- Email: [email protected]
Association of Body Mass with Dietary Restraint and Disinhibition
app p052
18-07-95 13:50:47
Appetite, 1995, 25, 31–41
Association of Body Mass with Dietary Restraint and Disinhibition
DONALD A. WILLIAMSON, OLGA J. LAWSON, ELLEN R. BROOKS, PATRICIA J. WOZNIAK, DONNA H. RYAN, GEORGE A. BRAY and ERICH G. DUCHMANN Louisiana State University, Pennington Biomedical Research Center
The relationship of disinhibition and dietary restraint with body mass was studied in a sample of 293 women. Results suggested that higher body mass was associated with an interaction of disinhibition and dietary restraint. The association of disinhibition with higher body mass was moderated by increased dietary restraint. Symptoms of an eating disorder were more strongly associated with disinhibition than with dietary restraint. These results suggest that dieting may moderate the increased body mass associated with overeating. Psychological and eating problems associated with dietary restraint were found to be of less significance than those associated with disinhibition. 1995 Academic Press Limited
A B M D R D The term dietary restraint has been used to describe a person’s intent to restrict dietary intake in order to control body weight (Herman & Mack, 1975). The construct of dietary restraint has been used to explain the common observation that obese persons intend to diet yet frequently overeat and thus maintain a higher body weight. A central hypothesis of dietary restraint theory is that the intent to diet may be disrupted or “disinhibited” by certain events, e.g. dysphoric emotions, alcohol or the availability of appetizing foods (Ruderman, 1986). This disinhibition hypothesis is important because it explains why persons trying to restrict eating frequently overeat. Numerous laboratory investigations have found support for the loss of control hypothesis (Lowe, 1993) and recently this hypothesis has been extended to account for the binge eating observed in bulimia nervosa and non-clinical subjects (Polivy & Herman, 1985; Polivy et al., 1994; Ruderman, 1986). One conclusion derived from this research is that dieting may have adverse psychological and behavioral effects upon certain individuals (Brownell, 1991; Brownell & Rodin, 1994). The adverse events associated with dieting range from binge eating (Polivy & Herman, 1985) to lowered total energy expenditure (Laessle et al., 1989). One consequence of these research findings has been increased concern about the benefits vs. costs of dieting to control obesity (Blundell, 1990; Brownell, 1991; Brownell & Rodin, 1994; French & Jeffery, 1994). Address correspondence to: Donald A. Williamson, Ph.D., Department of Psychology, Louisiana State University, Baton Rouge, LA 70803, U.S.A. 0195–6663/95/040031+11 $12.00/0
1995 Academic Press Limited
32
D. A. WILLIAMSON ET AL.
Other researchers have questioned the hypothesized relationship between dietary restraint and overeating (Cooper & Charnock, 1990; Treasure, 1990). In response to conflicting results in this research literature, Lowe (1993) observed that dietary restraint may be a multidimensional construct including a history of dieting, current dieting and degree of sustained weight loss. Also, Westenhoefer, Pudel and Maus (1990) reported the independence of dietary restraint and disinhibition in a large sample of European women. The investigation of Westenhoefer et al. (1990) is of special importance to the present investigation. Westenhoefer presented the results of two surveys of European women pertaining to body weight and caloric intake. Body mass index and caloric intake were analysed as a function of two scales of the Three Factor Eating Questionnaire (TFEQ; Stunkard & Messick, 1985). The two scales were Cognitive Restraint, which measures the intent to diet and actual dieting behavior (Allison, Kalinsky & Gorman, 1992), and Disinhibition, which measures overeating and binge eating in response to a variety of situations associated with loss of control of food intake (Stunkard & Messick, 1985). The results of this study showed that body mass index was predicted by an interaction of Cognitive Restraint and Disinhibition, such that low Restraint and high Disinhibition scores were associated with higher body mass. An interaction effect was also found for caloric intake, showing that high Restraint and low Disinhibition scores were associated with lower caloric intake and low Restraint and high Disinhibition scores were associated with higher caloric intake. One implication of these data is that Dietary Restraint and Disinhibition, as measured by the TFEQ, may have interactive influences upon obesity. Another implication is that dieting may serve to moderate the association of overeating with increased body mass. A second observation of Westenhoefer et al. (1990) was that dietary restraint was not necessarily associated with problematic eating behavior. In a follow-up study, Westenhoefer (1991), reported that the Cognitive Restraint scale of the TFEQ had two subscales, which he called Rigid and Flexible Control (of eating). Rigid Control was positively correlated with Disinhibition scores, whereas, flexible control was negatively correlated with Disinhibition scores. The present investigation was designed to replicate and extend the results of the earlier studies of Westenhoefer. The present study directly measured the body weight of subjects, as opposed to the self-reported survey data of the Westenhoefer et al. (1990) study, and used several measures of problematic eating habits and attitudes which have been validated for use with the eating disorders. A secondary purpose of the current study was to evaluate the association of Rigid and Flexible Control with Disinhibition and body mass. With this research design we sought to evaluate the association of Cognitive Restraint and Disinhibition with body mass and eating disorder symptoms.
M Subjects A total of 301 women were recruited from the community and university populations in the Baton Rouge region. These subjects were recruited via media advertisements for a cross-sectional study of eating behavior. They were not recruited
DIETARY RESTRAINT
33
as a part of a weight loss investigation or clinical weight loss program. From this sample of 301 women, 293 completed the study. Average age of these 293 women was 38.6 yrs. (SD=14·3). The age range was 17 to 78 years. Racial composition of the sample was 88% Caucasian, 12% African-American, and 1% other minority. The distribution of racial groups did not differ as a function of scores on the Dietary Restraint or Disinhibition scales of the TFEQ (p>0·30). Average height of the subjects was 1·6 m (SD=0·07) and average weight was 80·9 kg (SD=24·4). Weights ranged from 44 kg to 182 kg. It should be noted that the average weight of this sample was substantially higher than the average American woman. Assessment Measures Height and weight Height was measured in centimeters using a stadiometer. Weight was measured in kilograms using a digital scale calibrated to an accuracy of 0·10 kg. These measures were converted to body mass index (BMI), using the formula: BMI=weight (kg)/ height2 (m), which has been validated as an index of adiposity (Garrow, 1983). Three Factor Eating Questionnaire (TFEQ) The TFEQ was developed by Stunkard and Messick (1985) in response to psychometric problems that had been reported for other measures of dietary restraint. The three factors of the TFEQ were developed using factor analysis to derive a 51item questionnaire. These scales were labelled Cognitive Restraint, Disinhibition of Control, and Perceived Hunger. Later investigations have often referred to the first two scales as Dietary Restraint and Disinhibition. Given current knowledge about the validity of these two scales, we believe that Dietary Restraint is an appropriate label for the first factor. We believe that the Disinhibition scale is best viewed as a measure of overeating. These interpretations of the first two factors of the TFEQ are based upon recent findings that the first factor measures the intent to diet and dieting behavior (Allison et al., 1992; Laessle et al., 1989) and that the second factor measures episodic overeating that may be independent of dietary restraint (Westenhoefer et al., 1990). As noted by Heatherton et al. (1988), “without initial dietary inhibition (or restraint), it is difficult to imagine disinhibition” (p. 25). Examination of the items which form the second factor of the TFEQ show that overeating is the behavioral constant of this factor. For these reasons, we will refer to the three scales of the TFEQ as measuring: Dietary Restraint, Overeating, and Perceived Hunger. All three scales of the TFEQ have been found to have good test–retest reliability (Stunkard & Messick, 1985). Westenhoefer (1991) identified two subscales of the Dietary Restraint factor: (a) Rigid Control, which was positively correlated with Disinhibition scores; and (b) Flexible Control, which was negatively correlated with Disinhibition scores. Each subscale was composed of seven of the 21 original items of the Dietary Restraint scale. These two subscales have not been extensively studied and were included in several secondary analyses in this study. Eating Disorder Inventory (EDI) The EDI is a 64-item questionnaire developed by Garner, Olmstead and Polivy (1983). The questions of the EDI are scored using a six-point rating scale. The EDI is composed of eight scales: (1) Drive for Thinness (DFT), (2) Bulimia (B), (3) Body Dissatisfaction (BD), (4) Ineffectiveness (I), (5) Perfectionism, (6) Interpersonal
34
D. A. WILLIAMSON ET AL.
T 1 Intercorrelations of TFEQ Scales, Age and BMI Dietary Restraint Overeating Dietary Restraint Flexible Control Rigid Control Age
−0·07
Flexible Control −0·07 0·85∗∗
Rigid Control 0·04 0·85∗∗ 0·65∗∗
Age
BMI
0·26∗∗ 0·16∗ 0·04 0·16∗
0·38∗∗ −0·18∗ −0·23∗ −0·11 0·44∗∗
∗ p<0·01, ∗∗ p<0·001. TFEQ=Three Factor Eating Questionnaire, BMI=Body Mass Index.
Distrust (ID), (7) Interoceptive Awareness (IA) and (8) Maturity Fears (MF). The first three scales are considered to be primary symptoms of the eating disorders. Internal consistency of the scales has been reported to be high and test–retest reliability of the scales is satisfactory (Garner & Olmstead, 1984). Subjects diagnosed with anorexia and bulimia nervosa score higher than controls on the eight scales of the EDI (Garner & Olmstead, 1984; Gross et al., 1986). Eating Questionnaire-Revised (EQ-R) The EQ-R is a 15 item questionnaire developed by Williamson, Davis, Goreczny, McKenzie and Watkins (1989) as a measure of bulimic symptoms. It is a singlefactor test with a multiple-choice scoring format. It has been validated as a measure of bulimic symptoms and has good test–retest reliability and internal consistency. Procedure Subjects were scheduled for a single assessment session and all data were collected in that session. Consent forms were signed. Height and weight were measured and then the three questionnaires, TFEQ, EDI and EQ-R, were administered. R Statistical Methods Correlations among variables were evaluated using Pearson correlation. Multiple regression analysis was used to evaluate the relationship of BMI with Dietary Restraint, Overeating and Age. Multivariate analysis of variance (MANOVA) was used to evaluate the association of Dietary Restraint and Overeating with eating disorder symptoms. The associations among these variables were described using the canonical variates from MANOVA (Huberty & Morris, 1989). An alpha level of p<0·01 was used for interpretation of all statistical analyses. This relatively conservative alpha level was chosen because of the numerous correlations and regression analyses conducted in this study. Correlational Analysis Table I summarizes the intercorrelations among scales from the TFEQ, age and BMI. Overeating was uncorrelated with the original Dietary Restraint scale and
DIETARY RESTRAINT
35
with its subscales, Flexible and Rigid Control. Overeating was positively correlated with age and BMI. The subscales of Dietary Restraint, Flexible and Rigid Control, were strongly correlated with the original Dietary Restraint scale. Also, these subscales were positively and strongly correlated. Age was positively correlated with BMI, Dietary Restraint and Rigid Control. Dietary Restraint and Flexible Control were negatively correlated with BMI. Rigid Control was not significantly correlated with BMI. Restraint, Overeating and Body Mass The association of Dietary Restraint and Overeating with BMI was tested using a series of multiple regression analyses. Table 2 summarizes the six statistical models that were tested. All variables were treated as continuous variables. Model 1 represents a replication of the Westenhoefer study. In this regression model, a significant interaction of Dietary Restraint and Overeating with BMI was found. This interaction is shown in Fig. 1. For subjects with lower Overeating scores, level of Dietary Restraint was not significantly related to BMI. For subjects with average to high Overeating scores, low Dietary Restraint was associated with increased BMI, but as Restraint scores increased, BMI decreased. Since age was found to be positively correlated with BMI, Dietary Restraint and Overeating scores, age was added to the regression analysis as a concomitant variable in model 2. Age was found to be a significant predictor variable, and percentage of explained variance for BMI increased from 23% in model 1 to 37% in model 2. The primary effect of entering age as a predictor variable was a reduction in unexplained error. There was also a slight alteration of the regression slopes. Comparison of the two regression models shows that the basic relationships between Dietary Restraint, Overeating and BMI were not strongly affected by inclusion of age as a predictor variable. Rigid Control, Flexible Control, Overeating and Body Mass In regression models 3 and 4, Dietary Restraint was replaced by Rigid Control as a predictor variable. As can be seen in Table 2, these models were quite similar to models 1 and 2, which included Dietary Restraint as a predictor variable. Similarly, in models 5 and 6, Flexible Control replaced Rigid Control as a predictor variable. The results of models 5 and 6 were quite similar to corresponding earlier models (i.e. models 1 and 3, for model 5; and models 2 and 4, for model 6). In model 7, Rigid Control and Flexible Control were predictors of BMI. This model was statistically significant, but accounted for only 6% of the variance of BMI. Flexible Control was a significant predictor of BMI, but Rigid Control was not a significant predictor. Restraint, Overeating and Eating Disorder Symptoms The association of eating disorder symptoms with Dietary Restraint and Overeating was tested using MANOVA. The different measures of eating disorder symptoms were the dependent measures. Dietary Restraint, Overeating and Restraint×Overeating were treated as continuous predictors. The F statistic, based on Wilk’s Lambda, was used to test for statistical significance. The dependent
36
D. A. WILLIAMSON ET AL.
T 2 Summary of Multiple Regression Analyses with BMI as the Dependent Variable Model
Predictor variables
Parameter estimate t
p
R2
1 Restraint [F(3, 289)=28·40, Overeating p<0·0001] R×O Interaction Intercept
−0·43 0·80 −0·12 30·22
−4·02 6·29 −4·90
0·001 0·001 0·001
0·23
2 Restraint [F(4, 288)=42·09, Overeating p<0·0001] R×O Interaction Age Intercept
−0·56 0·54 −0·11 0·26 30·22
−5·72 4·58 −5·07 8·03
0·001 0·001 0·001 0·001
0·37
3 Rigid Control −0·78 [F(3, 289)=22·83, Overeating 0·88 p<0·0001] RC×O Interaction −0·22 Intercept 30·42
−2·94 6·85 −3·52
0·001 0·005 0·001
0·19
4 Rigid Control −1·04 [F(4, 288)=34·72, Overeating 0·65 p<0·0001] RC×O Interaction −0·20 Age 0·25 Intercept 30·40
−4·23 5·38 −3·43 7·56
0·001 0·001 0·001 0·001
0·33
5 Flexible Control [F(3, 289)=28·57, Overeating p<0·0001] FC×O Interaction Intercept
−1·39 0·79 −0·29 30·20
−4·74 6·28 −4·15
0·001 0·001 0·001
0·23
6 Flexible Control [F(4, 288)=39·72, Overeating p<0·0001] FC×O Interaction Age Intercept
−1·49 0·57 −0·27 0·24 30·20
−5·56 4·74 −4·17 7·53
0·001 0·001 0·001 0·001
0·36
7 [F(2, 290)=8·75, p<0·0002]
−1·54 0·35 30·32
−3·70 0·94
0·001 0·351
0·06
Flexible Control Rigid Control Intercept
F values describe the statistical model; t values reflect the significance of the variables within a model. Abbreviations: R=Dietary Restraint, O=Overeating, RC=Rigid Control, FC=Flexible Control.
variables were the Perceived Hunger scale, EQ-R and the eight scales of the EDI. Main effects for Dietary Restraint, F(10, 288)=15·26, p<0·0001, and Overeating, F (10, 288)=46·32, p<0·0001, were found. Also the interaction of Dietary Restraint and Overeating was statistically significant, F(10, 288)=4·03, p<0·0001). These effects were interpreted in terms of canonical structure, as suggested by Huberty and Morris (1989). The results of this analysis are summarized in Table 3. The canonical variate, Overeating, was positively correlated with Perceived Hunger, two measures of bulimic behavior (the EQ-R and EDI bulimia scale), and with three other EDI scales, Drive
37
DIETARY RESTRAINT
F 1. Relationship between BMI and Overeating at three values of Restraint (Μ, Mean; Χ, Mean +1 standard deviation; Ε, mean −1 standard deviation), unadjusted for Age.
T 3 Canonical correlations derived from MANOVA With Overeating
With Restraint
O×R interaction
Canonical correlation
0·79
0·59
0·35
Squared canonical correlation
0·62
0·35
0·12
0·68∗ 0·90∗ 0·56∗ 0·80∗ 0·55∗ 0·37 0·25 0·16 0·52∗ 0·13
−0·22 0·23 0·58∗ −0·33 0·01 −0·23 0·11 −0·06 0·00 −0·19
0·08 0·11 0·09 0·59∗ 0·60∗ 0·45 0·09 0·29 0·21 0·39
15·26
46·32
4·03
Total canonical structure
F value
Hunger EQ-R EDI DT EDI B EDI BD EDI I EDI P EDI ID EDI IA EDI MF
Canonical loadings greater than 0·50 were interpreted to define canonical variates as designated by ∗. The degrees of freedom were 10 and 288 for each F value. All three F values were statistically significant (p<0·001). Abbreviations: O=Overeating, R=Restraint, EQ-R=Eating QuestionnaireRevised, EDI=Eating Disorder Inventory, DT=Drive for Thinness, B=Bulimia, BD=Body Dissatisfaction, I=Ineffectiveness, P=Perfectionism, ID=Interpersonal Distrust, IA=Interoceptive Awareness, MF=Maturity Fears.
38
D. A. WILLIAMSON ET AL.
for Thinness, Body Dissatisfaction, and Interoceptive Awareness. The canonical variate, Dietary Restraint, had high loadings for increasing scores on the EDI Drive for Thinness scale. The canonical correlations for Overeating (0·79) and Dietary Restraint (0·59) were quite high. The canonical correlation for the interaction of Restraint and Overeating was somewhat smaller (0·35), but was statistically significant. The canonical variate associated with the interaction term was positively correlated with two EDI scales, Bulimia and Body Dissatisfaction. This interaction indicated that the symptoms of bulimia and body dissatisfaction were associated with a pattern of elevated scores on both the Dietary Restraint and Overeating scales.
D The results of this study supported the contention that the Dietary Restraint scale of the TFEQ was measuring the intent to diet and dieting behavior. The results also suggested that the scale previously labelled Disinhibition was measuring overeating. The Dietary Restraint scale was most strongly correlated with the Drive for Thinness scale of the EDI. Two recent factor-analytic studies (Gleaves, Williamson & Barker, 1993a; Williamson, Barker & Bertman, in press) have found the Drive for Thinness scale to load on a Dietary Restraint factor. Furthermore, as Dietary Restraint increased, body mass decreased, in persons reporting significant overeating. This pattern suggests that the Dietary Restraint scale is also measuring actual dieting behavior which results in weight suppression. These findings are consistent with the results of earlier studies of the Dietary Restraint scale. The Overeating scale was most positively correlated with measures of bulimic binge eating, i.e. EQ-R and EDI Bulimia scale. It was also positively correlated with the Perceived Hunger scale, of the TFEQ. For the purposes of further discussion, we will assume that Dietary Restraint and Overeating were the constructs measured by the TFEQ. We also found the Dietary Restraint and Overeating scales to be uncorrelated. Earlier studies which have investigated intercorrelations among the three scales of the TFEQ have reported some inconsistencies. In the original description of the TFEQ, Stunkard and Messick (1985) reported a moderate positive correlation (r= 0·43) between the Dietary Restraint and Overeating (Disinhibition) scales. Westenhoefer (1991) reported a moderate negative correlation (r=−0·37) between those two scales; our study found a non-significant relationship. These somewhat inconsistent findings are most likely to be due to differences in the samples recruited for each of the three studies. In a heterogeneous sample such as that of our study, the two scales were essentially independent. The independence of the Dietary Restraint and Overeating scales suggest that the causal link between restraint and overeating or binge eating may vary across individuals, with some individuals successfully dieting, some intermittently dieting and binge eating, and others who overeat or binge, but seldom attempt to restrain their eating (French & Jeffery, 1994; Lowe, 1993). The finding of an interactive association of Dietary Restraint and Overeating with body mass index is consistent with the survey results reported by Westenhoefer et al. (1990). This interaction, as illustrated in Fig. 1, suggests that increased dietary restraint moderates the association of overeating with obesity. Some of the subjects
DIETARY RESTRAINT
39
in this sample scored high on the Overeating scale, yet scored very low on a measure of dietary restraint. These unrestrained overeaters were very obese, which indirectly supports the conclusion that these women were overeating but were not actively dieting (Lowe, 1993). Women with low scores on the Overeating scale did not vary in body mass as a function of Dietary Restraint. This finding suggests that some women are restrained eaters, yet do not significantly overeat and maintain a weight level within the normal to moderately-overweight range. When age was added to this regression model, unexplained error was reduced, but the relationship of BMI with Dietary Restraint and Overeating was not substantially altered. Similarly, the replacement of Dietary Restraint with Rigid Control or Flexible Control did not substantially alter the relationships among the other variables. This finding suggests that both flexible and rigid approaches to dieting can moderate the association of BMI with Overeating. It should be noted that the Flexible and Rigid Control scales were highly correlated (r=0·65) in this study, and both scales were highly correlated (r=0·85) with the Dietary Restraint scale, from which they were derived (Westenhoefer, 1991). Also, in this study, the Rigid and Flexible Control scales were not significantly correlated with Overeating. In the development of these scales, Westenhoefer (1991) found the Rigid Control to be positively correlated with Overeating and the Flexible Control scale to be negatively correlated with Overeating. In this study, we found Flexible Control to be negatively correlated with BMI. Rigid Control was uncorrelated with BMI. The sample of the Westenhoeffer (1991) study was composed of persons seeking weight loss treatment. In the present study, lean and obese subjects were recruited and weight loss was not a purpose of the study. Given these conflicting results, we believe that further research on the association of Rigid and Flexible Control with Overeating and BMI is needed. The lack of association between high dietary restraint and overeating suggests that dietary restraint and overeating are independent behavioral phenomena when measured by the TFEQ. There are several possible explanations for this finding. One explanation is that overeating results in weight gain and that increasing dietary restraint is employed to attenuate this weight gain in some individuals. A second explanation is that subjects reporting frequent overeating and high dietary restraint are attempting to suppress their weight below that which is biologically normal, which causes episodic overeating in only these subjects. The finding of a significant canonical correlation of bulimic symptoms with the interaction of Dietary Restraint and Overeating is consistent with this explanation. A third explanation is that subjects scoring high on Dietary Restraint were using different criteria for judging whether they were “overeating”. Two recent studies (Gleaves, Williamson & Barker, 1993b; Williamson, Gleaves & Lawson, 1991) found that bulimics and compulsive binge eaters are biased in their judgements of overeating. It is possible that the restrained eaters in this study also inflated the perception of their problems with overeating, resulting in higher scores on the Overeating scale yet were not consuming the quantity of food being eaten by unrestrained overeaters. Further research will be required to determine which of these explanations is most valid. The results of the present study suggest that there was a psychological cost associated with overeating combined with high dietary restraint. The interaction of these two variables was correlated with high scores on the Bulimia and Body Dissatisfaction scales of the EDI. As can be seen in Fig. 1, subjects scoring high on both the Dietary Restraint and Overeating scales had the lowest average BMI. This
40
D. A. WILLIAMSON ET AL.
pattern of data suggests that it is the combination of high dietary restraint and overeating that is most highly correlated with bulimic symptoms in women of normal weight. The canonical correlation of Dietary Restraint with eating disorder symptoms showed that, by itself, dietary restraint was not strongly correlated with bulimic symptoms. Overeating, however, was highly correlated with measures of bulimia and hunger. In addition, overeating with low dietary restraint was associated with high BMI, as shown in Fig. 1. This pattern of binge eating and obesity is consistent with descriptions of binge eating disorder (American Psychiatric Association, 1994; Williamson, Gleaves & Savin, 1992). In summary, this investigation found that overeating was associated with increasing obesity unless the weight gain associated with overeating was moderated by dietary restraint. Overeating was associated with bulimic symptoms and increased body mass. High dietary restraint was associated with lower body mass. A combination of overeating and high dietary restraint was associated with bulimic symptoms and body dissatisfaction despite having a normal body weight. These findings are consistent with other recent studies which suggest that, for most persons, attempting to diet may not have adverse effects upon body mass and behavior, unless dieting is accompanied with overeating (French & Jeffery, 1994). R Allison, D. A., Kalinsky, L. B. & Gorman, B. S. (1992) A comparison of the psychometric properties of three measures of dietary restraint. Psychological Assessment, 4, 391–8. American Psychiatric Association. (1994) Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: American Psychiatric Association. Blundell, J. E. (1990) How culture undermines the biopsychological system of appetite control. Appetite, 14, 113–15. Brownell, K. D. (1991) Dieting and the search for the perfect body: where physiology and culture collide. Behavior Therapy, 22, 1–12. Brownell, K. D. & Rodin, J. (1994) The dieting maelstrom: is it possible and advisable to lose weight. American Psychologist, 49, 781–92. Cooper, P. J. & Charnock, D. (1990) From restraint to bulimic episodes: a problem of some loose connections. Appetite, 14, 120–2. French, S. A. & Jeffery, R. W. (1994) Consequences of dieting to lose weight: effects on physical and mental health. Health Psychology, 13, 195–215. Garner, D. M. & Olmstead, M. P. (1984) Manual for the Eating Disorder Inventory. Odessa, FL: Psychological Assessment Resources, Inc. Garner, D. M., Olmstead, M. P. & Polivy, J. (1983) Development and validation of a multidimensional eating disorder inventory for anorexia and bulimia nervosa. International Journal of Eating Disorders, 2, 15–34. Garrow, J. S. (1983) Indices of adiposity. Review of Clinical Nutrition, 53, 697–708. Gleaves, D. H., Williamson, D. A. & Barker, S. E. (1993a) Confirmatory factor analysis of a multidimensional model of bulimia nervosa. Journal of Abnormal Psychology, 102, 173–6. Gleaves, D. H., Williamson, D. A. & Barker, S. E. (1993b) Additive effects of mood and eating forbidden foods upon the perceptions of overeating and binging in bulimia nervosa. Addictive Behaviors, 18, 299–309. Gross, J., Rosen, J. C., Leitenberg, H. & Willmuth, M. (1986) Validity of Eating Attitudes Test and the Eating Disorder Inventory in bulimia nervosa, Journal of Consulting and Clinical Psychology, 54, 875–6. Heatherton, T. F., Herman, C. P., Polivy, J., King, G. A. & McGree, S. T. (1988) The (mis)measurement of restraint: an analysis of conceptual and psychometric issues. Journal of Abnormal Psychology, 97, 19–28.
DIETARY RESTRAINT
41
Herman, C. P. & Mack, D. (1975) Restrained and unrestrained eating. Journal of Personality, 43, 647–60. Huberty, C. J. & Morris, J. D. (1989) Multivariate analysis versus multiple univariate analyses. Psychological Bulletin, 105, 302–8. Laessle, R. G., Tuschl, R. J., Kotthaus, B. C. & Pirke, K. M. (1989) Behavioral and biological correlates of dietary restraint in normal life. Appetite, 12, 83–94. Lowe, M. R. (1993) The effects of dieting on eating behavior. Psychological Bulletin, 114, 100–21. Polivy, J. & Herman, C. P. (1985) Dieting and binging: a causal analysis. American Psychologist, 40, 193–201. Polivy, J., Zeitlin, S. B., Herman, C. P. & Beal, A. L. (1994) Food restriction and binge eating: a study of former prisoners of war. Journal of Abnormal Psychology, 103, 409–11. Ruderman, A. J. (1986) Dietary restraint: a theoretical and empirical review. Psychological Bulletin, 99, 247–62. Stunkard, A. J. & Messick, S. (1985) The Three-Factor Eating Questionnaire to measure dietary restraint, disinhibition, and hunger. Journal of Psychosomatic Research, 29, 71–83. Treasure, J. (1990) Comments on some theoretical considerations: dietary restraint to binge eating. Appetite, 14, 131–2. Westenhoefer, J. (1991) Dietary restraint and disinhibition: is restraint a homogenous construct? Appetite, 16, 45–55. Westenhoefer, J., Pudel, V. & Maus, N. (1990) Some restrictions on dietary restraint. Appetite, 14, 137–41. Williamson, D. A., Barker, S. E., Bertman, L. J. & Gleaves, D. H. (in press). Body image, body dysphoria, and dietary restraint: factor structure in non-clinical subjects. Behaviour Research and Therapy. Williamson, D. A., Davis, C. J., Goreczny, McKenzie, S. J. & Watkins, P. C. (1989) The eating questionnaire-revised: a new symptom checklist for bulimia. In P. A. Keller & L. G. Ritt (Eds.), Innovations in clinical practice: a sourcebook (pp. 321–326). Sarasota, FL: Professional Resource Exchange, Inc. Williamson, D. A., Gleaves, D. H. & Lawson, O. J. (1991) Biased perception of overeating in bulimia nervosa and compulsive binge eaters. Journal of Psychopathology and Behavioral Assessment, 13, 257–68. Williamson, D. A., Gleaves, D. H. & Savin, S. M. (1992) Empirical classification of eating disorder NOS: support for DSM-IV changes. Journal of Psychopathology and Behavioral Assessment, 14, 201–16. Received 12 August 1993, revision 30 October 1994
18-07-95 13:50:47
Appetite, 1995, 25, 31–41
Association of Body Mass with Dietary Restraint and Disinhibition
DONALD A. WILLIAMSON, OLGA J. LAWSON, ELLEN R. BROOKS, PATRICIA J. WOZNIAK, DONNA H. RYAN, GEORGE A. BRAY and ERICH G. DUCHMANN Louisiana State University, Pennington Biomedical Research Center
The relationship of disinhibition and dietary restraint with body mass was studied in a sample of 293 women. Results suggested that higher body mass was associated with an interaction of disinhibition and dietary restraint. The association of disinhibition with higher body mass was moderated by increased dietary restraint. Symptoms of an eating disorder were more strongly associated with disinhibition than with dietary restraint. These results suggest that dieting may moderate the increased body mass associated with overeating. Psychological and eating problems associated with dietary restraint were found to be of less significance than those associated with disinhibition. 1995 Academic Press Limited
A B M D R D The term dietary restraint has been used to describe a person’s intent to restrict dietary intake in order to control body weight (Herman & Mack, 1975). The construct of dietary restraint has been used to explain the common observation that obese persons intend to diet yet frequently overeat and thus maintain a higher body weight. A central hypothesis of dietary restraint theory is that the intent to diet may be disrupted or “disinhibited” by certain events, e.g. dysphoric emotions, alcohol or the availability of appetizing foods (Ruderman, 1986). This disinhibition hypothesis is important because it explains why persons trying to restrict eating frequently overeat. Numerous laboratory investigations have found support for the loss of control hypothesis (Lowe, 1993) and recently this hypothesis has been extended to account for the binge eating observed in bulimia nervosa and non-clinical subjects (Polivy & Herman, 1985; Polivy et al., 1994; Ruderman, 1986). One conclusion derived from this research is that dieting may have adverse psychological and behavioral effects upon certain individuals (Brownell, 1991; Brownell & Rodin, 1994). The adverse events associated with dieting range from binge eating (Polivy & Herman, 1985) to lowered total energy expenditure (Laessle et al., 1989). One consequence of these research findings has been increased concern about the benefits vs. costs of dieting to control obesity (Blundell, 1990; Brownell, 1991; Brownell & Rodin, 1994; French & Jeffery, 1994). Address correspondence to: Donald A. Williamson, Ph.D., Department of Psychology, Louisiana State University, Baton Rouge, LA 70803, U.S.A. 0195–6663/95/040031+11 $12.00/0
1995 Academic Press Limited
32
D. A. WILLIAMSON ET AL.
Other researchers have questioned the hypothesized relationship between dietary restraint and overeating (Cooper & Charnock, 1990; Treasure, 1990). In response to conflicting results in this research literature, Lowe (1993) observed that dietary restraint may be a multidimensional construct including a history of dieting, current dieting and degree of sustained weight loss. Also, Westenhoefer, Pudel and Maus (1990) reported the independence of dietary restraint and disinhibition in a large sample of European women. The investigation of Westenhoefer et al. (1990) is of special importance to the present investigation. Westenhoefer presented the results of two surveys of European women pertaining to body weight and caloric intake. Body mass index and caloric intake were analysed as a function of two scales of the Three Factor Eating Questionnaire (TFEQ; Stunkard & Messick, 1985). The two scales were Cognitive Restraint, which measures the intent to diet and actual dieting behavior (Allison, Kalinsky & Gorman, 1992), and Disinhibition, which measures overeating and binge eating in response to a variety of situations associated with loss of control of food intake (Stunkard & Messick, 1985). The results of this study showed that body mass index was predicted by an interaction of Cognitive Restraint and Disinhibition, such that low Restraint and high Disinhibition scores were associated with higher body mass. An interaction effect was also found for caloric intake, showing that high Restraint and low Disinhibition scores were associated with lower caloric intake and low Restraint and high Disinhibition scores were associated with higher caloric intake. One implication of these data is that Dietary Restraint and Disinhibition, as measured by the TFEQ, may have interactive influences upon obesity. Another implication is that dieting may serve to moderate the association of overeating with increased body mass. A second observation of Westenhoefer et al. (1990) was that dietary restraint was not necessarily associated with problematic eating behavior. In a follow-up study, Westenhoefer (1991), reported that the Cognitive Restraint scale of the TFEQ had two subscales, which he called Rigid and Flexible Control (of eating). Rigid Control was positively correlated with Disinhibition scores, whereas, flexible control was negatively correlated with Disinhibition scores. The present investigation was designed to replicate and extend the results of the earlier studies of Westenhoefer. The present study directly measured the body weight of subjects, as opposed to the self-reported survey data of the Westenhoefer et al. (1990) study, and used several measures of problematic eating habits and attitudes which have been validated for use with the eating disorders. A secondary purpose of the current study was to evaluate the association of Rigid and Flexible Control with Disinhibition and body mass. With this research design we sought to evaluate the association of Cognitive Restraint and Disinhibition with body mass and eating disorder symptoms.
M Subjects A total of 301 women were recruited from the community and university populations in the Baton Rouge region. These subjects were recruited via media advertisements for a cross-sectional study of eating behavior. They were not recruited
DIETARY RESTRAINT
33
as a part of a weight loss investigation or clinical weight loss program. From this sample of 301 women, 293 completed the study. Average age of these 293 women was 38.6 yrs. (SD=14·3). The age range was 17 to 78 years. Racial composition of the sample was 88% Caucasian, 12% African-American, and 1% other minority. The distribution of racial groups did not differ as a function of scores on the Dietary Restraint or Disinhibition scales of the TFEQ (p>0·30). Average height of the subjects was 1·6 m (SD=0·07) and average weight was 80·9 kg (SD=24·4). Weights ranged from 44 kg to 182 kg. It should be noted that the average weight of this sample was substantially higher than the average American woman. Assessment Measures Height and weight Height was measured in centimeters using a stadiometer. Weight was measured in kilograms using a digital scale calibrated to an accuracy of 0·10 kg. These measures were converted to body mass index (BMI), using the formula: BMI=weight (kg)/ height2 (m), which has been validated as an index of adiposity (Garrow, 1983). Three Factor Eating Questionnaire (TFEQ) The TFEQ was developed by Stunkard and Messick (1985) in response to psychometric problems that had been reported for other measures of dietary restraint. The three factors of the TFEQ were developed using factor analysis to derive a 51item questionnaire. These scales were labelled Cognitive Restraint, Disinhibition of Control, and Perceived Hunger. Later investigations have often referred to the first two scales as Dietary Restraint and Disinhibition. Given current knowledge about the validity of these two scales, we believe that Dietary Restraint is an appropriate label for the first factor. We believe that the Disinhibition scale is best viewed as a measure of overeating. These interpretations of the first two factors of the TFEQ are based upon recent findings that the first factor measures the intent to diet and dieting behavior (Allison et al., 1992; Laessle et al., 1989) and that the second factor measures episodic overeating that may be independent of dietary restraint (Westenhoefer et al., 1990). As noted by Heatherton et al. (1988), “without initial dietary inhibition (or restraint), it is difficult to imagine disinhibition” (p. 25). Examination of the items which form the second factor of the TFEQ show that overeating is the behavioral constant of this factor. For these reasons, we will refer to the three scales of the TFEQ as measuring: Dietary Restraint, Overeating, and Perceived Hunger. All three scales of the TFEQ have been found to have good test–retest reliability (Stunkard & Messick, 1985). Westenhoefer (1991) identified two subscales of the Dietary Restraint factor: (a) Rigid Control, which was positively correlated with Disinhibition scores; and (b) Flexible Control, which was negatively correlated with Disinhibition scores. Each subscale was composed of seven of the 21 original items of the Dietary Restraint scale. These two subscales have not been extensively studied and were included in several secondary analyses in this study. Eating Disorder Inventory (EDI) The EDI is a 64-item questionnaire developed by Garner, Olmstead and Polivy (1983). The questions of the EDI are scored using a six-point rating scale. The EDI is composed of eight scales: (1) Drive for Thinness (DFT), (2) Bulimia (B), (3) Body Dissatisfaction (BD), (4) Ineffectiveness (I), (5) Perfectionism, (6) Interpersonal
34
D. A. WILLIAMSON ET AL.
T 1 Intercorrelations of TFEQ Scales, Age and BMI Dietary Restraint Overeating Dietary Restraint Flexible Control Rigid Control Age
−0·07
Flexible Control −0·07 0·85∗∗
Rigid Control 0·04 0·85∗∗ 0·65∗∗
Age
BMI
0·26∗∗ 0·16∗ 0·04 0·16∗
0·38∗∗ −0·18∗ −0·23∗ −0·11 0·44∗∗
∗ p<0·01, ∗∗ p<0·001. TFEQ=Three Factor Eating Questionnaire, BMI=Body Mass Index.
Distrust (ID), (7) Interoceptive Awareness (IA) and (8) Maturity Fears (MF). The first three scales are considered to be primary symptoms of the eating disorders. Internal consistency of the scales has been reported to be high and test–retest reliability of the scales is satisfactory (Garner & Olmstead, 1984). Subjects diagnosed with anorexia and bulimia nervosa score higher than controls on the eight scales of the EDI (Garner & Olmstead, 1984; Gross et al., 1986). Eating Questionnaire-Revised (EQ-R) The EQ-R is a 15 item questionnaire developed by Williamson, Davis, Goreczny, McKenzie and Watkins (1989) as a measure of bulimic symptoms. It is a singlefactor test with a multiple-choice scoring format. It has been validated as a measure of bulimic symptoms and has good test–retest reliability and internal consistency. Procedure Subjects were scheduled for a single assessment session and all data were collected in that session. Consent forms were signed. Height and weight were measured and then the three questionnaires, TFEQ, EDI and EQ-R, were administered. R Statistical Methods Correlations among variables were evaluated using Pearson correlation. Multiple regression analysis was used to evaluate the relationship of BMI with Dietary Restraint, Overeating and Age. Multivariate analysis of variance (MANOVA) was used to evaluate the association of Dietary Restraint and Overeating with eating disorder symptoms. The associations among these variables were described using the canonical variates from MANOVA (Huberty & Morris, 1989). An alpha level of p<0·01 was used for interpretation of all statistical analyses. This relatively conservative alpha level was chosen because of the numerous correlations and regression analyses conducted in this study. Correlational Analysis Table I summarizes the intercorrelations among scales from the TFEQ, age and BMI. Overeating was uncorrelated with the original Dietary Restraint scale and
DIETARY RESTRAINT
35
with its subscales, Flexible and Rigid Control. Overeating was positively correlated with age and BMI. The subscales of Dietary Restraint, Flexible and Rigid Control, were strongly correlated with the original Dietary Restraint scale. Also, these subscales were positively and strongly correlated. Age was positively correlated with BMI, Dietary Restraint and Rigid Control. Dietary Restraint and Flexible Control were negatively correlated with BMI. Rigid Control was not significantly correlated with BMI. Restraint, Overeating and Body Mass The association of Dietary Restraint and Overeating with BMI was tested using a series of multiple regression analyses. Table 2 summarizes the six statistical models that were tested. All variables were treated as continuous variables. Model 1 represents a replication of the Westenhoefer study. In this regression model, a significant interaction of Dietary Restraint and Overeating with BMI was found. This interaction is shown in Fig. 1. For subjects with lower Overeating scores, level of Dietary Restraint was not significantly related to BMI. For subjects with average to high Overeating scores, low Dietary Restraint was associated with increased BMI, but as Restraint scores increased, BMI decreased. Since age was found to be positively correlated with BMI, Dietary Restraint and Overeating scores, age was added to the regression analysis as a concomitant variable in model 2. Age was found to be a significant predictor variable, and percentage of explained variance for BMI increased from 23% in model 1 to 37% in model 2. The primary effect of entering age as a predictor variable was a reduction in unexplained error. There was also a slight alteration of the regression slopes. Comparison of the two regression models shows that the basic relationships between Dietary Restraint, Overeating and BMI were not strongly affected by inclusion of age as a predictor variable. Rigid Control, Flexible Control, Overeating and Body Mass In regression models 3 and 4, Dietary Restraint was replaced by Rigid Control as a predictor variable. As can be seen in Table 2, these models were quite similar to models 1 and 2, which included Dietary Restraint as a predictor variable. Similarly, in models 5 and 6, Flexible Control replaced Rigid Control as a predictor variable. The results of models 5 and 6 were quite similar to corresponding earlier models (i.e. models 1 and 3, for model 5; and models 2 and 4, for model 6). In model 7, Rigid Control and Flexible Control were predictors of BMI. This model was statistically significant, but accounted for only 6% of the variance of BMI. Flexible Control was a significant predictor of BMI, but Rigid Control was not a significant predictor. Restraint, Overeating and Eating Disorder Symptoms The association of eating disorder symptoms with Dietary Restraint and Overeating was tested using MANOVA. The different measures of eating disorder symptoms were the dependent measures. Dietary Restraint, Overeating and Restraint×Overeating were treated as continuous predictors. The F statistic, based on Wilk’s Lambda, was used to test for statistical significance. The dependent
36
D. A. WILLIAMSON ET AL.
T 2 Summary of Multiple Regression Analyses with BMI as the Dependent Variable Model
Predictor variables
Parameter estimate t
p
R2
1 Restraint [F(3, 289)=28·40, Overeating p<0·0001] R×O Interaction Intercept
−0·43 0·80 −0·12 30·22
−4·02 6·29 −4·90
0·001 0·001 0·001
0·23
2 Restraint [F(4, 288)=42·09, Overeating p<0·0001] R×O Interaction Age Intercept
−0·56 0·54 −0·11 0·26 30·22
−5·72 4·58 −5·07 8·03
0·001 0·001 0·001 0·001
0·37
3 Rigid Control −0·78 [F(3, 289)=22·83, Overeating 0·88 p<0·0001] RC×O Interaction −0·22 Intercept 30·42
−2·94 6·85 −3·52
0·001 0·005 0·001
0·19
4 Rigid Control −1·04 [F(4, 288)=34·72, Overeating 0·65 p<0·0001] RC×O Interaction −0·20 Age 0·25 Intercept 30·40
−4·23 5·38 −3·43 7·56
0·001 0·001 0·001 0·001
0·33
5 Flexible Control [F(3, 289)=28·57, Overeating p<0·0001] FC×O Interaction Intercept
−1·39 0·79 −0·29 30·20
−4·74 6·28 −4·15
0·001 0·001 0·001
0·23
6 Flexible Control [F(4, 288)=39·72, Overeating p<0·0001] FC×O Interaction Age Intercept
−1·49 0·57 −0·27 0·24 30·20
−5·56 4·74 −4·17 7·53
0·001 0·001 0·001 0·001
0·36
7 [F(2, 290)=8·75, p<0·0002]
−1·54 0·35 30·32
−3·70 0·94
0·001 0·351
0·06
Flexible Control Rigid Control Intercept
F values describe the statistical model; t values reflect the significance of the variables within a model. Abbreviations: R=Dietary Restraint, O=Overeating, RC=Rigid Control, FC=Flexible Control.
variables were the Perceived Hunger scale, EQ-R and the eight scales of the EDI. Main effects for Dietary Restraint, F(10, 288)=15·26, p<0·0001, and Overeating, F (10, 288)=46·32, p<0·0001, were found. Also the interaction of Dietary Restraint and Overeating was statistically significant, F(10, 288)=4·03, p<0·0001). These effects were interpreted in terms of canonical structure, as suggested by Huberty and Morris (1989). The results of this analysis are summarized in Table 3. The canonical variate, Overeating, was positively correlated with Perceived Hunger, two measures of bulimic behavior (the EQ-R and EDI bulimia scale), and with three other EDI scales, Drive
37
DIETARY RESTRAINT
F 1. Relationship between BMI and Overeating at three values of Restraint (Μ, Mean; Χ, Mean +1 standard deviation; Ε, mean −1 standard deviation), unadjusted for Age.
T 3 Canonical correlations derived from MANOVA With Overeating
With Restraint
O×R interaction
Canonical correlation
0·79
0·59
0·35
Squared canonical correlation
0·62
0·35
0·12
0·68∗ 0·90∗ 0·56∗ 0·80∗ 0·55∗ 0·37 0·25 0·16 0·52∗ 0·13
−0·22 0·23 0·58∗ −0·33 0·01 −0·23 0·11 −0·06 0·00 −0·19
0·08 0·11 0·09 0·59∗ 0·60∗ 0·45 0·09 0·29 0·21 0·39
15·26
46·32
4·03
Total canonical structure
F value
Hunger EQ-R EDI DT EDI B EDI BD EDI I EDI P EDI ID EDI IA EDI MF
Canonical loadings greater than 0·50 were interpreted to define canonical variates as designated by ∗. The degrees of freedom were 10 and 288 for each F value. All three F values were statistically significant (p<0·001). Abbreviations: O=Overeating, R=Restraint, EQ-R=Eating QuestionnaireRevised, EDI=Eating Disorder Inventory, DT=Drive for Thinness, B=Bulimia, BD=Body Dissatisfaction, I=Ineffectiveness, P=Perfectionism, ID=Interpersonal Distrust, IA=Interoceptive Awareness, MF=Maturity Fears.
38
D. A. WILLIAMSON ET AL.
for Thinness, Body Dissatisfaction, and Interoceptive Awareness. The canonical variate, Dietary Restraint, had high loadings for increasing scores on the EDI Drive for Thinness scale. The canonical correlations for Overeating (0·79) and Dietary Restraint (0·59) were quite high. The canonical correlation for the interaction of Restraint and Overeating was somewhat smaller (0·35), but was statistically significant. The canonical variate associated with the interaction term was positively correlated with two EDI scales, Bulimia and Body Dissatisfaction. This interaction indicated that the symptoms of bulimia and body dissatisfaction were associated with a pattern of elevated scores on both the Dietary Restraint and Overeating scales.
D The results of this study supported the contention that the Dietary Restraint scale of the TFEQ was measuring the intent to diet and dieting behavior. The results also suggested that the scale previously labelled Disinhibition was measuring overeating. The Dietary Restraint scale was most strongly correlated with the Drive for Thinness scale of the EDI. Two recent factor-analytic studies (Gleaves, Williamson & Barker, 1993a; Williamson, Barker & Bertman, in press) have found the Drive for Thinness scale to load on a Dietary Restraint factor. Furthermore, as Dietary Restraint increased, body mass decreased, in persons reporting significant overeating. This pattern suggests that the Dietary Restraint scale is also measuring actual dieting behavior which results in weight suppression. These findings are consistent with the results of earlier studies of the Dietary Restraint scale. The Overeating scale was most positively correlated with measures of bulimic binge eating, i.e. EQ-R and EDI Bulimia scale. It was also positively correlated with the Perceived Hunger scale, of the TFEQ. For the purposes of further discussion, we will assume that Dietary Restraint and Overeating were the constructs measured by the TFEQ. We also found the Dietary Restraint and Overeating scales to be uncorrelated. Earlier studies which have investigated intercorrelations among the three scales of the TFEQ have reported some inconsistencies. In the original description of the TFEQ, Stunkard and Messick (1985) reported a moderate positive correlation (r= 0·43) between the Dietary Restraint and Overeating (Disinhibition) scales. Westenhoefer (1991) reported a moderate negative correlation (r=−0·37) between those two scales; our study found a non-significant relationship. These somewhat inconsistent findings are most likely to be due to differences in the samples recruited for each of the three studies. In a heterogeneous sample such as that of our study, the two scales were essentially independent. The independence of the Dietary Restraint and Overeating scales suggest that the causal link between restraint and overeating or binge eating may vary across individuals, with some individuals successfully dieting, some intermittently dieting and binge eating, and others who overeat or binge, but seldom attempt to restrain their eating (French & Jeffery, 1994; Lowe, 1993). The finding of an interactive association of Dietary Restraint and Overeating with body mass index is consistent with the survey results reported by Westenhoefer et al. (1990). This interaction, as illustrated in Fig. 1, suggests that increased dietary restraint moderates the association of overeating with obesity. Some of the subjects
DIETARY RESTRAINT
39
in this sample scored high on the Overeating scale, yet scored very low on a measure of dietary restraint. These unrestrained overeaters were very obese, which indirectly supports the conclusion that these women were overeating but were not actively dieting (Lowe, 1993). Women with low scores on the Overeating scale did not vary in body mass as a function of Dietary Restraint. This finding suggests that some women are restrained eaters, yet do not significantly overeat and maintain a weight level within the normal to moderately-overweight range. When age was added to this regression model, unexplained error was reduced, but the relationship of BMI with Dietary Restraint and Overeating was not substantially altered. Similarly, the replacement of Dietary Restraint with Rigid Control or Flexible Control did not substantially alter the relationships among the other variables. This finding suggests that both flexible and rigid approaches to dieting can moderate the association of BMI with Overeating. It should be noted that the Flexible and Rigid Control scales were highly correlated (r=0·65) in this study, and both scales were highly correlated (r=0·85) with the Dietary Restraint scale, from which they were derived (Westenhoefer, 1991). Also, in this study, the Rigid and Flexible Control scales were not significantly correlated with Overeating. In the development of these scales, Westenhoefer (1991) found the Rigid Control to be positively correlated with Overeating and the Flexible Control scale to be negatively correlated with Overeating. In this study, we found Flexible Control to be negatively correlated with BMI. Rigid Control was uncorrelated with BMI. The sample of the Westenhoeffer (1991) study was composed of persons seeking weight loss treatment. In the present study, lean and obese subjects were recruited and weight loss was not a purpose of the study. Given these conflicting results, we believe that further research on the association of Rigid and Flexible Control with Overeating and BMI is needed. The lack of association between high dietary restraint and overeating suggests that dietary restraint and overeating are independent behavioral phenomena when measured by the TFEQ. There are several possible explanations for this finding. One explanation is that overeating results in weight gain and that increasing dietary restraint is employed to attenuate this weight gain in some individuals. A second explanation is that subjects reporting frequent overeating and high dietary restraint are attempting to suppress their weight below that which is biologically normal, which causes episodic overeating in only these subjects. The finding of a significant canonical correlation of bulimic symptoms with the interaction of Dietary Restraint and Overeating is consistent with this explanation. A third explanation is that subjects scoring high on Dietary Restraint were using different criteria for judging whether they were “overeating”. Two recent studies (Gleaves, Williamson & Barker, 1993b; Williamson, Gleaves & Lawson, 1991) found that bulimics and compulsive binge eaters are biased in their judgements of overeating. It is possible that the restrained eaters in this study also inflated the perception of their problems with overeating, resulting in higher scores on the Overeating scale yet were not consuming the quantity of food being eaten by unrestrained overeaters. Further research will be required to determine which of these explanations is most valid. The results of the present study suggest that there was a psychological cost associated with overeating combined with high dietary restraint. The interaction of these two variables was correlated with high scores on the Bulimia and Body Dissatisfaction scales of the EDI. As can be seen in Fig. 1, subjects scoring high on both the Dietary Restraint and Overeating scales had the lowest average BMI. This
40
D. A. WILLIAMSON ET AL.
pattern of data suggests that it is the combination of high dietary restraint and overeating that is most highly correlated with bulimic symptoms in women of normal weight. The canonical correlation of Dietary Restraint with eating disorder symptoms showed that, by itself, dietary restraint was not strongly correlated with bulimic symptoms. Overeating, however, was highly correlated with measures of bulimia and hunger. In addition, overeating with low dietary restraint was associated with high BMI, as shown in Fig. 1. This pattern of binge eating and obesity is consistent with descriptions of binge eating disorder (American Psychiatric Association, 1994; Williamson, Gleaves & Savin, 1992). In summary, this investigation found that overeating was associated with increasing obesity unless the weight gain associated with overeating was moderated by dietary restraint. Overeating was associated with bulimic symptoms and increased body mass. High dietary restraint was associated with lower body mass. A combination of overeating and high dietary restraint was associated with bulimic symptoms and body dissatisfaction despite having a normal body weight. These findings are consistent with other recent studies which suggest that, for most persons, attempting to diet may not have adverse effects upon body mass and behavior, unless dieting is accompanied with overeating (French & Jeffery, 1994). R Allison, D. A., Kalinsky, L. B. & Gorman, B. S. (1992) A comparison of the psychometric properties of three measures of dietary restraint. Psychological Assessment, 4, 391–8. American Psychiatric Association. (1994) Diagnostic and statistical manual of mental disorders (4th ed.). Washington, DC: American Psychiatric Association. Blundell, J. E. (1990) How culture undermines the biopsychological system of appetite control. Appetite, 14, 113–15. Brownell, K. D. (1991) Dieting and the search for the perfect body: where physiology and culture collide. Behavior Therapy, 22, 1–12. Brownell, K. D. & Rodin, J. (1994) The dieting maelstrom: is it possible and advisable to lose weight. American Psychologist, 49, 781–92. Cooper, P. J. & Charnock, D. (1990) From restraint to bulimic episodes: a problem of some loose connections. Appetite, 14, 120–2. French, S. A. & Jeffery, R. W. (1994) Consequences of dieting to lose weight: effects on physical and mental health. Health Psychology, 13, 195–215. Garner, D. M. & Olmstead, M. P. (1984) Manual for the Eating Disorder Inventory. Odessa, FL: Psychological Assessment Resources, Inc. Garner, D. M., Olmstead, M. P. & Polivy, J. (1983) Development and validation of a multidimensional eating disorder inventory for anorexia and bulimia nervosa. International Journal of Eating Disorders, 2, 15–34. Garrow, J. S. (1983) Indices of adiposity. Review of Clinical Nutrition, 53, 697–708. Gleaves, D. H., Williamson, D. A. & Barker, S. E. (1993a) Confirmatory factor analysis of a multidimensional model of bulimia nervosa. Journal of Abnormal Psychology, 102, 173–6. Gleaves, D. H., Williamson, D. A. & Barker, S. E. (1993b) Additive effects of mood and eating forbidden foods upon the perceptions of overeating and binging in bulimia nervosa. Addictive Behaviors, 18, 299–309. Gross, J., Rosen, J. C., Leitenberg, H. & Willmuth, M. (1986) Validity of Eating Attitudes Test and the Eating Disorder Inventory in bulimia nervosa, Journal of Consulting and Clinical Psychology, 54, 875–6. Heatherton, T. F., Herman, C. P., Polivy, J., King, G. A. & McGree, S. T. (1988) The (mis)measurement of restraint: an analysis of conceptual and psychometric issues. Journal of Abnormal Psychology, 97, 19–28.
DIETARY RESTRAINT
41
Herman, C. P. & Mack, D. (1975) Restrained and unrestrained eating. Journal of Personality, 43, 647–60. Huberty, C. J. & Morris, J. D. (1989) Multivariate analysis versus multiple univariate analyses. Psychological Bulletin, 105, 302–8. Laessle, R. G., Tuschl, R. J., Kotthaus, B. C. & Pirke, K. M. (1989) Behavioral and biological correlates of dietary restraint in normal life. Appetite, 12, 83–94. Lowe, M. R. (1993) The effects of dieting on eating behavior. Psychological Bulletin, 114, 100–21. Polivy, J. & Herman, C. P. (1985) Dieting and binging: a causal analysis. American Psychologist, 40, 193–201. Polivy, J., Zeitlin, S. B., Herman, C. P. & Beal, A. L. (1994) Food restriction and binge eating: a study of former prisoners of war. Journal of Abnormal Psychology, 103, 409–11. Ruderman, A. J. (1986) Dietary restraint: a theoretical and empirical review. Psychological Bulletin, 99, 247–62. Stunkard, A. J. & Messick, S. (1985) The Three-Factor Eating Questionnaire to measure dietary restraint, disinhibition, and hunger. Journal of Psychosomatic Research, 29, 71–83. Treasure, J. (1990) Comments on some theoretical considerations: dietary restraint to binge eating. Appetite, 14, 131–2. Westenhoefer, J. (1991) Dietary restraint and disinhibition: is restraint a homogenous construct? Appetite, 16, 45–55. Westenhoefer, J., Pudel, V. & Maus, N. (1990) Some restrictions on dietary restraint. Appetite, 14, 137–41. Williamson, D. A., Barker, S. E., Bertman, L. J. & Gleaves, D. H. (in press). Body image, body dysphoria, and dietary restraint: factor structure in non-clinical subjects. Behaviour Research and Therapy. Williamson, D. A., Davis, C. J., Goreczny, McKenzie, S. J. & Watkins, P. C. (1989) The eating questionnaire-revised: a new symptom checklist for bulimia. In P. A. Keller & L. G. Ritt (Eds.), Innovations in clinical practice: a sourcebook (pp. 321–326). Sarasota, FL: Professional Resource Exchange, Inc. Williamson, D. A., Gleaves, D. H. & Lawson, O. J. (1991) Biased perception of overeating in bulimia nervosa and compulsive binge eaters. Journal of Psychopathology and Behavioral Assessment, 13, 257–68. Williamson, D. A., Gleaves, D. H. & Savin, S. M. (1992) Empirical classification of eating disorder NOS: support for DSM-IV changes. Journal of Psychopathology and Behavioral Assessment, 14, 201–16. Received 12 August 1993, revision 30 October 1994