How plausible is my model? Assessing model plausibility of structural equation models using Bayesian posterior...

Area of Science:

• Statistics
• Psychometrics
• Computational Statistics

Background:

• Structural Equation Modeling (SEM) is widely used for model comparison.
• Bayesian Posterior Probability (BPP) offers a method for model selection, computable from Bayesian Information Criterion (BIC) scores.
• Existing methods lack guidelines for selecting models to compare, potentially overlooking plausible alternatives.

Purpose of the Study:

• To propose a novel method for selecting neighboring models to facilitate the use of BPP in SEM.
• To provide researchers with a systematic approach for identifying a relevant set of alternative models for comparison.
• To enhance decision-making in SEM by offering a more comprehensive model evaluation strategy.

Main Methods:

• Developed a novel method for identifying 'neighboring models' to be included in BPP-based model comparison.
• Integrated this method into the typical SEM workflow, allowing researchers to assess a fitted model against its neighbors.
• Created a user-friendly R package, 'modelbpp', to automate the generation of neighboring models, model fitting, and BPP computation.

Main Results:

• The proposed neighboring model selection method facilitates the use of BPP for comparing SEMs.
• The 'modelbpp' R package streamlines the process of generating neighboring models and computing BPPs.
• This approach complements traditional goodness-of-fit indices and can reveal model evidence missed by other metrics.

Conclusions:

• The novel neighboring model selection method enhances the utility of BPP for SEM.
• The 'modelbpp' R package provides a practical tool for researchers to improve model evaluation.
• This facilitates more informed decisions in selecting the most plausible structural equation model.