A Traffic-Light Label Intervention and Dietary Choices in College Cafeterias

Abstract

Objectives. To examine whether traffic-light labeling and choice architecture interventions improved dietary choices among students at a northeastern US university.

Methods. In 6 cafeterias at Harvard University, in Cambridge, Massachusetts, we implemented a 7-week intervention including traffic-light labeling (red: least nutrient rich; yellow: nutrient neutral; green: most nutrient rich), choice architecture (how choices are presented to consumers), and “healthy-plate” tray stickers. During the 2014–2015 academic year, 2 cafeterias received all interventions, 2 received choice architecture only, and 2 were controls. We analyzed sales for 6 weeks before and 7 weeks during interventions. Using interrupted time-series analyses, we measured changes in red, yellow, and green items served. We collected 1329 surveys to capture perceptions of labeling.

Results. Among 2.6 million portions served throughout the study, we found no significant changes in red (–0.8% change/week; P = .2) or green (+1.1% change/week; P = .4) items served at intervention sites compared with controls. In surveys, 58% of students reported using traffic-light labels at least a few times per week, and 73% wanted them to continue.

Conclusions. Although many students reported using traffic-light labels regularly and wanted interventions to continue, cafeteria interventions did not demonstrate clear improvements in dietary quality.

Obesity is second only to tobacco as a leading cause of death among US adults.¹ Diet is a key factor in the obesity epidemic, and many young people develop dietary habits that increase later health risks.² College students in particular exhibit poor dietary intake and low levels of physical activity,^3–8 gain weight faster in their first year than average Americans at the same age,⁹ and see few improvements in health behaviors during college.¹⁰ In fact, the proportion of overweight and obese college students increased from 33% in 2010 to 40% in 2015 according to American College Health Association health assessments,^11,12 with students gaining an average of 2 to 7 pounds in the first 3 to 4 months of college.^13–15 Encouraging healthy eating in college could lead to long-lasting effects, and the university cafeteria could be an effective intervention site.

There has been very little research on interventions to encourage healthy eating among college students. Among 14 college student dietary interventions identified in a 2013 systematic review by Kelly et al., in-person interventions, such as educational nutrition courses and health consultations, or online nutrition educational activities produced short-term, minimal improvements to eating behaviors.¹⁶ Only 3 studies examined point-of-purchase interventions that displayed nutrition information to consumers while they made food choices. A 4-week intervention at a university cafeteria by Buscher et al. consisted of large posters that displayed encouraging phrases next to healthy food (e.g., “Boost yourself with fresh pineapple”) and found that sales of yogurt, pretzels, whole fruit, packaged salads, and candy increased.¹⁷ A 3-week intervention at a college cafeteria by Peterson et al. included healthy-choice indicator labels and found increased consumption of cottage cheese and low-fat salad dressing.¹⁸ Finally, a 5-week intervention at a university convenience store by Freedman and Connors added healthy-choice indicator labels but found no significant changes in the sales of labeled items.¹⁹

Compared with healthy-choice indicators that label only healthy items, traffic-light labels provide more information to consumers. Traffic-light labels classify food and beverages from green (most healthy) to red (least healthy) and provide information quickly and visibly at the point of sale.²⁰ A previous study conducted by Thorndike et al. in 2010 at a hospital cafeteria concluded that traffic-light labels, as well as a choice architecture intervention (improving the visibility and convenience of healthy foods), increased the number of healthy items purchased and decreased the number of unhealthy items purchased at a single hospital cafeteria.²⁰

In this study, we investigated the effect of a combined traffic-light labeling, choice architecture, and healthy-eating plate (HEP) intervention on the eating behaviors of undergraduates over 13 weeks. We hypothesized that the percentage of red-labeled items would decrease in the full-scale-intervention group compared with control and minimal-intervention (choice architecture only) groups, whereas the percentage of green-labeled items would increase relative to control and minimal-intervention groups. For surveys, we hypothesized that perceptions of the traffic-light labels would differ significantly by such factors as varsity athlete status or weight status because they might contribute to certain attitudes toward healthy eating.

METHODS

The study took place at Harvard University, a northeastern US college in Cambridge, MA, where more than 97% of undergraduates live on campus for all 4 years.²¹ Although the university publicly releases very limited demographic data, a survey that included responses from 80% of first-year students from the class of 2017 (n = 1311) revealed that 50% of students were female and 50% were male. Sixty-two percent of students were White, 25% Asian, 11% Hispanic or Latino, 10% Black or African American, 4% Indian, 2.5% American Indian, and 2.5% Alaska Native or Pacific Islander.²² More than 70% of Harvard students receive some form of financial aid,²¹ with 15% reporting family incomes less than $40 000, 15% between $40 000 and $80 000, 18% between $80 000 and $125 000, and 52% greater than $125 000.²²

At this college, sophomore, junior, and senior students live in 1 of 12 residential houses of approximately 400 students, and each residential house includes a cafeteria. The study setting was 6 of these cafeterias. Two of the cafeterias received the full intervention, including traffic-light labels on all foods and beverages and a choice architecture intervention combined with HEP tray stickers. Two of the cafeterias received the minimal intervention with the choice architecture intervention only. The final 2 cafeterias served as the control sites, with no interventions.

We designated cafeterias as intervention or control sites on the basis of the willingness of the cafeteria leadership to participate in the interventions. Although our unit of analysis was the cafeteria, students who ate in each cafeteria were representative of the entire student body because of the method of residential house assignment. After freshman year, the college randomly assigns groups of 1 to 8 students to residential houses, and students eat most of their meals in their assigned cafeterias. Although students may occasionally eat a meal outside of their residential house cafeterias, there are university policies that encourage eating at assigned residential houses.

Interventions

Full intervention: traffic-light labeling and healthy-plate stickers.

We evaluated each menu item against 5 positive and 6 negative nutritional criteria that we constructed on the basis of the prior study in a hospital cafeteria (see the box on page e3),²⁰ and by incorporating additional data linking nutrition to mortality or chronic disease outcomes (Note A, available as a supplement to the online version of this article at http://www.ajph.org). The labels followed university regulations prohibiting the posting of numerical nutrition facts, and they did not include calories as a criterion. Positive criteria included fruits, vegetables, whole grains, lean protein, and low-fat dairy, whereas negative criteria included saturated fat, added or high sugar, sodium, red meat, and refined starch.

Nutritional Criteria Used to Determine Traffic-Light Label Colors: Harvard University, Cambridge, MA, 2014

Positive Criteria

• Fruit source, or > 80% juice content

• Vegetable source

• Whole grain with carbohydrate–fiber ratio < 10

• Lean protein: < 5 g saturated fat with protein content at least 12 g

• Low-fat dairy: < 2 g saturated fat and > 200 mg calcium

Negative Criteria

• Saturated fat: > 5 g

• Added sugar: contains added sugar and > 8 g of total sugar

• High sugar: > 20 g

• High sodium: > 600 mg

• Red meat

• Refined starch: > 6 g and carbohydrate–fiber ratio > 10

Note. We used 5 positive and 6 negative criteria to evaluate both food and beverage items. Water, tea, and coffee were the only exceptions, which we labeled green.

We evaluated 467 menu items for labels, representing all menu items offered across all cafeterias during the study period. We determined health scores by adding positive and subtracting negative criteria, and assigned green labels to net positive scores, yellow labels to neutral scores, and red labels to net negative scores (Figure A, available as a supplement to the online version of this article at http://www.ajph.org). Labels included sidebars that listed special dietary needs, including “vegetarian,” as well as the nutritional criteria an item contained. Posters defined green labels as “nutrient-rich choice,” yellow labels as “nutrient-neutral choice,” and red labels as “more nutrient rich choice in [green circle] or [yellow circle].” The Harvard University Dining Services was responsible for displaying the labels in cafeterias; they did not label small items, including salad dressing, bagels, and condiments. We also attached stickers based on “Healthy Plate” materials, developed at the Harvard T. H. Chan School of Public Health, to cafeteria trays that visually displayed recommended portions of food types and the relative size of each food type on a plate.²³ The traffic-light labels began at the first full-intervention cafeteria on October 12, 2014, and continued until December 2, 2014, a total of 7 weeks and 3 days. The traffic-light labels started at the second full-intervention cafeteria a week later on October 19, 2014, and continued until December 1, 2014, a total of 6 weeks and 2 days. The HEP stickers ran at both full-intervention cafeterias from October 27, 2014 until December 2, 2014, a total of 5 weeks and 1 day. Delays in study approval at the intervention cafeterias led to staggered initiation.

Full- and minimal-intervention sites: choice architecture intervention.

Both the full- and minimal-intervention cafeterias received the choice architecture changes, including moving healthier food and beverage items to make them more accessible or convenient to reach. We organized serving lines so that healthier items including beans, whole grains, and vegetables were at the beginning. We added 4-gallon water pitchers to intervention cafeterias that dispensed cucumber- or fruit-infused water. The choice architecture changes were in place from the beginning of the interventions through the end (October 14, 2014 to December 2, 2014, a total of 7 weeks and 1 day).

Data Collection and Measures

Cafeteria servings.

We measured the primary outcome as the change in proportions of red, yellow, and green items served per week according to Dining Services records, overall and by subgroups of food categories, including entrees and beverages. Each cafeteria offered the same menu, which changed daily. Cafeteria staff determined menus for the study period well before the intervention began. Cafeterias served the food buffet-style, and staff constantly replenished food in the buffet line. Although we did not have the resources to measure food waste or consumption directly, Dining Services staff tracked how many servings of each food item students took from the buffet line. The executive chef and University Health Service dietitians set portion sizes. For most food items, Dining Services staff reported to us the number of servings taken from the buffet line. For cereals and beverages, staff provided the overall volume served to students, and we derived servings from the serving size on the product packaging.

Online surveys.

From October 6, 2014 to October 12, 2014, at baseline before any interventions were implemented, we conducted an online survey of students to query how and whether they used available nutrition information to guide choices and asked if they wanted to have nutritional labels and what information should be on those labels. At baseline and follow-up, we used student e-mail discussion lists to send the survey to all sophomore, junior, and senior students at Harvard, including the 6 residential houses that were involved in the study (control, minimal-intervention, and full-intervention cafeterias) and the 6 residential houses that were not. Students from the second through fourth years of college (sophomores to seniors) live in these houses. We chose these residential houses because we wanted a diverse range of students, across several years. First-year students live in residential houses separate from the rest of the student body and primarily use a first-year-only cafeteria; they were therefore not included in this study because there was no appropriate comparator site.

From December 5, 2014 to December 12, 2014, after the interventions had concluded, we repeated the survey and again sent it to non–first-year students. To capture differences between those living in the full-intervention residential houses (and dining in those cafeterias) and those living in the other residential houses, we stratified survey results between this group and all other students. The other students could have been in any of the 10 non–full-intervention residential houses, including those in the minimal-intervention and control sites. We did this because many of the questions pertained to labeling, and only full-intervention sites received labeling. After several students mentioned red meat as an important health concern in preintervention formative focus group discussions, we also added 1 question about health risks associated with red meat consumption to the postintervention survey.

Statistical Analyses

We examined the effect of the intervention on red, yellow, and green items served using interrupted time-series analyses.^24,25 This study met the requirements of interrupted time-series analysis because it had more than 3 data points before and after the interventions and clearly defined intervention points.²⁶ We compared the trends and slope of intervention and control groups. The outcome was the difference between items served in intervention cafeterias and control cafeterias. We ran Prais–Winsten regressions to account for serial correlation.²⁷

For surveys, we used χ² analysis to test for differences between pre- and postintervention survey responses, and between postintervention subgroups stratified by gender, varsity athlete status, self-reported body mass index status, and residency. The χ² test was appropriate to analyze pre- and postintervention survey responses because we were not able to fully match samples. For purposes of χ² analysis, we collapsed answers to questions with neutral responses into dichotomous outcomes in which neutral responses were considered negative responses for questions about change as a result of the interventions.

RESULTS

Over the 13-week study, 2 648 277 portions of food and beverages were served in 434 625 meals. Among items available in the cafeterias, 45% were labeled green, 21% yellow, and 34% red. Among entrees (both vegetarian and meat) available in the cafeterias, 38% were green, 20% yellow, and 42% red. Among beverages available in the cafeterias, 15% were green, 20% yellow, and 65% red.

Cafeteria Data

The proportions of all red-, yellow-, and green-labeled items (both food and beverages) served per week changed modestly over the course of the study (Figure 1). In the interrupted time-series comparison between the full-intervention and control groups, we saw nonsignificant change in slope for proportions of red (–0.8%; P = .199), yellow (–0.1%; P = .940) and green (+1.1%; P = .400) items before and after the intervention (Table 1). The only statistically significant change between any combination of intervention and control sites involved yellow entree items. When we compared the minimal-intervention and control sites, yellow entree items decreased 2.2% more per week at the minimal-intervention sites than at the control sites (P < .05). For beverages, there was no significant change in slope between full-intervention and control sites for proportions of red (–3.1%; P = .125), yellow (+4.0%; P = .176) and green (+1.4%; P = .757) items before and after the intervention. Although nonsignificant, red beverage portions served decreased an average of 4.1% (P = .069) per week at minimal-intervention sites compared with controls.

FIGURE 1—

Proportions of All Red-, Yellow-, and Green-Labeled Food and Beverages Served per Week at (a) Full-Intervention, (b) Minimal-Intervention, and (c) Control Cafeterias: Harvard University, Cambridge, MA, 2014

TABLE 1—

Regression Results of All Items Served per Week at Full-Intervention, Minimal-Intervention, and Control Cafeterias Before and After the Interventions (Percentage of Total Items Served): Harvard University, Cambridge, MA, 2014

Comparison	% Change (95% CI)
Red label
Full vs control
Difference in immediate change after intervention	0.73 (–4.25, 5.71)
Difference in change in slope after intervention	–0.78 (–2.04, 0.49)
Full vs minimal
Difference in immediate change after intervention	1.72 (–1.16, 4.60)
Difference in change in slope after intervention	–0.71 (–1.45, 0.02)
Minimal vs control
Difference in immediate change after intervention	–0.63 (–4.92, 3.66)
Difference in change in slope after intervention	–0.27 (–1.41, 0.86)
Yellow label
Full vs control
Difference in immediate change after intervention	4.61 (–1.96, 11.17)
Difference in change in slope after intervention	–0.06 (–1.74, 1.62)
Full vs minimal
Difference in immediate change after intervention	3.98 (–6.90, 14.87)
Difference in change in slope after intervention	–1.76 (–5.72, 2.19)
Minimal vs control
Difference in immediate change after intervention	0.97 (–5.59, 7.54)
Difference in change in slope after intervention	0.62 (–1.07, 2.31)
Green label
Full vs control
Difference in immediate change after intervention	–4.13 (–13.42, 5.16)
Difference in change in slope after intervention	1.10 (–1.72, 3.92)
Full vs minimal
Difference in immediate change after intervention	–5.09 (–15.56, 5.38)
Difference in change in slope after intervention	1.50 (–1.49, 4.48)
Minimal vs control
Difference in immediate change after intervention	–0.34 (–6.08, 5.41)
Difference in change in slope after intervention	–0.41 (–1.88, 1.07)

Note. CI = confidence interval.

Surveys

Five hundred and fifty students initiated the preintervention survey, with 95% completing the survey, and 779 students initiated the postintervention survey, with 88% completing the survey. In the preintervention survey, the mean age was 20 years, 65% identified as women, 15% were varsity athletes, 30% lived in full-intervention sites, and 18% lived in minimal-intervention sites. In the postintervention survey, the mean age was 20 years, 68% identified as women, 12% self-identified as overweight or obese, 23% were varsity athletes, 31% lived in full-intervention sites, and 17% lived in minimal-intervention sites (Table A, available as a supplement to the online version of this article at http://www.ajph.org).

Nutrition facts and traffic-light labels.

Prior to the intervention, nutrition information for items served in cafeterias was only available online, yet 91% of preintervention students and 87% of postintervention students either wanted nutrient information on labels or had no preference (Table B, available as a supplement to the online version of this article at http://www.ajph.org). Only 9% of pre- or postintervention students surveyed reported checking nutrition facts online daily. Sixty-two percent and 68% of postintervention students said nutrition information often or always affected food and beverage choices, respectively (Table 2). Postintervention women were more likely than men (56% vs 45%) to consider nutrition information daily when making choices. Ninety-three percent of postintervention athletes wanted nutrient information on labels or had no preference, compared with 85% for nonathletes (Table C, available as a supplement to the online version of this article at http://www.ajph.org).

TABLE 2—

Postintervention Survey Results (Percentages) for Total Sample and by Residency: Harvard University, Cambridge, MA, 2014

Survey Question	Total Sample (n = 764), %	Other Residential Houses (n = 535), %	Full Intervention (n = 229), %
Consider nutrition information daily.	53	54	51
Check nutrition facts online or on mobile app daily.	9	9	9
Nutrition information affects food choices often or always.	62	64	57
Nutrition information affects beverage choices often or always.	68	70	64
Use the traffic-light labels every meal.	17	15	20
Use the traffic-light labels at least daily.	25	22	29
Use the traffic-light labels at least a few times a week.	47	37*	58*
Traffic-light labels were helpful.	59	59	60
Traffic-light labels were not helpful.	21	16*	25*
Traffic-light labels should continue to be used or no preference.	73	78	69
Labels were not as effective at semester’s end as when the labels were introduced.	21	13*	29*
Traffic-light labels change food selected.	48	47	49
Traffic-light labels change amount consumed.	36	38	34
Traffic-light labels change perceptions about the healthfulness of specific foods.	62	61	63
HEP stickers were helpful.	34	48*	22*
HEP stickers were not helpful.	32	18*	45*
HEP stickers should continue to be used or no preference.	77	89*	67*
HEP stickers change food selected.	14	22*	7*
HEP stickers change amount consumed.	17	28*	7*
HEP stickers change perceptions about the healthfulness of specific foods.	20	27*	13*

Note. The “Other Residential Houses” group refers to students who lived at houses that included minimal-intervention cafeterias, control cafeterias, and cafeterias in residential houses not part of the study. These students did not live where the labeling intervention took place, but they could have occasionally seen or heard about the interventions. HEP = healthy-eating plate.

^*P < .05 for difference within a subgroup.

Fifty-nine percent of postintervention students who noticed the traffic-light labels thought they were helpful, and 73% said they should continue to be used after the study. Forty-eight percent of postintervention students reported that traffic-light labels changed the food items they selected, and 36% said they changed the amount of food consumed. Postintervention athletes were more likely than nonathletes (70% vs 55%) to report that traffic-light labels were helpful. Postintervention men and athletes were more likely to say traffic-light labels should continue to be used or had no preference (Table C).

In terms of nutrition knowledge, although most students correctly identified red meat as a source of micronutrients in postintervention surveys, just 43% correctly believed that red meat can increase risk of cancer and diabetes. Only 29% correctly stated that regularly eating small amounts of red meat can increase risk of heart disease and stroke (Table D, available as a supplement to the online version of this article at http://www.ajph.org).

Healthy Plate stickers.

Full-intervention residents were less likely than residents of houses not in the full intervention (22% vs 48%) to say that HEP stickers were helpful in postintervention surveys. Full-intervention residents were also less likely than other house residents (67% vs 89%) to say that HEP stickers should be available in cafeterias or had no preference in postintervention surveys. Only 7% of full-intervention residents said that HEP stickers changed the food selected or the amount consumed (Table 2).

DISCUSSION

To our knowledge, this is the first study of traffic-light labels and choice architecture interventions in a college population, and the first to look at both the efficacy and perceptions of HEP tray stickers. The size and scope of the study is substantial; it included more than 1300 survey responses over 2 surveys, 6 cafeterias, and more than 2.6 million food and beverage portions over 13 weeks. Traffic-light labeling and choice architecture interventions across several university cafeterias did not change portions of unhealthy foods (–0.8% change/week; P = .199) or healthy foods (+1.1% change/week; P = .400) served in intervention versus control sites. By contrast, labels were well received by participants. Only 25% of intervention residents reported in surveys that the labels were not helpful, and most students wanted labels and felt they were useful.

Label fatigue, the diverse needs and opinions of a university population, and a lack of knowledge about the health consequences associated with specific nutrition behaviors could have decreased the effect of the interventions at the university cafeterias. Label fatigue occurs when consumers pay less attention to labels over time. Twenty-nine percent of intervention residents surveyed reported that labels were less effective at the end of the study. In contrast to our results, a follow-up by Thorndike et al. on traffic-light labels at a hospital cafeteria found sustained healthier choices over 2 years after the interventions; that study did not use interrupted time-series analyses to examine trends before and after interventions.²⁸

Survey results revealed that students have diverse opinions and knowledge regarding nutrition. Women and athletes were significantly more likely to consider nutrition information daily, and women were twice as likely to check nutrition facts online. Athletes were significantly more likely to want nutrient information on labels. The success of a particular intervention could largely depend on what subgroup was exposed.

Recent research suggests that labels that focus on the health consequences of eating are more effective than simply listing nutrition facts of particular foods.²⁹ In our traffic-light labeling scheme, we used attributes including “high sugar” or “high salt,” but did not further make the connection of choices to health outcomes. To encourage healthy choices, perhaps it would be more effective to identify health consequences directly. Students’ responses regarding nutrition knowledge demonstrated the importance of this; 71% of postintervention students did not correctly identify the link between health consequences and red meat consumption. Future university food-labeling interventions should survey other nutrition knowledge outcomes such as the benefits of whole grains to investigate whether students with more knowledge about benefits associated with certain eating behaviors are more responsive to labeling interventions. Because university students are generally eating on their own for the first time, universities should incorporate nutrition education among the topics of health addressed on campus, similar to how universities now address alcohol and sexual assault. Additional education might reinforce connections between nutrition facts and associated health consequences. A study by Ha and Caine-Bish found that a college nutrition class emphasizing prevention of chronic diseases increased fruit and vegetable consumption among college students.³⁰ These educational programs have potential for enduring dietary change. One study of adults found that healthy dietary changes were evident 1 year after dietary advice was given.³¹

Despite these challenges, our surveys suggest that these interventions are valuable. First, 62% of postintervention survey respondents reported using nutrition information to help guide their choices often; similarly, a study at Yale University that found that 88% of students used nutrition information sometimes, often, or always to guide food choices, whereas just 4% of students thought cafeteria nutrition information never influenced choices.³² Second, most intervention residents (58%) reported in surveys that they used traffic-light labels at least a few times per week, and 73% of all students wanted them to continue after the study. In the Yale study, 96% of students thought it was a good idea to make nutrition information available.³² Third, since Harvard and many other universities do not display point-of-purchase nutrition information, students appear to receive almost no nutrition information for making food choices. Only 9% of students in pre- or postintervention surveys reported checking online nutrition information daily.

Our findings should be considered with the following caveats. Logistical delays postponed the start of interventions and limited postintervention data collection. We were unable to collect 9 days of data over the study period at all cafeterias. We also noted large week-to-week and site-to-site variations in portions served, making it more difficult to find significant differences between intervention and control sites. Another limitation of the study is the lack of individual-level data. Although students eat most of their meals in their resident cafeterias, they are still allowed to eat at any cafeteria, and it is possible that students in all cafeterias were affected by the intervention and ate differently in all cafeterias. This might bias our findings toward the null. The online surveys were limited by the use of a convenience sample and were conducted at 1 university, so results may differ for other students on the basis of school and location.

In this study, we found that a majority of college students at a large, northeastern US university are interested in learning more about nutrition and support nutrition labeling, and that a traffic-light labeling and choice architecture intervention at a college is feasible. The study did not demonstrate that traffic-light labels influenced choices. Future interventions in university cafeterias could use longer pre- and postintervention periods to increase the power to detect small changes in nutrition behaviors, track individual consumption to investigate whether interventions are more effective for certain subpopulations, and collect additional secondary outcomes such as self-reported consumption of healthy foods or willingness to try healthier foods. Future labeling interventions in university cafeterias should seek to include nutrition information, which students seem to appreciate, along with nutrition education to highlight the long-term health consequences of food choices.