Abstract
The hazard plotting method using a hazard paper has been widely used to analyze randomly censored lifetime data because of its simplicity and a graphical effectiveness. However, until now small sample properties of the method have not been investigated well. At first, this article explicitly examines biases of the estimators for a cumulative hazard function in case of there being a proportional hazard relationship between failure time and censoring time. Secondly, based on the fact that the method takes account of only the rank of censored time but not fully its length, a modified hazard plotting method will be proposed and its properties are compared with the ordinary one assuming Weibull distributions by Monte Carlo simulation. The results show that a modified method has excellent properties when the shape parameter of a failure time distribution is greater than that of a censoring time distribution.
