Does Dieting Lead to Weight Gain? A Four-Year Longitudinal Study of Middle School Girls
Department of Family and Community Medicine, University of Arizona, 1450 N. Cherry Avenue, Tucson, AZ 85724. E-mail: firstname.lastname@example.org
Objective: A majority of the published longitudinal research on children has reported that dieting is related to weight gain at a later point in time. The purpose of this study was to look at weight control behaviors and patterns of weight gain and loss, specifically whether dieting is related to weight gain.
Research Methods and Procedures: Baseline data were collected from 1358 female students in grades 6 to 9 from schools in Hayward, CA, and Tucson, AZ. Data were obtained annually over a 4-year period. Paper-and-pencil questionnaires and height and weight were obtained during the students’ regular classroom periods. Dieting was measured both with the single item, “In the past year, how often have you been on a diet to lose weight?” scored from “never” to “always,” and with a Dieting Behavior Scale including five items on weight control behaviors. Changes in BMI z-scores were analyzed.
Results: On average, girls who reported “never” dieting were most likely to have an increased BMI z-score at the next measurement, and those who reported “always” dieting were most likely to have a decreased BMI z-score. The same pattern was true for the Dieting Behavior Scale.
Discussion: Our finding that dieting and weight gain were not related, independently of initial BMI, does not mean that dieting to lose weight is appropriate, especially among young girls. Additional research is needed both to examine this relationship and to determine exactly what behaviors children are engaging in when they report that they are dieting.
The proportion of the adult population of the United States that is overweight continues to increase (1, 2), and this increase has been paralleled by an increase in the proportion of children who are overweight (BMI percentile ≥ 95th) or at-risk-of-overweight (BMI percentile ≥ 85th) (3, 4, 5). Overweight in childhood and adolescence was once considered a problem because overweight tends to track into adulthood (6). However, there is now concern over the increasing prevalence of weight-related diseases in children (7, 8, 9, 10), and children themselves are reporting dieting to lose weight (11, 12, 13, 14).
There is evidence that dieting is not particularly effective among adults (15, 16) and concern that dieting may actually promote weight gain (17, 18), although there is also disagreement with this view (19). Only a few longitudinal studies of the relationship between dieting and weight change among adolescents and preadolescents have been reported. In a 3-year study of adolescent girls, French et al. (20) found that dieting was not significantly related to changes in BMI or weight fluctuations of 8 lb or more. In a 9-month study of adolescent girls and boys by Stice (21), self-reported current dieting at the first measurement was predictive of a weight loss at the second measurement among girls. The results of two later 4-year studies of adolescent girls by Stice et al. (22, 23) showed that dieting predicted onset of obesity in both samples of girls, although dieting was not significantly related to an increase in relative weight. In a 3-year study of preadolescent and adolescent boys and girls, Field et al. (24) reported that dieting predicted weight gain in both boys and girls, even after adjustment for pubertal development, dietary intake, physical activity/inactivity, and BMI in the previous year. In summary, three of the five published longitudinal studies of the relationship between dieting and weight change among preadolescents and adolescents found dieting to be predictive of weight gain rather than weight loss.
The purpose of this analysis was to examine whether dieting and other weight-related behaviors were related to weight gain and loss among middle-school girls over a 4-year period. Based on perceived analytical problems in the studies reporting a positive relationship between dieting and weight gain, it was hypothesized that dieting would not be related to weight gain when the initial BMI was used as a control.
Research Methods and Procedures
The data reported in this analysis were originally collected for a study of eating disorders among adolescent girls. The methods of data collection have been described in detail in a prior publication (25) but are described briefly below.
Participants were 1358 female students in grades 6 to 9 at baseline, selected from schools in Hayward, CA, and Tucson, AZ, who completed the study questionnaire in English. The school districts were selected because they were diverse and had relatively stable populations. The school districts preferred, and the human subjects committee approved, the use of passive consent. Parents were mailed a letter informing them of the study and letting them know that their child could be withdrawn from the study by calling the school or returning an enclosed letter. Consent letters were mailed out in English and Spanish. In addition, the students were told that they could withdraw from the study at any time.
In the first year of the study, 25% of the girls were in the sixth grade, 27% in the seventh grade, 22% in the eighth grade, and 26% in the ninth grade. Age ranged from 10 to 17 years, with a mean of 12.7 years. The most frequent ethnicity was white non-Hispanic (38%), followed by Hispanic (34%), black (16%), and Asian (5%). All other ethnicities, including multiethnic, were less than 3%.
Data were obtained over a 4-year period. Paper-and-pencil questionnaires and structured clinical interviews were administered annually, and height and weight were also measured once a year during the students’ regular classroom periods. Data were available for 1081 students at the first follow-up, 924 students at the second follow-up, and 825 students at the third follow-up.
Compared with the girls who remained in the study, those lost to follow-up were more likely to be a minority (68% among those lost to follow-up vs. 60% in the study at the first follow-up, p = 0.02; 67% vs. 59% at the second follow-up, p = 0.03; and 66% vs. 58% at the third follow-up, p = 0.01). The girls lost to follow-up were not different in age or BMI z-score at the first follow-up but were older and heavier at the second and third follow-ups (mean age, 12.9 years among those lost at the second follow-up vs. 12.6 years among those still in the study, p = 0.001; 12.8 vs. 12.6 years at the third follow-up, p = 0.001; mean BMI z-score of 0.70 for those lost to follow-up at the second follow-up vs. 0.58 for those still in the study, p = 0.04; 0.70 vs. 0.57 at the third follow-up, p = 0.02). Girls in the study were not different from those lost to follow-up in primary language spoken in the home, dieting status, or residence in California or Arizona.
Attrition in this study was due primarily to relocation to another school district or to another city (84%), as well as chronic absence during data collection (13%) and student and parent refusal (4%).
McKnight Risk Factor Survey IV
The McKnight Risk Factor Survey IV is a revised version of the McKnight Risk Factor Survey III (26). On the basis of test–retest data, a small number of questions were rewritten, added, or dropped from the earlier (III) version for the later (IV) version. The McKnight Risk Factor Survey IV consists of 103 questions that assess demographics, age at onset of menstrual period and dating, appearance appraisal, effect of body changes, confidence, depressed mood, emotional eating, media modeling, concern with weight/shape, parental and peer concern with thinness, teasing, negative life events, perfectionism, changing eating around peers, substance use, support from others, dieting behaviors, “activities that make you feel good about yourself,” participation in athletic activities, and several miscellaneous items. Most items are scored on a five-point scale ranging from 1 (never) to 5 (always). Dieting was measured both with the single item, “In the past year, how often have you been on a diet to lose weight?” and with a Dieting Behavior Scale that included five items on how often in the past year girls were 1) trying to lose weight, 2) cutting back on what was eaten to lose weight, 3) skipping meals to lose weight, 4) eating fewer sweets to lose weight, and 5) dieting. This scale had a Cronbach's α of 0.89.
Standing height was measured to the nearest millimeter using a portable direct-reading stadiometer. Measurements were taken with shoes removed and the body positioned so that heels and buttocks were against the vertical support of the stadiometer. Weight was measured to the nearest 0.10 kg using a digital scale. BMI was computed using the formula weight in kilograms/(height in meters)2. z-Scores for BMI were computed using the Centers for Disease Control's formulas for sex and age in months for each year of the study (27).
The cross-sectional analysis included contingency table analysis for categorical variables and comparison of group means for continuous variables using one-way ANOVA. Comparison of BMI z-scores was done with a paired Student's t test within each category of dieting, which looked at the significance of the individuals’ change from 1 year to the next. Looking at change controlled for the initial year's BMI z-score. Stepwise forward logistic regression was used to examine the predictors from Year 1 for being overweight (>95th percentile) in Year 2.
The distribution of scores for the single dieting item “dieted in the past year” was quite consistent in each of the 4 years of measurement. A majority of the girls responded that they had not dieted (range, 50% to 53%), whereas very few indicated that they were “always” dieting (range, 1% to 3%). The proportion in each category declined as frequency of dieting increased. BMI z-scores for the first year ranged from −3.1 to 3.0, with a mean of 0.62 ± 1.02 (standard deviation).
In Year 1, the Diet Behavior Scale, which was the sum of responses for the five items, ranged from 5 to 25, with a median of 9 and an interquartile range from 5 to 14. Like the single dieting item, this scale was distributed in much the same way in each of the 4 years. In each year, the range was from 5 to 25 and the median was 9, except in Year 3, when it was 8. The interquartile range in Years 2 through 4 was 5 to 13.
Table 1 shows the cross-sectional relationship between the frequency of dieting and BMI z-scores for each year. In Years 1 through 3, the relationship from the category of “never” to the category of “always” is reverse U-shaped, with the highest mean BMI z-score in either the “sometimes” or “a lot” category. In the fourth year, the relationship is linear, with the lowest BMI z-scores in the “never” category and the highest in the “always” category; however, there are only three people in the last cell.
Table 1. Cross-sectional relationship between frequency of dieting and BMI z-scores
|Never||0.24 (692)||1.00||0.23 (556)||0.98||0.26 (491)||0.97||0.19 (385)||0.97|
|A little||0.90 (299)||0.87||0.88 (255)||0.87||0.97 (212)||0.85||0.86 (207)||0.89|
|Sometimes||1.17 (200)||0.83||1.16 (152)||0.80||1.03 (130)||0.85||1.12 (111)||0.81|
|A lot||1.09 (125)||0.91||1.22 (74)||0.79||1.12 (74)||0.73||1.22 (50)||0.77|
|Always||0.95 (31)||0.93||1.07 (30)||0.74||0.80 (8)||1.4||1.9 (3)||0.64|
|Total||0.62 (1347)||1.02||0.61 (1067)||1.00||0.61 (915)||0.99||0.58 (756)||1.00|
Data on changes in BMI z-scores by the single dieting item are shown in Table 2. In 2 of the 3 years, girls who reported not dieting had a statistically significant increase in BMI z-scores. Thus, the average BMI z-score for girls who reported never dieting in Year 1 increased from 0.22 in Year 1 to 0.25 in Year 2; those who reported never dieting in Year 2 increased from 0.23 in Year 2 to 0.27 in Year 3, both differences being statistically significant at p = 0.01. In 2 of the 3 years, girls who reported dieting “a lot” or “always” had a statistically significant decrease in BMI z-scores. The data from the Dieting Behavior Scale are shown in Table 3. The lowest quartile represents no dieting behavior in the past year, whereas the highest quartile represents the most frequent dieting behavior. The pattern is consistent with the results from the single item analysis. However, with additional items, the differences are larger. It is important to note that, although some mean z scores declined, on average, girls in all categories did gain weight. In Year 1, weight gain increased linearly (highest quartile, 3.3 kg; next, 3.2 kg; next, 3.8 kg; lowest, 3.9 kg) from Year 1 to Year 2 across quartiles. Thus, the girls reporting the most frequent dieting behavior gained the least weight. The same pattern was evident across quartiles in all years. The average increase dropped in each year; the overall increase was 3.6 kg from Year 1 to 2, 3.0 kg from Year 2 to 3, and 2.0 kg from Year 3 to 4.
Table 2. Change in BMI z-scores by single dieting item
|Diet to lose weight: Year 1||N = 552||N = 231||N = 161||N = 121|
|Diet to lose weight: Year 2||N = 465||N = 208||N = 111||N = 67|
|Diet to lose weight: Year 3||N = 381||N = 169||N = 90||N = 66|
Table 3. Change in BMI z-scores by dieting behavior scale
|Dieting Behavior Scale: Year 1||N = 290||N = 252||N = 266||N = 245|
|Dieting Behavior Scale: Year 2||N = 261||N = 223||N = 172||N = 198|
|Dieting Behavior Scale: Year 3||N = 236||N = 126||N = 206||N = 136|
Additional analyses were carried out to replicate prior research on dieting and the onset of obesity (22, 23). A stepwise binary logistic regression analysis was used to predict obesity (N = 38) at Year 2 follow-up using frequency of dieting in Year 1 as the predictor, excluding from the analysis those with a BMI already above the 95th percentile. When used without initial BMI z-score as a control, the frequency of dieting in Year 1 was significantly related to being obese at Year 2 (−2 log likelihood = 312.75; odds ratio for “always” compared with “never” was 9.8, with 95% confidence interval 2.9 to 33.0). When Year 1 BMI z-score was included in the model, dieting did not enter the equation (p = 0.55). This same pattern was also found for onset of obesity in Years 3 and 4 (data not shown).
The focus of this analysis of 4-year longitudinal data was the relationship between dieting and weight gain or loss. The results indicated that dieting was not related to increased weight gain among middle-school girls when weight change from initial BMI was analyzed, i.e., when initial BMI was controlled. Using either a single-item measure of dieting or a multi-item measure, weight gain was most likely to occur among those who never dieted, and weight loss was most likely to occur among those who dieted most often. If dieting led to weight gain, one would expect a dose–response relationship, with increasing weight gain as dieting increased. However, this pattern was not found in the sample of girls who were assessed in this study.
These results are consistent with the findings of French et al. (20) and Stice (21) but not with those of Stice et al. (22, 23) or Field et al. (24). The discrepancy with the findings of Stice et al. (22, 23), who found that dieting was related to the onset of obesity, is likely because of their lack of control for initial BMI. Although those who were already over the 95th percentile were excluded, there were likely a number of others who were just below the 95th percentile, who were both more likely to be dieting and more likely to slip over the 95th percentile cut-off point for obesity. We were able to replicate the findings of Stice of a significant relationship between dieting and obesity onset when not controlling for BMI, but found a lack of significance when BMI was controlled. When Stice et al. used controls for initial BMI in an analysis of weight gain, neither the dietary restraint scale nor self-labeled dieting was significantly related to weight gain (22). Field et al. (24) did use controls for initial BMI, among other factors, although the table for girls in which longitudinal results were reported did not indicate that initial BMI was controlled. However, in the study of Field et al., height and weight were self-reported rather than being measured by research personnel. Also, the weight difference between female non-dieters and frequent dieters, although statistically significant, was only slightly more than 0.5 lb, with an even smaller difference between girls who were non-dieters and infrequent dieters.1 A difference this small is probably not clinically significant.
One limitation of this study was that dieting behavior was self-reported and dietary intake was not measured, which would have helped to confirm lower caloric intake. However, Field et al. (24) included a self-report measure of dietary intake in their study but found that greater weight gain among dieters than non-dieters was not explained by differences in caloric intake, percentage of energy from fat or protein, and intake of snack foods. In addition, we did not have an adequate measure of physical activity or sedentary behaviors that could be related to individual dieting behavior. Ideally, measures of caloric intake, physical activity, dieting, and sedentary behaviors would lead to more definitive answers about the relationship between these behaviors and BMI changes.
Our finding that dieting and weight gain were not related, independently of initial BMI, does not mean that dieting to lose weight is appropriate, especially among young girls. Additional research is needed both to examine this relationship and to determine exactly what behaviors children are engaging in when they report that they are dieting. More information is also needed about exactly what girls do that leads to the achievement and maintenance of a healthy BMI. However, it would seem from our findings that dieting, in itself, does not promote weight gain among preadolescent and adolescent girls.
There was no funding/outside support for this study.
Girls who had a z change of 0 gained a mean of 9.4 lbs, and those with a z change of 1 gained a mean of 19.7 lbs. Therefore, 1 z unit is equivalent to 10.3 lbs. The difference between non-dieters and frequent dieters was 0.058 of a z, which represents 0.59 lbs (0.058 × 10.3).