Covariate inclusion in factor mixture modeling: Evaluating one-step and three-step approaches under model...

Area of Science:

• Behavioral and social sciences
• Quantitative psychology
• Statistical modeling

Background:

• Factor mixture modeling (FMM) is a statistical technique used to uncover unobserved heterogeneity within populations.
• Incorporating covariates into FMM helps elucidate the characteristics and formation of identified latent subgroups or classes.
• Evaluating different covariate inclusion strategies is crucial for accurate subgroup analysis.

Purpose of the Study:

• To compare the performance of one-step and three-step covariate inclusion approaches in Factor Mixture Modeling (FMM).
• To assess these methods under conditions of correct specification, misspecification, and overfitting of covariate effects.
• To provide guidance on selecting appropriate FMM covariate strategies based on study conditions.

Main Methods:

• A Monte Carlo simulation study was conducted to evaluate FMM covariate inclusion methods.
• Three scenarios were simulated: correct specification, model misspecification, and model overfitting regarding direct covariate effects on factors.
• Performance was assessed by comparing the one-step and three-step FMM approaches across these simulated conditions.

Main Results:

• The one-step and three-step FMM approaches performed comparably with large class separation and correct covariate effect specification.
• The one-step FMM demonstrated superior class enumeration performance when class separation was poor.
• The one-step FMM exhibited greater robustness against misspecification or overfitting of direct covariate effects.

Conclusions:

• The choice between one-step and three-step FMM covariate inclusion depends on the degree of class separation and sample size.
• The one-step FMM is recommended for situations with poor class separation or potential model misspecification.
• A large sample size (>=1000) and the use of sample size-adjusted BIC (saBIC) for class enumeration are advised for robust FMM analyses.