Date created: | Last Updated:

Category: Uncategorized

Description: This document contains the second version of the pre-analysis preregistration and the replication files for the project: ‘A New School, a Fresh Start? (Dis)continuation of Peer Attachment, Loneliness and Victimization across the Transition from Primary to Secondary School, and its Impact on Academic Achievement’. The preregistration is based on the template https://osf.io/t6m9v/, adapted for pre-analysis purposes, and contains A. introduction, theory, and hypotheses, B. methods, C. analysis plan, and final questions. The replication package zip file contains the code that was used for this study, also see the 'read me' file in this folder. Find the first version here https://osf.io/rdy34. The first preregistration version for this project contained hypotheses and an analytic plan for a person-centered approach. We pursued this approach yet concluded that a person-centered model does not fit our data. Specifically, first, we ran two cross-sectional LCAs, one for each time point. A two-class model fit the data best based on fit measures, yet the smallest class on the second timepoint consisted of 4.8% of the sample (n = 35) and thus violated the assumption that the smallest class should comprise at least 5.0% of the sample. When conducting a longitudinal LCA on both time points while constraining the means to be equal across time (Morin & Litalien, 2017), we ran into the error that the variance was zero for victimization at T2. This indicated estimation problems (i.e., the LCAs do not fit) and very low statistical power. After consultation with an expert in the SEM field, we conducted the cross-sectional LCAs again specifying different variance-covariance structures (Masyn, 2013) to check if this would improve model fit, rather than using the default setting for the variance-covariance structure. However, the best fitting model remained to find the smallest class to be smaller than 5.0%, namely 4.8%. This means the person-centered approach did not fit the model well. Moreover, a class of n = 35 would be too small to base conclusions on. Furthermore, the identification of only two profiles may theoretically be less interesting as the large group would represent students scoring high on attachment, and low on loneliness and victimization, and the very small group would represent students scoring low on attachment, and high on loneliness and victimization. This 'high-low distinction' could also be shown when using a variable-oriented approach. Therefore, in consultation with the professional and the co-authors, we decided to apply a variable-centered approach to the data instead of a person-centered approach. This aligns with ‘Option 2’ under Alternative Plan of the first version of the preregistration: https://osf.io/rdy34. We decided to preregister a new version of the paper as the hypotheses will substantially change using a variable-centered approach and as we have looked into the data more before adjusting the hypotheses than before the first version of the preregistration. That being said, the rationale, introduction, and theory largely overlap with the first version of the preregistration: https://osf.io/rdy34.