The Pearson and Johnson Systems
This chapter re-examines two of the best-known systems of parametric distributions: the Pearson and the Johnson. It is shown that, in the Pearson system, Pearson Types III and V are boundary embedded models of the main Types I, IV, and VI. A comprehensive way of finding the best type to fit is given using appropriate score statistics to guide a systematic search of all model types, including symmetric boundary models. Maximum likelihood estimation is used and details of its numerical implementation are given. Type IV can be a difficult model to fit. A method is discussed for this model that is reasonably robust, subject to certain restrictions on the parameter values. The same examination is made of the Johnson system where the lognormal, SL family is shown to be an embedded subsystem of both the main subsystems SB and SU. Two real data examples are given.