Oliver, Georgina PhD; Wardle, Jane PhD, and; Gibson, E. Leigh PhD
ANOVA = analysis of variance;, DBP = diastolic blood pressure;, DEBQ = Dutch Eating Behavior Questionnaire;, PANAS = Positive and Negative Affect Schedule;, SBP = systolic blood pressure;, STAI = State-Trait Anxiety Inventory.
There is increasing evidence that stress may affect health not only through its direct biological effects but also through changes in health behaviors that themselves influence health (1, 2). Clearly, one such health behavior is food choice: that is, stress may lead to ill health through unhealthy changes in diet as well as more general effects on appetite (3).
Stress and diet associations are particularly complex. Stress is associated with biological changes that might be expected to reduce food intake, at least in the short-term, such as adrenaline-induced glycogenolysis, slowed gastric emptying, autonomic shunting of blood from gut to musculature, and activation of the hypothalamic-pituitary-adrenal axis (4, 5). Yet the experimental results have been inconsistent. Animal studies have produced evidence of both hyperphagia and hypophagia in response to stress (5–7). Research on everyday food intake in human subjects under low- and high-stress conditions has also produced inconsistent results. Stress in the workplace has been associated with higher energy intake in two studies (8, 9), examination stress has produced mixed results (10, 11), and surgical stress, probably the most extreme stressor examined, has been found to have no consistent effect (12). These varying results may be related to the nature of the stressor; for example, mild stressors could induce hyperphagia, and more severe stressors, hypophagia (7, 13).
Alternatively, there could be significant individual differences in responses to stress, with the study samples varying in the proportions of the different response types. Pollard et al. (11) found that students who were high on anxiety and low on social support were more likely to show a hyperphagic response, and Wardle et al. (9) found that dietary restraint levels moderated the response to work stress. An individual difference model is supported by data from both prospective (14) and retrospective (15, 16) self-report studies, showing either increased, decreased, or no change in eating during stress but with consistent effects within individuals.
The importance of individual differences in the eating response to stress has also been borne out by a number of laboratory studies (13). Such studies typically induce stress through one of a number of standard procedures while assessing food intake, ostensibly as incidental to some other task, such as making taste ratings. A consistent pattern is that participants scoring highly on a measure of dietary restraint eat more under stress, whereas intake is the same or lower in unrestrained eaters (17–22).
This rather complex pattern of results suggests that more attention needs to be directed toward specifying the nature and intensity of the stress response, and the characteristics and motivational state of the participants (eg, hunger, restraint, and emotional eating tendency). Furthermore, in these studies, usually only a single food type is available, typically high in fat and/or sugar, such as ice cream. Thus, food intake has been conflated with food choice. Understanding which foods are selected or avoided under stress is a crucial issue, both because it is necessary for theoretical interpretation of the mechanisms involved and for prediction of harmful effects of stress on health.
In contrast to the laboratory studies, self-reported retrospective (15, 23) and prospective data (8, 9, 11) suggest that food choice does change under stress, with a tendency toward a relative increase in sugary, fatty (often snack-type) foods. Therefore, experimental tests of effects of stress on eating should take into account the sensory properties of foods.
Grunberg and Straub (24) extended the usual laboratory paradigm by providing participants with a range of foods differing in taste qualities (sweet, salty, and bland), although these were still snack foods presented incidentally to the main task of viewing a film (used for stress induction). They found that men in the stressed group ate less than men in the control group. In women there were no significant differences, although stressed women did show a trend toward a modest increase in consumption of sweet and bland foods with no change in intake of salty foods. These gender differences may have reflected differences in dietary restraint, which is known to be higher in women (16, 25, 26), but unfortunately, as the authors noted, they did not include a measure of restraint. The ecological validity of Grunberg and Straub’s (24) findings may also be limited by the fact that the amount of food eaten by their participants was very small (<20 g per person on average), whereas the variation was relatively high, suggesting that not all subjects were choosing to eat.
The present study was designed to extend the work of Grunberg and Straub (24), including foods from the sweet, salty, and bland taste categories and high- and low-fat examples within those sensory groups. Aside from replicating the work of Grunberg and Straub (24), analyzing effects by taste category may be important because stress has been shown to affect taste perception (27). A wider range of foods was provided to represent the kinds of foods that might be eaten in both meals and snacks. Participants were tested when they were moderately deprived of food and given a test meal around midday to increase the likelihood of eating beyond brief tasting.
Dietary restraint and emotional eating tendencies were assessed as possible explanatory variables. Previous studies that incorporated restraint used the Restraint Scale (28, 29) to measure such tendencies. However, that instrument does not clearly distinguish between cognitive restraint, which might be disrupted by stress; the tendency to overeat in the face of facilitating cues; and the tendency to find relief from emotional stress through eating. Thus, the DEBQ was used because it measures separate factors for dietary restraint and emotionally and externally influenced eating tendencies (30).
The literature suggests that women and restrained eaters consume more calories and fat under stress (8, 9) and shift their food choices away from meal-type foods, such as meat and vegetables, toward snack-type foods (15). In contrast, men and unrestrained eaters show either little difference or a reduction in food intake under stress (12, 24). Therefore, we hypothesized that stress would elicit greater preference for, and consumption of, highly palatable, snack-type foods, most especially in women and restrained and emotional eaters. In contrast, unrestrained, low emotional eaters (most likely to be men) were expected to show no change or even a decrease in consumption in response to stress.
Finally, hunger state and appetite for specific foods were assessed for potential mediation of effects of stress on food intake. In particular, hunger was expected to be lower in the stressed group and to be negatively correlated with biological signs of arousal, by means of which the impact of the stressor could be gauged.
Sixty-eight healthy, nonobese, nonsmoking volunteers (27 men and 41 women, students and staff of the University of London) agreed to participate in a study advertised as an investigation of “the effects of hunger on physiology, performance, and mood.” Volunteers were recruited through advertisements placed around the campus and were paid UK£5 ($7) on completion of the study. Participants were allocated to either a stress or control condition, during which they were provided with a buffet lunch in the laboratory. The study design was approved by the Research Committee on Ethics of University College London.
Anticipation of a speech performance was used as a stressor. Participants were told that they would be making a 4-minute speech that would be recorded by video equipment set up prominently in the laboratory and subsequently assessed. Written instructions for the speech task, based on those used by Kapczinski et al. (31), were given. Participants were invited to select 1 of 10 controversial topics for their speech and to prepare notes for a period of 10 minutes before receiving a meal. The speech was scheduled to take place immediately after the meal, and participants were led to believe that the study was about hunger and its effect on performance and that they had been placed in a “low-hunger” condition that necessitated them eating a meal before making their speech. This ensured that the anticipatory stress created by the threat of public speaking was sustained while subjects were exposed to food but that there was no active competing task to eating. Unknown to subjects at this stage was that they were not required actually to perform the speech. No mention of the speech task was made to participants in the control condition. Instead they were given a nonstressful task of comparable duration (10 minutes), which was to listen to a passage of emotionally neutral text (an excerpt from Under Milk Wood, Ref. 32). They were instructed to sit and relax while listening to the text, after which, they were told, they would receive a meal. They were led to believe that the study was concerned with changes in physiological measurements before and after a meal.
Assessment of Impact of Stress Manipulation
Both physiological and psychological indices of arousal were included, because desynchrony between self-reported anxious mood and physiological measures of arousal has been well documented (33). Blood pressure and heart rate were measured using a Copal digital sphygmomanometer UA-251 (baseline was an average of two readings taken over a 3- to 5-minute period). A self-reported measure of mood, the PANAS (34), was completed on arrival in the laboratory (baseline) and after the 10-minute stress induction. At the end of the study, as part of the debriefing, participants were asked to rate the perceived stressfulness of the study on a seven-point Likert scale (where 1 = “not at all stressful” and 7 = “extremely stressful”).
Assessment of Eating Behavior
Participants had been asked to refrain from eating for 4 hours before the study to ensure a reasonably standardized level of deprivation, resulting in a substantial intake from the meal. Ratings of current hunger (on a seven-point Likert scale, where 1 = “not at all” and 7 = “extremely”) were taken on arrival in the laboratory (baseline) and after the stress or control manipulation.
Two measures were used to assess the effect of the stress manipulation on eating behavior and food choice: 1) appetite for a range of foods immediately before eating the meal and 2) food intake during the meal. Foods had been selected on the basis of their nutrient content to represent three taste categories, sweet, salty, and bland, following the method of Grunberg and Straub (24). Within these taste categories, foods were additionally divided into low- and high-fat groups. A total of 34 foods were selected for the food preference tasks (Table 1). Fifteen similarly categorized foods were provided for the buffet meal (Table 2): 9 of these were represented by similar foods among the 34 foods, but in addition, 3 bland high-fat spreads and 3 sweet low-fat foods were chosen for appropriate use as small portions and practicality for meal construction. To test the assumption that taste was related to nutrient content, 34 adults (not those who took part in the main study) were asked to rate the taste of each of the 34 foods for sweetness, saltiness, and fattiness (where 1 = “not at all” and 7 = “extremely”). Perceptions of taste were found to correlate highly with actual nutrient content (sugar content and sweetness:r = 0.84, p < .001; salt content and saltiness:r = 0.76, p < .001; fat content and fatty taste:r = 0.77, p < .001), thus validating the use of nutritional composition data to generate categories based on taste.
Food Appetite Ratings.
Photographs of each of the 34 foods were presented one at a time. For each food, participants were asked, “How much do you fancy eating some of this food at the moment?,” and indicated their response on a scale from 1 (“I definitely don’t want to eat this food at all at the moment”) to 7 (“Right now I really want to eat this food”). The appetite ratings had previously been found to show adequate test-retest reliability when administered to 12 different adults on two occasions 30 minutes apart (mean r = 0.83, p < .01). The ratings were completed after the stress (or control) manipulations, immediately before the meal was served. The photographs were presented in random order to each subject to control for possible order effects between sequentially presented foods.
Participants were allowed to eat freely for 15 minutes from a buffet lunch consisting of, as far as was practicable, foods from each of the taste categories described above (Table 2). The foods were weighed to the nearest 0.1 g before and after the meal to determine the amount consumed.
Individual Difference Variables
A number of trait measures were completed before participation. Trait anxiety was assessed with the Trait scale of the STAI (35). Dietary restraint and the tendency to eat more when cognitive restraint on eating is disrupted by psychological, sensory, or emotional challenges (sometimes labeled “disinhibition”) were assessed with the Restraint, Emotional, and External Eating scales from the DEBQ (30). In addition, for each of the 34 foods listed in the appetite ratings, participants were asked to indicate how much they liked the food “in general.” Responses were recorded on a Likert scale (where −4 = “I really dislike this food” and +4 = “I really like this food”).
Procedures and Scheduling
The study was performed between 11:30 AM and 1:30 PM, that is, at a time when a meal would usually be eaten. This is in contrast to the usual laboratory eating paradigms in which snack consumption is measured without regard to meal times. On arrival at the laboratory, participants confirmed that they had eaten nothing in the previous 4 hours, and baseline measures of blood pressure, heart rate, mood (PANAS), and hunger were completed. Participants then received instructions for either the stress or control task, according to their random allocation, and were left alone for the 10-minute duration of the tasks. At the end of the 10-minute period, blood pressure, heart rate, and mood were reassessed in all subjects. A second hunger rating and the food appetite ratings were also completed. Participants then received a meal with the foods presented on separate plates on two trays, the position of the plates on the trays being varied for each participant. They were instructed that they could eat whatever they wished from the selection and in whatever quantities they desired, just as long as they ate something and were less hungry at the end of the meal. The experimenter explained that they would be left alone to eat for 15 minutes. At the end of the meal, the experimenter returned, and the true nature of the study was explained. The debriefing included a rating of the perceived stressfulness of the study. Finally, age, height, and weight were recorded.
The main hypotheses were tested by ANOVAs; the interactions between stress condition (group) and individual difference variables, such as gender, restraint, and emotional eating, were of primary interest. In most cases, there was an a priori prediction for the direction of the interaction effect, and so an α level of 0.05 was taken as significant, despite quite large numbers of statistical tests. Unexpected results that achieved this level of significance were interpreted cautiously. Multiple comparison tests on the same dependent variable were not required by this design. In addition, some relationships between physiological and psychological variables were assessed by using Pearson’s product-moment correlation.
The background characteristics of the group are summarized in Table 3. Participants were between 18 and 46 years old. The men were predictably heavier (F (1,64) = 117.17, p < .001) and taller (F (1,64) = 46.04, p < .001) than the women, but body mass index did not differ between the sexes. There were no differences between groups randomized to the stress or control condition on any of these measures. There were no significant group or gender differences in trait anxiety.
As expected, dietary restraint scores were significantly higher in women (F (1,64) = 13.50, p < .001), and women scored higher than men on the emotional eating scale (F (1,64) = 4.99, p < .05), but there were no gender differences in external eating. There were no differences between stress and control groups in dietary restraint, emotional eating, or external eating.
Ratings for liking of the foods to be used in the study showed that fatty sweet foods were most liked by the sample as a whole (see Table 3) and that salty low-fat foods were the least liked. Men reported liking fatty bland and fatty salty foods significantly more than women (F (1,64) = 6.65, p < .02 for fatty bland foods;F (1,64) = 15.12, p < .001 for fatty salty foods). There were no significant differences in general food preferences between stress and control groups.
Effectiveness of the Stress Manipulation
At baseline, there were no differences between the stress and control groups in heart rate, SBP, or DBP (see Table 4). Men in both groups had significantly higher SBP (F (1,64) = 25.98, p < .001) and DBP (F (1,64) = 4.28, p < .05) than women. The change in heart rate after the stress manipulation did not reach significance (group-by-time interaction:F (1,64) = 2.82, p < .10;Table 4). SBP increased over time (from baseline to after stress) in the stressed group and decreased in the control group (group-by-time interaction:F (1,64) = 14.41, p < .001;Table 4). The pattern was the same for men and women. DBP decreased in the control but not in the stress group (group-by-time interaction:F (1,64) = 4.42, p < .05).
There were no significant differences between stress and control subjects in positive or negative affect scores at baseline. Negative affect scores were log-transformed to produce a normal distribution. As predicted, subjects in the stress group showed a significant increase in negative affect from baseline, whereas those in the control group showed a reduction in negative affect relative to baseline (group-by-time interaction:F (1,64) = 11.77, p < .001). For positive affect, the control group showed a decrease, whereas in the stressed group positive affect remained constant (group-by-time interaction:F (1,64) = 10.00, p < .01). Gender had no effect on affect scores.
Overall, the manipulation achieved significant if modest differences between groups in both physiological and psychological indices of stress (Table 4). This was borne out by the post hoc subjective ratings of perceived stress. Participants in the stressed group rated their experience as significantly more stressful (mean rating = 4.26, SD = 1.4) than the control group (mean rating = 1.62, SD = 1.0) (F (1,64) = 69.12, p < .01). This between-group difference applied to both men and women.
Food Intake: Effects of Gender and Stress
Men ate significantly more weight of food (F (1,64) = 6.03, p < .02) and had a higher total energy intake (F (1,64) = 11.07, p < .001) than women (Table 5), as would be expected from their significantly higher body weights and consequently greater daily energy requirements. Controlling for energy requirements (covariate F (1,63) = 5.77, p < .02) removed this gender effect (for grams:F (1,63) = 1.16; for kilocalories:F (1,63) < 1; both NS), and so estimated energy requirement was included as a covariate in subsequent analyses involving total amounts of food eaten. However, analyses involving measures of selection of different food sensory categories were not adjusted for energy requirements. Daily energy requirements were estimated on the basis of published figures. 1This resulted in mean (SD) daily energy requirements of 3031.0 (236.9) kcal/d for the men and 2172.6 (185.1) kcal/d for the women (sex difference: F(1,64) = 284.6, p < .001). Both men and women consumed about one-third of their daily caloric requirements from the food presented, suggesting that this eating episode could realistically be considered a meal rather than merely a snack.
There were no significant main effects of stress group on weight of food consumed, total energy intake, or energy density of the meal (kcal/g), nor were there any interactions between group and gender (Table 5). Total intake was also analyzed in terms of the main macronutrients (carbohydrate, fat, and protein) and for starch and sugar separately, but no significant effects were found (data not shown).
To examine intake in relation to choice from the food sensory categories (sweet, salty, and bland), the amount of food eaten from each category, including high- and low-fat levels, was calculated (Table 5). Four-factor repeated-measures ANOVA of intake (g) with food category and fat level as within-subjects factors, and group and gender as between-subjects factors, revealed a number of significant effects (with ε-corrected F values adjusted for sphericity where necessary). There was no significant main effect or any interactions involving stress group (Table 5). However, intake differed by food category (F (2,128) = 67.47, p < .001), fat level (F (1,64) = 116.76, p < .001), and gender (F (1,64) = 6.03, p < .02). Furthermore, food category interacted with gender (F (2,128) = 3.28, p < .05) and fat level (F (2,128) = 194.55, p < .001), and all of these factors interacted (F (2,128) = 6.94, p < .001). Therefore, effects of fat level and gender were analyzed within each food category by two-factor ANOVA. Results for energy intake were essentially similar and so are not presented here.
Men ate significantly more bland food than women (F (1,66) = 11.69, p < .001), although this difference was more apparent for low-fat than high-fat bland foods (F (1,66) = 9.57, p < .01) (Table 5). This may not be surprising given that the high-fat bland foods were spreads, which were consumed in much smaller amounts than the low-fat bland foods (bread, carrots, and tomatoes) (F (1,66) = 381.17, p < .0001).
Men ate significantly more salty foods than women (F (1,66) = 4.48, p < .05). This effect was independent of fat level, although the one low-fat salty food available, yeast extract (Marmite spread), was eaten in far smaller amounts than the high-fat salty foods (F (1,66) = 135.12, p < .0001).
Unlike bland and salty foods, men did not eat any more sweet foods than women (F (1,66) < 1;Table 5). Both men and women ate significantly more weight of low-fat than high-fat sweet foods (F (1,66) = 20.44, p < .001, no interaction).
Effects of Dietary Restraint on Intake
A median split of restraint scores was used to distinguish restrained and unrestrained eaters. Gender was included as a covariate because there were significantly more female than male subjects in the high-restraint group (χ2 = 4.98, p < .05).
There were no significant differences in intake (as either weight or energy) between restrained and unrestrained eaters and no interaction between restraint level and stress condition (F values < 1, except the gender covariate, which had values of F (1,63) = 5.23, p < .05 for grams and F (1,63) = 9.82, p < .01 for kcal; results were not qualitatively different without gender as a covariate; see Table 6). Analyses grouping food intake in terms of energy density of overall intake (Table 6), sensory categories, and percentage of energy from carbohydrate, protein, and fat (data not shown) all failed to reveal any significant effects of restraint or stress or any interactions involving these factors.
Effects of Emotional Eating on Intake
Restrained eaters scored significantly higher on the emotional eating subscale of the DEBQ (t (66) = 3.23, p < .01), and women were also more emotional eaters than were men (t (66) = 2.26, p < .05) (Table 3). Thus, although gender might mediate the interaction between restraint and emotional eating, effects of the latter could give a clearer indication of individual differences in eating responses to stress. Again, gender was included as a covariate in these analyses.
Subjects were divided, on the basis of a median split, into high and low emotional eaters (ie, “emo-tional eaters” and “nonemotional eaters”). No significant effects of stress condition or any interaction with emotional eating status were seen when total intake was analyzed in terms of either weight or energy eaten (group effect and group-by-emotional eating interaction, all F values < 1; see Table 7). Gender was a significant covariate here for both grams of food eaten (F (1,63) = 7.18, p < .01) and energy intake (F (1,63) = 12.33, p < .005).
Unlike total intake, the energy densities of the meals eaten varied by stress condition and emotional eating status (stress-by-emotional eating interaction:F (1,63) = 6.17, p < .02;Table 7). In the stress group, the energy density of high emotional eaters’ intake was significantly greater than that of low emotional eaters (t (32) = 2.22, p < .05), whereas among control subjects the high emotional eaters ate less energy-dense meals on average (t (32) = 1.45, NS). Gender was not a significant covariate for energy density (F (1,63) = 1.56, NS;Table 7), possibly reflecting the fact that variation in energy density results from different choices rather than differences in overall intake.
Effects of stress and emotional eating, and any interactions with fat level, were investigated separately for intake of each food sensory category by using three-factor repeated-measures ANOVA. Intakes of bland and salty foods were unaffected by these factors.
Gender was not a significant covariate for intake of sweet foods, and so subsequent analyses did not include gender. Overall, weight of intake of sweet low-fat foods (fruit and jam) was higher than intake of high-fat foods (cake and chocolate biscuits) (F (1,64) = 18.66, p < .001), although the reverse was true for energy intake (F (1,64) = 35.12, p < .001), reflecting the far greater energy density of the sweet high-fat foods (Table 7). There were no main effects for stress group or emotional eating, but there were three-way interactions between fat level, group, and emotional eating (grams: (F (1,64) = 3.53, p < .07; kcal: (F (1,64) = 5.05, p < .05), which were examined further by separate analyses for low- and high-fat sweet foods.
High emotional eaters were found to eat almost twice the weight of sweet fatty foods on average than did low emotional eaters in the stress group (t (32) = 1.78, p < .05, one-tailed test;Table 7); in contrast, among controls, high and low emotional eaters did not differ significantly in their intake of sweet fatty foods (t (32) = 0.78). Although this interaction just failed to reach significance for grams eaten (stress-by-emotional eating interaction:F (1,64) = 3.60, p = .06), the effect on energy intake was significant (F (1,64) = 4.26, p < .05) for the group-by-emotional eating interaction, again with high emotional eaters eating nearly twice the energy intake of low emotional eaters among stressed subjects (Figure 1). In comparison, no effects of emotional eating or stress were seen for intake of sweet low-fat foods (Table 7).
Analyses of separate macronutrient intake did not reveal any significant effects of stress or emotional eating or any interactions for amount or percentage of energy eaten.
Effects on Hunger
Men and women did not differ in their hunger rating recorded at baseline, nor were there significant differences between participants allocated to the stress and control groups. For the sample as a whole, initial hunger ratings were reasonably high (mean rating = 4.87, SD = 1.23, of a maximum 7), so a substantial intake could be expected during the meal.
Hunger showed no significant change from baseline to after stress, nor was there any differential effect in the two groups; thus, there was no support for the prediction that stress affected hunger at the group level (data not shown). Data from the stress group alone were examined for evidence that greater physiological arousal was linked with lower hunger. A significant (negative) correlation emerged with heart rate in a partial correlation controlling for gender (r = −0.21, p < .05), with a higher heart rate being associated with lower hunger. There were no significant associations for hunger with blood pressure.
The association between rated hunger and desire to eat specified foods was examined using mean appetite ratings across all the 34 foods. This showed that hunger was positively correlated with appetite for the foods (r = 0.42, p < .001).
Desire to Eat Bland, Salty, and Sweet Foods
In the analyses of rated appetites for the three food sensory categories, general preference for foods in that category (based on the liking ratings made at the start of the study) was included as a covariate to assess appetite independently of variation due to differences in general liking. General preference ratings were consistently significant covariates for the rated desires to eat bland, salty, and sweet food (bland:F (1,65) = 12.57, p < .001; salty:F (1,65) = 10.22, p < .01; sweet:F (1,65) = 12.82, p < .001). When analyzing for effects of restraint and emotional eating on desire for foods, gender was not a significant covariate (largest F = 1.63), and so it was excluded from those analyses.
No significant main effects were found, but a group-by-gender interaction was seen for sweet foods (F (1,63) = 4.28, p < .05), and in an unexpected direction, with appetite for sweet foods being increased by stress in men but not in women (stressed vs. control men: mean (SD) = 3.57 (1.09) vs. 2.76 (1.10), F (1,24) = 4.64, p < .05; stressed vs. control women: mean (SD) = 3.32 (1.09) vs. 3.63 (1.10), F < 1; all means were adjusted for the effect of the general liking covariate). In contrast, among unstressed control subjects, women desired to eat sweet foods more than men (F (1,31) = 4.58, p < .05). This effect was independent of fat level (group-by-gender-by-fat level interaction, F < 1). No effects of gender or group, nor any interactions, were seen on appetite for either bland or salty foods (data not shown).
No significant effects of restraint, or any interactions with stress, on desire to eat foods were found (data not shown). This was the case for all food sensory categories.
The only notable effect was that, for salty foods only, stressed high emotional eaters expressed greater appetite than low emotional eaters, whereas control subjects did not differ (stress-by-emotional eating interaction:F (1,63) = 4.22, p < .05; mean (SD) desire to eat for high vs. low emotional eaters: stressed group, 4.58 (0.90) vs. 3.62 (0.91), respectively; control group, 3.73 (0.93) vs. 3.71 (0.90), respectively).
There was no evidence here of a general hypophagic effect of stress on men nor any stress-induced hyperphagia specifically in women, contrary to the findings of Grunberg and Straub (24). Previous studies of stress and eating typically gave snack-type foods to nonfood-deprived subjects, with eating being presented as an incidental activity while performing a more central task. The present study is unusual in explicitly providing mildly food-deprived participants with a meal to overcome the problem of low intake that encumbered earlier studies and in incorporating a wider range of foods to allow more valid assessment of food choice.
However, stress did increase intake of sweet fatty foods in emotional eaters. In addition, women scored more highly on emotional eating than men, as is typically found (30). Thus, Grunberg and Straub’s (24) sex difference in the appetitive response to stress may in fact have been mediated by a difference in emotional eating.
The cake and chocolate biscuits preferred here by stressed emotional eaters are typically eaten as snacks. There is evidence that snack consumption may be more susceptible to stress than meals (15, 37). Such foods may be preferred during stress through learning that small energy-dense snacks are more easily ingested and digested when gut activity is suppressed by sympathetic arousal. In comparison, a naturalistic study of the impact of surgical stress on food intake found no effect of stress on meal intake or composition (12).
Unlike a number of previous studies (18–20, 38, 39), the present study did not reveal a significant influence of dietary restraint on eating behavior under stress, although there was a trend toward greater consumption of sweet foods by highly restrained stressed subjects. In this study, restraint was measured using the restraint scale of the DEBQ, which contains items pertaining only to dietary restraint and thus is not a measure of vulnerability to dietary disinhibition, whether through emotional relief or other reasons. In contrast, the Restraint Scale (28, 29), which has been found to discriminate eating responses to stress (20, 38), paradoxically contains only 1 item (of 10) that explicitly refers to dietary restraint; the remaining items addressing weight fluctuation, preoccupation with food, tendency to binge when eating alone, and feeling guilty after overeating. The psychometric relevance of labeling such a questionnaire as measuring “restraint” has been discussed in detail elsewhere (30, 40, 41), but such a multifaceted instrument does not allow clear interpretation of the psychological mechanisms by which stress could be influencing food consumption and choice. In any event, it cannot be concluded from that measure that a critical factor in responding to stress is the tendency to restrain intake, and we find no support for such a conclusion. Furthermore, the “Disinhibition” scale of the Three-Factor Eating Questionnaire (41) (overeating elicited by social, sensory, and emotional cues) discriminates between women who report eating more during stress and those who do not (16), whereas the restraint-specific scale of that instrument did not predict stress-induced eating.
In comparison, the emotional eating scale of the DEBQ used here specifically defines individuals who have a tendency to eat more during negative emotional states (ie, emotion-induced disinhibition) (42). Just such an emotional eating effect was observed here: High emotional eaters ate more sweet, fatty, and thus energy-dense foods under stress. Thus, this study provides some evidence that a stress-induced change in food choice is a measurable behavioral phenomenon, at least in this laboratory environment, not just a subjective phenomenon confined to self-report measures. These results are particularly important because the increased eating is confined to certain foods, especially those that current health recommendations suggest should be limited. In contrast to intake, rated desire to eat the various categories of foods was less affected by stress, although desire to eat salty foods was specifically greatest in stressed emotional eaters. Even so, the increased desire to eat sweet foods when stressed seen here in men but not in women (irrespective of emotional eating status) is contrary to the (intake) results of Grunberg and Straub (24). These effects on desire for food sensory categories require replication and should be interpreted cautiously.
Why might some individuals be more susceptible to unhealthy shifts in food choice when under stress? The effect is not dependent on gender per se, but the emotional eaters were more likely to be female. From an intervention perspective, it will be important to understand how emotional eating tendencies develop. This characteristic has been discussed extensively elsewhere (13, 30), but its origin remains poorly understood. One consideration is that an initial experience of eating highly palatable energy-dense foods when upset may become habitual by reinforcement through sensory, nutritional, and neurohormonal routes (43, 44).
A recent study provides some support for this model: Markus et al. (45) found that neurotic (“stress-prone”) subjects were protected from depressed mood and raised cortisol induced by a psychological stressor task after eating a carbohydrate-rich/protein-poor breakfast and lunch but not after a carbohydrate-poor/protein-rich diet. In stable subjects, mood was depressed and cortisol increased equally after either diet. This result was interpreted as improved coping after a diet-induced increase in the supply of precursor amino acids to serotonin synthesis. That is, the carbohydrate-rich/protein-poor diet specifically allows greater uptake of the precursor amino acid tryptophan into the brain. The implication is that neurotic or stress-prone individuals may be particularly sensitive to dietary effects on brain pathways influencing mood and stress coping. Furthermore, to learn to “self-medicate” through eating in this manner would most likely require ingestion of unusually low-protein foods in isolation (46), as might be achieved by snacking on sweet and fatty foods when hungry. Also, within this theoretical framework, dietary restraint, found to be positively correlated with emotional eating or disinhibition here and elsewhere (16, 30, 41), may predispose an individual to learning such a dietary-induced relief of dysphoria; that is, dieting has been shown to lower plasma tryptophan levels in women and to sensitize serotoninergic function (47).
An alternative neurohormonal mechanism for stress-induced preferential selection of sweet fatty foods is suggested by evidence that such highly palatable foods can themselves relieve stress through release of endogenous opioids (44, 48).
The effects of stress on hedonic reactions to, and perception of, taste also need consideration. For instance, Dess and Edelheit (27) found that stress changed people’s perception of saccharin’s bitterness and sweetness, as it does in rats (6), but the direction of change depended on aspects of temperament such as trait arousability, pleasure (net affective valence), and dominance. Despite no suggestion of gender differences (27), it could be fruitful to determine the relationship of these traits to emotional eating tendencies.
As with any laboratory study carried out in this area, the impact of the stressor on the participants is likely to be less severe than is the case for real events occurring in a nonexperimental setting without ethical constraints. Caution is required in generalizing from these results to less controlled situations in daily life. Even so, anticipation of public speaking is known to be a fearful stimulus to students (49); actual performance might have led to subjects eating while in a relieved rather than stressed state. Here, certainly, subjects in the stress group rated their experience as significantly more stressful than did subjects in the control group, but this was in response to the question asked during the debriefing session, and so answers may have been influenced by demand effects. Nevertheless, on the premise that stress-induced changes in food choice might actually help to alleviate stress, group differences in the perceived stress level reported after the meal may have been less than would be the case without a meal. It should be noted that the (presumably adaptational) decrease from baseline in physiological arousal among control subjects is a well-recognized phenomenon in psychophysiological research (50) : Far from vitiating the use of “stress” and “control” group labels, it illustrates the justification for such a control group. The combination of this difference between groups in changes in physiological measures and the evidence of greater physiological and psychological arousal in the stressed group justifies the “stressed” vs. “control” group comparisons.
Despite the limitations described, this study is unique in its assessment of the effect of stress on food choice in the laboratory by presenting, in the form of an explicit meal, a range of foods varying in nutritional composition, taste and textural qualities, and dietary roles (ie, “snack” foods and “meal” foods). Susceptible individuals were found to select less healthy foods under stress, supporting the proposition that stress may damage health in part through unhealthy food choice. However, the variety of foods was limited, and so caution is needed in interpreting which properties of the foods are critical to the effect. The findings deserve replication and extension, for instance, under different stress conditions and eating contexts, together with further characterization of vulnerable traits (51).
1Estimated daily energy requirements were calculated, using published equations, from basal metabolic rate, which is dependent on age, sex, and weight, multiplied by physical activity level (PAL) (36). Because no data were available for activity levels, moderate levels were assumed for both occupational and nonoccupational activity in both men (PAL = 1.7) and women (PAL = 1.6). Cited Here...
1. Adler N, Matthews K. Health psychology: why do some people get sick and some stay well? Annu Rev Psychol 1994; 45: 229–59.
2. Steptoe A. The links between stress and illness. J Psychosom Res 1991; 35: 633–44.
3. Steptoe A, Lipsey Z, Wardle J. Stress, hassles and variations in alcohol consumption, food choice and physical exercise: a diary study. Br J Health Psychol 1998; 3: 51–63.
4. Johnson EO, Kamilaris TC, Chrousos GP, Gold PW. Mechanisms of stress: a dynamic overview of hormonal and behavioral homeostasis. Neurosci Biobehav Rev 1992; 16: 115–30.
5. York DA. Central regulation of appetite and autonomic activity by CRH, glucocorticoids and stress. Prog Neuroendocrinimmunol 1992; 5: 153–65.
6. Dess NK, Choe S, Minor TR. The interaction of diet and stress in rats: high-energy food and sucrose treatment. J Exp Psychol Anim Behav Process 1998; 24: 60–71.
7. Robbins TW, Fray PJ. Stress-induced eating: fact, fiction or misunderstanding? Appetite 1980; 1: 103–33.
8. McCann BS, Warnick GR, Knopp RH. Changes in plasma lipids and dietary intake accompanying shifts in perceived workload and stress. Psychosom Med 1990; 52: 97–108.
9. Wardle J, Steptoe A, Oliver G, Lipsey Z. Stress, dietary restraint and food intake. J Psychosom Res 2000; 48: 195–202.
10. Michaud C, Kahn JP, Musse N, Burlet C, Nicolas JP, Mejean L. Relationships between a critical life event and eating behaviour in high-school students. Stress Med 1990; 6: 57–64.
11. Pollard TM, Steptoe A, Canaan L, Davies GJ, Wardle J. The effects of academic examination stress on eating behaviour and blood lipid levels. Int J Behav Med 1995; 2: 299–320.
12. Bellisle F, Louis-Sylvestre J, Linet N, Rocaboy B, Dalle B, Cheneau F, L‘Hinoret D, Guyot L. Anxiety and food intake in men. Psychosom Med 1990; 52: 452–7.
13. Greeno CG, Wing RR. Stress-induced eating. Psychol Bull 1994; 115: 444–64.
14. Stone A, Brownell KD. The stress-eating paradox: multiple daily measurements in adult males and females. Psychol Health 1994; 9: 425–36.
15. Oliver G, Wardle J. Perceived effects of stress on food choice. Physiol Behav 1999; 66: 511–5.
16. Weinstein SE, Shide DJ, Rolls BJ. Changes in food intake in response to stress in men and women: psychological factors. Appetite 1997; 28: 7–18.
17. Rutledge T, Linden W. To eat or not to eat: affective and physiological mechanisms in the stress-eating relationship. J Behav Med 1998; 21: 221–40.
18. Cools J, Schotte DE, McNally RJ. Emotional arousal and overeating in restrained eaters. J Abnorm Psychol 1992; 101: 348–51.
19. Frost RO, Goolkasian GA, Ely RJ, Blanchard FA. Depression, restraint, and eating behavior. Behav Res Ther 1982; 20: 113–21.
20. Heatherton TF, Herman CP, Polivy J. Effects of physical threat and ego threat on eating behaviour. J Pers Social Psychol 1991; 60: 138–43.
21. Herman CP, Polivy J, Lank CN, Heatherton TF. Anxiety, hunger and eating behaviour. J Abnorm Psychol 1987; 96: 264–9.
22. Herman CP, Polivy J. Anxiety, restraint and eating behavior. J Abnorm Psychol 1975; 84: 666–72.
23. Lyman B. The nutritional values and food group characteristics of foods preferred during various emotions. J Psychol 1982; 112: 121–7.
24. Grunberg NE, Straub RO. The role of gender and taste class in the effects of stress on eating. Health Psychol 1992; 11: 97–100.
25. Johnson WG, Corrigan SA, Schlundt DG, Dubbert PM. Dietary restraint and eating behaviour in the natural environment. Addict Behav 1990; 15: 285–90.
26. Wardle J, Marsland L, Sheikh Y, Quinn M, Fedoroff I, Ogden J. Eating style and eating behaviour in adolescents. Appetite 1992; 18: 167–83.
27. Dess NK, Edelheit D. The bitter with the sweet: the taste/stress/temperament nexus. Biol Psychol 1998; 48: 103–19.
28. Herman CP, Mack D. Restrained and unrestrained eating. J Pers 1975; 43: 647–60.
29. Ruderman AJ. Dietary restraint: a theoretical and empirical review. Psychol Bull 1986; 99: 247–62.
30. Van Strien T, Frijters JER, Bergers GPA, Defares PB. Dutch Eating Behaviour Questionnaire for assessment of restrained, emotional and external eating behaviour. Int J Eating Disord 1986; 5: 295–315.
31. Kapczinski F, Curran HV, Gray J, Lader M. Flumazenil has an anxiolytic effect in simulated stress. Psychopharmacol 1994; 114: 187–9.
32. Thomas D. Under milk wood: a play for voices. London: Dent; 1954.
33. Rachman S, Hodgson R. Synchrony and desynchrony in fear and avoidance. Behav Res Ther 1974; 12: 311–8.
34. Watson D, Clark LA, Tellegen A. Development and validation of brief measures of positive and negative affect: the PANAS scales. J Pers Social Psychol 1988; 54: 1063–70.
35. Spielberger CD, Gorsuch RL, Lushene RE, Vagg PR, Jacobs JA. Manual for the State-Trait Anxiety Inventory. Palo Alto (CA): Consulting Psychologist Press; 1983.
36. Department of Health. Dietary reference values for food energy and nutrients for the United Kingdom. London: Her Majesty’s Stationery Office; 1991.
37. Lowe MR, Fisher EB. Emotional reactivity, emotional eating and obesity: a naturalistic study. J Behav Med 1983; 6: 135–48.
38. Polivy J, Herman CP, McFarlane. Effects of anxiety on eating: does palatability moderate distress-induced overeating in dieters? J Abnorm Psychol 1994; 103: 505–10.
39. Schotte DE, Cools J, McNally RJ. Film-induced negative affect triggers overeating in restrained eaters. J Abnorm Psychol 1990; 99: 317–20.
40. Heatherton TF, Herman CP, Polivy J, King GA, McGree ST. The (mismeasurement) of restraint. an analysis of conceptual and psychometric issues. J Abnorm Psychol 1988; 97: 19–28.
41. Stunkard AJ, Messick S. The three-factor eating questionnaire to measure dietary restraint, disinhibition and hunger. J Psychosom Res 1985; 29: 71–83.
42. Wardle J. Eating style: a validation study of the dutch eating behaviour questionnaire in normal subjects and women with eating disorders. J Psychosom Res 1987; 31: 161–9.
43. Gibson EL, Desmond E. Chocolate craving and hunger state: implications for the acquisition and expression of appetite and food choice. Appetite 1999; 32: 219–40.
44. Mercer ME, Holder MD. Food cravings, endogenous opioid peptides, and food intake: a review. Appetite 1997; 29: 325–52.
45. Markus CR, Panhuysen G, Tuiten A, Koppeschaar H, Fekkes D, Peters ML. Does carbohydrate-rich, protein-poor food prevent a deterioration of mood and cognitive performance of stress-prone subjects when subjected to a stressful task? Appetite 1998; 31: 49–65.
46. Fernstrom MH, Fernstrom JD. Brain tryptophan concentrations and serotonin synthesis remain responsive to food consumption after the ingestion of sequential meals. Am J Clin Nutr 1995; 61: 312–9.
47. Cowen PJ, Clifford EM, Walsh AE, Williams C, Fairburn CG. Moderate dieting causes 5-HT2C receptor supersensitivity. Psychol Med 1996; 26: 1155–9.
48. Blass EM, Shide DJ, Weller A. Stress-reducing effects of ingesting milk, sugars, and fats. a developmental perspective. Ann NY Acad Sci 1989; 575: 292–305.
49. Geer JH. The development of a scale to measure fear. Behav Res Ther 1965; 3: 45–53.
50. Russell DW. The analysis of psychophysiological data: multivariate approaches. In: Cacioppo JT, Tassinary LG, editors. Principles of psychophysiology: physical, social, and inferential elements. Cambridge: Cambridge University Press; 1990. p. 775–801.
51. Holland B, Welch AA, Unwin ID, Buss DH, Paul AA, Southgate DAT. McCance and Widdowson’s the composition of foods. 5th ed. Cambridge: Royal Society of Chemistry and Ministry of Agriculture, Fisheries and Food; 1994.