Journal of Abnormal Psychology 1992, Vol. 101, No. 4, 668-674
Copyright 1992 by the American Psychological Association, Inc. 002l-843X/92/$3,00
Relationship Between Dietary Restraint, Energy Intake, Physical Activity, and Body Weight: A Prospective Analysis Robert C. Klesges and Terry R. Isbell
Lisa M. Klesges
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Center for Applied Psychological Research Memphis State University
Department of Biostatistics/Epidemiology University of Tennessee, Memphis
Dietary, activity, and body weight differences in high- and low-restrained eaters and the independent impact of dietary restraint on body weight change were evaluated in 287 adults (141 men, 146 women) followed over a 1-year period. Analyses of measures of energy intake, physical activity, and dietary restraint indicated that high-restrained eaters did not differ in physical activity compared to low-restrained eaters at baseline but were ingesting significantly fewer kcal/lb and a higher percentage of the diet from fat. Body mass was significantly greater in both high-restrained men and high-restrained women than in their low-restrained counterparts. Regression modeling procedures revealed that weight and body mass at baseline were related to weight gain in men. In contrast, weight gain in women was predicted by baseline weight and higher restraint scores. These results indicate that dietary restraint is associated with weight gain in women but not in men.
There is tremendous interest in determining the factors associated with eating behavior, weight gain, and the etiology maintenance of obesity as it contributes to a number of physical health problems (Van Itallie, 1979). Additionally, given the apparently increasing incidence of other eating disorders such as anorexia and bulimia (Stunkard & Stellar, 1984), several psychological theories of eating behavior have been developed. Partly on the basis of the early internality-externality theory of obesity (Nisbett, 1972; Schachter, 1968), Herman and Mack (1975) have developed the more contemporary construct of restraint. According to their original formulation, eating patterns are influenced by both internal physiological cues that prompt the desire for food and cognitively mediated factors to resist the temptation to eat (or to overeat). The two major hypotheses regarding restrained eating were (a) when the self-control of restrained eaters was somehow violated, eating increased, and (b) restraint variations underlie obese/normal-weight differences in eating behavior. As a result of more than a decade of research (see Ruderman, 1986, for a review), there appears to be support for the hypothesis of counterregulatory eating among restrained eaters, at least in laboratory settings. That is, studies have noted that if restrained eaters are forced to "violate" their diet by eating highcaloric foods (Herman & Mack, 1975) or foods they think are high caloric (Polivy, 1976) or by drinking alcohol (Polivy &
Herman, 1976a, 1976b), their subsequent laboratory consumption of food is increased. Unrestrained (or nondieting) subjects, on the other hand, appropriately regulate for the high-caloric food or drink by reducing their laboratory consumption of food (Herman & Mack, 1975). Although the data support a counterregulatory mechanism in restrained and a regulatory mechanism in unrestrained eaters, the data do not support the hypothesis that differences in level of restraint underlie obese/normal-weight differences in eating behavior (Ruderman & Christensen, 1983; Ruderman & Wilson, 1979). For example, one study (Ruderman & Christensen, 1983) reported that obese people did not behave as restrained eaters. In fact, the opposite was true: The obese subjects ate significantly less after the preload than without it, whereas normal-weight people ate similar amounts both with and without the preload. Restrained overweight people did not counterregulate; they regulated their food intake better than normal-weight people. There are at least three possible reasons why a relationship between dietary restraint, via the mechanism of increased dietary intake as a result of disinhibition, and obesity has not been observed. One explanation is that there may not be stable differences in the eating patterns of obese compared with normal-weight individuals. The vast majority of studies of nutrition and obesity have failed to find overweight/normal-weight differences on measures of dietary intake (e.g., Berenson, Blonde, & Farris, 1979; Huenemann, 1974), and if differences were found (e.g., Johnson, Burke, & Mayer, 1956), they appeared in a paradoxical direction, with obese subjects eating less. Although these early studies can be criticized on methodological grounds (Wooley, Wooley, & Dyrenforth, 1979), more recent investigations using sophisticated methodologies and large sample sizes (e.g., Braitman, Adlin, & Stanton, 1985, N = 6,219) also have failed to find differences in habitual levels of dietary intake and obesity. A second reason for a lack of relationship between dietary
This study was supported by grants awarded to Robert C. Klesges (HL-36553, HL-45057, HL-46352) by the National Heart, Lung, and Blood Institute. Support was also received from a Centers of Excellence grant awarded to the Department of Psychology, Memphis State University, by the State of Tennessee. The authors would like to acknowledge Mary Lou Klem for her helpful suggestions on this manuscript. Correspondence concerning this article should be addressed to Robert C. Klesges, Center for Applied Psychological Research, Department of Psychology, Memphis State University, Memphis, Tennessee 38152. 668
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
DIETARY RESTRAINT AND BODY WEIGHT restraint and obesity may relate to the psychometric properties of the restraint scale. Studies that have examined the reliability of the restraint scale have found the scale's internal consistency to be lower in obese versus normal-weight groups. When used with normal-weight groups, the alpha reliability coefficients of the scale have been found to range from .78 (Klem, Klesges, Bene, & Mellon, 1990) to .86 (Ruderman, 1983). However, the reliability coefficients in obese subjects typically range from .50 to .68 (Johnson, Lake, & Mahan, 1984; Klem et al., 1990; Ruderman, 1983). Thus, it may be that there are indeed restrained obese eaters who counterregulate when disinhibited, but they respond differently to the restraint scale than do normal-weight restrained eaters. A third and perhaps most important reason why previous studies have failed to find a relationship between body weight and restraint relates to the cross-sectional nature of most designs. Obesity and weight gain are not static phenomena; they are processes that develop (usually gradually) over time. Moreover, in any cross-sectional survey of obesity, it is conceivable that subjects are either (a) becoming more overweight, (b) maintaining their obesity, or (c) losing weight. One would expect differences in dietary intake between overweight and normalweight subjects if overweight subjects were becoming more obese or if overweight subjects were trying to lose weight (i.e., might have lower intakes than normal-weight subjects; Klesges & Hanson, 1988). It is clear that longitudinal investigations determining the role of dietary restraint on prospective weight gain are needed. In the only longitudinal study to date (Klesges, Klem, Epkins, & Klesges, 1991), 305 adults (98 men, 207 women) were administered the restrained eating questionnaire, and self-reported height, weight, gender, race, and age were collected. Subjects were recontacted 2| years later and were reassessed on these same variables. Results indicated that restrained eating scores showed a high degree of consistency over a 2j-year period (intraclass correlation = .74). Over time, men gained more weight than women, and normal-weight subjects gained more weight over time than overweight subjects. No relationship was found between dietary restraint and weight gain over time. Unfortunately, this investigation relied on self-reports of height and weight and, most important, assessed neither energy intake nor physical activity as potential control variables in the prediction of changes in body weight. Thus, the purpose of the current investigation was to prospectively evaluate the role of dietary intake, physical activity, and dietary restraint on changes in body weight over a 1 -year period of time in a large cohort of adult men and women. By simultaneously evaluating dietary intake, physical activity, and dietary restraint, the independent contribution of each variable to weight gain, in both men and women, can be assessed. Method Subjects The cohort at baseline included Caucasian adults (141 men, 146 nonpregnant women) who participated in a longitudinal assessment of cardiovascular risk factor acquisition in parents and their young children. Average initial age of the participants was 35.7 years (SD - 4.51, range = 26-53 years); average initial weight for men was 191.29 Ib
669
(SD = 34.56), and for women, 148.62 Ib (SD = 32.27). After initial assessment, subjects were retested 1 year later. There was loss to attrition in the cohort of 13%, which resulted in a sample of 250 subjects (123 men, 127 women). Procedure Subjects were recruited through response forms distributed to local physicians' offices, day-care centers, and churches. Subjects were also recruited by advertisements placed in local newspapers, on local radio stations, and by recommendations from subjects already participating in the study. Data collected for this study were part of a large, longitudinal evaluation of cardiovascular risk factors in children (obesity onset and accelerated blood pressure increases) ages 3 to 4 years (at entry) and their parents. Families were intact, and the target child had to be the biological offspring of the parent. Approximately half of the children recruited were overweight at entry into the study, with the other half being normal weight. After agreeing to participate in the study, individuals received self-report questionnaires (described later) and consent forms to be filled out and signed. Subjects then participated in a laboratory assessment during which height, weight, physical activity, and restraint measurements were obtained. Subjects were recontacted after 1 year and brought to the laboratory, where dietary intake, height, and weight were obtained again. Measures Height and weight were measured by trained research assistants. Height was recorded to the nearest half-inch and weight to the nearest two-tenths of a pound. A hospital grade scale (Model #230 Health O Meter, Continental Scale Corporation, Bridgeview, IL), accurate to ±0.2 Ib and 0.25 in., was used for this procedure. Height and weight measurements were used to estimate body mass (kg/m2), which is a very good estimate of adiposity (Straw & Rogers, 1985). Dietary intake. Long-term assessment of dietary intake was obtained by the 1 -year Willett Food Frequency Questionnaire (Willett et al, 1985, 1988; Willett, Reynolds, Hoehner-Cottrell, Sampson, & Browne, 1987), which is a 61-item, self-administered questionnaire. The advantage of a long-term food frequency measure is its ability to take into account week-to-week and seasonal variations in dietary intake. Moreover, the Willett questionnaire strongly correlates to repeated dietary recalls (Willett et al, 1985) as well as daily dietary records administered for an entire year (Willett et al, 1988). For example, in a recent study (Rimm et al, 1992) of 127 male health professionals age 40-75 years, 1-year test-retest reliabilities (depending on the nutrient) ranged from .56 to .80. This study also compared the Willett questionnaire to the criteria of a 14-day dietary record and reported validity coefficients from .37 to .92. Administration of this measure is very straightforward, and subjects were able to direct questions about specific items to trained nutritionists. After subjects completed the food frequency, all items were checked to eliminate missing data. Subjects who failed to complete items were contacted, and missing data were obtained. All data were verified and analyzed using nutrient tapes provided by the Willett research group. Because this measure is retrospective over the previous year, the food frequency measure was obtained at the 1-year follow-up. Physical activity. Subjects were also administered the Baecke physical activity questionnaire, a factor-analyzed scale comprising 16 items that represent physical activity at work, sport, and nonsport leisuretime physical activity (Baecke, Burema, & Fritters, 1982). These items yield scores ranging from 1 (highly sedentary) to 5 (highly active in work, sport, and nonsport leisure time activity). The Baecke scale has been used extensively in recent years and has demonstrated a highly acceptable degree of reliability (3-month test-retest for the scales averages
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
670
R. KLESGES, T. ISBELL, AND L. KLESGES
.81) and construct validity (Baecke, van Stavern, & Burema, 1983; Washburn & Montoye, 1986). A recent study (Klesges et al, 1990) evaluated self-reports of physical activity against the criteria of surreptitious observation. Subjects were 44 adults who were required to engage in an hour of their preferred physical activity. During this period of time, subjects' physical activity levels were surreptitiously monitored by a trained observer and compared with self-reported estimates of physical activity obtained later. Results indicated that subjects were reasonably accurate in recalling their physical activity levels (average r=.62). Restraint. Subjects completed the Revised Restraint Scale (Herman & Polivy, 1980), which has been found to have generally high reliability, although its internal consistency varies as a function of the proportion of obese individuals in the sample (Johnson et al., 1984). When used with normal-weight groups, the alpha reliability coefficients of the scale have been found to range from .79 (Johnson et al., 1984) to .86 (Ruderman & Christensen, 1983). When administered to obese individuals, however, the alpha coefficients have typically been in the .50 range (Johnson et al, 1984; Ruderman & Christensen, 1983). A recent study (Klem et al, 1990) examining the impact of several demographic variables (e.g., race, gender, weight, and marital status) on psychometric characteristics of the scale found internal reliabilities of the total scale and its two subscales to be similar and generally high across the categories surveyed (alpha for total scale = .78).
Results Variables Assessed and Data Reduction Strategies The demographic and energy balance variables entered into the equation were based on previous reviews of the literature on the correlates of changes in body weight (e.g., Klesges & Hanson, 1988). As such, the demographic variables of age, sex, baseline weight, and body mass served as predictors for the prospective analyses. Both energy intake and levels of physical activity were also entered. Thus, the impact of dietary restraint was evaluated after adjustment for these important demographic and energy balance variables. Dietary intake was computed as average intake of total kilocalories per day, with fat and carbohydrate intake computed as a percentage of total kilocalories. For the multivariate analysis, kilocalories were adjusted for body weight and are expressed as kilocalories per pound of body weight. Cross-Sectional Analyses A multivariate analysis of variance (MANOV\) design was analyzed to determine differences in initial levels of dietary intake, physical activity, and body mass index between restrained and unrestrained eaters and for the possibility of a Restraint X Sex interaction. These results are summarized in Table I. Men with restraint scores of 14 or higher and women with restraint scores of 17 or higher were categorized as highrestrained individuals, representing a division of groups at the median. The MANOVA revealed no significant multivariate effect for a Restraint X Sex interaction, approximate F(275) = 1.06, p = .396, while the main effects for sex, approximate F(275) = 18.14, p < .0001, and restraint, approximate F(275) = 10.65, p < .0001, were statistically significant. Univariate analyses of variance for the restraint main effect revealed significant differences between high and low restraint categories for body mass index, F(\, 283) = 79.39, p < .0001;
energy ingested per pound of body weight, F(l, 283) = 23.84, p < .0001; percent of energy as fat, F(l, 283) = 12.95, p < .0001; and percent of energy as carbohydrate, F(l, 283) = 16.01, p < .0001. As can be seen in Table 1, high-restrained individuals had higher body mass compared to low-restrained individuals. In addition, high-restrained individuals ingested less energy per pound of body weight than low-restrained individuals. As is commonly observed in subjects with higher relative weights (Dreon et al, 1988; Romieu et al, 1988), high-restrained individuals had a dietary composition lower in carbohydrate and higher in fat related to higher body fat than low-restrained individuals. There were no differences in level of activity between restraint categories. Sex differences were also found to be marginally significant on univariate analyses of energy intake per pound of body weight, F(l, 283) = 3.77, p = .053, and significant for body mass, F(l, 283) = 19.62, p < .0001; sporting activity, F(l, 283) = 59.82, p < .0001; and leisure activity, F(l, 283) = 13.22, p < .0001. From Table 1 it can be seen that women had lower relative weights, higher energy intakes per pound of body weight, lower levels of sport activity, and higher levels of leisure activity compared with men. Cross-sectional predictors of body mass for both men and women are presented in Table 2. As Table 2 indicates, restraint scores were strong predictors of body mass in both men and women, with increasing restraint scores associated with higher body mass. Prospective Analyses Predictions of weight gain were investigated by means of a regression analysis with weight change from baseline to the 1-year follow-up as the dependent variable and independent variables of weight at baseline, body mass index, restraint score, age, total energy intake in kilocalories, percentage of daily kilocalories ingested from fats, percentage of daily kilocalories from carbohydrates, and work, sport, and leisure physical activity levels. Two-way interaction terms between sex and all other variables were also included in this analysis. The full regression model, F(19, 230) = 1.309, p = . 179, predicted 9.76% of the total variability in weight change. Because several of the interaction terms with gender were found to be significant in this model, follow-up analyses were conducted by submitting separate regression models for men and women. The results of these sex-specific follow-up models can be found in Table 3. For men, 13.02% of the total variability in their weight change was explained, F(10,112)= 1.68, p = .0948, a marginally significant amount. Significant predictors of weight change in men were baseline weight and body mass, whereas restraint, energy intake, and physical activity were of no statistically significant predictive value. Weight gain was higher with lower baseline body mass. Interestingly, a marginally significant, counterintuitive relationship was found with lowered percent fat intake associated with increased weight gain. For women, 21.22% of the total variability in weight change was explained, F(10,116) = 3.12, p = .0015. Significant predictors of weight change in women were baseline weight and re straint. Weight gain was associated with higher restraint score,-
671
DIETARY RESTRAINT AND BODY WEIGHT Table 1 Cross-Sectional Relationships Between Dietary Intake, Physical Activity, Restraint, and Body Weight
Men
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
High restraint (n = 68)
Women Low restraint (n = 73)
High restraint (n = 76)
Low restraint (n = 70)
Variable
M
SD
M
SD
M
SD
M
SD
Age" Body massa-b (kg/m2) Energy intake (kcal)8-1" per pound of body weight Intake from fat (% of diet)" Intake from carbohydrate*1 (% of diet) Level of work activity (1 = highly sedentary, 5 = highly active) Level of sport activity (1 = highly sedentary, 5 = highly active) Level of nonsport leisure" activity (1 = highly sedentary, 5 = highly active)
36.74 29.16
4.94 4.19
36.69 24.73
4.69 2.91
35.37 26.96
4.33 5.01
35.13 22.50
3.95 4.47
9.49 34.87
3.54 5.60
11.70 32.53
4.90 5.67
10.27 35.26
3.84 5.16
12.82 33.06
4.09 4.90
48.02
6.93
50.73
6.64
47.66
6.08
51.06
6.21
2.57
0.59
2.65
0.70
2.68
0.47
2.65
0.53
2.52
0.68
2.56
0.67
2.01
0.51
1.98
0.49
2.28
0.58
2.33
0.54
2.51
0.59
2.59
0.58
° Denotes a significant difference (p < .01) between men and women. b Denotes a significant difference (p < .01) between high-restrained and low-restrained subjects.
In contrast to the results with men, initial body mass was not a predictor of weight gain in women.1 Finally, energy intake was a marginally significant predictor of weight change in women.
Discussion The results of the current investigation indicate that high-restrained eaters have higher body mass and ingest fewer kilocalories per pound of body weight but ingest more energy as fat and less as carbohydrate compared to low-restrained individuals. Levels of physical activity were similar between the groups. In the prospective analyses of changes in body weight over a 1-year period in which both demographic and energy balance variables were controlled, dietary restraint was unrelated to changes in body weight in men but was significantly related to changes in body weight in women. The finding that high-restrained eaters were more likely to be overweight is consistent with several other investigations in the literature (e.g., Klem et al., 1990; Ruderman, 1986). However, this is the first investigation showing that restrained eaters maintain higher body weights despite lower levels of energy intake per pound of body weight and similar physical activity. The most likely explanation for this finding is that obese people are more likely to be restrained. In support of this notion, the pattern of dietary intake in restrained eaters is remarkably similar to that of obese people (Dreon et al., 1988; Romieu et al., 1988). That is, the restrained individuals had a lower total dietary intake, a higher percentage of the diet from fat, and a lower percentage of the diet from carbohydrate than unre-
strained individuals—a pattern identical to recent (Dreon et al., 1988; Romieu et al, 1988) evaluations of obesity. The results of the prospective analyses indicated that dietary restraint, controlled for demographic and energy balance variables, was clearly unrelated to changes in body weight in men. Higher body mass was associated with lower weight gain in men. No such relationship was observed in women. Of more interest, there was evidence that dietary restraint was positively related to changes in body weight in women. It should be noted that the regression equation controlled for the influence of other variables in the model. As such, the statistical procedure shows the independent effect of restraint and body mass, increasing confidence that a prospective relationship may indeed exist between dietary restraint and weight gain. The finding that higher body mass is associated with lower weight gain in men is consistent with other prospective studies of weight gain in adults (Klesges, Klesges, Haddock, & Eck, 1992; Williamson et al, 1991). Probably the most parsimonious explanation is that of a ceiling effect: Obese men are already at the extreme end of the weight distribution. The lack of a relationship between body mass and weight gain in women, however, is intriguing and is perhaps due to the increased prevalence of dieting in women relative to men (Klesges, Mizes, &
1 While it is noteworthy that baseline weight is positively related to weight gain in men and negatively related in women, this finding is uninterpretable because weight is uncorrected for height. Thus, body mass index, which is a measure of relative weight corrected for height, was used to assess the effects of relative overweight on weight gain.
672
R. KLESGES, T. ISBELL, AND L. KLESGES
Table 2 Cross-Sectional Predictors of Body Mass
Measure
% unique variability explained
(
Significance
27.94 1.87
7.222 1.547
Copyright 1992 by the American Psychological Association, Inc. 002l-843X/92/$3,00
Relationship Between Dietary Restraint, Energy Intake, Physical Activity, and Body Weight: A Prospective Analysis Robert C. Klesges and Terry R. Isbell
Lisa M. Klesges
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Center for Applied Psychological Research Memphis State University
Department of Biostatistics/Epidemiology University of Tennessee, Memphis
Dietary, activity, and body weight differences in high- and low-restrained eaters and the independent impact of dietary restraint on body weight change were evaluated in 287 adults (141 men, 146 women) followed over a 1-year period. Analyses of measures of energy intake, physical activity, and dietary restraint indicated that high-restrained eaters did not differ in physical activity compared to low-restrained eaters at baseline but were ingesting significantly fewer kcal/lb and a higher percentage of the diet from fat. Body mass was significantly greater in both high-restrained men and high-restrained women than in their low-restrained counterparts. Regression modeling procedures revealed that weight and body mass at baseline were related to weight gain in men. In contrast, weight gain in women was predicted by baseline weight and higher restraint scores. These results indicate that dietary restraint is associated with weight gain in women but not in men.
There is tremendous interest in determining the factors associated with eating behavior, weight gain, and the etiology maintenance of obesity as it contributes to a number of physical health problems (Van Itallie, 1979). Additionally, given the apparently increasing incidence of other eating disorders such as anorexia and bulimia (Stunkard & Stellar, 1984), several psychological theories of eating behavior have been developed. Partly on the basis of the early internality-externality theory of obesity (Nisbett, 1972; Schachter, 1968), Herman and Mack (1975) have developed the more contemporary construct of restraint. According to their original formulation, eating patterns are influenced by both internal physiological cues that prompt the desire for food and cognitively mediated factors to resist the temptation to eat (or to overeat). The two major hypotheses regarding restrained eating were (a) when the self-control of restrained eaters was somehow violated, eating increased, and (b) restraint variations underlie obese/normal-weight differences in eating behavior. As a result of more than a decade of research (see Ruderman, 1986, for a review), there appears to be support for the hypothesis of counterregulatory eating among restrained eaters, at least in laboratory settings. That is, studies have noted that if restrained eaters are forced to "violate" their diet by eating highcaloric foods (Herman & Mack, 1975) or foods they think are high caloric (Polivy, 1976) or by drinking alcohol (Polivy &
Herman, 1976a, 1976b), their subsequent laboratory consumption of food is increased. Unrestrained (or nondieting) subjects, on the other hand, appropriately regulate for the high-caloric food or drink by reducing their laboratory consumption of food (Herman & Mack, 1975). Although the data support a counterregulatory mechanism in restrained and a regulatory mechanism in unrestrained eaters, the data do not support the hypothesis that differences in level of restraint underlie obese/normal-weight differences in eating behavior (Ruderman & Christensen, 1983; Ruderman & Wilson, 1979). For example, one study (Ruderman & Christensen, 1983) reported that obese people did not behave as restrained eaters. In fact, the opposite was true: The obese subjects ate significantly less after the preload than without it, whereas normal-weight people ate similar amounts both with and without the preload. Restrained overweight people did not counterregulate; they regulated their food intake better than normal-weight people. There are at least three possible reasons why a relationship between dietary restraint, via the mechanism of increased dietary intake as a result of disinhibition, and obesity has not been observed. One explanation is that there may not be stable differences in the eating patterns of obese compared with normal-weight individuals. The vast majority of studies of nutrition and obesity have failed to find overweight/normal-weight differences on measures of dietary intake (e.g., Berenson, Blonde, & Farris, 1979; Huenemann, 1974), and if differences were found (e.g., Johnson, Burke, & Mayer, 1956), they appeared in a paradoxical direction, with obese subjects eating less. Although these early studies can be criticized on methodological grounds (Wooley, Wooley, & Dyrenforth, 1979), more recent investigations using sophisticated methodologies and large sample sizes (e.g., Braitman, Adlin, & Stanton, 1985, N = 6,219) also have failed to find differences in habitual levels of dietary intake and obesity. A second reason for a lack of relationship between dietary
This study was supported by grants awarded to Robert C. Klesges (HL-36553, HL-45057, HL-46352) by the National Heart, Lung, and Blood Institute. Support was also received from a Centers of Excellence grant awarded to the Department of Psychology, Memphis State University, by the State of Tennessee. The authors would like to acknowledge Mary Lou Klem for her helpful suggestions on this manuscript. Correspondence concerning this article should be addressed to Robert C. Klesges, Center for Applied Psychological Research, Department of Psychology, Memphis State University, Memphis, Tennessee 38152. 668
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
DIETARY RESTRAINT AND BODY WEIGHT restraint and obesity may relate to the psychometric properties of the restraint scale. Studies that have examined the reliability of the restraint scale have found the scale's internal consistency to be lower in obese versus normal-weight groups. When used with normal-weight groups, the alpha reliability coefficients of the scale have been found to range from .78 (Klem, Klesges, Bene, & Mellon, 1990) to .86 (Ruderman, 1983). However, the reliability coefficients in obese subjects typically range from .50 to .68 (Johnson, Lake, & Mahan, 1984; Klem et al., 1990; Ruderman, 1983). Thus, it may be that there are indeed restrained obese eaters who counterregulate when disinhibited, but they respond differently to the restraint scale than do normal-weight restrained eaters. A third and perhaps most important reason why previous studies have failed to find a relationship between body weight and restraint relates to the cross-sectional nature of most designs. Obesity and weight gain are not static phenomena; they are processes that develop (usually gradually) over time. Moreover, in any cross-sectional survey of obesity, it is conceivable that subjects are either (a) becoming more overweight, (b) maintaining their obesity, or (c) losing weight. One would expect differences in dietary intake between overweight and normalweight subjects if overweight subjects were becoming more obese or if overweight subjects were trying to lose weight (i.e., might have lower intakes than normal-weight subjects; Klesges & Hanson, 1988). It is clear that longitudinal investigations determining the role of dietary restraint on prospective weight gain are needed. In the only longitudinal study to date (Klesges, Klem, Epkins, & Klesges, 1991), 305 adults (98 men, 207 women) were administered the restrained eating questionnaire, and self-reported height, weight, gender, race, and age were collected. Subjects were recontacted 2| years later and were reassessed on these same variables. Results indicated that restrained eating scores showed a high degree of consistency over a 2j-year period (intraclass correlation = .74). Over time, men gained more weight than women, and normal-weight subjects gained more weight over time than overweight subjects. No relationship was found between dietary restraint and weight gain over time. Unfortunately, this investigation relied on self-reports of height and weight and, most important, assessed neither energy intake nor physical activity as potential control variables in the prediction of changes in body weight. Thus, the purpose of the current investigation was to prospectively evaluate the role of dietary intake, physical activity, and dietary restraint on changes in body weight over a 1 -year period of time in a large cohort of adult men and women. By simultaneously evaluating dietary intake, physical activity, and dietary restraint, the independent contribution of each variable to weight gain, in both men and women, can be assessed. Method Subjects The cohort at baseline included Caucasian adults (141 men, 146 nonpregnant women) who participated in a longitudinal assessment of cardiovascular risk factor acquisition in parents and their young children. Average initial age of the participants was 35.7 years (SD - 4.51, range = 26-53 years); average initial weight for men was 191.29 Ib
669
(SD = 34.56), and for women, 148.62 Ib (SD = 32.27). After initial assessment, subjects were retested 1 year later. There was loss to attrition in the cohort of 13%, which resulted in a sample of 250 subjects (123 men, 127 women). Procedure Subjects were recruited through response forms distributed to local physicians' offices, day-care centers, and churches. Subjects were also recruited by advertisements placed in local newspapers, on local radio stations, and by recommendations from subjects already participating in the study. Data collected for this study were part of a large, longitudinal evaluation of cardiovascular risk factors in children (obesity onset and accelerated blood pressure increases) ages 3 to 4 years (at entry) and their parents. Families were intact, and the target child had to be the biological offspring of the parent. Approximately half of the children recruited were overweight at entry into the study, with the other half being normal weight. After agreeing to participate in the study, individuals received self-report questionnaires (described later) and consent forms to be filled out and signed. Subjects then participated in a laboratory assessment during which height, weight, physical activity, and restraint measurements were obtained. Subjects were recontacted after 1 year and brought to the laboratory, where dietary intake, height, and weight were obtained again. Measures Height and weight were measured by trained research assistants. Height was recorded to the nearest half-inch and weight to the nearest two-tenths of a pound. A hospital grade scale (Model #230 Health O Meter, Continental Scale Corporation, Bridgeview, IL), accurate to ±0.2 Ib and 0.25 in., was used for this procedure. Height and weight measurements were used to estimate body mass (kg/m2), which is a very good estimate of adiposity (Straw & Rogers, 1985). Dietary intake. Long-term assessment of dietary intake was obtained by the 1 -year Willett Food Frequency Questionnaire (Willett et al, 1985, 1988; Willett, Reynolds, Hoehner-Cottrell, Sampson, & Browne, 1987), which is a 61-item, self-administered questionnaire. The advantage of a long-term food frequency measure is its ability to take into account week-to-week and seasonal variations in dietary intake. Moreover, the Willett questionnaire strongly correlates to repeated dietary recalls (Willett et al, 1985) as well as daily dietary records administered for an entire year (Willett et al, 1988). For example, in a recent study (Rimm et al, 1992) of 127 male health professionals age 40-75 years, 1-year test-retest reliabilities (depending on the nutrient) ranged from .56 to .80. This study also compared the Willett questionnaire to the criteria of a 14-day dietary record and reported validity coefficients from .37 to .92. Administration of this measure is very straightforward, and subjects were able to direct questions about specific items to trained nutritionists. After subjects completed the food frequency, all items were checked to eliminate missing data. Subjects who failed to complete items were contacted, and missing data were obtained. All data were verified and analyzed using nutrient tapes provided by the Willett research group. Because this measure is retrospective over the previous year, the food frequency measure was obtained at the 1-year follow-up. Physical activity. Subjects were also administered the Baecke physical activity questionnaire, a factor-analyzed scale comprising 16 items that represent physical activity at work, sport, and nonsport leisuretime physical activity (Baecke, Burema, & Fritters, 1982). These items yield scores ranging from 1 (highly sedentary) to 5 (highly active in work, sport, and nonsport leisure time activity). The Baecke scale has been used extensively in recent years and has demonstrated a highly acceptable degree of reliability (3-month test-retest for the scales averages
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
670
R. KLESGES, T. ISBELL, AND L. KLESGES
.81) and construct validity (Baecke, van Stavern, & Burema, 1983; Washburn & Montoye, 1986). A recent study (Klesges et al, 1990) evaluated self-reports of physical activity against the criteria of surreptitious observation. Subjects were 44 adults who were required to engage in an hour of their preferred physical activity. During this period of time, subjects' physical activity levels were surreptitiously monitored by a trained observer and compared with self-reported estimates of physical activity obtained later. Results indicated that subjects were reasonably accurate in recalling their physical activity levels (average r=.62). Restraint. Subjects completed the Revised Restraint Scale (Herman & Polivy, 1980), which has been found to have generally high reliability, although its internal consistency varies as a function of the proportion of obese individuals in the sample (Johnson et al., 1984). When used with normal-weight groups, the alpha reliability coefficients of the scale have been found to range from .79 (Johnson et al., 1984) to .86 (Ruderman & Christensen, 1983). When administered to obese individuals, however, the alpha coefficients have typically been in the .50 range (Johnson et al, 1984; Ruderman & Christensen, 1983). A recent study (Klem et al, 1990) examining the impact of several demographic variables (e.g., race, gender, weight, and marital status) on psychometric characteristics of the scale found internal reliabilities of the total scale and its two subscales to be similar and generally high across the categories surveyed (alpha for total scale = .78).
Results Variables Assessed and Data Reduction Strategies The demographic and energy balance variables entered into the equation were based on previous reviews of the literature on the correlates of changes in body weight (e.g., Klesges & Hanson, 1988). As such, the demographic variables of age, sex, baseline weight, and body mass served as predictors for the prospective analyses. Both energy intake and levels of physical activity were also entered. Thus, the impact of dietary restraint was evaluated after adjustment for these important demographic and energy balance variables. Dietary intake was computed as average intake of total kilocalories per day, with fat and carbohydrate intake computed as a percentage of total kilocalories. For the multivariate analysis, kilocalories were adjusted for body weight and are expressed as kilocalories per pound of body weight. Cross-Sectional Analyses A multivariate analysis of variance (MANOV\) design was analyzed to determine differences in initial levels of dietary intake, physical activity, and body mass index between restrained and unrestrained eaters and for the possibility of a Restraint X Sex interaction. These results are summarized in Table I. Men with restraint scores of 14 or higher and women with restraint scores of 17 or higher were categorized as highrestrained individuals, representing a division of groups at the median. The MANOVA revealed no significant multivariate effect for a Restraint X Sex interaction, approximate F(275) = 1.06, p = .396, while the main effects for sex, approximate F(275) = 18.14, p < .0001, and restraint, approximate F(275) = 10.65, p < .0001, were statistically significant. Univariate analyses of variance for the restraint main effect revealed significant differences between high and low restraint categories for body mass index, F(\, 283) = 79.39, p < .0001;
energy ingested per pound of body weight, F(l, 283) = 23.84, p < .0001; percent of energy as fat, F(l, 283) = 12.95, p < .0001; and percent of energy as carbohydrate, F(l, 283) = 16.01, p < .0001. As can be seen in Table 1, high-restrained individuals had higher body mass compared to low-restrained individuals. In addition, high-restrained individuals ingested less energy per pound of body weight than low-restrained individuals. As is commonly observed in subjects with higher relative weights (Dreon et al, 1988; Romieu et al, 1988), high-restrained individuals had a dietary composition lower in carbohydrate and higher in fat related to higher body fat than low-restrained individuals. There were no differences in level of activity between restraint categories. Sex differences were also found to be marginally significant on univariate analyses of energy intake per pound of body weight, F(l, 283) = 3.77, p = .053, and significant for body mass, F(l, 283) = 19.62, p < .0001; sporting activity, F(l, 283) = 59.82, p < .0001; and leisure activity, F(l, 283) = 13.22, p < .0001. From Table 1 it can be seen that women had lower relative weights, higher energy intakes per pound of body weight, lower levels of sport activity, and higher levels of leisure activity compared with men. Cross-sectional predictors of body mass for both men and women are presented in Table 2. As Table 2 indicates, restraint scores were strong predictors of body mass in both men and women, with increasing restraint scores associated with higher body mass. Prospective Analyses Predictions of weight gain were investigated by means of a regression analysis with weight change from baseline to the 1-year follow-up as the dependent variable and independent variables of weight at baseline, body mass index, restraint score, age, total energy intake in kilocalories, percentage of daily kilocalories ingested from fats, percentage of daily kilocalories from carbohydrates, and work, sport, and leisure physical activity levels. Two-way interaction terms between sex and all other variables were also included in this analysis. The full regression model, F(19, 230) = 1.309, p = . 179, predicted 9.76% of the total variability in weight change. Because several of the interaction terms with gender were found to be significant in this model, follow-up analyses were conducted by submitting separate regression models for men and women. The results of these sex-specific follow-up models can be found in Table 3. For men, 13.02% of the total variability in their weight change was explained, F(10,112)= 1.68, p = .0948, a marginally significant amount. Significant predictors of weight change in men were baseline weight and body mass, whereas restraint, energy intake, and physical activity were of no statistically significant predictive value. Weight gain was higher with lower baseline body mass. Interestingly, a marginally significant, counterintuitive relationship was found with lowered percent fat intake associated with increased weight gain. For women, 21.22% of the total variability in weight change was explained, F(10,116) = 3.12, p = .0015. Significant predictors of weight change in women were baseline weight and re straint. Weight gain was associated with higher restraint score,-
671
DIETARY RESTRAINT AND BODY WEIGHT Table 1 Cross-Sectional Relationships Between Dietary Intake, Physical Activity, Restraint, and Body Weight
Men
This document is copyrighted by the American Psychological Association or one of its allied publishers. This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
High restraint (n = 68)
Women Low restraint (n = 73)
High restraint (n = 76)
Low restraint (n = 70)
Variable
M
SD
M
SD
M
SD
M
SD
Age" Body massa-b (kg/m2) Energy intake (kcal)8-1" per pound of body weight Intake from fat (% of diet)" Intake from carbohydrate*1 (% of diet) Level of work activity (1 = highly sedentary, 5 = highly active) Level of sport activity (1 = highly sedentary, 5 = highly active) Level of nonsport leisure" activity (1 = highly sedentary, 5 = highly active)
36.74 29.16
4.94 4.19
36.69 24.73
4.69 2.91
35.37 26.96
4.33 5.01
35.13 22.50
3.95 4.47
9.49 34.87
3.54 5.60
11.70 32.53
4.90 5.67
10.27 35.26
3.84 5.16
12.82 33.06
4.09 4.90
48.02
6.93
50.73
6.64
47.66
6.08
51.06
6.21
2.57
0.59
2.65
0.70
2.68
0.47
2.65
0.53
2.52
0.68
2.56
0.67
2.01
0.51
1.98
0.49
2.28
0.58
2.33
0.54
2.51
0.59
2.59
0.58
° Denotes a significant difference (p < .01) between men and women. b Denotes a significant difference (p < .01) between high-restrained and low-restrained subjects.
In contrast to the results with men, initial body mass was not a predictor of weight gain in women.1 Finally, energy intake was a marginally significant predictor of weight change in women.
Discussion The results of the current investigation indicate that high-restrained eaters have higher body mass and ingest fewer kilocalories per pound of body weight but ingest more energy as fat and less as carbohydrate compared to low-restrained individuals. Levels of physical activity were similar between the groups. In the prospective analyses of changes in body weight over a 1-year period in which both demographic and energy balance variables were controlled, dietary restraint was unrelated to changes in body weight in men but was significantly related to changes in body weight in women. The finding that high-restrained eaters were more likely to be overweight is consistent with several other investigations in the literature (e.g., Klem et al., 1990; Ruderman, 1986). However, this is the first investigation showing that restrained eaters maintain higher body weights despite lower levels of energy intake per pound of body weight and similar physical activity. The most likely explanation for this finding is that obese people are more likely to be restrained. In support of this notion, the pattern of dietary intake in restrained eaters is remarkably similar to that of obese people (Dreon et al., 1988; Romieu et al., 1988). That is, the restrained individuals had a lower total dietary intake, a higher percentage of the diet from fat, and a lower percentage of the diet from carbohydrate than unre-
strained individuals—a pattern identical to recent (Dreon et al., 1988; Romieu et al, 1988) evaluations of obesity. The results of the prospective analyses indicated that dietary restraint, controlled for demographic and energy balance variables, was clearly unrelated to changes in body weight in men. Higher body mass was associated with lower weight gain in men. No such relationship was observed in women. Of more interest, there was evidence that dietary restraint was positively related to changes in body weight in women. It should be noted that the regression equation controlled for the influence of other variables in the model. As such, the statistical procedure shows the independent effect of restraint and body mass, increasing confidence that a prospective relationship may indeed exist between dietary restraint and weight gain. The finding that higher body mass is associated with lower weight gain in men is consistent with other prospective studies of weight gain in adults (Klesges, Klesges, Haddock, & Eck, 1992; Williamson et al, 1991). Probably the most parsimonious explanation is that of a ceiling effect: Obese men are already at the extreme end of the weight distribution. The lack of a relationship between body mass and weight gain in women, however, is intriguing and is perhaps due to the increased prevalence of dieting in women relative to men (Klesges, Mizes, &
1 While it is noteworthy that baseline weight is positively related to weight gain in men and negatively related in women, this finding is uninterpretable because weight is uncorrected for height. Thus, body mass index, which is a measure of relative weight corrected for height, was used to assess the effects of relative overweight on weight gain.
672
R. KLESGES, T. ISBELL, AND L. KLESGES
Table 2 Cross-Sectional Predictors of Body Mass
Measure
% unique variability explained
(
Significance
27.94 1.87
7.222 1.547
