ArticlePDF Available

Is Obesity Caused by Calorie Underestimation? A Psychophysical Model of Fast-Food Meal Size Estimation

Authors:

Abstract and Figures

Calorie underestimation is often alleged to contribute to obesity. By developing a psychophysical model of meal size estimation, the authors show that the association between body mass and calorie underestimation found in health science research is a spurious consequence of the tendency of high-body-mass people to choose--and thus estimate--larger meals. In four studies involving consumers and dieticians, the authors find that the calorie estimations of high- and low-body-mass people follow the same compressive power function; that is, they exhibit the same diminishing sensitivity to meal size changes as the size of the meal increases. The authors also find that using a piecemeal decomposition improves calorie estimation and leads people to choose smaller, but equally satisfying, fast-food meals. The findings that biases in calorie estimation are caused by meal size and not body size have important implications for allegations against the food industry and for the clinical treatment of obesity.
Content may be subject to copyright.
Journal of Marketing Research
Vol. XLIV (February 2007), 84–99
84
© 2007, American Marketing Association
ISSN: 0022-2437 (print), 1547-7193 (electronic)
*Pierre Chandon is Assistant Professor of Marketing, INSEAD (e-mail:
pierre.chandon@insead.edu). Brian Wansink is John S. Dyson Chair of
Marketing and of Nutritional Science, Applied Economics and Manage-
ment Department, Cornell University (e-mail: Wansink@Cornell.edu). The
authors thank Jill North, James E. Painter, and the American Association of
Diabetes Educators for help with data collection. Helpful comments on
various aspects of this research were provided by the anonymous JMR
reviewers; John Lynch; Brian Sternthal; Paul Bloom; Priya Raghubir;
Aradhna Krishna; Alex Chernev; Miguel Brendl; Gita Johar; Chris Moor-
man; and those who participated when the authors presented this research
at INSEAD, Wharton, the University of Illinois, Urbana–Champaign,
Kellogg, the University of North Carolina, Chapel Hill, Duke, New York
University, the University of Florida, Gainesville, and the University of
California, Berkeley.
PIERRE CHANDON and BRIAN WANSINK*
Calorie underestimation is often alleged to contribute to obesity. By
developing a psychophysical model of meal size estimation, the authors
show that the association between body mass and calorie
underestimation found in health science research is a spurious
consequence of the tendency of high-body-mass people to choose—and
thus estimate—larger meals. In four studies involving consumers and
dieticians, the authors find that the calorie estimations of high- and low-
body-mass people follow the same compressive power function; that is,
they exhibit the same diminishing sensitivity to meal size changes as the
size of the meal increases. The authors also find that using a piecemeal
decomposition improves calorie estimation and leads people to choose
smaller, but equally satisfying, fast-food meals. The findings that biases
in calorie estimation are caused by meal size and not body size have
important implications for allegations against the food industry and for
the clinical treatment of obesity.
Is Obesity Caused by Calorie
Underestimation? A Psychophysical Model
of Meal Size Estimation
1Following the guidelines of the World Health Organization, people are
classified as overweight if their body mass index (BMI) is greater than 25
and obese if their BMI is greater than 30. Body mass index is computed as
the ratio of weight, measured in kilograms, to squared height, measured in
meters.
Sixty-five percent of U.S. adults are either obese or over-
weight (Hedley et al. 2004).1Many policy makers and con-
cerned consumer groups have alleged that this epidemic is
being fueled by a combination of increasing portion sizes in
restaurant meals (Brownell and Horgen 2003; Nestle 2002;
Nielsen and Popkin 2003; Seiders and Petty 2004; Young
and Nestle 2002) coupled with “a virtual absence of intui-
tive understanding that larger portions contribute more calo-
ries” (Nestle 2003, p. 40). Because of the scale of this issue,
the food industry in general and fast-food restaurants in par-
ticular are being increasingly threatened by litigation, taxes,
and restrictions that promise to make it “the tobacco indus-
try of the new millennium” (Brownell and Horgen 2003;
Wansink and Huckabee 2005). The general question being
asked is, Is obesity really caused by the underestimation of
the number of calories contained in large fast-food meals,
and what can policy makers, food companies, and health
professionals do about it?
Evidence linking calorie underestimation and obesity is
strong and comes from health science research that com-
pares actual caloric intake (measured using “doubly labeled
water” [DLW] biomarkers) with self-reported estimates of
intake (measured in calories, volume, or frequency) for
people with high and low body masses (Lansky and
Brownell 1982; Livingstone and Black 2003; Tooze et al.
2004). In a pioneering DLW study, Lichtman and col-
leagues (1992) conclude that calorie underreporting is part
of the explanation for the failure to lose weight. In a meta-
analysis of 87 studies, Livingstone and Black (2003) find a
–.25 correlation between a person’s body mass index (BMI)
and the ratio of estimated to actual food intake, indicating
that people with a high BMI are significantly more prone to
underestimations than people with a low BMI.
We challenge the contention of dozens of studies in the
health sciences by hypothesizing that the evidence linking
BMI and calorie underestimation may be a spurious conse-
Is Obesity Caused by Calorie Underestimation? 85
quence of the tendency of high-BMI people to choose—and
thus estimate—larger meals. We show this by developing a
psychophysical model of meal size estimation, which
hypothesizes that estimation biases are caused by the size of
the meal, regardless of the body mass of the person doing
the estimation. Specifically, we hypothesize that the estima-
tions of low- and high-BMI people follow the same com-
pressive power function of actual meal size (i.e., increase at
a slower rate than actual changes in meal size). Thus,
people are more likely to underestimate the number of calo-
ries of larger meals than the number of calories of smaller
meals.
With the proposed psychophysical model of meal size
estimation, we address three unresolved issues. First, we
predict and find that after the natural association between
body mass and meal size is eliminated, low- and high-BMI
people have identical estimations. This suggests that the
higher body mass of overweight consumers is not caused by
their supposed tendency to underestimate the number of
calories of today’s large fast-food meals. It also rules out a
common assumption among dieticians that overweight
people underestimate their consumption because of motiva-
tional biases, such as denial or impression management
(Muhlheim et al. 1998).
Second, we predict and find that educating people about
meal size estimation biases and encouraging them to count
calories accurately does not reduce psychophysical biases.
This explains why general nutrition education efforts such
as the Food and Drug Administration’s “Count Calories”
campaign have shown insignificant results (Seiders and
Petty 2004). In comparison, our model predicts that a piece-
meal decomposition (Menon 1997; Srivastava and Raghubir
2002), in which people estimate the size of each component
of the meal rather than the size of the overall meal, reduces
psychophysical biases. As expected, we find that piecemeal
decomposition improves the accuracy of the estimations of
regular consumers and even those of certified dieticians and
that these improvements lead people to avoid choosing
unnecessarily large fast-food meals.
Third, and more general, we provide insights into con-
sumer research that directly address a pressing health sci-
ence question as to how people estimate intake and why it is
often, but not always, underestimated. This has been an
important question for researchers in marketing, epidemiol-
ogy, and nutrition, who, because of the prohibitive costs of
biomarker techniques, must rely on self-reports of con-
sumption. It is also a question that is poorly understood,
despite the considerable progress made in understanding
how people process and respond to changes in nutritional
information (Andrews, Netemeyer, and Burton 1998; Bala-
subramanian and Cole 2002; Moorman et al. 2004) and
product quantity (Chandon and Wansink 2002; Folkes and
Matta 2004; Krider, Raghubir, and Krishna 2001; Raghubir
and Krishna 1999; Wansink and Van Ittersum 2003). As
recent literature reviews have stated, “more fundamentally
still, we need to understand why people misreport food
intake” (Livingstone and Black 2003, p. 915S), and “our
inability to obtain good information on food intake is a
dilemma for nutrition and an enigma for psychology”
(Blundell 2000, p. 3).
We organize this article as follows: We begin by develop-
ing a psychophysical model of how people estimate the size
of fast-food meals and show that it provides a parsimonious
explanation of previously unresolved findings and an effec-
tive debiasing technique. In Study 1, we ask consumers
with a low or high body mass to estimate the number of
calories of eight fast-food meals of varying sizes, and we
show that the estimations of both groups follow the same
power function. In Study 2, we ask consumers to estimate
the number of calories of the fast-food meal that they would
choose to eat, and we compare the debiasing effectiveness
of the proposed piecemeal decomposition technique with
that of a typical disclosure-and-incentive technique. In
Study 3, we show that the estimations of people with a low
or high BMI and with a low or high involvement in nutri-
tion follow the same power function, even when they are
measured in the field immediately after people have fin-
ished consuming their fast-food meal. In Study 4, we exam-
ine certified dieticians’ own calorie estimations, their fore-
casts of the estimations of high- and low-BMI people, and
the effects of piecemeal decomposition on their calorie esti-
mation and fast-food consumption decisions. In the final
section, we discuss the implications of our findings for con-
sumption research and public policy.
A PSYCHOPHYSICAL MODEL OF MEAL SIZE
ESTIMATION
In this section, we review the psychophysics literature on
area and volume estimations. We then develop a model of
meal size estimation and examine its implications for the
current debate on the association between BMI and the
calorie underestimation bias. Our focus is on how people
(consumers and dieticians) estimate the number of calories
of fast-food meals because such food have been repeatedly
held responsible for the increasing obesity rates (Paeratakul
et al. 2003). In addition, unlike for packaged goods,
serving-size and calorie information are not mandatory for
the food served in fast-food restaurants. Therefore, con-
sumers cannot simply read meal size information or retrieve
this information from memory and must estimate it from
the actual size of the meal.
The objective of the model is not to describe how con-
sumers spontaneously estimate the size of their fast-food
meals—something that probably few consumers do—but
rather to test the argument that calorie underestimation is
one of the primary causes of obesity, an argument that is
backed by the evidence that high-BMI people tend to under-
estimate their calorie intake. Rather than studying whether
people with a tendency to underestimate meal size gain
more weight over time, we test one logical implication of
this argument: Are overweight people more likely to under-
estimate meal size than people who are of a lower weight?
The Power Law of Sensation
The “empirical law of sensation” (Stevens 1986)
describes the relationship between objective and subjective
magnitudes. It states that a percentage change in objective
magnitude leads to the same percentage change in subjec-
tive magnitude. For example, the subjective impact of
adding 100 calories to a meal depends on the size of the
meal; the difference between 100 and 200 calories is sub-
jectively different from that between 500 and 600 calories.
In contrast, the subjective impact of doubling the number of
calories of a meal is constant, regardless of the size of the
meal. The psychophysical function consistent with the
empirical law of sensation is a power function (S = aIβ),
where S is the subjective magnitude (or sensation), I is the
objective magnitude (or intensity), and ais a positive scal-
ing parameter. The exponent βof the power function cap-
tures its concavity. If β< 1, the power function is compres-
sive; that is, estimations are inelastic (they increase at a
slower rate than do actual magnitudes), and people become
more likely to underestimate objective magnitudes as they
increase. As a result, small intensities (below I* = a1/(1 – β))
are likely to be overestimated and are assimilated upward
toward I*, whereas large intensities (above I*) are likely to
be underestimated and are assimilated downward toward I*.
With a few exceptions (e.g., with the perception of the
intensity of electric shocks), sensations are always compres-
sive. As Krueger (1989, p. 264) states, “the true psy-
chophysical function is approximately a power function
whose exponent normally ranges from 0 to 1, and exceeds 1
only in rather special cases.” For example, people get the
impression that a second candle adds less brightness than a
first candle. This finding is robust across various estimation
tasks and measures (Chandon and Wansink 2007). Krueger
shows that the compressive power function holds whether
sensation is measured directly in a magnitude estimation
task or in a category rating task (e.g., a seven-point Likert
scale) or indirectly in an incremental detection task.
In the domain of size estimations, Teghtsoonian (1965)
finds that the exponent of the power function is approxi-
mately .8 when estimating two-dimensional objects and
approximately .7 when estimating three-dimensional
objects. Frayman and Dawson (1981) examine exponents of
power functions for different shapes (cubes, spheres, octa-
hedrons, cylinders, tetrahedrons) and find that they are all
approximately .6. In a review of psychophysics research on
size perception, Krishna (2005, p. 22) states that “the expo-
nent range of .5–1.0 appears fairly robust and generalizable
across shapes of the same dimensionality.
Implications for Meal Size Estimations
Drawing on the psychophysics literature, we hypothesize
that estimations of the size of a meal follow a compressive
power function of the actual size of the meal (i.e., a power
function with an exponent lower than 1). Drawing on the
robustness findings and in the absence of a theory suggest-
ing otherwise, we expect that the psychophysical function
holds, regardless of four factors: whether people have a low
or high BMI, whether the meal size is estimated before or
after intake, whether the meal size is self-selected, or
whether size is measured in calories, ounces, cups, or any
other volume unit.
The hypothesis that meal size estimations follow a com-
pressive power function leads to four testable predictions.
The first is that underestimations become more likely and
increase in magnitude as the size of the meal increases,
even when the magnitude of the bias is measured propor-
tionally to the actual size. The magnitude of estimation
biases is typically measured as the percentage deviation
from actual magnitude (PDEV = [{estimated – actual}/
actual] = [aIβ– I]/I = aIβ– 1 – 1) or as the log ratio of esti-
mated to actual size (LOGRATIO = ln[estimated/actual] =
ln[aIβ/I] = ln[aIβ– 1]). Both measures are closely related
(LOGRATIO = ln[PDEV + 1]), and therefore we use the
more intuitive measure, PDEV, to quantify the magnitude of
bias in descriptive analyses. It is easy to understand that if
S= aI
βand β< 1, the derivatives of PDEV and LOGRATIO
86 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
with respect to I are both negative (d[PDEV]/d[I] = [β– 1]
aIβ– 2 and d[LOGRATIO]/d[I] = [β– 1]/I). Therefore, if
meal size estimations follow a compressive power function,
the magnitude of the underestimation bias increases as the
actual size of the meal increases, even if the bias is meas-
ured in proportion to the observed meal.
The second prediction of the model is that when people
are asked to estimate self-selected meals, high-BMI people
are likely to be more prone to underestimations than low-
BMI people, even though they all follow the same psy-
chophysical function (and therefore have the same intrinsic
estimation biases). This prediction relies on the well-
established notion that high-BMI people tend to select
larger meals than low-BMI people (Subar et al. 2003). In
DLW studies, people are asked to estimate their own con-
sumption of the size of the meal they select. Therefore,
high-BMI people estimate the size of larger meals, which
can lead to a stronger underestimation bias. A corollary to
our rationale is that the estimations of low- and high-BMI
people should be identical when the natural association
between body mass and meal size is statistically controlled
for or is eliminated by asking high- and low-BMI people to
estimate the size of the same meals.
The third prediction of the model is that a piecemeal
decomposition estimation procedure, in which consumers
estimate the size of each individual component of the meal
rather than the size of the whole meal, should reduce psy-
chophysical biases and should be more effective than the
typical debiasing techniques, such as disclosing information
about the bias or trying to motivate consumers to be more
accurate. This is because the piecemeal decomposition esti-
mation procedure replaces a single estimation of a large
intensity (located on the flatter portion of the curve) with
multiple estimations of smaller intensities (located on the
steeper portion of the curve, where the slope is closer to 1).
This prediction is consistent with the work of Arkes (1991),
who argues that attempting to correct psychophysical biases
through information disclosure and incentives is ineffective
because the shape of the psychophysical function is driven
by automatic, low-level perceptual processes; this has been
documented in multiple psychophysical studies (Folkes and
Matta 2004; Krider, Raghubir, and Krishna 2001; Raghubir
and Krishna 1996, 1999; Wansink and Van Ittersum 2003).
It is also consistent with Arkes’s recommendation to exploit
the shape of the psychophysical function by changing the
location of the options or the location of a person’s refer-
ence point on the curve. Note that in addition to increasing
the sensitivity to changes in meal size, the piecemeal
decomposition strategy should lead to an overall increase in
meal size estimations, regardless of the size of the meal,
because it reduces the likelihood of forgetting a component
of the meal (Bolton 2003; Menon 1997; Srivastava and
Raghubir 2002).
The fourth prediction of the model is that the mean esti-
mated meal size will be lower than the mean observed meal
size when a representative sample of meal sizes is tested,
but it will be higher than the mean observed meal sizes
when only small meals are sampled. When a representative
sample of consumers and sizes is tested, the underestima-
tions of large meals are stronger than the overestimations of
small meals. As a result, the mean estimated size is lower
than the mean observed size. However, when only a subset
of small sizes is sampled, most of them are overestimated,
Is Obesity Caused by Calorie Underestimation? 87
and the mean estimated size is higher than the mean
observed size. This is consistent with the findings of Liv-
ingstone and Black’s (2003) meta-analysis of 77 DLW stud-
ies. Because these studies were conducted with randomly
selected people, they find that, on average, the mean esti-
mated food intake is 20% below the mean observed food
intake and that it is below the mean observed intake in 67 of
the 77 groups. It can also explain the few cases in which the
mean estimation is higher than the mean observed intake;
these tend to involve people with a very low BMI (e.g.,
anorexics), children ages 6–12 years, or parents estimating
the consumption of their children ages 1–6 years (Living-
stone and Black 2003; Williamson, Gleaves, and Lawson
1991). Consistent with the fourth prediction, a common
characteristic of these three groups is that they consume
smaller quantities than the average person, thus leading to
the average overestimation bias.
STUDY 1: BIASES IN CALORIE ESTIMATIONS: MEAL
SIZE OR BODY SIZE?
The objective of Study 1 is to test our hypothesis that
meal size estimations are independent of body mass but are
related to actual meal size through a compressive power
function. To achieve this goal, we asked 55 students with
low and high body masses to estimate the size of eight
meals that contained different sizes of a sandwich, fries, and
a beverage. Because this procedure ensures that meal size
and body mass are independent, we expect to find no differ-
ences between the estimations of low- and high-BMI
consumers.
Method
Eight typical fast-food meals were displayed on two
tables. All meals included the same three items (sandwich,
fries, and a soft drink), and only the quantity of each item
was varied (the total number of calories of the meals ranged
from 190 to 1480). Participants estimated the size of each
meal, and the order in which the meals were estimated was
varied randomly across participants. The participants then
provided their height and weight, which enabled us to
divide them into a low-BMI group (participants with a
healthful weight; i.e., BMI < 25, n = 39) and a high-BMI
group (overweight participants; i.e., BMI 25, n = 16). To
reduce the motivation to engage in impression management,
estimations were fully confidential and anonymous. In addi-
tion, we motivated participants to provide accurate estima-
tions by telling them that the names of the three most accu-
rate respondents would be announced to the rest of the
group, and they would each receive a $50 gift certificate to
a local bookstore.
In Study 1, as well as in subsequent studies, we asked the
participants to estimate meal size in number of calories
rather than in other measurement units for four reasons.
First, calories are common to all foods, whereas size units
(e.g., ounces, pounds, liters, cups) are valid only for some
of the foods in the meals. Second, calories are a metric unit
and thus are easier to add than ounces or cups. This
increases our confidence that we are measuring estimation
biases rather than computation biases. Third, calories are
more relevant than meal size for nutritional purposes. This
is one reason that calories are the first nutritional informa-
tion displayed on the mandatory nutrition labels established
for packaged goods by the U.S. Nutrition Labeling and
2Measuring calorie estimations rather than directly measuring meal size
estimations has some shortcomings because calories are not a sensation
and must be inferred from meal sizes. For example, comparing the power
exponents obtained with calorie estimations with those of previous psy-
chophysical studies (which used size estimations) requires estimating the
relationship between calorie and meal size estimations. For our purposes,
however, the use of calories as a proxy for meal size is appropriate for two
reasons. First, the caloric density of the meal is held constant across meal
sizes. As a result, errors in converting meal size into calories can only shift
all estimations up or down, leaving the exponent of the power function
unchanged. Second, collecting information on calories rather than meal
size is an alternative explanation to our results only if the relationship
between calorie and meal size estimations is different for low- and high-
BMI people. We empirically examined the latter issue by asking 45 low-
and high-BMI participants to estimate both the number of calories and the
size of three fast-food meals of different sizes. The stimuli and participants
were similar to those in Studies 1, 2, and 3, and the order of each measure
was counterbalanced across participants. We estimated the following
power model: ln(S) = α+ β×ln(SIZE) + γ×BM + δ×ln(SIZE) ×BM +
ε, where S are the calorie estimations, SIZE are the size estimations, and
BM measures whether participants have a high (25) or low (<25) BMI.
As we expected, the main effect of BM and its interaction were not statis-
tically significant (γ= –.06, t = –.24, p= .81, and δ= .01, t = .29, p= .78),
indicating that the relationship between size and calorie estimations is the
same for high- and low-BMI participants. Notably, the parameter for SIZE
was not statistically different from 1 (β= .9, t-test of difference with 1 =
–1.8, p= .08). This shows that in the range of meal sizes we study herein,
calorie estimations are directly proportional to meal size estimations.
Education Act of 1990 (21 U.S.C. § 343). If the underesti-
mation of meal size is indeed an important contributor to
obesity, we should be able to detect it more easily when
measuring size in calories than in ounces. Fourth, the num-
ber of calories in any meal is large enough for the biases not
to be driven by truncation at 0.2
Results
Descriptive results. The mean estimated meal size was
448 calories, whereas the mean actual meal size was 589
calories (to be consistent with the power model, we report
geometric means for the estimated and actual number of
calories and the arithmetic means for the bias magnitude
measures DEV and PDEV). The mean underestimation bias
is strongly statistically significant, regardless of whether it
is measured in absolute value (DEV = –139 calories; t =
–8.6, p< .001) or is relative to the real number of calories
(PDEV = –11.3%; t = –4.8, p< .001, where PDEV is the
arithmetic mean of individual-level PDEVs). As we
expected, the mean estimated number of calories and the
magnitude of the underestimation bias are the same for low-
BMI participants (M = 443 calories, DEV = –139 calories,
PDEV = –11.1%) as for high-BMI participants (M = 461
calories, DEV = –140 calories, PDEV = –12.0%). To test
for differences in the estimations of low- and high-BMI par-
ticipants, we used a repeated measure analysis of variance
with one within-subjects factor (MEALSIZE: with one level
for each of the eight meal sizes), one between-subjects fac-
tor (BM: low- versus high-BMI group), and their interac-
tion. The effect of MEALSIZE was statistically significant
(F(7, 371) = 89.6, p< .001). As we expected, the effect of
BM was not statistically significant (F(1, 53) < .1, p= .98),
nor was the interaction between MEALSIZE and BM
(F(7, 371) = .6, p= .72). We obtained similar results when
using each participant’s BMI as a covariate rather than the
dichotomous categorization of participants as either low or
high BMI.
To illustrate these results, we report in Figure 1 the geo-
metric mean and confidence interval of calorie estimations
88 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
Figure 1
STUDY 1: MEAL SIZE, NOT BODY SIZE, INFLUENCES CALORIE
ESTIMATIONS (OBSERVED GEOMETRIC MEANS, 95%
CONFIDENCE INTERVAL, AND MODEL PREDICTIONS)
of low- and high-BMI participants for four groups of meals
of increasing size (from the smallest two meals to the
largest two meals). Figure 1 shows that for each meal size,
the estimations of high-BMI participants are almost identi-
cal to those of low-BMI participants. In addition, it shows
that the mean estimation of the smallest meals is located on
the accuracy line, indicating that, on average, estimations of
the smallest meals are unbiased. However, because con-
sumers are not sensitive enough to the actual increase in
meal size, the mean estimations of the two largest meals are
only 62% of actual meal size.
Model results. We conducted several analyses to test the
hypothesis that the estimations of low- and high-BMI con-
sumers follow the same compressive power function. First,
we estimated a power model for each participant by fitting
the following linearized regression:
(1) ln(S) = α+ β×ln(I) + ε,
where S is estimated calories, I is observed calories (cen-
tered on their geometric mean), εis the error term, α=
ln(a), and αand βare parameters to be estimated with ordi-
nary least square (OLS). As we expected, 85% of the
individual-level exponents were below 1, and the mean
exponent across participants was well below 1 (M = .74). In
addition, the mean exponents (across participants) were
similar in the low-BMI group (M = .75, SD = .26) and in
the high-BMI group (M = .70, SD = .23), and a t-test shows
that they were not statistically different (t = 1.14, p= .46).
In a second analysis, we estimated a power model for
low-BMI (<25) and high-BMI (25) participants. To
account for the eight observations that each individual pro-
vided, we estimated Equation 1 using a fixed-effect model
(using the XTREG procedure in STATA 8.0). The exponent
we obtained in the low-BMI group was well below 1 (β=
.75, t-test of difference from 1 = –10.1, p< .001) and was
similar to the exponent in the high-BMI group (β= .70,
t-test of difference from 1 = –9.0, p< .001). We used these
parameter values to plot predicted estimations for each
group in Figure 1. As Figure 1 shows, the two predicted
power curves are almost indistinguishable over the whole
range of meals tested.
In a third analysis, we tested whether the power expo-
nents are similar in the low- and high-BMI groups by esti-
mating the following moderated regression using data from
all participants:
(2) ln(S) = α+ β×ln(I) + γ×BM + δ×ln(I) ×BM + ε,
where BM is a binary variable capturing whether the par-
ticipant is overweight or not (BM = –.5 if BMI < 25, and
BM = .5 if BMI 25). All coefficients were in the expected
direction. The coefficient for ln(I) was statistically below 1
(β= .73, t-test of difference from 1 = –9.2, p< .001), show-
ing that the power model is compressive. The simple effect
of BM was not statistically significant (γ= .04, t = .7, p=
.48), nor was its interaction with ln(I) (δ= –.05, t = –.6, p=
.55), indicating that the curvature of the power curve is the
same for both groups. We obtained similar results when
using BMI itself instead of categorizing consumers into a
low- or high-BMI group (in which δ= –.01, t = –.7, p=
.49). Note that the lack of association cannot be explained
by a lack of statistical power. The 440 observations in the
sample are significantly more than the number (n* = 139)
needed to detect the reported association between BMI and
estimation biases (r = –.25) at a .05 (two-tailed) significance
level with the conventional .80 power level.
Finally, we compared the fit of the power model shown in
Equation 1 with that of a linear model (S = α′ + β′ × I+ ε′).
Using data from all participants, we found that the power
model has a superior fit (R2= .42, F(1, 438) = 315.7, p<
.001, Akaike information criterion = 1.61) to the linear
model (R2= .39, F(1, 438) = 276.6, p< .001, Akaike infor-
mation criterion = 14.2). We also compared the predictive
accuracy of the power and linear models by computing the
mean average percentage error (MAPE) for each model.
The power model also outperformed the linear model on
this criterion (MAPE(Power) = .49 versus MAPE(Linear) = .85;
paired t-test = 8.37, p< .001). These results rule out the
alternative explanation that meal size estimations are due to
a simple regression to the mean or to Bayesian updating,
which would both predict a linear model.
Discussion
Study 1 shows that, on average, the number of calories of
familiar fast-food meals consisting of a sandwich, fries, and
a soft drink is well underestimated. Therefore, the under-
estimation bias holds even in a context in which some
researchers (Muhlheim et al. 1998) would expect none. This
is because consumers who were making multiple estima-
tions of familiar, simple meals should not have been moti-
vated to engage in impression management. First, they were
Is Obesity Caused by Calorie Underestimation? 89
not estimating meals they chose. Second, they knew that the
accuracy of their estimations would be checked.
Second, Study 1 shows that the calorie estimations of the
same meals by overweight (BMI 25) and healthful-
weight (BMI < 25) consumers are indistinguishable and
similarly influenced by the size of the meal. Estimations of
small meals tend to be unbiased (accurate on average),
whereas those of medium and large meals are well below
the actual number. Study 1 further shows that these biases
are caused by calorie estimations that follow a compressive
psychophysical power function of the actual number of
calories of the meal.
The results of Study 1 raise the question as to why prior
research has consistently found a stronger consumption
underestimation bias among overweight people than among
people with a lower body mass. Our explanation is that
these studies are biased by the natural association between
body size and meal size. In all these studies, consumers
were asked to estimate the number of calories contained in
meals they had consumed. Because people with a higher
body mass tend to consume larger meals, their greater
underestimation is caused by the meal they chose; it is not a
function of their body mass. A second question arising from
Study 1 is whether the strong compression of calories
would hold in a between-subjects design in which people
make only one estimation of a familiar meal, the meal of
their choice. Finally, Study 1 raises the question whether
biases can be corrected. We address these three questions in
Study 2.
STUDY 2: ESTIMATION BIASES AND CORRECTIVE
PROCEDURES FOR SELF-SELECTED FAST-FOOD
MEALS
Method
In contrast to Study 1, which manipulated meal size in a
within-subjects design, participants in Study 2 first chose
the size of a sandwich, portion of fries, and soft drink they
preferred, and then they were asked to estimate the number
of calories contained in the meal they had created. In addi-
tion to this, participants in Study 2 were randomly assigned
to one of three conditions. The control condition used the
same instructions as in Study 1 and simply asked partici-
pants to estimate the calories contained in the entire meal.
In the disclosure condition, participants were informed of
the biasing effects of meal size and then were asked to esti-
mate the number of calories contained in the entire meal. In
the piecemeal decomposition estimation condition, partici-
pants were not informed of the bias but were asked to esti-
mate the number of calories contained in each component
of their meals (i.e., the sandwich, the fries, and the soft
drink). The rest of the procedure was the same as in
Study 1.
Respondents were 156 university students who partici-
pated in the study to fulfill course requirements. To com-
pare the estimations of low- and high-BMI participants in
the control condition, twice as many participants were
assigned to the control condition (n = 79) than to either the
disclosure condition (n = 41) or the piecemeal estimation
condition (n = 36). On one table, we displayed three fast-
food meals, consisting of chicken, fries, and cola purchased
at a local fast-food restaurant. The first meal (Meal A) con-
sisted of 3 chicken nuggets, 1.45 ounces of fries, and a 10-
fluid-ounce glass marked “regular cola.” Meal B consisted
of 6 chicken nuggets, 2.90 ounces of fries, and a 20-fluid-
ounce glass of regular cola. Meal C contained 12 chicken
nuggets, 5.8 ounces of fries, and a 40-fluid-ounce glass of
regular cola. The food items were presented on white paper
plates or in glasses with no information about the name of
the restaurant or about their weight or volume (with the
exception of the beverages marked “regular cola”).
Participants were asked to imagine that they were going
to order a chicken nugget meal and were asked to indicate
which size (A, B, or C) of the chicken nuggets, fries, and
beverage they would order. Participants in the control con-
dition were simply asked, “What is the total number of
calories you think are contained in the meal you selected?”
Participants in the disclosure condition read this paragraph:
“When people estimate the number of calories in the food
they select, they nearly always underestimate how many
calories are in their food. The larger the meal, the more they
underestimate. For instance, for a 300-calorie meal, people
are fairly accurate, but if it is a 1500-calorie meal, they tend
to underestimate by 30%. Knowing this, what is the total
number of calories you think are contained in the meal you
selected?” Participants in the piecemeal estimation condi-
tion were asked three questions: “What is the number of
calories contained in the (chicken nuggets, fries, and bever-
age) size that you chose?” Finally, all participants indicated
their height and weight. In the control condition, 53 partici-
pants were in the low-BMI group, and 26 were in the high-
BMI group (because of the low sample size in the disclo-
sure and piecemeal estimation, we did not distinguish
between low- and high-BMI participants in these two
conditions).
Results
Control condition. We first examined the estimations of
low- and high-BMI participants in the control condition. As
in Study 1, there was a significant general calorie underesti-
mation (MS= 808 calories versus MI= 945 calories;
PDEV = –7.7%; t = –2.0, p< .05). As Figure 2 shows, the
estimations of smaller meals (categorized as such on the
basis of a median split) were unbiased, whereas those of
larger meals were well below the actual calorie content (the
identity line). Therefore, we were able to replicate the
results of Study 1 for self-selected meals and show a
stronger underestimation bias for large meals than for small
meals.
We found the expected association between BMI and
biases in calorie estimations for self-selected meals. The
mean estimations of low-BMI participants (MS= 799 calo-
ries) and of high-BMI participants (MS= 929 calories) were
not statistically different (F(1, 77) = .5, p= .48). In reality,
however, high-BMI participants selected meals containing
246 more calories (MI= 1117 calories) than those selected
by low-BMI participants (MI= 871 calories), a strongly sta-
tistically significant difference (F(1, 77) = 13.8, p< .001).
As a result, high-BMI participants underestimated calories
(PDEV = –17.9%; t = –2.5, p< .05), whereas low-BMI par-
ticipants were unbiased (PDEV = –2.6%; t = –.6, p= .56).
The difference between the two PDEV measures was mar-
ginally statistically significant (F(1, 77) = 3.6, p< .06).
To formally test our hypothesis that the stronger under-
estimation of high-BMI participants is a spurious conse-
quence of their selection of larger meals, we estimated the
90 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
Figure 2
STUDY 2: PIECEMEAL ESTIMATION, NOT DISCLOSURE AND
INCENTIVES, REDUCES THE BIASING EFFECTS OF MEAL
SIZE ON CALORIE ESTIMATIONS (OBSERVED GEOMETRIC
MEANS, 95% CONFIDENCE INTERVAL, AND MODEL
PREDICTIONS)
model represented in Equation 2 using the same variables as
for Study 1. As we expected, the coefficient for ln(I) was
statistically lower than 1 (β= .38, t = –4.1, p< .001), show-
ing that, on average, the power curve is compressive. The
coefficient for BM was not statistically significant (γ= –.05,
t = –.5, p= .60), indicating that low- and high-BMI partici-
pants have similar estimations after we control for the
effects of meal size. In addition, the interaction between the
effects of meal size and body mass was not statistically sig-
nificant (δ= –.11, t = –.4, p= .73), indicating that the cur-
vature of the power curve is the same for both groups. Fig-
ure 2 illustrates this by showing the estimations of low- and
high-BMI participants in the control condition on the same
curve.
Correcting psychophysical biases. We now examine the
ability of the piecemeal decomposition estimation proce-
dure and of the disclosure-and-incentive procedure to
debias psychophysical biases. Following the model predic-
tions, we expect that compared with the control condition
(in which participants estimate the calories contained in the
total meal), disclosing the bias and motivating respondents
to estimate accurately will not change the shape of the psy-
chophysical function (though it may have a main effect on
calorie estimations). Conversely, we expect that the piece-
meal decomposition estimation procedure will improve the
accuracy of the estimations by making the exponent of the
power curve closer to 1. We also expect two positive main
effects on calorie estimations. Disclosing that most meal
sizes are underestimated and motivating participants to be
accurate should lead to higher calorie estimations, regard-
less of meal size. The piecemeal decomposition should also
increase calorie estimations, regardless of meal size,
because it reduces the chances that participants will forget
one component of the meal (e.g., the regular cola). We for-
mally test these predictions by estimating the following
model:
(3) ln(S) = α+ β×ln(I) + γ×DISC + δ×PCM + θ×ln(I)
×DISC + λ×ln(I) ×PCM + ε,
where S is estimated calories; I is geometric mean-centered
observed calories; DISC is a binary variable capturing the
bias disclosure manipulation (DISC = .33 for participants in
the disclosure group, and DISC = –.67 otherwise); PCM is a
binary variable capturing the piecemeal estimation manipu-
lation (PCM = .33 for participants in the piecemeal estima-
tion group, and PCM = –.67 otherwise); εis the error term;
and α, β, γ, δ, θ, and λare the parameters to be estimated
(with OLS).
As in previous analyses, the coefficient for ln(I) was sta-
tistically lower than 1 (β= .53, t = –4.9, p< .001). As we
expected, both DISC and PCM had a positive and statisti-
cally significant main effect (γ= .24, t = 3.7, p< .001, and
δ= .15, t = 2.2, p< .05, respectively), indicating that, on
average, estimations are higher in the two debiasing condi-
tions than in the control condition. As we expected, how-
ever, the interaction between DIS and ln(I) was not statisti-
cally significant (θ= .05, t = .2, p= .83), indicating that the
disclosing-and-incentive procedure did not correct the
shape of the psychophysical function. In contrast, the inter-
action between PCM and ln(I) was positive and statistically
significant (λ= .48, t = 2.1, p< .05), indicating that the psy-
chophysical function is less compressive (the exponent is
higher) for piecemeal estimations than for holistic estima-
tions. The power exponent in the piecemeal estimation con-
dition was not statistically different from 1 (β= .83, t =
–1.1, p= .28). We illustrate these results in Figure 2, which
shows that the fitted psychophysical curve in the disclosure
condition is parallel to the fitted curve in the control condi-
tion. In contrast, the mean estimations and the fitted curve
for the piecemeal estimation condition are close to the accu-
racy line. (Figure 2 does not report results for low and high
BMI separately in the two debiasing groups because of the
limited number of observations in these conditions.)
Discussion
Study 2 shows that participants with a high BMI choose
larger meals than participants with a low BMI. As a result,
when high-BMI participants are asked to estimate the size
of self-selected meals, they are more prone to underestimat-
ing the size of the meal than those with a low BMI. How-
ever, after we statistically controlled for the size of the
meal, the estimations of both groups are identical. There-
fore, Study 2 reconciles the findings from Study 1 (that
low- and high-BMI consumers have similar estimations of
Is Obesity Caused by Calorie Underestimation? 91
meal sizes) with those of previous DLW studies (that BMI
and the underestimation bias are correlated).
Study 2 also shows that estimations of meal size follow a
compressive power function, even when consumers esti-
mate familiar meals in familiar sizes. Therefore, the find-
ings from Study 1 are not caused by respondent fatigue or
context effects due to the estimation of multiple meals.
Finally, Study 2 provides support for an implication of the
psychophysical model; namely, informing consumers about
psychophysical biases and motivating them to be accurate
does not eliminate these biases. However, asking consumers
to follow a simple piecemeal estimation procedure, in
which they make multiple estimations of small sizes rather
than estimate the size of the whole meal, is an effective pro-
cedure to correct psychophysical biases.
A possible limitation of Study 2 is that it does not rule
out the alternative explanation that people have a fixed calo-
rie estimation bias, regardless of meal size and body size.
This could produce the results found in Study 2 if high-BMI
people choose larger meals but report the same number of
calories as when they choose smaller meals. However, note
that this explanation is inconsistent with Study 1, which
showed that high- and low-BMI consumers adapt their calo-
rie estimations with the size of the meal. Taken together,
Studies 1 and 2 provide strong support for the hypothesized
psychophysical model in a laboratory setting in which con-
sumers make estimations before intake, the type of food is
held constant across different meal sizes, accuracy (not
underestimation) is rewarded, and response rate is 100%.
Will these results hold in a natural setting in which
people benefit from the additional sensory experience
obtained through intake? In a natural setting, high-BMI
people might also be more motivated to underestimate the
size of the meal for self-presentation reasons. Crandall
(1994) shows that antipathy toward obese people is wide-
spread and not as socially stigmatized as other forms of
prejudice. Similarly, it is possible that high-BMI people,
who have a more accurate estimation of the size of their
meals, may feel too embarrassed and simply decline to par-
ticipate in a field study. Finally, in a natural setting, larger
meals may not simply contain larger portions than smaller
items but may also contain different types of food (e.g.,
burgers versus salads), more items (larger meals may
include a dessert), or items that are more difficult to esti-
mate (e.g., multicomponent sandwiches). All these factors
would lead to a stronger underestimation among high-BMI
people in a natural setting than in a laboratory setting.
Therefore, failure to find reliable differences between low-
and high-BMI people in a natural setting would provide fur-
ther support for our hypothesis that calorie estimation
biases are driven by meal size and not by factors related to
body size.
Similarly, it might be asked whether the results of Studies
1 and 2 would hold for consumers with a deep personal
interest in health and nutrition. Using the same line of rea-
soning as for the influence of body mass, we expect that the
influence of nutrition involvement will be entirely mediated
by meal size. In other words, we expect that people who
pay attention to what and how much they eat choose smaller
fast-food meals. As a result, their meal size estimations
should be more accurate than those of consumers who do
not care about nutrition. Still, we expect that the estimations
of both groups of consumers equally underestimate changes
in meal size (i.e., follow the same psychophysical power
curve). As we noted previously, psychophysical biases are
automatic and driven low-level perceptual processes and
cannot be corrected by cognitive effort (Arkes 1991).
Therefore, they should apply equally to people with a high
or low interest in nutrition.
STUDY 3: A FIELD STUDY OF FAST-FOOD MEAL SIZE
ESTIMATIONS
Method
Trained interviewers approached every fourth person as
they were finishing their meals in food courts operated by
different fast-food restaurants in three medium-sized U.S.
cities in the Midwest and asked them if they would answer
some brief questions for a survey. No mention was made of
food at that time. Of the 200 people who were approached,
147 (73.5%) agreed to participate. They were first asked to
estimate the number of calories contained in their entire
meal. Then, they answered a short series of questions about
their eating habits and provided details of their height (in
feet and inches) and weight (in pounds), which were used to
compute their BMI. Of the 147 respondents, 91 were classi-
fied as low BMI (BMI < 25), and 56 were classified as high
BMI (BMI 25). During this process, the interviewer
unobtrusively recorded and confirmed the type and size of
the food and drinks from the wrappings left on the tray.
Nutritional information provided by the fast-food restau-
rants was then used to compute the actual number of calo-
ries of each person’s meal. In case of uncertainty (e.g., to
determine whether the drink was diet or regular), the inter-
viewer asked for clarification.
To create a reliable measure of involvement in nutrition,
we used a principal component analysis (α= .83) of partici-
pants’ responses to five rating scales (“I watch what I eat,
“I pay attention to what I eat,” “I pay attention to how much
I eat,” “Eating healthy is important to me,” and “Nutritional
information influenced me”) and to three binary questions
(“Was nutritional information readily available here?” “Did
you pay attention to the nutritional information available
here?” and “Did the nutritional information influence your
selection?”). In support of the validity of the measure, we
found a negative and statistically significant correlation
between nutrition involvement and BMI (r = –.23, p< .01),
indicating that high-BMI participants reported being less
involved in nutrition than low-BMI participants. We then
classified participants into a high-nutrition-involvement
group (n = 70) and a low-nutrition-involvement group (n =
70) on the basis of a median split.
Results
On average, participants underestimated the number of
calories of their meal (MS= 546 calories versus MI= 744
calories; PDEV = –17.5%; t = –7.45, p< .001). However,
this average underestimation hides large differences that
depend on the number of calories of the meal. After we
dichotomized meals with a median split, we found that the
number of calories in small meals was more accurately esti-
mated (MS= 433 calories versus MI= 484 calories;
PDEV = –.6%; t = –.1, p= .92), whereas the number of
calories of large meals was strongly underestimated (MS=
687 calories versus MI= 1144 calories; PDEV = –34.6%;
92 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
t = –10.4, p< .001). Respondents estimated that, on aver-
age, larger meals contained 254 more calories than did
smaller meals, though in reality, they contained 660 more
calories, more than twice the estimated number. Figure 3,
Panel A, shows the mean estimated and actual number of
calories for each quartile of the meals that low- and high-
BMI participants selected. As Panel A shows, mean estima-
tions are close to the accuracy line for small meals but grow
more slowly than the actual number of calories and quickly
fall below the accuracy line as consumption quantities
increase.
Effects of body mass. As we expected, Study 3 replicated
the findings of Study 2 about calorie underestimation for
self-selected meals. Because of the self-selection of meal
sizes, the calorie underestimation bias was more pro-
nounced among high-BMI participants (PDEV = –30.4%)
than among low-BMI participants (PDEV = –9.5%), and
the difference was statistically significant (F(1, 145) = 8.90,
p< .001). As in Study 2, the mean estimation of high-BMI
participants (MS= 560 calories) was not statistically differ-
ent from those of low-BMI participants (MS= 532 calories;
F(1, 145) = .44, p= .51), though the actual size of the meal
(MI= 900 calories) was 241 calories higher than the actual
size of the meals chosen by low-BMI participants (MI= 659
calories; F(1, 145) = 17.0, p< .001). We illustrate the ten-
dency for high-BMI participants to select larger meals in
Figure 3, Panel A, which shows that their mean estimations
(black dots) are higher (and more toward the right) than
those of low-BMI participants (white dots). Therefore, it is
possible that the stronger underestimation of high-BMI
people is due to their selection of larger meals and not to an
intrinsic tendency to underestimate meal size. Data in Fig-
ure 3, Panel A, support this hypothesis by showing that the
estimations of low- and high-BMI participants in the con-
trol condition are on the same curve.
To formally test whether the estimations of low- and
high-BMI participants follow the same power curve, we
conducted the same analysis as in Studies 1 and 2 and esti-
mated the model represented in Equation 2. As we
expected, the coefficient for meal size was statistically
lower than 1 (β= .56, t-test of difference from 1 = –5.8, p<
.001), showing that, on average, the power curve is com-
pressive. The coefficient for BM was not statistically sig-
nificant (γ= –.61, t = –.6, p= .54), indicating that low- and
high-BMI participants have similar estimations after we
control for the effects of meal size. In addition, the interac-
tion between the effects of meal size and body mass was not
statistically significant (δ= .07, t = .5, p= .62), indicating
that the curvature of the power curve is the same for both
groups. As in Study 1, we found that the hypothesized
power model fit the data better (R2= .28, F(1, 145) = 55.4,
p< .01) than a linear model (R2= .23, F(1, 145) = 43.7, p<
.01). The MAPE of the power model was also statistically
significantly better than that of the linear model
(MAPE(Power model) = .38 versus MAPE(Linear model) = .43;
t = 4.32, p< .01).
Figure 3
STUDY 3:THE BODY MASS (PANEL A) AND NUTRITION INVOLVEMENT (PANEL B) OF FAST-FOOD EATERS DO NOT CHANGE THE
EFFECTS OF MEAL SIZE ON CALORIE ESTIMATIONS (OBSERVED GEOMETRIC MEANS, 95% CONFIDENCE INTERVAL, AND MODEL
PREDICTIONS)
A: Body Mass B: Nutrition Involvement
Is Obesity Caused by Calorie Underestimation? 93
Effects of nutrition involvement. Figure 3, Panel B, shows
the mean estimated and actual calories for each quartile of
the meals selected by low- and high-nutrition-involvement
participants. As we expected, participants in the high-
nutrition-involvement group chose meals that contained
fewer calories (MI= 577 calories) than did participants in
the low-nutrition-involvement group (MI= 958 calories;
F(1, 138) = 42.9, p< .001). As a result, the estimations of
participants in the high-nutrition-involvement group were
more accurate (PDEV = –2.8%) than those of participants
in the low-nutrition-involvement group (PDEV = –31.4%;
F(1, 138) = 17.5, p< .001). To formally test our hypothesis
that the effects of nutrition involvement are entirely medi-
ated by meal size selection and that nutrition involvement
does not moderate psychophysical biases, we estimated the
model represented in Equation 4:
(4) ln(S) = α+ β×ln(I) + γ×INVOL + δ×ln(I) ×INVOL + ε,
where INVOL is a binary variable that captures whether the
individual is in the high- (INVOL = .5) or low- (INVOL =
–.5) involvement group. All parameters were in the
expected direction, and the parameter for meal size
remained unchanged from the previous analysis (β= .56).
As we expected, the coefficient for INVOL was not statisti-
cally significant (γ= –.31, t = –.3, p= .78), indicating that
participants with low and high involvement in nutrition
have similar estimations after we control for the effects of
meal size. In addition, the interaction between the effects of
meal size and nutrition involvement was not statistically
significant (δ= .06, t = .4, p= .71), indicating that the cur-
vature of the power curve was the same for both groups (see
Figure 3, Panel B). Again, calorie estimations are driven by
meal size, not by nutrition involvement.
Discussion
Summary. Study 3 replicated the findings of Studies 1
and 2 in a field setting in which people were asked to esti-
mate the size of their meal minutes after they had finished
consuming it. As in Study 2 and in previous DLW studies,
high-BMI consumers were more likely to underestimate the
true size of the meal than low-BMI consumers. However,
Study 3 shows that this result is a spurious consequence of
high-BMI people’s tendencies to eat larger meals and that
the estimations of low- and high-BMI consumers follow the
same compressive power function of the actual size of the
meal. We obtained these results in a natural setting in which
we would expect more intrinsic underestimation from high-
BMI consumers because of self-representation and selec-
tion of meals that were more difficult to estimate; this pro-
vides further support for our hypothesis that biases in meal
size estimation are driven mostly by psychophysical percep-
tual biases.
Study 3 also enabled us to test the robustness of psy-
chophysical biases by showing that they are not moderated
by nutrition involvement. Participants who reported paying
attention to nutritional information and eating healthfully
were as likely to underestimate increases in meal size as
participants who reported ignoring nutritional information
and healthful eating. In addition, this analysis provided
another illustration of the misleading results of naive analy-
ses that do not control for psychophysical effects. When
meal size is not controlled for, it appears that calorie esti-
3We thank the editor and one reviewer for encouraging us to address
these important questions.
mations are more accurate for people who are involved in
nutrition than for those who are not. However, after meal
size is controlled for, calorie estimations are identical across
both nutrition involvement groups.
Public health implications. The robustness of psy-
chophysical biases in meal size estimation across body
sizes, nutrition involvement, and estimation contexts raises
the question of their impact on public health. Three ques-
tions are particularly important in this regard: (1) Are dieti-
cians knowledgeable about people’s estimation biases? (2)
Are dieticians able to correct these biases in their own esti-
mations? and most important, (3) Do meal size estimation
biases influence fast-food consumption decisions?3
Studying whether dieticians are knowledgeable about
people’s estimation biases offers important implications for
the clinical treatment of obesity. We expect that dieticians
are aware of the health science research that shows that
high-BMI people underestimate their own food intake more
than low-BMI people but are unaware that it is caused by
meal size and not by body size. Therefore, we expect that
dieticians will predict (inaccurately) that high-BMI people
have lower meal size estimations than low-BMI people esti-
mating the same meals.
Studying whether dieticians are able to correct these
biases in their own estimations enables us to examine
whether professional training and practice eliminate, or at
least reduce, psychophysical biases. Because psychophysi-
cal biases are automatic and unconscious, we also expect
them to influence dieticians’ estimations, though the train-
ing and practice of professional dieticians may moderate
the strength of these biases. For these reasons, we also
expect that a piecemeal decomposition estimation will
improve dieticians’ estimations but that the improvement
will be less dramatic than for the regular consumers
involved in Study 2.
Finally, studying whether biases in meal size estimation
influence fast-food consumption decisions has obvious pub-
lic health implications. Underestimating the number of
calories contained in fast-food meals, especially in the
largest ones, conceals their negative long-term health conse-
quences. Improving the accuracy of calorie estimations
should lead people who value their health to choose smaller
fast-food meals. Therefore, we expect dieticians to choose a
smaller fast-food meal when they use a piecemeal decom-
position estimation than when they use a holistic estimation
procedure.
STUDY 4: ARE THE CALORIE ESTIMATIONS OF
DIETICIANS BIASED, AND DO THEY INFLUENCE
THEIR CONSUMPTION DECISIONS?
Method
We asked 405 certified dieticians attending an annual
conference of the American Association of Diabetes Educa-
tors to estimate the number of calories of three fast-food
meals that contained the same ingredients but in different
sizes. Meal A contained 480 calories (255 calories from a
three-inch ham sandwich, 125 calories from six chips, and
94 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
Figure 4
STUDY 4: MEAL SIZE BIASES DIETICIANS’ OWN CALORIE
ESTIMATIONS AND THEIR FORECAST OF THE ESTIMATIONS
OF A LOW- AND HIGH-BMI PERSON (OBSERVED GEOMETRIC
MEANS AND MODEL PREDICTIONS)
100 calories from a ten-ounce glass of regular cola). Meal B
contained double the amount of each ingredient, for a total
of 960 calories, and Meal C contained four times the
amount of each ingredient, for a total of 1920 calories.
These meals were described as consisting of “a ham, Genoa
salami, and pepperoni sandwich; regular chips; and Classic
Coke,” and participants saw pictures of the three meals, side
by side, on the same page.
The dieticians were randomly assigned to three condi-
tions. In the self-estimation condition, they were asked to
provide their own estimations of the number of calories of
three meals. In the forecast/low-BMI condition, they were
asked to forecast the calorie estimations of a low-BMI per-
son. To help them do this, we gave them a picture of a thin-
looking person with the following legend: “Bethany is a 25-
year-old, 5’7”, and 150-pound research assistant.” In the
forecast/high-BMI condition, they were asked to forecast
the calorie estimations of a high-BMI person. To help them
do this, we gave them a picture of an overweight-looking
person with the following legend: “Sarah Jo is a 25-year-
old, 5’7”, and 200-pound research assistant.” In addition,
half of the dieticians in the self-estimation condition were
asked to use a piecemeal decomposition estimation, and the
other half were asked to use a holistic estimation procedure
similar to the one used by dieticians in the two forecast con-
ditions. As in Study 2, the dieticians in the self/piecemeal
condition were asked to estimate separately the number of
calories of the sandwich, the chips, and the beverage and
then to add up these three numbers for each of the three
meals. Finally, the respondents in the self-estimation condi-
tion indicated which of the three meals they would order for
themselves and how satiated they expected to be after eat-
ing such a meal.
Results
Comparing dieticians’ self-estimations with their fore-
casts of the estimations of high- and low-BMI people. The
two questions motivating this study were, (1) Are dieticians
knowledgeable about people’s estimation biases? and (2)
Are dieticians able to correct these biases in their own esti-
mations? To answer these two questions, we first examined
the calorie estimations of dieticians in the self/holistic con-
dition and then compared them with those of dieticians in
the forecast/low-BMI condition and in the forecast/high-
BMI condition. On average, the estimations of dieticians in
the self/holistic condition were below the actual number of
calories (PDEV = –8.5%; t = –4.2, p< .01). As Figure 4
shows, however, this hides important differences across
meal sizes. Dieticians’ estimations were not statistically dif-
ferent from reality for the smaller meal (PDEV = 3.7%; t =
1.0, p= .33), but they were well below reality for the
medium meal (PDEV = –7.5%; t = –2.3, p< .05) and for the
large meal (PDEV = –21.4%; t = –7.5, p< .01).
To formally test whether psychophysical biases affect
dieticians’ own estimations and to compare dieticians’ self-
estimations with their forecasts, we estimated the following
model:
(5) ln(S) = α+ β×ln(I) + γ×F-LBMI + δ×F-HBMI
+ θ×ln(I) ×F-LBMI λ×ln(I) ×F-HBMI + ε,
where S is estimated calories; I is geometric mean-centered
observed calories; F-LBMI and F-HBMI are two binary
variables that capture the three conditions (F-LBMI = .33
for dieticians in the forecast/low-BMI condition and –.67
otherwise; F-HBMI = .33 for dieticians in the forecast/high-
BMI condition and –.67 otherwise); εis the error term; and
α, β, γ, δ, θ, and λare parameters estimated with OLS.
As we expected, the coefficient for meal size was statisti-
cally lower than 1 (β= .77, t-test of difference from 1 =
–10.3, p< .001), showing that across the three conditions,
dieticians’ calorie estimations are compressive. The main
effect of F-LBMI was not statistically significant (γ= –.02,
t = –.6, p= .56), showing that dieticians believe that the
estimations of a low-BMI person are similar to their own
estimations. In contrast, the main effect of F-HBMI was
negative and statistically significant (δ= –.11, t = –3.6, p<
.01), showing that dieticians erroneously believe that the
estimations of a high-BMI person are systematically lower
than their own estimations. Further contrast tests revealed
that they also believe that the estimations of a high-BMI
person are lower than the estimations of a low-BMI person
(t = –3.6, p< .01). None of the interaction terms were sig-
nificant (θ= –.02, t = –.4, p= .69, and λ= –.05, t = –.9, p=
.33). This shows that the magnitude of the psychophysical
bias is the same in the three conditions, providing further
Is Obesity Caused by Calorie Underestimation? 95
4In an additional analysis, we estimated the following power model:
ln(S) = α+ β×ln(I) + γ×PCM + δ×ln(I) ×PCM + ε, where S is the esti-
mated number of calories, I is the mean-centered actual number of calo-
ries, and PCM measures whether dieticians were in the self/piecemeal
(PCM = .5) or self/holistic (PCM = –.5) conditions. As we expected, the
main effect of PCM was positive and statistically significant (γ= .11, t =
4.3, p< .01). The interaction effect was in the expected direction, but
unlike in Study 2, it was not statistically significant (δ= .05, t = 1.3, p=
.2). This may be because the holistic estimations of dieticians were a lot
less compressive (β= .79) than the holistic estimations of the regular con-
sumers participating in Study 2 (β= .35). There is simply less room for
improvement for vigilant dieticians. Still, the analysis of variance results
show that the improvements brought about by using a piecemeal estima-
tion are statistically significant when examining medium and large meals
independently.
support that the tendency to underestimate meal size
changes and, thus, that to underestimate large meals more
strongly than small meals is robust because, in the tests, it
also influenced professional dieticians. It also shows that
estimation biases are not driven by self-presentation moti-
vation, because they also occur when forecasting other
people’s estimations.
Effects of piecemeal decomposition on dieticians’ estima-
tions and consumption decisions. We compared the estima-
tions of dieticians in the self-estimation/holistic condition
with those of dieticians in the self-estimation/piecemeal
condition. Recall that, on average, dieticians in the self/
holistic condition underestimated the number of calories of
the three meals (PDEV = –8.5%). In comparison, dieticians
in the self/piecemeal condition were more accurate
(PDEV = .0%; F(1, 551) = 11.7, p< .001). As Figure 4
shows, the effects of the piecemeal estimation were stronger
on large meals than on small meals. The estimations of
small meals were similar in the holistic and piecemeal con-
ditions (F(1, 182) = 1.6, p= .21), but the estimations of
medium and large meals were more accurate in the piece-
meal condition (F(1, 182) = 6.3, p< .05, and F(1, 182) =
5.7, p< .05, respectively). Overall, Study 4 provides further
evidence that piecemeal estimation reduces the calorie
underestimation bias even for dieticians and is particularly
effective for large meals.4
These results lead us to the third and final question: Do
biases in meal size estimation influence fast-food consump-
tion decisions? To examine this, we asked dieticians in the
self/holistic and self/piecemeal conditions to indicate which
of the three meals sizes they would order for lunch. As we
expected, the proportion of dieticians who chose a medium
or a large meal size was higher in the holistic condition
(M = 58.2%), when they tended to underestimate meal
sizes, than in the piecemeal condition (M = 43.8%; χ2= 4.1,
p< .05), when their estimations were more accurate. As a
result, the meals chosen by dieticians in the holistic condi-
tion contained more calories (M = 781 calories) than the
meals chosen by dieticians in the piecemeal condition (M =
690 calories; F(1, 182) = 5.7, p< .01). The correlation
between each dietician’s average calorie estimation and his
or her meal size choice was negative and statistically sig-
nificant (r = –.26, p< .01), further indicating that meal size
estimations drove meal size choices. Finally, we measured
dieticians’ expectations of their level of satiation with their
chosen meal using a nine-point scale (1 = “very hungry,”
and 9 = “very full”). Dieticians in the holistic condition
expected to be as full with their meal (M = 7.5) as dieticians
in the piecemeal condition (M = 7.3; F(1, 182) = .5, p=
.50). This shows that improving calorie estimations did not
make dieticians more willing to restrain their consumption
and choose meals too small to satisfy their hunger. Rather,
the piecemeal estimation made dieticians more aware of the
true number of calories of the meals, so they avoided choos-
ing meals that were unnecessarily large.
Discussion
Study 4 shows that psychophysical biases also apply to
professional dieticians, though they are less pronounced
than for the regular consumers who participated in Studies
1, 2, and 3. A comparison of the predictions made by the
best-fitting psychophysical models of calorie estimations in
Studies 1 and 4 (the two studies that provided multiple esti-
mations per respondent) shows that these models predict
that for a fast-food meal that contains 1000 calories, dieti-
cians’ mean estimations will be 857 calories, whereas regu-
lar consumers’ mean estimations will be only 664 calories.
On the one hand, these results show that professional edu-
cation and consistent practice improve calorie estimations.
On the other hand, they show that psychophysical biases are
difficult to eliminate, even when accompanied by such
diligence.
Study 4 also shows that dieticians wrongly predict that
people with a high BMI will have systematically lower
meal size estimations than people with a low BMI or than
themselves. We do not know whether this occurs because
dieticians believe that high-BMI people underestimate calo-
ries for self-presentation purposes or because they believe
that calorie underestimation explains why they have a high
BMI in the first place. However, because both theories are
wrong, this finding has important implications for the clini-
cal treatment of obesity. These results also rule out that
biases in meal size estimation are motivated by self-
presentation because they also occur when people are fore-
casting the estimations of other people, not just when they
are providing their own estimations.
Study 4 also shows that using a piecemeal decomposition
improved the calorie estimations of dieticians, though the
effect is less pronounced than for regular consumers, whose
holistic estimations tend to be more strongly compressive.
Our model predicts that the mean estimation of a 1000-
calorie meal by dieticians who use a piecemeal decomposi-
tion estimation is an impressively accurate 957 calories.
Finally, Study 4 shows that improving meal size estimations
has direct consequences on consumption decisions because
it can influence even professional dieticians to scale back to
smaller, but equally satisfying, meals. That piecemeal esti-
mation can influence the consumption decisions of profes-
sional dieticians suggests that its effects would be even
greater among average consumers, whose estimations bene-
fit even more from the use of a piecemeal decomposition.
This has important clinical implications because dieticians
often make recommendations about appropriate meal sizes
to their patients.
GENERAL DISCUSSION
This research is motivated by the often-cited allegation
that calorie underestimation, coupled with the increase in
restaurants’ meal sizes, is an important driver of obesity
(e.g., Nestle 2002). This argument is supported by consider-
96 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
able evidence that shows that people with a higher body
mass are more likely to underestimate their food intake than
people with a lower body mass (Livingstone and Black
2003). In this research, we develop and test a psychophysi-
cal model of meal size estimation and use it to show that the
association between body mass and biases in calorie esti-
mations is a spurious consequence of the tendency of over-
weight people to consume larger meals.
The key results of the three laboratory studies and of the
field study appear in Table 1. In all these studies, we find
that meal size estimations follow a compressive power
function of actual meal size. In other words, these estima-
tions exhibit diminishing sensitivity to meal size changes as
the size of the meal increases. We further show that the esti-
mations of low- and high-BMI consumers follow the exact
same psychophysical function, whether they are made
before or after intake, for self-selected or randomly selected
meals. As a result, the estimations of low- and high-BMI
consumers are identical, after the size of the meal is con-
trolled for or after the natural association between meal size
and body mass is eliminated. Calorie underestimation is
caused by meal size, not body size.
We also test two other predictions derived from the psy-
chophysical model. The first is that a piecemeal decomposi-
tion estimation procedure should reduce psychophysical
biases because it replaces the estimation of a whole meal (a
large quantity, which is likely to be underestimated) with
multiple estimations of the size of each component of the
meal (smaller quantities, which are likely to be estimated
more accurately). As we predicted, the piecemeal decompo-
Ta b l e 1
SUMMARY STATISTICS FOR STUDIES 1–4
Estimated Number Actual Number
of Calories of Calories Mean PDEV
Estimation Type Grouping (Geometric Mean) (Geometric Mean) (Arithmetic Mean)
Study 1
Whole meal All meals 448 589 –11.4%b
Small meals 308 351 2.0%
Large meals 654a980a–24.8%a, b
BMI < 25 443 589 –11.1%b
BMI 25 461 589 –12.0%b
Study 2
Whole meal All meals 808 945 –7.7%b
Small meals 784 755 9.0%
Large meals 835 1191a–24.7%a, b
BMI < 25 799 871 –2.6%
BMI 25 929 1117a–17.9%a, b
Whole meal with bias disclosure All meals 1030 942 17.6b
Small meals 929 755 32.7a, b
Large meals 1175a1251a–1.7
Piecemeal All meals 872 851 6.2
Small meals 703 671 8.4
Large meals 1081a1080a3.9
Study 3
Whole meal, post intake All meals 546 744 –17.5%b
Small meals 433 484 –.6%
Large meals 687a1144a–34.6%a, b
BMI < 25 532 659 –9.5%a
BMI 25 560 900a–30.4%a, b
Low involvement 601 958 –31.4b
High involvement 495a577a–2.8a
Study 4
Whole meal (for self) All meals 832 960 –8.5%b
Small meals 474 480 3.7%
Large meals 1100a1358a–14.5%a, b
Piecemeal (for self) All meals 925 960 0.0%
Small meals 510 480 10.5%b
Large meals 1246a1358a–5.1%a, b
Whole meal (forecasts for low-BMI patient) All meals 814 960 –9.6, a
Small meals 475 480 4.0
Large meals 1066a1358a–16.5a, b
Whole meal (forecasts for high-BMI patient) All meals 744 960 –15.3b
Small meals 439 480 –2.2
Large meals 968a1358a–21.9a, b
aStatistically different from the other group (p< .05).
bStatistically different from zero (p< .05).
Notes: We categorized small and large meals on the basis of a median split, except in Study 4, in which the small meal is the 480-calorie meal and the large
meals include the 960- and 1020-calorie meals.
Is Obesity Caused by Calorie Underestimation? 97
sition estimation reduces psychophysical biases not only
among regular consumers but also among certified dieti-
cians. In comparison, a common debiasing manipulation—
informing consumers about the bias and motivating them to
be accurate—does not improve people’s sensitivity to meal
size changes (though it leads to a general increase in calorie
estimations). The second prediction is that when a represen-
tative sample of consumers and consumption occasions is
surveyed, the mean estimated consumption is lower than the
mean observed consumption. This prediction is derived
from the nonlinear shape of the psychophysical function,
which leads to stronger underestimations of large quantities
than overestimations of small quantities. This prediction
explains why most studies, which use a representative sam-
ple, find an average underestimation bias, whereas the few
studies that focus on small consumption magnitudes (e.g.,
studying children or low-BMI consumers) find an average
overestimation bias.
Our final analyses address the public health implications
of psychophysical biases in meal size estimations by study-
ing the estimations, forecasts, and consumption decisions of
professional dieticians. We find evidence that psycho-
physical biases affect even highly educated, expert dieti-
cians, though to a lesser extent than regular consumers.
More worrying, we find that dieticians inaccurately expect
that high-BMI people underestimate meal size compared
with low-BMI people. Finally, we find that a piecemeal
decomposition also improves dieticians’ own calorie esti-
mations, which leads them to select smaller fast-food meals.
Public Policy and Health Practitioner Implications
As the availability and marketing of larger portion sizes
have increased, the calorie underestimation bias can explain
why average obesity rates are increasing over time. There-
fore, our findings do not exonerate the food industry’s role
of contributing to obesity. Still, they show that this role is
less than what has been suggested in the public health lit-
erature and in the popular press (Wansink and Huckabee
2005). The tendency for high-BMI consumers to underesti-
mate the amount of food they have consumed is not caused
by an inappropriate disclosure of nutritional information but
rather is a function of eating large meals. The reason some
people have a higher body mass than others cannot be
linked to their inability to estimate meal sizes. This implies
that the Food and Drug Administration–endorsed dieting
practice of counting calories (Food and Drug Administra-
tion 2004) may be less effective in fighting obesity than
expected because high-BMI people are not intrinsically
worse calorie estimators than low-BMI people. In addition,
counting the calories of whole meals is likely to lead to
severe underestimation biases. This strategy may even
backfire because the underestimation could suggest that
people can safely indulge in additional consumption (Chan-
don and Wansink 2007; Wansink and Chandon 2006).
Our results provide strong evidence that consumption
estimation biases have a perceptual origin and are not moti-
vational or personality based. Attributing biased calorie
estimations to denial or self-presentation motivations may
be unfair and ultimately counterproductive if people cope
with these accusations by avoiding treatment. Although the
focus is often on calorie underestimation by overweight
people, this also applies to calorie overestimations by
people with anorexia, which has often been attributed to an
“excessive concern with eating and dieting” (Williamson,
Gleaves, and Lawson 1991, p. 257). Our results indicate
that an important component of this overestimation bias
may have something to do with the small amounts of food
anorexics eat.
Our results also raise the question of what medical prac-
titioners, clinicians, and health policy professionals can do
to improve consumers’ meal size estimations. Our results
show the limitations of the traditional method the govern-
ment uses—namely, expensive educational efforts that
involve informing people of the bias. Information and
incentives can change average calorie estimations and
therefore can help reduce the general calorie underestima-
tion bias. However, they are not sufficient to change the
psychophysical bias that leads to the underestimation of the
increases in portion sizes that has occurred over the past 20
years (Nielsen and Popkin 2003).
A solution to raise average calorie estimates and to
improve the estimation of meal size change would be to dis-
play calorie information in restaurants (Seiders and Petty
2004). Bills requiring that the Nutritional Labeling and
Education Act be extended to restaurants have been exam-
ined in several U.S. states, but they face strong opposition
from the National Restaurant Association (Center for Sci-
ence in the Public Interest 2005). A less controversial solu-
tion would be to encourage people to use a piecemeal
decomposition rather than trying to estimate the number of
calories they consume in a meal or in a day. Research on the
effectiveness of a decomposition strategy has shown that
this would be particularly effective for single estimations,
when the meal components are not salient, and when people
have not yet made a holistic, top-down (e.g., brand-based)
judgment (Bolton 2003; Menon 1997). As a caveat, how-
ever, the tendency of piecemeal decomposition to increase
calorie estimations would make it undesirable for the treat-
ment of anorexia and bulimia, because anorexics and bulim-
ics are already prone to overestimations.
Research Implications
Biases in calorie estimations suggest that the results of
studies that use self-reported consumption data as an inde-
pendent or dependent variable can be biased. Unfortunately,
because of the prohibitive cost of the DLW technique, most
researchers in marketing, nutrition, and epidemiology are
likely to continue to rely on self-reported consumption data.
How can self-reported consumption data be debiased? One
technique is to eliminate data from overweight respondents
because they are more likely to underestimate their own
consumption, but this eliminates data from people who are
of the greatest interest. Another technique consists of apply-
ing a correction factor to all observations or to different
groups of respondents (e.g., high-BMI versus low-BMI
people). Our study suggests that better results can be
achieved by applying a different correction factor for each
meal size. In Study 3, because consumption was underesti-
mated by an average of 17.5%, a general correction factor
would be to multiply self-reports by 1.21 (1/[1 – .175]). We
compared the accuracy of the calorie estimations corrected
with (1) a single correction factor, (2) a different factor for
98 JOURNAL OF MARKETING RESEARCH, FEBRUARY 2007
low-BMI (<25) and high-BMI (25) people, and (3) a dif-
ferent factor for small and large meals (dichotomized on the
basis of a median split). The BMI-based correction
(MAPE = .40) was not more accurate than the single-factor
correction (MAPE = .38, t = 1.6, p= .10), and both were
less accurate than the meal size–based correction (MAPE =
.36, t = 2.2, p< .05).
A fruitful area for additional research would be to extend
our results to people with very high body mass. Because of
a small number of observations for this group, we could not
distinguish between obese (BMI 30) and simply over-
weight (25 BMI < 30) people. Extending the analyses to
obese people would also provide further evidence that the
results of Studies 1, 2, and 3 are not caused by the restricted
range of the BMI of the participants. Another area worthy
of research would be to examine the effects of expectations
(Chandon and Wansink 2007). Low-calorie expectations
might aggravate the underestimation bias so that consumers
would be more accurate when estimating a prototypical
high-calorie fast-food meal (e.g., a McDonald’s hamburger
and fries meal) than when estimating an objectively more
healthful meal (e.g., a Subway sandwich meal).
REFERENCES
Andrews, J. Craig, Richard G. Netemeyer, and Scot Burton
(1998), “Consumer Generalization of Nutrient Content Claims
in Advertising,Journal of Marketing, 62 (October), 62–75.
Arkes, Hal R. (1991), “Costs and Benefits of Judgment Errors:
Implications for Debiasing,” Psychological Bulletin, 110 (3),
486–98.
Balasubramanian, Siva K. and Catherine Cole (2002), “Con-
sumers’ Search and Use of Nutrition Information: The Chal-
lenge and Promise of the Nutrition Labeling and Education
Act,” Journal of Marketing, 66 (July), 112–27.
Blundell, John E. (2000), “What Foods Do People Habitually Eat?
A Dilemma for Nutrition, an Enigma for Psychology,Ameri-
can Journal of Clinical Nutrition, 71 (1), 3–5.
Bolton, Lisa E. (2003), “Stickier Priors: The Effects of Nonana-
lytic Versus Analytic Thinking in New Product Forecasting,
Journal of Marketing Research, 40 (February), 65–79.
Brownell, Kelly D. and Katherine Battle Horgen (2003), Food
Fight: The Inside Story of the Food Industry, America’s Obesity
Crisis, and What We Can Do About It. New York: McGraw-Hill.
Center for Science in the Public Interest (2005), Nutrition Label-
ing in Chain Restaurants: Pending State Bills. Washington, DC:
Center for Science in the Public Interest, (accessed October 9,
2006), [available at http://www.cspinet.org/nutritionpolicy/
pending_statebills.pdf].
Chandon, Pierre and Brian Wansink (2002), “When Are Stock-
piled Products Consumed Faster? A Convenience–Salience
Framework of Postpurchase Consumption Incidence and Quan-
tity,Journal of Marketing Research, 39 (August), 321–35.
——— and ——— (2007), “The Biasing Health Halos of Fast
Food Restaurant Health Claims: Lower Calorie Estimates and
Higher Side-Dish Consumption Intentions,” Journal of Con-
sumer Research, 34, forthcoming.
Crandall, Christian S. (1994), “Prejudice Against Fat People: Ide-
ology and Self-Interest,” Journal of Personality and Social Psy-
chology, 66 (5), 882–94.
Folkes, Valerie S. and Shashi Matta (2004), “The Effect of Pack-
age Shape on Consumers’ Judgments of Product Volume: Atten-
tion as a Mental Contaminant,” Journal of Consumer Research,
31 (2), 390–91.
Food and Drug Administration (2004), “HHS Unveils FDA Strat-
egy to Help Reduce Obesity: New ‘Calories Count’ Approach
Builds on HHS’ Education, Research Efforts,” press release,
United States Department of Health and Human Services,
(March 12), (accessed October 9, 2006), [available at http://
www.fda.gov/bbs/topics/news/2004/hhs_031204.html].
Frayman, Bruce J. and William E. Dawson (1981), “The Effect of
Object Shape and Mode of Presentation on Judgments of
Apparent Volume,Perception and Psychophysics, 29 (1),
56–62.
Hedley, Allison A., Cynthia L. Ogden, Clifford L. Johnson, Mar-
garet D. Carroll, Lester R. Curtin, and Katherine M. Flegal
(2004), “Prevalence of Overweight and Obesity Among US
Children, Adolescents, and Adults, 1999-2002,” Journal of the
American Medical Association, 291 (23), 2847–50.
Krider, Robert E., Priya Raghubir, and Aradhna Krishna (2001),
“Pizzas: Pi or Square? Psychophysical Biases in Area Compari-
sons,” Marketing Science, 20 (4), 405–425.
Krishna, Aradhna (2005), “Spatial Perception Research: An Inte-
grative Review of Length, Area, Volume and Number Percep-
tion,” working paper, Stephen M. Ross School of Business, Uni-
versity of Michigan.
Krueger, Lester E. (1989), “Reconciling Fechner and Stevens:
Toward a Unified Psychophysical Law,” Behavioral and Brain
Sciences, 12 (2), 251–320.
Lansky, David and Kelly D. Brownell (1982), “Estimates of Food
Quantity and Calories: Errors in Self-Report Among Obese
Patients, American Journal of Clinical Nutrition, 35 (4),
727–32.
Lichtman, S.W., K. Pisarska, E.R. Berman, M. Pestone, H. Dowl-
ing, E. Offenbacher, H. Weisel, S. Heshka, D.E. Matthews, and
S.B. Heymsfield (1992), “Discrepancy Between Self-Reported
and Actual Caloric Intake and Exercise in Obese Subjects,New
England Journal of Medicine, 327 (27), 1893–98.
Livingstone, M. Barbara E. and Alison E. Black (2003), “Markers
of the Validity of Reported Energy Intake,Journal of Nutrition,
133 (3), 895S–920S.
Menon, Geeta (1997), “Are the Parts Better Than the Whole? The
Effects of Decompositional Questions on Judgments of Fre-
quent Behaviors,Journal of Marketing Research, 34 (August),
335–46.
Moorman, Christine, Kristin Diehl, David Brinberg, and Blair
Kidwell (2004), “Subjective Knowledge, Search Locations, and
Consumer Choice,” Journal of Consumer Research, 31 (3),
673–80.
Muhlheim, Lauren S., David B. Allison, Stanley Heshka, and
Steven B. Heymsfield (1998), “Do Unsuccessful Dieters Inten-
tionally Underreport Food Intake?” International Journal of
Eating Disorders, 24 (3), 259–66.
Nestle, Marion (2002), Food Politics: How the Food Industry
Influences Nutrition and Health. Berkeley: University of Cali-
fornia Press.
——— (2003), “Increasing Portion Sizes in American Diets:
More Calories, More Obesity,Journal of the American Dietetic
Association, 103 (1), 39–40.
Nielsen, Samara Joy and Barry M. Popkin (2003), “Patterns and
Trends in Food Portion Sizes, 1977–1998,Journal of the
American Medical Association, 289 (4), 450–53.
Paeratakul, Sahasporn, Daphne P. Ferdinand, Catherine M. Cham-
pagne, Donna H. Ryan, and George A. Bray (2003), “Fast-Food
Consumption Among US Adults and Children: Dietary and
Nutrient Intake Profile,Journal of the American Dietetic Asso-
ciation, 103 (10), 1332–38.
Raghubir, Priya and Aradhna Krishna (1996), “As the Crow Flies:
Bias in Consumers’ Map-Based Distance Judgments,Journal
of Consumer Research, 23 (1), 26–39.
——— and ——— (1999), “Vital Dimensions in Volume Percep-
tion: Can the Eye Fool the Stomach?” Journal of Marketing
Research, 36 (August), 313–26.
Is Obesity Caused by Calorie Underestimation? 99
Seiders, Kathleen and Ross D. Petty (2004), “Obesity and the Role
of Food Marketing: A Policy Analysis of Issues and Remedies,
Journal of Public Policy & Marketing, 23 (Fall), 153–69.
Srivastava, Joydeep and Priya Raghubir (2002), “Debiasing Using
Decomposition: The Case of Memory-Based Credit Card
Expense Estimates,” Journal of Consumer Psychology, 12 (3),
253–64.
Stevens, Stanley Smith (1986), Psychophysics: Introduction to Its
Perceptual, Neural, and Social Prospects. Oxford: Transaction
Books.
Subar, Amy F., Victor Kipnis, Richard P. Troiano, Douglas
Midthune, Dale A. Schoeller, Sheila Bingham, Carolyn O. Shar-
baugh, Jillian Trabulsi, Shirley Runswick, Rachel Ballard-
Barbash, Joel Sunshine, and Arthur Schatzkin (2003), “Using
Intake Biomarkers to Evaluate the Extent of Dietary Misreport-
ing in a Large Sample of Adults: The OPEN Study,American
Journal of Epidemiology, 158 (1), 1–13.
Teghtsoonian, Martha (1965), “The Judgment of Size,” American
Journal of Psychology, 78 (3), 392–402.
Tooze, Janet A., Amy F. Subar, Frances E. Thompson, Richard
Troiano, Arthur Schatzkin, and Victor Kipnis (2004), “Psy-
chosocial Predictors of Energy Underreporting in a Large Dou-
bly Labeled Water Study,American Journal of Clinical Nutri-
tion, 79 (5), 795–804.
Wansink, Brian and Pierre Chandon (2006), “Can ‘Low-Fat’
Nutrition Labels Lead to Obesity?” Journal of Marketing
Research, 43 (November), 605–617.
——— and Mike Huckabee (2005), “De-Marketing Obesity,”
California Management Review, 47 (4), 6–18.
——— and Koert Van Ittersum (2003), “Bottoms Up! The Influ-
ence of Elongation on Pouring and Consumption Volume,
Journal of Consumer Research, 30 (3), 455–63.
Williamson, Donald A., David H. Gleaves, and Olga J. Lawson
(1991), “Biased Perception of Overeating in Bulimia Nervosa
and Compulsive Binge Eaters,Journal of Psychopathology and
Behavioral Assessment, 13 (3), 257–68.
Young, Lisa R. and Marion Nestle (2002), “The Contribution of
Expanding Portion Sizes to the US Obesity Epidemic,” Ameri-
can Journal of Public Health, 92 (2), 246–49.
... Prior research suggests that consumers tend to rely on consumption norms to decide how much to consume (Wansink, 2004). Consumption norms regulate how much one considers as "appropriate" to eat, and these norms can be influenced by several environmental factors (e.g., variety of food, package size, nutritional label) (Chandon & Wansink, 2007;Kahn & Wansink, 2004;Wansink & Chandon, 2006). For example, Kahn and Wansink (2004) show that a large assortment (e.g., more variety of colors of M&M candies) makes consumers believe that larger amounts are acceptable to consume than a small assortment does. ...
... For example, Kahn and Wansink (2004) show that a large assortment (e.g., more variety of colors of M&M candies) makes consumers believe that larger amounts are acceptable to consume than a small assortment does. Chandon and Wansink (2007) show that large-sized packages and portions suggest that a larger amount of food is reasonable to eat than small packages or portions would suggest (see also Wansink and Van Ittersum, 2007). This is particularly relevant in today's super-sized environment, where fast-food restaurants and coffee chains are increasingly introducing larger portion-sizes to their menus. ...
... Finally, they provided basic demographic information (age and gender) as well as their height and weight information (used to compute their body-mass index, BMI = weight in kilograms/ squared height in meters). Since consumers with a higher BMI or a higher liking for the food tend to choose a larger portion-size (Chandon & Wansink, 2007), we used these two variables as covariates. ...
Article
We propose and find that the extremeness aversion bias when choosing portion-sizes is stronger for healthy food as compared to unhealthy food items. In two studies (and a follow-up) we find that adding an extra-large option to a standard menu of small, medium, and large portions increases the choice share of the larger portion-sizes; but more so for healthy food than for unhealthy food. Furthermore, we find evidence for the lay belief that larger portions of healthy food do not have incremental health costs. When health costs of the larger portions of healthy food were made salient by providing calorie information, the above effects disappeared. These findings show (1) a boundary condition to the extremeness aversion effect when choosing portion sizes, and (2) imply that this bias can act as a nudge to increase the consumption of healthy food.
... Because being hungry is undesirable, it is possible that people show a below-average effect on perceived hunger, which then leads to choosing larger portions for others. Second and relatedly, another possibility is that consumers may believe that others weigh more than they do, and therefore need a larger amount of food than they do (Chandon and Wansink 2007). This perspective would again be underpinned by a below-average effect on weight, as overweight people are the object of various negative associations (Levine and Schweitzer 2015;Puhl and Brownell 2001). ...
... Finally, we also considered weight perceptions of the other person. One possibility is that people systematically think that others are heavier than they are, such that they choose larger portion sizes for others because heavier people tend to consume larger portion sizes (Chandon and Wansink 2007 Procedure. Participants were randomly assigned to one of the six conditions: self, typical participant, best friend, work or study colleague, celebrity, or family member. ...
Preprint
Full-text available
Consumers’ portion size choices are important, both as larger portion sizes can lead to overeating and as uneaten portions can contribute to food waste. Existing research has largely focused on consumers’ portion size choices for themselves—even though consumers often choose for others. Fourteen studies examine portion size choices for others, testing: Do consumers choose smaller, similar, or larger portion sizes for others, compared to various benchmarks—(1) how much they choose for themselves, (2) how much others want to receive, and (3) predictions about how much others actually want to eat? Studies show that consumers choose larger portion sizes for others across multiple choosing-for-others contexts, involving everyday favors, gift-giving, and joint consumption. Consumers’ goal to be considerate of others’ needs and desires given uncertainty about others’ consumption is one broad “baseline” driver of this multiply determined phenomenon. Consumers do not choose larger portion sizes for others when they lack a considerateness goal, when choosing larger portions is inconsiderate, or when a responsibility goal instead dominates (as in the choosing-for-others context of caregiving). This research offers theoretical implications for understanding choices for others and portion size choices and practical implications through identifying a potential cause of overeating and/or food waste.
... In another stream of papers, the caloric knowledge of consumers and the relationship with individual's food consumption has been vastly examined in highly cited publications. Chandon and Wansink (2007) argue that underestimation of calories is positively associated with meal size, which very likely stimulates obesity. However, knowledge of the consequences of calories and obesity has been found to have a curvilinear relationship with individual intention to purchase advertised high-calorie junk food such as snack bars (Andrews et al., 2009). ...
Article
Purpose In spite of wide civic and academic interest in obesity, there are no bibliometric records of this issue in the marketing corpus. Thus, this inquiry is conceived to address this shortcoming with a bibliometric analysis of Scopus indexed articles published on the subject. Design/methodology/approach The analysis followed a five-step science mapping approach of study design, data collection, data analysis, data visualisation and data interpretation. R programming software was used to review 88 peer reviewed journals published between 1987 and 2021. Findings A sizable stream of literature exploring obesity has accrued in the marketing area as authors have drawn parallels between the influence of persuasive communication and advertising on human wellbeing and child health. The United States of America is found to be by far the country with the highest number of publications on obesity, followed by Australia and the United Kingdom. The topic dendrogram indicates two strands of obesity discourse: (1) social and policy intervention opportunities and (2) the effects on social groups in the population. Research limitations/implications This review will shape future enquiries investigating obesity. Beyond the focus on children, males and females, an emerging focus on cola, ethics, food waste, milk, policy-making and students is highlighted. Originality/value This is the first bibliometric review of obesity in the marketing literature. This is especially timely for weighing up the utility of research aimed at understanding and reporting the trends, influences and role of stakeholders in addressing obesity.
... The pictorial warning label showed a food product, equivalent in caloric terms. As in other studies (Chandon and Wansink, 2007a, b), the panel of marketing experts selected a large consumption food product: The Big Mac McDonald's (509 Kcal). For the high caloric content stimulus, the pictorial warning label was used as a whole. ...
Article
Purpose –The purpose of this research is to demonstrate the effectiveness of pictorial warning labels that leverage the risk of obesity as a deterrent against alcohol abuse. It evaluates the impact of three different kinds of warning labels that can potentially discourage alcoholic drinking: (1) a claim, in text format, that cautions consumers about the product (i.e. a responsibility warning statement); (2) a textual warning label, text-format information on the content of the product or the consequences of excessive consumption (i.e. a synthetic nutritional table); (3) a pictorial warning label, an image depicting a food product with a caloric content equivalent to that of an alcoholic beverage. Design/methodology/approach – In Study 1, a 2 x 2 x 2 factorial design is used to evaluate the intention to buy different alcoholic cocktails. The stimuli comprised two cocktails that are similar in alcoholic volume, but different in their caloric content. The images of the products were presented across eight warning label conditions and shown to 480 randomly selected Italian respondents who quantified their intention to buy the product. In Study 2, a different sample of 34 Italian respondents was solicited with the same stimuli considered in Study 1, and neuropsychological measurements through Electroencephalography (EEG) were registered. A post hoc least significance difference (LSD) test is used to analyse data. Findings – The results show that only the presence of an image representing an alcoholic beverage’s caloric content causes a significant reduction in consumers’ purchase intentions. This effect is due to the increase in negative emotions caused by pictorial warning labels. Originality/value –The findings provide interesting insights on pictorial warning labels, which can influence the intention to purchase alcoholic beverages. They confirmed that the use of images in the warning labels has a greater impact than text, and that the risk of obesity is an effective deterrent in encouraging consumers to make healthier choices.
... Thus, for food manufacturers who are criticized for producing unhealthy foods with misleading nutrient-content claims, it might be more ethical and profitable to combine nutrient-content claims and smaller serving sizes rather than emphasizing nutrient-content claims only. This recommendation is consistent with past recommendations that unhealthy foods should be packaged in smaller portions to enhance health dietary behaviors (e.g., Chandon & Wansink, 2007a, 2007bMohr et al., 2012). ...
Article
Guided by the theoretical frameworks of the “health halo effect” and consumer expertise, this research was undertaken to determine how two individual factors, health consciousness and health literacy, differentially influence evaluations of nutrient-content claimed messaging for an unhealthy food (i.e., chocolate) and whether such evaluations are moderated by the reading of the food’s Nutrition Facts Label displaying different serving sizes. The research found that health consciousness positively influenced evaluative responses to unhealthy food messaging, and that the positive influence persisted following the reading of a Nutrition Facts label listing a large quantity of unhealthy ingredients per serving size. In contrast, health literacy negatively influenced perceived healthiness and purchase intention when the nutrition label communicated a higher serving size, indicating that subjective and objective expertise work differently. The results advance understanding of the information processing of nutrient-claimed unhealthy foods, and suggest implications for food marketing communication and public health policy.
Article
Purpose Building on the social marketing theory, this study aims to examine the relationship between family units and obesity in Nigeria; and the social marketing interventions used to reduce and prevent obesity in the Nigerian society. Design/methodology/approach This study adopted a semi-structured interview research design with 42 obese individuals in Nigeria. Findings The study findings show that the family unit an individual grows up in influences their consumption behaviour, which drives their obesity. The findings reveal that obese Nigerian citizens are willing to live a healthier lifestyle due to the direct and indirect medical costs associated with obesity. Furthermore, the findings disclose the social marketing interventions – local celebrity endorsements, healthy lifestyle promotions, reduced gym membership and affordable access to healthy foods and services – used to prevent and reduce the rising obesity rates in the Nigerian society. Research limitations/implications The findings have important theoretical implication given the focus on consumption behaviour and obesity. Practical implications The study findings provide an avenue to guide government officials, policymakers and social marketers in shaping their public policy and social marketing interventions to encourage healthier consumption and lifestyle behaviours among families and individuals in the Nigerian society. Originality/value To the best of the authors’ knowledge, this is the first research study to investigate how family units in the emerging market of sub-Saharan Africa drive obesity and the social marketing interventions used to reduce and prevent obesity. Theoretical and practical implications are discussed.
Article
Consumers seeking to maintain a healthy diet commonly estimate their daily caloric intake. Despite the importance of accurate calorie estimation, contextual and situational factors significantly influence consumers’ calorie estimations. The current research investigates how the number of items listed on a menu systematically influences consumers’ calorie estimations. Four experiments show that consumers estimate an item’s caloric content as greater (less) in a menu listing more (fewer) items. The results indicate that consumers are more likely to form a wider (vs. narrower) range of calorie estimates for a larger (vs. smaller) menu, leading them to estimate a target item’s caloric content as greater in a larger menu. We suggest that the number of menu items can prime a sense of magnitude, which in turn affects one’s calorie estimation process.
Article
The construct of involvement has been shown to affect brand loyalty, diffusion of innovations, responses to advertising, and finally, food purchase choices. Despite the recognized importance of this construct in the food market, there is no agreement about the best way to define and measure it. This systematic review provides an overview of currently available involvement in food measures in order to understand the psychological domains covered and to assess their psychometric properties. Comprehensive searches of three electronic databases were conducted in October 2020. Studies were considered that aimed at developing a measure of involvement in food (FI) or that assessed at least one measurement property of involvement in food measures. Methodological quality of studies was assessed with the COnsensus-based Standards for the selection of health Measurement INstruments checklist. The titles and abstracts of 4,160 articles were screened, with 258 full-text articles assessed for eligibility. Of these, 36 studies were identified as meeting the study criteria, 19 of which were measure development studies. A range of FI entries was captured by included measures. These were classified into five psychological domains: affective, self-expressive, situational, cognitive and behavioral. Regarding the psychometric quality of measures, the results highlight that the scientific quality of most instruments is doubtful or inadequate. Future research should focus on reaching a shared definition of FI, oriented by a solid psychological analysis of the phenomenon, in order to develop a comprehensive scale able to generate rigorous, comparable and readable results.
Article
Objective: We identify individuals who set daily intake budgets and examine if an intervention making people estimate their calorie intake up to a certain point in the day helps those setting daily budgets to regulate their calorie intake for the remainder of the day, after high prior consumption. Design: We conducted an online experiment in five countries: Australia, China, Germany, India, and the UK (n = 3,032) using a 2 (setting calorie budget: yes vs. no, measured) x 2 (intervention: intake reminder vs. control, manipulated) between-subjects design, with the amount of prior consumption measured. Participants were contacted in the afternoon. Those in the intervention condition were asked to estimate their prior calorie intake on that day. Main outcome measures: We measured the individual characteristics of those who set daily calorie budgets and the intended calorie intake for the remainder of the day. Results: Among people who set daily calorie budgets, the intervention reduced intended calorie intake for the remainder of the day by 176 calories if they had already consumed a high amount of calories that day. Conclusion: A timely intervention to estimate one's calorie intake can lower additional intended calorie intake among those who set daily calorie budget.
Thesis
The consequences of unhealthy eating are one of today’s most important societal issues. Accordingly, a growing area of research has started to examine how marketer-controlled variables impact food consumption. In this dissertation, I first highlight how food consumption research may benefit from more targeted research from a theoretical lens of motivated reasoning. Then, I empirically examine how two specific marketing actions—serving food to consumers versus letting them serve themselves, and serving portions that lead to larger versus smaller amounts of food leftovers—influence the extent to which consumers can downplay unhealthy eating, which in turn encourages unhealthier choices and behaviors. Focusing on processes that take place when consumers obtain their food, I find that whether oneself (versus a server) serves the food determines the opportunity for self-serving attribution of responsibility for one’s eating, such that being served enables, but serving oneself disables, rejection of responsibility. Through rejecting responsibility, and consequently feeling better about oneself, being served food encourages consumers to choose unhealthy options as well as larger portions. Examining the period after consumers have completed their meal, I find that larger (versus smaller) amounts of food leftovers reduce perceived consumption, which improves consumers’ self-evaluative feelings and dampens their motivation to compensate for their food consumption, as manifested in greater consumption and lesser exercise effort subsequently. Theoretical contributions and managerial and policy implications are discussed.
Article
Full-text available
This paper describes the Observing Protein and Energy Nutrition (OPEN) Study, conducted from September 1999 to March 2000. The purpose of the study was to assess dietary measurement error using two self-reported dietary instruments-the food frequency questionnaire (FFQ) and the 24-hour dietary recall (24HR)-and unbiased biomarkers of energy and protein intakes: doubly labeled water and urinary nitrogen. Participants were 484 men and women aged 40-69 years from Montgomery County, Maryland. Nine percent of men and 7% of women were defined as underreporters of both energy and protein intake on 24HRs; for FFQs, the comparable values were 35% for men and 23% for women. On average, men underreported energy intake compared with total energy expenditure by 12-14% on 24HRs and 31-36% on FFQs and underreported protein intake compared with a protein biomarker by 11-12% on 24HRs and 30-34% on FFQs. Women underreported energy intake on 24HRs by 16-20% and on FFQs by 34-38% and underreported protein intake by 11-15% on 24HRs and 27-32% on FFQs. There was little underreporting of the percentage of energy from protein for men or women. These findings have important implications for nutritional epidemiology and dietary surveillance.
Article
Full-text available
Given the number of volume judgments made by consumers, for example, deciding which package is larger and by how much, it is surprising that little research pertaining to volume perceptions has been done in marketing. In this article, the authors examine the interplay of expectations based on perceptual inputs versus experiences based on sensory input in the context of volume perceptions. Specifically, they examine biases in the perception of volume due to container shape. The height of the container emerges as a vital dimension that consumers appear to use as a simplifying visual heuristic to make a volume judgment. However, perceived consumption, contrary to perceived volume, is related inversely to height. This lowered perceived consumption is hypothesized and shown to increase actual consumption. A series of seven laboratory experiments programmatically test model predictions. Results show that perceived volume, perceived consumption, and actual consumption are related sequentially. Furthermore, the authors show that container shape affects preference, choice, and postconsumption satisfaction. The authors discuss theoretical implications for contrast effects when expectancies are disconfirmed, specifically as they relate to biases in visual information processing, and provide managerial implications of the results for package design, communication, and pricing.
Article
Objective A bogus pipeline paradigm was utilized to assess whether food intake underreporting by unsuccessful dieters is intentional. Method: Twenty‐eight subjects completed 1‐week food diaries. Then, 17 subjects in the experimental condition kept 2‐week food diaries while being told the researcher was verifying their report. Eleven subjects in the control group were asked merely to self‐monitor for two more weeks. Results: Results indicate that subjects in the experimental group reported significantly greater intake than control subjects, when controlling for reported intake during the screening phase and weight change. Discussion: Thus, the belief that the researcher could verify their report improved the accuracy of patients' self‐report. However, all subjects continued to underreport their dietary intake. In summary, underreporting may be an intentional attempt to manage presentation to others in a society that is increasingly critical of overweight persons. © 1998 by John Wiley & Sons, Inc. Int J Eat Disord 24: 259–266, 1998.
Article
Background: Underreporting of energy intake is associated with self-reported diet measures and appears to be selective according to personal characteristics. Doubly labeled water is an unbiased reference biomarker for energy intake that may be used to assess underreporting. Objective: Our objective was to determine which factors are associated with underreporting of energy intake on food-frequency questionnaires (FFQs) and 24-h dietary recalls (24HRs). Design: The study participants were 484 men and women aged 40-69 y who resided in Montgomery County, MD. Using the doubly labeled water method to measure total energy expenditure, we considered numerous psychosocial, lifestyle, and sociodemographic factors in multiple logistic regression models for prediction of the probability of underreporting on the FFQ and 24HR. Results: In the FFQ models, fear of negative evaluation, weight-loss history, and percentage of energy from fat were the best predictors of underreporting in women (R(2) = 0.09); body mass index, comparison of activity level with that of others of the same sex and age, and eating frequency were the best predictors in men (R(2) = 0.10). In the 24HR models, social desirability, fear of negative evaluation, body mass index, percentage of energy from fat, usual activity, and variability in number of meals per day were the best predictors of underreporting in women (R(2) = 0.22); social desirability, dietary restraint, body mass index, eating frequency, dieting history, and education were the best predictors in men (R(2) = 0.25). Conclusion: Although the final models were significantly related to underreporting on both the FFQ and the 24HR, the amount of variation explained by these models was relatively low, especially for the FFQ.
Article
The Centers for Disease Control has declared obesity a public health epidemic: More than 30% of U.S. adults are obese, and obesity now equals smoking as the leading preventable cause of disease and death. The authors explore policy issues associated with the accelerated growth of obesity in the U.S. population, particularly policy related to the debated influence of food marketing practices on obesity. The authors discuss possible market failures that influence consumer food choices and address the role of existing informational and regulatory policies in moderating the alleged threat of food marketing practices to public health. They consider various types of policy remedies that have been proposed as ways to reduce societal obesity costs, and they offer an agenda for further research to address knowledge gaps that represent barriers to effective public policy decisions.
Article
This paper describes the Observing Protein and Energy Nutrition (OPEN) Study, conducted from September 1999 to March 2000. The purpose of the study was to assess dietary measurement error using two self-reported dietary instruments-the food frequency questionnaire (FFQ) and the 24-hour dietary recall (24HR)-and unbiased biomarkers of energy and protein intakes: doubly labeled water and urinary nitrogen. Participants were 484 men and women aged 40-69 years from Montgomery County, Maryland. Nine percent of men and 7% of women were defined as underreporters of both energy and protein intake on 24HRs; for FFQs, the comparable values were 35% for men and 23% for women. On average, men underreported energy intake compared with total energy expenditure by 12-14% on 24HRs and 31-36% on FFQs and underreported protein intake compared with a protein biomarker by 11-12% on 24HRs and 30-34% on FFQs. Women underreported energy intake on 24HRs by 16-20% and on FFQs by 34-38% and underreported protein intake by 11-15% on 24HRs and 27-32% on FFQs. There was little underreporting of the percentage of energy from protein for men or women. These findings have important implications for nutritional epidemiology and dietary surveillance.
Article
Although considerable research exists on consumer processing of nutrition labeling and package claims, less is known about consumer interpretation of nutrient content claims in advertising. This is important because product advertising often provides a significant first step for consumers in learning new nutrition information. Yet, unlike package claims, Nutrition Facts Panels are often not available for consumers during the processing of such advertising claims. Therefore, the authors examine the following research questions: (1) Do consumers misinterpret (i.e., overgeneralize) common nutrient content claims in advertising? If so, under what conditions does this occur? and (2) Can various types of disclosure statements remedy this problem? To address these questions, the authors interview a total of 365 primary food shoppers in three geographically dispersed malls in the United States in a between-subjects experiment. Misleading generalizations, beyond those of control ad claims, are found for general and specific nutrient content claims. Ad disclosure type, and claim type, and nutrition knowledge all separately influence nutrient content and disease risk measures. Evaluative disclosures reduce misleading generalizations to a greater extent than do absolute or relative disclosures. The authors offer implications for public policy and food marketers.
Article
The author investigates whether using a decompositional question (which decomposes an event into subcategories and elicits frequencies at the subcategory level) is effective in increasing the accuracy of frequency judgments elicited in consumer surveys. Results of a study show that the decompositional question makes the process of eliciting frequencies less effortful and enhances the accuracy of the elicited frequencies for frequent, irregular behaviors (i.e., occurring at sporadic intervals), but not for frequent, regular behaviors. Mediational analyses confirm that these effects manifest because the decompositional question triggers an episodic recall strategy, which enhances the efficiency of the judgment formulation process for irregular behaviors but interferes with the normal process for regular behaviors.