Maartje Boer

SMU AND MENTAL HEALTH 165 6 extend the measurement, a fourth item was added to the scale, including ‘How often are your friends at your home?’ (1 never to 6 very often ). Cronbach’s alpha varied between 0.85 and 0.87 across the waves. Preliminary Analyses Prior to our main analyses, we conducted three preliminary analyses using Mplus 8.4 (L. K. Muthén & Muthén, 2017b). Attrition Analyses First, we carried out attrition analyses by predicting dropout in T2 and T3 with the study measures at the previous wave using multivariate logistic regressions with the measures modelled as latent variables. Results showed that adolescents with higher levels of SMU problems, lower life satisfaction, lower upward social comparisons, less face-to-face contact, or lower school achievements in T1 were more likely to drop out in T2 (OR range = 1.166 to 1.408). Adolescents with higher levels of SMU problems, higher SMU intensity, higher levels of cybervictimization, or lower levels of upward social comparisons in T2 were more likely to drop out in T3 (OR range = 1.115 to 1.626). The magnitudes of these associations varied from very small (OR < 1.5) to small (OR < 2) (Sullivan & Feinn, 2012). Thus, although there were relatively many pre-vocational level educated adolescents that dropped out (particularly in T3), associations between our study variables and dropout were limited. Measurement Invariance Second, we conducted measurement invariance analysis, because in order to draw valid conclusions about changes over time, our measures should have the same measurement properties across the investigated years (F. F. Chen, 2007; Cheung & Rensvold, 2002). Hence, we examined whether the magnitudes of the item factor loadings and intercepts (or thresholds in case of categorical items) were consistent over time. Measurement invariance was established when applying equality constraints to the item factor loadings and intercepts (or thresholds) did not substantially deteriorate model fit in terms of change in Comparative Fit Index ( Δ CFI = decrease of ≤ .010) and Root Mean Square Error of Approximation ( Δ RMSEA = increase of ≤ .015;