Esther Mertens

| 153 General Discussion engagement, adjustments to the program, monitoring change in a control group, and program reach. Based on both R&W trainers’ self-reports and observations by R&W experts, the intervention seemed implemented with moderate to high fidelity to the intervention manual indicating that in general all aspects of the intervention were administered. A limitation concerns the extent to which the findings regarding R&W can be generalized. I evaluated the intervention only in Dutch schools and only for prevocational students. Due to differences between countries and student populations, one should be careful with generalizing the intervention effects beyond this specific context and population (Bonell et al., 2012). Thus, more research is needed to replicate my findings in different settings to determine the generalizability of the present results. Furthermore, notwithstanding the strong research design of an RCT and the comparison of the intervention conditions to CAU, i.e., an active control group, the RCT in this dissertation had some limitations. One school dropped-out after randomization but before data collection started and was therefore replaced by another school. This slightly decreased the true RCT design of the study since the new school was placed in a condition rather than randomly assigned. Furthermore, allocation to conditions was conducted at school level, but the data were analyzed on the level of individual students. This is common practice in research examining school-based interventions (e.g., Bonell et al., 2018; Cross et al., 2016) as the interest is in change in individuals rather than in schools. I accounted for clustering at school level in the analyses to prevent underestimation of standard errors and exaggeration of statistically significant findings (Muthén & Muthén, 2017). An exception to these analyses are the analyses conducted for examining modeling and reinforcement as mediators. I analyzed these mechanisms at classroom level rather than the level of individual students since the research questions concerned changes in modeling and reinforcement in the classroom, hence, cluster-level variability (Preacher, Zyphur, & Zhang, 2010). Future research The present dissertation sheds light on the effectiveness of universal school-based interventions, in particular R&W, and raises new questions about intervention effects for vulnerable students, about working mechanisms, and about effectiveness of separate components. Can the beneficial effects for vulnerable students, found in the present study, be generalized to other subgroups of vulnerable students? Given that classmates’ modeling and reinforcement were not mechanisms of change, what are other possible mechanisms of change? Are the components associated with stronger, and weaker, intervention effects indeed effective, or ineffective, components? And can school-based interventions be optimized by adding or deleting components? By 7