Abstract
Objectives. To examine US adults’ reactions to health warnings with strong versus weak causal language.
Methods. In 2018, we randomly assigned 1360 US adults to answer an online survey about health warnings for cigarettes, sugar-sweetened beverages, or alcohol. Participants rated 4 warning statements using different causal language variants (“causes,” “contributes to,” “can contribute to,” and “may contribute to”) displayed in random arrangement.
Results. Most participants (76.3%) selected the warning that used “causes” as the 1 that most discouraged them from wanting to use the product. “Causes” was also selected most often (39.0% of participants) as the warning that participants most supported implementing. By contrast, most (66.1%) chose “may contribute to” as the warning that least discouraged them from wanting to use the product. We found few demographic differences in these patterns.
Conclusions. Warnings with stronger causal language are perceived to be effective and are supported by the public.
Requiring health warnings on unhealthy products—including cigarettes, sugar-sweetened beverages (SSBs), and alcohol—is a promising health policy. The evidence base supporting cigarette pack warnings is strong, showing that pictorial warnings decrease smoking.1–3 Several experiments have shown that health warnings for SSBs and alcohol may change precursors to behavior change such as risk perceptions,4,5 negative affective responses,6 and intentions to purchase these products.4,7–10
As product health warnings become an increasingly popular policy option,11 more research is needed about how to maximize their effectiveness. One understudied area is whether consumers respond differently to warnings with stronger or weaker causal language. Product warnings in the United States include a wide variety of causal language variants. For example, 4 of 9 mandated, but not yet implemented, cigarette pack warning statements use strong causal language (i.e., “causes”) when describing the link between smoking and health effects (e.g., “WARNING: Cigarettes cause cancer”).12 Although the United States does not yet require warnings on SSB containers, San Francisco, California, passed a 2015 ordinance requiring SSB advertisements to display a warning using weaker causal wording than cigarette warnings (i.e., “WARNING: Drinking beverages with added sugar(s) contributes to obesity, diabetes, and tooth decay”).13,14 Mandated alcohol warnings in the United States also use weaker causal wording than cigarette warnings, stating, “Alcoholic beverages . . . may cause health problems.”
Product warnings in the United States are more likely to withstand court challenges if they are factual and uncontroversial. Moreover, all warnings must advance a government interest such as improving public health. With the goal of establishing an evidence base that product warnings improve public health, we explored whether causal language variants used in product health warnings elicit differential reactions in consumers. Specifically, we examined whether strength of causal language used in health warnings affects perceived message effectiveness and public support, outcomes that have been shown to predict warnings’ actual behavioral impact.15,16
METHODS
In April 2018, we recruited a convenience sample of 1413 adults to participate in an online experiment. Inclusion criteria were currently residing in the United States and being aged at least 18 years. Recruitment occurred through the online platform Amazon Mechanical Turk (MTurk). Experiments conducted on MTurk largely replicate findings from studies conducted via probability-based samples.17
Procedures
Participants provided informed consent and took a 10- to 15-minute online survey. After completing 2 other experimental tasks (1 about text-based warnings for SSBs6 and 1 about graphic warnings for cigarettes, SSBs, and alcohol), we randomly assigned participants, using simple randomization (i.e., no stratification), to view 4 health warnings for either cigarettes, SSBs, or alcohol. Warning statements for each product used 4 causal language variants presented simultaneously in a random order: “causes,” “contributes to,” “can contribute to,” and “may contribute to.” The cigarette warnings read: “WARNING: Smoking cigarettes [causal language variant] lung cancer.” The SSB warnings read: “WARNING: Drinking beverages with added sugar [causal language variant] tooth decay.” And the alcohol warnings read: “WARNING: Drinking alcohol [causal language variant] liver disease.” We selected these health effects because they are commonly known consequences of consuming each product,18–20 ensuring warnings would be similarly believable across products. We also verified that the warning statements were scientifically accurate.21–23 Participants received $2.20 for completing the survey. Before data collection, we preregistered the study on AsPredicted.org (https://aspredicted.org/7dz9i.pdf).
Measures
Participants responded to 3 questions about warnings for their randomly assigned product. To assess perceived message effectiveness, we asked, “Which of these warnings would most discourage you from wanting to [use product]?” To assess perceived message ineffectiveness, we asked, “Which of these warnings would least discourage you from wanting to [use product]?” Finally, to assess public support, we asked, “Which of these warnings would you most support being on [product]?” We randomized the order of these 3 questions. For each question, the response options were warnings with each of the 4 causal language variants. Participants also provided information on their demographic characteristics and health behaviors.
Data Analysis
To avoid repeat respondents, we identified duplicate IP addresses and retained only the record with the least amount of missing data. When the amount of missing data was equivalent, we kept only the first record. This resulted in dropping 32 cases from the data set. We repeated this process for duplicate MTurk usernames, dropping 8 cases. Finally, we excluded 13 participants who had recently completed a pilot test of our survey instrument, yielding an analytic sample of n = 1360.
We performed our analyses using Stata/SE version 14.1 (StataCorp LP, College Station, TX) with 2-tailed tests, a critical α of 0.05, and listwise deletion for missing data. We first ran descriptive statistics for each outcome separately for each product type. Then, we conducted pairwise comparisons using the unpaired z test to compare proportions of participants selecting each causal language variant for each outcome, collapsed across product type. Next, we examined predictors of which causal language variant participants selected for each outcome. Although our preregistration specified ordinal or multinomial models, for simplicity we dichotomized outcomes into selection of the most common choice for that outcome (vs selection of any other variant) and used logistic regression. For all 3 outcomes, dichotomizing into most common choice compared with all other choices maintained the natural ordering of the categories. We conducted sensitivity analyses controlling for participants’ randomly assigned condition in the 2 experimental tasks that appeared earlier in the survey. These analyses revealed a nearly identical pattern of findings, indicating that the experimental tasks did not influence our results.
In prespecified exploratory moderation analyses, we examined whether the findings varied by product type. We fully interacted the 3 models with the “product type” variable and examined whether any of the interaction coefficients were statistically significant. For significant moderators, we then visually plotted the predicted probabilities for each level of the moderators. In these analyses, we used a Bonferroni correction24 because of the large number of exploratory hypotheses being tested. We divided the critical α of 0.05 by the number of coefficients estimated in the interacted model, yielding a corrected α of 0.002.
RESULTS
Participants’ mean age was 37.4 years, and 10.4% identified as gay, lesbian, or bisexual (Table 1). The sample was mostly non-Hispanic (90.0%) and White (81.8%). Most participants held college or graduate degrees (64.3%), and about half (51.2%) had an annual household income of at least $50 000. In terms of health behaviors, 21.9% were current smokers (defined as having smoked at least 100 cigarettes in their lifetime and now smoking some days or every day), about a third (36.0%) consumed SSBs at least once a day, and most (62.3%) consumed alcohol at least once a month. The sample was younger; more likely to identify as gay, lesbian, or bisexual; less likely to be Hispanic; and more likely to smoke than were nationally representative samples (Table A, available as a supplement to the online version of this article at http://www.ajph.org).
TABLE 1—
Participant Characteristics: United States, 2018
| Characteristic | No. (%) or Mean ±SD |
| Age, y | |
| 18–29 | 361 (26.7) |
| 30–39 | 547 (40.5) |
| 40–54 | 295 (21.8) |
| ≥ 55 | 149 (11.0) |
| Mean ±SD | 37.4 ±11.6 |
| Gender | |
| Male | 704 (52.1) |
| Female | 639 (47.3) |
| Transgender or other | 9 (0.7) |
| Gay, lesbian, or bisexual | 141 (10.4) |
| Hispanic ethnicity | 122 (9.0) |
| Race | |
| White | 1106 (81.8) |
| Black/African American | 127 (9.4) |
| Asian | 63 (4.7) |
| Other/multiracial | 47 (3.5) |
| American Indian/Alaska Native | 8 (0.6) |
| Native Hawaiian/Pacific Islander | 1 (0.1) |
| Education | |
| High school diploma or less | 170 (12.6) |
| Some college | 313 (23.2) |
| College graduate or associate’s degree | 699 (51.7) |
| Graduate degree | 170 (12.6) |
| Annual household income, $ | |
| 0–24 999 | 234 (17.3) |
| 25 000–49 999 | 425 (31.5) |
| 50 000–74 999 | 322 (23.8) |
| ≥ 75 000 | 370 (27.4) |
| Low income (≤ 150% of FPL) | 224 (16.6) |
| Current smoker | 298 (21.9) |
| Frequency of sugar-sweetened beverage consumption | |
| < 1 time per d | 866 (64.0) |
| 1 to < 3 times per d | 312 (23.1) |
| ≥ 3 times per d | 175 (12.9) |
| Frequency of alcohol consumption | |
| < 1 time per mo | 510 (37.7) |
| 1–3 times per mo | 265 (19.6) |
| 1–2 times per wk | 310 (22.9) |
| 3–7 times per wk | 268 (19.8) |
Note. FPL = federal poverty level. n = 1360. Missing demographic data range from 0.6% to 0.9%.
When asked which warning would most discourage them from consuming the product, most (76.3%) participants selected the warning that used “causes,” followed by “contributes to” (13.9%), “can contribute to” (5.3%), and “may contribute to” (4.5%; Figure 1). When asked which warning would least discourage them from consuming the product, most (66.1%) participants selected the warning that used “may contribute to.” Finally, “causes” was selected most often (39.0% of participants) as the warning that participants would most support. We found few differences in these patterns by demographic characteristics (Tables B, C, and D, available as supplements to the online version of this article at http://www.ajph.org). Descriptive data for each product type appear in Table 2.
FIGURE 1—
Percentage of Participants Selecting Each Causal Variant for (a) Most Discouraged Use of the Product, (b) Least Discouraged Use of the Product, and (c) Most Supported Being on Product: United States, 2018
Note. NS = not significant (P > .05). Error bars are SEs. The sample size was n = 1353.
*P < .05; **P < .001.
TABLE 2—
Warning Selection by Product Type: United States, 2018
| Most Discouraged Use |
Least Discouraged Use |
Most Supported Being on Product |
|||||||
| Warning | Cigarettes (n = 451), No. (%) | SSBs (n = 448), No. (%) | Alcohol (n = 454), No. (%) | Cigarettes (n = 451), No. (%) | SSBs (n = 448), No. (%) | Alcohol (n = 454), No. (%) | Cigarettes (n = 451), No. (%) | SSBs (n = 448), No. (%) | Alcohol (n = 454), No. (%) |
| Causes | 367 (81.4) | 335 (74.8) | 330 (72.7) | 59 (13.1) | 45 (10.0) | 71 (15.6) | 252 (55.9) | 144 (32.1) | 131 (28.9) |
| Contributes to | 45 (10.0) | 69 (15.4) | 74 (16.3) | 16 (3.6) | 32 (7.1) | 25 (5.5) | 108 (24.0) | 162 (36.2) | 125 (27.5) |
| Can contribute to | 22 (4.5) | 23 (5.1) | 27 (6.0) | 73 (16.2) | 68 (15.2) | 70 (15.4) | 52 (11.5) | 57 (12.7) | 95 (20.9) |
| May contribute to | 17 (3.8) | 21 (4.7) | 23 (5.1) | 303 (67.2) | 303 (67.6) | 288 (63.4) | 39 (8.7) | 85 (19.0) | 103 (22.7) |
Note. SSB = sugar-sweetened beverage.
Product type influenced participants’ likelihood of selecting “causes” as the most discouraging and most supported variant. Participants randomized to rate SSB warnings (odds ratio [OR] = 0.63; 95% confidence interval [CI] = 0.46, 0.88) and alcohol warnings (OR = 0.58; 95% CI = 0.42, 0.80) were less likely to select “causes” as the most discouraging warning than were those randomized to view cigarette warnings. Similarly, participants randomized to rate SSB warnings (OR = 0.37; 95% CI = 0.28, 0.49) or alcohol warnings (OR = 0.32; 95% CI = 0.24, 0.42) were less likely to select “causes” as the warning they most supported than were those who saw cigarette warnings. This relationship was even more pronounced among heavy alcohol consumers: among those who drank at least once per week, rating alcohol instead of cigarette warnings reduced the predicted probability of selecting “causes” as the warning with highest support from 55% to 29% (OR = 0.39), compared with a reduction from 43% to 42% (OR = 0.07) among lighter drinkers (< 1 time/week; P interaction < .001; Figure A, available as a supplement to the online version of this article at http://www.ajph.org). None of the other preregistered interactions between demographic or behavioral predictors and product type were statistically significant.
DISCUSSION
In our study of US adults, about 3 in 4 participants perceived a health warning with strong causal language (“causes”) as being the most effective. About two thirds perceived a health warning with weak causal language (“may contribute to”) as being the least effective. When asked to choose which warning they most supported appearing on products, most participants selected warnings with strong or relatively strong causal language (“causes” or “contributes to”). Cigarette warnings that used stronger causal language were viewed more positively than were SSB and alcohol warnings with equivalent wording, perhaps because public support for tobacco control policies tends to be higher than for other health policies.16 We observed few demographic differences in our results, suggesting that warnings with strong causal language are equally compelling to and supported by consumers with diverse characteristics.
In the United States, the decision to use stronger causal language in health warnings can have important legal implications. The First Amendment protects corporations against unconstitutional government-compelled speech, meaning that health warnings are often challenged in court. In 2017, a 3-judge panel of the US Court of Appeals for the Ninth Circuit enjoined enforcement of San Francisco’s proposed SSB warnings, taking issue with the phrase “contributes to,” instead proposing weaker wording (“may contribute to”).25 A 2019 en banc decision of the full Ninth Circuit upheld the injunction for a different reason but did not specifically take issue with the phrasing “contributes to.”26
The ultimate fate of the San Francisco SSB warnings has yet to be decided. Given the uncertainty about the legal viability of different causal language variants used in warnings, researchers and policymakers should consider consulting legal experts familiar with the latest developments related to the First Amendment to discuss the best way of maximizing warnings’ impact while anticipating possible legal challenges. In any case, weak causal language such as “may contribute to” may not be as effective as stronger language such as “causes” or “contributes to.” Thus, in countries where governments have more leeway to implement effective warnings without the threat of industry litigation, we recommend using the strongest causal language that is supported by the evidence.
Strengths and Limitations
Strengths of this study include the evaluation of warnings for 3 products and the large sample from across the United States. Limitations include the use of a convenience sample, meaning that our data cannot be used to infer population-level estimates. Our measurement approach did not allow us to estimate the magnitude of differences in perceived effectiveness and public support across causal language variants. We did not assess behavioral outcomes, although perceived message effectiveness has been shown to predict behavior change.15 Because of survey space constraints, we did not assess other constructs, such as believability of the warning statements. Future studies should replicate these findings with nationally representative samples and with behavioral outcomes.
Conclusions
This study found that US adults rated warnings that demonstrated a clear causal link between products and health consequences as being more effective than warnings that used weaker causal language. Participants were also more likely to support warnings with stronger causal language across all product types (i.e., cigarettes, sugar-sweetened beverages, and alcohol). When justified by the scientific evidence, warnings that use strong causal language might ultimately have a larger public health impact than do those with weaker wording.
ACKNOWLEDGMENTS
Research reported in this publication was supported by the National Cancer Institute, National Institutes of Health (NIH; grant T32-CA057726). The National Heart, Lung, and Blood Institute, NIH, supported M. G. Hall’s time writing the article (grant K01HL147713). The National Cancer Institute and the Food and Drug Administration (FDA), Center for Tobacco Products supported M. R. Kameny’s time writing the article (grant P50CA180907). General support and training support were provided by the Carolina Population Center (grants P2C HD050924, T32 HD007168). O. M. Maynard was supported by the Economic and Social Research Council (grant ES/R003424/1).
We thank Cathy Zimmer for statistical consultation.
Note. The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH or the FDA.
CONFLICTS OF INTEREST
The authors have no conflicts of interest to disclose.
HUMAN PARTICIPANT PROTECTION
This research was approved by the University of North Carolina, Chapel Hill institutional review board. Written informed consent was obtained.
Footnotes
See also Chiolero, p. 1319.
REFERENCES
- 1.Brewer NT, Hall MG, Noar SM et al. Effect of pictorial cigarette pack warnings on changes in smoking behavior: a randomized clinical trial. JAMA Intern Med. 2016;176(7):905–912. doi: 10.1001/jamainternmed.2016.2621. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Noar SM, Francis DB, Bridges C, Sontag JM, Ribisl KM, Brewer NT. The impact of strengthening cigarette pack warnings: systematic review of longitudinal observational studies. Soc Sci Med. 2016;164:118–129. doi: 10.1016/j.socscimed.2016.06.011. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Levy DT, Mays D, Yuan Z, Hammond D, Thrasher JF. Public health benefits from pictorial health warnings on US cigarette packs: a SimSmoke simulation. Tob Control. 2017;26(6):649–655. doi: 10.1136/tobaccocontrol-2016-053087. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Wigg S, Stafford LD. Health warnings on alcoholic beverages: perceptions of the health risks and intentions towards alcohol consumption. PLoS One. 2016;11(4):e0153027. doi: 10.1371/journal.pone.0153027. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Al-hamdani M, Smith S. Alcohol warning label perceptions: emerging evidence for alcohol policy. Can J Public Health. 2015;106(6):e395–e400. doi: 10.17269/CJPH.106.5116. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Grummon AH, Hall MG, Taillie LS, Brewer NT. How should sugar-sweetened beverage health warnings be designed? A randomized experiment. Prev Med. 2019;121:158–166. doi: 10.1016/j.ypmed.2019.02.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.VanEpps EM, Roberto CA. The influence of sugar-sweetened beverage warnings: a randomized trial of adolescents’ choices and beliefs. Am J Prev Med. 2016;51(5):664–672. doi: 10.1016/j.amepre.2016.07.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Bollard T, Maubach N, Walker N, Ni Mhurchu C. Effects of plain packaging, warning labels, and taxes on young people’s predicted sugar-sweetened beverage preferences: an experimental study. Int J Behav Nutr Phys Act. 2016;13(1):95. doi: 10.1186/s12966-016-0421-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Roberto CA, Wong D, Musicus A, Hammond D. The influence of sugar-sweetened beverage health warning labels on parents’ choices. Pediatrics. 2016;137(2) doi: 10.1542/peds.2015-3185. [DOI] [PubMed] [Google Scholar]
- 10.Grummon AG, Taillie LS, Golden SD, Hall MG, Ranney LM, Brewer NT. Sugar-sweetened beverage health warnings and purchases: a randomized controlled trial. Am J Prev Med. In press. [DOI] [PMC free article] [PubMed]
- 11.Falbe J, Madsen K. Growing momentum for sugar-sweetened beverage campaigns and policies: costs and considerations. Am J Public Health. 2017;107(6):835–838. doi: 10.2105/AJPH.2017.303805. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12. Family Smoking Prevention and Tobacco Control Act of 2009. Pub L No. 111-31, HR 1256. 111th Congress, 1st session (2009).
- 13.City and County of San Francisco. Health Code—Sugar-Sweetened Beverage Warning for Advertisements. Ordinance No. 100-15 (2015).
- 14.Pomeranz JL, Mozaffarian D, Micha R. Can the government require health warnings on sugar-sweetened beverage advertisements? JAMA. 2018;319(3):227–228. doi: 10.1001/jama.2017.19209. [DOI] [PubMed] [Google Scholar]
- 15.Baig SA, Noar SM, Gottfredson NC, Boynton MH, Ribisl KM, Brewer NT. UNC perceived message effectiveness: validation of a brief scale. Ann Behav Med. 2019;53(8):732–742. doi: 10.1093/abm/kay080. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Diepeveen S, Ling T, Suhrcke M, Roland M, Marteau TM. Public acceptability of government intervention to change health-related behaviours: a systematic review and narrative synthesis. BMC Public Health. 2013;13(1):756. doi: 10.1186/1471-2458-13-756. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Jeong M, Zhang D, Morgan JC et al. Similarities and differences in tobacco control research findings from convenience and probability samples. Ann Behav Med. 2019;53(5):476–485. doi: 10.1093/abm/kay059. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Park S, Lundeen EA, Pan L, Blanck HM. Impact of knowledge of health conditions on sugar-sweetened beverage intake varies among US adults. Am J Health Promot. 2018;32(6):1402–1408. doi: 10.1177/0890117117717381. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Cotter T, Perez D, Dunlop S, Kite J, Gaskin C. Knowledge and beliefs about alcohol consumption, longer-term health risks, and the link with cancer in a sample of Australian adults. N S W Public Health Bull. 2013;24(2):81–86. doi: 10.1071/NB12089. [DOI] [PubMed] [Google Scholar]
- 20.Weinstein N, Slovic P, Waters E, Gibson G. Public understanding of the illnesses caused by cigarette smoking. Nicotine Tob Res. 2004;6(2):349–355. doi: 10.1080/14622200410001676459. [DOI] [PubMed] [Google Scholar]
- 21.US Department of Health and Human Services. The Health Consequences of Smoking—50 Years of Progress: A Report of the Surgeon General. Atlanta, GA: Centers for Disease Control and Prevention; 2014. [Google Scholar]
- 22.Li H, Zou Y, Ding G. Dietary factors associated with dental erosion: a meta-analysis. PLoS One. 2012;7(8):e42626. doi: 10.1371/journal.pone.0042626. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.O’Shea RS, Dasarathy S, McCullough AJ. Alcoholic liver disease. Hepatology. 2010;51(1):307–328. doi: 10.1002/hep.23258. [DOI] [PubMed] [Google Scholar]
- 24.Bonferroni CE. Teoria statistica delle classi e calcolo delle probabilita [Statistical Class Theory and Probability Calculation]. Florence, Italy: Seeber; 1936.
- 25. Am Bev Ass’n v City & Cnty of San Francisco, US App LEXIS 18150 (9th Cir. 2017).
- 26. Am Bev Ass’n v City & Cnty of San Francisco, No. 16-16072 (9th Cir. 2019).