Maartje Boer

VALIDATION OF THE SMD-SCALE 37 2 Subgroups of Users We explored whether we could identify subgroups with specific item score patterns by means of Latent Class Analysis (LCA) on the nine items. Specifically, we evaluated different class (i.e., subgroup) solutions on their model fit and classification accuracy (Nylund et al., 2007). Model fit was examined using the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), and the Lo-Mendell-Rubin adjusted Likelihood Ratio Test (LMR-LRT). Classification accuracy was based on the Entropy. After the best class solution was established, we compared adolescents’ observed item scores across the empirically identified classes. In addition, the LCA- models assume by default that the items are independent within each class, that is, that there are no correlations between the residuals of the items (Asparouhov & Muthén, 2015). This assumption of ‘conditional independence’ is often too restrictive, because it typically does not comply with the data. Therefore, imposing the assumption may lead to biased results and wrong model selection (Uebersax, 1999). Hence, a sensitivity analysis was conducted where the LCA was repeated while allowing for conditional dependence. Particularly, for each model, we consulted the ‘bivariate fit information’ to inspect the pairs of items that violated the assumption based on the bivariate Pearson Chi-Square ( > 10), after which we modified the respective model by adding correlations between the pairs of items that violated the assumption (Asparouhov & Muthén, 2015). We applied this procedure to all class solutions and evaluated whether it yielded a similar model selection as the initial analysis that assumed conditional independence. Criterion Validity Criterion validity defines the extent to which test scores relate to outcomes they should theoretically be related to. We examined whether higher levels of problematic SMU were associated with more mental health problems (emotional problems, conduct problems, hyperactivity, and peer problems), school problems (school dissatisfaction, school pressure), and sleep problems (less hours of sleep than recommended, low sleep quality). Problematic SMU was measured by the sum-score of the nine items of endorsed problematic SMU criteria (min. 0, max. 9). Due to the dichotomous nature of the outcome variables, analyses were conducted using logistic regression. In these