ABSTRACT
In this chapter, we address censored data in the context of regression analysis, and discuss the biases associated with non-model-based analyses. Likelihood-based methods are shown to reduce such biases. The chapter starts by estimating the distribution of the response variable (Y), without any predictor variable (X), in the presence of censoring. This method is then extended to estimate the conditional distributions of Y, given X variables, by using regular and semi-parametric likelihood-based methods, including the Cox Proportional Hazards model. The Tobit regression model is presented as another censored data model, and interval-based censoring is introduced.
