- Email: [email protected]
Flavor–nutrient learning in restrained and unrestrained eaters
Physiology & Behavior 90 (2007) 133 – 141
Flavor–nutrient learning in restrained and unrestrained eaters Jeffrey M. Brunstrom a,⁎, Gemma L. Mitchell b a
Department of Experimental Psychology, University of Bristol, 12a Priory Road, Bristol, BS8 1TU, England, UK b Department of Human Sciences, Loughborough University, England, UK Received 26 May 2006; received in revised form 14 September 2006; accepted 14 September 2006
Abstract After we consume a novel food an association can form between its sensory characteristics (e.g., taste properties) and the effect it has on the body (rewarding). Associations of this kind underpin much of our everyday dietary behavior because they mediate both the affective quality of food (flavor-preference learning) and the amount that we choose to consume (learning satiation). Notwithstanding this fact, very few studies have successfully demonstrated the process of dietary learning in human adults. In addition, based on evidence from related research, we explored whether learning is less likely to occur in individuals who have high scores on a measure of dietary restraint. Female participants (N = 44) consumed two differently flavored desserts. Each was presented three times on separate days. One was formulated with a high-energy content (1882 kJ) and the other with a low-energy content (226 kJ). After training, we found little evidence for learned satiation. However, we did observe flavor-preference learning. Specifically, participants acquired a greater liking and desire-to-eat the dessert flavor that was paired with a higher energy density during training. Further analysis revealed that this effect on liking is qualified by dietary restraint. As predicted, unrestrained eaters demonstrated greater differential responding to the two desserts than did restrained eaters. These data provide further evidence for flavor–nutrient learning in adults and they highlight a hitherto unexplored and potentially important difference between restrained and unrestrained eaters. © 2006 Elsevier Inc. All rights reserved. Keywords: Dietary restraint; Appetite; Flavor preference; Dietary learning; Human; Conditioned satiety; Satiation; Associative learning
1. Introduction A critical component of ‘normal’ dietary control is an ability to learn about the consequences of consuming a particular food [1]. As reviewed recently [2], in both animals and humans, ‘flavor–nutrient’ associations can be formed based on the temporal contiguity between the sensory characteristics of a novel tasting food (the ‘conditioned stimulus’ [CS]) and the post-ingestive consequences of its consumption (the ‘unconditioned stimulus’ [US]). When a US is linked to a high-energy density this kind of ‘flavor–nutrient’ association can bring about an increased affection for the CS [3]. In animals, learning tends to be evidenced either by preferential intake in a twobottle test, a “conditioned-taste preference,” or as an increase in intake in a one-bottle test, a “conditioned-flavor acceptance.” By contrast, human learning has tended to be assessed using a variety of different rating and ranking procedures, often in
⁎ Corresponding author. Tel.: +44 117 928 88574; fax: +44 117 928 88588. E-mail address: [email protected] (J.M. Brunstrom). 0031-9384/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.physbeh.2006.09.016
response to questions relating to pleasantness and desire to consume the CS (e.g., [4–6]). Unfortunately, it remains unclear how these measures map conceptually onto a conditioned-taste preference or a conditioned flavor acceptance, and a resolution of this issue is beyond the scope of this study. Here, for simplicity, and to establish a useful collective term, we will refer to all learned changes in affective measures as an example of a “conditioned flavor preference.” However, in so doing, the reader should note that we are not referring to a specific process that might otherwise be observed in animal behavior. Flavor–nutrient associations can also modify meal size. During a meal, there may be insufficient post-ingestive feedback to terminate eating behavior [7]. Instead, meal size may become governed by an anticipation of the likely effects of eating food. This response appears to be based on an association between the flavor of a food and visceral cues (US) that are present towards the end of a meal [5,8], and it is referred to as ‘learned satiation’ [8]. In addition to flavor–nutrient learning, animals and humans also acquire preferences via a more immediate process. ‘Flavor–flavor learning’ occurs when a neutral flavor (CS) is
134
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
presented simultaneously and repeatedly (in a single session) with a substance (e.g., sucrose) that is already liked (US). After this exposure, the valance of the CS shifts in the direction of the US [9,10]. Together, these various forms of dietary learning are important, because they account for many of our everyday decisions about food choice and portion size. Yet despite various speculation [2,11], it is surprising that very little is known about whether individual differences in dietary learning are predicted by everyday dietary behaviors and attitudes. One of the most researched individual difference is the tendency to engage in ‘dietary restraint.’ This term refers to a conscious attempt to limit food intake by overriding the processes that are responsible for dietary control. In some societies, dietary restraint is very common, especially amongst women [12]. In some cases, this behavior is appropriate, because it can form part of a strategy to prevent obesity. However, it is also the case that many normal-weight individuals restrict their intake for other reasons, often in response to concerns about body image [13]. In relation to this group, it should be noted that chronic dietary restriction is related to a range of undesirable outcomes, including low bone-mineral content [14], high cortisol excretion [15], cognitive impairments [16], eating disorders [17], and anxiety [18]. Given these problems, one solution might be to find ways to encourage ‘normal’ or ‘natural’ eating behavior [19]. However, this approach presupposes that restrained eaters have access to the controls of dietary behavior that are otherwise available to their unrestrained counterparts. This study explores aspects of this proposition. In a recent study, flavor–flavor learning was compared in groups of high- and low-restrained eaters [20]. Novel flavors were paired several times with the presentation of a sweet reward. After this period, unrestrained eaters reported a relative increase in liking for the flavors that had been paired most often with a sweet reward. By contrast, restrained eaters reported a greater preference for the flavors that were paired less often. Although the reason for this difference remains to be clarified, the notion that flavor–flavor learning is disrupted is consistent with an earlier study indicating that restrained eaters fail to acquire this type of association [21]. One possibility is that this difference generalizes to the formation of associations based on flavor– nutrient pairings (i.e., flavor–nutrient learning). This is important because evidence of this kind would support the proposition that dieting undermines a range of processes that are otherwise important to the maintenance of normal dietary control. Accordingly, an aim of this study was to determine whether a measure of dietary restraint predicts evidence for either learned satiation or flavor-preference learning, after participants have been exposed to a set of flavor–nutrient pairings. Previously, flavor-preference learning has been observed in preschool children [4,22,23]. However, evidence in adults has been extremely limited [5,6,24]. Similarly, following initial evidence for learned satiation in adults [5,8], successful learning has been observed only in children [25,26] and failures have been reported in studies using adults [27–29] (but see Ref. [11]). Previously, it has been suggested that learning might be limited to childhood because a critical period exists during which flavor–nutrient associations can only be acquired [2].
Therefore, in addition to exploring issues relating to dietary restraint, this research was motivated by a more general concern to explore the extent to which successful dietary learning can be demonstrated in adult humans. 2. Method 2.1. Experiment overview Female participants were exposed to a distinctly flavored high- and low-energy dessert on alternating days, with three exposures to each flavor CS paired with energy US. After this training (days 1–6), the participants received each flavored dessert once more, again on separate days (day 7 and day 8). This time, they were formulated with an intermediate and identical energy density. Learned satiation was assessed by comparing intake of pasta salad after consuming each flavor. Flavor-preference learning was assessed by comparing changes in the affective and sensory quality of the desserts throughout the experiment. On the final day (day 9), we explored the extent to which the participants had acquired an explicit appreciation of the relative consequences of consuming each dessert. 2.2. Participants Female participants were recruited via email from the student population at Loughborough University (UK). After inspecting the data, we were concerned to eliminate participants who started the experiment with a very different preference for the two desserts. Evidence of this kind indicates a degree of familiarity that could limit the extent to which learning is observed [30]. Therefore, post hoc, we applied a crude cut-off point and removed those participants who rated their liking for one of the desserts more than 50 points (on a 100-point scale) higher than their rating for the other dessert at the beginning of the experiment (details of the ratings that were used are given below). On this basis, four participants were removed from the data set leaving a final sample of 44 participants (mean age = 20.5 years [S.D. = 2.00]). Of the remaining participants, two were currently dieting to lose weight, one smoked tobacco, and three were taking medication for asthma. At the outset, all participants offered signed consent and were informed of their right to withdraw from the experiment. The Loughborough University Ethics Committee gave approval for the general protocol. All participants were paid £25 for their assistance. 2.3. Materials The two desserts were designed to be very different and highly novel. Both were comprised of flavored yogurt mixed with small chunks of Jell-O® (gelatin-based dessert). The fruitflavored dessert was served in a tall dessert glass and the chocolate dessert was served in a relatively squat glass bowl. Both glasses had a similar capacity. The fruit-flavored dessert was served with dried fruit flakes sprinkled on the top and the chocolate dessert was topped with whipped cream. Both
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
toppings weighed approximately 2 g (∼ 10 kJ). The fruitflavored dessert was formulated using a combination of mandarin-flavored Jell-O® and vanilla-flavored yogurt, and the chocolate-flavored dessert was formulated using a combination of raspberry-flavored Jell-O® and chocolate-flavored yogurt. The Jell-O® in the low-calorie version was formed by mixing 4.44 g (57 kJ) of ‘Sugar Free Jelly Flakes’ (Rowntrees) with 185 ml of water (as per manufacturer's instructions). The Jell-O® in the high-calorie version was produced by mixing 33.7 g (419 kJ) of ‘regular’ Jell-O® (Tesco stores own brand) with 94 ml of water (30 ml less than recommended by manufacturer, but an otherwise identical process). We added 60 g (1005 kJ) of maltodextrin (Cerestar C⁎DRY 01910, supplied by Albion Chemical Distribution, Coalville, Leicestershire) to this mixture before the Jell-O® had set. The desserts in the lowcalorie condition contained 70 g of low-calorie yogurt (Müllerlight, 159 kJ). In the high-calorie condition, the desserts contained 50 g of the same yogurt (114 kJ), to which 20 g of maltodextrin was added (448.5 kJ in total). Both flavors (fruit and chocolate) of the high-energy and the low-energy dessert had identical energy content, 1882 kJ and 226 kJ, respectively. The high-energy desserts weighed 273 g and their composition was 1% fat, 96% carbohydrate, and 3% protein. The low-energy desserts weighed 260 g; their composition was 6% fat, 49% carbohydrate, and 45% protein. Informal pilot testing indicated that high- and low-energy versions of each variety tasted and looked very similar. Based on taste tests, we found little evidence that participants could reliably identify one particular version that they thought would be more filling. This issue was also explored in some detail at the end of the experiment—see Section 3.4. The pasta salad was served in a large bowl containing more food than could be consumed at a single sitting. A 100 g serving (approx. 774 kJ) contained 75 g of cooked pasta, 8 g of mayonnaise, 5 g of cucumber, 5 g of green pepper, and 7 g of cherry tomatoes. 2.4. Evaluating the desserts Visual-analogue rating scales (100 mm) were used to explore how attitudes towards the desserts changed during the course of the experiment. Three ratings scales were issued before the participants consumed a dessert. The first of these assessed the participants' liking for its taste. After sampling the dessert, the participants were then encouraged to think about their most and least favorite dessert. They then completed a rating that was headed “How much do you like the taste of that dessert?” with anchor points “The same as my least favorite dessert” and “The same as my most favorite dessert,” on the left- and right-hand side, respectively. Participants then completed two further ratings, one headed “How sweet is that food?” and the other headed “How strong is your desire to eat this dessert right now?” with anchor points “Not at all sweet/strong” and “Extremely sweet/strong.” After consuming a full portion of a dessert, the participants completed a final rating headed “How much did you enjoy eating that food?” anchored “not at all” and “very much.”
135
These four measures are referred to here as (a) ‘liking,’ (b) ‘sweet,’ (c) ‘desire,’ and (d) ‘enjoyment.’ A measure of sweetness was included to explore whether the sensory characteristics of the desserts changed or remained stable during the experiment. 2.5. Procedure 2.5.1. Pre-exposure session At a preliminary screening session the participants were told to sample the pasta salad. All confirmed that they would be willing to eat this food for lunch. Participants also completed the restraint section of the Dutch-Eating Behavior Questionnaire (DEBQ) [31]. This measure of restraint was selected because it has been used previously to identify different patterns of flavor–flavor learning in restrained and unrestrained eaters [21]. Finally, a measure of height and weight was recorded and the participants were given detailed instructions on how to complete a visual-analogue rating scale. 2.5.2. Training (days 1–6) Six training sessions were conducted in the Ingestive Behavior Laboratory at Loughborough University, between 12:00 p.m. and 2:00 p.m., on non-consecutive days. All of the sessions were spaced no more than 3 days apart. Across the six sessions, the high- and the low-energy desserts were consumed on alternate days. The allocation of a particular flavor (chocolate or fruit) to the high- and the low-energy condition, together with the ordering of the high- and low-energy desserts, was counterbalanced across participants. (Note that with four participants removed from the data set perfect counterbalancing was lost. However, the most important consideration is the balanced allocation of each flavor to the high- and low-energy condition across participants. This aspect of the counterbalancing was preserved in the final data set.) All participants were instructed to consume their normal breakfast during the morning of each test day. They were also told to refrain from eating after 9:00 a.m. in the morning. To increase compliance with this request, all participants were told to complete a short food diary throughout each test day. On arrival, the participants rated their initial hunger and fullness (100 mm visual-analogue scales). They then sampled and consumed the appropriate dessert for that session. As described above, we also required the participants to produce four ratings. The participants then left the laboratory and they were told that they should refrain from eating or drinking for 90 min. Entries in the food diaries indicated compliance with this request. 2.5.3. Test of learned satiation (days 7 and 8) Each dessert was given to the participants once more, again on separate days. The ordering of the desserts followed the alternating pattern established during training. However, following earlier work [8], both desserts were served with an identical and intermediate energy density (1054 kJ). Ostensibly, the procedure did not change. However, after making the final evaluative rating the participants were instructed to provide a
136
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
second hunger and fullness rating. Again, following earlier studies [5,27], learned satiation was assessed by comparing intake of a second food after consuming a fixed portion of the food that was presented during training. Here, we chose to use pasta salad because this food is generally liked and it is relatively uniform. It has been also been shown to be a sensitive test food in basic preloading studies [32,33]. After this period, the participants provided a final rating of their hunger and fullness. 2.5.4. Assessment of demand and contingency awareness (day 9) On the day immediately after the final test session, the participants were asked to complete an ‘awareness questionnaire.’ Initially, to assess demand characteristics, the participants were asked to guess the purpose of the experiment. Then, after confirming that they remembered the two desserts, they were given two questions that assessed contingency awareness. The first of these was worded, “The desserts differed in their taste, but did they differ in other ways?” (options: ‘yes they differed’ and ‘no they seemed the same.’). If they responded ‘yes,’ then they were asked to write down the way in which they desserts differed. On a separate sheet of paper, the participants were then told, “One of the desserts contained an ingredient that makes you fell less hungry and more full” and they were asked, “Can you guess which one?” (options: ‘chocolate’ and ‘fruit’). Finally, the exact purpose of the experiment was explained, the participants were thanked, and they received payment for their participation.
2.6. Data analysis To explore evidence for preference learning, we derived a set of change scores based on the difference between ratings made at the beginning and the end of the experiment (exposure 4–exposure 1). Thus, for each participant and each type of rating, we calculated two change scores, one relating to the highenergy condition and the other relating to the low-energy condition. The common approach of taking median splits of continuous predictors such as restraint scores prior to ANOVA was avoided as it reduces power and may produce spurious interaction effects [34]. Instead, as in previous studies [35], we entered restraint scores as a continuous variable into a multipleregression analysis. Our aim was to explore whether our change scores are predicted by condition (high/low energy), restraint, and the interaction between restraint and condition. In all regression analyses, ratings made during exposure 1 were entered as a controlling variable. The possibility of co-linearity was addressed by centering the restraint scores before calculating the restraint ⁎ condition interaction term. To explore evidence for learned satiation, we compared the amount of pasta consumed during the two test sessions using a paired t-test. When relatively greater intake is observed after consuming the (previously) low-energy-paired dessert then this can be taken as evidence that learning has occurred. In all of our statistical analyses, we used the conventional critical p-value of 0.05.
Fig. 1. Mean (±S.E.M.) ratings (mm) of liking, desire to eat, enjoyment, and sweetness (see panels a–d, respectively) as a function of trial-pair number (1–4) and dessert type: “low-energy” dessert (□); “high-energy” dessert (n). Trial pairs 1–3 indicate results elicited during the training sessions. Results from the test days are presented as trial-pair 4.
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
3. Results 3.1. Change in ratings: evidence for flavor-preference learning
Table 1 Summary of multiple regression analyses exploring predictors of the change in liking, desire, enjoyment, and sweetness Dependent measure/predictor
Fig. 1 shows the average ratings of the participants as they progressed through the experiment. Separate panels show the results for the four types of rating: (a) liking, (b) desire, (c) sweetness, and (d) enjoyment. Ratings for the low- and the high-energy desserts are shown in open and closed squares, respectively. At the beginning of the experiment, the high- and the low-energy dessert received very similar ratings on measures of liking, desire, and sweetness. Ratings of enjoyment were slightly higher for the high-energy dessert. However, this difference did not reach significance (F[1,43] = 2.02, p = 0.162). The outcome of our multiple-regression analysis is shown in Table 1. This analysis considered predictors of the change in ratings (exposure 4–exposure 1) during the experiment. In terms of changes in desire, condition emerged as a significant predictor ( p = 0.004). This confirms the reliability of the differences seen in Fig. 1 (panel b), and it can be taken as evidence that learning occurred. With respect to liking, the effect of condition did not approach significance ( p = 0.288). Instead, we found that liking was predicted by the interaction between condition and restraint ( p = 0.015). All other main effects and interaction terms failed to reach significance. To provide a graphical representation of how this change in liking is predicted by the interaction between restraint and condition we derived two groups of participants based on a median split of their DEBQ-restraint scores. Fig. 2 shows the change in liking in these groups separately. Unrestrained and restrained eaters are presented in panels a and b, respectively. Relative to baseline, unrestrained eaters developed a pattern of differential responding that is consistent with flavor-preference learning (increasing liking for the high-energy-paired dessert). By contrast, restrained eaters developed an increase in liking for both desserts, indicating that they failed to learn to discriminate based on their respective energy density. In relation to this interaction, we were interested to understand why the same term failed to predict changes in ratings of desire to eat. After inspecting the raw data, we found a single, highly anomalous, rating made by a participant during her final exposure session. This participant gave a rating of 99 for her liking of one of the desserts but rated her desire to eat the same dessert at 12. This was the case even though her hunger rating was 92 (on a 100-point scales). Given this, we explored the effect of removing this participant from the data set. The results relating to the revised interaction term are entered in parenthesis in Table 1. As indicated, the associated term narrowly missed significance (p = 0.082). Consistent with ratings of liking, unrestrained eaters tended to demonstrate greater evidence for learning. 3.2. Differences in pasta intake: evidence for learned satiation After consuming the (previously) low- and the high-energypaired dessert, the participants ate on average 234 g (S.D. = 125) and 251 g (S.D. = 141) of pasta, respectively (t = 1.04, p = 0.3).
137
Liking Condition Restraint Condition ⁎ Restraint
t
S.E.
β
p
1.07 1.75 − 2.49
1.65 1.80 1.81
1.77 3.16 − 4.51
0.288 0.083 0.015
2.97 − 0.03 − 0.61 (−1.77)
2.24 2.44 2.45 (2.34)
6.65 − 0.06 − 1.49 (− 4.11)
0.004 0.979 0.543 (0.082)
Enjoyment Condition Restraint Condition ⁎ Restraint
0.07 0.12 − 0.49
2.18 2.36 2.36
0.16 0.27 −1.17
0.941 0.908 0.622
Sweetness Condition Restraint Condition ⁎ Restraint
1.54 0.33 −1.17
1.80 2.02 1.98
2.78 0.66 − 2.31
0.127 0.745 0.246
Desire Condition Restraint Condition ⁎ Restraint
Parentheses are used to indicate the outcome of an analysis after the removal of anomalous data from a single participant.
On this basis, we have little evidence that learned satiation occurred (we anticipated less intake following the high-energypaired dessert). For each participant, we also calculated a score based on the difference between the amount of pasta consumed in each condition. These scores failed to correlate significantly with the DEBQ-restraint scores (r = − 0.025, p = 0.808), indicating that restraint is a poor predictor of differences in the amount consumed in each condition. 3.3. Hunger and fullness ratings One possibility is that our findings are associated with differences in the level of hunger and fullness experienced during the experiment. Therefore, we explored ratings of initial hunger and initial fullness firstly using a two-way ANOVA with exposure (1–3) and condition entered as factors, and then using a set of two-sample t-test exploring the effect of condition across the test sessions. All main effects and interactions failed to reach significance. We also considered the possibility that our restraint scores co-varied with a differential appetite for the desserts, and that this might influence the outcome of the CS-US pairings. To test this proposition, we examined the relationship between DEBQ-restraint scores and ratings of hunger and fullness, taken both before and after consuming each dessert. A separate comparison was made for each occasion that the participants attended the laboratory. In addition, we also compared hungerand fullness-change scores, derived from the difference between pre- and post-dessert ratings. This set of analyses revealed only one significant correlation. On day 3, individuals with higher DEBQ-restraint scores tended to report being more full when they arrived for testing (r = 0.455, p = 0.002). All other comparisons failed to reach significance (all p N 0.05). The same comparison on the other test days did not yield significant
138
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
Fig. 2. Mean (±S.E.M.) ratings (mm) of liking. Panels a and b show data from unrestrained and restrained eaters, respectively (distinction based on median split of DEBQ-restraint scores). Results are plotted as a function of trial-pair number (1–4) and dessert type: “low-energy” dessert (□); “high-energy” dessert (n). Trial-pairs 1–3 indicate results elicited during the training sessions. Results from the test days are presented as trial-pair 4.
relationships (all p N 0.185). Therefore, it would seem likely that this relationship is spurious and we have little evidence that differences in hunger and fullness can account for the pattern of results that we report in Fig. 2. 3.4. Contingency awareness, demand awareness, and dietary restraint In relation to our question assessing demand awareness (“What was the purpose of the experiment?”), several participants mentioned taste, fullness, and the effects of repeated exposure. However, none commented on or alluded to the possibility that the desserts might have different postingestive effects and that this might influence the measures that we were taking. In response to the first of our measures of contingency awareness (“The dessert differed in their taste, but did they differ in other ways?”), 33 (75%) of the participants confirmed that they did differ. When asked how they differed, 24 of these respondents referred to taste, texture, or visual differences. One participant did not respond, and eight (including two who also mentioned sensory differences) mentioned or alluded to post-ingestive effects by referring to differences in ‘filling’ or ‘calories.’ The second contingency awareness question explored the participants' ability to identify these differences (“One of the desserts contained an ingredient that makes you feel less hungry and more full. Can you guess
Table 2 Summary of responses to the contingency awareness questions Response
N
Mean (S.D.) restraint score
Question 1. The desserts differed in their taste. But did they differ in other ways? Yes 33 2.445 (0.72) No 11 2.439 (0.99) Question 2. One of the desserts contained an ingredient that makes you feel less hungry and more full. Can you guess which one it was? Incorrect 26 2.26 (0.87) Correct 18 2.70 (0.96) For each response, a corresponding mean (S.D.) DEBQ−restraint score is provided.
which one?”). Of the 44 participants tested, 26 (59%) selected the wrong dessert. Interestingly, of the eight respondents who mentioned post-ingestive effects in the previous question, only four answered this second question correctly (chance responding). Based on the above, it would seem that the level of contingency awareness was similar to what might otherwise be expected from chance. Nevertheless, we decided to test a final proposition that the evidence for learning in our unrestrained eaters can be attributed to a disproportionately higher level of awareness in this group. The results from this comparison are shown in Table 2. In relation to question 1, the DEBQ-restraint scores of Yes and No responders are almost identical. For question 2, although our comparison failed to reach significance (t = 1.54, p = 0.133), we found that ‘correct’ responders tended to have higher DEBQ-restraint scores than did ‘incorrect’ responders. Thus, for both questions, we have little evidence that contingency awareness was associated with lower restraint scores, which makes it unlikely that contingency awareness is a necessary feature of the preference learning that is associated with low levels of dietary restraint. 4. Discussion A general aim of this research was to explore evidence for learned satiation and flavor-preference learning in adult humans. In addition, based on related research, we were concerned to test the prediction that restrained eaters are less likely to acquire associations of this kind. In relation to learned satiation, irrespective of dietary restraint, participants consumed a similar amount of pasta on both test days. This failure to demonstrate learning is consistent with other studies [27–29] and reasons for this lack of reliability have been discussed elsewhere [2]. Nevertheless, Yeomans et al. have recently reported data that might be interpreted as learned satiation in a group of normalweight male participants [11]. Future studies should consider whether this outcome is related to procedural, gender, or attitudinal differences that are peculiar to the sample that was tested. In relation to potential methodological differences, it
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
might be relevant that learned satiation has been assessed in different ways. In the present study, learning was measured by comparing intake of a previously unpaired food (pasta) across conditions. Although this general approach has proved successful in the past [8] a potential criticism is that pasta portion sizes are likely to be determined largely by habit, i.e., learned responses that reflect ‘my average portion of pasta.’ In essence, evidence for learned satiation might fail to emerge because consumption of the second food is, itself, governed by learned satiation. Yeomans et al. [11] overcame this problem by measuring ad lib intake of their novel foods at test. However, in this context, the foods remained in their high- and low-energy formulations (we used the same intermediate energy values). These authors acknowledge that this decision complicates any interpretation about flavor–nutrient associations since learning is never assessed in the context of an absence of energy difference across conditions. In future, researchers should consider comparing ad lib intake of equal-energy versions of the foods that are otherwise presented in fixed quantities during training (for a detailed discussion on methodological issues, see Ref. [36]). It might also be relevant that our participants consumed rather less pasta than in other studies exploring basic ad lib consumption of this food [37,38]. This is likely to be because they consumed a dessert before being offered the pasta. However, we cannot exclude the possibility that learned satiation is observed, but only when the participants consume a larger portion size. Although we failed to observe learned satiation, we did find convincing evidence that flavor-preference learning occurred. After conditioning, participants expressed a relative increase in their desire to eat and their liking for the hitherto high-energypaired dessert. This finding is consistent with recent research by Appleton et al. [24] which, together with our results, represent relatively rare examples of this kind of learning in humans. Importantly, Appleton et al. independently manipulated the effect of energy deprivation and found that learning was evident only in individuals who had been energy deprived during exposure. It remains to be determined whether deprivation state will modulate our results in the same way and this represents a potentially interesting area for future research. In relation to liking, the effect of condition was qualified by dietary restraint as predicted. Specifically, unrestrained eaters developed a differential liking for the high- and the low-energypaired desserts, whereas relatively restrained eaters developed a liking for both desserts. After the removal of an outlier, a similar pattern was observed in relation to desire to eat. However, in this case, the associated interaction term narrowly missed significance. One explanation is that long-term restriction brings about a reduced sensitivity to appetite and hunger [39]. It might also limit responsiveness to the consequences of changes in energy density [40], thereby making people less sensitive to the rewarding characteristics of the US. Consistent with this ‘sensitivity’ hypothesis, it has been noted [2,41] that most successful demonstrations of dietary learning can be found in studies involving young children who have not yet begun to restrict their intake [22,25]. Children are also found to be more sensitive to the energy density of foods, and they seem to lose
139
this ability to self-regulate and respond to internal hunger cues as they adopt restraint-related behaviors [42]. The possibility that individuals differ in their responsiveness to internal cues is also closely related to the concept of ‘externality’ [43]. Pioneering research in the late 1960s and 1970s revealed that obese humans exhibit greater sensitivity to their external environment (i.e., the sight, taste, and smell of food) and that lean people are relatively more sensitive to internal hunger and satiation cues (e.g., [44,45]). However, researchers soon came to realize that externality might not be limited to obese individuals [46], because it is also found in normal-weight restrained eaters [47]. Since that time, issues relating to externality have been discussed primarily in terms of examples of ‘counter regulatory’ behavior observed in restrained eaters [48]. However, questions about the importance of this basic underlying trait remain, and the relationship between externality and dietary learning has not been considered to date. Thus, one possibility is that our measure of restraint operates as a proxy for externality, and that differences between learners and non-learners should otherwise be attributed to variation along this behavioral dimension. There are, however, aspects of the data that complicate an interpretation based solely on insensitivity to internal cues. In particular, it is curious that restrained eaters developed an increase in liking for both desserts. One possibility is that restrained eaters exhibit a greater tendency to generalize their liking across desserts. In the future, it might be interesting to see whether restrained eaters generalize in other contexts, for example, when confronted with the kind of flavor–caffeine pairings that are also known to alter flavor preferences [49,50]. Alternatively, the tendency for restrained eaters to acquire a liking for both the high- and the low-energy-paired dessert might reflect the fact that flavor–flavor learning occurred. As outlined in the Introduction, this kind of associative learning can be considered a shortcut for transferring important information about existing like or dislike from one food to another. Liking for both the high- and the low-energy dessert could be acquired via the same process. This is because both desserts comprised a novel flavor combined with sweetness. At present, this interpretation is difficult to reconcile with evidence, indicating that flavor–flavor learning is limited in individuals who tend to restrict their intake [21]. Nevertheless, more recently, it has been suggested that the outcome of a similar pairing (picture–food) might be influenced both by restraint and by more general beliefs and attitudes about a food US [20]. In the future, it would be interesting to extend this work by exploring dietary restraint in the context of a causal relationship that might be established between attitudes towards the CS and US and the outcome of both flavor–flavor and flavor–nutrient pairings. One way to do this is to manipulate the way that particular foods are labeled, thereby altering expectations about whether they should be regarded as healthy or unhealthy [51]. A more mundane explanation for our findings is that restrained eaters respond differently to the aspartame that was included in some of the desserts. Since restrained eaters are more likely to use dieting products they may be more accepting of the taste of this ingredient. At present, we are unable to rule out this possibility. However, in relation to the specific
140
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
formulations that we used, this explanation would seem unlikely. This is because restrained and unrestrained eaters tended to report a similar liking for the high- and the low-energy desserts at the beginning of the training period. Another general concern might be that our evidence for learning is based on the acquisition of particular kinds of explicit representation. For example, since we alternated the presentation of the CSs, participants had an opportunity to figure out the dessert that was most filling. This explicit information might then color judgments about questions relating to liking for the desserts. Fortunately, we found little evidence to support this proposition. Indeed, taken together, our measures of contingency awareness indicate that levels of awareness did not deviate from chance responding, suggesting that concerns about ecological validity are unfounded. Instead, for now, our results would seem to indicate that flavor–nutrient conditioning represents an automatic and involuntary process. The prospect that this form of Pavlovian learning occurs outside awareness sets it apart from related forms of learning such as drug procurement behaviors [52] and other forms of evaluative conditioning [53]. Indeed, our result helps to confirm recent speculation that dietary learning is so important that it makes sense to learn in a spontaneous way that is facilitated by learning outside conscious control [54]. In the present context, this interpretation is relevant because it would suggest that an ‘elemental’ difference exists between the way in which restrained and unrestrained eaters adapt to their dietary environment. More generally, the idea that differences exist between restrained and unrestrained eaters is supported by studies exploring a range of psychological and behavioral trait characteristics. For example, it has been noted that restrained eaters are more likely to act on impulse [55], and they habituate faster to repeated presentation of neutral stimuli [56]. Our results add to this literature. For now, it remains to be established whether a failure to learn to discriminate based on flavor–nutrient pairings is ‘caused’ directly by the momentary cognitive activity associated with dietary restraint, or whether it represents a more fundamental characteristic that predates or possibly promotes this dietary behavior. Either way, this research merits attention because it represents the first of its kind to indicate that restrained eaters lack appropriate sensitivity to a process that is otherwise regarded as important to everyday dietary behavior. Acknowledgement This research was funded by the Biotechnology and Biological Sciences Research Council (BBSRC); Grant Reference D15238. References [1] Mela DJ. Why do we like what we like? J Sci Food Agric 2001;81:10–6. [2] Brunstrom JM. Dietary learning in humans: directions for future research. Physiol Behav 2005;85:57–65. [3] Sclafani A. Learned controls of ingestive behaviour. Appetite 1997;29:153–8. [4] Kern DL, McPhee L, Fisher J, Johnson S, Birch LL. The postingestive consequences of fat condition preferences for flavors associated with high dietary fat. Physiol Behav 1993;54:71–6.
[5] Booth DA, Mather P, Fuller J. Starch content of ordinary foods associatively conditions human appetite and satiation, indexed by intake and eating pleasantness of starch-paired flavors. Appetite 1982;3:163–84. [6] Gibson EL, Wainwright CJ, Booth DA. Disguised protein in lunch after low-protein breakfast conditions food-flavor preferences dependent on recent lack of protein-intake. Physiol Behav 1995;58:363–71. [7] Le Magnen J. Effets sur la prise alimentaire du rat blanc des administrations postprandiales d'insuline et le mecanisme des appetits caloriques. J Physiol (Paris) 1956;48:789–802. [8] Booth DA, Lee M, McAleavey C. Acquired sensory control of satiation in man. Br J Psychol 1976;67:137–47. [9] Zellner DA, Rozin P, Aron M, Kulish C. Conditioned enhancement of humans liking for flavor by pairing with sweetness. Learn Motiv 1983;14:338–50. [10] Lavin MJ. Establishment of flavor–flavor associations using a sensory preconditioning training procedure. Learn Motiv 1976;7:173–83. [11] Yeomans MR, Weinberg L, James S. Effects of palatability and learned satiety on energy density influences on breakfast intake in humans. Physiol Behav 2005;86:487–99. [12] Wardle J, Johnson F. Weight and dieting: examining levels of weight concern in British adults. Int J Obes 2002;26:1144–9. [13] Ackard DM, Croll JK, Kearney-Cooke A. Dieting frequency among college females: association with disordered eating, body image, and related psychological problems. J Psychosom Res 2002;52:129–36. [14] Van Loan MD, Keim NL. Influence of cognitive eating restraint on totalbody measurements of bone mineral density and bone mineral content in premenopausal women aged 18–45 y: a cross-sectional study. Am J Clin Nutr 2000;72:837–43. [15] McLean JA, Barr SI, Prior JC. Cognitive dietary restraint is associated with higher urinary cortisol excretion in healthy premenopausal women. Am J Clin Nutr 2001;73:7–12. [16] Brunstrom JM, Davison CJ, Mitchell GL. Dietary restraint and cognitive performance in children. Appetite 2005;45:235–41. [17] Hsu LKG. Can dieting cause an eating disorder? Psychol Med 1997;27:509–13. [18] Rosen JC, Gross J, Vara L. Psychological adjustment of adolescents attempting to lose or gain weight. J Consult Clin Psychol 1987;55:742–7. [19] Steinhardt MA, Bezner JR, Adams TB. Outcomes of a traditional weight control program and a nondiet alternative: a one-year comparison. J Psychol 1999;133:495–513. [20] Brunstrom JM, Higgs S, Mitchell GL. Dietary restraint and us devaluation predict evaluative learning. Physiol Behav 2005;85:524–35. [21] Brunstrom JM, Downes CR, Higgs S. Effects of dietary restraint on flavour–flavour learning. Appetite 2001;37:197–206. [22] Birch LL, McPhee L, Steinberg L, Sullivan S. Conditioned flavor preferences in young children. Physiol Behav 1990;47:501–5. [23] Johnson SL, McPhee L, Birch LL. Conditioned preferences—young children prefer flavors associated with high dietary fat. Physiol Behav 1991;50:1245–51. [24] Appleton KM, Gentry RC, Shepherd R. Evidence of a role for conditioning in the development of liking for flavours in humans in everyday life. Physiol Behav 2006;87:478–86. [25] Birch LL, Deysher M. Conditioned and unconditioned caloric compensation—evidence for self-regulation of food-intake in young children. Learn Motiv 1985;16:341–55. [26] Birch LL, McPhee L, Shoba BC, Steinberg L, Krehbiel R. Clean up your plate—effects of child feeding practices on the conditioning of meal size. Learn Motiv 1987;18:301–17. [27] Zandstra EH, Stubenitsky K, De Graaf C, Mela DJ. Effects of learned flavour cues on short-term regulation of food intake in a realistic setting. Physiol Behav 2002;75:83–90. [28] Specter SE, Bellisle F, Hemery-Veron S, Fiquet P, Bornet FRJ, Slama G. Reducing ice cream energy density does not condition decreased acceptance or engender compensation following repeated exposure. Eur J Clin Nutr 1998;52:703–10. [29] Van Wymelbeke V, Beridot-Therond ME, de la Gueronniere V, Fantino M. Influence of repeated consumption of beverages containing sucrose or intense sweeteners on food intake. Eur J Clin Nutr 2004;58:154–61.
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141 [30] Boakes RA, Rossiarnaud C, Garciahoz V. Early experience and reinforcer quality in delayed flavor–food learning in the rat. Appetite 1987;9:191–206. [31] Van Strien T, Frijters JER, Vanstaveren WA, Defares PB, Deurenberg P. The predictive-validity of the Dutch restrained eating scale. Int J Eat Disord 1986;5:747–55. [32] Gray RW, French SJ, Robinson TM, Yeomans MR. Dissociation of the effects of preload volume and energy content on subjective appetite and food intake. Physiol Behav 2002;76:57–64. [33] Gray RW, French SJ, Robinson TM, Yeomans MR. Increasing preload volume with water, reduces rated appetite but not food intake in healthy men even with minimum delay between preload and test meal. Nutr Neurosci 2003;6:29–37. [34] MacCallum RC, Zhang S, Preacher KJ, Rucker DD. On the practice of dichotomization of quantitative variables. Psychol Methods 2002;7:19–40. [35] Brunstrom JM, Mitchell GL, Baguley TS. Potential early-life predictors of dietary behaviour in adulthood: a retrospective study. Int J Obes 2005;29:463–74. [36] Gibson EL, Booth DA. Food-conditioned odour rejection in the late stages of the meal, mediating learnt control of meal volume by aftereffects of food consumption. Appetite 2000;34:295–303. [37] Yeomans MR. Palatability and the micro-structure of feeding in humans: the appetizer effect. Appetite 1996;27:119–33. [38] Yeomans MR, Gray RW, Mitchell CJ, True S. Independent effects of palatability and within-meal pauses on intake and appetite ratings in human volunteers. Appetite 1997;29:61–76. [39] LeGoff DB, Leichner P, Spigelman MN. Salivary response to olfactory food stimuli in anorexics and bulimics. Appetite 1988;11:15–25. [40] Rolls BJ, Kimharris S, Fischman MW, Foltin RW, Moran TH, Stoner SA. Satiety after preloads with different amounts of fat and carbohydrate— implications for obesity. Am J Clin Nutr 1994;60:476–87. [41] Brunstrom JM. Does dietary learning occur outside awareness? Conscious Cogn 2004;13:453–70. [42] Birch LL, Fisher JO, Davison KK. Learning to overeat: maternal use of restrictive feeding practices promotes girls' eating in the absence of hunger. Am J Clin Nutr 2003;78:215–20.
141
[43] Schachter S. Obesity and eating—internal and external cues differentially affect eating behavior of obese and normal subjects. Science 1968;161:751. [44] Schachter S. Some extraordinary facts about obese humans and rats. Am Psychol 1971;26:129–44. [45] Schachter S, Rodin J. Obese humans and rats. Washington DC: Erlbaum/ Halsted; 1974. [46] Nisbett RE. Hunger, obesity, and ventromedial hypothalamus. Psychol Rev 1972;79:433–53. [47] Herman CP, Mack D. Restrained and unrestrained eating. J Pers 1975;43:647–60. [48] Ruderman AJ. Dietary restraint—a theoretical and empirical review. Psychol Bull 1986;99:247–62. [49] Rogers PJ, Richardson NJ, Elliman NA. Overnight caffeine abstinence and negative reinforcement of preference for caffeine-containing drinks. Psychopharmacology (Berl) 1995;120:457–62. [50] Yeomans MR, Jackson A, Lee MD, Steer B, Tinley E, Durlach P, et al. Acquisition and extinction of flavour preferences conditioned by caffeine in humans. Appetite 2000;35:131–41. [51] Yeomans MR, Lartamo S, Procter EL, Lee MD, Gray RW. The actual, but not labelled, fat content of a soup preload alters short-term appetite in healthy men. Physiol Behav 2001;73:533–40. [52] Hogarth L, Duka T. Human nicotine conditioning requires explicit contingency knowledge: is addictive behaviour cognitively mediated? Psychopharmacology (Berl) 2006;184:553–66. [53] Lovibond PF, Shanks DR. The role of awareness in pavlovian conditioning: empirical evidence and theoretical implications. J Exp Psychol Anim Behav Processes 2002;28:3–26. [54] Field AP. I like it, but i'm not sure why: can evaluative conditioning occur without conscious awareness? Conscious Cogn 2000;9:13–36. [55] Nederkoorn C, van Eijs Y, Jansen A. Restrained eaters act on impulse. Pers Individ Differ 2004;37:1651–8. [56] Jansen A, Klaver J, Merckelbach H, Vandenhout M. Restrained eaters are rapidly habituating sensation seekers. Behav Res Ther 1989;27:247–52.
Flavor–nutrient learning in restrained and unrestrained eaters Jeffrey M. Brunstrom a,⁎, Gemma L. Mitchell b a
Department of Experimental Psychology, University of Bristol, 12a Priory Road, Bristol, BS8 1TU, England, UK b Department of Human Sciences, Loughborough University, England, UK Received 26 May 2006; received in revised form 14 September 2006; accepted 14 September 2006
Abstract After we consume a novel food an association can form between its sensory characteristics (e.g., taste properties) and the effect it has on the body (rewarding). Associations of this kind underpin much of our everyday dietary behavior because they mediate both the affective quality of food (flavor-preference learning) and the amount that we choose to consume (learning satiation). Notwithstanding this fact, very few studies have successfully demonstrated the process of dietary learning in human adults. In addition, based on evidence from related research, we explored whether learning is less likely to occur in individuals who have high scores on a measure of dietary restraint. Female participants (N = 44) consumed two differently flavored desserts. Each was presented three times on separate days. One was formulated with a high-energy content (1882 kJ) and the other with a low-energy content (226 kJ). After training, we found little evidence for learned satiation. However, we did observe flavor-preference learning. Specifically, participants acquired a greater liking and desire-to-eat the dessert flavor that was paired with a higher energy density during training. Further analysis revealed that this effect on liking is qualified by dietary restraint. As predicted, unrestrained eaters demonstrated greater differential responding to the two desserts than did restrained eaters. These data provide further evidence for flavor–nutrient learning in adults and they highlight a hitherto unexplored and potentially important difference between restrained and unrestrained eaters. © 2006 Elsevier Inc. All rights reserved. Keywords: Dietary restraint; Appetite; Flavor preference; Dietary learning; Human; Conditioned satiety; Satiation; Associative learning
1. Introduction A critical component of ‘normal’ dietary control is an ability to learn about the consequences of consuming a particular food [1]. As reviewed recently [2], in both animals and humans, ‘flavor–nutrient’ associations can be formed based on the temporal contiguity between the sensory characteristics of a novel tasting food (the ‘conditioned stimulus’ [CS]) and the post-ingestive consequences of its consumption (the ‘unconditioned stimulus’ [US]). When a US is linked to a high-energy density this kind of ‘flavor–nutrient’ association can bring about an increased affection for the CS [3]. In animals, learning tends to be evidenced either by preferential intake in a twobottle test, a “conditioned-taste preference,” or as an increase in intake in a one-bottle test, a “conditioned-flavor acceptance.” By contrast, human learning has tended to be assessed using a variety of different rating and ranking procedures, often in
⁎ Corresponding author. Tel.: +44 117 928 88574; fax: +44 117 928 88588. E-mail address: [email protected] (J.M. Brunstrom). 0031-9384/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.physbeh.2006.09.016
response to questions relating to pleasantness and desire to consume the CS (e.g., [4–6]). Unfortunately, it remains unclear how these measures map conceptually onto a conditioned-taste preference or a conditioned flavor acceptance, and a resolution of this issue is beyond the scope of this study. Here, for simplicity, and to establish a useful collective term, we will refer to all learned changes in affective measures as an example of a “conditioned flavor preference.” However, in so doing, the reader should note that we are not referring to a specific process that might otherwise be observed in animal behavior. Flavor–nutrient associations can also modify meal size. During a meal, there may be insufficient post-ingestive feedback to terminate eating behavior [7]. Instead, meal size may become governed by an anticipation of the likely effects of eating food. This response appears to be based on an association between the flavor of a food and visceral cues (US) that are present towards the end of a meal [5,8], and it is referred to as ‘learned satiation’ [8]. In addition to flavor–nutrient learning, animals and humans also acquire preferences via a more immediate process. ‘Flavor–flavor learning’ occurs when a neutral flavor (CS) is
134
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
presented simultaneously and repeatedly (in a single session) with a substance (e.g., sucrose) that is already liked (US). After this exposure, the valance of the CS shifts in the direction of the US [9,10]. Together, these various forms of dietary learning are important, because they account for many of our everyday decisions about food choice and portion size. Yet despite various speculation [2,11], it is surprising that very little is known about whether individual differences in dietary learning are predicted by everyday dietary behaviors and attitudes. One of the most researched individual difference is the tendency to engage in ‘dietary restraint.’ This term refers to a conscious attempt to limit food intake by overriding the processes that are responsible for dietary control. In some societies, dietary restraint is very common, especially amongst women [12]. In some cases, this behavior is appropriate, because it can form part of a strategy to prevent obesity. However, it is also the case that many normal-weight individuals restrict their intake for other reasons, often in response to concerns about body image [13]. In relation to this group, it should be noted that chronic dietary restriction is related to a range of undesirable outcomes, including low bone-mineral content [14], high cortisol excretion [15], cognitive impairments [16], eating disorders [17], and anxiety [18]. Given these problems, one solution might be to find ways to encourage ‘normal’ or ‘natural’ eating behavior [19]. However, this approach presupposes that restrained eaters have access to the controls of dietary behavior that are otherwise available to their unrestrained counterparts. This study explores aspects of this proposition. In a recent study, flavor–flavor learning was compared in groups of high- and low-restrained eaters [20]. Novel flavors were paired several times with the presentation of a sweet reward. After this period, unrestrained eaters reported a relative increase in liking for the flavors that had been paired most often with a sweet reward. By contrast, restrained eaters reported a greater preference for the flavors that were paired less often. Although the reason for this difference remains to be clarified, the notion that flavor–flavor learning is disrupted is consistent with an earlier study indicating that restrained eaters fail to acquire this type of association [21]. One possibility is that this difference generalizes to the formation of associations based on flavor– nutrient pairings (i.e., flavor–nutrient learning). This is important because evidence of this kind would support the proposition that dieting undermines a range of processes that are otherwise important to the maintenance of normal dietary control. Accordingly, an aim of this study was to determine whether a measure of dietary restraint predicts evidence for either learned satiation or flavor-preference learning, after participants have been exposed to a set of flavor–nutrient pairings. Previously, flavor-preference learning has been observed in preschool children [4,22,23]. However, evidence in adults has been extremely limited [5,6,24]. Similarly, following initial evidence for learned satiation in adults [5,8], successful learning has been observed only in children [25,26] and failures have been reported in studies using adults [27–29] (but see Ref. [11]). Previously, it has been suggested that learning might be limited to childhood because a critical period exists during which flavor–nutrient associations can only be acquired [2].
Therefore, in addition to exploring issues relating to dietary restraint, this research was motivated by a more general concern to explore the extent to which successful dietary learning can be demonstrated in adult humans. 2. Method 2.1. Experiment overview Female participants were exposed to a distinctly flavored high- and low-energy dessert on alternating days, with three exposures to each flavor CS paired with energy US. After this training (days 1–6), the participants received each flavored dessert once more, again on separate days (day 7 and day 8). This time, they were formulated with an intermediate and identical energy density. Learned satiation was assessed by comparing intake of pasta salad after consuming each flavor. Flavor-preference learning was assessed by comparing changes in the affective and sensory quality of the desserts throughout the experiment. On the final day (day 9), we explored the extent to which the participants had acquired an explicit appreciation of the relative consequences of consuming each dessert. 2.2. Participants Female participants were recruited via email from the student population at Loughborough University (UK). After inspecting the data, we were concerned to eliminate participants who started the experiment with a very different preference for the two desserts. Evidence of this kind indicates a degree of familiarity that could limit the extent to which learning is observed [30]. Therefore, post hoc, we applied a crude cut-off point and removed those participants who rated their liking for one of the desserts more than 50 points (on a 100-point scale) higher than their rating for the other dessert at the beginning of the experiment (details of the ratings that were used are given below). On this basis, four participants were removed from the data set leaving a final sample of 44 participants (mean age = 20.5 years [S.D. = 2.00]). Of the remaining participants, two were currently dieting to lose weight, one smoked tobacco, and three were taking medication for asthma. At the outset, all participants offered signed consent and were informed of their right to withdraw from the experiment. The Loughborough University Ethics Committee gave approval for the general protocol. All participants were paid £25 for their assistance. 2.3. Materials The two desserts were designed to be very different and highly novel. Both were comprised of flavored yogurt mixed with small chunks of Jell-O® (gelatin-based dessert). The fruitflavored dessert was served in a tall dessert glass and the chocolate dessert was served in a relatively squat glass bowl. Both glasses had a similar capacity. The fruit-flavored dessert was served with dried fruit flakes sprinkled on the top and the chocolate dessert was topped with whipped cream. Both
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
toppings weighed approximately 2 g (∼ 10 kJ). The fruitflavored dessert was formulated using a combination of mandarin-flavored Jell-O® and vanilla-flavored yogurt, and the chocolate-flavored dessert was formulated using a combination of raspberry-flavored Jell-O® and chocolate-flavored yogurt. The Jell-O® in the low-calorie version was formed by mixing 4.44 g (57 kJ) of ‘Sugar Free Jelly Flakes’ (Rowntrees) with 185 ml of water (as per manufacturer's instructions). The Jell-O® in the high-calorie version was produced by mixing 33.7 g (419 kJ) of ‘regular’ Jell-O® (Tesco stores own brand) with 94 ml of water (30 ml less than recommended by manufacturer, but an otherwise identical process). We added 60 g (1005 kJ) of maltodextrin (Cerestar C⁎DRY 01910, supplied by Albion Chemical Distribution, Coalville, Leicestershire) to this mixture before the Jell-O® had set. The desserts in the lowcalorie condition contained 70 g of low-calorie yogurt (Müllerlight, 159 kJ). In the high-calorie condition, the desserts contained 50 g of the same yogurt (114 kJ), to which 20 g of maltodextrin was added (448.5 kJ in total). Both flavors (fruit and chocolate) of the high-energy and the low-energy dessert had identical energy content, 1882 kJ and 226 kJ, respectively. The high-energy desserts weighed 273 g and their composition was 1% fat, 96% carbohydrate, and 3% protein. The low-energy desserts weighed 260 g; their composition was 6% fat, 49% carbohydrate, and 45% protein. Informal pilot testing indicated that high- and low-energy versions of each variety tasted and looked very similar. Based on taste tests, we found little evidence that participants could reliably identify one particular version that they thought would be more filling. This issue was also explored in some detail at the end of the experiment—see Section 3.4. The pasta salad was served in a large bowl containing more food than could be consumed at a single sitting. A 100 g serving (approx. 774 kJ) contained 75 g of cooked pasta, 8 g of mayonnaise, 5 g of cucumber, 5 g of green pepper, and 7 g of cherry tomatoes. 2.4. Evaluating the desserts Visual-analogue rating scales (100 mm) were used to explore how attitudes towards the desserts changed during the course of the experiment. Three ratings scales were issued before the participants consumed a dessert. The first of these assessed the participants' liking for its taste. After sampling the dessert, the participants were then encouraged to think about their most and least favorite dessert. They then completed a rating that was headed “How much do you like the taste of that dessert?” with anchor points “The same as my least favorite dessert” and “The same as my most favorite dessert,” on the left- and right-hand side, respectively. Participants then completed two further ratings, one headed “How sweet is that food?” and the other headed “How strong is your desire to eat this dessert right now?” with anchor points “Not at all sweet/strong” and “Extremely sweet/strong.” After consuming a full portion of a dessert, the participants completed a final rating headed “How much did you enjoy eating that food?” anchored “not at all” and “very much.”
135
These four measures are referred to here as (a) ‘liking,’ (b) ‘sweet,’ (c) ‘desire,’ and (d) ‘enjoyment.’ A measure of sweetness was included to explore whether the sensory characteristics of the desserts changed or remained stable during the experiment. 2.5. Procedure 2.5.1. Pre-exposure session At a preliminary screening session the participants were told to sample the pasta salad. All confirmed that they would be willing to eat this food for lunch. Participants also completed the restraint section of the Dutch-Eating Behavior Questionnaire (DEBQ) [31]. This measure of restraint was selected because it has been used previously to identify different patterns of flavor–flavor learning in restrained and unrestrained eaters [21]. Finally, a measure of height and weight was recorded and the participants were given detailed instructions on how to complete a visual-analogue rating scale. 2.5.2. Training (days 1–6) Six training sessions were conducted in the Ingestive Behavior Laboratory at Loughborough University, between 12:00 p.m. and 2:00 p.m., on non-consecutive days. All of the sessions were spaced no more than 3 days apart. Across the six sessions, the high- and the low-energy desserts were consumed on alternate days. The allocation of a particular flavor (chocolate or fruit) to the high- and the low-energy condition, together with the ordering of the high- and low-energy desserts, was counterbalanced across participants. (Note that with four participants removed from the data set perfect counterbalancing was lost. However, the most important consideration is the balanced allocation of each flavor to the high- and low-energy condition across participants. This aspect of the counterbalancing was preserved in the final data set.) All participants were instructed to consume their normal breakfast during the morning of each test day. They were also told to refrain from eating after 9:00 a.m. in the morning. To increase compliance with this request, all participants were told to complete a short food diary throughout each test day. On arrival, the participants rated their initial hunger and fullness (100 mm visual-analogue scales). They then sampled and consumed the appropriate dessert for that session. As described above, we also required the participants to produce four ratings. The participants then left the laboratory and they were told that they should refrain from eating or drinking for 90 min. Entries in the food diaries indicated compliance with this request. 2.5.3. Test of learned satiation (days 7 and 8) Each dessert was given to the participants once more, again on separate days. The ordering of the desserts followed the alternating pattern established during training. However, following earlier work [8], both desserts were served with an identical and intermediate energy density (1054 kJ). Ostensibly, the procedure did not change. However, after making the final evaluative rating the participants were instructed to provide a
136
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
second hunger and fullness rating. Again, following earlier studies [5,27], learned satiation was assessed by comparing intake of a second food after consuming a fixed portion of the food that was presented during training. Here, we chose to use pasta salad because this food is generally liked and it is relatively uniform. It has been also been shown to be a sensitive test food in basic preloading studies [32,33]. After this period, the participants provided a final rating of their hunger and fullness. 2.5.4. Assessment of demand and contingency awareness (day 9) On the day immediately after the final test session, the participants were asked to complete an ‘awareness questionnaire.’ Initially, to assess demand characteristics, the participants were asked to guess the purpose of the experiment. Then, after confirming that they remembered the two desserts, they were given two questions that assessed contingency awareness. The first of these was worded, “The desserts differed in their taste, but did they differ in other ways?” (options: ‘yes they differed’ and ‘no they seemed the same.’). If they responded ‘yes,’ then they were asked to write down the way in which they desserts differed. On a separate sheet of paper, the participants were then told, “One of the desserts contained an ingredient that makes you fell less hungry and more full” and they were asked, “Can you guess which one?” (options: ‘chocolate’ and ‘fruit’). Finally, the exact purpose of the experiment was explained, the participants were thanked, and they received payment for their participation.
2.6. Data analysis To explore evidence for preference learning, we derived a set of change scores based on the difference between ratings made at the beginning and the end of the experiment (exposure 4–exposure 1). Thus, for each participant and each type of rating, we calculated two change scores, one relating to the highenergy condition and the other relating to the low-energy condition. The common approach of taking median splits of continuous predictors such as restraint scores prior to ANOVA was avoided as it reduces power and may produce spurious interaction effects [34]. Instead, as in previous studies [35], we entered restraint scores as a continuous variable into a multipleregression analysis. Our aim was to explore whether our change scores are predicted by condition (high/low energy), restraint, and the interaction between restraint and condition. In all regression analyses, ratings made during exposure 1 were entered as a controlling variable. The possibility of co-linearity was addressed by centering the restraint scores before calculating the restraint ⁎ condition interaction term. To explore evidence for learned satiation, we compared the amount of pasta consumed during the two test sessions using a paired t-test. When relatively greater intake is observed after consuming the (previously) low-energy-paired dessert then this can be taken as evidence that learning has occurred. In all of our statistical analyses, we used the conventional critical p-value of 0.05.
Fig. 1. Mean (±S.E.M.) ratings (mm) of liking, desire to eat, enjoyment, and sweetness (see panels a–d, respectively) as a function of trial-pair number (1–4) and dessert type: “low-energy” dessert (□); “high-energy” dessert (n). Trial pairs 1–3 indicate results elicited during the training sessions. Results from the test days are presented as trial-pair 4.
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
3. Results 3.1. Change in ratings: evidence for flavor-preference learning
Table 1 Summary of multiple regression analyses exploring predictors of the change in liking, desire, enjoyment, and sweetness Dependent measure/predictor
Fig. 1 shows the average ratings of the participants as they progressed through the experiment. Separate panels show the results for the four types of rating: (a) liking, (b) desire, (c) sweetness, and (d) enjoyment. Ratings for the low- and the high-energy desserts are shown in open and closed squares, respectively. At the beginning of the experiment, the high- and the low-energy dessert received very similar ratings on measures of liking, desire, and sweetness. Ratings of enjoyment were slightly higher for the high-energy dessert. However, this difference did not reach significance (F[1,43] = 2.02, p = 0.162). The outcome of our multiple-regression analysis is shown in Table 1. This analysis considered predictors of the change in ratings (exposure 4–exposure 1) during the experiment. In terms of changes in desire, condition emerged as a significant predictor ( p = 0.004). This confirms the reliability of the differences seen in Fig. 1 (panel b), and it can be taken as evidence that learning occurred. With respect to liking, the effect of condition did not approach significance ( p = 0.288). Instead, we found that liking was predicted by the interaction between condition and restraint ( p = 0.015). All other main effects and interaction terms failed to reach significance. To provide a graphical representation of how this change in liking is predicted by the interaction between restraint and condition we derived two groups of participants based on a median split of their DEBQ-restraint scores. Fig. 2 shows the change in liking in these groups separately. Unrestrained and restrained eaters are presented in panels a and b, respectively. Relative to baseline, unrestrained eaters developed a pattern of differential responding that is consistent with flavor-preference learning (increasing liking for the high-energy-paired dessert). By contrast, restrained eaters developed an increase in liking for both desserts, indicating that they failed to learn to discriminate based on their respective energy density. In relation to this interaction, we were interested to understand why the same term failed to predict changes in ratings of desire to eat. After inspecting the raw data, we found a single, highly anomalous, rating made by a participant during her final exposure session. This participant gave a rating of 99 for her liking of one of the desserts but rated her desire to eat the same dessert at 12. This was the case even though her hunger rating was 92 (on a 100-point scales). Given this, we explored the effect of removing this participant from the data set. The results relating to the revised interaction term are entered in parenthesis in Table 1. As indicated, the associated term narrowly missed significance (p = 0.082). Consistent with ratings of liking, unrestrained eaters tended to demonstrate greater evidence for learning. 3.2. Differences in pasta intake: evidence for learned satiation After consuming the (previously) low- and the high-energypaired dessert, the participants ate on average 234 g (S.D. = 125) and 251 g (S.D. = 141) of pasta, respectively (t = 1.04, p = 0.3).
137
Liking Condition Restraint Condition ⁎ Restraint
t
S.E.
β
p
1.07 1.75 − 2.49
1.65 1.80 1.81
1.77 3.16 − 4.51
0.288 0.083 0.015
2.97 − 0.03 − 0.61 (−1.77)
2.24 2.44 2.45 (2.34)
6.65 − 0.06 − 1.49 (− 4.11)
0.004 0.979 0.543 (0.082)
Enjoyment Condition Restraint Condition ⁎ Restraint
0.07 0.12 − 0.49
2.18 2.36 2.36
0.16 0.27 −1.17
0.941 0.908 0.622
Sweetness Condition Restraint Condition ⁎ Restraint
1.54 0.33 −1.17
1.80 2.02 1.98
2.78 0.66 − 2.31
0.127 0.745 0.246
Desire Condition Restraint Condition ⁎ Restraint
Parentheses are used to indicate the outcome of an analysis after the removal of anomalous data from a single participant.
On this basis, we have little evidence that learned satiation occurred (we anticipated less intake following the high-energypaired dessert). For each participant, we also calculated a score based on the difference between the amount of pasta consumed in each condition. These scores failed to correlate significantly with the DEBQ-restraint scores (r = − 0.025, p = 0.808), indicating that restraint is a poor predictor of differences in the amount consumed in each condition. 3.3. Hunger and fullness ratings One possibility is that our findings are associated with differences in the level of hunger and fullness experienced during the experiment. Therefore, we explored ratings of initial hunger and initial fullness firstly using a two-way ANOVA with exposure (1–3) and condition entered as factors, and then using a set of two-sample t-test exploring the effect of condition across the test sessions. All main effects and interactions failed to reach significance. We also considered the possibility that our restraint scores co-varied with a differential appetite for the desserts, and that this might influence the outcome of the CS-US pairings. To test this proposition, we examined the relationship between DEBQ-restraint scores and ratings of hunger and fullness, taken both before and after consuming each dessert. A separate comparison was made for each occasion that the participants attended the laboratory. In addition, we also compared hungerand fullness-change scores, derived from the difference between pre- and post-dessert ratings. This set of analyses revealed only one significant correlation. On day 3, individuals with higher DEBQ-restraint scores tended to report being more full when they arrived for testing (r = 0.455, p = 0.002). All other comparisons failed to reach significance (all p N 0.05). The same comparison on the other test days did not yield significant
138
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
Fig. 2. Mean (±S.E.M.) ratings (mm) of liking. Panels a and b show data from unrestrained and restrained eaters, respectively (distinction based on median split of DEBQ-restraint scores). Results are plotted as a function of trial-pair number (1–4) and dessert type: “low-energy” dessert (□); “high-energy” dessert (n). Trial-pairs 1–3 indicate results elicited during the training sessions. Results from the test days are presented as trial-pair 4.
relationships (all p N 0.185). Therefore, it would seem likely that this relationship is spurious and we have little evidence that differences in hunger and fullness can account for the pattern of results that we report in Fig. 2. 3.4. Contingency awareness, demand awareness, and dietary restraint In relation to our question assessing demand awareness (“What was the purpose of the experiment?”), several participants mentioned taste, fullness, and the effects of repeated exposure. However, none commented on or alluded to the possibility that the desserts might have different postingestive effects and that this might influence the measures that we were taking. In response to the first of our measures of contingency awareness (“The dessert differed in their taste, but did they differ in other ways?”), 33 (75%) of the participants confirmed that they did differ. When asked how they differed, 24 of these respondents referred to taste, texture, or visual differences. One participant did not respond, and eight (including two who also mentioned sensory differences) mentioned or alluded to post-ingestive effects by referring to differences in ‘filling’ or ‘calories.’ The second contingency awareness question explored the participants' ability to identify these differences (“One of the desserts contained an ingredient that makes you feel less hungry and more full. Can you guess
Table 2 Summary of responses to the contingency awareness questions Response
N
Mean (S.D.) restraint score
Question 1. The desserts differed in their taste. But did they differ in other ways? Yes 33 2.445 (0.72) No 11 2.439 (0.99) Question 2. One of the desserts contained an ingredient that makes you feel less hungry and more full. Can you guess which one it was? Incorrect 26 2.26 (0.87) Correct 18 2.70 (0.96) For each response, a corresponding mean (S.D.) DEBQ−restraint score is provided.
which one?”). Of the 44 participants tested, 26 (59%) selected the wrong dessert. Interestingly, of the eight respondents who mentioned post-ingestive effects in the previous question, only four answered this second question correctly (chance responding). Based on the above, it would seem that the level of contingency awareness was similar to what might otherwise be expected from chance. Nevertheless, we decided to test a final proposition that the evidence for learning in our unrestrained eaters can be attributed to a disproportionately higher level of awareness in this group. The results from this comparison are shown in Table 2. In relation to question 1, the DEBQ-restraint scores of Yes and No responders are almost identical. For question 2, although our comparison failed to reach significance (t = 1.54, p = 0.133), we found that ‘correct’ responders tended to have higher DEBQ-restraint scores than did ‘incorrect’ responders. Thus, for both questions, we have little evidence that contingency awareness was associated with lower restraint scores, which makes it unlikely that contingency awareness is a necessary feature of the preference learning that is associated with low levels of dietary restraint. 4. Discussion A general aim of this research was to explore evidence for learned satiation and flavor-preference learning in adult humans. In addition, based on related research, we were concerned to test the prediction that restrained eaters are less likely to acquire associations of this kind. In relation to learned satiation, irrespective of dietary restraint, participants consumed a similar amount of pasta on both test days. This failure to demonstrate learning is consistent with other studies [27–29] and reasons for this lack of reliability have been discussed elsewhere [2]. Nevertheless, Yeomans et al. have recently reported data that might be interpreted as learned satiation in a group of normalweight male participants [11]. Future studies should consider whether this outcome is related to procedural, gender, or attitudinal differences that are peculiar to the sample that was tested. In relation to potential methodological differences, it
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
might be relevant that learned satiation has been assessed in different ways. In the present study, learning was measured by comparing intake of a previously unpaired food (pasta) across conditions. Although this general approach has proved successful in the past [8] a potential criticism is that pasta portion sizes are likely to be determined largely by habit, i.e., learned responses that reflect ‘my average portion of pasta.’ In essence, evidence for learned satiation might fail to emerge because consumption of the second food is, itself, governed by learned satiation. Yeomans et al. [11] overcame this problem by measuring ad lib intake of their novel foods at test. However, in this context, the foods remained in their high- and low-energy formulations (we used the same intermediate energy values). These authors acknowledge that this decision complicates any interpretation about flavor–nutrient associations since learning is never assessed in the context of an absence of energy difference across conditions. In future, researchers should consider comparing ad lib intake of equal-energy versions of the foods that are otherwise presented in fixed quantities during training (for a detailed discussion on methodological issues, see Ref. [36]). It might also be relevant that our participants consumed rather less pasta than in other studies exploring basic ad lib consumption of this food [37,38]. This is likely to be because they consumed a dessert before being offered the pasta. However, we cannot exclude the possibility that learned satiation is observed, but only when the participants consume a larger portion size. Although we failed to observe learned satiation, we did find convincing evidence that flavor-preference learning occurred. After conditioning, participants expressed a relative increase in their desire to eat and their liking for the hitherto high-energypaired dessert. This finding is consistent with recent research by Appleton et al. [24] which, together with our results, represent relatively rare examples of this kind of learning in humans. Importantly, Appleton et al. independently manipulated the effect of energy deprivation and found that learning was evident only in individuals who had been energy deprived during exposure. It remains to be determined whether deprivation state will modulate our results in the same way and this represents a potentially interesting area for future research. In relation to liking, the effect of condition was qualified by dietary restraint as predicted. Specifically, unrestrained eaters developed a differential liking for the high- and the low-energypaired desserts, whereas relatively restrained eaters developed a liking for both desserts. After the removal of an outlier, a similar pattern was observed in relation to desire to eat. However, in this case, the associated interaction term narrowly missed significance. One explanation is that long-term restriction brings about a reduced sensitivity to appetite and hunger [39]. It might also limit responsiveness to the consequences of changes in energy density [40], thereby making people less sensitive to the rewarding characteristics of the US. Consistent with this ‘sensitivity’ hypothesis, it has been noted [2,41] that most successful demonstrations of dietary learning can be found in studies involving young children who have not yet begun to restrict their intake [22,25]. Children are also found to be more sensitive to the energy density of foods, and they seem to lose
139
this ability to self-regulate and respond to internal hunger cues as they adopt restraint-related behaviors [42]. The possibility that individuals differ in their responsiveness to internal cues is also closely related to the concept of ‘externality’ [43]. Pioneering research in the late 1960s and 1970s revealed that obese humans exhibit greater sensitivity to their external environment (i.e., the sight, taste, and smell of food) and that lean people are relatively more sensitive to internal hunger and satiation cues (e.g., [44,45]). However, researchers soon came to realize that externality might not be limited to obese individuals [46], because it is also found in normal-weight restrained eaters [47]. Since that time, issues relating to externality have been discussed primarily in terms of examples of ‘counter regulatory’ behavior observed in restrained eaters [48]. However, questions about the importance of this basic underlying trait remain, and the relationship between externality and dietary learning has not been considered to date. Thus, one possibility is that our measure of restraint operates as a proxy for externality, and that differences between learners and non-learners should otherwise be attributed to variation along this behavioral dimension. There are, however, aspects of the data that complicate an interpretation based solely on insensitivity to internal cues. In particular, it is curious that restrained eaters developed an increase in liking for both desserts. One possibility is that restrained eaters exhibit a greater tendency to generalize their liking across desserts. In the future, it might be interesting to see whether restrained eaters generalize in other contexts, for example, when confronted with the kind of flavor–caffeine pairings that are also known to alter flavor preferences [49,50]. Alternatively, the tendency for restrained eaters to acquire a liking for both the high- and the low-energy-paired dessert might reflect the fact that flavor–flavor learning occurred. As outlined in the Introduction, this kind of associative learning can be considered a shortcut for transferring important information about existing like or dislike from one food to another. Liking for both the high- and the low-energy dessert could be acquired via the same process. This is because both desserts comprised a novel flavor combined with sweetness. At present, this interpretation is difficult to reconcile with evidence, indicating that flavor–flavor learning is limited in individuals who tend to restrict their intake [21]. Nevertheless, more recently, it has been suggested that the outcome of a similar pairing (picture–food) might be influenced both by restraint and by more general beliefs and attitudes about a food US [20]. In the future, it would be interesting to extend this work by exploring dietary restraint in the context of a causal relationship that might be established between attitudes towards the CS and US and the outcome of both flavor–flavor and flavor–nutrient pairings. One way to do this is to manipulate the way that particular foods are labeled, thereby altering expectations about whether they should be regarded as healthy or unhealthy [51]. A more mundane explanation for our findings is that restrained eaters respond differently to the aspartame that was included in some of the desserts. Since restrained eaters are more likely to use dieting products they may be more accepting of the taste of this ingredient. At present, we are unable to rule out this possibility. However, in relation to the specific
140
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141
formulations that we used, this explanation would seem unlikely. This is because restrained and unrestrained eaters tended to report a similar liking for the high- and the low-energy desserts at the beginning of the training period. Another general concern might be that our evidence for learning is based on the acquisition of particular kinds of explicit representation. For example, since we alternated the presentation of the CSs, participants had an opportunity to figure out the dessert that was most filling. This explicit information might then color judgments about questions relating to liking for the desserts. Fortunately, we found little evidence to support this proposition. Indeed, taken together, our measures of contingency awareness indicate that levels of awareness did not deviate from chance responding, suggesting that concerns about ecological validity are unfounded. Instead, for now, our results would seem to indicate that flavor–nutrient conditioning represents an automatic and involuntary process. The prospect that this form of Pavlovian learning occurs outside awareness sets it apart from related forms of learning such as drug procurement behaviors [52] and other forms of evaluative conditioning [53]. Indeed, our result helps to confirm recent speculation that dietary learning is so important that it makes sense to learn in a spontaneous way that is facilitated by learning outside conscious control [54]. In the present context, this interpretation is relevant because it would suggest that an ‘elemental’ difference exists between the way in which restrained and unrestrained eaters adapt to their dietary environment. More generally, the idea that differences exist between restrained and unrestrained eaters is supported by studies exploring a range of psychological and behavioral trait characteristics. For example, it has been noted that restrained eaters are more likely to act on impulse [55], and they habituate faster to repeated presentation of neutral stimuli [56]. Our results add to this literature. For now, it remains to be established whether a failure to learn to discriminate based on flavor–nutrient pairings is ‘caused’ directly by the momentary cognitive activity associated with dietary restraint, or whether it represents a more fundamental characteristic that predates or possibly promotes this dietary behavior. Either way, this research merits attention because it represents the first of its kind to indicate that restrained eaters lack appropriate sensitivity to a process that is otherwise regarded as important to everyday dietary behavior. Acknowledgement This research was funded by the Biotechnology and Biological Sciences Research Council (BBSRC); Grant Reference D15238. References [1] Mela DJ. Why do we like what we like? J Sci Food Agric 2001;81:10–6. [2] Brunstrom JM. Dietary learning in humans: directions for future research. Physiol Behav 2005;85:57–65. [3] Sclafani A. Learned controls of ingestive behaviour. Appetite 1997;29:153–8. [4] Kern DL, McPhee L, Fisher J, Johnson S, Birch LL. The postingestive consequences of fat condition preferences for flavors associated with high dietary fat. Physiol Behav 1993;54:71–6.
[5] Booth DA, Mather P, Fuller J. Starch content of ordinary foods associatively conditions human appetite and satiation, indexed by intake and eating pleasantness of starch-paired flavors. Appetite 1982;3:163–84. [6] Gibson EL, Wainwright CJ, Booth DA. Disguised protein in lunch after low-protein breakfast conditions food-flavor preferences dependent on recent lack of protein-intake. Physiol Behav 1995;58:363–71. [7] Le Magnen J. Effets sur la prise alimentaire du rat blanc des administrations postprandiales d'insuline et le mecanisme des appetits caloriques. J Physiol (Paris) 1956;48:789–802. [8] Booth DA, Lee M, McAleavey C. Acquired sensory control of satiation in man. Br J Psychol 1976;67:137–47. [9] Zellner DA, Rozin P, Aron M, Kulish C. Conditioned enhancement of humans liking for flavor by pairing with sweetness. Learn Motiv 1983;14:338–50. [10] Lavin MJ. Establishment of flavor–flavor associations using a sensory preconditioning training procedure. Learn Motiv 1976;7:173–83. [11] Yeomans MR, Weinberg L, James S. Effects of palatability and learned satiety on energy density influences on breakfast intake in humans. Physiol Behav 2005;86:487–99. [12] Wardle J, Johnson F. Weight and dieting: examining levels of weight concern in British adults. Int J Obes 2002;26:1144–9. [13] Ackard DM, Croll JK, Kearney-Cooke A. Dieting frequency among college females: association with disordered eating, body image, and related psychological problems. J Psychosom Res 2002;52:129–36. [14] Van Loan MD, Keim NL. Influence of cognitive eating restraint on totalbody measurements of bone mineral density and bone mineral content in premenopausal women aged 18–45 y: a cross-sectional study. Am J Clin Nutr 2000;72:837–43. [15] McLean JA, Barr SI, Prior JC. Cognitive dietary restraint is associated with higher urinary cortisol excretion in healthy premenopausal women. Am J Clin Nutr 2001;73:7–12. [16] Brunstrom JM, Davison CJ, Mitchell GL. Dietary restraint and cognitive performance in children. Appetite 2005;45:235–41. [17] Hsu LKG. Can dieting cause an eating disorder? Psychol Med 1997;27:509–13. [18] Rosen JC, Gross J, Vara L. Psychological adjustment of adolescents attempting to lose or gain weight. J Consult Clin Psychol 1987;55:742–7. [19] Steinhardt MA, Bezner JR, Adams TB. Outcomes of a traditional weight control program and a nondiet alternative: a one-year comparison. J Psychol 1999;133:495–513. [20] Brunstrom JM, Higgs S, Mitchell GL. Dietary restraint and us devaluation predict evaluative learning. Physiol Behav 2005;85:524–35. [21] Brunstrom JM, Downes CR, Higgs S. Effects of dietary restraint on flavour–flavour learning. Appetite 2001;37:197–206. [22] Birch LL, McPhee L, Steinberg L, Sullivan S. Conditioned flavor preferences in young children. Physiol Behav 1990;47:501–5. [23] Johnson SL, McPhee L, Birch LL. Conditioned preferences—young children prefer flavors associated with high dietary fat. Physiol Behav 1991;50:1245–51. [24] Appleton KM, Gentry RC, Shepherd R. Evidence of a role for conditioning in the development of liking for flavours in humans in everyday life. Physiol Behav 2006;87:478–86. [25] Birch LL, Deysher M. Conditioned and unconditioned caloric compensation—evidence for self-regulation of food-intake in young children. Learn Motiv 1985;16:341–55. [26] Birch LL, McPhee L, Shoba BC, Steinberg L, Krehbiel R. Clean up your plate—effects of child feeding practices on the conditioning of meal size. Learn Motiv 1987;18:301–17. [27] Zandstra EH, Stubenitsky K, De Graaf C, Mela DJ. Effects of learned flavour cues on short-term regulation of food intake in a realistic setting. Physiol Behav 2002;75:83–90. [28] Specter SE, Bellisle F, Hemery-Veron S, Fiquet P, Bornet FRJ, Slama G. Reducing ice cream energy density does not condition decreased acceptance or engender compensation following repeated exposure. Eur J Clin Nutr 1998;52:703–10. [29] Van Wymelbeke V, Beridot-Therond ME, de la Gueronniere V, Fantino M. Influence of repeated consumption of beverages containing sucrose or intense sweeteners on food intake. Eur J Clin Nutr 2004;58:154–61.
J.M. Brunstrom, G.L. Mitchell / Physiology & Behavior 90 (2007) 133–141 [30] Boakes RA, Rossiarnaud C, Garciahoz V. Early experience and reinforcer quality in delayed flavor–food learning in the rat. Appetite 1987;9:191–206. [31] Van Strien T, Frijters JER, Vanstaveren WA, Defares PB, Deurenberg P. The predictive-validity of the Dutch restrained eating scale. Int J Eat Disord 1986;5:747–55. [32] Gray RW, French SJ, Robinson TM, Yeomans MR. Dissociation of the effects of preload volume and energy content on subjective appetite and food intake. Physiol Behav 2002;76:57–64. [33] Gray RW, French SJ, Robinson TM, Yeomans MR. Increasing preload volume with water, reduces rated appetite but not food intake in healthy men even with minimum delay between preload and test meal. Nutr Neurosci 2003;6:29–37. [34] MacCallum RC, Zhang S, Preacher KJ, Rucker DD. On the practice of dichotomization of quantitative variables. Psychol Methods 2002;7:19–40. [35] Brunstrom JM, Mitchell GL, Baguley TS. Potential early-life predictors of dietary behaviour in adulthood: a retrospective study. Int J Obes 2005;29:463–74. [36] Gibson EL, Booth DA. Food-conditioned odour rejection in the late stages of the meal, mediating learnt control of meal volume by aftereffects of food consumption. Appetite 2000;34:295–303. [37] Yeomans MR. Palatability and the micro-structure of feeding in humans: the appetizer effect. Appetite 1996;27:119–33. [38] Yeomans MR, Gray RW, Mitchell CJ, True S. Independent effects of palatability and within-meal pauses on intake and appetite ratings in human volunteers. Appetite 1997;29:61–76. [39] LeGoff DB, Leichner P, Spigelman MN. Salivary response to olfactory food stimuli in anorexics and bulimics. Appetite 1988;11:15–25. [40] Rolls BJ, Kimharris S, Fischman MW, Foltin RW, Moran TH, Stoner SA. Satiety after preloads with different amounts of fat and carbohydrate— implications for obesity. Am J Clin Nutr 1994;60:476–87. [41] Brunstrom JM. Does dietary learning occur outside awareness? Conscious Cogn 2004;13:453–70. [42] Birch LL, Fisher JO, Davison KK. Learning to overeat: maternal use of restrictive feeding practices promotes girls' eating in the absence of hunger. Am J Clin Nutr 2003;78:215–20.
141
[43] Schachter S. Obesity and eating—internal and external cues differentially affect eating behavior of obese and normal subjects. Science 1968;161:751. [44] Schachter S. Some extraordinary facts about obese humans and rats. Am Psychol 1971;26:129–44. [45] Schachter S, Rodin J. Obese humans and rats. Washington DC: Erlbaum/ Halsted; 1974. [46] Nisbett RE. Hunger, obesity, and ventromedial hypothalamus. Psychol Rev 1972;79:433–53. [47] Herman CP, Mack D. Restrained and unrestrained eating. J Pers 1975;43:647–60. [48] Ruderman AJ. Dietary restraint—a theoretical and empirical review. Psychol Bull 1986;99:247–62. [49] Rogers PJ, Richardson NJ, Elliman NA. Overnight caffeine abstinence and negative reinforcement of preference for caffeine-containing drinks. Psychopharmacology (Berl) 1995;120:457–62. [50] Yeomans MR, Jackson A, Lee MD, Steer B, Tinley E, Durlach P, et al. Acquisition and extinction of flavour preferences conditioned by caffeine in humans. Appetite 2000;35:131–41. [51] Yeomans MR, Lartamo S, Procter EL, Lee MD, Gray RW. The actual, but not labelled, fat content of a soup preload alters short-term appetite in healthy men. Physiol Behav 2001;73:533–40. [52] Hogarth L, Duka T. Human nicotine conditioning requires explicit contingency knowledge: is addictive behaviour cognitively mediated? Psychopharmacology (Berl) 2006;184:553–66. [53] Lovibond PF, Shanks DR. The role of awareness in pavlovian conditioning: empirical evidence and theoretical implications. J Exp Psychol Anim Behav Processes 2002;28:3–26. [54] Field AP. I like it, but i'm not sure why: can evaluative conditioning occur without conscious awareness? Conscious Cogn 2000;9:13–36. [55] Nederkoorn C, van Eijs Y, Jansen A. Restrained eaters act on impulse. Pers Individ Differ 2004;37:1651–8. [56] Jansen A, Klaver J, Merckelbach H, Vandenhout M. Restrained eaters are rapidly habituating sensation seekers. Behav Res Ther 1989;27:247–52.