Bayesian analysis of multivariate mixed longitudinal ordinal and continuous data

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Joint analysis of multivariate longitudinal ordinal and continuous data is challenging due to complex correlations and lack of suitable distributions.
• The multivariate probit model is a natural choice but faces identifiability constraints, restricting covariance matrix elements.
• These constraints limit the development of classical and Bayesian methods for mixed data analysis.

Purpose of the Study:

• To propose novel Markov chain Monte Carlo (MCMC) methods for the joint analysis of mixed multivariate longitudinal data.
• To overcome the identifiability issues associated with the multivariate probit model in longitudinal settings.
• To provide robust analytical tools for researchers dealing with complex mixed-type longitudinal datasets.

Main Methods:

• Developed three MCMC algorithms: Metropolis-Hastings within Gibbs (identifiable model), Gibbs sampling (non-identifiable model), and parameter-expanded data augmentation (non-identifiable model).
• Utilized simulation studies to evaluate the performance and efficiency of the proposed methods.
• Applied the methods to a real-world dataset to demonstrate practical utility.

Main Results:

• The proposed MCMC methods effectively handle the joint analysis of multivariate longitudinal ordinal and continuous data.
• The non-identifiable model-based MCMC sampling methods show promise for developing advanced analytical techniques.
• Performance evaluation through simulations and real data application confirms the viability of the developed approaches.

Conclusions:

• The study successfully extends MCMC methodologies for analyzing complex mixed-type longitudinal data.
• The developed methods offer flexible and powerful tools for statistical modeling in various scientific fields.
• The use of non-identifiable models provides a valuable avenue for future methodological advancements in MCMC sampling.