of schools
nationally (CDC 2000). Until 2003, foods in the main cafeteria line were
easy-to-prepare and a` la carte items, included pizza from Pizza Hut and
Domino’s, cheeseburgers, chips, and high-calorie beverages. All foods
were served on disposable trays. Vending operations were contracted
out and machines were stocked with soda, candy, and potato chips. By
the end of school year 2002, the district superintendent, with the support
of the school board, set goals of providing quality foods in an appealing
environment in the district’s cafeterias, which serve a total of about 9,000
students daily. To implement a new program, in July 2003, the district hired
a new food service director who had a professional restaurant and hotel
food service management background.
The innovations began at the high school, opening an optional window
called the Health Nut Cafe´, which served only foods free of trans-fats, high
in fiber with low levels of sugars, including high-fructose corn syrup. The
offerings included organic whole-grain cookies, 100% juice drinks, and
268 THE JOURNAL OF CONSUMER AFFAIRS
freshly made salads and sandwiches. Later that year, the vending contract
with Pepsi was canceled, which had been a lucrative source of unrestricted
cash. Rather than eliminating the machines, the Hopkins Schools bought
them, filling them with water and 100% juices. After the voters approved
a bonding initiative, a new food service kitchen and lunchroom area were
built at the high school. The new program, named Royal Cuisine, dramat-
ically changed food service operations in all schools in the district, although
the most profound changes were visible at Hopkins High School.
The new program focused on preparing more nutritious foods on site. No
longer were meals simply warmed prior to service; cooking from scratch
became the norm and students ate from plates with flatware. In the high
school, in addition to the Health Nut Cafe´, a new food service window,
the Diner, offered full meals beginning in 2003–2004. In 2004–2005,
the Tuscan Oven and Ethnic Adventures, both of which are NSLP meal
options, were introduced in the high school cafeteria. Starting in fall
2004, the switch was made to low-fat salad dressings and cheese (including
pizzas) and whole-wheat breads and pizza crusts. In addition, more vege-
tables were added to many dishes, including the pizza toppings.
Because preparing foods from scratch and providing more fresh fruits
and vegetables is more expensive, prices increased. The price at Hopkins
of the NSLP meal was $2.05 in 2002–2003, $2.30 in 2003–2003, and $2.50
in 2004–2005. The food service director’s experience in the private sector
led him to focus on increasing efficiency and productivity in the kitchens to
keep costs down. Although the capital expenditure required to start such
a program was large, variable costs such as labor did not increase substan-
tially due to efficiency gains. Hopkins had a ‘‘closed-campus’’ policy,
which means that students could not go off campus to buy food and the
school food service did not have to compete against fast food outlets
and others. Nationally, 94% of elementary schools and 72% of high schools
have closed campuses (CDC, National Center for Chronic Disease Preven-
tion and Health Promotion, Division of Adolescent and School Health
2006).
METHODS
Each student in the Hopkins School District had a personal spending
account with an associated personal identification number used for purchas-
ing food. Students and their families deposited money into their accounts
on a rolling basis. On any given day, a student purchasing food items in the
school entered his or her personal identification number at the point of pur-
chase and a cashier entered the item numbers of the foods chosen. Sales
WINTER 2007 VOLUME 41, NUMBER 2 269
data were then kept electronically. The data used in this analysis can be
grouped into four main categories: sales, expenditure, student demographic
data, and nutritional ratings.
Sales Data
Sales data came from the Hopkins food service program. POS reports
were run for a random sample of students using their personal identification
numbers for November 2002, 2003, and 2004 as well as April 2003, 2004,
and 2005. Because of the demands on the school district staffs’ time to
assemble the data, we were only able to obtain data for a sample of students,
who were sophomores in 2002–2003, juniors in 2003–2004, and seniors in
2004–2005. Since the major changes implemented in Hopkins came after
the new food service director started in July, 2003, the school year 2002–
2003 was treated as a baseline.
Daily data on the item number and the quantity purchased by each stu-
dent were provided in the POS report. The item number corresponded to the
food purchased for some items and represented a food category in other
cases. For example, item number 647 was a hot dog, but item number
643 represented ‘‘$2.00 miscellaneous,’’ a category designed to capture
all items costing $2.00 that did not have a separate key on the cash register.
For this reason, it was not possible to determine what item was purchased if
643 appeared in the POS report for a student. The cash registers had a lim-
ited number of keys, and minimizing the number of key strokes speeded up
the checkout process. However, most item numbers corresponded to a unique
food product, providing a rich data set of historical a` la carte purchases.
The full NSLP meals also appeared in the POS reports. However, only
for academic year 2004–2005 was it possible to distinguish which entre´e
the student purchased. The side items that supplement the entre´e to create
a full NSLP meal did not appear in the data since they were not keyed in by
the cashier. In the first two years (i.e., 2002–2003 and 2003–2004), the data
indicated whether the student purchased an NSLP lunch but not which one.
Expenditure
NSLP meal and a` la carte item prices were obtained from the food service
and entered into a database. In the analysis to follow, ‘‘expenditure’’ refers
to the total value of the foods purchased by a given student for the specified
time frame. Students receiving free or reduced-price lunches paid less than
the expenditure shown. In these cases, expenditures correspond to the total
value of the foods purchased.
270 THE JOURNAL OF CONSUMER AFFAIRS
Demographic Data
Demographic data for each student were also gathered. With the coop-
eration of the school district office, access to student demographic data was
granted. Using only the student’s ID number to preserve privacy, students’
purchases were matched to his or her demographic characteristics. Table 1
summarizes the demographic variables by month and year, together with
the number of observations in each category.
Female and male dummy variables (with the latter omitted) were used in
the statistical analysis. With cohort data, age or grade was not a variable in
a given year since the students were all sophomores in 2002–2003, juniors
in 2003–2004, and seniors in 2004–2005. For each student, the school dis-
trict held data regarding ethnicity/race. Ethnicity/race was divided into five
groups: white (omitted), Asian, black, Hispanic, and American Indian/
Alaskan/Pacific Islander, which we term other for the sake of brevity. These
were used as indicator variables.
Finally, the district provided data on the federal assistance each student
received for school lunch through the NSLP, that is, if the student received
a free or reduced-price lunch. These were combined to create the low-income
indicator variable. This served both as an indication of the subsidy received per
student and as a proxy for that family’s household income, on which eligibility
for free or reduced-price meals was based. An average of 6.9% of the students
across the six time periods received free or reduced-price NSLP lunches.
Nationwide, 9.8% of NSLP participants received free or reduced-price meals
in 2003, 9.9% in 2004, and 10.0% in 2005 (U.S. Census Bureau 2007).
Hopkins High School is thus somewhat lower than the national average.
TABLE 1
Demographic Data Summary for Hopkins High School Students
November
2002
April
2003
November
2003
April
2004
November
2004
April
2005
Sample size 515 561 566 555 523 480
Female 254 274 283 280 250 233
Male 261 287 283 275 273 247
Indian 666432
Asian 26 28 24 27 27 26
Hispanic 17 16 11 13 16 16
Black 33 32 51 47 34 34
White 433 479 474 464 443 402
Reduced 11 12 13 14 15 9
Free 22 17 33 22 26 28
Note: The demographic data cover students from Hopkins High School in Hopkins, MN. These samples
were randomly drawn for each month and year combination.
WINTER 2007 VOLUME 41, NUMBER 2 271
Nutritional Ratings and Relative Healthiness Index
The fourth type of data concerned nutritional ratings of the foods pur-
chased. For this measure, we constructed two indexes of the healthiness of
student food choices. The first relative healthiness index (RHI) was based
on the ratings of the school dietitian, and referred to as the ‘‘dietitian-
developed index’’ (RHI-DDI). The second ranked foods using the USDA’s
HEI (RHI-HEI). The two indexes are described below.
RHI-DDI
One of the key dependent variables was developed in conjunction with
the school district’s dietitian/nutritionist. This index was constructed by
considering every a` la carte item in each of the three years and every NSLP
lunch entre´e in the 2004–2005 school year. Each item was rated ‘‘more
healthy’’ and given a rating of 11, ‘‘less healthy’’ (21), or not clearly either
(0), based on the nutritional content of the foods. The dietitian worked in
the high school over the entire time period and used a computerized nutri-
tional menu planner, so was able to verify the nutritional content ratings. As
discussed in the sales section, because each entr
´ee could be paired with
multiple side items, we considered the entre´e’s rating only in the case
of the NSLP meals.
Since the data covered food purchases, what the students actually con-
sumed was not known with certainty. However, in multiple visits to the
school district lunchrooms, the authors observed very little food being
thrown out. The food service staff confirmed this observation. Because stu-
dents had a wide range of choices, even for the full NSLP meals, they sat-
isfied their food preferences and had relatively little plate waste.
Both positive and negative criteria were used to rank foods. Positive fac-
tors included whether the food was made with whole grains, contained mul-
tiple fruits and vegetables, and was minimally processed. Negative factors
included trans-fatty acids content, high other fats content, and high levels
of sugars including high-fructose corn syrup. In general, foods rated health-
ier were lower in fat and had more nutrients per calorie, whereas less health-
ful foods were higher in fat, more energy dense, and had relatively low
nutrient value. These ratings were then tied to the item numbers from
the sales data.
To provide a feel for how the index worked, consider some food items
that were given a healthier rating of 11: milk (1/2 pint), yogurt (4 oz.), veg-
etables with dip, fresh fruit, and a mixed salad. Items rated a 21included
ice cream, milk shakes, cinnamon rolls, chocolate chip cookies, and shrimp
272 THE JOURNAL OF CONSUMER AFFAIRS
poppers. Examples of foods given a neutral rating included a super pretzel,
coffee, apple cider, an English muffin, and a ham and cheese sandwich.
Once each food was ranked, the RHI-DDI was constructed. For each day
in each month, the average rating of the items purchased by a student was
calculated and defined as the ‘‘daily average rating’’ for student i. To
smooth out day-to-day fluctuations in choices, a measure was needed that
included multiple days in a given month. For each month, the 10-day sum
of these ‘‘daily average ratings’’ was calculated for each student, over the
first 10 days of a student’s purchases.
1
This sum is the RHI-DDI, which
ranged from –10 to 110. In one specification, this included only a` la carte
items; in the specification for the 2004–2005 school year, it included NSLP
meal entre´es and a` la carte items. To differentiate between the two spec-
ifications, we denoted the latter as RHI-DDI
meal
.
It is important to note that the actual observed days for each student may
vary because to constrain this measure to the same days for all students
would result in a drastic reduction in sample size. Although the RHI-
DDI used the first 10 days in each month in which students purchased food,
a` la carte choices available on any day of the month did not change, so that
the choices on any given day varied only in the entre´e selection. All a`la
carte items as well as grill, deli, and salad options were consistently offered
over each time period, although the food service steadily improved the
quality of foods offered. It is important to emphasize that while the food
offerings generally became more healthful over time, there were still foods
available at the high school with very low nutritional ratings in 2004–2005.
Students were not forced to choose healthier foods simply because less
healthy choices were eliminated.
Because the RHI-DDI score provided by school district’s dietitian may
be open to challenge, its validity was checked by using an alternative mea-
sure based solely on the nutritional components of the foods developed by
USDA. We employed both specifications when possible, but when consid-
ering NSLP meals, the RHI-DDI was used because of data limitations.
RHI-HEI
For this measure, we used the 12 components from the USDA’s HEI to
rate each food on a 0–100 scale, based on the nutritional content of the
1. The choice of 10 days was somewhat arbitrary—summing over five days did not create as much
variation in the constructed dependent variable as 10 days, and more than 10 days reduced the number of
students considered each month considerably. The same analyses were conducted using different time
periods, and the number of days did not affect the results significantly.
WINTER 2007 VOLUME 41, NUMBER 2 273
foods (USDA, CNPP 2005; see the appendix). The HEI was used to give
a score to a person’s lunch diet, but because information on what students
were eating outside the school was unavailable, the HEI could not rate a stu-
dent’s total diet as traditionally done (for an example of its traditional use,
see Bowman, Gerrior, and Basiotis 1998, and Variyam, Blaylock, and
Smallwood 1998). Given the detailed data on a` la carte purchases, these
foods could be rated using the HEI. Information available on the NSLP
entre´es in 2004–2005 allowed only the RHI-DDI to be used.
Each a` la carte item was given a score based on the 12 components of the
HEI. If an item met the standard, it was given full points for that category.
For example, a serving of pure orange juice was given 5 points for the fruit
category, 0 points for the whole fruit category (because it is a juice), 0 for
most other categories, but 10 points for low sodium and saturated fat and 20
points for calories, for a total of 45 points. Every a` la carte item received an
HEI score. Scoring criteria are outlined in the appendix.
After rating each food, the RHI-HEI was constructed. For each day in
each month, the average rating of the a` la carte items purchased by a student
was calculated as the ‘‘daily average rating’’ for student i. For each month,
the 10-day average of these ‘‘daily average ratings’’ was calculated for the
first 10 days the student made a` la carte purchases. This 10-day average is
the RHI-HEI utilized.
Descriptive statistics by month are shown in Table 2. Both indexes
clearly trend upward on average over time, showing that students were
making healthier food choices. In November 2002, the average RHI-
DDI was –1.237, which improved to 2.571 by April 2005. Likewise,
the mean RHI-HEI increased from 23.062 to 27.656. Both indexes fell
in April 2004 without explanation. The minimum RHI-DDI in each time
period was –10 suggesting substantial variation in student eating behavior,
including continued purchases of less healthy food. For three periods, the
maximum value was 10, again supporting a wide range of behavior. The
minimum RHI-HEI was less than 10 in every month and as low as 0.75 in
April 2004. The maximum was 45 or more in three of the months. The
correlation between the two indexes across all months was 0.6554, indi-
cating that RHI-DDI was a reasonable measure of the healthiness of a stu-
dents’ food choices.
Given the grade cohort sample, it is possible that improvements in nutri-
tion could be related to increases in age and maturity. However, USDA
research showed that the nutritional quality of children’s and adolescents’
diets deteriorates as they grow older (Devaney, Gordon, and Burghardt
1993). The steady improvement in the RHIs in Table 2, rather than a sharp
improvement between 2002–2003 (prior to the new program) and
274 THE JOURNAL OF CONSUMER AFFAIRS
2003–2004 (after the new program commenced), should not be surprising
given that the changes in food service operations were introduced gradually.
Econometric Methodology
As noted above, for each month/year combination, repeated random
samples were drawn from the same cohort of students who were sopho-
mores in 2002–2003, juniors in 2003–2004, and seniors in 2004–2005.
Hence, the number of observations for each student varies by individual
and forms an unbalanced panel. Although desirable methodologically, suf-
ficient research funding was not available to identify and then obtain and
analyze data from a ‘‘control’’ school, in which the food program remained
unchanged during this period. The RHI-DDI and RHI-HEI were calculated
only for students purchasing one or more a` la carte items on at least 10 days
during a given month, and the first 10 days for each student were used in the
calculation. Data contained entries for each of the six months under con-
sideration, with 550 individuals appearing at least once, and the average
student observed in 4.3 time periods.
The central research question is whether the improvements in the school
lunch offerings had an effect on the overall nutritional quality of the
TABLE 2
Descriptive Statistics for the RHIs for A
`La Carte Foods
November
2002
April
2003
November
2003
April
2004
November
2004
April
2005
RHI-DDI
Mean 21.237 20.791 0.559 0.469 1.930 2.571
SD 2.883 3.154 3.568 3.317 3.779 3.681
Min 210 210 210 210 210 210
Max 8.5 9.17 10 9.5 10 10
RHI-HEI
Mean 23.062 23.303 24.166 19.976 27.807 27.656
SD 4.551 4.666 6.082 6.009 6.697 6.259
Min 8.5 8.458 9.417 0.75 6.5 3.5
Max 40.8 43.708 45 33.833 45.25 45
n334 349 324 313 283 257
Note: The RHI is a measure of the healthiness of a student’s a` la carte choices over a 10-day period. The
description of its construction is discussed in detail in the data section. RHI-HEI is based on the criteria
used in the USDA’s HEI and RHI-DDI is based on the school-dietitian-developed index. Min and Max
refer to the minimum and maximum observed value of RHI, respectively. The possible range for RHI-
DDI is 210 to 110 and for RHI-HEI is from 0 to 100. However, given the construction of RHI-HEI and
the food components evaluated, it would be very unlikely to have an actual food item with an RHI-HEI
greater than 45–50. The sample includes all students who purchased a` la carte items on at least 10 days
during the given month.
WINTER 2007 VOLUME 41, NUMBER 2 275
students’ lunch choices. In addition, how did factors such as gender,
ethnicity/race, and expenditure affect the healthiness of students’ school
lunch choices? Finally, did different groups of students, such as males
and females, respond differently to the lunch program changes? To an-
swer these questions, the two measures of RHI were used as the dependent
variables in a series of regressions.
The regression equation specified was:
RHIit ¼a1x0
itb1cTrendt1ui1eit ð1Þ
where x
it
is a vector of demographic characteristics of student iin period t,
and Trend
t
is a variable that assigns a value of 1 to the first month of the
analysis (November 2002), a 2 to the second, 3 to the third, and so on, to test
whether there was a constant (linear) improvement in RHI over time. We
also included multiplicative interaction terms, trend Asian, trend
black, trend Indian, trend Hispanic, and trend male. These tested
for trends for each of the respective demographic groups in terms of eating
healthier foods. In an alternative specification (not reported here), we
included month dummies rather than the trend variable and interactions.
The results using that model were very similar to those reported below
and are available from the authors upon request.
Equation (1) was estimated under two sets of assumptions. First, we
assumed that u
i
is uncorrelated with the explanatory variables. This is
the random-effects specification. We also estimated equation (1) assuming
that u
i
can be arbitrarily correlated with the explanatory variables (i.e., it is
a student-specific fixed effect). In the case of student-specific fixed effects,
time-invariant explanatory variables are excluded from the estimation
(Wooldridge 2002). Equation (1) was estimated under both sets of assump-
tions using Stata 8.1.
Ninety-six students were observed in just one month, which in the fixed-
effects model is equivalent to running the regression for the panel without
these observations. If the random-effects assumptions are valid, these
observations could still be used in the estimation. The panel was unbal-
anced due to randomly sampling individuals each month, so these students
were unlikely to be different from those who appear more often. As a check
on the specifications presented, regressions were run omitting students with
fewer than three months of observations and results did not change signif-
icantly. The sample was also restricted to create a balanced panel and we
again found that results did not change substantially.
276 THE JOURNAL OF CONSUMER AFFAIRS
EMPIRICAL RESULTS
Table 3 presents results for the two indexes of relative healthiness for a`
la carte purchases over the six month/year periods. For RHI-HEI, results for
the random- and fixed-effects models of equation (1) are shown in columns
1 and 2. For the RHI-DDI, results for the random- and fixed-effects models
are shown in columns 3 and 4. R
2
are low, which is not unusual in the
analysis of individual consumer behavior. Lower R
2
with fixed effects
are also expected. To test for the presence of random effects, the Hausman
test was performed by comparing each pair of estimates (i.e., columns 1 and
2, and columns 3 and 4) (Hausman 1978). The test for the regressions using
RHI-HEI (columns 1 and 2) in Table 3 did not reject random effects
(pvalue ¼.5003), nor did the test reject random effects when comparing
the results of the RHI-DDI regressions in columns 3 and 4 (pvalue ¼.1034).
Since the random-effects specifications allowed us to examine the role of time-
invariant demographic variables, discussion is focused on these results.
Results summarized in Table 3 show that expenditure (the amount spent
by a student over the 10-day period) was positively correlated with RHI in
the random-effects regression in all specifications, plus in the fixed-effects
regression using RHI-HEI in column 2. This was expected since more
healthful foods were generally more expensive. The coefficient for female
was positive and significant in both random-effects specifications; females
were more likely to make healthier food choices than white males, the omit-
ted group reflected in the intercept. This finding is consistent with previous
studies tying gender to healthy eating (USDHHS 2000). In terms of the
ethnicity/race variables, Asians made healthier food choices than white
males. In column 1, the variable black had a negative effect, but was sig-
nificant only at the 10% level and not significant in column 3. Low income,
reflecting whether students received a free or reduced-price NSLP meal,
also did not have a significant effect. This result was encouraging because
it means that poorer students, who receive assistance buying lunch, were
able to make food choices that are as healthy as other students.
The linear time trend variable (trend) was positive and significant at the
1% level in every specification in Table 3. This provides clear support for
the hypothesis that there was an improvement in students’ diets over time
and that these improvements were associated with the changes imple-
mented by the school district. However, trend might also reflect other fac-
tors changing over the period.
In both random-effects equations, the coefficient for Asian was positive
and significant as noted, while the interaction variable trend Asian was
negative and significant. This combination suggests that although Asians
WINTER 2007 VOLUME 41, NUMBER 2 277
TABLE 3
RHI Regression Results for A
`La Carte Foods for 2002–2005
Dependent
Variable
Random Effects,
RHI-HEI
Fixed Effects,
RHI-HEI
Random Effects,
RHI-DDI
Fixed Effects,
RHI-DDI
Expenditure 0.05840 (.01090)*** 0.04246 (.01507)*** 0.02252 (.00579)*** 0.00867 (.00724)
Female 3.95769 (.64318)*** — 2.13771 (.34937)*** —
Asian 3.12371 (1.59618)** — 1.68108 (.86480)* —
Black 23.03510 (1.80164)* — 21.41100 (.95381) —
Hispanic 2.40425 (2.15368) — 1.87835 (1.16891) —
Other 22.17893 (3.56607) — 0.48096 (1.93920) —
Low income 21.21654 (2.15294) 1.90922 (4.22034) 0.35664 (1.15515) 1.84900 (2.02705)
Trend 0.66077 (.11623)*** 0.61460 (.12903)*** 0.69286 (.05789)*** 0.66417 (.06197)***
Trend male 0.65722 (.15577)*** 0.57164 (.17135)*** 0.30392 (.07738)*** 0.27010 (.08230)***
Trend Asian 20.96956 (.37003)*** 20.98694 (.41363)** 20.50828 (.18425)*** 20.53342 (.19867)***
Trend black 0.46913 (0.45109) 0.87531 (0.53273)* 0.34966 (0.22771) 0.31546 (0.25587)
Trend Hispanic 20.60228 (0.49517) 20.46348 (0.55653) 20.60028 (0.24654)** 20.58525 (0.26730)**
Trend Indian 0.77030 (0.94695) 0.90439 (1.00868) 20.04340 (0.46756) 0.09130 (0.48447)
Trend low income 0.36840 (0.60069) 20.53841 (0.89081) 20.39613 (0.31687) 20.70142 (0.42786)
Intercept 17.08412 (0.66227)*** 19.83373 (0.68597)*** 24.14117 (0.35462)*** 22.40320 (0.32948)***
Rho 0.26393 0.48705 0.44255 0.58891
Wald v
2
,F(9, 1301) 181.41 12.19 454.75 40.13
R
2
within 0.0764 0.0778 0.2147 0.2173
R
2
between 0.1195 0.0057 0.1635 0.0363
R
2
overall 0.0997 0.0212 0.1738 0.0841
N1,860 1,860 1,860 1,860
Note: The RHI is a measure of the healthiness of a student’s a` la carte choices over a 10-day period. The description of its construction is discussed in detail in the data
section. RHI-HEI is based on the criteria used in the USDA’s HEI and RHI-DDI is based on the school-dietitian-developed index. SEs are given in parentheses. *, **,
and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Recall this is an unbalanced panel; there are 1,860 total observations and 550 students,
with observations every November and April from November 2002 to April 2005. The Hausman test pvalues are .5003 and .102, respectively.
278 THE JOURNAL OF CONSUMER AFFAIRS
might have made healthier choices than whites, the improvement in RHI for
Asian students was less dramatic. The variable trend Hispanic was also
negative and significant in both the fixed- and random-effects estimations
using RHI-DDI but not significant using RHI-HEI as a dependent variable.
This result suggests that Hispanics improved less than other groups but was
not robust across dependent variables. It is important to stress that these
interaction variables do not imply that minority students experienced
a decline in their RHI scores; rather, because trend was positive, it suggests
that the healthiness of their choices improved less dramatically than that of
white students.
The interaction term trend male was positive and significant at the 1%
level across specifications, implying that males had a greater improvement
in the healthiness of their choices than females over the period (not that
females saw a decrease in RHI). Recall that female was positive and sig-
nificant in both equations. This means that males improved their RHI scores
at a greater rate than females. Getting teenage males to eat healthier food is
known to be particularly difficult. Therefore, it is encouraging that males
saw an improvement in the healthiness of their food choices after the inno-
vations in the school feeding program.
RecallthatRHI-DDI
meal
was calculated only for 2004–2005 due to
data restrictions. During 2004–2005, the entre´e purchased with a full
NSLPmealwascapturedinthedataandcouldberatedbythedietitian
for healthiness. For this reason, we estimated this year separately, con-
structing the dependent variable for student iif he or she purchased at
least two a` la carte items or an NSLP meal on at least 10 days during that
month. As before, if the student met these criteria, the student’s first 10
days with such purchases were used. Because the data only cover two
months in this specification, we included only students who were
observed in both periods. There were 292 students in this sample for a total
of 584 observations in November and April.
Table 4 presents the results for the random- and fixed-effects estimates
for 2004–2005. Interaction terms with the ‘‘trend’’ variable were not
included because the time period considered was only two months, and
no food service changes were implemented between November and April.
As seen in Table 4, by including full NSLP meals in the calculation, the
previous positive relationship between expenditure and RHI no longer held.
The full NSLP meals during the 2004–2005 academic year generally
received high healthiness ratings and meals were a relatively inexpensive
alternative to a` la carte lunches.
Interestingly, female was not significant in the random-effects regression
in column 1. Given the negative and significant coefficients, black and
WINTER 2007 VOLUME 41, NUMBER 2 279
especially other (i.e., Native American) students made substantially less
healthy food choices than white males (the omitted group). This suggests
that the food service needed to give particular attention to getting these
groups to eat healthier foods. Receiving a reduced-price or free meal
did not affect the healthiness of the students’ school lunches, which is
important to note. Since the November dummy variable was negative
and significant, the improvement in the overall healthiness of the school
lunches chosen continued over the 2004–2005 school year. Again, this
effect could be reflecting other factors that changed between November
2004 and April 2005.
The food service found it necessary to raise prices of NSLP meals. These
might have adversely affected participation. Therefore, tests were made to
determine whether there was a statistically significant change in the partic-
ipation rate for full NSLP lunches. Food service records indicated that par-
ticipation actually increased over the three years. Paired t-tests comparing
participation in November 2002, November 2003, and November 2004
rejected the null hypothesis that there had been no increase in participation,
TABLE 4
RHI Regression Results for A
`La Carte Foods and NSLP Meal Entre´es for 2004–2005
(Dependent Variable: RHI-DDImeal)
Random Effects Fixed Effects
Expenditure 20.00549 (0.01501) 20.00118 (.02432)
Female 0.30702 (0.39305) —
Asian 20.78645 (20.78646) —
Black 22.01814 (0.73864)*** —
Hispanic 21.0577 (1.10863) —
Other 27.3618 (2.3350)*** —
Reduced 0.68893 (1.40198) —
Free 20.178034 (0.72800) —
November 2004 dummy 21.57706 (0.22544)*** 21.56903 (0.22725)***
Intercept 4.88328 (0.70143)*** 4.45346 (0.98544)***
Rho 0.70143 0.60053
Wald v
2
,F(2, 290) 73.68 24.22
R
2
within 0.1430 0.1431
R
2
between 0.0815 0.0037
R
2
overall 0.0987 0.0410
N584 584
Note: The RHI is a measure of the healthiness of a student’s a` la carte and NSLP meal entre´e choices
over a 10-day period. The sample includes observations for November 2004 and April 2005. The
description of its construction is discussed in detail in the data section. For 2004–2005, RHI is based
on the school-dietitian-developed index only. SEs are given in parentheses. *, **, and *** indicate
significance at the 10%, 5%, and 1% levels, respectively. Rho is the fraction of variance that is
due to u
i
, the individual disturbance term. There are 584 total observations, and the panel is restricted
to students with observations in both periods.
280 THE JOURNAL OF CONSUMER AFFAIRS
with pvalues near zero. Similarly, tests for participation in April 2003 ver-
sus April 2004 and April 2005 were significant at the 5% level. There is
strong statistical evidence that NSLP meal participation increased with
implementation of the innovative program.
Not only did participation increase but also the average expenditure (the
total value of all foods purchased) decreased over the time period. Even if
the price of full NSLP meals increased, average expenditure would have
decreased if more students bought meals and fewer students purchased
multiple and/or expensive a` la carte items. Mean daily expenditures of stu-
dents making at least one purchase during the month were measured. Using
Welch’s approximation (Welch 1938, 1947) for the t-test, we tested the null
hypothesis that the mean daily expenditure had not changed versus the
alternative that it decreased. We tested November 2002 versus November
2004, as well as April 2003 with April 2005. Test statistics followed a
t-distribution approximately with 792 and 851 df and were 152.52 and
130.95, respectively. Both had pvalues of .0000, suggesting clear rejection
of the null hypothesis. Even though NSLP meal prices had increased, stu-
dent expenditures on lunch actually declined.
Finally, food service staff also reported that there was no increase in
students bringing food from home to avoid the new food choices. School
food service managers reported that bringing food from home is not
‘‘cool.’’ For an age group so sensitive to peer pressure, this factor may
be very important.
CONCLUSIONS
Promoting healthier eating habits in schools is a major issue across the
United States. However, rigorous analysis of the factors that determine
whether student behavior can change has been limited. Casual empiricism
or isolated and time-limited interventions cannot substitute for careful
observation in a cafeteria in real life and real time. This study analyzed
students’ food choices in an actual high school cafeteria against a backdrop
of major changes in its food program. A focus of the analysis was whether
students made healthier food choices after the innovative program was
introduced. The insights gained suggest approaches to bringing about
improvements in school lunch programs elsewhere.
Hopkins School District dramatically revamped its school food program
with a commitment to providing healthy and appealing food choices to stu-
dents. As the new program was gradually introduced, there was a steady
improvement in the nutritional quality of students’ food choices. Moreover,
the trend toward choosing healthier a` la carte food choices strengthened the
WINTER 2007 VOLUME 41, NUMBER 2 281
longer the program operated. When both a` la carte and full meals were
analyzed together, students were clearly making healthier food choices
in April 2005 than in November 2004. This suggests that over time, stu-
dents will develop healthier eating habits if given a choice of appealing
nutritious foods, at least on a closed campus. The argument that schools
have to serve foods, such as processed pizzas and deep-fried French fries,
or face dissatisfied students is unsupported. Without a control group, cau-
sality cannot be claimed definitively but an association between the food
service reforms and the relative healthiness of students’ food choices can be
inferred.
Helping students develop healthier eating habits at school can be a key
part of the fight against childhood obesity and potentially help improve life-
long health. The Hopkins food service reforms were associated with an
improvement in the nutritional quality of students’ food choices, while
operating on a financially sound basis. Hopkins is a typical, older,
inner-ring suburb with an increasingly diverse middle- and lower-income
population. Its innovations can therefore serve as a model for reforms in
many school districts nationwide.
Admittedly, these results for one school district, and the lack of a control
group in another high school, are less than conclusive. Further, only limited
information was available on NSLP meals. Researchers might conduct sim-
ilar studies using POS data from school lunchroom cash registers, including
a control group. Nevertheless, the innovations we have studied offer useful
lessons for improving school food service operations.
APPENDIX
HEI Criteria Used in Constructing RHI-HEI
Component Points Standard for Points
Total fruit (includes 100% juice) 5 At least one serving
Whole fruit (not juice) 5 At least one serving
Vegetables 5 At least one serving
Dark green and orange vegetables/legumes 5 At least one serving
Total grains 5 At least one serving
Whole grains 5 At least one serving
Milk products 10 At least one serving
Meat and beans 10 At least one serving
Oils 10 $12 g per 1,000 kcal
Saturated fat 10 Constitutes #7% of calories
Sodium 10 #0.7 g per 1,000 calories
Calories from solid fat and added sugar 20 #20% of total calories
Note: This is an adapted version of the HEI, 2005 criteria (USDA, CNPP 2005). Each a` la carte item was
rated based on the components in the table above. Points were given if the a` la carte item met the
corresponding standard; otherwise the item received zero points.
282 THE JOURNAL OF CONSUMER AFFAIRS
REFERENCES
Akin, John S., David K. Guilkey, and Barry M. Popkin. 1983. The School Lunch Program and Nutri-
tional Intake: A Switching Regression Analysis. American Journal of Agricultural Economics,
65 (3): 477–485.
American Obesity Association. 2005. AOA Fact Sheets—Obesity in Youth. http://www.obesity.org/
subs/fastfacts/obesity-youth.shtml.
Anderson, Patricia M. and Kristin Butcher. 2005. Reading, Writing and Raisinets: Are School Finances
Contributing to Children’s Obesity? National Bureau of Economic Research. Working Paper
11177. http://www.nber.org/papers/w11177.
Becker, Elizabeth and Marian Burros. 2003. Eat Your Vegetables? Only at a Few Schools. The New
York Times, January 13.
Bowman, Shanthy, Mark Lino, Shirley Gerrior, and P. Peter Basiotis. 1998. The Healthy Eating Index:
1994–96. Washington, DC: Center for Nutrition Policy and Promotion, U.S. Department of Agri-
culture. CNPP-5.
Centers for Disease Control and Prevention (CDC), National Center for Chronic Disease Prevention and
Health Promotion, Division of Adolescent and School Health. 2006. Fact Sheet: Foods and
Beverages Sold Outside of the School Meal Programs, 2000. http://www.cdc.gov/Healthy-
Youth/shpps/factsheets/pdf/outside_food.pdf.
Cooper, Ann and Lisa M. Holmes. 2006. Lunch Lessons: Changing the Way We Feed Our Children.
New York: Collins.
Devaney, Barbara, Anne Gordon, and John Burghardt. 1993. The School Nutrition Dietary Assessment:
Dietary Intakes and Program Participants and Nonparticipants. Washington, DC: Food and Nutri-
tion Service, Office of Analysis and Evaluation, U.S. Department of Agriculture.
Gleason, Philip M. and Carol W. Suitor. 2003. Eating at School: How the National School Lunch Pro-
gram Affects Children’s Diets. American Journal of Agricultural Economics, 85 (4): 1047–1061.
Guthrie, Joanne. 2003. A Healthy School Meal Environment. Food Assistance and Nutrition Research
Report Number 34-6. Washington, DC: U.S. Department of Agriculture.
Hausman, James. 1978. Specification Tests in Econometrics. Econometrica, 46 (6): 1251–1271.
Perry, Cheryl L., Donald B. Bishop, Gretchen L. Taylor, Marsha Davis, Mary Story, Clifton Gray,
Susan C. Bishop, Rita A. Warren Mays, Leslie A. Lytle, and Lisa Harnack. 2004. A Randomized
School Trial of Environmental Strategies to Encourage Fruit and Vegetable Consumption among
Children. Health Education and Behavior, 31 (1): 65–76.
Ralston, Katherine, Jean Buzby, and Joanne Guthrie. 2003. A Healthy School Meal Environment. Food
Assistance and Nutrition Research Report Number 34-5. Washington, DC: U.S. Department of
Agriculture.
Royal Cuisine. 2006. Royal Cuisine Business Plan: School Year 2006-07. Hopkins, MN: Hopkins
School District.
Simone French, Mary Story, Robert Jeffery, Patricia Synder, Marla Eisenberg, Abbey Sidebottom,
and David Murray. 1997. Pricing Strategy to Promote Fruit and Vegetable Purchases in High
School Cafeterias. Journal of the American Dietetic Association, 97 (9): 1008–1010.
U.S. Bureau of the Census. 2007. Statistical Abstract of the United States, 2007. Washington, DC: U.S.
Government Printing Office.
U.S. Department of Agriculture (USDA), Center for Nutrition Policy and Promotion (CNPP). 2005.
The Healthy Eating Index, 2005. CNPP Fact Sheet No. 1. Washington, DC: U.S. Department of
Agriculture.
U.S. Department of Agriculture (USDA), Food and Nutrition Service (FNS). 2001. School Nutrition
Dietary Assessment Study II, Report No. CN-01SNDAII. Washington DC: U.S. Department of
Agriculture.
———. 2005a. Making It Happen! School Nutrition Success Stories, Washington, DC. http://
www.fns.usda.gov/cnd/lunch/aboutlunch\nslpfactsheet.pdf.
WINTER 2007 VOLUME 41, NUMBER 2 283
———. 2005b. Healthy Schools, Washington, D.C. http://www.fns.usda.gov/tn/Healthy/wellness_
policyrequirements.html
———. 2005c. National School Annual Summary. Washington, DC. http://www.fns.usda.gov/pd/
slsummar.htm
U.S. Department of Health and Human Services (USDHHS). 2000. Healthy People 2010. Washington,
DC: U.S. Department of Health and Human Services.
Variyam, Jayachandran N., James Blaylock, and David Smallwood. 1998. USDA’s Healthy Eating
Index and Nutrition Information. Technical Bulletin No. 1866. Washington, DC: U.S. Department
of Agriculture.
Welch, Bernard L. 1938. The Significance of the Difference between Two Means When the Population
Variances Are Unequal. Biometrica, 29 (3–4): 350–362.
——— 1947. The Generalization of ‘‘Student’s’’ Problem When Several Different Population Varian-
ces Are Unknown. Biometrica, 29 (1–2): 28–35.
Wooldridge, Jeffrey M. 2002. Econometric Analysis of Cross Section and Panel Data. Cambridge,
MA: MIT Press.
284 THE JOURNAL OF CONSUMER AFFAIRS
