Chapter 3 92 PREREGISTERED STATISTICAL ANALYSES We conducted the same preregistered Random-Intercepts Cross-Lagged Panel Models (RI-CLPMs; see Hamaker et al., 2015; for graphical representation, see Figure 1) with Mplus 8.3 (Muthén & Muthén, 2020) for all five datasets. In contrast to a standard CLPM, a RI-CLPM disentangles the within-family variance (i.e., over-time fluctuations within the same family) from the stable between-family variance (i.e., relative mean differences between families) by modeling the between-family variance as latent factors (Hamaker et al., 2015; Keijsers, 2016). Consequently, the remaining residual variances represent the fluctuations of the families around their own stable mean levels, which are used to estimate the carry-over stability paths, within-family correlations at T1 and correlated errors at T>1, and within-family cross-lagged paths. For the preregistration, see https:// osf.io/bfyst/. Prior to the data-analyses, we ran the preregistered checks on our data-structures. We established that within-family variance of parental support and adolescent depressive symptoms was in all cases at least 10%, specifically within the range from 20% to 56% (see Tables A1-A5 in appendix), justifying the use of RI-CLPM. The pattern of the missing data was considered Missing at Random (i.e., χ2/df of Little’s MCAR test higher than 3). Therefore, full information maximum likelihood (FIML) was used to handle missing data. Moreover, we used maximum likelihood with robust standard errors (MLR estimator) (Kline, 2016) with the daily dataset, because skewness of parental support and adolescent depressive symptoms was higher than 3. In all other cases, Maximum Likelihood estimation (ML estimator) was used. To test H1-3, we applied the preregistered time-constrained models, in which the stability paths, correlated errors at T>1, and cross-lagged paths were constrained to be equal across measurement occasions. Because the comparative fit index and the Tucker Lewis index values were all higher than 0.90, and the root-mean-squared error of approximation lower than 0.08 (Schumacker & Lomax, 2016), the model fit of all single-group timeconstraint models was acceptable (see Table 2). Subsequently, we conducted multi-group analyses to test whether girls compared to boys (H4) and adolescents scoring high on neuroticism compared to adolescents scoring low (H5; based on the median-split approach), would differ in lagged effects and thus show different transactional processes between parental support and adolescent depressive symptoms. Specifically, we estimated two types of constrained models, one in which we
RkJQdWJsaXNoZXIy MTk4NDMw