Edita Chvojka

Faculty of Social and Behavioural Sciences
Methodology and Statistics/
Interdisciplinary Social Science
Utrecht University

Email
Website

Project
Rethinking Model Fit Guidelines for Longitudinal Structural Equation Modeling (LSEM) in Research on Youth.

Evaluating model fit is a key ingredient in (L)SEM. Commonly reported model-
fit indices are the CFI, TLI, and RMSEA. Current model-fit guidelines (Hu &Bentler, 1999) do not consider how certain model characteristics (number of indicators, measurement occasions and/or latent variables, reliability and distribution of indicators) influence model-fit indices (Fan & Sivo, 2007; Chen et al., 2008.; McNeish et al., 2018). This is particularly problematic in the context of bullying among youth, as this phenomenon involves many underlying social and environmental factors (e.g., Bowes et al., 2013. Smith et al., 2003; Zych et al., 2020). Relying on the current fit guidelines for drawing conclusions on bullying trajectories is problematic. We aim to provide a tool to create unique model-fit intervals taking these model characteristics into account. We will evaluate our method using applied data on bullying.
A simulation study will evaluate the sensitivity, bias, and error rates based on the model characteristics mentioned previously of LSEMs to establish reasonable intervals of model fit that can be used instead of arbitrary cutoffs. A tool (calculator) for future researchers will be developed and tested on the TRAILS (Tracking Adolescents’ Individual Lives Survey; Ormel et al., 2012) data. We see this project going even further as we replicate and re-evaluate its use as we teach other researchers the technique and extend its capability to other model characteristics as necessary.
This project fits the IOPS scope as its main goal is to provide new informative guidelines for fit indices in SEM, one of the most prominent modeling techniques not only in psychometrics but in psychology in general. Our primary focus is on LSEM; however, the knowledge can also inform, for instance, the assessment of the fit of psychometric measurement models or model fit guidelines for SEM on cross-sectional data. Moreover, as we aim to develop a calculator for adjusted model-fit intervals, IOPS meetings and conferences provide a great opportunity to familiarize other researchers with such a tool and disseminate the knowledge among those who often mention the model-fit guidelines when teaching or use SEM as a part of their own modeling toolbox.

Supervisors
Prof. dr. A.G.J. van de Schoot
Prof. dr. G.W.J.M. Stevens
Dr. E.M. Grandfield
Dr. G. Ouvrein

Financed by
Faculty Social and Behavioural Sciences – Utrecht University

Period
October 2022 – April 2027