The measurement of dietary restraint, disinhibition and hunger: an ...

International Journal of Obesity (2001) 25, 900±906 ß 2001 Nature Publishing Group All rights reserved 0307±0565/01 $15.00 www.nature.com/ijo

PAPER The measurement of dietary restraint, disinhibition and hunger: an examination of the factor structure of the Three Factor Eating Questionnaire (TFEQ) MJ Bond1*, AJ McDowell1 and JY Wilkinson2 1 Behavioural Sciences and Biostatistics, Department of Psychiatry, School of Medicine, Flinders University, Adelaide, South Australia, Australia; and 2Applied Sciences, School of Nursing, Flinders University, Adelaide, South Australia, Australia

OBJECTIVE: To conduct separate factor analyses of the Three Factor Eating Questionnaire (TFEQ-R, TFEQ-D and TFEQ-H) scales and provide initial evidence of the construct validity of the obtained solutions. DESIGN: A cross-sectional survey with a 12 month retest of a subsample of subjects. SUBJECTS: A total of 553 undergraduate university women with a mean age of 25.0 y. The retest sample comprised 64 subjects with a mean age at retest of 25.7 y. MEASUREMENTS: In addition to the TFEQ, age, body mass index (BMI), satisfaction with current weight, nutrition knowledge and current exercise level were recorded. RESULTS: Three restraint (strategic dieting behaviour, attitude to self-regulation, avoidance of fattening foods), three disinhibition (habitual susceptibility, emotional susceptibility, situational susceptibility) and two hunger constructs (internal locus for hunger, external locus for hunger) were identi®ed. Initial evidence of the validity of these constructs was provided. CONCLUSION: The explanation of disordered eating behaviour is likely to be re®ned more by speci®c constructs, such as those presented, rather than by the more general constructs measured by the original TFEQ-R, TFEQ-D and TFEQ-H scales. Further examination of the factor structures presented is therefore encouraged. International Journal of Obesity (2001) 25, 900 ± 906 Keywords: TFEQ; restraint; disinhibition; hunger; factor analysis; construct validity

Introduction

Eating behaviour is generally accepted to be the outcome of internalised multidimensional constructs that include behavioural, cognitive and affective components.1 Restraint theory provides one attempt to synthesise these components. Unusual eating patterns are proposed to develop as a result of the stress associated with chronic self-control,2 with food intake determined by a balance between the desire to eat and the aspiration to diet.3 That is, cognitive processes override physiological hunger and satiety cues. While

*Correspondence: MJ Bond, Behavioural Sciences and Biostatistics, Department of Psychiatry, School of Medicine, Flinders University, GPO Box 2100, Adelaide, SA 5001, Australia. E-mail: malcolm.bond@¯inders.edu.au Received 5 July 2000; revised 11 September 2000; accepted 5 January 2001

restrained eaters may diet successfully for a time, certain disinhibiting events (eg consumption of forbidden foods, alcohol, negative emotional states) tend to interfere with self-control and result in overeating.4 Comparisons of alternative operational de®nitions of restraint theory5 ± 7 reveal considerable differences in interpretations of restraint, leading to an inevitable divergence in the philosophy and stated purpose of instruments derived to measure restrained eating.1,8,9 Therefore it is not surprising that the original Restraint Scale (RS),5 and the restraint scales included in the Dutch Eating Behaviour Questionnaire (DEBQ-R)6 and the Three Factor Eating Questionnaire (TFEQ-R)7 have all received critical evaluations.10 For example, the validity of the RS has been questioned due to con¯icting results being reported for the weight ¯uctuation and concern for dieting subscales.7,11 ± 15 The irrelevance of some items for different target groups (ie general vs clinical samples) has also been raised as a major criticism.16,17 With

Factor structure of the TFEQ MJ Bond et al

regard to the DEBQ-R, debate has arisen regarding whether it is unidimensional,18 or whether it in fact measures two aspects of restrained eating: intention to diet and actual dieting behaviour.19 By far the most attention, however, has been given to the TFEQ-R. In particular, there remains doubt about the number and nature of the dimensions embedded in the TFEQ-R. While early analyses determined the TFEQ-R to be unidimensional,20,21 later studies suggested the need for both a ¯exible control subscale that measures an adaptive strategy not resulting in an eating disorder, and a rigid control subscale that measures a maladaptive dieting strategy likely to be associated with recurrent restraint and disinhibition episodes.22 ± 24 An alternative two-factor model (cognitive restraint and behavioural restraint) was offered by Allison et al,19 although this was later rejected on statistical grounds as it was determined that cognitive and behavioural restraint were more probably two poles of a single dimension.10 Most recently, Ricciardelli and Williams25 have presented a three-factor model of restraint consisting of emotional= cognitive concern for dieting, behavioural dieting control and calorie knowledge. These three factors all correlated positively with current and past dieting behaviour, although only emotional=cognitive concern was associated with disinhibition. Further, behavioural dieting control was negatively related to body mass index (BMI). It was also noted that emotional=cognitive concern and behavioural dieting control were conceptually similar to constructs identi®ed in the restrained drinking literature, suggesting an underlying communality in restrained behaviour. However, only weak positive associations were found between emotional= cognitive concern for dieting and cognitive emotional preoccupation with drinking and cognitive behavioural control of drinking, respectively. The TFEQ-R factor structure reported by Ricciardelli and Williams was derived from a sample obtained from a general population of undergraduate university students. In their evaluation of the various alternate factorial models that have been proposed, Ricciardelli and Williams question whether samples of differing composition should give rise to identical factor structures.25 That is, body image as a hypothetical construct cannot be assumed to be uniform across different subgroups.26 For example, there can be no a priori expectation that measures developed for a clinical application will be automatically appropriate to a non-clinical, epidemiological application.27 In acknowledgment of these comments, the ®rst objective of the current study was to factor analyse the TFEQ-R. As the available sample was similar to that of Ricciardelli and Williams in terms of culture, age, gender and education, the speci®c aim was to determine whether their three-factor model could be replicated. Preliminary data regarding the construct validity of the factor solution obtained in the current study is also presented. It was hypothesised that TFEQ-R dimensions would be differentially associated with cognitive (nutrition knowledge), affective (satisfaction with current weight) and behavioural

(current exercise level) variables. Contextual variables (age and BMI) were also examined. Given the considerable interest in the constructs embedded in the TFEQ-R, it is surprising that little attention has been given to the other TFEQ scales. This is despite the observation that the construct validity, and the predictive validity in particular, of all TFEQ scales (restraint, disinhibition and hunger) has yet to be unequivocally demonstrated.28 There is certainly no evidence on which to base the conclusion that disinhibition (TFEQ-D) and hunger (TFEQ-H) are unidimensional, and yet potential multiple constructs may assist in explicating the manifestation of disinhibition and=or hunger. The second objective of the current study was therefore to examine the factorial structure of the TFEQ-D and TFEQ-H scales. Further, in the event that either of these scales should suggest multiple dimensions, the construct validity of these dimensions will be examined using the variables introduced to test the construct validity of the TFEQ-R factor solution.

901

Methods

Subjects and procedure Data for 553 undergraduate university women studying either nursing or physical education at Flinders University (Adelaide, Australia) were obtained from a standing database, to which new students are added each academic year. Data were collected as part of a laboratory practical on anthropometric measurement. The data to be reported were collected over 2 y and, although participation was voluntary, the overall refusal rate was low ( < 5%). The mean age of subjects was 25.0 y (s.d. 7.1 y). The modal age, however, was 19 y. A subset of nursing students from the ®rst year of data collection (n 64) were retested in the second year, at which time their mean age was 25.7 y (s.d. 7.0 y, mode 20 y).

Measures Three Factor Eating Questionnaire (TFEQ). The TFEQ,7 also known as the Eating Inventory,29 incorporates measures of restraint (21 items), disinhibition (16 items), and hunger (14 items). Responses are scored 0 or 1 and summed. Higher scores denote higher levels of restrained eating, disinhibited eating and predisposition to hunger, respectively. Nutrition knowledge. A series of questions designed by Burns, Richman and Caterson30 were used to test subjects' understanding of basic nutrition concepts. Responses are scored as correct or incorrect, summed and reported as the percentage of correct responses. Other data. Subjects recorded their age, height, current weight, whether they were satis®ed with their current weight (`yes'=`no'), and their current exercise level (®ve-point scale: 1 `none' to 5 `frequent vigorous exercise'). International Journal of Obesity


902

Statistical analysis Data analysis was conducted using SPSS for Windows (Release 6.0).31 Items from the restraint, disinhibition and hunger scales were analysed separately as follows. First, a principle component analysis (PC) was conducted in order to replicate Ricciardelli and Williams.25 Unfortunately, factor solutions obtained using this procedure either did not converge or did not produce interpretable factors. Therefore, principal axis factoring was conducted to screen for redundant items. Initial communality estimates (squared multiple correlations) were inspected and items with a communality less than 0.10 were omitted under the assumption that they would not load on any factors that might be derived.32 The remaining items were then analysed using PC followed by varimax rotation. The number of factors to retain for rotation was determined using parallel analysis criteria.33 This strategy takes account of both the sample size and the number of items being analysed. Again, while this strategy diverges from that used by Ricciardelli and Williams,25 use of the eigenvalues greater than one rule34 with the current sample produced more factors than could be realistically interpreted. Finally, items attaining a loading of 0.45 or better on any retained factor were examined to determine their suitability for scale construction (ie, internal reliability) using alpha (a).35

Results

Summaries of the obtained factors and the subscales derived from them, are shown in Table 1. Item numbers in the following descriptions correspond to the original version of the TFEQ.7

Restraint PAF identi®ed item 4 as redundant (communality < 0.10). The PC analysis of the remaining 20 items had a Kaiser ± Meyer ± Olkin measure of sampling adequacy of 0.86, which is very good ( > 0.80).36 Three factors were retained, accounting for 39.0% of the total variance. The subscales derived Table 1 Summary of scales derived from analyses of TFEQ items Eigenvalue

Percentage variance

Restraint R1: strategic dieting behaviour R2: attitude to self-regulation R3: avoidance of fattening foods

4.97 1.56 1.27

24.8 7.8 6.4

4 5 4

0.67 0.64 0.63

Disinhibition D1: habitual susceptibility D2: emotional susceptibility D3: situational susceptibility

3.84 1.39 1.06

29.6 10.7 8.2

5 3 5

0.67 0.78 0.60

Hunger H1: internal locus for hunger H2: external locus for hunger

3.76 1.17

28.9 9.0

6 6

0.73 0.63

Scale

International Journal of Obesity

Items > 0.45 Alpha

from these factors were de®ned as follows. In all cases higher scores indicate higher levels of restraint. Strategic dieting behaviour comprised items 6, 23, 28 and 48, which are concerned with speci®c behaviours that might be used to control weight (eg `deliberately take small helpings', `consciously holding back at meals'). Attitude to self-regulation (of eating) comprised items 10, 21, 30, 32 and 37, which assess subjects' overarching perspective on eating and weight control (eg `life is too short to worry about dieting', Ì eat anything I want, anytime I want'). Avoidance of fattening foods consisted of items 33, 42, 43 and 44, which are predominantly behavioural (eg, `How frequently do you avoid stocking up on tempting foods?', `How likely are you to shop for low calorie foods?').

Disinhibition Items 1, 25 and 31 were omitted due to low communalities. The remaining items had a sampling adequacy of 0.85. Three factors, accounting for 48.5% of the variance, were retained. For all subscales, higher scores indicate higher levels of disinhibition. The ®rst disinhibition subscale was termed habitual susceptibility (to disinhibition). Including items 11, 36, 45, 49 and 51, it appears to be the core component of disinhibition. The items describe circumstances that may predispose to recurrent disinhibition (eg Ì start dieting in the morning, but because of any number of things that happen during the day, by evening I have given up and eat what I want, promising myself to start dieting again tomorrow'). Emotional susceptibility (to disinhibition) consisted only of items 9, 20 and 27. They describe circumstances in which disinhibition is associated with negative affective states (ie Ì feel anxious', Ì feel blue', Ì feel lonely'). Situational susceptibility (to disinhibition) comprised items 2, 7, 13, 15 and 16. These items refer to disinhibition that is initiated by speci®c environmental cues (eg àt social occasions', `with someone else who is overeating').

Hunger Item 17 had a communality less than 0.10. A sampling adequacy of 0.84 was noted for the remaining 13 items, which split into two factors accounting for 37.9% of the variance. For both subscales, higher scores indicate higher levels of hunger. Internal locus for hunger included items 3, 5, 12, 24, 34 and 39. They describe hunger that is interpreted and regulated internally (eg Ì am usually so hungry that I eat more than three times a day', Ì often feel so hungry that I just have to eat something'). External locus for hunger contained items 8, 19, 22, 26, 41 and 47 that describe hunger triggered by external cues (eg `being with someone who is eating often makes me hungry enough to eat also', `when I see a real delicacy, I often get so hungry that I have to eat it right away').


Correlations among TFEQ scales Correlations among the original TFEQ scales and the subscales derived in the current study are presented in Table 2. Moderate positive correlations were noted among the restraint, disinhibition and hunger subscales, respectively. Relatively high correlations were noted between the subscales and their respective parent scale. Only habitual susceptibility (to disinhibition) demonstrated a consistent association with restraint scores. Internal, but not external, locus for hunger was negatively correlated with all restraint scores. All hunger scores were positively correlated with all disinhibition scores.

Age. The relationship between age and satisfaction with current weight approached signi®cance (t(551) 1.91, P 0.057 two-tailed). Women dissatis®ed with their weight were marginally more likely to be older (M 25.5, s.d. 7.3) than those who were satis®ed (M 24.4, s.d. 6.7). Positive correlations were found between age and TFEQ-R, strategic dieting behaviour and attitude to self-regulation for the women in the satis®ed subsample. These results were not re¯ected in the dissatis®ed subsample. Hunger scores were almost universally associated with age regardless of satisfaction with current weight. The total TFEQ-D score was negatively correlated with age for the satis®ed subsample. Of most interest was the negative correlation between emotional susceptibility (to disinhibition) and age for the satis®ed subsample that was coupled with a positive correlation for the dissatis®ed subsample.

Relationships between the TFEQ and other variables Over half of the sample (n 313, 56.6%) reported being dissatis®ed with their weight. A comparison of this group with women satis®ed with their weight produced a consistent pattern of results: those not satis®ed reported higher TFEQ scores. Only one subscale (internal locus for hunger) failed to achieve signi®cance (t(536) 1.76, P 0.079, twotailed). All further analyses (see Table 3) were therefore conducted independently for subjects satis®ed with their weight and those not satis®ed to explore the possibility that different relationships might be evident for the two subsamples. Table 2

903

Body mass index. The mean body mass index (BMI) score for the total sample was 23.1 (s.d. 3.9), which is within the normal range (20 ± 25).37 Those satis®ed with their current weight had a mean score of 21.0 (s.d. 2.4; 5.5% overweight or obese), compared with 24.6 (s.d. 4.1; 40.6% overweight or obese) for those dissatis®ed with their current weight (t(545) 12.00, P < .001). Signi®cant positive correlations were noted for the dissatis®ed subsample between BMI and

Intercorrelations among TFEQ scales

Scale

TFEQ-R

R1

TFEQ-R: restraint R1: strategic dieting behaviour R2: attitude to self-regulation R3: avoidance of fattening foods

0.69*** 0.77*** 0.73***

TFEQ-D: disinhibition D1: habitual susceptibility D2: emotional susceptibility D3: situational susceptibility

0.14*** 0.22*** Ð Ð

TFEQ-H: hunger H1: internal locus for hunger H2: external locus for hunger

R2

0.37*** 0.32*** Ð Ð Ð 7 0.09*

7 0.12** 7 0.17*** Ð

7 0.12** 7 0.13** 7 0.09*

R3

TFEQ-D

D1

D2

0.80*** 0.69*** 0.81***

0.40*** 0.50***

0.37***

0.55*** 0.41*** 0.57***

0.37*** 0.25*** 0.41***

0.38*** 0.29*** 0.36***

D3

TFEQ-H

H1

0.57*** 0.41*** 0.59***

0.87*** 0.84***

0.53***

0.46*** 0.13** 0.20*** Ð Ð 7 0.12** 7 0.19*** Ð

0.11* 0.15*** Ð Ð Ð 7 0.10* Ð

*P < 0.05; ** P < 0.01; *** P < 0.001; Ð not signi®cant.

Table 3

Relationships between TFEQ scales and satisfaction with current weight, age, BMI, current exercise level and nutrition knowledge

Variable=sample

TFEQ-R

R1

R2

R3

Satis®ed with current weight?

6.55***

2.38*

7.93***

3.83***

Age=satis®ed Age=dissatis®ed

0.24*** Ð

0.21*** Ð

0.19** Ð

Ð Ð

BMI=satis®ed BMI=dissatis®ed

0.18** Ð

0.15* Ð

Ð Ð

Exercise level=satis®ed Exercise level=dissatis®ed

Ð 0.27***

Ð 0.20***

Nutrition knowledge=satis®ed Nutrition knowledge=dissatis®ed

0.27*** Ð

0.13* Ð

0.19** 7 0.14**

Ð 0.19*** Ð Ð

0.13* 0.11* 0.29*** Ð

TFEQ-D 9.44*** 7 0.17** Ð Ð 0.31*** Ð 7 0.11* Ð Ð

D1 10.20*** Ð Ð Ð 0.26*** Ð Ð Ð Ð

D2 6.23*** 7 0.15* 0.15** Ð 0.20*** Ð 7 0.13* Ð Ð

D3 4.44***

TFEQ-H 3.26***

H1

H2

Ð

4.06***

Ð Ð

7 0.28*** 7 0.11*

7 0.26*** Ð

7 0.22*** 7 0.14**

Ð 0.20***

Ð Ð

Ð Ð

Ð Ð

Ð Ð

Ð Ð

Ð Ð

Ð Ð

Ð Ð

7 0.16* Ð

7 0.14* Ð

7 0.16* Ð

*P < 0.05; ** P < 0.01; *** P < 0.001; Ð not signi®cant.



904

all disinhibition scores. TFEQ-R, strategic dieting behaviour and attitude to self-regulation were positively correlated with BMI for the satis®ed subsample. However, BMI and strategic dieting behaviour were negatively correlated for the dissatis®ed subsample. Current exercise level. The relationship between current exercise level and satisfaction with current weight approached signi®cance (t(549) 1.84, P 0.067), with those dissatis®ed with their weight (M 2.7, s.d. 1.0) marginally less likely to exercise than those who were satis®ed (M 2.9, s.d. 1.0). A single signi®cant correlation was identi®ed for the satis®ed subsample: those exercising more reported higher scores for avoidance of fattening foods. In contrast, all restraint scores were positively correlated with exercise level for the dissatis®ed subsample. The dissatis®ed group also recorded negative correlations for disinhibition and emotional susceptibility (to disinhibition). Nutrition knowledge. The mean nutrition knowledge score for the women satis®ed with their weight was 59.7% (s.d. 8.3%); those dissatis®ed with their weight scored 60.2% (s.d. 8.4%). These means were not signi®cantly different (t(529) 0.68). Despite this, signi®cant correlations between nutrition knowledge and TFEQ scores were only noted for the satis®ed subsample. There were positive associations with TFEQ-R, strategic dieting behaviour and avoidance of fattening foods, and negative associations with all hunger scores.

Analysis of test ± retest data Table 4 presents the results of the analyses conducted for the 64 subjects who completed the TFEQ for a second time after an interval of 12 months. Note that both correlations and repeated measures t-tests were conducted as it is possible to achieve a high correlation in the presence of signi®cantly different means. Therefore, reliability is indicated by a high and signi®cant correlation coupled with a non-signi®cant mean difference.38 Table 4 Results of test-retest reliability analysis Scale

r

t

TFEQ-R: restraint R1: strategic dieting behaviour R2: attitude to self-regulation R3: avoidance of fattening foods

0.81 0.85 0.77 0.53

0.20 0.18 3.35** 1.11

TFEQ-D: disinhibition D1: habitual susceptibility D2: emotional susceptibility D3: situational susceptibility

0.86 0.80 0.70 0.79

2.00 2.41* 1.98 0.87

TFEQ-H: hunger H1: Internal locus for hunger H2: external locus for hunger

0.75 0.75 0.76

1.36 0.74 2.40*

All correlations signi®cant at P < 0.001; ** P < 0.01; * P < 0.05.


All correlations were high enough ( 0.70)38 for the measured behaviours to be considered stable traits with the exception of avoidance of fattening foods. However, three signi®cant mean differences were noted. Attitude to selfregulation was signi®cantly lower at retest, while habitual susceptibility (to disinhibition) and external locus for hunger were signi®cantly higher. These results may suggest that at least some aspects of eating behaviour are not stable traits. For example, behaviours may change with ¯uctuations in BMI. To test this hypothesis, changes in BMI from test to retest were correlated with test ± retest change scores for all TFEQ scales. A signi®cant result was noted only for external locus for hunger (r 0.29, P < 0.05), indicating that decreases in BMI were associated with decreases in external locus for hunger.

Discussion

The ®rst objective was to attempt to replicate the TFEQ-R factors recently reported by Ricciardelli and Williams.25 While the samples used in the two studies shared many similarities, the statistical techniques chosen for the current study were slightly different. We were forced to omit items that were determined statistically not to be central to the restraint construct for our subjects. We also had to use a relatively conservative strategy for the retention of nonrandom factors33 which no doubt reduced the overall percentage of variance accounted for in our analysis (only 39%). Further, relatively few items achieved a factor loading appropriate for their consideration in the derived subscales. Nevertheless, a three-factor solution was obtained that was recognisably similar to that of Ricciardelli and Williams.25 The greatest equivalence was between strategic dieting behaviour and the behavioural dieting control factor of Ricciardelli and Williams. These factors, measuring a general behavioural construct, are virtually identical. The addition of the term `strategic', however, was chosen to re¯ect the purposeful choices that restrained eaters make about food intake. The items comprising avoidance of fattening foods are a subset of calorie knowledge. However, this subset measures a behavioural construct that targets a speci®c method of weight control, rather than being indicative of the more cognitive construct assessed by the complete set of calorie knowledge items included in the Ricciardelli and Williams factor. Finally, there was little similarity between attitude to self-regulation (of eating) and emotional=cognitive concern for dieting in terms of shared items. However, in both studies the construct measured by these factors represents the core cognitive component of the restraint scale. The naming of our factor acknowledges the importance of self-regulation, de®ned as conscious, internally directed behaviour for the promotion of health and homeostasis.39 There was also some similarity between the proposed three-factor model and previously suggested two-factor models. For example, it is clear that there are both cognitive and behavioural issues inherent in restraint, as suggested by


Allison et al.19 Our analysis revealed both a general and a speci®c behavioural dimension, along with a key cognitive dimension. Further, there is some overlap between our factors and the ¯exible and rigid control concepts proposed by Westenhoefer.22 Strategic dieting behaviour shares items with the ¯exible control subscale, while attitude to self-regulation and avoidance of fattening foods share elements with the rigid control subscale. The pattern of correlations obtained between our restraint factors and TFEQ-D also support this observation. Our second objective was to examine factor solutions for the TFEQ-D and TFEQ-H scales. The disinhibition items split logically into three factors, each of which described a set of situations under which people may be susceptible to disinhibited eating. Indeed, we chose to label these factors using the term susceptibility after the comment of Westenhoefer et al23 that the TFEQ-D itself would be better named `susceptibility to eating problems'. The core component of disinhibition appears to be habitual susceptibility, implying as it does a cyclical behaviour pattern involving restraint and disinhibition. The TFEQ-H was found to include both internal and external cues for hunger, which we chose to label after Rotter's locus of control theory that emphasises the importance of perceived cues for reinforcement.40 A useful distinction between internal and external reinforcement in studies of weight control has also been described by other authors.41 One test of whether the constructs proposed by our analyses usefully expand the understanding of restraint, disinhibition and hunger is the degree to which they provide discriminating results, both among themselves and with other measures. Correlations between TFEQ scores demonstrated little discriminability among disinhibition and hunger scores, although only internal locus for hunger was consistently associated with restraint. Further, there were few signi®cant associations between disinhibition and restraint, implying that only certain aspects of disinhibition suggest disordered eating. Relationships between TFEQ scores and the measures included to determine construct validity demonstrated little differentiation until the sample was divided according to satisfaction with current weight, after which considerable variability emerged. In general, among subjects who were satis®ed with their weight, age and nutrition knowledge were signi®cantly associated with our proposed TFEQ constructs. In contrast, among the dissatis®ed sample, BMI and exercise level were more likely to demonstrate signi®cant associations with the proposed constructs. It must be acknowledged that the results we have presented, and the above interpretations, are based purely on correlational evidence and as such no comment regarding causation can be made. In addition, there were relatively few signi®cant relationships noted, and none of any great size (ie > 0.30). One explanation for these observations is the small range of values possible for the subscales we have proposed. Due to the low number of items loading on each factor, maximum subscale values ranged from only 3 for emotional susceptibility (to disinhibition) to 6 for both hunger subscales.

Such truncated ranges are known to produce lower bound estimates of linear relationships.42 An alternative explanation is that the relationships between the TFEQ subscales and age, BMI, exercise level and nutrition knowledge are nonlinear. This is a suggestion not made by previous investigations in this ®eld yet evidence from locus of control39 and self-regulation research,40 eg, would certainly lend support to such a proposition. Relatively larger coef®cients were obtained when TFEQ subscales were correlated among themselves (eg the full sample coef®cients presented in Table 2 and the test ± retest coef®cients in Table 4). Indeed, as previously noted, the test ± retest correlations were all impressively large, being suggestive of considerable stability over the 12-month interval. These results (in conjunction with the repeated measures t-tests) generally support the notion that the TFEQ subscales are stable traits rather than state-speci®c appraisals. However, it is acknowledged that further evidence is required before such a conclusion can be made. In conclusion, the factor structures presented for the TFEQ-R, TFEQ-D and TFEQ-H offer some promise of a more complex picture of eating behaviour than that provided by the original scales. In fact, the modest number of items loading on each factor, and the modest amount of variance accounted for in each factor analysis (particularly for restraint and hunger items), are suggestive of more, not fewer, factors. These observations further reinforce the need to clearly distinguish constructs relevant to nonclinical populations from those relevant to clinical samples. While our data are suggestive rather than conclusive, the further examination of factor structures such as ours is encouraged by the comment that general factors are no longer useful in forwarding knowledge in the area of body image and eating disorders.43 Rather, it has been argued that domain-speci®c factors, measuring constructs at a more precise level, are now more likely to assist in explaining these very complex issues.

905

References

1 Ruderman AJ. Dietary restraint: a theoretical and empirical review. Psychol Bull 1986; 99: 247 ± 262. 2 Herman CP, Polivy J. Restrained eating. In: Stunkard AJ (ed.). Obesity. WB Saunders: Philadelphia, PA; 1980. pp 208 ± 225. 3 Ruderman AJ. Dysphoric mood and overeating: a test of restraint theory's disinhibition hypothesis. J Abnorm Psychol 1985; 94: 78 ± 85. 4 Lowe MR. The effects of dieting on eating behavior: a three-factor model. Psychol Bull 1993; 114: 100 ± 121. 5 Herman CP, Mack D. Restrained and unrestrained eating. J Person 1975; 43: 647 ± 660. 6 van Strein T, Frijters JER, Bergers GPA, Defares PB. The Dutch Eating Behavior Questionnaire (DEBQ) for assessment of restrained, emotional, and external eating behavior. Int J Eat Disord 1986; 5: 295 ± 315. 7 Stunkard AJ, Messick S. The Three-Factor Eating Questionnaire to measure dietary restraint, disinhibition and hunger. J Psychosom Res 1985; 29: 71 ± 83.



906

8 Laessle RG, Tuschl RJ, Kotthaus BC, Prike KM. A comparison of the validity of three scales for the assessment of dietary restraint. J Abnorm Psychol 1989; 98: 504 ± 507. 9 Heatherton TF, Herman CP, Polivy J, King GA, McGree ST. The (mis)measurement of restraint: an analysis of conceptual and psychometric issues. J Abnorm Psychol 1988; 97: 19 ± 28. 10 Gorman BS, Allison DB. Measures of restrained eating. In: Allison DB (ed). Handbook of assessment methods for eating behaviors and weight-related problems. Sage: Thousand Oaks, CA; 1995. pp 149 ± 184. 11 Meyers AW, Stunkard AJ, Coll M. Food accessibility and food choice: a test of Schachter's externality hypothesis. Arch Gen Psychiat 1980; 37: 1133 ± 1135. 12 Rodin J. Current status of the internal ± external hypothesis for obesity: what went wrong? Am Psychol 1981; 36: 361 ± 372. 13 Spitzer L, Rodin J. Human eating behaviour: a critical review of studies in normal weight and overweight individuals. Appetite 1981; 2: 293 ± 329. 14 Johnson WG, Lake L, Mahan JM. Restrained eating: measuring an elusive construct. Addict Behav 1983; 8: 413 ± 418. 15 Blanchard FA, Frost RO. Two factors of restraint: concern for dieting and weight ¯uctuation. Behav Res Ther 1983; 21: 259 ± 267. 16 Wardle J. Dietary restraint and binge eating. Behav Anal Modif 1980; 4: 201 ± 209. 17 Wardle J. The assessment of restrained eating. Behav Res Ther 1986; 24: 213 ± 215. 18 Ogden J. The measurement of restraint: confounding success and failure? Int J Eat Disord 1993; 13: 69 ± 76. 19 Allison DB, Kalinsky LB, Gorman BS. A comparison of the psychometric properties of three measures of dietary restraint. Psychol Assess 1992; 4: 391 ± 398. 20 Ganley RM. Emotional eating and how it relates to dietary restraint, disinhibition, and perceived hunger. Int J Eat Disord 1988; 7: 635 ± 647. 21 Hyland ME, Irvine SH, Thacker C, Dann PL, Dennis I. Psychometric analysis of the Stunkard-Messick Eating Questionnaire (SMEQ) and comparison with the Dutch Eating Behavior Questionnaire (DEBQ). Curr Psychol Res Rev 1989; 8: 228 ± 233. 22 Westenhoefer J. Dietary restraint and disinhibition: is restraint a homogeneous construct? Appetite 1991; 16: 45 ± 55. 23 Westenhoefer J, Broeckmann P, Muench AK, Pudel V. Cognitive control of eating behaviour and the disinhibition effect. Appetite 1994; 23: 27 ± 41. 24 Shearin EN, Russ MJ, Hull JW, Clarkin JF, Smith GP. Construct validity of the Three-Factor Eating Questionnaire: ¯exible and rigid control subscales. Int J Eat Disord 1994; 16: 187 ± 198.


25 Ricciardelli LA, Williams RJ. A two-factor model of dietary restraint. J Clin Psychol 1997; 53: 123 ± 131. 26 Keeton WP, Cash TF, Brown TA. Body image or body images? Comparative, multidimensional assessment among college students. J Person Assess 1990; 54: 213 ± 230. 27 Ben-Tovim DI, Walker MK. A quantitative study of body-related attitudes in patients with anorexia and bulimia nervosa. Psychol Med 1992; 22: 961 ± 969. 28 Hetherington MM, Burnett L. Ageing and the pursuit of slimness: dietary restraint and weight satisfaction in elderly women. Br J Clin Psychol 1994; 33: 391 ± 400. 29 Stunkard AJ, Messick S. The eating inventory. Psychological Corporation: San Antonio, TX; 1988. 30 Burns CM, Richman R, Caterson ID. Nutrition knowledge in the obese and overweight. Int J Obes 1987; 11: 485 ± 492. 31 NorusÏis MJ. SPSS for Windows: base system user's guide, release 6.0. SPSS Inc.: Chicago, IL; 1993. 32 Gorsuch RL. Factor analysis. WB Saunders: Philadelphia, PA; 1974. 33 Lautenschlager GJ. A comparison of alternatives to conducting Monte Carlo analyses for determining parallel analysis criteria. Multivar Behav Res 1989; 24: 365 ± 395. 34 Kaiser HF. The application of electronic computers to factor analysis. Educ Psychol Meas 1960; 20: 141 ± 151. 35 Cronbach LJ. Coef®cient alpha and the internal structure of tests. Psychometrika 1951; 16: 297 ± 334. 36 Hair JF, Anderson RE, Tatham RL, Black WC. Multivariate data analysis, 4th edn. Prentice Hall: Englewood Cliffs, NJ; 1995. 37 Garrow JS, Webster J. Quetelet's Index (W=H2) as a measure of fatness. Int J Obes 1985; 9: 147 ± 153. 38 Streiner DL, Norman GR. Health measurement scales: a practical guide to their development and use, 2nd edn. Oxford University Press: Oxford; 1995. 39 Green JA, Shellenberger R. The dynamics of health and wellness: a biopsychosocial approach. Holt Rinehart and Winston: Fort Worth, TX; 1991. 40 Rotter JB. Some problems and misconceptions related to the construct of internal versus external control of reinforcement. J Consult Clin Psychol 1975; 43: 56 ± 67. 41 Tiggemann M, Rothblum ED. Gender differences in internal beliefs about weight and negative attitudes towards self and others. Psychol Women Q 1997; 21: 581 ± 593. 42 Cohen J, Cohen P. Applied multiple regression=correlation analysis for the behavioral sciences, 2nd edn. Lawrence Erlbaum: Hillsdale, NJ; 1983. 43 Ben-Tovim DI, Walker MK. Women's body attitudes: a review of measurement techniques. Int J Eat Disord 1991; 10: 155 ± 167.