抄録
Optimal sample size for analyzing doubly exponential data is discussed from the standpoint of variance and efficiency of the estimator for mode. The proposed concept is available for the characteristics which follow the first asymptotic distribution of the largest values as well as the smallest values. A numerical analysis of variances and efficiencies of the above estimator draws the conclusion that the optimal sample sizes are n=5,10 and 20. n=10 is recommended as the standard sample size, n=5 for the reduced test or routine purpose, and n=20 for the tightened test of analytical purpose. The very procedure which constructs the body of the present paper verifies also that the estimator for mode proposed by this author is valid when not only 2≦n≦20 but 21≦n.
