Discussion of paper by D. M. Hershfield and M. A. Kohler, ‘An empirical appraisal of the Gumbel Extreme-Value Procedure’

Abstract

In Tables 1, 2, and 3 Hershfield and Kohler [1960] show a comparison of the estimated number of events that have a return period of 2, 5, 10, ... years with the corresponding computed number of events based upon the Gumbel method of curve fitting to extreme-value data. They point to a bias in the results of Table 1 using the dependent data and to an apparent lack of bias in Tables 2 and 3 when they test their results on independent data. Ordinarily one expects that an equation of estimate will perform better upon the data sample from which it is derived than it would upon an independent sample. But the authors apparently suggest that the reverse is true.