Appetite 58 (2012) 277–284
Contents lists available at SciVerse ScienceDirect
Appetite
journal homepage: www.elsevier.com/locate/appet
Research report
Moderation of distress-induced eating by emotional eating scores q
Tatjana van Strien a,b,⇑, C. Peter Herman c, Doeschka J. Anschutz a, Rutger C.M.E. Engels a,
Carolina de Weerth a
a
Behavioural Science Institute, Radboud University Nijmegen, The Netherlands
Institute for Gender Studies, Radboud University Nijmegen, The Netherlands
c
Department of Psychology, University of Toronto, Toronto, Ontario, Canada M5S 3G3
b
a r t i c l e
i n f o
Article history:
Received 17 May 2011
Received in revised form 4 October 2011
Accepted 10 October 2011
Available online 15 October 2011
Keywords:
Emotional eating
Moderator effect
Snack intake
Predictive validity
Distress-induced food intake
Self-report
a b s t r a c t
Earlier studies assessing the possible moderator effect of self-reported emotional eating on the relation
between stress and actual food intake have obtained mixed results. The null findings in some of these
studies might be attributed to misclassification of participants due to the use of the median splits and/
or insufficient participants with extreme scores. The objective of the two current studies was to test
whether it is possible to predict distress-induced eating with a self-report emotional eating scale by using
extreme scorers. In study 1 (n = 45) we used a between-subjects design and emotional eating was
assessed after food intake during a negative or a neutral mood (induced by a movie). In study 2
(n = 47) we used a within-subjects design and emotional eating was assessed well before food intake,
which occurred after a control or stress task (Trier Social Stress Task). The main outcome measure was
actual food intake. In both studies self-reported emotional eating significantly moderated the relation
between distress and food intake. As expected, low emotional eaters ate less during the sad movie or after
stress than during the neutral movie or after the control task, whereas high emotional eaters ate more. No
such moderator effect was found for emotional eating in the entire sample (n = 124) of study 1 using the
median-split procedure or the full range of emotional eating scores. We conclude that it is possible to predict distress-induced food intake using self-reports of emotional eating provided that the participants
have sufficiently extreme emotional eating scores.
Ó 2011 Elsevier Ltd. All rights reserved.
Introduction
Negative mood or distress is associated with both increased and
decreased food intake (Greeno & Wing, 1994) with eating less
being the typical and predominant response (Gold & Chrousos,
2002; Heatherton, Herman, & Polivy, 1991; Stone & Brownell,
1994). Negative mood or distress is normally associated with a
hyperactive hypothalamic–pituitary–adrenal (HPA) axis, with
physiological adaptations that are biologically designed to prepare
the individual for a fight-or-flight response. These adaptations include inhibition of gastric motility and the release of sugar into
the bloodstream, thereby suppressing feelings of hunger (Gold &
Chrousos, 2002). However, so-called emotional eaters show an
atypical response and eat the same amount or even more during
distress (Oliver, Wardle, & Gibson, 2000; Van Strien & Ouwens,
2003).
q
Acknowledgment: The two studies were financially supported by Radboud
University Nijmegen. Tatjana van Strien has copyright and royalty interest in the
DEBQ. No conflicts of interest are reported by C. Peter Herman, Doeschka J.
Anschutz, Rutger C.M.E. Engels, and Carolina de Weerth.
⇑ Corresponding author.
E-mail address: [email protected] (T. van Strien).
0195-6663/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.appet.2011.10.005
Emotional eating is thought to be a learned response (Bruch,
1973). It was found to emerge in adolescence in association with
depressive feelings and inadequate parenting (e.g., Ouwens, Van
Strien, & Van Leeuwe, 2009; Snoek, Engels, Janssens, & Van Strien,
2007) in interaction with a genetic vulnerability (Van Strien, Snoek,
van der Zwaluw, & Engels, 2010; Van Strien, van der Zwaluw, &
Engels, 2010). Emotional eating was found to occur in people with
poor coping skills, poor interoceptive awareness, and high alexithymia (Larsen, Van Strien, Eisenga, & Engels, 2006; Ouwens
et al., 2009; Spoor, Bekker, Van Strien, & Van Heck, 2007; Van Strien & Ouwens, 2007). Emotional eaters have been observed to be
poor at recognizing whether they are hungry or satiated or suffer
from some other discomfort, and to overeat in response to emotional agitation rather than in response to internal hunger cues.
If emotional eating occurs frequently, it may ultimately lead to
weight gain and obesity.
Studies assessing the possible moderator effect of self-reported
emotional eating on the relation between stress and actual food
intake have obtained mixed results. In a laboratory study on the
effects of a distress manipulation (the anticipation of a public
speaking task), Oliver et al. (2000) found a significant moderator
effect of self-reported emotional eating on food choice and food
278
T. van Strien et al. / Appetite 58 (2012) 277–284
intake. Stressed emotional eaters ate more sweet, high-fat foods
and a more energy-dense meal than did unstressed eaters and
non-emotional eaters in the distress condition. Also Wallis and
Hetherington (2004) found that emotional eating was associated
with greater chocolate intake after an ego-threatening stressor relative to a control task. No such moderator effect for emotional eating on actual food intake, however, was found by Evers, de Ridder,
and Adriaanse (2009) and Raspopow, Abizaid, Matheson, and
Anisman (2010) (n = 48), or Wallis and Hetherington (2009) (n = 26).
Evers et al. (2009) used four different emotion–induction procedures in four different experiments (vignettes, n = 30; film excerpts, n = 60; recall, n = 37; false feedback, n = 57). Failing to find
effects for emotional eating, Evers et al. (2009) argued that assessing oneself as an emotional eater is a ‘‘mission impossible’’ because
it may be too difficult for individuals to assess their own emotionrelated eating behavior.
Another possible explanation for the null results in some studies
is misclassification of ‘low’ vs. ‘high’ emotional eaters due to (a) the
use of a median split on the emotional-eating scale or (b) insufficient
participants with extreme emotional eating scores or (c) both. Median splits are notoriously vulnerable to misclassification of research
participants and spurious effects (Maxwell & Delaney, 1993).
Oliver et al. (2000) and Wallis and Hetherington (2004), who
did find a significant moderator effect for emotional eating, also
used a median-split procedure for their classification of participants into low and high emotional eaters, though, so a median split
approach to participant classification is not necessarily fatal. Further, some studies with no significant moderator effect for emotional eating (additionally) used multiple regression analyses on
the full range of scores. Another possible explanation for the null
results in some studies may be that there had been insufficient
participants with extreme scores on the moderating variable.
According to McClelland and Judd (1993), pp. 382–383 ‘‘jointly extreme observations are crucial for detecting interactions.’’ In comparison with the ‘four-corners design,’ in which 25% of cases are
allocated to each extreme, ‘‘a normal like distribution of the two
variables has a relative efficiency of only .06 for detecting an interaction and requires nearly 17 times as many observations to have
comparable efficiency’’ (Whisman & McClelland, 2005, p. 117).
Interestingly, most studies with no moderator effect for emotional
eating that (additionally) used multiple regression analyses only
had a small number of subjects in their study.1
In the following two studies we experimentally tested whether
emotional eating moderates the effect of distress on eating by
using only subjects with scores from the extreme ends of the DEBQ
emotional eating subscale (Van Strien, 2011). In the first study, ‘‘an
examination of general TV viewing behavior in daily life’’, we analyzed data from a study (Anschutz, Van Strien, & Engels, 2008) of
participants’ snack intake during exposure to a negative vs. a neutral film. In addition to analyses of extreme scores, we also conducted analyses on the entire sample using a median split
classification and the full range of scores on the DEBQ emotional
eating subscale. This study used a between-subjects design, and
emotional eating was assessed after the experimental procedure.
In the second study, an investigation of ‘‘health and physiology’’,
we analyzed food intake following a laboratory control task or
stress task (Trier Social Stress Test; TSST). In this study we used a
within-subjects design in which all subjects performed the control
task and the acute stress task on two consecutive days. A further
difference was that emotional eating was assessed well before
the experiment, by recruiting participants from a pool of female
students taking introductory psychology or pedagogy courses
who had completed the emotional eating scale and then inviting
only those with scores at the extreme ends to participate in a study
on ‘health and physiology’.
Both studies included only females, because there is a greater
prevalence of stress-induced food intake in females (O’Connor
et al., 2008). We hypothesized that low emotional eaters (i.e., people who score low on the DEBQ emotional eating subscale) would
eat less in the negative mood or acute stress condition than in the
neutral or control condition, whereas high emotional eaters (i.e.,
those who score high on the DEBQ emotional eating subscale)
would eat more.
1
This observation also provides a possible explanation for O’Connor, Jones, Conner,
McMillan, and Ferguson’s (2008) finding a significant moderator effect for emotional
eating in a regression analysis even though they used the entire continuum of
emotional eating. O’Connor et al.’s diary study of the effects of daily hassles and
eating style on food intake had 422 participants! This line of reasoning would also
apply to the results of Conner, Fitter, and Fletcher (1999), who found no moderator
effect for emotional eating in a highly similar study with only 60 participants.
O’Connor and O’Connor (2004), used 155 participants in their diary study and a
median split classification for low vs. high emotional eating but their results were in
the predicted direction (For the present article we restrict ourselves to studies that
assessed the moderation of distress-induced eating by emotional eating scores).
2
Preliminary results with other cut-of points of the first (of three) analyses of
Study 1 have been reported in a Letter to the Editor of Health Psychology (Van Strien,
2010).
3
In their fMRI study, Bohon, Stice, and Spoor (2009) used 1.6 and 3.6 as qualifying
scores for low vs. high DEBQ emotional eaters. Some other researchers prefer the
median split ±1SD. However with a median of 2.68 ± .78 for emotional eating in Study
1, the cut-off score for low emotional eaters would have been exactly the same as our
present cut-off point, and the score for high emotional eaters would have been even
higher than the score that we used. So we preferred to use cut-off scores derived from
the DCOTAN (Dutch Committee of Test and Testing) representative Dutch norm
sample.
Study 1
Methods
Participants
The sample consisted of 124 female students at the Radboud
University Nijmegen. They participated in exchange for money or
course credits. We also performed analyses on those participants
with scores at the extreme ends of the emotional eating subscale
of the DEBQ (Van Strien, 2011), using scores below the 20th and
above the 80th percentile in the Dutch norm group of females.2
Corresponding cut-points are <1.82 and >3.25.3 A total of 45 female
students fulfilled these criteria, 23 low and 22 high emotional eaters.
The mean and standard deviations (SD) of age, body mass index
(BMI = weight (kg)/height (m2)) and scores on emotional, external
and restrained eating of the total sample and the subsamples may
be found in Table 1.
Procedure
The study was presented as an examination of general TV viewing behavior in daily life. To increase ecological validity, we tested
the subjects in a specially equipped, relaxing room at our lab, with
a comfortable couch and a big TV screen. On the side table in front
of the couch stood a glass of water, two pre-weighed bowls with
crisps and M&Ms, and a glass of water. Participants were informed
that they would watch TV for about 45 min and were invited to
drink or eat whatever they liked while watching. To make sure that
everyone would eat something, the experimenter asked them to
eat at least one piece of one type of snack food.
Half of the participants viewed a sad movie, ‘‘Breaking the
Waves,’’ and the other half viewed a neutral movie, ‘‘Travelling
Birds.’’ After 14 and 33 min, the movies were interrupted by a
3.5-min commercial break, containing either neutral ads (e.g., promoting a car or a video camera) or ads with slim models or diet
products (e.g., promoting shaving gel or low-fat cheese). For the
present study, these ads were not of relevance but in the statistical
279
T. van Strien et al. / Appetite 58 (2012) 277–284
Table 1
Characteristics of total sample and the subsamples of low and high emotional eaters.
Age
BMI
Emotional eating
External eating
Restrained eating
a
Total sample (n = 124)
Low emotional eating (n = 23)
High emotional eating (n = 22)
Mean
SD
Mean
SD
Mean
SD
21.8
23.3
2.6
3.2
2.8
3.6
3.7
.8
.5
.8
23.3
21.9
1.4
2.7
2.5
6.1
3.4
.3
.5
.7
21.3
24.7
3.7
3.4
3.3
2.9
5.0
.4
.6
.6
Cohen’s da
0.44
0.66
6.76
1.29
1.23
Effect size of the difference between low and high emotional eaters by Cohen’s d (0.20 = small, 0.50 = medium and 0.80 = large).
analyses the possible influence of ad-type was controlled for. After
the end of the movie clip, participants completed several questionnaires concerning: the manipulation checks; their attitudes towards the movie and the ads; recall; and an assessment of their
eating behavior. Weight and height were measured, (Larsen,
Ouwens, Engels, Eisinga, & Van Strien, 2008) and a checklist
concerning the purpose of the experiment was presented. Finally,
participants were thanked and paid. They were fully debriefed
after data collection was finished.
interrelations between emotional eating, external eating and
dietary restraint (.45 < rs < .54, p < .001), we corrected for external
eating and dietary restraint in all analyses. Further, the possible
confounding effect of ad type, BMI, time since eating and hunger
before the experiment were controlled for. To avoid multicollinearity in the regression analyses, all variables were centered before
computing interaction terms (Aiken & West, 1991).
Results
Measures
Mood: Mood was measured at the end of the study. To check
whether the emotion manipulation had the desired outcome, we
used a visual analog scale (VAS) to measure sadness. We also measured sadness with the item ‘sad’ (a 5-point rating scale). We further asked the participants directly whether they felt better or
worse after the movie by asking them to state how well the following statements applied to them: I felt worse after seeing the movie
and I felt better after the movie (5-point rating scales).
Eating behavior: Eating behavior was assessed with the DEBQ
(Van Strien, 2011). The DEBQ has 33 items, 13 on emotional eating
(e.g., ‘‘Do you have a desire to eat when you are irritated?’’), 10 on
external eating (e.g., ‘‘If food smells and looks good, do you eat
more than usual?’’) and 10 on restrained eating (e.g. ‘‘Do you try
to eat less at mealtimes than you would like to eat’’?). All items
have to be rated on a 5-point scale with response categories that
range from 1 ‘never’ to 5 ‘very often.’ The scales have good internal
reliability and good construct and predictive validity (Van Strien,
Herman, & Anschutz, in press; Van Strien & van de Laar, 2008).
Hunger: Because food intake may be affected by deprivation, we
measured individual differences in hunger by having participants
complete a visual analog scale (VAS) measuring the extent to
which they felt hungry or satiated before the experiment. Additionally, we asked the respondents at what time they had last
eaten.
Food intake: While watching the movie, participants were allowed to eat freely from two pre-weighed bowls of crisps or
M&Ms. The amount of food eaten during the experiment was measured with a professional balance (Mettler PM3000) to the nearest
0.1 g. We used total caloric intake as the dependent variable instead of total grams of food consumed, because chips and M&Ms
differ in weight and caloric value.
Analytic strategy
All analyses were carried out using SPSS version 15.0 (SPSS Inc.,
Chicago). Success of randomization and the manipulation was assessed by using one-way ANOVAs and calculating effect sizes (Cohen’s d). Second, with ANCOVAs and hierarchical regression
analyses we tested the moderating effect of emotional eating on
the relation between movie condition and total intake of food. A
significant moderator effect for emotional eating would be demonstrated by a significant interaction between emotional eating and
the movie condition, whether or not there is a main effect for the
moderator variable (emotional eating). Because of the high
Randomisation check, check of mood induction and outliers
In both the extreme-score subsample and the total sample there
were no significant main effects of the movie condition for time
since last having eaten before the experiment or for BMI, indicating
that random assignment to the movie conditions had been successful (extreme score subsample: mean [sd]:time since last having eaten before the experiment was 1.64 [1.02] vs. 1.49 [1.24], p = .66;
Cohen’s d = 0.13; BMI: 23.88 [5.72] vs. 22.84 [2.89], p = .44; Cohen’s
d = 0.24; total sample: mean [sd]:time since last having eaten before the experiment was 1.89 [1.40] vs. 1.62 [1.26], p = .26; Cohen’s
d = 0.19; BMI: 23.67 (4.32) vs. 22.98 (2.85), p = .0.299; Cohen’s
d = 0.19). There were, however, significant main effects of the movie condition on measures of negative moods, indicating that the
mood manipulation was successful: extreme score subsample:
mean [sd]: feeling worse after the movie: 0.27 [0.55] vs. 1.48
[1.44], p < .001; Cohen’s d = 1.21; feeling better after the movie:
1.41 [1.14] vs. 0.30 [0.56], p < .001; Cohen’s d = 1.32; VAS sadness:
2.64 [2.88] vs. 5.59 [2.89], p < .001; Cohen’s d = 1.05) (total sample:
mean [sd]: feeling worse after the movie: 0.26 [0.65] vs. 1.77
[1.45], p < .001; Cohen’s d = 1.43; feeling better after the movie:
1.50 [1.18] vs. 0.35 [0.66], p < .001; Cohen’s d = 1.25; VAS sadness:
2.56 [2.52] vs. 6.28 [2.73], p < .001; Cohen’s d = 1.41).
Next, food intake data were scrutinized for outliers, defined as
>mean + 3SD, and for skewness. There was one outlier with a kcal
intake value of 1510.57. This value was winsorised by replacing the
outlying value with the value of 3SD above the mean
(kcal = 1361.90). After this, no problems with skewness were
observed.
Moderator effects
In an ANCOVA on the extreme score subsample (controlling for
BMI, hunger, time since last eaten, ad type, dietary restraint and
external eating), we tested the moderating effect of emotional eating on the relation between movie condition and food intake. The
control variables BMI and hunger had no significant main effects
on food intake (kcal), so a reduced ANCOVA was run (controlling
for time since last having eaten, ad type, dietary restraint and
external eating). There was a significant moderator effect of emotional eating on the relation between movie condition and food intake (kcal), F(1,37) = 9.34, p = .004, partial eta squared = .202. As
expected, low emotional eaters ate less during the sad movie than
during the neutral movie (adjusted means [SE]: 440.32 [68.46] vs.
667.16 [87.68]) whereas high emotional eaters ate more (441.13
280
T. van Strien et al. / Appetite 58 (2012) 277–284
[74.83] vs. 239.28 [77.41]). In this analysis, 57.6% of the variance
was explained by the model.
As recommended by Holmbeck (2002) we used a post-hoc hierarchical regression analysis on the final model to determine the
nature of this significant interaction effect between the dichotomous variable low vs. high emotional eating (dummy coded as 0
and 1) and movie condition (neutral movie = 0; sad movie = 1).
Prior to these analyses, we computed conditional moderator variables, where low emotional eating was assigned a value of 0 in
one analysis and high emotional eating was assigned a value of 0
in the other analysis. The regression for low levels of emotional
eating indicated that movie condition was significantly negatively
associated with total food intake, B = 44.52, p = .038. In contrast,
results of the regression for high levels of emotional eating indicated that movie condition was borderline significantly positively
associated with total food intake, B = 37.43; p = .062. In other
words, low emotional eaters ate less during the sad movie than
during the neutral movie whereas high emotional eaters tended
to eat more during the sad movie than during the neutral movie.
This moderator effect for emotional eating remained significant
when a reduced ANCOVA without the control variable ad type was
run. F(1,38) = 6.92, p = .012, partial eta squared = .154. The moderator effect of emotional eating remained significant when additionally the control variable dietary restraint was removed from the
model (F(1,39) = 4952, p = .032, partial eta squared = .113). However,
the moderator effect of emotional eating became insignificant
when additionally the control variable external eating was removed as a covariate from the model, F(1,40) = 1581, p = .216, partial
eta squared = .038.
In additional analyses we tested the notion that the absence of a
moderator effect for emotional eating in some studies can be
attributed to misclassification of ‘‘low emotional eaters’’ vs. ‘‘high
emotional eaters’’ due to the use of median splits. To demonstrate
this problem, we used the full sample (n = 124). Using the mediansplit procedure (cut point = 2.615), 58 females were classified as
‘‘low emotional eaters’’ and 66 females were classified as ‘‘high
emotional eaters.’’ In an ANCOVA (controlling for BMI, hunger, time
since last having eaten, ad type, dietary restraint and external eating), we tested the moderating effect of emotional eating on the
relation between movie condition and food intake. The controlling
variables BMI, time since last having eaten and ad type had no significant main effects on total intake of food (tot kcal), so for our
final analysis a reduced ANCOVA was run (controlling for hunger–satiety, dietary restraint and external eating). There was no
moderator effect of emotional eating on the relation between movie condition and food intake, F(1117) = 0.121, p = .65. This analysis
supports the possibility that the absence of a moderator effect in
earlier studies might be explained by misclassification of ‘‘low’’
and ‘‘high’’ emotional eaters due to the use of median splits.
Alternatively, perhaps scores near the median, while not exactly
‘‘misclassified,’’ may still be inadequately extreme to capture the
essence of a ‘‘high’’ or ‘‘low’’ emotional eater. (see Scatterplot 1
and 2 of Supplementary data).
There also was no significant moderator effect of emotional eating when we used, in a hierarchical regression analysis, the full
range of scores of emotional eating (controlling for BMI, hunger,
time since last having eaten, ad type, dietary restraint and external
eating), B = 23.47, p = .675. This analysis supports the possibility
that the absence of a moderator effect in earlier studies might also
be explained by an insufficient number of participants with extreme values on both variables contributing to the interaction.
Discussion
The main finding was that self-reported emotional eating significantly moderated the relation between stress and snack-food
intake when we used extreme scores of the emotional eating scale
to dichotomize people into ‘‘low emotional eaters’’ vs. ‘‘high emotional eaters.’’ As expected, the food intake of the low emotional
eaters was lower during the sad movie than during the neutral movie, whereas the food intake of the high emotional eaters was
slightly higher during the sad movie than during the neutral movie.
It should be noted that the moderator effect of emotional eating
disappeared when the control variable external eating was removed from the model. This would suggest that a possible additional explanation for the mixed results regarding the moderator
effect of self-reported emotional eating would be that external eating must be taken into account.
We found no moderator effect for emotional eating when we
classified subjects with the commonly used median-split procedure, even though the sample was in that case more than twice
as large (n = 124 vs. n = 45) or when we used the full range of
scores for emotional eating in the total sample.
Thus it seems that the null findings in some earlier studies
might be attributed to the use of the median-split procedure. Another possibility, not incompatible with the first, is that these studies had insufficient participants with extreme values on emotional
eating included in the interaction.
A strength of Study 1 is that the moderator effect of emotional
eating was tested in an experiment where mood was successfully
manipulated by a sad or a neutral movie in a natural setting. Further strengths were the use of actual food intake as our dependent
variable and the semi-naturalistic setting used in the study.
A limitation of Study 1 is that the DEBQ was administered to
participants after the movie; the amount eaten during the movie
may have influenced responses to the DEBQ items. Administration
of the questionnaire immediately before the movie, however,
might have sensitized participants to the purpose of our study. A
further limitation is that the negative mood induction (a sad film)
may not have been the most effective distress manipulation at eliciting enhanced food intake in high emotional eaters.
Study 2
For the second study we used a distress manipulation that involved a social-evaluative threat, one that has been shown to induce extreme negative affect (Appelhans, Pagoto, Peters, &
Spring, 2010; Newman, O’Connor, & Commer, 2007; Raspopow
et al., 2010). The Trier Social Stress Test (TSST) involves public
speaking before a jury coupled with an arithmetic challenge. In
an earlier study on dieters and non-dieters, anticipation of having
to give a speech in front of an evaluative audience, a so-called egothreat, was found to be effective at suppressing eating in non-dieters and increasing eating in dieters (Heatherton et al., 1991). For
the present study we were interested in finding out whether a similar effect would also be obtained for low and high emotional eaters challenged with the TSST. Also, in the present study only
participants with emotional eating scores at the extreme ends of
the scale were used, but this time emotional eating was assessed
at least 1 month before the experiment was conducted. We used
a within-subjects design in which all participants were subjected
to a control task and a stress task (TSST) on two consecutive days.
In addition to a possible emotional eating by stress interaction effect on food intake, we were also interested in a possible emotional
eating by stress interaction effect on post-task hunger. We expected that low emotional eaters would show the typical reaction
to stress and report less hunger after the stress task than after the
control task, but we were unsure whether high emotional eaters
would report similar or higher amounts of hunger after the stress
task than after the control task.
T. van Strien et al. / Appetite 58 (2012) 277–284
Method
Participants
Participants were recruited from a pool of female students taking introductory psychology or pedagogy courses who had completed the emotional eating scale. Females with scores below
1.82 or above 3.25 (corresponding to the 20th and 80th percentiles
of the Dutch norm group of females) were invited to participate in
a study on ‘health and physiology.’ A total of 47 female students
agreed to participate, 23 low and 24 high emotional eaters. Their
mean age was 19 years (range 18–27 years) and they had a mean
body mass index of 21.27 (SD = 2.66). The study protocol was approved by the ethical board of the Faculty of Social Sciences of
the Radboud University Nijmegen.
Procedure
Two laboratory sessions were completed on consecutive weekdays. Participants were instructed to wake up at least two and a
half hours before the experiment and to refrain from intake of alcohol or drugs. For 1 h prior to the experiment they were not allowed
to smoke, to engage in physical exercise (including cycling), to eat
or to drink (an exception was made for water) or to brush their
teeth. Experimental sessions were scheduled between 11 and
15 h to minimize the effects of diurnal rhythms on HPA-axis reactivity.4 Upon arrival on the first day participants were asked to fill
out an informed consent form.
On the first test day, participants were subjected to the control
condition, in which they had to rate six different fabrics (wool, fur,
felt, silk, linen and cotton) on various attributes (e.g., softness,
pleasantness, warmth). The participants were left alone in the
experimental room for 15 min, after which the experimenter took
them to a separate room to fill out several questionnaires at a table
which also held a glass of water and four bowls filled with, respectively, white grapes, pieces of carrot, M&Ms and pieces of butter
cake. The experimenter left the room, saying ‘‘Please help yourself
to the water and the food. You have earned it.’ After 20 min the
experimenter returned to take the participants to another room
to perform a 15-min computer task (not relevant to the present
study).
On the second test day the subjects were subjected to a modified version of the Trier Social Stress Task (Kirschbaum, Pirke, &
Hellhammer, 1993). Briefly, the task consisted of preparing
(5 min) and delivering (5 min) a videotaped speech, followed by
a serial subtraction task (5 min). The speech and subtraction task
were presented in front of a two-person jury who sat behind a table and wore white doctors’ coats. The participant had to stand in
stocking feet on a WiiÓ balance board, presumably to measure kinesthetic activity. After the stress task, the experimenter asked the
participant to wait for the jury’s judgment of the participant’s performance—in this manner the stressfulness of the public speaking
task was extended by a prolonged period of waiting for the results—and to fill out a set of questionnaires. After 15 min the
experimenter returned to communicate a positive judgment by
the jury after which the experimenter took them to a separate
room to fill out a further set of questionnaires. Once again the table
held a glass of water and bowls with food, and participants were
invited to help themselves to the water and the food in the same
words as the previous day.5 After 20 min the experimenter returned
4
As part of an ongoing different and larger study we also collected at various time
points data on cortisol (by taking samples of saliva), but the cortisol measures are
beyond the scope of the present paper.
5
Like Appelhans et al. (2010), who also used the TSST to induce stress, we
measured eating after the threatening experience had concluded, rather than while
the threat was still anticipated. This decision is justified by Appelhans et al.’s finding
that the highest cortisol levels were detected after than just before the speech and
math tasks.
281
to measure the weight and height of the participant in light clothing
and stocking feet. Finally the participant was debriefed, thanked and
paid with course credits.
It should be noted that the experimenter was kept blind to the
emotional eating status of the participants and that all participants
were unaware that their food intake was being measured. Furthermore, we deliberately chose to always start with the control condition and not to counterbalance the order of the two conditions
because the stress condition was perceived by some subjects as extremely stressful (some participants even burst into tears). We
were afraid that the control condition would suffer from carry-over
effects if we started with the stress condition and also that we
would lose too many participants because they would refuse to
come back the following day for the control condition.
Measures
Food intake: Before and after participants ate, the bowls with
grapes, carrots, M&Ms and butter cakes were weighed with a professional balance (Kern 200). We used total caloric intake as the
dependent variable because the various types of food differed in
weight and caloric value.
Affect and hunger: On both days, ratings of affect and hunger
upon arrival and post-task (but before food intake) were measured.
The Positive and Negative Affect Schedule (PANAS; Watson, Clark,
& Tellegen, 1988) was used to measure, on a 5-point (‘not at all’ to
‘extremely’) scale, the degree to which participants experienced 10
positive and 10 negative affects. The descriptor ‘hungry’ was inserted among the PANAS items so that hunger could be evaluated
without alerting the participants to the true nature of the study.
For details on the Dutch Eating Behavior Questionnaire, see
Study 1.
Analyses
Intake of food (kcal) in the control and stress condition showed
skewness and kurtosis. Only by computing the difference between
intake in the stress condition and the control condition could we
achieve normality, a positive value meaning a higher intake in
the stress condition. A General Linear Model (GLM) was used with
emotional eating as the between-subjects factor and a difference
score for each type of food as the dependent variable while controlling for pre-task hunger (the mean of pre-task hunger on the control and on the stress day), external eating and dietary restraint. In
a further repeated measures ANCOVA, we assessed changes in
post-task hunger as the dependent variable (controlling for pretask hunger) with emotional eating as a between-subjects variable
and condition (control vs. stress) as a within-subjects factor.
Results
We first report the results of various manipulation checks.
Immediately after participants’ arrival for the study, there were
no differences between the control day and the stress day on hunger or any of the affect items of the PANAS. After the task, there
were, however, differences on several PANAS items between the
control day and the stress day: reports of irritation, shame, feeling
upset and feeling scared were higher on the stress day compared to
the control day (Mean [SD]: irritation: 2.43 [1.29] vs. 1.94 [1.17],
p = .003; Cohen’s d = 0.38; shame: 2.13 [1.21] vs. 1.23 [.48],
p < .001; Cohen’s d = 0.74; feeling upset: 1.66 [1.15] vs. 1.40 [.99],
p = .038, Cohen’s d = 0.22; feeling scared:1.55 [1.06] vs. 1.23 [.73],
p = .006; Cohen’s d = 0.30). Additionally, reports of hunger were
lower on the stress day compared to the control day (4.62 [2.31]
vs. 5.85 [3.25], p = .003; Cohen’s d = 0.53).
The GLM (controlling for pre-task hunger, dietary restraint and
external eating) showed a significant moderator effect for emotional eating F(1,42) = 4.83, p = .033, partial eta squared = .103.This
282
T. van Strien et al. / Appetite 58 (2012) 277–284
moderator effect was significant when the possible confounding
variables dietary restraint and external eating were removed from
the model, F(1,44) = 5.80, p = .020, partial eta squared = .116. As expected, low emotional eaters tended to eat less after the stress task
than after the control task (adjusted means, standard error [SE]:
121.29 [43.25] vs. 180.34 [33.41], t(22) = 1.68, p = .056, Cohen’s
d = .35) whereas high emotional eaters tended to eat more
(225.91 [42.09] vs. 152.20 [32.50], t(23) = 1.70, p = .051, Cohen’s
d = .35).
There also was a significant emotional eating stress interaction effect on post-task hunger (controlling for pre-task hunger),
F(1,44) = 4.27, p = .045, partial eta squared = .088. As expected, low
emotional eaters reported less hunger after the stress task than
after the control task (adjusted means [SE]: 3.96 [.32] vs. 6.00
[.33], t(22) = 3.99, p < .001, Cohen’s d = .99), whereas high emotional eaters reported similar amounts of hunger after the stress
and control tasks (5.24 [3.16] vs. 5.70 [.32], t(23) = .82, p = .21, Cohen’s d = .16).
Discussion
Study 2 demonstrated that the interaction effect of emotional
eating is also found when using a within-subjects design and when
emotional eating is assessed well before the actual study was conducted. This indicates that the moderator effect of emotional eating is robust for experimental design and assessment timing. A
further interesting finding, which will be discussed in the general
discussion, was the emotional eating stress-condition interaction effect on post-task hunger.
General discussion
Earlier studies that assessed the possible moderator effect of
self-reported emotional eating on the relation between stress and
actual food intake have obtained mixed results. In two studies,
we tested whether it is possible to predict distress-induced eating
with a self-report emotional-eating scale by using only participants with extreme scores on the emotional-eating scale. The main
finding of both studies was a significant moderator effect for selfreported emotional eating on the relation between stress and food
intake. As expected, the food intake of the low emotional eaters
was lower during the sad movie or in the stress condition than during the neutral movie or in the control condition, whereas the food
intake of the high emotional eaters was (slightly) higher during the
sad movie or in the stress condition than during the neutral movie
or in the control condition.
This moderator effect of emotional eating was found when we
assessed emotional eating after the food intake while watching
the movie (Study 1) and when we only used females with scores
at the extreme ends of the emotional eating scale for our main
analyses. This moderator effect of emotional eating was also found
in Study 2 when emotional eating was assessed at least 1 month
before the experiment, and only people with extreme scores were
invited to participate. It should further be noted that the moderator effect of emotional eating was also robust for the design of the
study, because it was found when we used a between-subjects design (Study 1) or a within-subjects design (Study 2).
A remarkable finding in Study 2 was that high emotional eaters,
in contrast to the low emotional eaters, did not report a substantial
reduction of hunger immediately after the stress task compared to
the control task. The question that now arises is: Do emotional eaters display an entirely different stress response in that they simply
do not experience the typical post-stress reduction of hunger (because of a hypoactivation rather than a hyperactivation of the HPA
axis)? In this case they would have neurovegetative symptoms that
are the reverse of those associated with the typical stress response:
they would display increased food intake (hyperphagia) and
weight gain, instead of hypophagia and weight loss (Gold &
Chrousos, 2002).6 Alternatively, it is possible that high emotional
eaters did not report the normal post-stress reduction of hunger because they suffer from poor interoceptive awareness, i.e. difficulty in
recognizing whether they are hungry or satiated or suffering from
some other discomfort. The present finding that the increased food
intake of the high emotional eaters in the stress condition was not
paralleled by high post-stress hunger—in fact, high emotional eaters
reported similar degrees of hunger after the control task and the
stress task—suggests that high emotional eaters ate in response to
stress-related emotional agitation rather than in response to internal
hunger. This, in turn, would suggest support for poor interoceptive
awareness. However, more research on this issue is needed.
A further possibility, suggested by results of a functional magnetic resonance imaging study by Bohon et al. (2009), may be that
emotional eaters differ from non-emotional eaters in their activation of neural circuitry in response to food. For high emotional eaters, food may be more rewarding when they are in a state of
negative mood. It as yet unclear, however, whether emotional eating reflects a deficit of reward (Volkow et al., 2003) or a heightened
sensitivity to reward (Davis, Strachan, & Berkson, 2004). In a study
on brain dopamine measures, emotional eating was negatively
associated with baseline dopamine D2-receptor availability in the
dorsal striatum (Volkow et al., 2003). This finding was taken to
support the view that hypo-dopaminergic functioning underlies
the development of emotional eating which would explain why
emotional eaters are more receptive to the reinforcing value of
food (Volkow et al., 2003), and use food as ‘self-medication’ to
blunt effects of negative emotions. Reduced brain dopamine of
emotional eaters has, however, also been explained as possible
outcome of an adaptive down-regulation of the dopaminergic system, a reflection of a neuroadaptation secondary to overstimulation with food as result of a heightened sensitivity to reward
(Bohon et al., 2009; Davis et al., 2004).
In Study 1 there was a moderator effect for emotional eating
only when we examined participants with extreme scores on emotional eating and controlled for external eating as possible confounder. The fact that we found no such moderator effect when
we used the entire sample and classified people into low or high
emotional eaters by using the commonly used median-split procedure or when we used the full range of scores for emotional eating
(even though the sample was in those cases more than twice as
large, n = 124 vs. n = 45) would suggest that emotional eating
may be best described as a categorical rather than a continuous
variable. This notion would be consistent with the results of a functional magnetic resonance imaging study where extreme scores on
self-reported emotional eating predicted important individual differences in reward response during negative moods (Bohon et al.,
2009).
It should be noted that the validity of the emotional eating scale
for predicting emotional eating should not be considered in isolation from the emotional cues to which it is related (Macht,
2008). Earlier it was found that experimental manipulations of
physical fear (e.g., threats of electrical shock) reduced eating of
non-dieting normal-weight individuals but did not markedly increase the eating of obese or dieting individuals (Herman & Polivy,
1975; McKenna, 1972; Schachter, Goldman, & Gordon, 1968).
However, when anxiety was experienced as diffuse or when individuals were not able to label the source of their emotional arousal,
6
In this respect it should be noted that Raspopow et al. (2010) found some
interesting differences between low vs. high (DEBQ) emotional eaters (median split
classification) in their hormonal (cortisol and ghrelin) responses following the Trier
Social Stress Test.
T. van Strien et al. / Appetite 58 (2012) 277–284
obese individuals were found to eat significantly more when distressed than when calm (e.g., Slochower, 1983). A highly similar
pattern of results was found for emotional eaters with an experimental manipulation that involved ego threat (the anticipation of
a public speaking task (Oliver et al., 2000; see also Wallis & Hetherington, 2004)). So it would seem that the moderator effect of emotional eating takes the form of low emotional eaters eating
appreciably less and high emotional eaters eating appreciably
more only in response to certain types of threats. This is also nicely
illustrated by the results of the present two studies. Strictly speaking, distress did not produce a statistically significant increase in
food intake among emotional eaters in either of the two studies;
in both studies the effect of high emotional eating on intake was
only borderline significant (ps = .063 and .051). Results did, however, suggest that high emotional eaters tended to eat more in response to the ego threat as induced by the Trier Social Stress Task
(Study 2), and only slightly more in response to the negative mood
(induced by the sad movie) (Study 1).
Alternatively, demand characteristics and socially desirable
responding may have been responsible for the fact that emotional
eaters are rarely observed to significantly eat more in response to
stress in the laboratory. According to Krantz (1978), people struggling with overeating or weight problems may be concerned with
their self-presentation and therefore less inclined to (over)eat in
the laboratory. 7
Limitations
A possible limitation is that we cannot rule out the possibility
that, for reasons of social desirability, people may have disavowed
emotional eating. Still, scores on the emotional eating scale (at
least extreme scores) showed predictive validity in these studies,
reducing this concern. A further limitation is that the present
experiments were conducted on predominantly normal-weight female students, so the present results need replication in overweight subjects. Other limitations may be the lack of assessment
of palatability ratings and the presentation of a relative small number of food items.8 Finally, the present findings would need replication outside the laboratory.
Strengths
In the present studies we avoided the possibility that individuals
with certain scores might be classified as high in one study but low in
the other by using as cut-off points for low vs. high emotional eating
the 20th and 80th percentiles of the latest norm group of Dutch females (Van Strien, 2011). A further strength is that in all analyses
we controlled for restrained eating and external eating, so the
7
Support for this possibility in the present study was obtained by inspection of the
results of a post-hoc analysis of the intakes of the individual food items in Study 2;
only intake of buttercake yielded a significant interaction effect in the predicted
direction, and no such interaction effect was obtained with intake of M&Ms. In spite
of the chocolate in the M&Ms, emotional eaters may have been reluctant to overeat
M&Ms. in Study 2, possibly because eating M&Ms, in comparison to eating the spongy
buttercake, makes relatively more noise. Participants may have been reluctant to
indulge themselves on M&Ms. before the experimenter whom they knew to be within
hearing distance in the other room (there was no background noise). In Study 2 they
ate altogether only a mean of 12,5 grams of M&Ms, which amounts to 14 pieces of
M&Ms. (In comparison, in Study 1 where great care was taken to create a context in
which people could relax and display their own natural behavior, and where we
measured the subjects’ snack intake while they watched TV (background noise), and
where they only had a choice between noisy food (crisps and M&Ms), the total mean
intake of M&Ms. was 44.45 g. This is nearly thrice as much as the total intake of
M&Ms in Study 2).
8
Following our observation in Footnote 7, for future research it would be also
advisable to replace M&Ms with a type of chocolate that permits eating without
making a noise.
283
moderator effect of emotional eating seems robust for the possible
confounding effects of external and restrained eating.
Conclusion
By using only participants with extreme scores on emotional
eating, it is possible to predict actual distress-induced food intake.
References
Aiken, L. S., & West, S. G. (1991). Multiple regression. Testing and interpreting
interactions. Sage Publications, Inc: Thousand Oaks, CA, USA.
Anschutz, D. J., Van Strien, T., & Engels, R. C. M. E. (2008). Exposure to slim images in
mass media. Television commercials as reminders of restriction in restrained
eaters. Health Psychology, 27, 401–408.
Appelhans, B. M., Pagoto, S. L., Peters, E. N., & Spring, B. J. (2010). HPA axis response
to stress predicts short term snack intake in obese women. Appetite, 54,
217–220.
Bohon, C., Stice, E., & Spoor, S. (2009). Female emotional eaters show abnormalities
in consummatory and anticipatory food reward. A functional magnetic
resonance imaging study. International Journal of Eating Disorders, 42, 210–221.
Bruch, H. (1973). Eating disorders. New York: Basic Books.
Conner, M., Fitter, M., & Fletcher, W. (1999). Stress and snacking. A diary study of
daily hassles and between-meal snacking. Psychology & Health, 14, 51–63.
Davis, C., Strachan, S., & Berkson, M. (2004). Sensitivity to reward. Implications for
overeating and overweight. Appetite, 42(2), 131–138.
Evers, C., de Ridder, D. T. D., & Adriaanse, M. A. (2009). Assessing yourself as an
emotional eater. Mission impossible? Health Psychology, 28, 717–725.
Gold, P. W., & Chrousos, G. P. (2002). Organization of the stress system and its
dysregulation in melancholic and atypical depression. High vs low CRH/NE
states. Molecular Psychiatry, 7(3), 254–275.
Greeno, C. G., & Wing, R. R. (1994). Stress-induced eating. Psychological Bulletin, 115,
444–464.
Heatherton, T. F., Herman, C. P., & Polivy, J. (1991). Effect of physical threat or ego
threat on eating behavior. Journal of Personality and Social Psychology, 60,
138–143.
Herman, C. P., & Polivy, J. (1975). Anxiety, restraint and eating behavior. Journal of
Abnormal Psychology, 84, 666–672.
Holmbeck, G. N. (2002). Post-hoc probing of significant moderational and
meditational effects in studies of pediatric populations. Journal of Pediatric
Psychology, 27, 87–96.
Kirschbaum, C., Pirke, K. M., & Hellhammer, D. H. (1993). The Trier Social Stress Test.
A tool for investigating psychosocial stress responses in a laboratory setting.
Neuropsychobiology, 28, 76–81.
Krantz, D. S. (1978). The social context of obesity research. Another perspective on
its place in the field of social psychology. Personality and Social Psycholology
Bulletin, 4, 177–184.
Larsen, J. K., Van Strien, T., Eisenga, R., & Engels, R. C. M. E. (2006). Gender
differences in the association between alexithymia and emotional eating in
obese individuals. Journal of Psychosomatic Research, 60, 237–243.
Larsen, J. K., Ouwens, M., Engels, R. C. M. E., Eisinga, R., & Van Strien, T. (2008).
Validity of self-reported weight and height and predictors of weight bias in
female college students. Appetite, 50, 386–389.
Macht, M. (2008). How emotions affect eating. A five-way model. Appetite, 50, 1–11.
Maxwell, S. E., & Delaney, H. D. (1993). Bivariate median splits and spurious
statistical significance. Psychological Bulletin, 113, 181–190.
McClelland, G. H., & Judd, C. M. (1993). Statistical difficulties of detecting
interactions and moderator effects. Psychological Bulletin, 114, 376–390.
McKenna, R. J. (1972). Some effects of anxiety level and food cues on the eating
behaviour of obese and normal subjects. A comparison of the Schachterian and
psychosomatic conceptions. Journal of Personality and Social Psychology, 22,
311–319.
Newman, E., O’Connor, D. B., & Commer, M. (2007). Daily hassles and eating
behaviour. The role of cortisol reactivity status. Psychoneuroendocrinology, 32,
125–132.
O’Connor, D. B., & O’Connor, R. C. (2004). Perceived changes in food intake in
response to stress. The role of conscientiousness. Stress and Health, 20, 279–291.
O’Connor, D. B., Jones, F., Conner, M., McMillan, B., & Ferguson, E. (2008). Effects of
daily hassles and eating style on eating behavior. Health Psychology, 27, s20–s31.
Oliver, G., Wardle, J., & Gibson, E. L. (2000). Stress and food choice. A laboratory
study. Psychosomatic Medicine, 62, 853–865.
Ouwens, M. A., Van Strien, T., & Van Leeuwe, J. F. J. (2009). Possible pathways
between depression, emotional and external eating. A structural equation
model. Appetite, 53, 245–249.
Raspopow, K., Abizaid, A., Matheson, K., & Anisman, H. (2010). Psychosocial stressor
effects on cortisol and ghrelin in emotional and non-emotional eaters. Influence
on anger and shame. Hormones and Behaviour, 58, 677–684.
Schachter, S., Goldman, R., & Gordon, A. (1968). Effects of fear, food deprivation and
obesity on eating. Journal of Personality and Social Psychology, 10, 91–97.
Slochower, J. (1983). Excessive eating. New York: Human Sciences Press.
Snoek, H. M., Engels, R. C. M. E., Janssens, J. M. A. M., & Van Strien, T. (2007). Parental
behaviour and adolescents’ emotional eating. Appetite, 49, 223–230.
284
T. van Strien et al. / Appetite 58 (2012) 277–284
Spoor, S. T. P., Bekker, M. J., Van Strien, T., & Van Heck, G. L. (2007). Relations
between negative affect, coping and emotional eating. Appetite, 48, 368–376.
Stone, A. A., & Brownell, K. D. (1994). The stress-eating paradox. Multiple daily
measurements in adult males and females. Psychology & Health, 8, 425–436.
Van Strien, T. (2011). Dutch Eating Behaviour Questionnaire. Manual. Amsterdam:
Boom Test Publishers.
Van Strien, T. (2010). Predicting distress-induced eating with self-reports. Mission
impossible or a piece of cake? Health Psychology, 29, 343.
Van Strien, T., Herman, C.P., & Anschutz D. (in press). The predictive validity of the
DEBQ-external eating scale for eating in response to food commercials while
watching television. International Journal of Eating Disorders. doi:10.1002/
eat.20940.
Van Strien, T., & Van de Laar, F. (2008). Intake of energy is best predicted by overeating tendency and consumption of fat is best predicted by dietary restraint. A
four-year follow up of patients with newly diagnosed type-2 diabetes. Appetite,
50, 544–547.
Van Strien, T., & Ouwens, M. A. (2003). Counterregulation in female obese emotional
eaters. Schachter, Goldman, and Gordon’s (1968) test of psychosomatic theory
revisited. Eating Behaviors, 3, 329–340.
Van Strien, T., & Ouwens, M. (2007). Effects of distress, alexithymia and impulsivity
on eating. Eating Behaviors, 8, 251–257.
Van Strien, T., Snoek, H. M., van der Zwaluw, C. S., & Engels, R. C. M. E. (2010).
Parental control and the Dopamine D2 Receptor Gene (DRD2) interaction on
emotional eating in adolescence. Appetite, 54, 255–261.
Van Strien, T., van der Zwaluw, C. S., & Engels, R. C. M. E. (2010). Emotional eating in
adolescence. A gene (SLC6A4/5-HTT)-depressive feelings interaction analysis.
Journal of Psychiatric Research., 44, 1035–1042.
Volkow, N. D., Wang, G. J., Maynard, L., Jayne, M., Fowler, J. S., Zhu, W., et al. (2003).
Brain dopamine is associated with eating behaviors in humans. International
Journal of Eating Disorders, 33(2), 136–142.
Wallis, D. J., & Hetherington, M. M. (2004). Stress and eating. The effects of egothreat and cognitive demand on food intake in restrained and emotional eaters.
Appetite, 43, 39–46.
Wallis, D. J., & Hetherington, M. M. (2009). Emotions and eating. Self-reported and
experimentally induced changes in food intake under stress. Appetite, 52,
355–362.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and validation of brief
measures of positive and negative affect. The PANAS scales. Journal of
Personality and Social Psychology, 54, 1063–1070.
Whisman, M. A., & McClelland, G. H. (2005). Designing, testing and interpreting
interactions and moderator effects in family research. Journal of Family
Psychology, 19, 111–120.