Censored Quantile Regression Redux

This vignette is a slightly modified version of Koenker (2008a). It was written in plain latex not Sweave, but all data and code for the examples described in the text are available from either the JSS website or from my webpages. Quantile regression for censored survival (duration) data offers a more flexible alternative to the Cox proportional hazard model for some applications. We describe three estimation methods for such applications that have been recently incorporated into the R package quantreg: the Powell (1986) estimator for fixed censoring, and two methods for random censoring, one introduced by Portnoy (2003), and the other by Peng and Huang (2008). The Portnoy and Peng-Huang estimators can be viewed, respectively, as generalizations to regression of the Kaplan-Meier and NelsonAalen estimators of univariate quantiles for censored observations. Some asymptotic and simulation comparisons are made to highlight advantages and disadvantages of the three methods.