U. S. Food and Drug Administration
Center for Food Safety and Applied Nutrition
Consumer Impacts of Health Claims
January 1997

JANUARY 24, 1997 SUBJECT OF NOTICE OF AVAILABILITY, FOOD LABELING: HEALTH CLAIMS.

Consumer Impacts of Health Claims: An Experimental Study

Alan S. Levy, Brenda M. Derby and Brian E. Roe: Division of Market Studies, Center for Food and Applied Nutrition (HFS-727), Food and Drug Administration.

The purpose of this experimental study of health claims was to evaluate, in a controlled setting, some of the model health claims presented in the Food and Drug Administration (FDA)'s 1993 regulations (FDA, 1993a) as well as several alternatives suggested by the Keystone National Policy Dialogue on Food, Nutrition, and Health (Keystone, 1996) to improve the communication effectiveness of health claims. The stated purpose of the health claim regulations for food labels, mandated by the 1990 Nutrition Labeling and Education Act (NLEA), was to help consumers achieve healthier diets. Because the communication processes assumed to underlie this effect are not well understood, there has been considerable uncertainty about how health claims on food labels should be designed to optimize the intended benefits.

In the 1993 regulations, the FDA promulgated certain principles and provided examples of health claim wordings (FDA, 1993b) to guide the use of approved diet-disease health claims on food labels that would facilitate the stated NLEA objective. When the Keystone National Policy Dialogue on Food, Nutrition and Health reviewed implementation issues raised by the 1993 food labeling regulations, they noted that some of the labeling requirements for health claims had never been subjected to consumer testing and they identified the rules governing the wording and presentation of approved health claims as an area that required further study.

The current study grew out of the Dialogue's discussions of ways to maximize the effectiveness of health claims on product labels (Keystone Center, 1996). A clear theme of these discussions was that the communication processes likely to produce desired health claim effects are complex, leading to different views about how health claims on food labels are supposed to help consumers achieve healthier diets. Health claims on food labels refer simultaneously to a product and to a scientifically grounded (i.e., generic) diet-disease relationship, and as such t hey stand posed between a product-specific claim and general dietary guidance.

There are a number of things that consumers might be expected to learn from health claims on food labels--the likely health benefits of a given product, the state of current nutrition science for the relevant diet-disease relationship, an understanding of how a given product fits into a balanced diet, or general principles of good nutrition. Over the course of the Dialogue discussions, two viewpoints emerged, emphasizing different consumer impacts as theoretically and practically important.

One view of the way health claims have an effect is that they serve to create a favorable impression of the food product, leading to a greater likelihood of product purchase, and ultimately increased consumption of healthful products. The key consumer impact of health claims from this product promotion perspective is how compelling health claims are to consumers (i.e., effective at getting people to buy the product). Another view is that health claims have an important educational impact, serving to educate consumers about the healthful effects of certain kinds of food in the diet as well as the type of person most likely to benefit from eating these kinds of food, which leads to more informed food selections and more healthful consumption patterns. The key consumer impact of health claims from the educational perspective is how accurately health claims communicate to consumers the specific health benefits of the particular product and the characteristics of consumers most likely to need such benefits.

The Keystone Dialogue Group concluded that the appropriate standard for evaluating the effectiveness of health claims was that they be "compelling, but not misleading." Such a standard combines the manufacturer's interest in making the product appear more attractive, and more likely to be purchased, with the nutrition educator's interest in assuring that consumers do not attribute exaggerated benefits to the product that may lead consumers to make unwise or inappropriate dietary choices, the "magic bullet" effect. Health claims that meet the "compelling, but not misleading standard" are presumed to help consumers achieve healthier diets.

To address the concerns of the Keystone Dialogue the study utilized several types of measures. Attitudes towards the product, (e.g., purchase intentions, ratings of product healthfulness, beliefs that some people in particular would benefit from eating the product) were used as indicators of how compelling a given label condition was to respondents. Indicators of the communication impacts of a health claim and whether it might be misleading were less straightforward. In order to distinguish misleading communication from poor communication, the impacts of label conditions on perceptions of product health benefits and nutrient characteristics were assessed by a series of context-specific accuracy measures based on b oth open-ended questions about perceived product health benefits and ratings of specific health effects and nutrient characteristics. Context-specific accuracy measures distingui shed between perceived characteristics that follow from message points in the claim (i.e., direct hits) and perceived characteristics that might be true about the product, but were not mentioned on the front label, and perceived characteristics that might be considered fanciful or even incorrect. Direct hits were considered likely indicators of how well or poorly a health claim communicated, while the other types of accuracy measures were considered likely indicators of possible misleading communication effects associated with health claims.

The final report of the Dialogue suggested a number of different wordings and different presentation styles for improving the communication effectiveness of approved health claims on product packages. One of the major recommendations was that health claims on food labels should be shorter, containing only a statement of the relevant diet-disease relationship without additional qualifying information as called for in the current regulations. They also suggested greater flexibility in terms of placement of the health claim information by allowing for split messages on package labels (i.e., short messages on the front, continued in more detail on the back). Finally, they suggested that different ways of enhancing the credibility of health claims on food labels be explored such as allowing for explicit endorsements from reputable public health organizations or the use of logos certifying that a health claim has been approved by the FDA.

These suggestions were incorporated into the experimental label conditions investigated in this study. Eight different executions of product-appropriate health claims were examined for each of three different food products to evaluate the impacts of message length, placement and endorsements on several practically important aspects of consumer experience with health claims on food labels: attitudes toward the product, perceptions of the health benefits of the product, and label reading behavior.

Method

Experimental Design and Procedure

Data were gathered from 1403 primary food shoppers recruited at shopping malls at eight sites around the country (Birmingham, AL; Charlotte, NC; Dayton, OH; Houston, TX; Philadelphia, PA; Phoenix, AZ; San Jose, CA; Waterbury, CT). Respondents were screened for how much food shopping they did, age, literacy and whether they required eyeglasses to read. Only those who reported they did half or more of the household food shopping, were 18 years of age or older, who had required eyeglasses with them, and who could read a short written passage, were considered eligible for the study.

The study was described to respondents as "a study for the Food and Drug Administration about how people use food label information". Respondents were brought to a central facility for a face-to face interview. Each respondent was given a realistic mock-up of a product package, with brand identity information deleted. The package was available for inspection throughout the interview. The interviewer asked a prescribed set of questions, recorded responses, and observed and recorded the label reading behavior of respondents as they answered the questions. The experimental questionnaire, containing interviewer instructions, response codes and the question wordings, is included in Appendix 1. The same procedure was repeated for each of the three types of product (raisin bran cereal, lasagna frozen dinner, strawberry yogurt) seen be every respondent. Experimental conditions (product, order of presentation, label condition) were appropriately randomized and counterbalanced among participants to avoid confounds between main effects.

There were ten label conditions for each product consisting of eight different versions of a product-appropriate health claim as well as a condition containing only the product-appropriate nutrient content claims (Content) and a condition without a health or nutrient content claim (Control). Appendix 2 contains graphic representations of the 30 label conditions used in the study. The eight versions of health claims consisted of four different presentation styles, each with a long/short version of the health claim. The generic presentation was the current style for approved model health claims, a statement of the diet-disease relationship without attribution or endorsement. Two presentation styles involved some form of endorsement; in the NIH style the health claim was stated as a recommendation from a reputable public health organization, either the American Heart Association or the National Institutes of Health. In the FDA style the generic claim was accompanied by a logo certifying that the claim was approved by FDA. The SBP (see back panel) presentation style had a short form of the generic claim on the front and a note to see the back panel for additional information. In the SBP-Long condition there was additional health claim information on the back panel, but in the SBP-Short condition there was no additional health claim information on the back. Table 1 presents the design scheme for the label conditions.

To accomplish counterbalancing between product, order of presentation, and label condition, each respondent was randomly assigning to a row of one of four different 3 X 3 Greco-Latin squares (Product X Label Condition). Each respondent saw either the Content or Control label condition and two of the eight remaining label conditions with the restrictions that the two health claim conditions seen by a respondent were not both long or both short or both with endorsements. Appendix 3 presents the scheme for the Greco-Latin squares.

Table 1. Design Scheme for Label Conditions

Label Condition	Content Claim	Health Claim	Length of Claim	Endorsed Claim	Split Message
Control	No	No
Content	No	Yes
Generic-Short	Yes	Yes	Short	No	No
Generic-Long	Yes	Yes	Long	No	No
FDA-Short	Yes	Yes	Short	Yes	No
FDA-Long	Yes	Yes	Long	Yes	No
NIH-Short	Yes	Yes	Short	Yes	No
NIH-Long	Yes	Yes	Long	Yes	No
SBP-Short	Yes	Yes		No	Yes
SBP-Long	Yes	Yes		No	Yes

Each type of product was associated with a product-appropriate health claim that was consistent with the nutrient profile of the product and acceptable under the current regulations. The raisin bran cereal carried a folic acid/neural tube defects message. The strawberry yogurt carried a calcium/osteoporosis message, and the frozen lasagna entree carried a fat/heart disease message. Appendix 4 presents the wording for the short and long health claim messages for each type of product. The nutrition panel information for each product was veridical, and it did not change across label conditions.

Questionnaire Items and Variable Construction

The interview sequence was as follows: First, respondents were asked how often they purchased the type of product, then they were handed the product package, and given as much time as they desired to inspect the package before the interview continued. Then, respondents rated the information on the package (Q2a-2d), responded to open-ended questions about the perceived health benefits of the product and the type of person most likely to realize these benefits (Q3, 3a-3c), rated the magnitude of effect the product would have on seven possible health effects (Q4a-4f), rated the overall healthfulness of the product (Q5), rated the nutrient levels of the product (Q6a-6e), and finally, indicated how likely they would be buy the particular product (Q7). As respondents answered each question the interviewer was instructed to record the part of the label they looked at: front, Nutrition Facts panel, both, or neither.

After answering the set of questions, the product was returned to the interviewer and removed from sight before repeating the question series for the next product. This procedure was repeated three times as respondents saw three different product packages during the interview.

Variables based on respondents' ratings

The variables used in analyses were defined by a corresponding questionnaire item or items. Some variables were based on product-contingent scoring of respondents' answers to questions. Other variables were scored the same for all products. Variable definitions are given below:

GOODJOB = Q2a. (The package does a good job of telling me about the health benefits of this product. Agree Strongly to Disagr ee Strongly)

ENUFINFO= Q2b. (The package gives me all the nutrition information I need. Agree Strongly to Disagree Strongly)

TRUSTINFO= Q2c. (I trust the information on the nutrition label. Agree Strongly to Disagree Strongly)

NOTICE = Q2d. (Do you see any health information on the front of this package? Yes/No.)

TRUST = Q2e. (If Yes to NOTICE: Do you agree strongly, agree somewhat, disagree Somewhat or disagree strongly with th is statement? I trust the health information on the front of this package.)

EVERYONE = 1, IF Q3=1 (Do you think everyone would get the same benefits from eating this product as a regular part of their di et? Respond "Everyone")

EVERYONE = 0, Otherwise.

SOMEPEOPLE = 1, IF Q3=2(...only some people, but not others [would get a benefit|. Response "Some People")

SOMEPEOPLE = 0, Otherwise.

NODIFF = 1, IF Q3=3 (...would eating this product regularly not make that much difference to most people? Response "Not Much Difference")

NODIFF = 0, Otherwise.

CORRECT = Q4a (Lowering your chances of having a heart attack), IF PRODUCT = LASAGNA and respond "Large Effect" OR

CORRECT = Q4d (Lowering your chances of having osteoporosis), IF PRODUCT = YOGURT and respond "Large Effect" OR

CORRECT = Q4g (If a woman ate this food regularly, how much of an effect, if any would you expect it to have on her chances of having a healthy baby?), IF PRODUCT = CEREAL and respond "Large Effect."

INCORRECT = AVG(Q4b + Q4e).

(Q4b. Lowering your chances of having high blood pressure;

Q4e. Reducing how often you get sick with a cold or the flu)

MARGINAL = AVG(Q4c + Q4f).

(Q4c. Lowering your chances of getting cancer;

Q4f. Helping you get all the vitamins and minerals you need without taking pills or supplements)

HEALTHY = Q5. (Based on what you see on this package, how important would this food be as part of a healthy diet? Very Impor tant to Not Important At All)

TRUELOW = AVG(Q6a + Q6e), IF PRODUCT =CEREAL OR

(Q6a. Fat and Q6e. Cholesterol)

TRUELOW = AVG(Q6a + Q6e), IF PRODUCT = YOGURT OR

(Q6a. Fat and Q6e. Cholesterol)

TRUELOW = AVG(Q6a + Q6e), IF PRODUCT = LASAGNA.

(Q6a. Fat and Q6e. Cholesterol)

TRUEHIGH = Q6b, IF PRODUCT = YOGURT OR

(Q6b. Calcium)

TRUEHIGH = AVG(Q6b + Q6d), IF PRODUCT = LASAGNA.

(Q6b. Calcium and Q6d. Fiber)

TRUECONT = AVG(TRUEHIGH + (4-TRUELOW).

PURCHASE= Q7. (Assume for the moment that you are going to buy some [type of product| and that this product costs about the sa me as other [products of this type| on the shelf. Based on what you see on this package, how likely would you be to buy this product? Very L ikely to Very Unlikely)

Variables based on open-ended responses

For the analysis of the open-ended questions we created variables based upon the context-specific accuracy of the open-ended responses. Six binary variables were created which reflected the accuracy of the respondents' responses to the open-ended question about product health benefits (Q3a or Q3b2) and a four level variable was created to reflect the accuracy of respondents' responses to the open-ended question about who would be most likely to get health benefits from the product (Q3b1).

Three categories of accuracy (direct hits, true, and false) and two 'vocabularies' (nutritional content and dietary disease) were distinguished for responses to Q3a and Q3b2. The respondent could choose to describe the product health benefits by mentioning some dietary disease(s) associated with the product nutrient characteristics (D), or they could choose to mention some nutrient characteristic of the product (N) without explicitly mentioning a disease effect. A 'direct hit' refers to open-ended responses that repeated the health benefit(s) (DDHIT) or nutrient content characteristic(s) (NDHIT) specifically mentioned on the long versions of the product's front label (e.g., high folic acid and/or birth defects for the cereal). A 'true' response refers to health (DTRUE) or nutrient content (NTRUE) responses that were accurate for the product, but were not part of that product's health or content claim (e.g., responses of calcium or osteoporosis for lasagna). An open-ended response was categorized 'false' if it identified a disease effect (DFALSE) or nutrient characteristic (NFALSE) that would not be appropriate for the product (e.g., low sugar for the low-fat yogurt). The coding scheme for open-ended responses to Q3a and Q3b2 are shown in Table 2.

For responses to the open-ended question about what type of people would be most likely to derive a health benefit from the particular product (Q3b2), the coding distinguished between responses that mirrored the explicit content of the long health claim (RISKGRP=4), responses that were consistent with the health claim content but were stated more broadly than in the health claim (RISKGRP=3), responses that were appropriate responses but not plausibly based on the content of the health claim (RISKGRP=2), and responses that seemed inappropriate (RISKGRP=1).

Open-ended responses to Q3b2 are coded according to the scheme presented in Table 3.

Table 2. Coding Scheme for Open-ended Responses for Q3a and Q3b2.

Open-ended Response	DDHIT	DTRUE	DFALSE	NDHIT	NTRUE	NFALSE
reduce the risk of heart disease	L	Y, C
reduce the risk of osteoporosis	Y	L, C
reduce the risk of birth defects	C		L, Y
reduce the risk of neural tube birth defects	C		L, Y
a healthy part of their diet
a nutritional part of their diet
low fat				L, Y	C
low cholesterol				L	Y, C
high calcium				Y	L
high fiber				C	L
high folic acid/folate				C
vitamins/minerals					L, C	Y
low sugar						L, Y, C
low salt/sodium					Y	L, C
builds strong bones		L, Y, C
reduce risk of cancer		C	L, Y
regularity, relieve constipation		C	L, Y
low calorie						L, Y, C
contains iron
contains potassium
protein
energy/gives you energy			L, Y, C
low carbohydrates/starches						L, Y, C
has carbohydrates/high carbohydrates/starches
good for dieting/losing weight
fills you up/fills stomach/filling
helps with yeast infections			Y
contains fruit
contains active cultures/yeast
helps with digestion/good for digestive system
good for blood pressure/helps to lower high blood pressure			L, C
quick/easy to fix/eat
controls portion size/won't overindulge
would help people gain weight
intake of dairy products (i.e., milk, cheese)
It's natural/no artificial ingredients

NOTE: L = lasagna, Y = yogurt, and C = cereal.

Table 3. Coding Scheme for Open-ended Responses to Q3b2.

Open-ended Response	RISKGRP=1	RISKGRP=2	RISKGRP=3	RISKGRP=4
women/all women		L	C,Y
Caucasian/Asian women only	C,L			Y
Teen/young adult women	L			C,Y
Cauc/Asian/young adult women	C,L			Y
Women in childbearing years	L		Y	C
Men/all men	C,Y	L
Adults only		C,Y,L
Children only		C,Y,L
Older people		C,Y,L
Young people		L	C,Y
Dieters/trying to lose weight		C,Y,L
Babies	C,L	Y
Pregnant women	L		Y	C
Older women	C.L	Y
People with health problems		C,Y	L
People watching fat/low fat		C,Y		L
Health/fitness conscious		C,Y,L
Busy people/in a hurry		C,Y,L
Don't have time to cook		C,Y
People who need fiber/high fiber	Y,L	C
Not allergic to dairy/lactose	C,L	Y
Allergic/lactose intolerant	C,Y,L
No health problems		C,Y,L
People who need calcium		C,L		Y
High cholesterol/on a low chol. diet		C,Y		L
People with poor eating habits		C,Y,L
People with cardiac/heart disease		C,Y		L
People with high/active metabolism	C,Y,L
People not watching their weight	C,Y,L
People who are thin/slender/small	C,L	Y
People who don't like milk	C,L	Y

NOTE: L = lasagna, Y = yogurt, and C = cereal.

Variables based on interviewer observations

The observational variables provided a rough measure of how respondents obtained information from different parts of the product package (front label vs. nutrition facts panel) as they answered interviewer questions. The original observation variable recorded whether the respondent consulted the packages' front or nutrition facts panel, looked at both, or looked at neither for every question and question subpart. From this we calculate the number of front and nutrition panel 'looks' by the following schematic:

Table 4. Coding Scheme for Observation Measures.

Observation Code	Front Looks	Nutrition Panel Looks	Total Looks
Neither	0	0	1
Front	1	0	1
Nutrition Panel	0	1	1
Both	1	1	2

Then for any particular question or group of questions we calculated a 'percent front'(FRONT), a 'percent nutrition panel'(BACK), and a 'percent neither'(NEITHER) variable that were used in the analyses. These were calculated as the sum of front looks over the questions of interest divided by the sum of total looks over the same group of questions. These scores can then be roughly interpreted as the percent of time the respondent spent looking at the front or nutrition panel of the package.

This constructed variable deviated from an ideal measure of percent time looking at each respective panel because there was no clock time measurement of how long the individual spent looking at each panel. The scoring scheme effectively assumed that the respondent spent the same amount of time looking at both the front and nutrition panels whenever they were coded as looking at one or the other. Furthermore, we have no way to compare total amount of time spent looking at the package between question blocks or between individuals. We effectively assumed that all individuals spent the same amount of time looking at a panel each time they were marked as having looked at that panel.

ANALYSIS

All variables were analyzed using SAS GLM (SAS Institute, 1989). Independent variables were PRODUCT (3 levels: Cereal, Yogurt, Lasagna) and LABEL (10 levels: Control, Content, Short, Long, FDA1, FDA2, NIH1, NIH2, SBP1 and SBP2). Both main effects and the interaction were included in the model. A number of contrast effects are estimated to test specific hypotheses as follows:

Hypothesis 1:

Content claims communicate positive attitudes about the product more effectively and communicate health information about the product more accurately than no claim at all.

Planned contrast: (Control vs. Content).

Hypothesis 2:

Health claims communicate positive attitudes about the product more effectively than a content claim and communicate health information about the product more accurately than no claim at all.

Planned contrast: (Content vs. AVG(Short, Long, FDA1, FDA2, NIH1, NIH2, SBP1, SBP2)) or (Control vs. AVG(Short, Long, FDA1, FDA2, NIH1, NIH2, SBP1, SBP2)).

Hypothesis 3:

Short health claims communicate positive attitudes about the product more effectively and communicate health information about the product more accurately than long health claims.

Planned contrast: (AVG(Short, FDA1, NIH1) vs. AVG(Long, FDA2, NIH2)).

Hypothesis 4:

Endorsed health claims communicate positive attitudes about the product more effectively and communicate health information about the product more accurately than non-endorsed claims.

Planned contrast: (AVG(Short, Long) vs. AVG(FDA1, FDA2, NIH1, NIH2)).

Hypothesis 5:

Split message health claims communicate positive attitudes about the product more effectively and communicate health information about the product more accurately than non-split claims.

Planned contrast: (SBP2 vs. Long))

The set of hypotheses were tested for all products combined and for each product individually.

The analyses were organized to examine the "compelling, but not misleading" standard set by the Keystone Dialogue report. The "compelling" variables constructed from measures of positive attitudes about the product, are PURCHASE, HEALTHY and NODIFF. The "communication accuracy" variables constructed from open-ended measures are DDHIT, DTRUE, DFALSE, NDHIT, NTRUE, NFALSE, and RISKGRP, and the communication accuracy variables constructed from closed-ended measures are CORRECT, INCORRECT, MARGINAL , and TRUECONT. The variables of label reading behavior constructed from interviewer observations are FRONT, BACK and NEITHER. The variables constructed from respondent ratings of the information on the package are GOODJOB, ENUFINFO, TRUSTINFO, NOTICE AND TRUST.

RESULTS

Measures of Compellingness

Table 5 presents the results for indicators of compellingness by label condition across all three product categories. Tables 5a, 5b, and 5c, present the results for the cereal, lasagna and yogurt products, respectively. Table 6 presents the results of the planned contrasts testing the five hypotheses.

The hypothesis that content claims are more compelling than no claims at all was partially supported. Content claims were significantly more compelling than no claims at all for the lasagna product, but this effect was not found for either the cereal or yogurt products (Table 6: Hypothesis 1; PURCHASE, HEALTHY, NODIFF). The hypothesis that health claims on product labels are more compelling than content claims alone, was not supported by these results. A folic acid/birth defect claim on the cereal product had a significant positive impact on attitudes toward the product compared to the high fiber/high folic acid content claim by itself, but the fat/heart disease claim on the lasagna product had a significant negative effect on respondents' attitudes toward the product compared to a low fat claim by itself. For the yogurt product, neither health nor content claims had a significant effect on product attitudes compared to the control condition (Table 6: Hypothesis 2; PURCHASE, HEALTHY, NODIFF). Short claims were not generally more compelling than long claims, but in the case of cereal, where health claims had the expected positive effect on respondents' attitudes toward the product, the shorter claim did have a significant positive effect compared to the long claim on one of the three measures of compellingness (Table 6: Hypothesis 3; PURCHASE).

The hypothesis that health claims on product labels which are endorsed or certified as approved would be more compelling than health claims without endorsements, was not supported. In fact, the opposite seemed to be true. Endorsements on the product label seemed to have a significantly negative impact on respondents' attitudes toward the product, particularly in the case of the lasagna product (Table 6: Hypothesis 4; PURCHASE, HEALTHY, NODIFF). The hypothesis that a split message health claim would be more compelling than the same health claim on the front of the package was not supported (Table 6: Hypothesis 5; PURCHASE, HEALTHY, NODIFF). There was some indication, however, that the split message presentation style worked better in the case of the yogurt product than for the other types of products, perhaps because the yogurt packaging was smaller and more densely printed than the other products.

Table 5. Indicators of Compellingness by Label Condition.

Label Condition	Purchase(Q.7)	Healthy (Q.5)	Nodiff (Q.3)
	Mean	Mean	Percent
Control	2.90	2.93	28.0
Content	3.04	3.06	22.7.
All Health Claims	3.04	3.04	22.6
Short	3.21	3.13	21.9
Long	3.10	3.10	22.9
FDA-Short	3.13	3.08	23.9
FDA-Long	2.94	3.00	22.7
NIH-Short	2.91	2.96	22.9
NIH-Long	2.95	3.02	21.1
SBP-Short	2.97	2.95	23.4
SBP-Long	3.12	3.08	21.6
	1-4, 4=very likely	1-4,4=very healthy	% saying nodiff

Table 5a. Indicators of Compellingness by Label Condition: Cereal.

Label Condition	Purchase(Q.7)	Healthy (Q.5)	Nodiff (Q.3)
	Mean	Mean	Percent
Control	3.08	3.14	23.8
Content	3.06	3.24	21.8
All Health Claims	3.24	3.38	14.1
Short	3.52	3.49	11.0
Long	3.28	3.47	9.4
FDA-Short	3.45	3.47	16.4
FDA-Long	3.10	3.36	17.9
NIH-Short	3.19	3.36	11.9
NIH-Long	3.18	3.39	9.4
SBP-Short	3.00	3.22	19.7
SBP-Long	3.17	3.31	17.5
	1-4, 4=very likely	1-4,4=very healthy	% saying nodiff

Table 5b. Indicators of Compellingness by Label Condition: Lasagna.

Label Condition	Purchase(Q.7)	Healthy (Q.5)	Nodiff (Q.3)
	Mean	Mean	Percent
Control	2.68	2.55	35.9
Content	3.07	2.84	28.2
All Health Claims	2.81	2.64	35.1
Short	3.05	2.87	33.3
Long	2.99	2.72	37.3
FDA-Short	2.75	2.67	35.1
FDA-Long	2.72	2.57	29.9
NIH-Short	2.61	2.50	38.1
NIH-Long	2.68	2.67	36.8
SBP-Short	2.82	2.43	36.8
SBP-Long	2.88	2.66	33.3
	1-4, 4=very likely	1-4,4=very healthy	% saying nodiff

Table 5c. Indicators of Compellingness by Label Condition: Yogurt.

Label Condition	Purchase(Q.7)	Healthy (Q.5)	Nodiff (Q.3)
	Mean	Mean	Percent
Control	2.92	3.10	24.3
Content	2.99	3.08	18.2
All Health Claims	3.07	3.09	19.4
Short	3.04	3.02	21.4
Long	3.03	3.12	22.0
FDA-Short	3.18	3.08	20.5
FDA-Long	2.98	3.06	20.3
NIH-Short	2.94	3.03	18.8
NIH-Long	2.98	2.99	17.1
SBP-Short	3.10	3.19	13.7
SBP-Long	3.31	3.26	21.4
	1-4, 4=very likely	1-4,4=very healthy	% saying nodiff

Table 6. Results of Planned Contrasts for Indicators of Compellingness.

	Hypothesis 1	Hypothesis 2	Hypothesis 3	Hypothesis 4	Hypothesis 5
Purchase	F(1,4130)	F(1,4130)	F(1,4130)	F(1,4130)	F(1,4130)
All	6.95 **	ns	3.85	(13.07) ***	ns
Cereal	ns	5.35 *	6.73 **	(4.06) *	ns
Lasagna	16.37 ***	(11.27) ***	ns	(11.27) ***	ns
Yogurt	ns	ns	ns	ns	4.14 *
Healthy	F(1,4140)	F(1,4140)	F(1,4140)	F(1,4140)	F(1,4140)
All	8.76 **	ns	ns	(7.45) **	ns
Cereal	ns	6.11 *	ns	ns	ns
Lasagna	16.06 ***	(12.50) ***	ns	(8.52) **	ns
Yogurt	ns	ns	ns	ns	ns
Nodiff	F(1,4178)	F(1,4178)	F(1,4178)	F(1,4178)	F(1,4178)
All	5.81 *	ns	ns	ns	ns
Cereal	ns	6.50 *	ns	ns	ns
Lasagna	3.73	(4.66) *	ns	ns	ns
Yogurt	2.72	ns	ns	ns	ns
NOTE. H1: Control v. Content, H2: Content v. Health Claims, H3: Short v. Long, H4: Generic v. Endorsed, H5: See Back Panel v. Generic Long. For all tables: * p<.05, **p<.01, ***p<.001, ns F<1.0

Communication of Risk Group Information

If respondents thought that the product had some health benefits, they were asked whether those benefits would be enjoyed by everyone ( EVERYONE) or whether only some people would benefit from the product (SOMEPEOPLE). These two variables were used to identify how well health claims communicated information about who would derive health benefits from the product. The long versions of the product-appropriate health claims used in the study contained messages identifying certain risk groups as the primary beneficiaries of the product health benefits, so it was hypothesize that respondents in long health claim conditions would be more likely to recognize limitations on the health benefits from the product . At this level of analysis we will not be concerned about the accuracy of the perceived health benefits. Table 7 presents the results across all products for these variables, and Tables 7a, 7b and 7c present the results for the cereal, lasagna and yogurt products respectively. Table 8 presents the results of the planned comparisons testing the five experimental hypotheses.

The comparison between long and short health claims provided the most direct test of the ability of a health claim to communicate risk group information. Across products the effect of long versus short messages on perceptions of who would benefit from the given product was not significant. There was a significant effect of a long versus short health claim for the yogurt product. In the case of yogurt, respondents seeing the long health claim were more likely to recognize limitations on the health benefits of yogurt, but this effect was not found for either the cereal or lasagna product (Table 8: Hypothesis 3; SOMEPEOPLE). For the most part, respondents' beliefs about who will benefit from eating the product seemed to depend much more on type of product, than on the information provided by the health claim.

Table 7. Measures of Risk Group Communication.
Label Condition	Somepeople(Q.3)	Everyone(Q.3)
	Percent	Percent
Control	36.2	31.5
Content	38.6	34.5
All Health Claims	38.8	34.9
Short	33.2	42.0
Long	39.7	33.4
FDA-Short	37.8	35.2
FDA-Long	41.5	31.5
NIH-Short	39.1	32.9
NIH-Long	40.2	35.9
SBP-Short	38.7	34.2
SBP-Long	40.1	34.0

Table 7a. Measures of Risk Group Communication: Cereal.
Label Condition	Somepeople(Q.3)	Everyone(Q.3)
	Percent	Percent
Control	31.5	42.6
Content	34.2	39.3
All Health Claims	34.9	47.0
Short	26.3	61.0
Long	28.2	53.8
FDA-Short	30.2	49.1
FDA-Long	41.0	37.6
NIH-Short	36.4	48.3
NIH-Long	35.0	53.0
SBP-Short	42.7	32.5
SBP-Long	39.5	40.4

Table 7b. Measures of Risk Group Communication: Lasagna.
Label Condition	Somepeople(Q.3)	Everyone(Q.3)
	Percent	Percent
Control	42.7	14.5
Content	42.7	25.2
All Health Claims	37.9	23.1
Short	41.0	23.1
Long	40.7	20.3
FDA-Short	39.5	24.6
FDA-Long	35.9	25.5
NIH-Short	36.4	20.3
NIH-Long	40.2	19.7
SBP-Short	29.1	31.6
SBP-Long	40.2	19.7

Table 7c. Measures of Risk Group Communication: Yogurt.

Label Condition	Somepeople(Q.3)	Everyone(Q.3)
	Percent	Percent
Control	34.4	37.4
Content	39.0	39.0
All Health Claims	43.5	34.5
Short	32.5	41.9
Long	50.0	26.3
FDA-Short	43.6	31.6
FDA-Long	47.5	30.5
NIH-Short	44.4	29.9
NIH-Long	45.3	35.0
SBP-Short	44.4	38.5
SBP-Long	40.5	42.2

Table 8. Results of Planned Contrasts for Measures of Risk Group Communication.

	Hypothesis 1	Hypothesis 2	Hypothesis 3	Hypothesis 4	Hypothesis 5
Somepeople	F(1,4179)	F(1,4179)	F(1,4179)	F(1,4179)	F(1,4179)
All	ns	ns	3.1	2.0	ns
Cereal	ns	ns	ns	4.7 *	3.1
Lasagna	ns	(1.9)	ns	ns	ns
Yogurt	ns	6.6 *	4.1 *	ns	ns
Everyone	F(1,4179)	F(1,4179)	F(1,4179)	F(1,4179)	F(1,4179)
All	ns	ns	ns	(3.3)	ns
Cereal	ns	1.7	ns	(7.9) **	(4.9) *
Lasagna	(6.2) *	(6.6) *	ns	ns	ns
Yogurt	ns	ns	ns	ns	7.0 **
NOTE. H1: Control v. Content, H2: Control v. Health Claims, H3: Short v. Long, H4: Generic v. Endorsed, H5: See Back Panel v. Generic Long.

Accuracy measures based on open-ended questions

Table 9 presents the results for seven accuracy measures of perceived product health benefits based on respondents' answers to open-ended questions about whether the product has health benefits, what they are, and who would be most likely to enjoy them. Respondents answered different questions depending on their previous answers, (e.g., respondents who said that eating the product wouldn't have any health benefits were not asked further about the perceived health benefits of the product), but there was no attempt in the present analyses to take account of the self-selection process that results in different numbers of respondents answering different questions. Respondents who were not asked a question were simply omitted from the analysis. Table 9a, 9b, and 9c present the results for the cereal, lasagna and yogurt products, respectively. Table 10 presents the results of the planned contrasts testing the five experimental hypotheses.

The overall frequency distributions for the six dichotomous variables based on Q3a and Q3b2 showed basic tendencies in how respondents' talked about product health benefits. The dominant "vocabulary" respondents used was one of nutrient characteristics rather than disease effects. Nutrient characteristics of products were by far the most frequent type of reported product health benefits, both nutrients that were mentioned in the front panel claims and those that were not. Although respondents were asked specifically about "health benefits", and product packages were continuously available for inspection, respondents were at least four times more likely to describe the health benefits of the product in terms of its nutrient characteristics than in terms of its specific disease effects. Few respondents identified a diet/disease effect that was not true for the product. On average, fewer than three percent of respondents mentioned a false disease effect as a possible health benefit of a product.

There were significant effects of label condition on perceived product health benefits. Compared to a control label, a content claim increased the likelihood that respondents said the nutrient characteristic(s) mentioned in the claim were product health benefits (Table 10; Hypothesis 1; NDHIT). However, respondents who saw a content claim were also less likely to recognize that other nutrient characteristics not mentioned in the claim would provide health benefits compared to those who saw only a control label (Table 10; Hypothesis 1; NTRUE). The same pattern of effects, increased accuracy with respect to direct hits, but reduced accuracy with respect to 'trues', was found when the average health claim was compared to a control claim, although in this case the effects included mentions of specific diseases effects as well as nutrient characteristics (Table 10: Hypothesis 2; DDHIT, DTRUE, NDHIT, NTRUE). Health claims, on average, increased the likelihood that respondents recognized the specific nutrients or dietary diseases mentioned in the claim as product health benefits, but decreased the likelihood that other possible health benefits of the product were recognized.

Short health claims were better than long health claims in communicating specific disease effects as product health benefits (Table 10: Hypothesis 3, DDHIT). On several other measures, the results indicated that short health claims communicated product health benefits better than longer messages, although the effects did not always reach conventional significance levels (Table 10: Hypothesis 3; DTRUE, DFALSE, NFALSE). Endorsed health claims compared to non-endorsed claims increased the likelihood that respondents recognized true product characteristics not mentioned by the health claim, (Table 10: Hypothesis 4; DTRUE, NTRUE), but the results did not support the hypothesis that endorsed health claims would be better at accurately communicating about product health benefits mentioned in the health claim. The split health claim message was actually less effective than the long generic claim at communicating product nutrient characteristics mentioned in the claim, but it increased the likelihood that respondents would notice true product nutrient characteristics not mentioned in the claim (Table 10: Hypothesis 5; NDHIT, NTRUE).

The RISKGRP variable measured the accuracy of respondents reported perceptions about who would be most likely to get health benefits from eating the given product. Those who answered this question agreed in a prior question that only some people would get benefits from eating the food (SOMEPEOPLE=1). They had the product package with its health claim in front of them when they answered the question. The results demonstrated the limited potential of product labels to communicate about specific risk groups for specific dietary diseases. Short and long health claims, but particularly long health claims, were more likely to communicate information about relevant risk groups for certain dietary diseases than the control claim (Table 10: Hypothesis 2 and Hypothesis 3; RISKGRP), but the magnitude of this effect was fairly modest. Table 11 presents the frequency distribution of RISKGRP by type of product and length of health claim. The large majority of respondents in every condition were in the RISKGRP=2 category, meaning that respondents were relying more on prior beliefs about the product to answer the question than what was on the product label. Even in the most favorable situation of the folic acid/ birth defect claim found on cereal, when the nature of the dietary disease essentially identified the relevant risk group, fewer than a quarter of respondents gave a clear indication that they had comprehended the health claim information about who was most likely to get a health benefit.

Table 9. Open End Measures of Perceived Health Benefits by Label Condition.

Label Condition	DDHIT	DTRUE	DFALSE	NDHIT	NTRUE	NFALSE	RISKGRP
	%	%	%	%	%	%	Mean
Control	4.4	17.8	2.7	60.8	44.7	28.9	2.11
Content	7.8	15.3	2.5	74.6	36.0	30.3	2.19
All Health Claims	17.4	13.8	2.8	71.4	36.2	28.5	2.35
Short	20.4	9.8	1.9	72.8	36.6	29.1	2.29
Long	15.5	11.2	5.4	75.6	31.0	32.2	2.39
FDA-Short	22.1	20.2	3.2	73.5	34.4	24.9	2.32
FDA-Long	15.2	14.4	3.5	70.0	37.4	30.0	2.48
NIH-Short	14.6	19.3	0.8	66.9	39.4	28.0	2.22
NIH-Long	13.5	14.6	2.2	70.8	41.9	24.3	2.34
SBP-Short	18.8	12.1	2.3	73.4	30.8	28.5	2.26
SBP-Long	19.5	9.3	2.7	68.5	38.1	31.1	2.46
	%	%	%	%	%	%	1-4,4= direct hit

Table 9a. Open End Measures of Perceived Health Benefits by Label Condition: Cereal.

Label Condition	DDHIT	DTRUE	DFALSE	NDHIT	NTRUE	NFALSE	RISKGRP
	%	%	%	%	%	%	Mean
Control	0.0	24.7	3.4	54.6	58.6	17.2	2.04
Content	4.1	22.7	2.3	70.3	53.5	12.8	2.12
All Health Claims	14.8	20.6	0.9	64.2	61.7	17.8	2.50
Short	18.4	12.6	1.0	68.9	58.3	17.5	2.45
Long	16.7	13.5	1.0	72.9	63.5	25.0	2.52
FDA-Short	15.2	31.3	0.0	69.6	59.8	16.3	2.48
FDA-Long	14.1	13.0	2.2	54.3	62.0	16.3	2.59
NIH-Short	14.0	31.0	1.0	65.0	61.0	19.0	2.37
NIH-Long	7.8	22.3	0.0	61.2	72.8	13.6	2.51
SBP-Short	11.4	18.2	2.3	64.8	58.2	15.9	2.25
SBP-Long	20.9	10.0	0.0	57.1	58.3	18.7	2.79
	%	%	%	%	%	%	1-4,4= direct hit

Table 9b. Open End Measures of Perceived Health Benefits by Label Condition: Lasagna.

Label Condition	DDHIT	DTRUE	DFALSE	NDHIT	NTRUE	NFALSE	RISKGRP
	%	%	%	%	%	%	Mean
Control	3.7	8.2	0.7	48.5	37.3	29.1	2.00
Content	6.9	6.2	1.9	69.2	22.0	42.8	2.15
All Health Claims	15.4	5.6	2.6	73.2	19.7	39.1	2.17
Short	18.7	5.3	1.3	73.3	17.3	45.3	2.15
Long	11.1	4.2	4.2	81.9	4.2	48.6	2.26
FDA-Short	24.7	2.7	1.4	74.0	23.3	35.6	2.16
FDA-Long	8.2	8.2	4.1	78.1	24.7	46.6	2.21
NIH-Short	14.9	7.5	0.0	58.2	25.4	29.9	2.03
NIH-Long	12.8	4.3	7.1	77.1	22.8	41.4	2.05
SBP-Short	16.9	5.6	0.0	69.0	15.5	38.0	2.23
SBP-Long	15.7	7.1	2.9	74.3	24.3	27.1	2.28
	%	%	%	%	%	%	1-4,4= direct hit

Table 9c. Open End Measures of Perceived Health Benefits by Label Condition: Yogurt.

Label Condition	DDHIT	DTRUE	DFALSE	NDHIT	NTRUE	NFALSE	RISKGRP
	%	%	%	%	%	%	Mean
Control	9.5	18.3	3.6	76.9	36.1	40.8	2.30
Content	12.2	16.1	3.3	83.3	31.7	36.1	2.29
All Health Claims	21.6	13.4	4.8	77.3	23.5	31.2	2.38
Short	24.1	10.3	3.4	77.0	27.6	28.7	2.33
Long	17.8	14.4	11.1	73.0	17.8	26.7	2.42
FDA-Short	27.3	12.5	8.0	77.3	17.0	25.0	2.36
FDA-Long	21.7	20.7	4.3	79.3	22.8	30.4	2.60
NIH-Short	14.9	14.9	1.1	75.9	25.3	36.8	2.24
NIH-Long	20.2	13.8	1.1	76.6	22.3	23.4	2.47
SBP-Short	26.8	11.3	4.1	84.5	25.8	33.0	2.30
SBP-Long	20.1	9.4	5.2	75.0	29.2	45.8	2.34
	%	%	%	%	%	%	1-4,4= direct hit

Table 10. Results of Planned Contrasts for Open-ended Measures of Perceived Health Benefits.

	Hypothesis 1	Hypothesis 2	Hypothesis 3	Hypothesis 4	Hypothesis 5
DDHIT	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)
All	2.37	55.4 ***	7.16 **	ns	ns
Cereal	ns	27.1 ***	ns	(2.61)	ns
Lasagna	ns	12.8 ***	7.06 **	ns	ns
Yogurt	ns	17.9 ***	ns	ns	ns
DTRUE	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)
All	ns	(5.0) *	2.2	10.1 * *	ns
Cereal	ns	(2.2)	17.58 ***	20.1 ***	ns
Lasagna	ns	ns	ns	ns	ns
Yogurt	ns	(2.8)	ns	ns	ns
DFALSE	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)
All	ns	ns	5.8 *	ns	ns
Cereal	ns	3.43	ns	ns	ns
Lasagna	ns	ns	7.4 **	ns	ns
Yogurt	ns	ns	ns	6.1 *	6.2 *
NDHIT	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)
All	24.7 ***	25.5 ***	ns	(2.8)	(4.6) *
Cereal	10.6 **	6.5 *	ns	(4.6) *	(5.8) *
Lasagna	15.4 ***	32.9 ***	(5.9) *	ns	ns
Yogurt	ns	ns	ns	ns	ns
NTRUE	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)
All	(8.22) **	(16.6) ***	ns	4.4 *	4.7 *
Cereal	ns	ns	(3.0)	ns	ns
Lasagna	(8.32) **	(16.5) ***	ns	8.4 **	7.0 **
Yogurt	ns	(10.7) **	ns	ns	3.0
NFALSE	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)	F(1,3025)
All	ns	ns	ns	3.0	ns
Cereal	ns	ns	ns	ns	ns
Lasagna	(7.0) **	(5.5) *	4.1 *	3.7	8.4
Yogurt	ns	6.5 *	ns	ns	(8.7)
RISKGRP	F(1, 1500)	F(1, 1500)	F(1, 1500)	F(1, 1500)	F(1, 1500)
All	ns	24.8 ***	(6.1) *	ns	ns
Cereal	ns	27.6 ***	ns	ns	3.2
Lasagna	ns	4.8 *	ns	ns	ns
Yogurt	ns	ns	(5.9) *	ns	ns

NOTE. H1: Control v. Content, H2: Control v. Health Claims, H3: Short v. Long, H4: Generic v. Endorsed, H5: See Back Panel v. Generic Long.

Table 11: Frequency distribution of RISKGRP by product type and length of health claim.

Type of Product	Type of Label	Riskgrp = 1	Riskgrp = 2	Riskgrp = 3	Riskgrp = 4
		%	%	%	%
Cereal	Control (N=148)	4.7	84.5	8.8	2.0
Cereal	Short (N=153)	2.0	74.5	7.8	15.7
Cereal	Long (N=158)	1.9	59.5	14.6	24.1
Lasagna	Control (N=182)	9.3	75.8	12.6	2.2
Lasagna	Short (N=160)	8.8	72.5	15.0	3.8
Lasagna	Long (N=176)	6.3	72.7	15.9	5.1
Yogurt	Control (N=164)	2.4	69.5	24.4	3.7
Yogurt	Short (N=186)	2.2	69.9	23.1	4.8
Yogurt	Long (N=203)	1.5	63.6	22.2	12.8

Accuracy measures based on closed-ended questions

The results for the four accuracy measures based on closed-ended questions about possible health benefits (CORRECT, INCORRECT, MARGINAL, and TRUECONT) are presented averaged across all products (Table 12), and for the cereal, lasagna and yogurt products individually (Tables 12a, 12b and 12c). The results of the planned comparisons testing the five experimental hypotheses are presented in Table 13.

Compared to control claims, both content and health claims had similar effects. Both types of claims increased respondents' ratings of the magnitudes of expected health effects of the product, regardless of whether the expected health effect was explicitly mentioned in the health claim (CORRECT), implausible for the product (INCORRECT), or possible health effects that were not mentioned in the health claim but could be partially supported by available scientific evidence (MARGINAL) (Table 13, Hypotheses 1 and 2, CORRECT, INCORRECT AND MARGINAL). Apparently, the positive impressions conveyed by either a content or health claim generalized to other possible health benefits of the product, not just those explicitly identified in the message. It is hard to conclude that health claims increased the overall accuracy of respondents perceptions of health benefits since the impact was indiscriminate with respect to the product-specific appropriateness of the given health effect.

Neither the length of the health claim nor whether it was split between the front and back of the package had a consistent effect on the accuracy of respondents' ratings of product health benefits. Endorsements made a significant difference on respondents' ratings, albeit in the direction opposite to predictions (Table 13: Hypothesis 4, CORRECT, INCORRECT, MARGINAL). Respondents rated the magnitude of product health effects lower after seeing endorsed compared to non-endorsed health claims. Again, the effect occurred regardless of the product-specific appropriateness of the given health effect.

Both content and health claims increased the accuracy of respondents' ratings of product nutrient levels (Table 13: Hypothesis 1 and Hypothesis 2, TRUECONT).

Table 12. Closed End Measures of Perceived Health Benefits and Nutrient Profile.

By Label Condition.

Label Condition	CORRECT	INCORRECT	MARGINAL	TRUECONT
	Mean	Mean	Mean	Mean
Control	2.40	1.94	2.17	2.36
Content	2.59	2.07	2.25	2.44
All Health Claims	2.68	2.05	2.25	2.46
Short	2.84	2.13	2.30	2.52
Long	2.81	2.13	2.35	2.46
FDA-Short	2.69	2.07	2.25	2.49
FDA-Long	2.59	2.03	2.22	2.44
NIH-Short	2.61	2.00	2.18	2.45
NIH-Long	2.65	2.03	2.22	2.48
SBP-Short	2.55	2.03	2.19	2.41
SBP-Long	2.65	2.08	2.25	2.43
	1-4,4=lrge eff.	1-4,4=lrge eff.	1-4,4=lrge eff.	1-3,3=accur.

Table 12a. Closed End Measures of Perceived Health Benefits and Nutrient Profile.

By Label Condition: Cereal.

Label Condition	CORRECT	INCORRECT	MARGINAL	TRUECONT
	Mean	Mean	Mean	Mean
Control	2.61	2.07	2.46	2.61
Content	2.79	2.18	2.59	2.69
All Health Claims	2.93	2.25	2.61	2.68
Short	3.08	2.31	2.71	2.75
Long	3.22	2.38	2.86	2.72
FDA-Short	2.87	2.31	2.62	2.66
FDA-Long	2.91	2.15	2.47	2.69
NIH-Short	2.79	2.17	2.45	2.67
NIH-Long	2.92	2.25	2.60	2.69
SBP-Short	2.74	2.19	2.57	2.62
SBP-Long	2.93	2.25	2.59	2.61
	1-4,4=lrge eff.	1-4,4=lrge eff.	1-4,4=lrge eff.	1-3,3=accur.

Table 12b. Closed End Measures of Perceived Health Benefits and Nutrient Profile.

By Label Condition: Lasagna.

Label Condition	CORRECT	INCORRECT	MARGINAL	TRUECONT
	Mean	Mean	Mean	Mean
Control	1.92	1.63	1.90	2.01
Content	2.29	1.87	2.00	2.14
All Health Claims	2.20	1.84	1.92	2.16
Short	2.49	1.93	1.95	2.19
Long	2.33	1.93	1.99	2.19
FDA-Short	2.16	1.81	1.87	2.18
FDA-Long	2.16	1.94	1.99	2.14
NIH-Short	2.10	1.75	1.90	2.11
NIH-Long	2.13	1.82	1.92	2.14
SBP-Short	1.98	1.66	1.73	2.17
SBP-Long	2.21	1.89	2.03	2.16
	1-4,4=lrge eff.	1-4,4=lrge eff.	1-4,4=lrge eff.	1-3,3=accur.

Table 12c. Closed End Measures of Perceived Health Benefits and Nutrient Profile.

By Label Condition: Yogurt.

Label Condition	CORRECT	INCORRECT	MARGINAL	TRUECONT
	Mean	Mean	Mean	Mean
Control	2.67	2.12	2.13	2.44
Content	2.66	2.18	2.15	2.49
All Health Claims	2.91	2.11	2.21	2.53
Short	2.96	2.16	2.23	2.61
Long	2.90	2.10	2.20	2.46
FDA-Short	3.04	2.09	2.25	2.61
FDA-Long	2.71	2.00	2.19	2.49
NIH-Short	2.98	2.09	2.20	2.56
NIH-Long	2.90	2.04	2.13	2.58
SBP-Short	2.92	2.24	2.25	2.44
SBP-Long	2.85	2.13	2.19	2.52
	1-4,4=lrge eff.	1-4,4=lrge eff.	1-4,4=lrge eff.	1-3,3=accur.

Table 13. Results of Planned Contrasts for Closed Ended Accuracy Measures.

	Hypothesis 1	Hypothesis 2	Hypothesis 3	Hypothesis 4	Hypothesis 5
Correct	F(1,4035)	F(1,4035)	F(1,4035)	F(1,4035)	F(1,4035)
All	11.7 ***	44.2 ***	ns	(17.1) ***	(4.2) *
Cereal	3.9 *	19.4 ***	(2.0)	(12.0) ***	(4.8) *
Lasagna	16.2 ***	14.4 ***	ns	(11.8) ***	ns
Yogurt	ns	11.1 ***	4.5 *	ns	ns
Incorrect	F(1,3927)	F(1,3927)	F(1,3927)	F(1,3927)	F(1,3927)
All	(10.4) **	(14.5) ***	ns	7.2 **	ns
Cereal	(2.22)	(10.0) **	ns	3.7	ns
Lasagna	(10.4) **	(13.0) ***	ns	2.4	ns
Yogurt	ns	ns	ns	ns	ns
Marginal
All	(3.9)	(5.9)*	ns	8.6 **	ns
Cereal	(3.1)	(6.9)**	ns	15.3 ***	6.5 *
Lasagna	(1.8)	ns	ns	ns	ns
Yogurt	ns	(1.6)	ns	ns	ns
Truecont	F(1,3994)	F(1,3994)	F(1,3994)	F(1,3994)	F(1,3994)
All	15.3 ***	37.2 ***	2.4	1.9	ns
Cereal	4.3 *	5.1 *	ns	3.1	(4.0) *
Lasagna	10.8 **	25.0 ***	ns	2.0	ns
Yogurt	1.9	10.7 **	7.6 **	ns	ns

NOTE. H1: Control v. Content, H2: Control v. Health Claims, H3: Short v. Long, H4: Generic v. Endorsed, H5: See Back Panel v. Generic Long.

Observational Measures of Label Reading Behavior

Table 14 presents the distributions of label reading variables by question series. The distributions of the label reading variables averaged across all questions and all products are in Table 15, and the distributions averaged across all questions for the cereal, lasagna and yogurt products individually are in Tables 15a, 15b and 15c, respectively. The results of the planned comparisons testing the five experimental hypotheses for the label reading variables are presented in Table 16.

As shown in Table 14, respondents tended to consult the back of the label when answering questions about the product or its packaging. This was particularly true when they answered specific (i.e., closed-ended) questions about the nutrient characteristics or health benefits of the product. Respondents tended to consult the front of the package more when they answered open-ended questions about product health benefits or when they made global judgments about the products such as whether or not they would buy it.

Respondents were more likely to read the front of the package, and less likely to read the back of the package, when the front of the package had either a health or content claim (Table 16: Hypothesis 1 and Hypothesis 2, FRONT, BACK). However, this effect was found only for the cereal and lasagna products, not for the yogurt product. For the yogurt product, label condition had no effect at all on reading behavior.

For the most part, the effects of health claim devices such as message length, endorsements, or placement on reading behavior depended very much on the specific product that respondents were looking at. Message length did not have a consistent effect on label reading behavior (Table 16: Hypothesis 3). Endorsements of the cereal health claims caused respondents to look more at the front, and less at the back of the package, but endorsements of the lasagna health claim caused respondents to do the opposite; look less at the front and more at the back (Table 16: Hypothesis 4). The effect of the split message presentation of health claims on both cereal and yogurt products was paradoxically that respondents were more likely to look at the front label than the back (Table 16: Hypothesis 5).

Table 14. Observational Measures of Label Reading Behavior by Question Series.

Question Series	FRONT	BACK	NEITHER
Q2a-Q2b (Global Ratings)	31.8	64.7	3.5
Q3-Q3c (Health Benefits, open-end)	41.0	50.0	9.0
Q4a-Q4g (Health Benefits, closed-end)	28.7	56.8	14.5
Q5 (Healthy Rating)	35.1	41.8	23.1
Q6a-Q6b (Nutrient Ratings)	15.6	79.7	4.7
Q7 (Purchase Intentions)	38.0	42.5	19.5

Table 15. Observational Measures of Label Reading Behavior By Label Condition.

Label Condition	FRONT	BACK	NEITHER
	%	%	%
Control	24.6	61.5	13.9
Content	31.4	56.4	12.2
All Health Claims	33.1	54.7	12.2
Short	33.7	54.3	12.0
Long	32.5	56.6	10.9
FDA-Short	34.2	54.3	11.5
FDA-Long	34.6	52.9	12.5
NIH-Short	31.5	56.6	11.9
NIH-Long	31.0	55.5	13.5
SBP-Short	31.3	54.0	14.7
SBP-Long	35.7	53.2	11.1

Table 15a. Observational Measures of Label Inspection Behavior By Label Condition: Cereal.

Label Condition	FRONT	BACK	NEITHER
	%	%	%
Control	22.7	59.7	17.6
Content	31.7	53.5	14.8
All Health Claims	34.1	54.6	11.3
Short	29.2	59.6	11.2
Long	28.6	60.8	10.6
FDA-Short	35.4	54.1	10.5
FDA-Long	35.4	49.7	14.9
NIH-Short	35.0	55.8	9.2
NIH-Long	32.2	56.9	10.9
SBP-Short	38.6	47.3	14.1
SBP-Long	38.2	52.5	9.3

Table 15b. Observational Measures of Label Inspection Behavior By Label Condition: Lasagna.

Label Condition	FRONT	BACK	NEITHER
	%	%	%
Control	21.6	67.4	11.0
Content	31.8	58.5	9.7
All Health Claims	33.9	54.5	11.6
Short	35.3	49.5	15.2
Long	40.3	51.2	8.5
FDA-Short	34.8	56.0	9.2
FDA-Long	37.5	54.2	8.3
NIH-Short	28.3	61.0	10.7
NIH-Long	34.4	54.4	11.2
SBP-Short	27.4	55.1	17.5
SBP-Long	33.1	54.9	12.0

Table 15c. Observational Measures of Label Inspection Behavior By Label Condition: Yogurt.

Label Condition	FRONT	BACK	NEITHER
	%	%	%
Control	29.5	57.7	12.8
Content	30.7	57.4	11.9
All Health Claims	31.2	55.1	13.7
Short	36.7	53.8	9.5
Long	28.2	58.0	13.8
FDA-Short	32.3	52.8	14.9
FDA-Long	31.2	54.9	13.9
NIH-Short	31.2	53.8	15.0
NIH-Long	26.3	55.3	18.4
SBP-Short	28.0	59.6	12.4
SBP-Long	35.8	52.3	11.9

Table 16. Results of Planned Contrasts for Observational Measures.

	Hypothesis 1	Hypothesis 2	Hypothesis 3	Hypothesis 4	Hypothesis 5
FRONT	F(1, 4043)	F(1, 4043)	F(1, 4043)	F(1, 4043)	F(1, 4043)
All	33.3 ***	82.4 ***	ns	ns	3.9 *
Cereal	19.6 ***	50.4 ***	ns	10.0 **	10.8 **
Lasagna	24.9 ***	57.0 ***	(7.6) **	(5.4) *	(6.3) *
Yogurt	ns	ns	8.4 **	ns	6.9 **
BACK	F(1, 4043)	F(1, 4043)	F(1, 4043)	F(1, 4043)	F(1, 4043)
All	(15.1) ***	(43.1)	ns	ns	(3.4)
Cereal	(7.3) **	(7.9) **	ns	(9.3) **	(6.4) *
Lasagna	(15.0) ***	(49.4) ***	ns	9.4 **	ns
Yogurt	ns	ns	ns	ns	(3.1)

NOTE. H1: Control v. Content, H2: Control v. Health Claims, H3: Short v. Long, H4: Generic v. Endorsed, H5: See Back Panel v. Generic Long.

Ratings of Package Information

Table 17 presents the means for ratings of package information by label condition averaged across all products, and Tables 17a, 17b and 17c present the means

by label condition for each product separately. Table 18 presents the results of the planned comparison tests of experimental hypotheses for these measures.

The variable NOTICE was essentially a manipulation check of whether or not respondents recognized health information on the label. Ninety five percent of respondents recognized that labels with health claims provided health information. Different ways of presenting health claim information had no effect on this recognition level (Table 18: Hypothesis 3, Hypotheses 4 and Hypothesis 5, NOTICE). More than 90 percent of respondents recognized that content claims also provided health information. It was surprising how many respondents considered a label with no claims at all to provide health information. Around 15 percent of respondents who saw the control versions of the cereal or lasagna packages considered the front labels of these packages to provide health information, and more than 70 percent of respondents who saw the control version of the yogurt package thought that its front label provided health information. Packaging characteristics were probably responsible for these unintended communications about the product's health characteristics. The yogurt package, for example, had a standard of identity classification, "low-fat yogurt", which could easily be seen as a content claim, and the cereal product was called Raisin Bran, which many consumers might have associated with high fiber. The yogurt product also carried a "contains live cultures" statement that was probably seen as health information. The results showed how easy it is to communicate about product health characteristics without using explicit health or content claims.

The variables GOODJOB and ENUFINFO were evaluative measures of the information quality of the product labels. Overall, respondents were highly positive about the labeling on these packages, reflecting the realistic and professional quality of the package executions. Content claims produced significantly more positive evaluations than the control claim only for the yogurt product (Table 18: Hypothesis 1; GOODJOB, ENUFINFO). Health claims produced significantly more positive evaluations than the control claim for both the yogurt and cereal products, but there were significant negative effects of health claims relative to both the control and content claim conditions for the lasagna product (Table 18: Hypothesis 2; GOODJOB, ENUFINFO). The message length, endorsements or placement of health claims had little effect on respondents' evaluations of the labeling, except in the case of the lasagna product. In the case of lasagna, the experimental hypotheses were not just disconfirmed, they were reversed. Short messages, endorsed messages and split messages tended to have negative impacts on respondents evaluations of the labeling when they were used on the lasagna product (Table 18: Hypothesis 3, Hypothesis 4, Hypothesis 5; GOODJOB, ENUFINFO).

The variables TRUSTINFO and TRUST were, respectively, measures of the perceived credibility of the front and back label information. Perceived credibility was high across all products and label conditions. Label condition had no effects on these variables..

Table 17. Evaluative Ratings of Package Information by Label Condition.

Label Condition	GOODJOB	ENUFINFO	TRUSTINFO	NOTICE	TRUST
	Mean	Mean	Mean	Mean	Mean
Control	4.16	4.50	4.26	1.67	NA
Content	4.33	4.53	4.31	1.10	NA
All Health Claims	4.46	4.55	4.29	1.04	4.30
Short	4.49	4.57	4.28	1.04	4.29
Long	4.53	4.64	4.38	1.06	4.36
FDA-Short	4.52	4.53	4.31	1.03	4.26
FDA-Long	4.42	4.52	4.27	1.03	4.25
NIH-Short	4.39	4.56	4.33	1.04	4.31
NIH-Long	4.45	4.52	4.26	1.05	4.36
SBP-Short	4.44	4.58	4.25	1.04	4.27
SBP-Long	4.46	4.51	4.28	1.06	4.27
	1-5,5=agree	1-5,5=agree	1-5,5=agree	1=yes,2=no	1-5,5=agree

Table 17a. Evaluative Ratings of Package Information by Label Condition: Cereal.

Label Condition	GOODJOB	ENUFINFO	TRUSTINFO	NOTICE	TRUST
	Mean	Mean	Mean	Mean	Mean
Control	4.35	4.57	4.29	1.84	NA
Content	4.44	4.63	4.44	1.09	NA
All Health Claims	4.58	4.72	4.43	1.04	4.37
Short	4.63	4.71	4.47	1.04	4.47
Long	4.56	4.74	4.50	1.05	4.51
FDA-Short	4.70	4.77	4.41	1.03	4.26
FDA-Long	4.54	4.69	4.37	1.02	4.30
NIH-Short	4.65	4.72	4.53	1.04	4.42
NIH-Long	4.60	4.73	4.51	1.03	4.47
SBP-Short	4.46	4.74	4.28	1.06	4.32
SBP-Long	4.47	4.62	4.36	1.07	4.24
	1-5,5=agree	1-5,5=agree	1-5,5=agree	1=yes,2=no	1-5,5=agree

Table 17b. Evaluative Ratings of Package Information by Label Condition: Lasagna.

Label Condition	GOODJOB	ENUFINFO	TRUSTINFO	NOTICE	TRUST
	Mean	Mean	Mean	Mean	Mean
Control	4.15	4.53	4.18	1.87	NA
Content	4.29	4.54	4.13	1.09	NA
All Health Claims	4.34	4.40	4.17	1.05	4.18
Short	4.36	4.48	4.08	1.07	4.06
Long	4.55	4.65	4.26	1.07	4.26
FDA-Short	4.34	4.29	4.19	1.05	4.20
FDA-Long	4.41	4.38	4.19	1.04	4.25
NIH-Short	4.11	4.35	4.11	1.02	4.15
NIH-Long	4.33	4.35	4.09	1.04	4.16
SBP-Short	4.28	4.32	4.13	1.04	4.11
SBP-Long	4.34	4.34	4.30	1.04	4.25
	1-5,5=agree	1-5,5=agree	1-5,5=agree	1=yes,2=no	1-5,5=agree

Table 17c. Evaluative Ratings of Package Information by Label Condition: Yogurt.

Label Condition	GOODJOB	ENUFINFO	TRUSTINFO	NOTICE	TRUST
	Mean	Mean	Mean	Mean	Mean
Control	3.97	4.41	4.30	1.29	NA
Content	4.27	4.42	4.36	1.10	NA
All Health Claims	4.47	4.56	4.29	1.04	4.31
Short	4.49	4.51	4.28	1.01	4.34
Long	4.50	4.61	4.38	1.05	4.31
FDA-Short	4.52	4.54	4.32	1.03	4.32
FDA-Long	4.30	4.50	4.26	1.04	4.19
NIH-Short	4.42	4.61	4.36	1.07	4.38
NIH-Long	4.41	4.49	4.17	1.06	4.25
SBP-Short	4.57	4.67	4.34	1.02	4.38
SBP-Long	4.58	4.58	4.20	1.07	4.31
	1-5,5=agree	1-5,5=agree	1-5,5=agree	1=yes,2=no	1-,5=agree

Table 18. Results of Planned Contrasts for Evaluative Ratings of Package Information.

	Hypothesis 1	Hypothesis 2	Hypothesis 3	Hypothesis 4	Hypothesis 5
Goodjob	F(1,4160)	F(1,4160)	F(1,4160)	F(1,4160)	F(1,4160)
All	12.1 ***	59.3 ***	ns	(2.5)	ns
Cereal	ns	10.5 **	1.7	ns	ns
Lasagna	2.5	7.8 **	(5.1) *	(4.4)*	ns
Yogurt	11.9 ***	54.1 ***	ns	ns	ns
Enufinfo	F(1,4161)	F(1,4161)	F(1,4161)	F(1,4161)	F(1,4161)
All	ns	2.6	ns	(3.7)	(4.6) *
Cereal	ns	6.5 *	ns	ns	ns
Lasagna	ns	(4.8) *	ns	(11.6) ***	(8.5) **
Yogurt	ns	5.9 *	ns	ns	ns
Trustinfo	F(1,4153)	F(1,4153)	F(1,4153)	F(1,4153)	F(1,4153)
All	ns	ns	ns	ns	ns
Cereal	ns	3.9 *	ns	ns	ns
Lasagna	ns	ns	ns	ns	ns
Yogurt	ns	ns	ns	ns	ns
Notice	F(1,4179)	F(1,4179)	F(1,4179)	F(1,4179)	F(1,4179)
All	1684.1 ***	3213.4 ***	ns	ns	ns
Cereal	968.8 ***	1770.5 ***	ns	ns	ns
Lasagna	1043.7 ***	1877.1 ***	ns	ns	ns
Yogurt	59.1 ***	162.8 ***	ns	ns	ns
NOTE. H1: Control v. Content, H2: Control v. Health Claims, H3: Short v. Long, H4: Generic v. Endorsed, H5: See Back Panel v. Generic Long.

Summary of Label Effects by Type of Effect and Claim

The effects of health and content claims and the three message devices on respondents' attitudes toward the product and their perceptions of product health benefits were complicated. Table 19 tallies the number of measures that showed significant product-appropriate health claim or content claim effects on the three measures of product attitudes (PURCHASE, HEALTHY, NODIFF) and the thirteen measures of communication accuracy (DDHIT, DTRUE, DFALSE, NDHIT, NTRUE, NFALSE, RISKGRP, CORRECT, INCORRECT, MARGINAL, TRUECONT, FRONT, BACK) scored according to experimental hypotheses for each type of product.

Content Claims. There were no significant effects of a content claim for the yogurt product on either product attitudes or communication measures. The content claim for the cereal product had no significant attitude effects and only five significant communication effects. The content claim for the lasagna product, by contrast, had three significant attitude effects and seven communication effects. Clearly the low fat content claim for the lasagna product had a more practically important impact on product attitudes and perceptions than the content claims for cereal or yogurt.

Health Claims. Health claims tended to have more impact on respondents than content claims, but the impact was not necessarily positive. The folic acid/birth defect health claim for the cereal product had a significant impact on all three measures of attitudes and nine out of the 13 measures of communication accuracy. The low fat/heart disease health claim for lasagna produced three significant negative attitude effects and ten significant communication effects out of thirteen possible measures. The calcium/osteoporosis claim for yogurt had less of an impact than the other health claims.

Message Devices. None of the message features, with the possible exception of endorsements, had much impact on respondents' attitudes toward the product. Endorsements tended to produce more negative attitudes toward the product, calling into question the assumption that endorsements might allay latent consumer skepticism about the credibility of health claims.

The impact of the various message devices on communication accuracy was modest. None of the devices produced more than five significant effects out of the thirteen communication measures. Short messages appeared to improve communication relative to longer measures, particularly for the lasagna and yogurt products. Short messages, particularly for the lasagna product, reduced the likelihood that respondents attributed inappropriate health benefits to the product. The communication effects of endorsements were paradoxical. Endorsed claims did not communicate key message elements as well as non-endorsed claims, but they were better than non-endorsed claims at getting respondents to notice other information on the product label. It may be that respondents were skeptical of endorsements and paid more attention to the back of the label when they saw endorsed health claims. Split presentation health claims had few effects on accuracy of communicated product benefits.

Mixed Character of Communication Effects

In all cases, the communication effects of health or content claims were mixed. Some measures, principally measures of direct hits, showed effects indicating greater accuracy of perceived health benefits associated with the presence of a health claim on the product package. But other measures, principally those intended to gauge a broader range of perceived health benefits (i.e., those that were not explicitly mentioned in the health claim) tended to show negative effects of health and content claims on communication accuracy.

Table 19. Scorecard of Label Effects by Type of Effect and Type of Claim.

PRODUCT	Attitude Effects	Communication Effects
	3 Measures	13 Measures
Cereal
Content (H1)	0	4+, 1-
Health (H2)	3+	6+, 3-
Length (H3)	1+	1+
Endorse (H4)	1-	3+, 2-
Split (H5)	0	2-
Lasagna
Content (H1)	3+	4+, 3-
Health (H2)	3-	6+, 4-
Length (H3)	0	3+, 1-
Endorse (H4)	2-	3+, 2-
Split (H5)	0	3+
Yogurt
Content (H1)	0	0
Health (H2)	0	4+, 1-
Length (H3)	0	3+, 1-
Endorse (H4)	0	3+
Split (H5)	1+	1+, 2-

DISCUSSION

The study was designed to evaluate the consumer impacts of different possible versions of health claims in order to inform recommendations about how to fashion more effective health claims on food product labels. On balance, the evaluation of specific health claim devices indicated that short messages were more effective than long messages, that endorsed health messages have liabilities compared to non-endorsed messages, and that splitting messages between the front and back label made little difference. However, the pattern of findings was so complex that it raised questions about some of our basic working assumptions about the purpose and value of health claims on product labels. Product-appropriate health claims can have no effect, a positive effect, or a negative effect on respondents' attitudes toward a product depending on the particular product and type of health claim. Content claims can be as effective, and sometimes more effective, than product-appropriate health claims in encouraging positive attitudes toward the product.

Health claims appear to have limited ability to communicate risk group or health benefit information. More than twenty percent of respondents did not acknowledge that a product had any health benefits even when the package carried an explicit health claim. Less than forty percent of respondents recognized that there would be distinctive product health benefits for certain kinds of people even when such a message was a major element of the health claim on the product package. Fewer than 20 percent of these people (less than ten percent of the sample) indicated that they had read and/or correctly understood the information about the risk group that would be most likely to get a health benefit from the product.

Health claims communicated specific diet/disease related information about product health benefits better than content claims or control labels. Health claims increased the likelihood that respondents would repeat or "playback" the key message points from the health claim when asked about product health benefits, but this increase in perceived accuracy of product health benefits was associated with certain costs. One apparent cost was that consumers were less likely to acquire other relevant health information from the product label as a consequence of seeing a health claim on the front of the package. Related to this, and a possible explanation for the effect, was the finding that health claims decreased the likelihood of consumers reading the nutrition information on the back of the package. Another apparent cost was that health claims led consumers to believe that the product was likely to have health effects that it did not have. This effect recalls the "halo" effect seen in many judgment situations. A "halo effect" occurs when people generalize from positive information about a person or thing by attributing other positive characteristics that were not mentioned in the original information. This pattern of findings makes it hard to conclude that the impact of health claims is to produce more accurate perceptions of product benefits.

Product Type as a Boundary Condition on Health Claim Effects

The findings that health or content claims had impacts on product attitudes and perceptions of product health benefits for some products (e.g., cereal and lasagna), but not others (e.g., yogurt) calls for an explanation. The manipulation check indicated that respondents were generally aware of the presence of a health or content claim, but it was not clear how carefully they paid attention to the substance of the various claims. The health and content claims on yogurt, for example, had no effect on label reading behavior compared to the control label for yogurt which suggested they did not pay much attention to these claims. Presumably, respondents saw little added value in the information provided by the calcium/osteoporosis claims. In general, the various label claims provided different amounts of new information about the product that the person was not likely to already know. It may be that the presence of added value information is an important boundary condition on whether a label claim will have an impact.

All of the products in the study were familiar to respondents so that they almost certainly had prior beliefs about the nutritional value and healthfulness of these products. Yogurt is widely known to be a high calcium dairy product, and the health benefits of calcium are well known. It was likely that most of the information provided in the health and content claims about yogurt was known to respondents and had little added value for them.. By contrast, the two conditions in the study which showed strong positive label claim effects, the low-fat content claim for lasagna and the folic acid/birth defects claim for cereal, were both conditions where the information provided by the claim was most likely to be new to respondents. Lasagna is not typically a low fat food, and relatively few people are currently aware of the health benefits of folic acid or the folic acid/birth defects relationship (Levy and Derby, 1996). An explanation consistent with these findings, and consistent with much advertising research, is that for a health or content claim to have a significant consumer impact, it must provide information that a consumer considers "new" (i.e., as providing added value). Unless health or content claims provide added value information to consumers, they may not have a practical consumer impact regardless of how they are designed or presented.

The other finding that needs to be explained was that the product-appropriate low fat/heart disease claim for the lasagna product actually produced negative reactions toward the product compared to the low fat content claim by itself. Reinforcing this paradoxical result was the finding that adding endorsements of the health claims from reputable public health authorities increased the negative impact of a health claim. The fact that product-appropriate health claims may sometimes have negative effects is counterintuitive and calls into question the assumption that health claims have purely informational value. Situations in which positive information produce negative effects are usually found in social contexts where the person receiving the information has reasons to question the motives or interests of the information source. In the case of the lasagna product, it is not plausible that respondents doubted the credibility of the low fat/heart disease relationship, which can hardly be in doubt. It must be that respondents felt the health claim was put on the package for a purpose other than providing them with truthful information.

The qualitative research conducted previously by the Keystone dialogue (Levy, 1996) suggested that consumers sometimes considered health claims on product labels to be a form of advertising rather than a credible source of health information. Advertising is susceptible to criticism that it has ulterior motives and manipulative intent. Of the three products in the study, the lasagna product was recognized as the least healthful product. It may be that when a health claim was applied to a product like lasagna, which was not seen as deserving the designation of a healthful product, respondents viewed the message as an inappropriate or manipulative influence attempt. The fact that health claims are presented as public health motivated information messages may make them particularly susceptible to this kind of criticism. We assumed, for example, that endorsements would be a device for enhancing the credibility of health claims. Apparently, respondents took them as evidence of ulterior motives or manipulative intent. As a practical matter, consumers' prior beliefs about the healthful characteristics of foods may constitute effective limits on the potential utility of health claims.

2. Food and Drug Administration, U. S. Department of Health and Human Services. Food Labeling; Final Rules, Health Claims, Federal Register 58: 2537-2849. U. S. Government Printing Office, Washington, D.C., January 6, 1993b.

3. The Keystone Center, The Keystone National Policy Dialogue on Food, Nutrition, and Health: Final Report, Keystone, CO and Washington, D.C., March 1996.

4. SAS Institute, Inc. The GLM Procedure, in SAS/STAT Users Guide: Volume 2, GLM-VARCOMP, Version 6, Fourth Edition, Cary , NC: SAS Institute Inc., 1989, pp. 891-996.

5. Levy, A. S. and Derby, B. M. The Impact of the NLEA on Consumers: Recent Findings from FDA's Food Label and Nutrition Tracking System. Internal Report, January 23, 1996. Food and Drug Administration, Center for Food Safety and Applied Nutrition, Division of Market Studies, Consumer Studies Branch.

6. Levy, A. S. Summary Report on Health Claims Focus Groups, in The Keystone National Policy Dialogue on Food, Nutrition, and Health: Final Report, Keystone, CO and Washington, D.C., March 1996, pp. 141-150.