Discussion by Walter Liggett

Abstract

The author describes applications to astronomy of a variety of established statistical methods for censored data, concludes that such applications are improvements over past practice, and raises the issue of the appropriate use of established and innovative statistical methods in astronomy. The issue of appropriateness is the focus of this discussion. There is little question that an astronomer can do much worse than using established statistical methods. However, the issue of which methods are entirely appropriate is more difficult and leads to examination of the basis for judging a method appropriate, that is, for judging when the assumptions underlying the method are satisfied. In many cases of censored data, the underlying assumptions cannot be verified without additional information. For example, the censored data alone are not adequate for verification of the assumptions of the Kaplan-Meier estimator (Tsiatis 1975). Such inadequacies in the data result in qualification of the conclusions drawn from astronomical data. From the astronomy viewpoint, such qualification might be quite different from the qualification encountered in other applications of censored data methods.

National Institute of Standards and Technology, Admin A377, Gaithersburg, MD 20899.