ArticlePDF Available

Using Field Experiments to Encourage Healthy Eating in Schools

Authors:

Abstract and Figures

Schools provide a unique opportunity to influence healthy eating decisions in children. Field experiments provide a practical tool for evaluating the types of interventions that can have the largest impact on these decisions in the short and long run. This article provides some insights on conducting field experiments in schools; the issues it covers are related to data collection, randomization, heterogeneous treatment effects, and statistical inference.
Content may be subject to copyright.
Using Field Experiments to Encourage
Healthy Eating in Schools
Joseph Price
Schools provide a unique opportunity to inuence healthy eating decisions in
children. Field experiments provide a practical tool for evaluating the types of
interventions that can have the largest impact on these decisions in the short and
long run. This article provides some insights on conducting eld experiments
in schools; the issues it covers are related to data collection, randomization,
heterogeneous treatment eects, and statistical inference.
Key words:behavioral economics, healthy eating, school meal program
In the United States, children from a low-income family are less likely to
consume the daily recommended amount of fruits and vegetables than are
children from a high-income family (Rasmussen et al. 2006). Some of these
dierences may stem from access (Larson, Story, and Nelson 2009) or the
fact that fruits and vegetables are perceived as more expensive (Cassady,
Jetter, and Culp 2007). However, schools provide a unique opportunity to
provide children with free and easy access to fruits and vegetables, and there
has been considerable eort to include more fruits and vegetables in the
breakfast and lunch oerings and in snack programs supported by USDA. In
fact, most children in the United States do not meet dietary recommendations
and would benet from an improved diet, including high-income children
(Story 2009).
Over 70% of children eat school lunch three or more times per week
(Wansink et al. 2013), and recent estimates indicate that about 77% of
children eat school lunch three or more times per week (CDC and NCHS
2018). Many children also eat breakfast at school, receive meals during the
summer, are given food to take home over the weekend, and are sometimes
even oered dinner when they stay for after-school programs. The
combination of these programs means that some kids have the potential to
eat more of their meals at school than at home. This provides a unique
opportunity for schools to inuence the amount of fruits and vegetables that
children eat. However, the habits that children form at home may aect their
willingness to consume fruits and vegetables even when they are oered for
free at school. Thus, additional approaches are required to help children tap
into the access they have to fruits and vegetables at school.
Joseph Price, Brigham Young University, Economics, Economics, 162 FOB, Provo, Utah, 84602, Email:
joe_price@byu.edu
Agricultural and Resource Economics Review (2019) 114
© The Author(s) 2019. This is an Open Access article, distributed under the terms of the Creative
Commons Attribution licence (http://creativecommons.org/licenses/by/4.0/), which permits
unrestricted re-use, distribution, and reproduction in any medium, provided the original work is
properly cited.
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Behavioral economics provides a number of tools that can be used to
encourage children to consume more fruits and vegetables at school. Many of
these tools fall under the category of nudges, as described by Thaler and
Sunstein (2008), and the Healthy Lunchroom Initiative provides an even
more expanded list of these approaches (Just and Wansink 2009). Possible
interventions can be based on incentives, default options, choice architecture,
competition, messaging, and pre-commitment. Interventions can also include
removing competing foods (like snacks that can be purchased individually),
increasing the variety of fruit and vegetable options, and embedding fruits
and vegetables as an integral part of other items (like smoothies), associating
fruits and vegetables with something cool (like the Food Dudes program
[Wengreen et al. 2013]), and moving recess to before lunch. There is a
growing amount of research that documents the ecacy of many of these
interventions. The focus of this article is on providing some practical advice
about how to conduct a research study in schools to evaluate the impact of a
particular intervention.
I start with an intervention that has not been rigorously evaluated by
academic researchers but was piloted by a school nutrition company called
Nutrislice (www.nutrislice.com). It is an interesting intervention that helps
frame the issues related to evaluating the impact of a specic intervention.
Nutrislice ran a competition during lunch between adjacent grades in which
it recorded and reported in real time the number of servings of broccoli that
had been consumed. The grade that won the competition would have each
student be given a small smoothie. Figure 1 shows the setup of the
intervention. The two grades were closely matched during the competition,
Figure 1. Nutrislice Broccoli Competition
Agricultural and Resource Economics Review22019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
and the gure provides a picture of the kids as they began to chant Eat your
broccoli!while watching updates in the numbers on a large screen that had
been set up. In the end, the children at that school consumed several times
more broccoli that day than had been observed on any other previous days.
This was a one-time pilot study that had a huge eect on the amount of broccoli
that children ate. The question is what to do with this information. Should every
school have a broccoli competition? How often should the school do it? Would we
want to do it every day? The ability to answer these questions relies on the
degree to which a particular intervention will have a sustained eect over
time, the cost of implementing the intervention, and whether a one-day
intervention will have positive or negative eects in the long run when the
intervention is concluded. Additionally, the answers to these questions depend
on the external validity of the experiment and whether or not the results are
generalizable to a broader population. Whether or not a specic intervention
should be implemented in a large number of schools depends on if the
intervention will have a similar eect in most schools where it is implemented.
This particular intervention includes some interesting elements of competition,
real-time feedback, peer pressure, and incentives.
It would be helpful to design variants of the experience that remove dierent
aspects of the intervention to see which underlying behavioral motivations
seem to be the most important. For example, it would be interesting to see if
the same approach would work even if there hadnt been any smoothies
oered to the winners. This would certainly improve the cost-eectiveness of
the program. Providing real-time feedback is a logistical challenge and
requires some technology to be in place. It would be helpful to test how
much the impact of the intervention drops without that component. Finally, it
would be important to see what happens in the school the next time broccoli
is served so that we can see what the long-run eects of the intervention are.
The exciting thing is that schools will nd creative ways to encourage children to
eat in a more healthy way. Because of this, it is important to have a rigorous way to
evaluate the impact of the intervention andtobeabletoframethedegreetowhich
the intervention aects student behavior both during the intervention and after it
is removed. This provides a unique opportunity for academics who develop tools
and empirical strategies that allow them to collaborate with schools to evaluate
school-based interventions. There might even be ways that students themselves
could be equipped with the tools to carry out a well-designed experiment as
part of a science fair project and collect data in ways in which they can be
aggregated and evaluated by researchers.
The sections that follow describe several issues that are important to consider
when conducting a eld experiment in a school. Each of these issues arose in the
course of my own eld experiments in schools, and the hope is that the lessons
we learned will be helpful to scholars considering doing research in schools.
The focus of my experiments has been on encouraging children to eat more
fruits and vegetables, but the issues addressed are relevant to interventions
that inuence other behaviors or decisions that children make at school.
Joseph Price Using Field Experiments to Encourage Healthy Eating in Schools 3
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Data Collection
All empirical studies will require some form of data collection. When I rst
started working in this area, I met with the nutrition directors of dierent
school districts to better understand the type of data that they were
gathering. Nearly all schools collect production records on the number of
items that were served during each lunch. This is usually based on calculating
the amount of each item that was used in preparing the meals for that meal
and then subtracting out the amount of each item that is left over. This will
provide a good aggregate measure of what is served but does not account for
any waste that occurs when children discard their trays. In addition, it doesnt
provide information at the individual level, which can be important if we want
to see how the eect of an intervention diers by grade or gender.
Many schools also have data from a point-of-sale (POS) system. The advantages
of this data are that they will provide information on the item-level choices that
children make. For example, many POS systems will record the exact fruit and
vegetable items that children take, which entrée they choose, and whether
they take white or chocolate milk. Since these data are already collected by
the school, if researchers can access these data, then they can get a very long-
run look at the behavior before, during, and after the intervention, which is a
huge advantage. The main disadvantage is that the data dont provide any
information about whether the child actually ate the items.
In addition to the POS and aggregate data, school administrative data that
contain information about the characteristics of the schools involved in your
experiment can be useful. The Common Core of Data (CCD) provided by the
National Center for Education Statistics (NCES) contains useful demographic
data such as gender, ethnicity, and grade level distribution. This data source
also contains information relevant to school lunch participation, such as the
fraction of students who are eligible to receive a free or reduced-price lunch
(FRPL). The Food Research and Action Center (FRAC) provides statistics on
the Identied Student Population (ISP), which is the percentage of students
in a school that come from households receiving government welfare
assistance. These administrative data can be used to compare characteristics
between treatment and control schools.
In order to collect consumption data, you need to have some way of observing
the childs tray at the end of lunch so you can see which items were not
consumed. There have been several methods developed to provide an
accurate measure of the amount of each item that the child consumed. Some
of these involve weighing items from the tray at the start and end of lunch
(Getlinger et al. 1996). Others involve taking a picture of the tray at the start
and end of lunch and then using researchers to evaluate the amount
consumed based on the pictures (Swanson 2008). All of these methods
involve collecting data at the start and end of lunch and generally require
that you take a random sample of students to follow, because these methods
can be very time-intensive. In addition, since you are gathering data on
Agricultural and Resource Economics Review42019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
specic students, this approach generally requires active parental consent for
the child to participate.
Based on some pilots we did, we ultimately decided to take a dierent
approach to gathering data that has some nice features that would allow the
data to be easily used by other researchers. One of the advantages of
measuring consumption of fruits and vegetables is that they come in pre-
portioned amounts and leave behind a cup or peel that allows us to observe
the numbers of items that were taken. This makes it possible to gather the
data using only observations taken at the end of lunch. We would have our
research assistants stand in the cafeteria near the place where students
discard their trays and record informationthe childs grade, age, entrée
choice, and the number of servings of each fruit and vegetable that they took
and consumed. While this approach worked for the schools in our
experiment, it would likely need to be adjusted for dierent schools, such as
schools with a salad bar where students can serve themselves.
Figure 2 is a picture of the iPhone app that we developed to gather data. In the
particular case shown, the child had taken two servings of broccoli and one
serving of green beans. They had consumed 1.5 servings of broccoli but none
of the green beans. The disadvantage of this approach is that the data are
gathered rather quickly, and the consumption amounts are less precisely
measured than in other approaches. The key advantage is that our method
requires about 40 hours of research assistant time to collect 10 days of data
at one school. Thus, for a cost of about $340 in research assistant wages, you
can gather 3,000 student-day observations. We were even able to nd ways
to hire parents from the school to gather data for us, which reduced the
travel costs of collecting data and allowed us to expand the geographic scope
of the schools we work with.
In addition to collecting data on what is served, it is also helpful to take
pictures each day of the setup of the cafeteria, the layout of the way food is
served, and the specic items that are served that day. These pictures allow
future researchers to create measures from your experiment that expand the
use of the data you collect to answer other questions. It is also helpful in
framing the issue to public audiences, because they often have a misperception
about the quality of fruits and vegetables being served in schools. Figure 3
provides some examples of photos that we took in the schools that we were
working with. Most of the parents that weve shared these photos with have
been surprised by the quality and attractiveness of the items being served in
schools. With advances in machine learning, there might even be innovative
ways these pictures could be used in future research to provide quantitative
measures of the items being served each day.
Baseline Data
A key component of your study will be to establish what the baseline pattern of
consumption is at the school. For our experiments, we have tried to collect at
Joseph Price Using Field Experiments to Encourage Healthy Eating in Schools 5
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
least 510 days of baseline data at each school. The larger baseline period will
be important in terms of statistical power, since the intervention eect will be
evaluated relative to this baseline measure. For our original experiment on
incentives (Just and Price 2013), we collected ve days of baseline data at
each of the 15 schools. This provided 30,550 child-day observations and
made it possible to establish a reasonable baseline at each school. From this
baseline data, we learned that 33.5% of children ate at least one serving of
fruits and vegetables each day, and this rate was higher for girls and older
children. We also learned that 42.7% of the fruits and vegetables being
served were ending up in the trash; this was a surprise to the nutrition
directors we were working with, as all of their previous estimates of
consumption and waste had been based on production records.
The baseline data also provide an important role in mitigating concerns about
a Hawthorne eect. It is true that every intervention we evaluate has the
possibility that having people in the cafeteria gathering data will inuence
childrens behavior. One way to reduce this concern is to have researchers
Figure 2. VProject iPhone App for Data Collection
Agricultural and Resource Economics Review62019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
located in as inconspicuous a place as possible and to have them answer
questions about what they are doing in non-specic ways that dontreect
the fact that they are gathering data on fruits and vegetables. In one of our
experiments (OBryan, Price, and Riis 2017), the students place their trays on
a conveyor belt, and so we were able to gather the data on the other side of
the wall, out of sight. Having a longer baseline data collection period helps to
reduce the novelty eect you might pick up the rst day data collectors are
present. You can also test to see if there is a trend during the baseline period,
and we nd that there is a slight negative trend. This means that your
baseline data are likely to reect the true behavior once you have been
collecting data for enough days.
Control Schools
Another related issue is the use of control schools. The dierence-in-dierence
model is a commonly used empirical strategy in policy evaluation. This involves
having pre-intervention and post-intervention data for a treatment and a
control group. The control group serves two important functions in this
Figure 3. Example Pictures of Fruit and Vegetable Items Served in Our Sample
Joseph Price Using Field Experiments to Encourage Healthy Eating in Schools 7
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
approach. First, it accounts for any other changes that might have occurred at
the same time the treatment started. Second, it helps account for any
Hawthorne Eect of being in the treatment group. The disadvantages of
including a control group is that it adds to the cost of running the
experiment. It can also make it hard to recruit schools to participate in the
study. It can often be dicult to recruit schools, especially when they are told
there is a chance they will participate without any possible benets for their
students.
One approach we used in a large eld experiment with 40 schools
(Loewenstein, Price, and Volpp 2016) was randomizing the timing of when
the treatment began. As a result, some schools started in October, others in
November, and so on through the school year. This allowed us to use the
staggered baseline periods to capture any changes over time in other factors
that might inuence the outcome. To address the concern about the
Hawthorne Eect, we used a 10-day baseline period so the students could
fully adjust to having data collectors in the cafeteria before we started the
treatment. Another benet of the staggered approach is that it allowed us to
work with 40 schools and not have to hire nearly as many research
assistants as we would have needed if we had done all the schools at the
same time. When research assistants completed their experiment at one
school, they would move on to the next school, which allowed the inclusion
of data from about four times as many schools with the same number of
research assistants.
Level of Randomization
Another important decision is the level of randomization to use. The unit of
observation that you use for your randomization will have a large eect on
the statistical power you have for your experiment. If you can randomize at
the individual level, then you could have over 500 randomization units in one
school. In contrast, if you randomize at the classroom level, this could drop to
20, or at the grade level, it could drop to six. When you do inference, you will
want to cluster your standard errors at the level of randomization that you
used in your experiment. Thus youll want to use the randomization unit that
is the most reasonable for your experiment. When choosing between options
with similar benets, opt for the approach that will give you the most
randomization units.
All of the experiments that I have done have involved school-level
randomization. There were four reasons we chose to use school-level
randomization instead of student-level randomization. First, in many cases,
we wanted to be able to get maximum advertising for the intervention once
it started. For an incentive experiment, we wanted to be able to include a
morning announcement for the whole school about the new program. Second,
we wanted to avoid any contamination eects. This could occur if children
learned of others getting a reward for eating a serving of fruits and
Agricultural and Resource Economics Review82019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
vegetables and then changed their behavior because they thought the same
would be true for themselves. Third, we wanted to avoid any reactance
eects that might occur if members of the control group felt it was unfair
that they were being treated dierently. Fourth, we wanted to be able to use
an opt-out consent form for participation in our experiment, which was only
justied on the basis that the entire school was getting the same treatment.
A modied approach would cost much less to implement and would preserve
many of the nice features of school-level randomization. When the
randomization unit is the school, then you dont get a lot of extra statistical
power from having more student observations within the school. Having
more students does improve the precision of your estimates a bit, but not
nearly as much as having additional schools. Thus, with a xed budget it
might make sense to consider only included students from a specic grade or
lunch period in the experiment. It might also make sense to focus on schools
that have a smaller student body. Both of these decisions will mean you will
have a smaller subject pool within each school, which will make it less costly
to implement some experiments (like those that involve incentives) or to
collect the data (since at big schools, we often needed to send two to three
research assistants to collect the data). School size is an important factor to
consider, because it is possible that focusing on smaller schools potentially
excludes urban schools or secondary schools. In our experiments, we focus
on fruit and vegetable consumption in elementary schools rather than
secondary schools, so our results are more generalizable to the elementary
school population. Whether experiment results are generalizable to urban
schools depends on whether or not the key dierence between urban and
non-urban schools is school size. The key concern with these approaches
depends on whether you think the eect of the treatment diers by grade or
school size. In the work weve done, we nd that baseline levels of fruit and
vegetable consumption dier by grade but that the treatment eect does not
dier by grade or school size. Thus, if the key dierence between urban and
non-urban schools is size, it appears that our experiment results are
generalizable to urban schools.
Heterogeneous Treatment Eects
There will be some experiments, however, where you will want to know how
the treatment eect diers based on student or school characteristics. These
heterogeneous treatment eects can help in understanding which groups to
target dierent interventions toward and can provide insight into how an
intervention might create dierences in outcomes between groups. In our
experiments about the eect of incentives, we did not nd any dierences in
the eect of incentives based on gender or grade. However, we did nd very
large dierences based on income level.
We had originally explored some ways to identify the free or reduced-price
lunch (FRPL) status of individual students, but it just never worked very
Joseph Price Using Field Experiments to Encourage Healthy Eating in Schools 9
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
well, given the way we were collecting our data. As a result, instead of using
individual-level data on FRPL status, we used measures we had about the
fraction of students at the school that received a FRPL as a proxy for the
income level of the school. In our estimation models, we then included an
interaction between treatment and the FRPL rate for the school.
This turned out to be a very informative interaction term that indicated that
the eect of the incentives at the poorest schools in our sample (FRPL ¼77%)
was over twice as large as at the richest schools in our sample (FRPL ¼17%).
This suggests that incentives do a very good job of specically targeting the
students who are most likely to benet from the intervention, since they are
the ones least likely to be consuming fruits and vegetables at home.
Standard Errors
In a previous section, I mentioned the importance of clustering your standard
errors at the level of randomization (student, class, grade, school, etc.). There
is a real challenge with having too few randomization units, because the
ability to cluster the standard errors can become problematic. The
experiment in Just and Price (2013) had only 15 schools. The editor of the
rst journal we submitted our article to commented that it just didnt seem
possible to do inference with only 15 clusters. There are other approaches to
doing inference when you have a small number of clusters. One option is to
collapse your data down to the level of variation.
Thus, instead of doing your inference on student-level data about what they
consumed, you can get an observation for each school from its pre period and
one from its post period. This approach would not require you to cluster your
standard but would leave you with a much smaller sample than you originally
thought. This has the benet of not letting a large sample size give the illusion
that you have a lot of insightful variation in your data. With your collapsed data
you can even do a Fishers exact test, where you take each of your schools and
label them as treatment or control, based on their status. You then give schools
individually an indicator based on whether they had an increase in behavior
above some threshold. Then you use the Fishers exact test to determine the
probability of observing by chance the pattern that you see in your data. In
Just and Price (2013), we found that all 13 treatment schools had a positive
increase, while the two control schools did not. The probability of observing
this pattern by chance is 0.0095, which provides insight about inference
similar to a p-value.
Another approach in these settings is to use permutation inference. In
Loewenstein, Price, and Volpp (2016), we had 40 schools in our sample; 19
were given incentives for ve weeks, and 21 were given incentives for three
weeks. We wanted to test if the long-run eects of incentives were larger
when the incentives were in place for a longer period of time. We had used
several methods for calculating the standard errors of one of our key
estimates, including generalized estimating equations, wild bootstrap, and
Agricultural and Resource Economics Review10 2019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
paired bootstrap. All of these provided a p-value of our parameter of interest
right near 0.05, but some were above that threshold and some were below.
We decided to use permutation inference because it has a few features that
provided a more transparent way of testing the likelihood that there was a
real dierence in outcomes between the two treatment groups. First, we
randomly generated permutations of each of the possible labels for each
school and then reestimated our model under 10,000 of these permutations.
This means that whether a school was labeled as in the ve-week treatment
group or three-week treatment group was randomly assigned and then the
model was estimated as though that had been the correct assignment. This
provided a distribution of coecients from our model under all of the
permutations. We then compared this distribution to the coecient that we
obtained with the correctly labeled schools. Of the 10,000 permutations,
there were 246 that had a coecient higher than what we found, providing a
p-value of 0.049 (see Figure 4).
In addition, when we put the coecients into categories based on the number
of schools that had been labeled correctly by accident during the random
assignment, we found the largest coecients for those permutations where
the highest fraction of schools had been labeled correctly (see Figure 4). If
there had been no true treatment eect, then we would expect the average
coecients to be similar across the dierent bins.
An approach similar to permutation inference can play a particularly
important role in experiments where there is a very low cost of collecting
baseline or control data but a very high cost of collecting treatment data. For
example, if the outcome you are interested in could be collected using data
from a POS system, then it might be possible to obtain tens of thousands of
observations for nearly free. You can use these data to establish a well-
Figure 4. Graphs Based on Permutation Inference
Notes: The graph on the left provides the distribution of coecients that we obtained with 10,000
samples where we randomly assigned the labels assigned to each school. The red line is the coecient
that we obtained when the correct labels are assigned. The graph on the right provides the average of the
coecients that we get based on the fraction of schools that receive the correct label (put in decline bins).
The red line is the coecient we obtain when we correctly label all of the schools.
Joseph Price Using Field Experiments to Encourage Healthy Eating in Schools 11
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
dened distribution for what you would expect the outcomes to be on typical
data. Then, if you implement your treatment for even one day, you can
compare that observation to this well-dened distribution to obtain the
probability of observing that outcome by chance. While most would not want
to recognize a result based on one treatment observation, the principle is
that permutation inference can provide a relatively transparent way to
observe the likelihood of observing a particular outcome by chance, and this
insight can be pretty clear even with only a few treatment observations.
Aggregation of Marginal Eects
Some of the interventions that we will use might provide only a small nudge. It
is important to look for ways to aggregate these marginal eects in a way to
amplify the nal impact. One example of this is the type of incentive that we
used in the habit formation experiment with 40 schools. In my earlier
experiment with David Just, we had used cash prizes. Cash prizes turned out
to be the most eective incentive to use in our setting but did have the
downsides of the kids using the money mostly to buy candy and of an uptick
in bullying in the school. As a result, for our larger experiment, we decided to
use a veggie coin instead of cash. The veggie coin was worth a quarter but
could only be used at the school store (which was not allowed to sell candy).
Figure 5 shows the image that was on the coin. It turns out that the use of the
veggie coin had three important benets over cash. First, the image on the coin
provided a tangible reminder of the action the children had taken to receive the
reward. Second, since all of the veggie coins were redeemed at the school store,
it made it easier logistically to make the cash disbursements. We would simply
Figure 5. Veggie Coin Used in Habit Formation Experiment
Agricultural and Resource Economics Review12 2019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
write the PTA a check at the end of the experiment based on the number of
veggie coins that were redeemed, which we could also check against the data
collection on the number of kids who received a coin. Third, many schools
only had the school store open once a week or at the end of the incentive
period. The kids would then accumulate their coins over time, which would
make the perceived marginal benet of consuming a serving of fruits and
vegetables each day seem larger as well. This technique is often used by
nancial planners who emphasize that small changes in spending each day
can aggregate into very large changes in spending over time.
Conclusion
The goal of this article was to point out some issues that are helpful to consider
when designing a eld experiment to encourage healthy eating in schools. These
experiments have the added benet of providing general behavioral insights
about interventions that might work for behaviors outside the specic focus
of healthy eating, such as academic performance or exercise. My hope is that
more researchers will consider collaborating with schools to test the eect of
dierent interventions that can inuence positive behaviors in children. This
research can be very personally rewarding and have an impact on schools.
The study I did with David Just about the impact of moving recess before
lunch was picked up by the media and has led lots of schools and districts to
adopt this simple change. In addition, many of the interventions you identify
that work in schools might have applications in your own home or in your
own daily choices.
Another important aspect to consider when examining the ecacy
of interventions in schools is external validity, how well results generalize to
a broader population. Schools serve students who range widely in age, from
6 to 18, and schools are often very dierent along socioeconomic and
demographic characteristics. This should be considered when examining how
well an intervention may work when implemented on a broader scale. In
particular, one must consider how representative of the general school
population the schools involved in an experiment are. Interventions should
only be implemented on a broad scale if the experimental units are
representative of the general population and as long as the intervention has
been implemented multiple times with similar results. Ultimately, the most
eective interventions that inuence children to eat fruits and vegetables in
school are those that are tested in internally valid experiments with a sample
from which externally valid conclusions can be drawn.
References
Cassady, D., K. M. Jetter, and J. Culp. 2007. Is Price a Barrier to Eating More Fruits
and Vegetables for Low-Income Families?Journal of the American Dietetic Association
107(11): 19091915.
Joseph Price Using Field Experiments to Encourage Healthy Eating in Schools 13
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
Centers for Disease Control and Prevention (CDC) and National Center for Health Statistics
(NCHS) 2018. National Health and Nutrition Examination Survey Data. U.S. Department
of Health and Human Services, Centers for Disease Control and Prevention. Hyattsville,
MD.
Getlinger, M. J., C. V. Laughlin, E. Bell, C. Akre, and B. H. Arjmandi. 1996. Food Waste Is
Reduced When Elementary-School Children Have Recess before Lunch.Journal of the
Academy of Nutrition and Dietetics 96(9): 906.
Just, D. R., and B. Wansink. 2009. Smarter Lunchrooms: Using Behavioral Economics to
Improve Meal Selection.Choices 24(3): 17.
Just, D. R., and J. Price. 2013. Using Incentives to Encourage Healthy Eating in Children.
Journal of Human Resources 48(4): 855872.
Larson, N. I., M. T. Story, and M. C. Nelson. 2009. Neighborhood Environments: Disparities in
Access to Healthy Foods in the US.American Journal of Preventive Medicine 36(1): 7481.
Loewenstein, G., J. Price, and K. Volpp. 2016. Habit Formation in Children: Evidence from
Incentives for Healthy Eating.Journal of Health Economics 45: 4754.
OBryan, A., J. Price, and J. Riis. 2017. Using Default Options to Increase Healthy Add-ons to a
Meal.Journal of Marketing Behavior 3(1): 3950.
Rasmussen, M., R. Krølner, K. I. Klepp, L. Lytle, J. Brug, E. Bere, and P. Due. 2006.
Determinants of Fruit and Vegetable Consumption among Children and Adolescents:
A Review of the Literature.International Journal of Behavioral Nutrition and Physical
Activity 3(1): 22.
Story, M. 2009. The Third School Nutrition Dietary Assessment Study: Findings and Policy
Implications for Improving the Health of US Children.Journal of the Academy of
Nutrition and Dietetics 109(2): S7-S13.
Swanson, M. 2008. Digital Photography as a Tool to Measure School Cafeteria Consumption.
Journal of School Health 78(8): 432437.
Thaler, R. H., and C. R. Sunstein. 2008. Nudge. New York: Knopf.
Wansink, B., D. R. Just, R. W. Patterson, and L. E. Smith. 2013. Nutrition Report Cards: An
Opportunity to Improve School Lunch Selection.PLOS One 8(10): e72008.
Wengreen, H. J., G. J. Madden, S. S. Aguilar, R. R. Smits, and B. A. Jones. 2013. Incentivizing
Childrens Fruit and Vegetable Consumption: Results of a United States Pilot Study of
the Food Dudes Program.Journal of Nutrition Education and Behavior 45(1): 5459.
Agricultural and Resource Economics Review14 2019
https://doi.org/10.1017/age.2019.5
Downloaded from https://www.cambridge.org/core. IP address: 158.115.230.131, on 13 Jul 2019 at 01:53:47, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms.
ResearchGate has not been able to resolve any citations for this publication.
Article
Full-text available
There is growing interest in the situations in which incentives have a significant effect on positive behaviors, particularly in children. Using a randomized field experiment, we find that incentives increase the fraction of children eating a serving of fruits or vegetables during lunch by 80 percent and reduce the amount of waste by 33 percent. At schools with a larger fraction of low-income children, the increase in the fraction of children who eat a serving of fruits or vegetables is even larger, indicating that incentives successfully target the children who are likely to benefit the most from the increased consumption.
Article
Full-text available
To explore the feasibility and implementation efficiency of Nutritional Report Cards(NRCs) in helping children make healthier food choices at school. Pilot testing was conducted in a rural New York school district (K-12). Over a five-week period, 27 parents received a weekly e-mail containing a NRC listing how many meal components (fruits, vegetables, starches, milk), snacks, and a-la-carte foods their child selected. We analyzed choices of students in the NRC group vs. the control group, both prior to and during the intervention period. Point-of-sale system data for a-la-carte items was analyzed using Generalized Least Squares regressions with clustered standard errors. NRCs encouraged more home conversations about nutrition and more awareness of food selections. Despite the small sample, the NRC was associated with reduced selection of some items, such as the percentage of those selecting cookies which decreased from 14.3 to 6.5 percent. Additionally, despite requiring new keys on the check-out registers to generate the NRC, checkout times increased by only 0.16 seconds per transaction, and compiling and sending the NRCs required a total weekly investment of 30 minutes of staff time. This test of concept suggests that NRCs are a feasible and inexpensive tool to guide children towards healthier choices.
Article
We present findings from a field experiment conducted at 40 elementary schools involving 8000 children and 400,000 child-day observations, which tested whether providing short-run incentives can create habit formation in children. Over a 3- or 5-week period, students received an incentive for eating a serving of fruits or vegetables during lunch. Relative to an average baseline rate of 39%, providing small incentives doubled the fraction of children eating at least one serving of fruits or vegetables. Two months after the end of the intervention, the consumption rate at schools remained 21% above baseline for the 3-week treatment and 44% above baseline for the 5-week treatment. These findings indicate that short-run incentives can produce changes in behavior that persist after incentives are removed.
Article
Background Poor dietary patterns and obesity, established risk factors for chronic disease, have been linked to neighborhood deprivation, neighborhood minority composition, and low area population density. Neighborhood differences in access to food may have an important influence on these relationships and health disparities in the U.S. This article reviews research relating to the presence, nature, and implications of neighborhood differences in access to food. Methods A snowball strategy was used to identify relevant research studies (n=54) completed in the U.S. and published between 1985 and April 2008. Results Research suggests that neighborhood residents who have better access to supermarkets and limited access to convenience stores tend to have healthier diets and lower levels of obesity. Results from studies examining the accessibility of restaurants are less consistent, but there is some evidence to suggest that residents with limited access to fast-food restaurants have healthier diets and lower levels of obesity. National and local studies across the U.S. suggest that residents of low-income, minority, and rural neighborhoods are most often affected by poor access to supermarkets and healthful food. In contrast, the availability of fast-food restaurants and energy-dense foods has been found to be greater in lower-income and minority neighborhoods. Conclusions Neighborhood disparities in access to food are of great concern because of their potential to influence dietary intake and obesity. Additional research is needed to address various limitations of current studies, identify effective policy actions, and evaluate intervention strategies designed to promote more equitable access to healthy foods.
Article
Objective: Preliminary evaluation in the United States (US) of a school-based fruit and vegetable (F/V) intervention, known as the Food Dudes (FD) program, developed in the United Kingdom. Methods: Over 16 days (Phase 1), elementary-school children (n = 253) watched short videos featuring heroic peers (the FD) eating F/V and received a reward for eating F/V served at lunchtime. In the 3 months that followed (Phase 2), children received increasingly intermittent rewards for eating F/V. Consumption was measured by photo analysis and assessment of skin carotenoids. Results: Fruit and vegetable intake increased significantly after Phases 1 and 2 (P < .001 for both). This effect was most discriminable among children who consumed no fruit (n = 100) or no vegetables (n = 119) at pre-intervention baseline. Among these children, F/V intake (combined) increased by 0.49 (0.53) cups per day. Conclusions and implications: The FD program can increase F/V intake in US elementary schools.
Article
Much progress has been made in the past 5 to 10 years in measuring and understanding the impact of the food and physical activity environments on behavioral outcomes. Nevertheless, this research is in its infancy. A work group was convened to identify current evidence gaps and barriers in food and physical activity environments and policy research measures, and develop recommendations to guide future directions for measurement and methodologic research efforts. A nominal group process was used to determine six priority areas for food and physical activity environments and policy measures to move the field forward by 2015, including: (1) identify relevant factors in the food and physical activity environments to measure, including those most amenable to change; (2) improve understanding of mechanisms for relationships between the environment and physical activity, diet, and obesity; (3) develop simplified measures that are sensitive to change, valid for different population groups and settings, and responsive to changing trends; (4) evaluate natural experiments to improve understanding of food and physical activity environments and their impact on behaviors and weight; (5) establish surveillance systems to predict and track change over time; and (6) develop standards for adopting effective health-promoting changes to the food and physical activity environments. The recommendations emanating from the work group highlight actions required to advance policy-relevant research related to food and physical activity environments.