Are fit indices used to test psychopathology structure biased? A simulation study

Area of Science:

• Psychometrics
• Psychopathology Research
• Statistical Modeling

Background:

• Structural models offer dimensional alternatives to categorical systems in psychopathology.
• Bifactor and correlated factors models are common, with fit indices used for comparison.
• Previous research indicated potential bias in fit indices within psychological domains.

Purpose of the Study:

• To investigate bias in fit indices for structural models of psychopathology.
• To extend research on fit index bias to the bifactor model, which lacks thorough characterization.
• To examine the impact of model misspecifications on fit index bias using different estimators and data types.

Main Methods:

• Monte Carlo simulations were employed to generate data.
• Binary and positively skewed continuous indicators simulated psychiatric diagnoses and symptom counts.
• Two common estimators, WLSMV and MLR, were used to assess model fit under various misspecifications.

Main Results:

• Complex patterns of bias were observed across estimators, indicator distributions, and misspecifications.
• Fit indices frequently failed to identify the correct data-generating model (correlated factors model).
• No single fit index consistently provided unbiased results across all tested scenarios.

Conclusions:

• Fit index comparisons between bifactor and alternative models in psychopathology may be unreliable.
• Previous studies favoring the bifactor model based solely on fit indices warrant re-evaluation.
• Emphasize substantive interpretability, model equivalence, and robust statistical metrics for evaluating structural models.