Detection and Classification of Misspecifications in Structural Equation Modeling using Machine Learning

In this DFG-funded project (Detection and Classification of Misspecifications in Structural Equation Modeling using Machine Learning, DFG GO 3499/1-1), modern approaches from computational statistics such as machine learning and meta-heuristic optimization algorithms are used to develop novel methods for detecting and classifying misspecifications in structural equation models as well as automatically re-specifying measurement models in a fully data-driven manner. By integrating various nuisance parameters (e.g., sample and model size) as features in the prediction model, the new approach promises to consider their interactions with model fit indices. Accordingly, the ML-based model fit evaluation addresses shortcomings in the current practice that often relies on simple rules of thumb combining fixed cutoffs and model fit indices.

Figure 1: Abstract visualization of a tree-based ML approach taking different nuisance parameters into account when identifying model misspecifications in SEM. Created with DALL-E.

Selected Publications and Preprints:

Goretzko, D., Siemund, K., & Sterner, P. (2024). Evaluating Model Fit of Measurement Models in Confirmatory Factor Analysis. Educational and Psychological Measurement. https://doi.org/10.1177/00131644231163813

Partsch, M. V., Sterner, P., & Goretzko, D. (2025). A simulation study on the interaction effects of underfactoring and nuisance parameters on model fit indices. Structural Equation Modeling: A Multidisciplinary Journal. https://doi.org/10.1080/10705511.2025.2514592

Partsch, M. V., & Goretzko, D. (2025). Detecting model misfit in structural equation modeling with machine learning—A proof of concept. Multivariate Behavioral Research. https://doi.org/10.1080/00273171.2025.2552304

Goretzko, D., & Partsch, M. V. (2025). Predicting measurement model misfit with machine learning while accounting for nuisance parameters — An illustration. Psychological Test Adaptation and Development. https://doi.org/10.1027/2698-1866/a000111

Partsch, M. V., & Goretzko, D. (2025). A Machine Learning Pipeline to Classify the Type and Severity of Misspecifications in Latent Measurement Models. Preprint at https://doi.org/10.31234/osf.io/84ftk_v1.

Goretzko, D., & Wetzel, E. (2026). Stop Using Fixed Cutoffs to Evaluate Latent Variable Models! European Journal of Psychological Assessment. https://doi.org/10.1027/1015-5759/a000943