Bayesian semiparametric analysis of structural equation models with mixed continuous and unordered categorical...

Area of Science:

• Biostatistics
• Medical Informatics
• Statistical Modeling

Background:

• Structural Equation Models (SEMs) are widely used in biological and medical research.
• Traditional SEMs often assume normally distributed latent variables, limiting their applicability.
• Handling mixed continuous and categorical data within SEMs presents analytical challenges.

Purpose of the Study:

• To develop a Bayesian semiparametric SEM capable of analyzing data with mixed variable types.
• To model latent variables without the restrictive assumption of normality.
• To incorporate covariates into the SEM framework for enhanced explanatory power.

Main Methods:

• A Bayesian semiparametric SEM approach was employed.
• Latent variables were modeled using a truncated Dirichlet process with a stick-breaking procedure.
• The methodology accommodates mixed continuous and unordered categorical variables and includes covariates.

Main Results:

• The proposed semiparametric SEM demonstrated flexibility in modeling latent variables.
• Simulation studies confirmed the methodology's performance.
• The model was successfully applied to a real-world medical dataset, showcasing its practical utility.

Conclusions:

• The Bayesian semiparametric SEM offers a robust alternative to traditional methods for complex biological and medical data.
• This approach enhances the analysis of datasets with non-normally distributed latent variables and mixed data types.
• The methodology provides a valuable tool for researchers investigating intricate relationships in health sciences.