Impact of prior specifications on performance of Bayesian factor mixture modeling

Area of Science:

• Social and behavioral sciences
• Health sciences
• Quantitative psychology
• Statistical modeling

Background:

• Factor mixture modeling (FMM) integrates continuous latent factors and categorical latent classes to reveal population heterogeneity.
• FMM's complexity presents methodological challenges, particularly in classification accuracy and parameter recovery.
• Bayesian estimation offers a flexible framework for addressing these FMM challenges.

Purpose of the Study:

• To evaluate the effectiveness of Bayesian estimation, specifically prior specifications, in improving FMM's classification accuracy and parameter recovery.
• To investigate the impact of subjective prior beliefs about class separation on FMM performance.
• To provide guidance for applied researchers using FMM.

Main Methods:

• Comprehensive Monte Carlo simulations were employed to assess FMM performance under various conditions.
• Bayesian estimation was utilized, with a focus on the influence of prior specifications on model outcomes.
• Scenarios manipulated subjective class separation beliefs relative to true population class separation.

Main Results:

• Adequate model performance was achieved using a moderately informative prior.
• Prior specifications where subjective class separation exceeded true class separation demonstrated adequate results.
• The study identified conditions under which Bayesian FMM yields reliable estimates.

Conclusions:

• Bayesian estimation, particularly with carefully chosen prior specifications, is a viable approach for addressing methodological challenges in FMM.
• Researchers can leverage prior information to enhance classification accuracy and parameter recovery in FMM.
• The findings offer practical guidance for the application of FMM in social, behavioral, and health sciences.