Interval estimation of slope with right-censored data

Abstract

SUMMARY A new method for point and interval estimation of a slope that uses randomly right- censored data is presented. The new method is based on inverting a censored data version of a test for association. A simulation study shows that the new method compares favourably with a method due to Buckley & James (1979). In this paper, we consider nonparametric estimation of the slope parameter in a simple linear regression model when the data are randomly right censored. The three main nonparametric methods for the estimation of slope in a linear regression model are due to Miller (1976), Buckley & James (1979), and to Koul, Susarla & Van Ryzin (1981). A fourth method due to Cox (1972), though used extensively in the regression analysis of right-censored data, is not directly comparable to the first three methods, because it is based on a proportional hazards model. Miller & Halpern (1982) compare the four methods and come out generally in favour of the Cox and Buckley & James methods, finding that the other two methods have some methodological weaknesses. Miller & Halpern conclude that the choice between the Cox and Buckley- James methods must be based on the appropriateness of the proportional hazards model or the linear model. In this paper we propose a new method for point and interval estimation of the slope parameter in a simple linear regression model. Under suitable regularity conditions, the