Efficient estimation of the weibull shape parameter based on a modified profile likelihood

Abstract

The maximum likelihood estimator of the Weibull shape parameter can be very biased. An estimator based on the modified profile likelihood is proposed and its properties are studied. It is shown that the new estimator is almost unbiased with relative bias being less than 1% in most of situations, and it is much more efficient than the regular MLE. The smaller the sample or the heavier of the censoring, the more efficient is the new estimator relative to the regular MLE.