Efficiencies of MLE and BLUE for parameters of an accelerated life-test model

Maximum-likelihood estimates (MLE) and best-linear-unbiased estimates (BLUE) of parameters of an accelerated life-test model were compared according to the mean-square error criterion in small and moderate sample size cases. A factorial experiment was used to investigate the effects of the: (1) number of levels of the acceleration variable, (2) percent censorship, and (3) sample size. Overall, the two kinds of estimators were of comparable efficiencies. However, where there were two or less uncensored observations, at any level(s) of the acceleration variable, the MLE was clearly favored. This is in accord with findings for single sample cases, where MLE and BLUE for location and scale parameters of the Weibull and the extreme-value distributions were compared. >