Bayesian inference for two-part mixed-effects model using skew distributions, with application to longitudinal...

Area of Science:

• Statistics
• Biostatistics
• Longitudinal Data Analysis

Background:

• Semicontinuous data, common in substance abuse research, often present excess zeros and right-skewed positive values.
• Traditional statistical models may not adequately capture the complexities of such data in longitudinal studies.
• Two-part mixed-effects models offer a framework for analyzing repeated measures of semicontinuous outcomes.

Purpose of the Study:

• To propose a flexible two-part mixed-effects model incorporating skew distributions for correlated semicontinuous alcohol data.
• To apply a Bayesian approach for robust statistical inference.
• To analyze longitudinal alcohol abuse/dependence symptoms data.

Main Methods:

• A two-part mixed-effects model was developed, comprising a generalized logistic model for the occurrence of positive values (Part I) and a linear mixed-effects model for the intensity of positive values (Part II).
• Correlated random effects linked the two parts of the model.
• Skew distributions, specifically skew-t and skew-normal, were employed for the model errors in Part II.
• A Bayesian framework was utilized for model estimation and comparison.

Main Results:

• The proposed flexible two-part mixed-effects model was successfully applied to longitudinal alcohol abuse/dependence symptoms data.
• Model performance was evaluated by comparing various model specifications under different random-effects structures.
• Simulation studies confirmed the effectiveness of the proposed methodology.

Conclusions:

• The flexible two-part mixed-effects model with skew distributions provides a robust approach for analyzing correlated semicontinuous longitudinal data, particularly in substance abuse research.
• The Bayesian framework facilitates comprehensive analysis and model comparison.
• This methodology enhances the understanding of factors influencing alcohol use patterns over time.