Bayesian Kernel Mixtures for Counts.

Area of Science:

• Statistics
• Biostatistics
• Computational Statistics

Background:

• Existing Bayesian nonparametric mixture models are limited for count data.
• Common approaches like Poisson mixtures cannot handle underdispersion.
• Alternative methods have limitations in flexibility and support.

Purpose of the Study:

• To propose a flexible class of Bayesian nonparametric mixture models for count data.
• To address the limitations of existing methods in capturing complex count data distributions.
• To develop efficient computational methods for posterior inference.

Main Methods:

• Utilizing nonparametric mixtures of rounded continuous kernels.
• Developing an efficient Gibbs sampler for posterior computation.
• Extending the framework to multivariate count data and joint modeling.

Main Results:

• The proposed models demonstrate superior performance in simulation studies.
• The rounded Gaussian mixture model effectively handles various count data characteristics.
• The framework successfully accommodates multivariate and joint modeling scenarios.

Conclusions:

• Nonparametric mixtures of rounded continuous kernels provide a flexible alternative for count data modeling.
• The developed methods are applicable to diverse real-world data, including toxicity and marketing studies.
• This approach expands the toolkit for Bayesian analysis of count data.