1 INTRODUCTION

Changes to the National School Lunch Program (NSLP), the United States' second largest food assistance program, under the Healthy, Hunger-Free Kids Act of 2010 aimed to improve the dietary quality of participants (United States Department of Agriculture [USDA], 2012). Although consumption of fruits and vegetables is linked to various health benefits including lower risk of obesity, chronic noncommunicable diseases, and some cancers (Fung et al., 2001; Kranz et al., 2012; Maynard et al., 2003; Rolls et al., 2004), only 2% of adolescents consume the USDA recommended amounts of vegetables and only 7% of adolescents consume the USDA recommended amounts of fruits (Lange et al., 2021).

Recent research suggests employing behavioral economic principles such as nudging and precommitment can lead to students selecting and consuming healthier foods (Lai et al., 2020; Marcano-Olivier et al., 2019; Miller et al., 2016; World Health Organization, 2022). Nudges or subtle behavioral cues aim to influence the choices made by individuals without taking away their freedom of choice (Thaler & Sunstein, 2009). Prior studies conducted in school settings indicate nudges can increase the selection and consumption of fruits, vegetables, and white milk, as well as lower the selection and consumption of less healthful foods, during and after the various interventions (Lai et al., 2020; Marcano-Olivier et al., 2019; Miller et al., 2016). A recent systematic review of nudge interventions in schools by Metcalfe et al. (2020) found that behavioral nudges generally have a positive relationship with selection of meal items and mixed relationship with consumption.

Previous studies examining precommitment, which involve making choices in advance of mealtime in the absence of stimuli such as smell and visualization of the foods, found these methods to be an effective means of encouraging healthier choices (Just et al., 2008; Miller et al., 2016; Stites et al., 2015). Miller et al. (2016), combining preordering and nudging, found that students in fourth and fifth grade who preordered their lunches were more likely to select the fruit, vegetable, and low-fat milk NLSP meal components than students who did not preorder. Students who received a nudge based on USDA MyPlate recommendations while preordering were 51.4%, 29.7%, and 37.3% more likely to select a fruit, vegetable, and low-fat milk, respectively, than students who went through the normal lunch line. Outside the school setting, Van Epps et al. (2016) found calorie content of selected meals had a strong negative relationship with the time between preordering and receiving the meal.

Other studies suggest peers are a powerful influence. Lowe et al. (2004) showed primary-school students' videos of peer-aged “heroes” eating and enjoying produce and gave small prizes to students who selected and consumed fruits and vegetables during school lunch and snack times. During the intervention, students ate more fruits and vegetables at school and at home. The students also expressed increased enjoyment from fruits and vegetables following the intervention. Trogdon et al. (2008) highlighted a positive correlation between mean peer weight and individual weight in adolescents. Holford (2015) found peers influence participation in free-school school meal programs in the United Kingdom. Peer effects and social comparisons on individual choices also affect energy conservation, voting, retirement savings, and charitable giving (Allcott, 2011; Ayres et al., 2013; Costa & Kahn, 2013; Frey & Meier, 2004; Gerber & Rogers, 2009; Nolan et al., 2008; Schultz et al., 2007). Furthermore, the effects of peer influence on adolescent risk behaviors such as the use of alcohol, drugs or tobacco, unprotected sexual activity, crime, and reckless driving are well documented (Albert et al., 2013). While not still fully understood why, adolescents are more susceptible to peer influences than adults. Peer influences are linked to the brain's socioemotional reward system; as individuals age and the cognitive-control system develops, individuals develop additional capacity to resist peer influence (Albert et al., 2013).

In this study, we combine the behavioral economics principles of nudging, precommitment, and social comparisons to encourage the selection of fruits and vegetables by middle school children participating in the NSLP in two Florida schools. Miller et al. (2016) and Allcott (2011) provide the motivation for this study. Allcott (2011) provides Home Energy Report letters to residential customers. These reports present bar graph depictions of the customers' electricity use as well as information pertaining to the electricity use of the customers' neighbors. Allcott found that the letters reduced energy consumption by 2.0%. In the current study, similar to Miller et al., we invited participants to preorder their school lunches each morning during their homeroom class for 9 weeks. During Weeks 4 through 7, all participants received informational nudges related to their own prior selections of fruits and vegetables when they preordered their meals. In addition to receiving information pertaining to their own selections, some participants (treatment B) also received information pertaining to the prior selections of their peers defined as other study participants in the same grade and school. A final group of participants (treatment C) received information pertaining to their selections, peer selections, and USDA-recommended selections. Like Allcott (2011), we provide participants with information regarding their own and peers' behaviors using bar graphs.

Economic theory predicts students should select the foods that maximize their utility. However, this is not always the case. Economists have identified numerous biases and cognitive errors that lead individuals to make suboptimal decisions (The Decision Lab, 2020). For example, present biasedness may prevent an individual from selecting a healthier food option even if they value sound nutrition as the benefits of healthy eating are received in the future while the cost (eating a less desirable food) is incurred up front (Sunstein, 2015). Furthermore, to reduce the mental burden of the decision-making process, individuals often take shortcuts by making habitual decisions, which can lead to poor dietary choices (Cohen & Babey, 2012). The ability to employ shortcuts in some decision-making processes is known as the dual-process theory prevalent in both cognitive and social psychology (Evans & Stanovich, 2013). Dual-process theory draws a distinction between two kinds of thinking, one fast and intuitive, the other slow and deliberative. Often, fast and intuitive thinking is optimal—like when driving a car. However, at times, habitual decisions can lead to suboptimal outcomes, such as habitually snacking on a bag of chips while watching TV. Precommitment can prevent impulsive decisions by eliminating sensory cues that trigger individuals to choose less healthy choices (Stites et al., 2015). Similarly, informational nudges can encourage decision makers to overcome biases and cognitive errors by causing them to switch from fast and intuitive thinking to slow and deliberative thinking as depicted in Figure 1.

By providing students with information regarding their prior selection of fruits and vegetables while preordering, we are making these choices more salient in hopes of interrupting habitual behaviors. Hence, we anticipate students who receive information regarding prior selections will be more likely to select the vegetable and fruit meal components when nudged during preordering. While peer pressure and influence frequently lead adolescents to make unhealthy or risky choices, we anticipate that providing students with information about their peers' selections of fruits and vegetables will enable them to make social comparisons and encourage the selection of these items. We recognize the potential for boomerang effects (Schultz et al., 2007), in which individuals who are consuming more fruits and vegetables than their peers may decrease their selection of these items following social comparisons. We also hypothesize that students who receive informational nudges containing the USDA recommendations will be reminded of the importance of consuming a healthy diet and be more likely to select vegetable and fruit components when nudged. Thus, we anticipate the effect of each treatment will be positive with the largest effect occurring when the three nudges are combined.

To the best of our knowledge, this is the first study that uses informational nudges related to own or peer groups' prior behaviors to encourage the selection of fruits and vegetables among middle school children while preordering. While we hypothesize that informational and social comparison nudges will positively impact the selection of fruits and vegetables, we find negative effects on the selection of vegetables and no significant effects on the selection of fruits. At the first study school, we found that students in the treatment groups who received own order history nudges and who received own and peer order history nudges were significantly less likely to select a vegetable during the postintervention phase of the study. At the second study school, we found students in the second treatment group, who received own and peer order history nudges, were significantly less likely to select a vegetable during both the intervention and postintervention phases of the study relative to their selections during the baseline period, suggesting receiving information on peer selection backfired and negatively impacted the selection of vegetables with the impacts persisting after nudging ceased.

2 METHODS

2.1 Experimental design

The study was conducted during the 2013–2014 school year in Alachua County, Florida, following the University of Florida Institutional Review Board (IRB) approval (#2012-U-1245). Students at two middle schools (grades 6–8) were recruited to participate in early September through a demonstration of the preordering program in their homeroom classrooms. While participants did not receive monetary compensation, they were allowed to pick up their lunches before other students.

Potential study schools (from the set of seven middle schools in the district) were identified by the district's Director of Food and Nutrition Services in coordination with the district's Superintendent and the school principals; budget constraints only allowed the study to be implemented in two schools. We sought schools that collectively represented the district's demographics and schools that were not participating in other nutrition-related interventions during the study period. The Alachua County Public School system is made up of both schools that receive Title I funding and those that do not. Thus, the study was conducted at one school that did not receive Title I funding (school 1) and one lower income, Title 1 school (school 2). While both schools have diverse student bodies with respect to race and ethnicity, school 2 has a higher proportion of students from underrepresented groups.

During the study, participants preordered their lunches each morning for nine consecutive weeks using a web-based computer program designed for this study. For the first 3 weeks (baseline phase), students preordered their lunches without receiving any nudges. Then, students were randomly assigned to one of three treatment groups, and for the next 4 weeks (intervention phase) students received informational nudges while preordering. The nudges received by the students depended on their assigned treatment group, which remained the same throughout the study.

Students in treatment A received information pertaining to the average number of servings of fruits and vegetables they selected in the past five school days when they ate an NSLP lunch, students in treatment B received information on their own selections for the past five school days as well as information pertaining to the average selections of their peers for the same period, and students in treatment C received information on their own selections, peer selections, and the USDA recommended number of servings. Peer selection information was the average number of servings of fruits and vegetables selected per day by students who ate an NSLP lunch and were in the same grade and school. Recommended information was based on USDA MyPlate guidelines, which recommend consuming a protein, grain, fruit, vegetable, and dairy at each meal (USDA, n.d.). Thus, the recommended servings of fruits and vegetables were one serving of each item per meal. The information pertaining to own, peer, and recommended servings was presented using bar graphs. Teachers, administration, and students were consulted at both study schools before finalizing the design to ensure participants could correctly understand and interpret the graphical information. An example of a graph received by participants in treatment C is shown in Figure 2.

Following the intervention phase, participants preordered for another 2 weeks without receiving nudges. This allowed us to determine if the nudges had effects that lasted past the end of the intervention.

Randomization into treatment groups was done by school and strata based on the number of times the student preordered during the 3-week baseline phase of the study; students who ordered less than 40% of the time during the baseline period formed one stratum and those who ordered more frequently were a different stratum. At school 1, one-third of the students in each stratum were randomly assigned to each of the three treatments, while at school 2 half of the students in each stratum were assigned to treatment A and the other half were assigned to treatment B. Once assigned to a treatment, the student remained in the same treatment group for the duration of the study. Students at the second study school were only randomized into treatments A and B due to high rates of attrition during the baseline period of the study. In accordance with IRB regulations, children participating in the study had the option to withdraw from the study at any time. Approximately 27% of the students enrolled in grades 6–8 at school 2 signed the assent form agreeing to participate in the study, yet only 62% of the students who signed the assent form preordered a lunch during the baseline period of the study. In contrast, approximately 14% of the students enrolled in grades 6–8 at school 1 signed the assent form and 79% of these students preordered a lunch during the baseline period of the study.1

We analyze attrition bias by constructing an indicator variable that takes the value of 1 if the participant preordered in all three phases of the study, and 0 otherwise. We then model attrition as a function of assigned treatment, school, and student-level demographic variables using logit regression analysis. We find no evidence of attrition bias across treatment groups; however, students at school 2 were significantly more likely to exit the study than students at the first study school.2

During the study, on school days when participants choose to eat the school lunch (as opposed to bringing a lunch from home), participants preordered their lunches before 9:00 a.m. Schools participating in the NSLP program are required to offer the five NLSP meal components (meat/meat alternative, grain, fruit, vegetable, and milk) each school day. During preordering, available items were displayed by meal component categories (e.g., broccoli and carrots as vegetable options and apple wedges and melon as fruit options). Each offering was categorized based on discussions with the food service director and in accordance with the NSLP guidelines. While the menu was set at the district level, each school had some flexibility in the specific fruit and vegetable items offered. French fries were served more frequently as a vegetable option at school 1. The cafeteria staff at the study schools referred to the meat/meat alternative lunch component as the main dish and the grain component as starch; hence, the preordering program employed the same terminology to avoid confusing the participants.

Students must select at least three of the five offered components and one of the selected components must be a fruit or vegetable for the school to receive reimbursement for the lunch under the NSLP (USDA, 2012). Thus, participants who selected less than three items were prompted to select additional items and given the opportunity to change their orders. If a student ordered more than one item from any NSLP meal component category, they were prompted that they would need to pay extra for each additional item and given the opportunity to change their order. These prompts were similar to the verbal prompts that students received from cafeteria staff. Participants received these prompts less than 1% of the time when preordering. We analyze students' final selections.

At lunchtime, study participants picked up their preordered meals and completed their transactions as normal using their personal identification numbers (PINs). These PINs were linked to account information indicating if the student qualified to receive a free or reduced-price lunch. Students from households with incomes less than 130% of the federal poverty level can qualify for free lunches, while students from households with incomes between 130% and 185% of the federal poverty level can qualify for reduced-price lunches; students who do not qualify for free or reduced-price lunches may purchase full-price lunches, which are minimally subsidized (USDA, 2015). The preordering software allowed students to order only NSLP lunch components. Additional à la carte items could be purchased in the cafeteria at lunch.

Participants also used their PIN to log on to the preordering software program, allowing us to track students throughout the study and match their orders with race, gender, grade level, and NSLP income-eligibility status (free, reduced price, or full price) information provided by the school district's administration.

3 RESULTS

4 DISCUSSION

School administrators and policymakers are showing interest in innovative ways to increase the consumption of healthier items in school cafeterias while also improving the efficiency and profitability of school meal programs. Previous studies found preordering and nudging can be effective means of achieving these objectives. We contribute to this growing body of literature by combining preordering with nudging. To our knowledge, Miller et al. (2016) is the only prior study that analyzes the impacts of preordering with nudging among NSLP participants. We expand on their study by adding an element of social comparison.

Although Miller et al. (2016) found nudges received during preordering increased the likelihood of students selecting a fruit, vegetables, and low-fat milk, the analysis for school 2 presented here suggests that receiving information regarding peer behaviors while preordering backfired leading to a negative impact on the selection of vegetables with the impact persisting after nudging ceased. We found no statistically significant effects of the other two nudging treatments during the intervention phase. Unlike the study conducted by Miller et al. (2016), we cannot compare the effects of our nudging interventions on students who did not preorder because we do not have a record of selections made by students who did not preorder. We also note selection rates of the fruit and vegetable components during the baseline period of our study are much higher than the selection rates reported by Miller et al. (2016) during the baseline phase of their study hence leaving less room for improvement.

We hypothesized providing students with information regarding their prior sections would nudge them toward making healthier choices. However, we found that this nudge (treatment A) did not have a significant impact during the nudging phase. Surprisingly, students who received information about their prior selections were significantly less likely to select a vegetable once nudging ceased at school 1. We also found that providing information regarding peer behaviors had a negative impact on selection of vegetables during the nudging phase of the study and the effects intensified when the nudging ceased at school 2. The negative effects observed during the postintervention phase of the study are similar to widely cited rebound effects in the medical literature in which a patient's symptoms reappear after the use of a medication is ceased. Similarly, in the psychology literature, there is some evidence that behaviors also return when efforts to suppress these behaviors are ceased (Denzler et al., 2010; Follenfant & Ric, 2010; Leckman, 1986). However, unlike the rebound effects found in the medical and psychology literatures, our treatments had null and negative effects during the nudging phase of the study.

One possible justification for observed negative impacts is that differences in vegetable offerings across the study phases are driving selection behaviors. Supporting Information S1: Table A.27 in the supporting information summarizes the most offered vegetables by school and study phase. French fries were offered more frequently during the baseline than in the other two phases of the study at both schools. To determine if our results are driven by differences in the availability of French fries across study phases, we constructed a control variable that took the value of 1 when French fries were offered, and 0 otherwise. Using this additional control variable, we repeated the analyses for the selection of vegetable by treatment. The results of these analyses presented in Supporting Information S1: Tables A.28 and A.29 in the supporting information are similar to the results in Tables 2 and 3, respectively. The only noteworthy exception is that the effect of treatment A at school 1 is no longer significant during the postintervention phase. The similarity of other treatment effects when controlling for offering of French fries suggests that our findings are not driven by offerings of French fries.

The negative estimated treatment effect for treatment B suggests that eating vegetables may be stigmatized or perceived as uncool; we speculate that students receiving peer information may have been reminded that eating vegetables was uncool and hence selected this lunch component less frequently. Further investigation indicates that the results were not driven by the so-called boomerang effect (Schultz et al., 2007), in which students who were selecting vegetables more frequently than their peers reduced their selection of vegetables. To determine this, we constructed a variable that took the value of 1 when the participants received graphical information that their own prior selections exceeded that of their peers, and 0 otherwise. We interacted this indicator variable with the treatment variable and repeated the analyses, presented in Tables 2 and 3, and Supporting Information S1: A.8 for treatment B to conduct a heterogeneous treatment analysis. We find that participants whose average selection of vegetables for the prior 5 days exceeded that of their peers were more likely to select the vegetable component during the intervention phase of the study as shown in Supporting Information S1: Tables A.30 and A.31.

Receiving information about the USDA recommendations seemed to mitigate the negative effects of receiving information on peer selections. Perhaps, the inclusion of the USDA recommendations reminded the students of the importance of eating healthy, providing a nudge that was sufficient to offset the negative effect of the peer information. Prior studies suggest that MyPlate nudges positively influence behavior; Miller et al. (2016) found MyPlate nudges positively impacted the selection of fruits, vegetables, and low-fat milk while preordering. Similarly, Brown et al. (2014) found undergraduate students who received biweekly text messages of the MyPlate icon and the USDA's Dietary Guidelines for 7 weeks self-reported increased consumption of fruits. However, we are cautious to draw strong conclusions about the impact of the MyPlate nudge in our study due to the relatively small number of students assigned to this treatment group.

We find no effect of the interventions on the selection of fruit. This may be due to strong preferences for the fruit component as suggested by approximately 92% of the meals selected during the baseline phase contained a fruit.

A limitation of our study is the lack of a control group in which students either ordered in the cafeteria or preordered without receiving nudges throughout the duration of the study. The decision to treat all study participants stemmed from our desire to minimize spillover effects. Students ordered during their homeroom class, taking turns on the available computers. While efforts were taken to provide privacy while ordering, we felt that if students observed only some students received graphs on their screens while preordering, they would be more likely to discuss the study with other participants. While efforts were taken to minimize spillover effects such as using privacy screens and reminding the students not to discuss the study with other students, given the length of the study, we recognize that spillover effects were likely and may have impacted the findings. Unfortunately, we have no means of estimating these effects. Originally, we intended to add a network analysis component to the study and collect information about participants' friend network and who they ate lunch with. However, the district did not approve this part of the study due to concerns over the ability to maintain students' confidentially. Social network analysis to better understand spillover effects and the role of peers in food choices is an area for future research.

We also found the attrition rates varied by school. While this is purely speculation, differences in the attrition rates across schools may have been driven by differences in the availability of computers for preordering at the two study schools. At school 1, students preordered their lunches using computers within their homeroom classrooms. At school 2, students had to go to the library or computer lab to access a computer to complete the preordering process. It is likely that this additional burden of preordering at school 2 contributed to the higher attrition rate. A recent study that asks participants to narrate their thinking aloud as they navigate placing hypothetical preorders using a web-based program suggests that restrictive timeframes within which preorders could be placed are a potential barrier to engaging in preordering; furthermore, the study finds some participants did not like forecasting in the morning what they would feel like eating for lunch (Breathnach et al., 2020). These factors may have contributed to attrition. Following our study, the study district conducted a pilot program allowing high school students to preorder. The district administrators informed us the pilot program was abandoned because students preferred the social aspect of waiting in line.

While we took care to confirm with teachers and administrators that the students in our sample could understand and interpret bar graphs, some students may have had limited ability to understand them. A lack of ability to understand the information conveyed in the nudges could render them less effective.

Furthermore, our sample was relatively small. Future research could explore if similar results are obtained from larger samples or in other districts.

5 CONCLUSIONS

Preordering can create opportunities to actively involve parents in children's decisions. The software used in this study could easily be converted into a phone application, allowing parents to track their kid's selection of various meal items and use this information to educate and influence them.

Interest in the use of preordering programs by administrators and policymakers centers around it being an easily implemented, low-cost intervention (Migliavada et al., 2021). Preordering may help reduce food waste by allowing food service personnel to prepare only the necessary number of servings; however, uncollected orders can be a significant source of food waste (Migliavada et al., 2021). Metcalfe et al. (2020) found nudging interventions in schools generally increase food waste. More research is needed to fully understand the implications of preordering on reducing food waste and the cost of serving school lunches.