56 Chapter 2 Moderator analyses Given the significant between- and/or within-study variation in mean effect sizes of outcomes (Table 3), moderator analyses were conducted to explore heterogeneity of effect sizes in each main outcome. In addition, three-level forest plots (Figure 2) were generated to provide an overview of the variability in mean effect sizes per study (Fernández-Castilla et al., 2020). None of the outcome, sample, intervention, and study characteristics significantly moderated the effect size of the primary outcome social network. Several characteristics significantly moderated effect sizes of the following four secondary outcomes: (1) general functioning, (2) mental health treatment adherence, (3) psychiatric functioning, and (4) substance use. First, assessment time significantly moderated the effect size of general functioning (F(1,27) = 6.307, p = 0.018), showing larger effect sizes on follow-up assessments than on post-assessments. Furthermore, age significantly moderated the effect size of general functioning (F(1,30) = 4.731, p = 0.038), indicating larger effects in younger patients. Second, assessment time (F(2,67) = 3.902, p = 0.025) and personal network involvement (F(1,68) = 4.574, p = 0.036) significantly moderated the effect size of mental health treatment adherence. Effect sizes on post-assessments were significantly larger than those on follow-up assessments. Further, effect sizes were significantly larger when personal network members were not involved in the intervention. Third, sex (F(1,78) = 6.402, p = 0.013) significantly moderated the effect size of psychiatric functioning. The effect sizes of samples with relatively more females were significantly larger than those with more males. Fourth, outcome subcategories significantly moderated the mean effect size of substance use (F(2,100) = 8.111, p < 0.001). The effect size for abstinence was significantly larger than for the quantity of substance use. Furthermore, sex (F(1,118) = 20.151, p < 0.001) and intervention duration (F(1,118) = 5.049, p = 0.026) significantly moderated the effect size of substance use. Effect sizes of samples with relatively more females were significantly larger compared to samples with respectively more males and longer interventions showed larger effect sizes. Additionally, moderator analyses showed that the risk of bias did not significantly moderate the effect sizes of outcome categories. A complete overview of moderator analyses results is presented in Appendix A, Table 3.1-3.7. Next, models with multiple moderators were analyzed to examine the unique effect of significant moderating variables, see Table 4. In general, results show that most moderating variables were not significant after adjusting for the other variables in the model. Specifically, only assessment time significantly moderated the effect size of general
RkJQdWJsaXNoZXIy MjY0ODMw