Esther Mertens

38 | Chapter 2 actively involves the students in the lessons and the extent to which the trainer can activate students physically (i.e., exercises) and mentally (i.e., reflection; e.g., “Do students respond to questions of the trainer?”). Adaptations are clear deviations from the manual. These can be adaptations to the exercises, structure of the exercise, instructions and adding steps to an exercise (e.g., “What percentage of the exercises of the lesson are adapted?”). Statistical analyses The power calculations are based on the N:q rule for structural equation models (Kline, 2015). This rule states that for each free parameter (q) 10 to 20 participants (N) are needed. We took the conservative approach by taking 20 participants per free parameter for our power calculations. In our multigroup LGC model there are 9 free parameters per condition, 36 free parameters in total. Thus a total sample of 720 participants is needed for our analysis, 180 participants in each condition. Since not only students can drop out but also classes (about 20 students) and schools (about 60 to 90 students), we will include three to four schools per condition. Missing data will be handled in M plus . In our data, students are nested in classes which are nested in schools. Therefore, we will examine whether there is significant intra-class correlation on one of the levels (i.e., school, and class) and we will calculate the design effect. Each level with a design effect larger than 2.0 will be modeled in the analyses which allows us to correct for the nested data, that is multilevel analyses (Muthén & Satorra, 1995). The first aim is to examine the effectiveness of R&W in the conditions which differ in the number of parties involved in the intervention. This will be examined using an analysis of covariance (ANCOVA) for the outcomes of socio-emotional adjustment and social safety, in case the design effect is smaller than 2.0. The dependent variables will be the post-measurements after the second year (8th Grade), the independent variables the condition, and the covariates the premeasurements (7th Grade). If needed, due to large design effects, multilevel regression analyses will be used (This also holds for the other aims). Then, we will analyze the trajectories of change in socio- emotional adjustment and social safety during R&W with multigroup Latent Growth Curve (LGC) modeling in M plus . We will examine if these trajectories of change differ significantly between the four conditions. The second aim, the effect of potential moderators on the effectiveness of R&W on socio-emotional adjustment and social safety, will be examined using ANCOVAs for categorical moderators and regression analyses for the continuous moderators. The interaction effects of the concerned outcome measure with the student, trainer or parent characteristics will be added as an interaction term. The third aim, studying the working mechanisms of R&W, will be examined by analyzing multiple mediators. We will analyze whether the R&W intervention