Bayesian variable selection for latent class models.

Area of Science:

• Statistics
• Biostatistics
• Machine Learning

Background:

• Latent class models are useful for identifying unobserved subgroups in data.
• Predicting these latent classes based on subject characteristics is crucial for understanding underlying phenomena.
• Variable selection uncertainty can complicate the identification of important predictors.

Purpose of the Study:

• To develop a latent class model where class probabilities are influenced by subject-specific covariates.
• To identify significant predictors of these latent classes, accounting for variable selection uncertainty.
• To apply the proposed methodology to real-world data, specifically pregnancy weight gain.

Main Methods:

• Developed a latent class model incorporating subject-specific covariates.
• Proposed a Bayesian variable selection approach to handle uncertainty in predictor identification.
• Implemented a stochastic search Gibbs sampler for posterior computation.
• Calculated model-averaged estimates, including marginal inclusion probabilities for predictors.

Main Results:

• The Bayesian variable selection approach effectively identified important predictors of latent classes.
• The methodology demonstrated robustness in simulation studies.
• Application to pregnancy weight gain data revealed key predictors associated with different weight gain patterns.

Conclusions:

• The proposed Bayesian latent class modeling framework with variable selection is effective for identifying key predictors.
• This approach provides reliable model-averaged estimates, enhancing interpretability.
• The findings offer insights into factors influencing pregnancy weight gain, with potential implications for clinical practice.