A new method of fitting the JohnsonSBdistribution using maximum likelihood (ML) procedures is described. Commencing with assumed values for the location and scale parameters, estimates of all four parameters are found iteratively. This does not have the disadvantages of the method of moments for which estimates of the higher moments are required and the ambiguities of the method of percentiles in which estimates of parameters depend on the choice of percentiles. Application is made to sequences of annual maximum daily rainfalls in which the kurtosis is low when compared with that corresponding to the theoretical lognormal distribution for the observed skewness. In some cases the ML procedure is not feasible. Problems encountered with ML estimation and inadequacies arising from short samples on the choice of distribution are discussed.
展开▼
