6 Like no other? 181 Pre-registered Analytical Approach S-GIMME To answer the research question whether subgroups of families exist who share similar daily parent-adolescent dynamics, a pre-registered (see https://osf.io/a4rzm) Subgrouping Group Iterative Multiple Model Estimation (S-GIMME; Gates et al., 2017; Lane et al., 2019) was conducted, by using the R package gimme version 0.7-10 (Gates & Molenaar, 2012). Figure 1 visualizes how this study used GIMME, a data-driven statistical technique, to estimate sparse unit-specific (here: family-specific) temporal networks. GIMME is particularly well-suited for estimating the heterogeneity of associations in intensive longitudinal data from heterogeneous samples (Gates & Molenaar, 2012). To estimate a family-specific network, as well as (sub)groups of similarly functioning families, GIMME implements family-specific unified structural equation models (see Gates et al., 2010). These models are a type of structural vector autoregressive (VAR) model that combines traditional VAR and structural equation models (SEM) to simultaneously estimate directed lagged (i.e., first-order next-day) and contemporaneous (i.e., same-day) associations. GIMME implements these family-specific models within a grouping algorithm that prioritizes the estimation of relationships that are common across participants (if any exist). All technical steps are summarized in Figure 2. GIMME handles missing data using full information maximum likelihood (FIML; Beltz & Gates, 2017). To achieve this, GIMME begins with empty “null” network models. Then, group-level associations are iteratively added to all empty family-specific networks if they significantly (after Bonferroni correction of .05/N, thus the alpha level here was .004) improve the model fit for the majority of the sample (here, 51%; see Gates et al., 2020) according to Lagrange multiplier equivalence tests (i.e., modification indices; Sörbom, 1989). To improve path recovery, autoregressive effects were estimated for every family-specific network, by default (Lane et al., 2019). Subsequently, the subgrouping option within the GIMME algorithm clusters individual families using Walktrap community detection, based on similarities in family-specific estimates of (1) group-level associations and (2) associations that are likely to emerge at the individual family level. Subgroup-level associations are iteratively added to the family-specific networks of the subgroup members if they significantly (Bonferroni corrected) improve the model fit for the majority of the subgroup members (again 51%; Gates et al., 2017; Lane et al., 2019) according to modification indices. Finally, individuallevel associations (i.e., associations unique to each family) are iteratively added to a family’s network by evaluating whether they significantly (p < .01) improve model fit. At the group,
RkJQdWJsaXNoZXIy MTk4NDMw