Esther Mertens

| 17 General Introduction intervention’s effectiveness, but also for theory building and implementation of the program (Bonell et al., 2012). Contexts and populations are by nature heterogeneous which could influence intervention effects. What works in one context may not work in another context. Similarly, some individuals with certain characteristics may improve, or deteriorate, while the total population might show no or weak changes in the outcome of interest (Bonell et al., 2012; Farrell, Henry, & Bettencourt, 2013; Greenberg & Abenavoli, 2017). Knowing whether characteristics of contexts and participants affect intervention’s effectiveness gives an indication to what extent the intervention effects can be generalized (Rowe & Trickett, 2018). This is especially important in universal interventions as these are typically implemented in a broad context targeting a heterogeneous population (Farrell et al., 2013). In addition, insights in how characteristics of the context and participants affect intervention effects are eminent for theory development. If intervention effects are affected by certain characteristics, this suggests differences in the underlying working mechanism (Kazdin, 2007). The importance of studying under what circumstances and for whom an intervention is effective has been long acknowledged and is even incorporated in the “Standards of Evidence” of the Society for Prevention Research (Flay et al., 2005). However, most studies neglect characteristics of the context and only examine routinely collected characteristics of participants such as sex and ethnicity. Moreover, most studies determine moderators in post hoc analyses rather than based on theory and a priori hypotheses (Farrell et al., 2013; Kazdin, 2007). For instance, Rowe and Trickett (2018) showed in their meta-analysis that only 20 of the 50 included moderation analyses were supported by literature or a priori hypotheses. In the present dissertation, I attempted to overcome these limitations by determining moderators a priori and by examining characteristics of the context as well as characteristics of participants. This enabled me to move beyond moderation analyses based on convenience (i.e., on routinely collected characteristics) and test theoretically relevant characteristics (Kazdin, 2007). As characteristics of the context, I focused on two aspects of intervention dosage. The first dosage characteristic was the ecological width of an intervention, that is, the involvement of multiple systems in the intervention. The social ecological model of Bronfenbrenner (1979) states that behavior is determined by the interactions of multiple systems such as the individual, family, and school systems. In some interventions stronger effects were found when more systems were actively involved in the interventions (e.g., Flay, Graumlich, Segawa, Burns, & Holliday, 2004), while in other interventions no effects of involving multiple systems were found on their effectiveness (e.g., Durlak et al., 2011). In the present dissertation, I manipulated the extent to which systems were involved in R&W resulting in three levels of ecological width, i.e., 1) “Light condition”: Only R&W trainers (core teamof teachers) were involved in the intervention (i.e., class system), 2) “Standard condition”: The entire teaching staff of the school was involved (i.e., class and school systems), 3) “Plus condition”: The entire teaching staff of the school and parents were involved (i.e., class, school, and family 1