Participants
Individuals responded to print advertisements soliciting volunteers for research on the use of exercise and nutrition instruction for weight loss to be completed in YMCA centers in the southeastern U.S. Inclusion criteria were: a) minimum age of 21 y, b) BMI of 35–50 kg/m2, and c) no regular exercise (less than a self-reported mean of 30 min/week). Exclusion criteria were: a) current enrollment in a commercial or medical weight-loss program, b) soon-planned or current pregnancy, and c) taking medications prescribed for weight loss or a psychological condition that might affect survey responses (e.g. anxiety disorder, depression). A signed statement of adequate health to participate was required from a physician. Approval was received from the institutional review board of Kennesaw State University, and written informed consent was appropriately obtained from all participants.
Figure 1 outlines the flow of participants through the recruitment and study processes. There was no significant difference in proportion of women (overall 83%), age (overall 43.0 ± 9.5 y), BMI (overall M = 40.5 ± 4.1 kg/m2), and racial make-up (overall 48% White, 48% African American, and 4% of other racial/ethnic groups) between participants assigned to a treatment of cognitive-behavioral exercise support plus nutrition education (nutrition education group; n = 145) and the same cognitive-behavioral exercise support plus cognitive-behavioral nutrition methods (cognitive-behavioral nutrition group; n = 149) through simple random sampling. Nearly all participants were middle-class.
Measures
Exercise behavior
Volume of exercise was measured by the Godin-Shephard Leisure-Time Physical Activity Questionnaire [31]. It incorporates estimates of metabolic equivalent of tasks (METs), or the physiological energy cost based on physical activity intensity. Respondents enter weekly frequencies of strenuous (“heart beats rapidly”, e.g. running), moderate (“not exhausting”, e.g. fast walking), and light (“minimal effort”, e.g. easy walking) exercise for “more than 15 minutes” per session. The responses are multiplied by 9, 5, and 3 METs, respectively, and then summed. Reported test-retest reliability over 2 weeks was .74 [32]. Construct validity was previously indicated by significant correlations with accelerometer and maximum volume of oxygen uptake assessments [33, 34].
Eating behavior
Intake of number of fruits and vegetable servings “in a typical day” (“looking back over the last month”) was based on the U.S. Food Guide Pyramid’s descriptions of foods and their corresponding portion sizes, and how to count “mixed foods” (e.g. cereal with fruit; salads) [35]. Responses from the two items (one each for fruits and vegetables) were summed. Reported test-retest reliability over 2 weeks averaged .82, and concurrent validity was indicated through significant correlations of the present measure with lengthier food frequency questionnaires [35]. In pilot research, predictive validity was supported in severely obese adults through a significant inverse relationship (r = -.42, p < .001) between change in this measure and weight change. Research suggests that fruit and vegetable consumption, alone, is an accurate predictor of quality of the diet and overall caloric consumption [36, 37], and in contexts such as the present research, single-item scales may not possess a disadvantage [38].
Self-efficacy
Self-efficacy for controlled eating was measured by the Weight Efficacy Lifestyle Scale [39]. It incorporates items from five factors (negative emotions, availability, physical discomfort, positive activities, and social pressure) (e.g. “I can resist eating even when others are pressuring me to eat”), that are summed for a total score. Responses to the scale’s 20 items range from 0 (not confident) to 9 (very confident). Internal consistency was reported to range from α = .70-.90 [39]. Internal consistency for the present sample was α = .82.
Exercise self-efficacy was measured by the Exercise Self-Efficacy Scale [40]. Five items begin with the stem, “I am confident I can participate in regular exercise when …” (e.g. “I feel I don’t have the time)”, with responses ranging from 1 (not at all confident) to 11 (very confident). Reported internal consistency ranged from α = .76-.82, and test-retest reliability over 2 weeks was .90 [41]. Internal consistency for the present sample was α = .84.
Mood
Mood was assessed by the Profile of Mood States Short Form’s measure of Total Mood Disturbance [42]. It is an aggregate of six subscales (Depression, Tension, Fatigue, Vigor, Confusion, and Anger) incorporating a total of 30, one- to three-word items (e.g. “sad”, “worn out”, “tense”) that are rated from 0 (not at all) to 4 (extremely) based on, “how you have been feeling during the past week including today”. Internal consistency ranged from α = .84-.95, and test-retest reliability at 3 weeks averaged .69 [42]. Internal consistency for the present sample was α = .79-.91.
Self-regulation
For the measurement of self-regulatory skill usage for exercise and self-regulatory skill usage for controlled eating, a scale by Saelens et al. [43], where items are to be based on the content of the intervention presently being used, was adapted. Examples of the 10-item scales are, “I set physical activity goals”, and “I purposefully address my barriers to eating appropriately”, respectively. Responses range from 1 (never) to 5 (often). Reported internal consistency was α = .79 and .81, respectively; and test-retest reliability over 2 weeks was .78 and .74, respectively [22]. Internal consistency for the present sample was α = .76 and α = .79, respectively.
Procedure
Each participant reported to his/her assigned YMCA center, received a group orientation to the study’s processes, and was provided access to the facility for the duration of the study. The cognitive-behavioral exercise support component was identical for each of the two groups. It consisted of a computer-supported protocol of six, 45- to 60-min meetings (approximately monthly) with a trained wellness specialist over 26 weeks [12]. These sessions included an orientation to exercise areas and apparatus, but most time was spent in individual consultation in an office. Long-term goals were identified and broken down into process-oriented short-term goals. Goal progress was tracked graphically on the computer at each meeting. Self-regulatory skills instructions addressed cognitive restructuring, stimulus control, behavioral contracting, and relapse prevention. Exercise modalities (e.g. treadmill; walking on an indoor track) were based on each participant’s preference. Standard exercise recommendations (i.e. 150 minutes/week of moderate cardiovascular activity [16]) were described; however, the benefit from any volume of exercise was also indicated.
The nutrition components differed by group, with one emphasizing education in healthy eating practices, and one emphasizing the use of self-regulation methods to control inappropriate eating. Each had six, 1-hour sessions administered by a certified wellness specialist in group format of 10–15 participants over 12–14 weeks. In the nutrition education group, the standardized protocol [44] included: a) understanding macronutrients and calories, b) healthy recipes, c) menu planning, d) low-fat snacking, and e) stocking a healthy kitchen. In the cognitive-behavioral nutrition group, the protocol included: a) establishing caloric goals based on weight, b) logging foods and associated calories, c) cognitive restructuring, d) relapse prevention training, and e) cues to overeating. In both groups, increasing exercise and fruit and vegetable intake was emphasized during each session.
Wellness specialists that administered the treatments had YMCA and other national health and fitness certifications, and were blind to the purposes the study. Approximately 15% of treatment sessions were monitored for fidelity by study staff members. Deviations from the assigned protocols rarely occurred. However, when indicated, corrective measures were immediately taken by YMCA supervisors in cooperation with study administrators. Assessments were administered in a private area at baseline and week 26.
Data analyses
An intention-to-treat format was incorporated where data from all participants initiating treatment were included in the analyses. The 16% of missing measure scores (all missing at week 26) were imputed using the expectation-maximization algorithm [45, 46]. Statistical significance was set at α = .05 (two-tailed), throughout. Effect sizes are expressed as either Cohen’s d or partial eta-square (η2p) where .20, .50, and .80; and .01, .06, and .14 denote small, moderate, and large effects, respectively. For the planned multiple regression analysis, to detect a small-to-moderate effect (f2 = .08) at the statistical power of .90, α = .05, a minimum of 180 participants was required [47]. Oversampling was conducted to secure the statistical power.
Scores of all measures were approximately normally distributed. Change scores in measures of exercise volume, fruit and vegetable intake, mood, and exercise- and eating-related self-regulation and self-efficacy were calculated as differences between scores at baseline and scores at week 26. As suggested for the present research context [48], change scores were unadjusted for their baseline value. Mixed model repeated measure ANOVAs (time × treatment type) were incorporated to simultaneously assess whether the changes in each variable were significant over 26 weeks, and whether those changes differed across the two treatment types.
Using aggregated data, intercorrelations of scores were computed, and multiple regression models were fitted to separately assess the independence of changes in the measures of self-regulation, self-efficacy, and mood in the prediction of changes in exercise volume and fruit and vegetable intake. Collinearity was tested, and both the variance inflation factors (1.27-1.74) and tolerances (.43-.79) were within accepted limits for regression analyses. Inspection of residual scatterplots indicated homogeneity of variance and linearity in the data.
Mediation models (see Figure 2) were specified using a bias-corrected bootstrapping procedure incorporating 10,000 re-samples [28]. Within this procedure, R2 is used to assess significance of overall mediation, and if the relationship of the predictor and outcome variable (path c) is no longer significant after entry of the mediator (path c′), then complete mediation is identified. Utilizing this mediation analysis method [28], a series of reciprocal effects analyses, based on recent related research [29], were computed that assessed: a) the reciprocal effects of changes in exercise and fruit and vegetable intake, resulting from the two treatment types, b) the reciprocal effects of changes in the three psychosocial variables of interest (i.e. self-efficacy, mood, and self-regulation) and fruit and vegetable change, resulting from change in exercise volume, and c) the reciprocal effects of changes in the three psychosocial variables and exercise change, resulting from change in fruit and vegetable intake. Considering the rationale presented by Marsh and Craven [49], a reciprocal effect was considered to be present when mediation is significant in each of two complementary models having a consistent predictor -- one model where a variable is specified as an outcome, and the other where that same variable is specified as a mediator [29]. For example, in the initial reciprocal effects analysis, treatment type was the predictor in both mediation models where, in the first equation of that analysis, change in exercise volume was the outcome variable and change in fruit and vegetable intake was the mediator. In the second and complementary equation of that analysis, fruit and vegetable intake change was the outcome variable and change in exercise was the mediator. The same procedure was then used to assess the presence of six additional reciprocal relationships that incorporated psychosocial factors that emerged from both social cognitive theory [18, 19] and previous research [22–25]. In the initial three of these reciprocal effects analyses, change in exercise volume was the predictor, fruit and vegetable intake change was the outcome measure, and either change in the measure of self-efficacy, mood, or self-regulation was the mediator in its initial equation, and then was the outcome variable in its complementary equation. In the final three reciprocal effects analyses, change in fruit and vegetable intake was the predictor and volume of exercise change was the outcome variable in its initial equation, with the same pattern of entry of the psychosocial variables as in the previous three reciprocal effects analyses.