|
Abstract:
|
For many biomedical studies with skewed multivariate responses, the main goals are to estimate the covariate effects and the degree of skewness to describe their effects on the response and the predictive distributions. To address this, we propose a new class of semiparametric multivariate skewed response models with associated Bayesian method. Our proposed model enjoys several desirable properties, including a flexible structure for the within subject association, meaningful physical interpretations of marginal covariate effects and skewness, and assurance of consistent Bayesian estimates of parameters and nonparametric error density under a set of practical assumptions on the priors. We examine the finite sample performance and robustness of our proposed method through simulation studies. We also illustrate the practical advantages of our methods over existing parametric methods by analyzing a motivating dataset from a periodontal disease study.
|
