86 Chapter 5 example: ‘How well can you motivate uninterested students?’. Participants could give answers on a nine-point Likert scale ranging from (1) ‘nothing’ to (9) ‘great deal’. A higher score on these items indicates a higher amount of teacher self-efficacy (Nie et al., 2010). In this study, the TSES-SF at baseline had a Cronbach’s alpha of 0.90 and a McDonald’s omega of 0.93. Data analyses All analyses were conducted in R (R Core Team, 2022). Before running mediation analyses, Pearson correlations were calculated between all outcome and mediator variables. Mediation models were fitted using structural equation modelling implemented in the Lavaan package (Rosseel, 2012) and followed recommendations by Preacher and Hayes (Hayes & Rockwood, 2017; Preacher & Hayes, 2008). Condition (0=control group, 1=intervention group) was used as independent variable (X) in all models. First, simple mediation models with mindfulness, self-compassion, teacher self-efficacy, or emotion regulation as mediator (M) at post-test and perceived stress, well-being, or classroom climate quality as dependent variable (Y) at follow-up were conducted (Figure 1). In addition to the three previously described mediators in the study protocol (Lensen et al., 2021), we decided to also include teacher self-efficacy based on the literature. Second, three multiple mediation models were conducted including all mediators that had been shown to be significant factors in the simple mediation models. One multiple mediation model was performed per dependent variable. In all models, the mediator at post-test was additionally regressed on baseline scores of the mediating variables to account for possible baseline differences (Preacher & Hayes, 2008). Further, regression coefficients for the a-path, b-path, as well as direct (c’), total (c), and indirect effects were calculated for all models (Figure 1). The a-path represents the effect of X on M, while the b-path is the effect of M on Y while controlling for X. The total effect is the effect of X on Y, without partialling out the indirect effect and the direct effect represents the effect of X on Y after partialling out the indirect effect of the mediator. The direct effect therefore represents the effect of X on Y after taking into account the indirect effect of M. To test for mediation, indirect effects were calculated as the product of a and b. Bootstrapped confidence intervals (95%) with 5000 iterations were calculated for these indirect effects. If the corresponding confidence interval did not contain zero, indirect effects were seen as significant. All models were analyzed using per-protocol data including only participants that attended at least four of the eight group sessions.
RkJQdWJsaXNoZXIy MTk4NDMw