Confidence limits for normal type I censored regression

A new method for calculating confidence limits for regression analysis using normal Type I censored data is presented. The new method is compared to two existing methods via a simulation study. Over a variety of examples, the new method consistently shows good agreement with the simulation results. The other two methods do well in some cases but not well in others. Key aspects of the new method (which is an amalgam of existing methods) are its handling of the loss of degrees of freedom for error due to censoring and the effect of censoring on the regression (X′X) matrix. Examples discussed include data from Quantitative Structure Activity Relationship studies, as well as from chemical process studies.