Box and Cox power-transformation to confined and censored nonnormal responses in regression

We describe a Box and Cox power-transformation to confined and censored nonnormal responses in regression. We consider the most general situation where responses include four types of observations, i.e., right, left, interval-censored and exactly specified ones. To evaluate the performance of the estimates from the proposed approach, we perform the simulation study with several simple regression models. Also, one real example is given to illustrate some aspects of the approach. Finally, we discuss an application of the approach to discrete and ordered categorical responses. (c) 2007 Elsevier B.V. All rights reserved.