A parametric competing risks regression model with unknown dependent censoring

Abstract

A typical situation in survival analysis is that there is only interest in one risk and some prior information about its distribution is available. At the same time, other risks are not of interest and no information about risk dependence is available. A parametric regression model with unknown dependent censoring would be suitable here, but existing approaches require restrictions on all marginal survivals, the censoring distribution or the degree of dependence. This article introduces a model without these restrictions. An application to employment duration demonstrates that it avoids sizable bias of the estimated gender effect on employment duration.