Maartje Boer

CHAPTER 5 136 to the intercepts of two items from the hyperactivity-scale. Hyperactivity was thereby sufficiently invariant over time for the purposes of our analyses (Van de Schoot et al., 2012). Table 5.1 Measurement Invariance Analysis: Multigroup CFA (n = 1,629) Overall model fit constrained model 1 Change in model fit 2 CFI TLI RMSEA ΔCFI ΔRMSEA SMU intensity 0.989 0.989 0.047 0.009 -0.010 SMU problems 0.963 0.957 0.034 -0.007 0.006 Attention deficits 0.932 0.935 0.073 0.009 0.007 Impulsivity 0.987 0.987 0.031 0.004 0.002 Hyperactivity 0.874 0.879 0.122 0.019 0.026 Notes . SMU = social media use; CFA = confirmatory factor analysis; CFI = comparative fit index; TLI = Tucker-Lewis index; RMSEA = root mean square error of approximation. 1 Multigroup CFA model where item loadings and intercepts/thresholds were constrained to be equal over time. 2 Compared to multigroup CFA model where item loadings and intercepts/thresholds were free to vary over time. Generating Factor Scores Modelling the RI-CLPM using latent variables for our measures was not feasible, given the complexity of our model related to the large number of latent variables. We therefore considered using the sum-scores of the observed items, which is the most common practice in applications of the RI-CLPM (Hamaker et al., 2015). However, the distribution of the sum-score of SMU problems is heavily skewed (Van den Eijnden et al., 2016), which often leads to biased results in statistical analyses (Hox et al., 2010). Moreover, sum- scores do not consider that items have different contributions to their latent measure, as reflected by their different factor loadings, which may lead to inaccurate representations of latent measures (Distefano et al., 2009). We addressed these shortcomings by using factor scores instead of sum-scores, which are imputed values that reflect plausible values of latent measures based on the CFA-model (Distefano et al., 2009). Factor scores were computed using Mplus 8.1 (L. K. Muthén & Muthén, 2017b). For all fivemeasures separately, CFA-models with three latentmeasures were specified, referring to the three repeated measures in wide format ( n = 543). In these models, measurement invariance over time was imposed, and means of the latent measures were freely estimated. Factor scores according to these CFA-models were computed and saved. The saved factor scores were subsequently used as observed variables for the RI-CLPM. Factor scores of