Modeling a mixture of ordinal and continuous repeated measures

Area of Science:

• Statistics
• Biostatistics
• Psychometrics

Background:

• Analyzing mixed data types (ordinal and continuous) presents statistical challenges.
• Repeated measures designs are common in health sciences but require specialized modeling techniques.
• Existing methods may not adequately capture the complex correlation structures in combined data.

Purpose of the Study:

• To develop a Bayesian framework for modeling correlation structures in mixed ordinal and continuous repeated measures.
• To address the computational challenges associated with sampling restricted covariance matrices.
• To provide a flexible methodology applicable to various research fields, including neuroscience.

Main Methods:

• Utilized a Bayesian approach with a multivariate probit model for ordinal variables and normal linear regression for continuous variables.
• Incorporated latent normal variables to link ordinal and continuous data, allowing for correlation.
• Employed parameter-extended data augmentation and the Metropolis-Hastings algorithm for posterior distribution sampling of the restricted covariance matrix.

Main Results:

• Successfully developed and demonstrated a computational method for analyzing mixed-type repeated measures.
• The Bayesian approach effectively handles the correlation structure between ordinal and continuous variables.
• Validated the methodology through simulation studies and a real-world application.

Conclusions:

• The proposed Bayesian methodology provides a robust framework for analyzing mixed ordinal and continuous repeated measures.
• Parameter-extended data augmentation is an effective computational strategy for sampling restricted covariance matrices.
• This approach offers valuable insights into complex data structures, as shown in the UCLA Brain Injury Research Center application.