|
Abstract:
|
We investigate Maximum Likelihood Estimation (MLE) and Median Rank Regression (MRR) for estimating parameters of a Weibull Distribution under Type I censoring. Through simulation, we compare both methods using the empirical mean and RMSE of estimates as performance criteria. The simulation experiment evaluates the effects of several factors, such as the Weibull shape parameter value, number of censoring times, fraction failing and expected number of failures before a given censoring time. Both MLE and MRR tend to overestimate the Weibull scale parameter when the number of expected failures is small, but the MLE procedure typically has smaller bias RMSE. The same behavior generally holds for shape parameter and quantile estimates. However, under certain conditions when the expected number of failures is small, the MRR procedure may yield quantile estimates with better accuracy.
|
