Maartje Boer

VALIDATION OF THE SMD-SCALE 43 2 Figure 2.2 Item Information Curves (A) and Total Information Curve (B), n = 6,626 Notes. SMU = social media use. Items in legend were sorted on their discrimination parameter . Δ -CFI = 0.000; Δ -RMSEA = -0.002; Δ -SRMR = 0.002), because imposing equality constraints did not substantially deteriorate model fits (F. F. Chen, 2007). Thus, the factor loadings and thresholds of all nine items were equal across all group comparisons, which implies measurement invariance across all investigated subpopulations. Subgroups of Users Table 2.3 shows the results of the LCA. We examined five class-solutions, because the five-class solution did not improve model fit relative to the four- class solution (LMR-LRT p = 0.122), which makes estimating additional class- solutions redundant (Nylund et al., 2007). TheAICandBICdecreasedwith each number of increasing classes, indicating that model fit improved with the number of classes (Nylund et al., 2007). However, the classification accuracy of the four- and five-class solutions was lower than 0.7, which is often considered as unacceptable (Meeus et al., 2010; Reinecke, 2006a). This means that there was substantial overlap in adolescents’ item scores between the classes in