Detecting and Classifying Misspecification in Structural Equation Modeling using Machine Learning

• Applicant: Dr. David Goretzko
• Subject Area: Personality Psychology, Clinical and Medical Psychology, Methodology
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 457716512

The central goal of this project is to develop a new method for the detection and classification of model misspecifications in confirmatory factor analyses (CFA as an example for structural equation modeling). Conventional methods for the identification of misspecified models have various disadvantages and cannot be generalized to all application contexts. Therefore, in this project a new method is to be tested promising to detect and classify misspecifications in CFA models for broad data conditions. By combining comprehensive data simulation and modern machine learning (ML) methods, a predictive model will be developed which is able to detect misspecifications of a structural equation model using different data properties and model characteristics as input features. The general idea - combining ML models and extensive data simulation to develop a predictive model that supports the application of a classical statistical method - has already been tested in the factor retention process of exploratory factor analysis (see Goretzko & Bühner, 2020). Accordingly, it can be assumed that the new approach may also work properly as a tool for model evaluation in CFA and can make an important contribution to improving research practice.

• DFG Programme: Research Grants
• International Connection: Netherlands