To reduce added-sugar consumption, jurisdictions are considering requiring restaurant menu labels to identify high-added-sugar items. This study examined the impacts of added-sugar warning labels on hypothetical choices, knowledge of items’ added-sugar content, and perceptions of high-added-sugar items.
Study design
The design was an online RCT.
Setting/participants
National sample of adults (N=15,496) was recruited to approximate the U.S. distribution of sex, age, race, ethnicity, and education.
Intervention
Participants viewed fast-food and full-service restaurant menus displaying no warning labels (control) or icon-only added-sugar warning labels next to high-added-sugar items (containing >50% of the daily recommended limit).
Main outcome measures
The main outcome measures were hypothetical ordering of ≥1 high-added-sugar item, grams of added sugar ordered, and knowledge of items’ added-sugar content assessed in 2021 and analyzed in 2021–2022.
Results
Warning labels reduced the relative probability of ordering ≥1 high-added-sugar item by 2.2% (probability ratio=0.978, 95% CI=0.964, 0.992; p=0.002); improved knowledge of added-sugar content (p<0.001); and led to a nonstatistically significant reduction of 1.5 grams of added sugar ordered, averaged across menus (p=0.07). The label modestly reduced the appeal of high-added-sugar items, increased perceptions that consuming such items often will increase Type 2 diabetes risk, increased perceived control over eating decisions, and increased injunctive norms about limiting consumption of high-added-sugar items (ps<0.001). However, in the warning condition, only 47% noticed nutrition labels, and 21% recalled seeing added-sugar labels. When restricting the warning condition to those who noticed the label, the result for grams of added sugar ordered was significant, with the warning condition ordering 4.9 fewer grams than the controls (95% CI= –7.3, –2.5; p<0.001).
Conclusions
Added-sugar warning labels reduced the probability of ordering a high-added-sugar menu item and increased participants’ knowledge of whether items contained >50% of the daily value for added sugar. The modest magnitudes of effects may be due to low label noticeability. Menu warning labels should be designed for noticeability.
Registration
This study was registered at AsPredicted.org #65655
INTRODUCTION
Most U.S. children and adults consume added sugar in excess of the Dietary Guidelines for Americans’ recommended limit of 10% of daily calories,
U.S. Department of Agriculture, HHS, Dietary Guidelines for Americans 2020–2025, 2020; Washington, DC: U.S., https://www.dietaryguidelines.gov. Published December 2020. Accessed January 29, 2020.
increasing the population’s risk for cardiometabolic diseases.
U.S. Department of Agriculture, HHS, Dietary Guidelines for Americans 2020–2025, 2020; Washington, DC: U.S., https://www.dietaryguidelines.gov. Published December 2020. Accessed January 29, 2020.
,,
Individuals’ ability to reduce added-sugar intake relies largely on the environment in which they make food decisions.
U.S. Department of Agriculture, HHS, Dietary Guidelines for Americans 2020–2025, 2020; Washington, DC: U.S., https://www.dietaryguidelines.gov. Published December 2020. Accessed January 29, 2020.
Trends in food sources and diet quality among U.S. children and adults, 2003–2018.
making restaurants an important target for public health intervention. One barrier to informed choice in restaurants is the lack of added-sugar information.
An in-depth exploration of knowledge and beliefs associated with soda and diet soda consumption.
Although the U.S. mandates added-sugar labeling on packaged foods, calorie labeling on chain-restaurant menus, and disclosure of several nutrients upon request in chain restaurants, chain restaurants are not required to disclose added sugar let alone label high-added-sugar items. Therefore, the New York City (NYC) Council passed a bill in 2021 requiring added-sugar menu labels in chain restaurants to indicate prepackaged items that exceed the daily recommended limit. This policy is similar to NYC and Philadelphia laws requiring sodium labeling in chain restaurants.,
Online experiments have found that restaurant menu sodium warnings and multinutrient warnings (such as Chile’s octagonal nutrient labels) on food-ordering websites reduced hypothetical ordering of labeled items,
Do nutritional warnings encourage healthier choices on food ordering websites? An exploratory experimental study in Uruguay.
but there is a lack of research on added-sugar menu warning labels. The only study to examine such warnings found that compared with a control label, icon-only and icon-plus-text added-sugar warning labels were perceived as more effective and increased knowledge about items’ added-sugar content, with both label types performing similarly.
Perceived effectiveness of added-sugar warning label designs for U.S. restaurant menus: an online randomized controlled trial.
However, research with behavioral outcomes is needed to better understand the potential effectiveness of added-sugar menu labels. The goal of this study was to examine the effect of icon-only added-sugar warning labels displayed next to restaurant menu items high in added sugar on (1) ordering ≥1 high-added-sugar item in a menu ordering task, (2) grams of added sugar ordered in that task, and (3) knowledge about menu items’ added-sugar content. A secondary objective was to examine warning-label effects on perceptions predictive of behavior.
RESULTS
Appendix Table 2 (available online) shows the participant characteristics. The distribution of race, Hispanic ethnicity, sex (measured in this study vs. sex measured by the ACS), and education were similar to the 2018 ACS estimates. For annual household income before taxes, 27% reported ≤$35,000; 27% reported $35,001–$65,000; 18% reported $65,001–$95,000; and 28% reported >$95,000. There were no significant differences by condition.
Appendix Table 3 (available online) shows the number of items ordered by restaurant and item category. Figure 3A and B and Appendix Table 4 (available online) show primary ordering outcomes. In total, 81.5% in the warning condition ordered ≥1 high-added-sugar item from either menu, compared with 83.4% in the control condition, for an absolute difference of –1.9 percentage points (pps) and relative difference of –2.2% (PR=0.978, 95% CI=0.964, 0.992; p=0.002; NNT=53). Effects were larger for the full-service than for the fast-food menu (–2.7% vs –1.3%; PR=0.973, 95% CI=0.953, 0.994 vs PR=0.987, 95% CI=0.966, 1.008; NNT=53 vs NNT=112).
Figure 3(A) Percentage of participants who ordered ≥1 HAS item, (B) added sugar ordered, and (C) the percentage of items correctly classified as HAS or not.
***p<0.001, **p<0.01, *p<0.05, and ap<0.10 from (A) Poisson regression models with robust SEs and (B, C) linear regression models in which the outcome was regressed on an indicator for the warning label condition. Means and 95% CIs (indicated by error bars) were generated using the Stata margins command.
HAS, high-added-sugar.
Although not a statistically significant difference, the amount of added sugar ordered, averaged across menus, was lower in the warning condition by 1.5 grams (95% CI= –3.0, 0.1; p=0.07), a relative difference of 2.1%. Although also not significant, label effect sizes for grams of added sugar ordered were similar by restaurant menu (fast food= –1.4 grams vs –1.5 grams for full service) and larger for beverages than for mains or desserts when averaged across menus (–0.7 vs –0.3 and –0.4 grams) (Appendix Table 4, available online).
In the sensitivity analysis excluding 47 outliers for grams of added sugar ordered (Appendix Table 5, available online), results were statistically significant: the warning group ordered 1.6 fewer average grams of added sugar than the controls (95% CI= –3.1, –0.03; p=0.046). Second, in adjusted analyses that restricted the warning group to participants who reported noticing added-sugar warnings (Appendix Table 6, available online), effect sizes for ordering outcomes were stronger than in the full sample, and results for grams of added sugar ordered were statistically significant. The relative percentage of participants who ordered ≥1 high-added-sugar item was lower by 4.1% (3.4 pp) among those who noticed the warning than for controls (PR=0.959, 95% CI=0.935, 0.984; p=0.001), and the amount of added sugar ordered was significantly lower by 4.9 grams (6.8%) among those who noticed the warning than among the controls (95% CI= –7.3, –2.5 grams; p<0.001). Results for both ordering outcomes were robust to the inclusion of speeders, those who failed the attention check, and those without complete ordering data (Appendix Table 7, available online). Finally, there were no significant warning effects when examining other dichotomous ordering thresholds for high-added-sugar items (Appendix Table 8, available online).
For the knowledge outcomes, participants’ ability to correctly identify the 5 high-added-sugar items did not differ between conditions (Figure 3C). However, warning group participants correctly classified a significantly higher percentage of the 3 not-high-added-sugar items (3.5 pp [95% CI=2.5, 4.5; p<0.001]) and all the 8 menu items by their added-sugar content (1.4 pp [95% CI=0.8, 2.1; p<0.001]) than the controls. In sensitivity analyses restricting warning group participants to those who noticed the added-sugar warnings (Appendix Table 6, available online), associations were stronger. For example, warning group participants who noticed warnings correctly classified 6.9 pp more of the 8 items by added-sugar content than the controls (95% CI=5.8, 8.0 pp; p<0.001). Warning effects on knowledge were also robust to the inclusion of speeders and those who failed the attention check or did not complete the ordering task (Appendix Table 7, available online).
Table 1 displays the perception and behavioral intention results. Warnings modestly reduced perceived healthfulness and appeal of high-added-sugar items (ps<0.001) and increased the relative probability of feeling more in control of eating decisions by 6% (PR=1.06, 95% CI=1.03, 1.10; p<0.001). The warnings did not significantly affect intentions to reduce added-sugar consumption in the next month. The warning group was also modestly more likely to agree that people who are important to them would want them to limit consumption of high-added-sugar items (i.e., injunctive norm) and that consuming such items often would increase Type 2 diabetes risk (ps<0.001).
Table 1Perception, Intention, and Process Outcomes From a Randomized Experiment of Added-Sugar Menu Warning Labels
The p-values show statistical significance, including after the Holm–Bonferroni correction.
Table 1 shows process outcomes. A total of 47% of participants in the warning condition (n=3,617) reported noticing nutrition labels other than calories, and 21% of the warning group (44% of the 47% who noticed nutrition labels, n=1,603) correctly recalled that the labels were for added sugar. Among those who noticed added-sugar warnings, 79% (n=1,265) reported perceived knowledge gain, and 55% (n=877) reported using the labels when ordering. Upon viewing the label again, 51% of the warning group (n=3,923) perceived that the label grabbed their attention and caused them to think about added sugar quite a bit or a great deal. A total of 38% (n=2,889) reported being very or extremely likely to talk about the labels with others.
Upon viewing the warning, the majority (72%) of participants supported a law requiring warning labels on chain-restaurant menus (39% strongly and 33% somewhat supported), whereas 19% had no opinion, and 9% opposed the law (4% strongly and 5% somewhat opposed). Finally, there were no significant differences between control and warning groups in dollar amounts of orders (i.e., amount of money hypothetically spent) from fast-food ($12.78 vs $12.64; p=0.16) or full-service ($25.21 vs $25.19; p=0.93) menus.
The only outcome for which there was significant moderation was knowledge: the warning was more effective in helping lower-income (≤$35,000 and >$35,000–65,000/year) than high-income (>$95K/year) participants to correctly classify high-added-sugar items (ps=0.02) (Appendix Table 9, available online).
DISCUSSION
To the authors’ knowledge, this is the first study to test the effect of restaurant menu added-sugar warning labels on hypothetical orders. This online RCT found that added-sugar warnings significantly reduced the relative probability of ordering ≥1 high-added-sugar item by 2.2% (absolute difference: 1.9 pps [control=83.4% vs warning=81.5%]). Warnings also led to a nonsignificant 1.5-gram (2%) reduction in average added sugar ordered across both menus and a statistically significant 1.6-gram reduction after excluding outliers. Given the frequency of restaurant food consumption, these relatively small effects may lead to meaningful changes in intake at the population level. Such warnings might also motivate restaurants to reduce the added-sugar content of menu items.
Testing the efficacy of and parents’ preferences for nutrition labels on children’s menus from a full-service chain restaurant: results of an online experiment.
and the directions of effects are consistent with those of other experiments testing sugar-related warnings on packages and signage.
The impact of front-of-package claims, fruit images, and health warnings on consumers’ perceptions of sugar-sweetened fruit drinks: three randomized experiments.
Online randomized controlled trials of restaurant sodium warning labels.
found that icon-only sodium warnings reduced the percentage of participants ordering ≥1 high-sodium item by 2.4 pp and reduced sodium ordered by 25 mg or 2%, although effects were not statistically significant, potentially owing to sample size. However, the icon-plus-text sodium warnings examined in that study resulted in larger and statistically significant results compared to the icon-only label: 4.1–6.4 pp reduction in ordering ≥1 high-sodium item and 46–68 fewer mg (3%–5%) of sodium ordered.
Online randomized controlled trials of restaurant sodium warning labels.
Thus, it is possible that icon-plus-text added-sugar warnings could result in larger effects than observed in this study for icon-only added-sugar warnings.
Results also found that added-sugar warnings increased participants’ knowledge of whether menu items contained >50% of the added-sugar daily recommended limit and modestly reduced perceptions of healthfulness and appeal for high-added-sugar items. They also increased perceptions that eating such items frequently would increase Type 2 diabetes risk, and they shifted norms about consuming high-added-sugar items. Support was high (72%) for a law that would require chain restaurants to display added-sugar warnings on menus. Finally, there were no warning effects on (hypothetical) dollar amounts of orders.
The observed effects on primary outcomes may be modest because only 47% of the warning group reported noticing a nutrition label (similar to a previous finding for sodium icons),
Online randomized controlled trials of restaurant sodium warning labels.
and only 21% of the warning group reported noticing an added-sugar warning. When analyses were restricted to those who noticed the added-sugar warning, the adjusted effect sizes were significant and larger than in the full sample. The relative probability of ordering ≥1 high-added-sugar item was lower by 4.1%; average added sugar ordered was lower by 4.9 grams; and the percentage of items correctly classified by added-sugar intake was higher by 6.9 pp. The icon-only label that was tested was black, the same size as the menu-item text, and not the more commonly used warning triangle shape. Larger warning size, bright and contrasting colors,
Online randomized controlled trials of restaurant sodium warning labels.
could increase noticeability and efficacy.
The icon-only warning in this study had a much smaller effect on knowledge than that of a previous study that examined 6 icon-only added-sugar warnings, including the icon in this study. This may be because the previous study
UNC perceived message effectiveness: validation of a brief scale.
which drew attention to the labels, or because the knowledge tasks differed. In the previous study, participants were presented with 8 items on a single menu and asked to click the items that had more than half the daily added-sugar limit. In this study, only 2 items were presented at a time, and there may have been more items that participants typically associate with high-added-sugar content.
Future research should explore whether the effectiveness of added-sugar warnings can be increased by including bright colors, making the warning larger than the menu text, changing the label text (e.g., added-sugar warning versus sugar warning), and adding text next-to icons. It is also worth exploring whether the placement of labels (e.g., on the left side of item names) or grouping labeled items together or near an unlabeled counterpart to increase salience and attract attention
A salient sugar tax decreases sugary-drink buying.
could boost label effectiveness. It would also be valuable to investigate the effects of different added-sugar warning thresholds; alternative labeling approaches for combo meals; and the potential interactive effects of added-sugar warnings, sodium warnings, and calorie labels. Real-world policy evaluations should also examine whether added-sugar warnings spur reformulation. Furthermore, requiring chain restaurants to disclose added-sugar content would facilitate the implementation and enforcement of added-sugar labeling policies. Study strengths include recruiting a large national sample, basing the label design on results of a previous experiment testing multiple warning designs,
Perceived effectiveness of added-sugar warning label designs for U.S. restaurant menus: an online randomized controlled trial.
and examining warnings on fast-food and full-service menus.
Limitations
Limitations include measuring hypothetical, not actual, behavior (although incentives were offered to try to increase realistic decision making); the possibility for social desirability bias (which was likely small given participant anonymity); and the potential that although the sample was recruited to match the U.S. distribution of sex, age, race, Hispanic ethnicity, and education, there may have been differences in other characteristics that limit generalizability. However, online convenience samples do typically produce internally valid experimental results.
Evaluating online labor markets for experimental research: amazon.com’s Mechanical Turk.
In addition, because the Food and Drug Administration does not require restaurants to disclose added sugar, the added-sugar content of many items had to be estimated. Finally, the study examined only 1-time label exposure. Research on repeated exposures, especially in real-world settings, is needed to understand long-term impacts.
CRediT authorship contribution statement
Jennifer Falbe: Conceptualization, Data curation, Funding acquisition, Methodology, Supervision, Formal analysis, Writing – original draft, Writing – review & editing. Aviva A. Musicus: Conceptualization, Methodology, Visualization, Writing – review & editing. Desiree M. Sigala: Methodology, Writing – review & editing. Christina A. Roberto: Conceptualization, Methodology, Writing – review & editing. Sarah E. Solar: Methodology, Data curation, Writing – review & editing. Brittany Lemmon: Data curation, Visualization, Formal analysis, Writing – review & editing. Sarah Sorscher: Conceptualization, Writing – review & editing. DeAnna Nara: Conceptualization, Writing – review & editing. Marissa G. Hall: Conceptualization, Methodology, Writing – review & editing.
ACKNOWLEDGMENTS
The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views or policy of Bloomberg Philanthropies, the Center for Science in the Public Interest (CSPI), the NIH, or the U.S. Department of Agriculture. Bloomberg Philanthropies played no role in the study design, label design, analytic plan, data collection, data curation or analysis, or the drafting or revising of the manuscript. CSPI, a nonprofit organization, played no role in the data collection, experimental design, or analysis of the data.
This study was funded by Bloomberg Philanthropies (2019-71208) directly and through a subaward from CSPI. JF is supported by NIH/National Institute of Diabetes and Digestive and Kidney Diseases (K01DK113068) and the U.S. Department of Agriculture/National Institute of Food and Agriculture Hatch project 1016627. DMS is supported by the NIH/National Heart, Lung, and Blood Institute Postdoctoral Diversity Supplement R01HL137716. MGH is supported by NIH/National Heart, Lung, and Blood Institute K01HL147713. AAM is supported by grants T32HL098048 and T32CA057711 from the NIH. UC Davis IRB approval number: 1641776. SS and DN are employed by CSPI and contributed to the study conceptualization, label design, and reviewing and editing of the manuscript.
Contents from this article were previously presented at the 2022 American Public Health Association Annual Meeting.
No other financial disclosures were reported by the authors of this paper.
