Maartje Boer

THE COMPLEX ASSOCIATION BETWEEN SMU AND WELLBEING 245 8 Modelling Next, the data were exported to Mplus 8.6 (L. K. Muthén & Muthén, 2017b) to conduct a series of multilevel models. First, fixed effects models were conducted to test the associations between the six different SMU activities and life satisfaction. Specifically, we estimated the within-person and between-person associations between adolescents’ intensity of SMU activities and their life satisfaction, for each SMU activity separately (M1a-f). In these models, the within-person associations were constrained to be equal across adolescents. The models included ‘wave’ as a level-1 control variable to account for common time trends (Hox, 2010a; Wang & Maxwell, 2015). Also, we included gender, educational level, and immigrant background as level-2 control variables. This first series of models are referred to as the baseline models. To test whether the associations were confounded by SMU problems, we subsequently extended the baseline models with SMU problems as additional level-1 and level-2 control variable (M2a-f). In the next step, we extended the baseline models with quadratic terms for the SMU activities on both levels (M3a-f). Thereafter, we extended the quadratic models with SMU problems as additional control variable on both levels (M4a-f). Further, we extended the baseline models with random slopes for the within-person associations between the six SMU activities and life satisfaction (M5a-f). As recommended for multilevel modeling, a covariance between the randomslope and random intercept was specified (Hox, 2010d). When adding the random parameters significantly improved model fit, this indicated that the respective within-person association varied across adolescents. Model fit was evaluated based on the deviance , where lower values indicated better model fit. The difference in deviance was evaluated using a chi- square difference test, with a corrected p- value as appropriate for testing slope variance (Hox, 2010b; Stoel et al., 2006). Also, lower Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) values indicated better model fit (Hox, 2010b). In addition, 95% prediction intervals (PIs) were computed, which express the estimated range of the associations across adolescents (Hox, 2010d). Subsequently, these random effects models were extended with SMU problems as additional control variable on both levels (M6a-f). Continuing on the random effects models, we examined whether the variances of the within-person associations between the six SMU activities