An asymptotic measure of accuracy effect from censorship in parametric estimation

Abstract

An asymptotic measure is provided to evaluate the effect on loss of accuracy for censored data in parametric estimation of location and scale parameters. With this measure, it is shown that the amount of effect from censored data relative to noncensored data is invariant of the actual values of the location and scale parameters, but is only dependent on the form of underlying distributions which the data are originated. In addition, among the most well-known distributions, obtained results for the measure show that two censored data values together usually may possess more information than one noncensored data value in the parametric estimation for location and scale parameters.