The Biasing Health Halos of Fast-Food Restaurant Health Claims: Lower Calorie Estimates and Higher Side-Dish Consumption Intentions

The Biasing Health Halos of Fast-Food
Restaurant Health Claims: Lower Calorie
Estimates and Higher Side-Dish
Consumption Intentions
PIERRE CHANDON
BRIAN WANSINK*
Why is America a land of low-calorie food claims yet high-calorie food intake? Four
studies show that people are more likely to underestimate the caloric content of
main dishes and to choose higher-calorie side dishes, drinks, or desserts when
fast-food restaurants claim to be healthy (e.g., Subway) compared to when they
do not (e.g., McDonald’s). We also ?nd that the effect of these health halos can
be eliminated by simply asking people to consider whether the opposite of such
health claims may be true. These studies help explain why the success of fast-
food restaurants serving lower-calorie foods has not led to the expected reduction
in total calorie intake and in obesity rates. They also suggest innovative strategies
for consumers, marketers, and policy makers searching for ways to ?ght obesity.
As the popularity of healthier menus increases, so does way’s television commercial starring Jared Fogle showing
the weight of many Americans. Between 1991 and
that Subway’s turkey sandwich has only 280 calories, half
2001, the proportion of obese U.S. adults has grown from
the 560 calories of a Big Mac, was the most recalled tele-
23% to 31% of the population, a 3% annual compound rate
vision commercial during the 2004 holidays (Advertising
(National Center for Health Statistics 2002). In the same
Age 2005). This parallel increase in obesity rates and in the
period, the proportion of U.S. adults consuming low-calorie
popularity of healthier foods with lower calorie and fat den-
food and beverages grew from 48% to 60% of the population
sity has been noted in consumer research (Seiders and Petty
(a 2.3% annual compound rate), and the proportion of U.S.
2004) and in health sciences as “the American obesity par-
consumers trying to eat a healthy diet grew at a 6% annual
adox” (Heini and Weinsier 1997).
rate (Barrett 2003; Calorie Control Council National Con-
The original explanation of the American obesity paradox
sumer Surveys 2004; Food Marketing Institute 2005). In the
was that people burn fewer calories than they used to be-
past 5 years, fast-food restaurants positioned as healthy (e.g.,
cause of technological progress and changing lifestyles
Subway) have grown at a much faster rate than those not
(Heini and Weinsier 1997). However, this explanation is now
making these claims (e.g., McDonald’s). For example, Sub-
contested. First, the last 4 decades have actually seen an
increase in leisure-time physical activity and a decline in
*Pierre Chandon is associate professor of marketing at INSEAD, Boule-
the proportion of sedentary people (Talbot, Fleg, and Metter
vard de Constance, 77300 Fontainebleau, France (pierre.chandon@insead
2003). Second, Heini and Weinsier relied on self-reported
.edu). Brian Wansink is the John S. Dyson Chair of Marketing and of Nu-
data, which strongly underestimate increases in actual cal-
tritional Science in the Applied Economics and Management Department,
Cornell University, 110 Warren Hall, Ithaca, NY 14853-7801 (Wansink@
orie intake (Chandon and Wansink 2007; Livingstone and
Cornell.edu). Correspondence: Pierre Chandon. The authors wish to thank
Black 2003). In fact, the U.S. Department of Agriculture
Jill North and James E. Painter for help with data collection. The data in
data on food supply (Putnam, Allshouse, and Kantor 2002)
these studies were collected at the expense of the authors, and the studies
show that calorie supply and calorie intake (computed by
were not sponsored by any outside source. Helpful comments on various
aspects of this research were provided by the editor, the JCR reviewers, and
subtracting food losses at home and at all levels of the supply
Vicki Morwitz.
chain) have both increased by 18% since 1983 (reaching,
respectively, 3,900 and 2,800 calories per person and per
John Deighton served as editor and Stephen Hoch served as associate
day in 2000). As a result, most recent reviews of obesity
editor for this article.
research, from ?elds as diverse as economics and epide-
Electronically published June 29, 2007
miology, attribute rising obesity rates to increased calorie
301
? 2007 by JOURNAL OF CONSUMER RESEARCH, Inc. ? Vol. 34 ? October 2007
All rights reserved. 0093-5301/2007/3403-0010$10.00

---

302
JOURNAL OF CONSUMER RESEARCH
intake and not to decreased calorie expenditures (Cutler,
Although it does not elucidate which speci?c mechanism is
Glaeser, and Shapiro 2003; Kopelman 2000).
responsible for health halos, the fourth study demonstrates
In this article, we propose and test a halo-based expla-
how asking a consumer to “consider the opposite” eliminates
nation for a speci?c facet of the American obesity paradox:
the biasing effects of health halos on calorie estimation and
the simultaneous increase in obesity and in the popularity
on side-dish orders. Finally, we discuss the implications of
of restaurants serving lower-calorie foods and claiming to
our ?ndings for research and for reducing the negative ef-
be healthier. We argue that the health claims made by these
fects of health claims in away-from-home and in-home con-
restaurants lead consumers to (1) underestimate the number
sumption.
of calories contained in their main dishes and (2) order
higher-calorie side dishes, drinks, or desserts. Taken to-
CONCEPTUAL FRAMEWORK
gether, these two effects can lead to more overeating (de?ned
as undetected excessive calorie intake) when ordering from
How Health Claims In?uence Calorie Estimations
restaurants positioned as healthy than from restaurants not
making this claim. Health halos can therefore explain why
Restaurants are exempted from the U.S. 1990 Nutrition
the increased popularity of healthier fast-food restaurants
Labeling and Education Act, which made calorie and other
has not led to the expected reduction in total calorie intake
nutrition information mandatory for packaged goods. In the
and in obesity rates.
absence of nutrition information, it is very dif?cult to es-
Studying how health claims in?uence calorie estimations
timate calorie content through visual inspection or sensory
and the choice of side dishes helps bridge the multidisciplin-
satiation (Chandon and Wansink 2007; Livingstone and
ary obesity research efforts in health sciences and consumer
Black 2003). Even when consumers know the list of ingre-
research. The Food and Drug Administration has singled out
dients included in a meal, they have dif?culty estimating
away-from-home consumption as a critical contributor to
portion sizes (Nestle 2003). Consumers asked to estimate
overeating (Food and Drug Administration 2006). Still, bi-
the number of calories contained in a meal must therefore
ased calorie estimations of restaurant foods are less fre-
make inferences based on internal and external cues, such
quently noted in health sciences than the other factors con-
as the health positioning of the restaurant’s brand. The am-
tributing to overeating, such as the increase in portion size
biguity of sensory experience also increases the chances that
(Ledikwe, Ello-Martin, and Rolls 2005; Nielsen and Popkin
calorie estimations are in?uenced by the activation of spe-
2003), the higher availability of ready-made foods (Cutler
ci?c consumption goals, by feelings of guilt, or by self-
et al. 2003), or the lower prices of calorie-rich, nutrient-
presentation motives (Wansink and Chandon 2006).
poor foods (Hill et al. 2003).
Inferential Mechanisms. Consumers frequently draw
Consumer researchers have extensively studied biased nu-
inferences about missing attributes from the brand position-
trition inferences (e.g., Andrews, Netemeyer, and Burton
ing or from the attributes of comparable products (for a
1998; Moorman et al. 2004), but they have focused on nu-
review, see Kardes, Posavac, and Cronley [2004]). For ex-
trition evaluation and purchase decisions rather than calorie
ample, Ross and Creyer (1992) found that, if an attribute is
estimations or consumption decisions. Our health halo re-
missing, consumers rely on the same attribute information
sults also contribute to the literature on consumer trade-offs
from other brands in the same category. This suggests that
between vice and virtue goals by providing evidence (based
consumers may make inferences about the number of cal-
on real choices rather than on scenarios) that people balance
ories in a particular food from the health positioning of the
health and taste goals in single consumption episodes (e.g.,
restaurant brand or from other food items on the restaurant’s
Dhar and Wertenbroch 2000; Kivetz and Simonson 2002;
menu.
Okada 2005; Osselaer et al. 2005). More generally, our ?nd-
Selective accessibility is one of the models that can ex-
ings that healthy eaters underestimate calories more than
plain the assimilation of calorie estimations to the health
unhealthy eaters show the limits of a purely motivational
claims of the restaurant. Selective accessibility contends
perspective, which would instead predict the opposite based
that, unless consumers are speci?cally asked to consider the
on guilt or self-presentation goals.
opposite, they will spontaneously test whether the target
In this article, we start by reviewing the various inferential
food is similar to the healthy standards or to the speci?c
and self-regulatory mechanisms that may explain how health
calorie anchor advertised by the restaurant. This increases
claims in?uence calorie estimations and a consumer’s choice
the accessibility of standard-consistent information, leading
of complementary food and beverages. In one ?eld study,
to the assimilation of calorie estimations to the anchor (for
we show that calorie estimations are signi?cantly lower for
a review, see Mussweiler [2003]). Another explanation is
Subway meals than for comparable meals eaten at Mc-
provided by a Brunswikian model (e.g., Fiedler 1996),
Donald’s. These results are con?rmed in a within-subjects
which assumes that consumers normatively aggregate the
laboratory study, which also shows that nutrition involve-
information provided by the intrinsic and extrinsic cues
ment improves the accuracy of calorie estimations but does
available. In a noisy environment, extrinsic cues such as
not reduce the halo effects of health claims. A third study
quantity anchors can bias estimations even if a consumer is
shows that health claims lead consumers to unknowingly
not directly in?uenced by motivational or memory-based
order beverages and side dishes containing more calories.
biases (Chandon and Wansink 2006). Conversational norms

---

HEALTH HALOS AND FAST-FOOD CONSUMPTION
303
can also contribute to the in?uence of health claims because
consumption, from 55% to 61% of actual food intake.
consumers typically assume that the advertised information
McKenzie et al. (2002) manipulated guilt and self-presen-
is required by law to be truthful and would therefore see
tation motives by using either an obese interviewer or one
no reason not to draw inferences from it (Johar 1995).
with a normal weight to conduct in-person food intake in-
terviews. They found that the body mass of the interviewer
Self-Regulatory Mechanisms. Two con?icting goals
had no impact on food intake estimations. Given these re-
are salient when making food consumption decisions: the
sults, we expect that calorie estimations are primarily driven
hedonic goal of taste enjoyment and the more utilitarian goal
by inferential mechanisms and are thus assimilated toward
of maintaining good health (Dhar and Simonson 1999; Fish-
the health claims made by the restaurant.
bach, Friedman, and Kruglanski 2003). Many studies have
shown that health primes can activate different consumption
How Health Claims In?uence Complementary
goals. Priming hedonic goals and concepts, such as sweetness,
Food Decisions
increases the intensity of desire for hedonic food (such as
cookies) and leads consumers to choose this better-tasting but
Complementary food decisions are those pertaining to the
less healthy option over a less tasty but healthier option (e.g.,
choice of side orders, drinks, or desserts ordered following
Ramanathan and Menon 2006; Shiv and Fedorikhin 1999).
one’s choice of a main course (Dhar and Simonson 1999).
Health primes can also in?uence guilt and self-presentation
Existing research has only examined the effects of health
goals. Okada (2005) found that restaurant diners were more
claims on the choice and consumption of the advertised food,
likely to order “Cheesecake deLite,” a relatively healthy des-
and its evidence is mixed. Kozup et al. (2003) found that
sert, than “Bailey’s Irish Cream Cheesecake,” a relatively
adding a “heart-healthy” claim to a menu increased con-
unhealthy dessert, when they were presented side by side on
sumers’ intentions to order the food. However, Raghunathan,
the menu but preferred the unhealthy dessert to the healthy
Naylor, and Hoyer (2006) found that labeling food as
one when each was presented alone. She attributes these ?nd-
“healthy” reduced the likelihood that it would be chosen
ings to the fact that joint presentation increases guilt and the
because of negative taste inferences. Other studies have
dif?culty of social justi?cation.
found that the preference for healthy foods depends on the
The effects of health primes on goal activation and guilt
degree of ego depletion (Baumeister 2002), cognitive load
predict a contrast effect for calorie estimation rather than
(Shiv and Fedorikhin 1999), guilt and the need for justi-
the assimilation effect predicted by inferential mechanisms.
?cation (Okada 2005), individual differences in body mass
To reduce their feelings of guilt and to justify their activated
(Wansink and Chandon 2006), comparison frames (Wan-
hedonic goal, consumers should report lower calorie esti-
sink 1994), and the accessibility of chronic hedonic goals
mations in the unhealthy prime condition than in the healthy
(Ramanathan and Menon 2006; Ramanathan and Williams
prime condition. Supporting this argument, studies in nu-
2007).
trition and epidemiology have found that the individual trait
In contrast, the evidence regarding the effects of health
of fear of attracting a negative evaluation is correlated with
claims on complementary food decisions is more consistent.
the tendency to underreport calories (Tooze et al. 2004).
In a series of vignette studies, Dhar and Simonson (1999)
found that consumers predict that people prefer to balance
Hypotheses. Support for the inferential arguments can
an unhealthy main course with a healthy dessert, or a healthy
be found in the many studies showing that consumers gen-
main course with an unhealthy dessert, rather than choosing
eralize health claims inappropriately (Balasubramanian and
two healthy or unhealthy main courses and desserts. Fish-
Cole 2002; Garretson and Burton 2000; Keller et al. 1997;
bach and Dhar (2005) found that increasing perceived prog-
Moorman 1996). For example, Andrews et al. (1998) found
ress toward the goal of losing weight activates the hedonic
that consumers believe that foods low in cholesterol are also
taste goals and increases the likelihood that people choose
low in fat, and consumers eating an energy bar they believed
a chocolate bar over an apple. Guilt is one of the expla-
to contain soy rated it higher in nutritional value but lower
nations why consumers tend to balance health and taste goals
in taste (Wansink 2003). These halo effects also apply to
within a single consumption episode. Ramanathan and Wil-
restaurant menus. Kozup, Creyer, and Burton (2003) found
liams (2007) found that some consumers are able to launder
that adding a “heart-healthy” sign on a menu reduced the
the guilt created by their choice of an indulgent cookie by
perceived risk of heart disease when objective nutritional
choosing the utilitarian option in a subsequent choice. We
information was absent, even though it was placed next to
therefore expect that, once the choice of the main course
an objectively unhealthy menu item (lasagna).
has been made, consumers will choose side orders, desserts,
In contrast, the few studies attempting to manipulate mo-
and beverages containing more calories if the main course
tivational factors have found little impact on calorie esti-
is positioned as healthy than if it is not.
mations. Muhlheim et al. (1998) directly manipulated guilt
and self-presentation motives through a “bogus pipeline”
Moderating Factors
procedure, which consisted of warning some of the study
participants that the accuracy of their calorie estimations
Clearly, not all consumers base their food consumption
would be objectively assessed. They found that the bogus
decisions on health or nutrition considerations. One might
pipeline manipulation only slightly increased self-reported
expect that consumers highly involved in nutrition would

---

304
JOURNAL OF CONSUMER RESEARCH
be more knowledgeable about it and less likely to be in?u-
contained in their meal, and we then compared their esti-
enced by health claims (Wansink 2005). Yet, past research
mates to the actual calorie content of the meals. Study 1
suggests that nutrition involvement may not moderate the
was conducted on 9 weekdays in three medium-sized Mid-
effects of health claims. Moorman (1990) found that nutri-
western U.S. cities. As they completed their meal, every
tion involvement increases the self-assessed ability to pro-
fourth person was systematically approached and asked if
cess nutrition information but does not improve nutrition
they would answer some brief questions for a survey. No
comprehension or the nutrition quality of food choices in
mention was made of food at that point. During this process,
two product categories. Two studies (Andrews, Burton, and
the interviewer unobtrusively recorded the type and size of
Netemeyer 2000; Andrews et al. 1998) found that objective
the food and drinks from the wrappings left on the person’s
nutrition knowledge improves the accuracy of some nutri-
tray. In case of uncertainty (e.g., to determine if the beverage
tion evaluations but does not signi?cantly reduce erroneous
was diet or regular), the interviewer asked for clari?cation
inferences across nutrients or the effectiveness of objective
from the respondents.
nutrient information in reducing these overgeneralizations.
Nutrition information provided by the restaurants was
More generally, studies have found that association-based
then used to compute the actual number of calories of each
errors, such as those resulting from priming, cannot be cor-
person’s meal. Of the 392 people who were approached
rected by increasing incentives and the degree of elaboration
while they were ?nishing a Subway meal, 253 (65%) agreed
(Arkes 1991). In fact, Johar (1995) found that highly in-
to participate. Of the 379 people who were approached while
volved consumers are more likely to be deceived by implied
they were ?nishing a McDonald’s meal, 265 (70%) agreed
advertising claims because involvement increases the like-
to participate.
lihood of making invalid inferences from incomplete-com-
To pretest the health positioning of McDonald’s and Sub-
parison claims, such as “this brand’s sound quality is better.”
way, we asked 49 regular customers of both restaurants who
Chapman and Johnson (1999) showed that cognitive elab-
were eating at Subway or McDonald’s to indicate their agree-
oration, one of the consequences of involvement, actually
ment with the sentence: “The food served here is healthy”
enhances anchoring effects because it facilitates the selective
on a nine-point scale anchored at 1 p strongly disagree and
retrieval of anchor-consistent information. For these reasons,
9 p strongly agree. As expected, Subway meals were rated
we expect that nutrition involvement increases the overall
as signi?cantly more healthy (M p 6.2) than McDonald’s
accuracy of calorie estimations but does not moderate the
meals (M p
;
2.4 F(1, 49) p
,
80 p ! .001).
effects of health claims on calorie estimations and on com-
plementary food decisions.
Results
How can health halos be reduced? If calorie inferences are
partly caused by priming and selective activation, one solution
To increase the comparability of McDonald’s and Subway
is to encourage consumers to question the validity of the
meals, we restricted the analysis to the meals consisting of
health prime. Drawing attention to the priming source reduces
a sandwich, a soft drink, and a side order. This yielded a
priming effect even if the activation of information in memory
total of 320 meals (193 for McDonald’s and 127 for Sub-
occurred nonconsciously (Strack et al. 1993). The effective-
way). To test the hypothesis that calorie estimations are
ness of the debiasing strategy is enhanced if people are asked
lower for Subway than for McDonald’s meals containing
to consider evidence inconsistent with the prime. Mussweiler,
the same number of calories, we estimated the following
Strack, and Pfeiffer (2000), working on the estimation of the
regression via ordinary least squares:
value of a used car, showed that instructing people to consider
whether a claim opposite to the one primed may be true
ESTCAL p a + b # HEALTHCLAIM + d # ACTCAL
increases the accessibility of claim-inconsistent knowledge
and therefore reduces selective-accessibility biases.
+ l # HEALTHCLAIM # ACTCAL + ?,
In summary, we predict that health claims reduce calorie
estimations for the main dishes served by fast-food restaurants
(1)
and lead consumers to order high-calorie complementary food
or drinks. We also expect that asking consumers to consider
where ESTCAL is the estimated number of calories,
whether opposite health claims may be equally valid elimi-
HEALTHCLAIM is a binary variable taking the value of
nates the effects of health halos on main-dish calorie esti-
1/2 for Subway meals and ?1/2 for McDonald’s meals,
mation and side-dish choices. We test these predictions in one
ACTCAL is the mean-centered actual number of calories
?eld study and in three laboratory experiments.
of the meals, and ? is the error term. We included ACTCAL
as a covariate because consumers tend to underestimate the
STUDY 1: CALORIE ESTIMATIONS BY
calories of large meals (Chandon and Wansink 2007) and
because McDonald’s meals tend to be bigger than Subway
SUBWAY AND MCDONALD’S DINERS
meals.
Method
As expected, the coef?cient for HEALTHCLAIM was neg-
ative and statistically signi?cant (b p ?151, t p ?3.6, p !
We asked consumers who had just ?nished eating at
.001). These participants believed that the meals from Subway
McDonald’s or Subway to estimate the number of calories
contained an average of 151 fewer calories than a same-calorie

---

HEALTH HALOS AND FAST-FOOD CONSUMPTION
305
FIGURE 1
STUDY 1: CALORIE ESTIMATIONS OF SUBWAY AND MCDONALD’S DINERS
meal at McDonald’s. The regression parameters enable us to
Discussion
predict that, for a meal containing 1,000 calories, the mean
calorie estimation will be 744 calories for someone eating at
Study 1 examines the general health halo that leads people
McDonald’s and only 585 calories (21.3% lower) for some-
to believe that a 1,000-calorie Subway meal contains 21.3%
one eating at Subway. The coef?cient for ACTCAL and
fewer calories than same-calorie McDonald’s meals. It also
for the interaction (respectively, d
shows that calorie estimations are not primarily driven by
p .29, t p 4.7, p ! .001
and l
guilt or by self-presentation goals, as this would have pre-
p ?.12, t p ? ,
.9 p p .34) indicated that consumers
tended to underestimate calories more signi?cantly for large
dicted lower-calorie estimations by McDonald’s customers
meals than for small meals but that the effect of meal size is
than by Subway customers. These results nonetheless raise
two important questions that need to be addressed in sub-
similar for both Subway and McDonald’s meals. The same
sequent studies. First, the results of study 1 might be caused
results were obtained when using the percentage deviation
by intrinsic differences between self-selected Subway and
([estimated ? actual]/actual) as the dependent variable (b p
?
McDonald’s diners.1 A second issue is that participants in
19.2, t p ?
,
3.9 p !
;
.001 d p ?
,
.06 t p ?
,
7.8 p !
;
.001
study 1 evaluated only one McDonald’s or Subway meal.
and l p ?
,
.03 t p ?
,
1.8 p p .0 ),
7 indicating that the mean
Their estimations might have been better calibrated if they
percentage deviation is more negative (more biased) for Sub-
had been asked to make multiple estimates or asked to com-
way meals than for McDonald’s meals containing the same
pare meals instead of estimating a single meal. This is be-
number of calories.
cause consumers pay more attention to hard-to-evaluate at-
To illustrate the effects of health claims on calorie esti-
tributes (such as calories) in joint evaluations than in
mations for comparable meals, we computed the mean cal-
separate evaluations (Hsee 1996).
orie estimate for small, medium, and large meals (catego-
We address these issues in study 2 by using a within-
rized on the basis of actual number of calories). As shown
in ?gure 1, mean calorie estimates were lower for Sub-
1To explore this issue, we recontacted 58 participants who provided their
way meals than for comparable McDonald’s meals in each
telephone numbers and asked them to report their height and weight, which
size tier (for small meals, 473 vs. 563 calories, F(1, 106)
we used to compute their body mass index (BMI). Although we found no
p
difference in body mass (M p 23.4kg/m2 for McDonald’s customers vs.
4.0, p ! .05; for medium meals, 559 vs. 764 calories,
M p 23.6 kg/m2 for Subway customers, F(1, 56) p .1, p p .76), we can-
F(1, 105) p 9.1, p ! .01; and for large meals, 646 vs. 843
not rule out that the groups may be different on other dimensions, such
calories, F(1, 103) p
,
4.1 p !
)
.05 .
as involvement in nutrition.

---

306
JOURNAL OF CONSUMER RESEARCH
subjects design in which respondents estimate the calories
involvement group (similar results were obtained when us-
contained in two small and large Subway and McDonald’s
ing the continuous scale). We included all two-way and
sandwiches containing the same number of calories. Study
three-way interactions. Because the order of estimations had
2 also enables us to examine whether nutrition involvement
no effect on calorie estimations and did not interact with
can mitigate the biasing effects of health claims on calorie
any of the other factors, we excluded this factor from the
estimations.
analysis reported here.
The main effects of HEALTHCLAIM and ACTCAL and
STUDY 2: CAN NUTRITION
their interactions were all statistically signi?cant (respec-
tively, F(1, 314) p 158, p ! .001; F(1, 314) p 468, p !
INVOLVEMENT MITIGATE THE HALO
.001; and F(1, 314) p 72.5, p ! .001). As shown in ?gure
EFFECTS OF HEALTH CLAIMS ON
2, calorie estimations were lower for Subway sandwiches
CALORIE ESTIMATIONS?
than for McDonald’s sandwiches that contained the same
number of calories. Furthermore, the halo effects of health
Method
claims were stronger for the sandwiches containing 600 cal-
ories (M p ?200 calories, a 33% underestimation) than for
Study 2 used a 2 (health claims: Subway vs. McDon-
smaller sandwiches containing 330 calories (M p ?80 cal-
ald’s) # 2 (actual number of calories: 330 vs. 600) within-
ories, a 24% underestimation). In addition, the main effect
subjects design. It was conducted among University of Il-
of nutrition involvement and its interaction with ACTCAL
linois students and staff members, who were given the
were both statistically signi?cant (respectively, F(1, 314) p
opportunity to win a series of raf?e prizes in exchange for
9.8, p ! .01 and F(1, 314) p
,
6.1 p !
)
.05 , indicating that
their participation. We asked 316 of these consumers who
respondents highly involved in nutrition had higher (more
had eaten at least three times at Subway and McDonald’s
accurate) calorie estimations, especially for the larger sand-
in the previous year to estimate the number of calories con-
wiches. As also expected, the interaction between NUTINV
tained in two Subway sandwiches (a 6-inch ham and cheese
and HEALTHCLAIM and the three-way interaction were not
sandwich containing 330 calories and a 12-inch turkey sand-
statistically signi?cant (respectively, F(1, 314) p .9, p p
wich containing 600 calories) and in two McDonald’s burg-
.34 and F(1, 314) p
,
.4 p p
)
.55 . This indicates that nutri-
ers (a cheeseburger containing 330 calories and a Big Mac
tion involvement did not reduce the biasing effects of the
containing 600 calories). The ordering of the restaurants was
restaurant brands’ health positioning on consumers’ calorie
counterbalanced across participants. Unlike in study 1, in
estimations.
which participants had ordered and consumed the food, par-
ticipants in study 2 knew that they would not consume the
food.
Discussion
To measure their nutrition involvement, we used a ?ve-
Study 2 shows that even consumers familiar with both
item scale and asked respondents to indicate their agreement
restaurants estimate that Subway sandwiches contain sig-
with these statements: “I pay close attention to nutrition
ni?cantly fewer calories than McDonald’s sandwiches con-
information,” “It is important to me that nutrition infor-
taining the same number of calories. Study 2 therefore rep-
mation is available,” “I ignore nutrition information” (re-
licates the ?ndings from study 1 in a repeated-measures
verse coded), “I actively seek out nutrition information,”
context. The within-subjects design of study 2 allows us to
and “Calorie levels in?uence what I eat” on a nine-point
rule out the alternative explanation that the results of study
scale anchored at 1 p strongly disagree and 9 p strongly
1 were caused by self-selection or by unobserved differences
agree. The mean, median, and standard deviation of the scale
in the type of meals consumed in the two restaurants. Study
were, respectively, 4.6, 4.5, and 2.1. After verifying the
2 also shows that, although nutrition involvement improves
reliability (a p .85) and unidimensionality of the scale
the quality of calorie estimations, it does not reduce the halo
(62% of the variance was extracted by the ?rst principal
effects of the restaurant brand’s health positioning.
component), we averaged the responses to the ?ve items
Taken together, studies 1 and 2 provide converging evi-
and categorized respondents into a low or high nutrition
dence that Subway and McDonald’s health claims bias con-
involvement group via a median split.
sumers’ calorie estimations. In study 3, we examine the
effects of these claims on consumers’ complementary food
Results
decisions. This also allows us to test the alternative expla-
nation that the results of studies 1 and 2 are caused by simple
We analyzed the data using a repeated-measures ANOVA
response scaling biases, that is, that the health positioning
with two within-subjects factors and one between-subject
of Subway and McDonald’s in?uenced only consumers’ cal-
factor. The two within-subject factors were HEALTH-
orie ratings, not their general estimation of the healthiness
CLAIM (which indicates whether food was from Subway
of the food. This would predict that health claims would
or McDonald’s) and ACTCAL (which measured the ac-
have no impact on the decision to choose low- or high-
tual number of calories of the food—330 or 600 calories).
calorie side orders and drinks. Finally, by collecting calorie
The between-subject factor was NUTINV, which indicates
estimation data after the consumption decision task, study
whether respondents belonged to the high or low nutrition
3 tests whether health claims in?uence side-dish purchase

---

HEALTH HALOS AND FAST-FOOD CONSUMPTION
307
FIGURE 2
STUDY 2: HOW NUTRITION INVOLVEMENT INFLUENCES CALORIE ESTIMATIONS FOR SUBWAY AND MCDONALD’S SANDWICHES
intentions even when people are not explicitly asked to es-
to estimate the number of calories contained in their sand-
timate the caloric content of their main dishes.
wich, beverage, and cookies. Finally, we measured how im-
portant eating healthily is to them by asking them to indicate
STUDY 3: CAN HEALTH CLAIMS LEAD
their agreement with three sentences (“Eating healthily is
CONSUMERS TO UNKNOWINGLY
important to me,” “I watch how much I eat,” and “I pay
attention to calorie information”) on a nine-point scale an-
CHOOSE HIGHER-CALORIE SIDE
chored at 1 p strongly disagree and 9 p strongly agree.
ORDERS AND DRINKS?
Method
Results
Forty-six undergraduate students were recruited on the
We ?rst examine the total number of calories contained
campus of Northwestern University and were paid $2 to
in the beverages and cookies that were ordered in the Sub-
participate in this and another unrelated study. Half were
way and McDonald’s coupon condition. Compared to those
given a coupon for a McDonald’s Big Mac sandwich, and
who had received a Big Mac coupon, participants who re-
the other half were given a coupon for a Subway 12-inch
ceived the Subway coupon were less likely to order a diet
Italian BMT sandwich. To provide a more conservative test
soda, more likely to upgrade to a larger drink, and more likely
of the effects of health claims on consumption decisions,
to order cookies. As a result, participants receiving a Subway
the “healthy” food used in study 3 has actually 50% more
coupon ordered side dishes and beverages containing more
calories than the “unhealthy” food (a 12-inch Subway Italian
calories (M p 111 calories) than participants receiving
BMT sandwich has 900 calories, and a Big Mac has 600
a McDonald’s coupon (M p 48 calories; F(1, 44) p 4.0,
calories).
p ! .05; see ?g. 3). Because the Subway sandwich also con-
We then gave the participants a menu and asked them to
tained more calories than the McDonald’s sandwich, partic-
indicate what they would like to order with their sandwich,
ipants ended up with a meal containing 56% more calories
if anything. The menu included a small, medium, or large
(M p 1,011 calories) in the Subway coupon condition than
regular fountain drink (containing 155, 205, and 310 cal-
in the McDonald’s coupon condition (M p 648 calories;
ories, respectively); a small, medium, or large diet fountain
F(1, 44) p 132.9, p ! .001).
drink containing no calories; and one or two chocolate chip
We now examine whether participants receiving the Sub-
cookies (containing 220 calories per cookie). These items
way coupon realized they were ordering calorie-rich side
were chosen because they are the only side orders common
orders and whether they ended up with a much larger com-
to both McDonald’s and Subway. We then asked participants
bined meal than those receiving the McDonald’s cou-

---

308
JOURNAL OF CONSUMER RESEARCH
FIGURE 3
STUDY 3: HOW SUBWAY AND MCDONALD’S COUPONS INFLUENCE THE ESTIMATED AND ACTUAL NUMBER
OF CALORIES (FOR THE MAIN SANDWICH, SIDE ORDERS, AND THE WHOLE MEAL)
pon. As shown in ?gure 3, calorie estimations for the side
number of calories of the main sandwich. When entered
orders were similar for participants with the Subway coupon
alone in a regression of the actual number of calories con-
(M p 48 calories) and for participants with the Big Mac
tained in side dishes, the parameter of the binary variable
coupon (M p 43 calories; F(1, 44) ! .1, p p .43). Simi-
capturing the coupon manipulation was statistically signif-
larly, calorie estimations for the main sandwich were sim-
icant (B p 63.3, t p 2.0, p ! .05). However, this parameter
ilar in both conditions (M p 439 calories for the 12-inch
becomes insigni?cant when the calorie estimation bias is
Subway sandwich vs. M p 557 calories for the Big Mac;
entered in the regression as a covariate (B p 23.7, t p
,
.6
F(1, 44) p 2.4, p p .13). As a result, calorie estimations
p p .56). A Sobel test shows that the mediation effect is
for the total meal were similar in the healthy prime con-
statistically signi?cant (z p 2.32, p ! .05). Of course, this
dition (M p 487 calories) and in the unhealthy prime con-
analysis cannot rule out the opposite causality link, that is,
dition (M p 600 calories; F(1, 44) p 1.9, p p .17). Be-
that participants adjusted their main-dish calorie estimations
cause the actual number of calories of the meal was
to justify their side-dish orders. In contrast, the analysis of
signi?cantly higher in the Subway (healthy prime) con-
the healthy eating data shows that health claim manipulation
dition than in the McDonald’s (unhealthy prime) condition,
did not activate the goal of eating healthily. Respondents were
the calorie underestimation was signi?cantly larger in the
as likely to agree with the three sentences (“Eating healthily
healthy prime condition (M
p
is important to me,” “I watch how much I eat,” and “I pay
(est.?act. cal.)
?524 calories, a
52% underestimation) than in the unhealthy prime condi-