At Time 1 (T1; baseline), 1297 adolescents (58.2% girls) aged 13–20 years (M 16.54, SD 0.90) with BMIs ranging from 12.63 to 41.21 (M 22.30, SD 3.77) participated in the study, of whom 33 (2.6%) adolescents were underweight, 888 (69.1%) had normal body weight, 257 (19.8%) were overweight, and 107 (8.2%) obese (calculated according to WHO growth reference for children and adolescents; de Onis et al. ). At Time 2 (T2; 11 months later), a total of 911 (52.3% girls) adolescents aged 14–20 years (M 17.19, SD 0.93) with BMIs ranging from 14.24 to 39.66 (M 22.13, SD 3.97) participated in the study. At T2, 33 (3.6%) participants were underweight, 681 (74.8%) had normal body weight, 145 (15.9%) were overweight, 52 (5.7%) or obese. Overweight (n = 257) and obesity (n = 107) were identified as an exclusion criterion in the study, since the mechanisms related to their higher weight might be different than in the groups with underweight or normal weight. All participants were white. The majority (64%) lived in urban areas, with 36% living in rural areas.
The total attrition rate was 29.8%. Missing data from those who dropped out at T2 were imputed. Therefore, data collected from N = 1042 adolescents (63.8% girls) aged 13–20 years (M 16.54, SD 0.88) with BMIs ranging from 12.63 to 24.89 (M 22.57, SD 1.98) were included in the analyses.
The study was conducted in 16 public middle and high schools in Central and Eastern Poland. For all participants T2 data were collected at 11 months after T1. All potential respondents lived with their parents (98.9%) or other legal guardians (1.1%) at T1 and T2. Participants and parents of individuals younger than 18 years provided informed consent before the study. Adolescents were informed about the objectives, the procedure and the possibility of refusing to participate in the study (in this case access to school’s library or dayroom was provided). Personal codes were assigned to secure anonymity and identification across the measurement points. Participants were asked to provide their data referring to their nutrition behaviors and body satisfaction. At T1, participants completed a questionnaire measuring their body satisfaction, and at T2 participants (70.2% of those who completed T1) filled in a questionnaire regarding their eating behaviors. At both T1 and T2, participants have their height and weight measured. Researchers were available for consultations after the study completion and made multiple efforts to reduce attrition. The study was approved by the Institutional Review Board and Ethics Committee at University of Social Sciences and Humanities, Poland. The procedures of the study were described in more detail in a paper by Zarychta et al. .
The hypotheses in the present study are tested using two measurement points to establish temporal precedence . Thus, the independent variable was measured at T1, and the mediators and dependent variable at T2.
Body satisfaction (T1)
Body satisfaction variable consisted of seven items, based on The Multidimensional Body-Self Relations Questionnaire’s Appearance Evaluation Subscale (MBSRQ; Cash ). The measure assesses feelings of physical attractiveness or unattractiveness with high scores indicating higher satisfaction with one’s appearance and low scores indicating general unhappiness with one’s appearance. In order to assess it, the respondents were asked to read seven statements (e.g., “I like my looks just the way they are”, “Most people would consider me good-looking” and “I like the way I look without my clothes on”) and decide how much each statement pertains to them. The responses ranged from 1 (definitely disagree) to 5 (definitely agree).
Healthy eating (T2)
In order to evaluate healthy eating, two questions adopted from Lally, Bartle and Wardle  were used: “How often did you eat a portion of fresh fruit in the last two weeks?” and “How often did you eat a portion of vegetables in the last two weeks (fresh, boiled or fried without fat)?”. The portion was defined as the amount fitting into a cupped hand. The responses were given on a 6-point scale, ranging from 1 (once a week or less) to 6 (four or more times a day).
Unhealthy eating (T2)
In order to evaluate unhealthy eating, adolescents answered two questions, adopted from Lally, Bartle and Wardle : “How often did you eat fatty foods (e.g., pizza, chips, foods with dressings) in the last two weeks?” and “How often did you eat sweets (e.g., chocolate bars or wafers, cakes) in the last two weeks?”. The responses were given on a 6-point scale, ranging from 1 (once a week or less) to 6 (four or more times a day).
Restrictive dieting (T2)
In line with research conducted so far on overall diet quality (cf. Loftfield et al. ), the measurement of dietary restrictions was based on a single item adopted from MBSRQ : “I am on a restrictive weight-loss diet”. The responses ranged from 1 (definitely disagree) to 5 (definitely agree).
Body weight and height (T1 and T2)
Biometric measures were assessed with standard medically approved telescopic height measuring rods and floor scales (scale type: BF-100 or BF-25). Age- and sex-specific BMI percentiles were calculated with WHO AnthroPlus macro , based on the WHO growth reference  for children and adolescents. BMI z-scores were calculated and used as independent variables in both analyses. Also, two weight status categories were created based on BMIs’ SD cut-offs (− 1—underweight [less than or equal to 2 SD], + 1—normal weight) .
Data were analyzed using SPSS version 24. Multiple mediation analysis was performed to test the relation between body satisfaction, healthy eating, unhealthy eating or restrictive dieting, and adolescents’ BMIs with the use of PROCESS macro (Model 4) with 10,000 bootstraps . Analyses were conducted accounting for the covariates (T1 BMI, age and sex [coded as 1 for males and 2 for females]). Results are presented using two types of coefficients: (1) the regression coefficient for each parameter (see Fig. 1), and (2) the indirect effect coefficient (B) for each indirect pathway between the independent variable (T1 body satisfaction) and the dependent variable (T2 BMI), accounting for respective mediators (see Fig. 1; Table 2). Furthermore, sex, age, and BMI (T1) were included into each regression model as predictors of the dependent variable, BMI at T2. This way potential confounders and BMI at T1 were controlled for.
Additionally, moderated multiple mediation analysis was conducted using PROCESS macro (Model 58). In particular, these analyses tested if the associations between (1) the independent variable and the mediator, and (2) the mediator and the dependent variable are moderated by participants’ weight status.
In the current study the independent variable (IV) was body satisfaction (T1); the dependent variable (DV) was BMI z-score measured at T2; the mediators were healthy eating (T2), unhealthy eating (T2) and restrictive dieting (T2); moderator was weight status (T1) coded as − 1 for underweight and + 1 for normal body weight.
In order to establish temporal precedence , the variables were measured at different time points (T1 and T2). Missing data were imputed with multiple imputation method . A total of 5.9% of the completers’ data were missing. The attrition analysis is presented below.