GENERALIZATION OF THE METHOD OF MAXIMUM-LIKELIHOOD

The method of 'maximum likelihood' (MLH), frequently used to study experimental findings in accordance with Weibull statistics, is more biased, the smaller the number of samples. Semi-empirical corrective coefficients are proposed to attempt to correct the deficiency, which is unsatisfactory from a theoretical standpoint. In this paper, it is demonstrated that MLH contains an implicit hypothesis, which validity is discussed. Generalization of the basic MLH formulae enables other hypotheses to be chosen. In particular, if the hypothesis chosen is derived from the 'Median Rank Function', the whole becomes coherent.