Maartje Boer

SMU AND MENTAL HEALTH 167 6 analyses. Table 6.1 shows the descriptive statistics of the plausible values over the imputed datasets in long format. The plausible values were used for our main analyses. Table 6.1 Descriptive Statistics (Long Format, n = 6,327) Variable Mean SD Min. Max. SMU problems -0.013 0.663 -2.411 2.530 SMU intensity 0.120 1.278 -4.470 4.176 Depressive symptoms 0.089 0.618 -1.890 2.554 Life satisfaction -0.212 1.044 -4.723 3.069 Upward social comparison 0.006 0.645 -2.121 2.604 Face-to-face contact with friends -0.042 1.042 -3.712 3.533 School satisfaction -0.006 0.946 -3.508 3.240 Cybervictimization 0.030 1.003 -3.340 6.405 Notes. SD = standard deviation; Min. = minimum; Max. = maximum. Results denote the descriptive statistics of the computed plausible values averaged over 20 imputed datasets. Descriptive statistics were computed with data structured in long format (i.e. each row in the dataset represents an observation). Main Analyses Analytical Approach Directionality can be established by studying whether adolescents’ increases in, for example, SMU problems precede or follow from increases in, for example, depressive symptoms. Grasping such dynamic processes that occur within adolescents requires separating within-person variance from between-person variance. Hence, we investigated our research questions using the ‘random intercept cross-lagged panel model’ (RI-CLPM), which is an innovative modelling technique that examines bidirectional processes within persons (Hamaker et al., 2015). By disentangling within- and between- person variance, the RI-CLPM controls for all possible stable characteristics, providing more accurate estimates of directionality (Hamaker et al., 2015). Modelling the RI-CLPM Figure 6.2 illustrates a RI-CLPM with SMU intensity, SMU problems, and depressive symptoms. The between-person part of themodel consisted of the random intercepts (light gray circles), which are latent variables that denote the time-invariant levels of the respective behaviors. The random intercepts were extracted from three repeated plausible values (white squares), with factor loadings constrained to one. The RI-CLPM also included correlations