Right-censoring and replicates effects for proportional intensitymodels

摘要

This paper not only reviews the robustness of the proportional intensity (PI) models (Prentice-Williams-Petersongaprntime (PWP-GT) and Andersen-Gill (AG)) for right-censored failure data, but examines the replicates effect forrnthe PWP-GT model. The case of examining two shape parameters of 1.0 and 1.5 with severe censoring is presentedrnfor replicates effect by doubling the number of experimental runs. The results indicate using replicates effectsrnimproves estimation accuracy. Experimental runs are associated with distinct computer seed numbers. An increasedrnnumber of experimental runs improves the model accuracy when data scarcity exists due to heavy censoring.