Bayesian nonparametric inference on stochastic ordering.

Area of Science:

• Statistics
• Bayesian Inference
• Computational Statistics

Background:

• Statistical analysis often requires comparing unknown distributions across different groups.
• Partial stochastic ordering provides a framework for comparing distributions with some constraints.
• Existing methods may lack flexibility in handling complex group structures and ordered distributions.

Purpose of the Study:

• To develop Bayesian inference methods for collections of distributions under partial stochastic ordering.
• To propose restricted dependent Dirichlet process priors for flexible mixture modeling.
• To enable robust testing of equalities and estimation of group-specific distributions.

Main Methods:

• Development of restricted dependent Dirichlet process priors with full support on stochastically ordered distributions.
• Application of Bayesian inference for collections of unknown mixture distributions.
• Utilizing Markov chain Monte Carlo (MCMC) for efficient posterior computation.

Main Results:

• The proposed priors provide a flexible class of mixture models for stochastically ordered data.
• Demonstrated theoretical properties of the new Bayesian approach.
• Efficient computational methods were developed for practical implementation.

Conclusions:

• The novel Bayesian approach effectively handles Bayesian inference for distributions under partial stochastic ordering.
• Restricted dependent Dirichlet process priors offer a powerful tool for flexible mixture modeling in comparative studies.
• The methods are validated through application to DNA damage and repair data analysis.