ABSTRACT
This paper developed the so-called ‘product limit’ estimator for a survival function when the data are subject to non-informative censoring. In this way all the data could be used in estimating survival curves, e.g. individuals followed for only 2 years still make a contribution to the estimated 5-year survival rates. The product-limit estimate takes the form of a product of terms of the form (1 di/ni), where there is one term for each failure time t1 t2 . . . tk, indexed in increasing order, di is the number of failures at time ti and ni is the ‘number at risk’ just before time ti .
