A. Experimental Design and Data Collection
An experimental study over the Internet was designed to gather data on respondent reactions to products bearing health claims and disclaimers. Stimuli included product labels for various conventional foods and dietary supplements, with each product associated with at least one health claim.
Each respondent was randomly assigned to one of 124 experimental conditions. Each condition consisted of a front label and a back label (containing the Nutrition Facts or Dietary Supplement Facts). The front label varied by condition, the back label varied by product. Some front labels displayed health claims and others displayed no claims (control labels). Non-control labels displayed one variant of a health claim, such as a qualified version of the health claim, or an unqualified version of the claim. A sample size of 7,440 respondents was selected to ensure a minimum of 60 respondents for each of the 124 possible conditions.
The sample, all adults of 18 years of age or older, was selected from an Internet panel and was diverse in age, education, gender, and race. Harris Interactive Service Bureau (HISB), through the Harris Poll Online (HPOL) cooperative respondent database, provided the sample. The HPOL is a voluntary respondent sample, stratified by age, gender and region. Data collection occurred from March to May 2007.
After responding to questions about various health attitudes, dietary supplement usage and prior knowledge of diet-disease relationships, each respondent reviewed the front label of a single product and responded to questions about the product claims. Respondents could elect to view and use the additional information on the back label such as the Nutrition Facts, if desired, by clicking on a link. The questionnaire used closed-ended questions about the respondent’s rating of scientific evidence about the health claims, and perceptions about the product’s likelihood of reducing the risk of diseases, importance for a healthy diet, quality, and safety.
B. Claims and Disclaimers
A full list of the claims, products and disclaimers used in the study are listed in Table 1. The study tests disclaimer schemes using a mixture of currently approved SSA health claims and allowed QHCs. SSA claims in this study discuss the relationships between soluble fiber and heart disease and stanol and sterol esters from vegetable oil and heart disease. The claims meet the highest level of scientific evidence and are referred to in this study as A-level claims. To facilitate discussion of the separate claims they are labeled “A(1)” and “A(2),” but the level of evidence should be considered equal for both claims. The QHCs in the study discuss the relationships: (1) between EPA/DHA omega-3 fatty acids and heart disease, labeled “B” and referred to as a B-level claim with supportive evidence; (2) between monounsaturated fats from olive oil and heart disease, labeled “C” and referred to as a C-level claim with limited evidence; (3) between green tea and breast cancer; and (4) between green tea and prostate cancer, labeled “D(1)” and “D(2)” and referred to as D-level claims with weak evidence. Similar to the A(1) and A(2) claims, D(1) and D(2) claims should be considered equivalent with respect to the level of scientific evidence supporting the claims. Each claim is paired with one food product and one dietary supplement.
The study assesses the impacts of three disclaimer schemes for conveying levels of scientific evidence for the claims. One scheme is the current scheme employed by FDA, called “Scheme 1” for the purposes of this report. In Scheme 1, health claims meeting the SSA standard are authorized by FDA and may be displayed on products without the use of qualifying language (i.e., they are unqualified claims). Claims not meeting the SSA standard may be allowed by FDA but must feature language that discusses the level of scientific evidence supporting the claim (i.e. they are qualified claims). FDA crafts the language of these claims before allowing their use by manufacturers.
The two other schemes assessed in the study employ SSA claims modified so that they conform to the structure of qualified claims within each scheme. In Scheme 2, SSA claims include a disclaimer about the level of scientific evidence. These SSA claims are prefaced by “very strong scientific evidence suggests” to make them more congruent with qualified health claims which have prefacing language about scientific evidence. In Scheme 3, all claims are presented without any qualified language but accompanied by report card grades. SSA claims display a grade “A” for strong evidence and qualified claims display grades “B”, “C”, or “D” for progressively weaker levels of evidence.
In addition to the three schemes, the study also tests the effect of an additional disclaimer to Scheme 3 (report card scheme) that states that “this rating is not a grade for the product’s quality, safety, or overall healthfulness” as a means of mitigating some of the misperceptions created by report card schemes observed in the earlier study and Kapsak et al. (2008).3
Click on the following link to see Table 1. Products and Health Claims Used in the Study.
C. Evaluation of Schemes to Convey the Level of Scientific Evidence for Health Claims
Derby and Levy (2005) put forth three performance standards against which disclaimer schemes should be evaluated. Each of the three schemes in this study is evaluated against these standards to determine their effectiveness for conveying the level of scientific evidence for health claims. The three standards are:
- Disclaimer Effect: Perceived level of scientific evidence for an unqualified health claim should be decreased by a disclaimer.
- Distinct Levels of Scientific Evidence: Perceived level of scientific evidence should decrease as disclaimers state that the evidence is weaker.
- Perceived Health Benefits and Other Attributes: Inferences about a product’s health benefits and other attributes should be consistent with the effects of disclaimers on perceived level of scientific evidence.
Generalized linear models are used to analyze the relationships between each dependent measure (e.g., rating of scientific evidence) and the characteristics of the experimental condition (e.g., product and claim type) and the demographic traits of the respondent (e.g., income and gender). Planned comparisons between means of a given dependent measure are used to determine the magnitude and statistical significance of the effect of claims and disclaimers for the dependent measure. For all tests, statistically significant results achieve a 5% critical value (p<0.05), and marginally significant results achieve a 10% critical value (0.05≤p<0.10).4 The following section presents graphs of the means for the various dependent measures, tables of the results of planned comparisons, and a discussion of the implications of those results for the effectiveness of each scheme against the performance standards.
3 Kapsak WR, Schmidt D, Childs NM, Meunier J, White C. 2008. “Consumer perceptions of graded, graphic and text label presentations for qualified health claims.” Critical Reviews in Food Science and Nutrition. 48(3):248-256.
4 Throughout the report, “p” refers to the level of statistical significance of a test result.