|
Abstract:
|
Censoring challenges any regression analysis, including the quantile regression. However, the conditional quantile functions often have better identifibility than the conditional mean function when the response variable is censored. In this talk, we examine the impact of fixed and random censoring on the estimation of quantile regression models. A new approach to doubly censored response will be outlined, and a discussion of existing algorithms on censored quantile regression will be attempted. In particular, we demonstrate that some commonly used censored regression quantile estimators that do not share the spirit of Kaplan-Meier could be poor in efficiency, but a global approach to redistributing the probability mass of censored observations is both challenging and rewarding.
|
