A0389
Title: Polya trees for survival data
Authors: Liqun Diao - University of Waterloo (Canada) [presenting]
Yixing Zhao - University of Waterloo (Canada)
Abstract: Polya trees are commonly used as priors in nonparametric Bayesian analysis. Approaches for utilizing Polya trees to characterize the distribution of time-to-event data are discussed, which may be subjected to different forms of censoring, such as right censoring or interval censoring. Different aspects of Polya trees are covered, including partitions, prior strength, and choices of prior distributions. Comparisons of the proposed methods to existing approaches for estimating survival probabilities are provided in both simulated settings and through applications to real datasets. It is shown that the proposed methods either improve upon or remain competitive with existing nonparametric estimation methods.
