The problem of selecting the size of the tails from a sample drawn from a distribution of test scores for item analysis study is discussed. In the past the selected tails were viewed as independent samples. However, this is not the case. The tails are dependent. Given this we find that each tail should contain about 21% of the sample and not the traditional 27%. Use of 27%, however, is not far from optimal.
Get full access to this article
View all access options for this article.
References
1.
Chernoff, H., Gastwirth, J. L. and Johns, M. V.Asymptotic distribution of linear combinations of order statistics. The Annals of Mathematical Statistics, 1967, 31, 52-72.
2.
Cureton, E. E.The upper and lower twenty-seven percent rule. Psychometrika, 1957, 22, 293-296.
3.
Feldt, L. S.The use of extreme groups to test for the presence of a relationship. Psychometrika, 1961, 26, 307-316.
4.
Kelley, T. L.The selection of upper and lower groups for the validation of test items. Journal of Educational Psychology, 1939, 30, 17-24.
5.
Mosteller, F.On some useful "inefficient" statistics. The Annals of Mathematical Statistics, 1946, 17, 377-408.
6.
Ross, J. and Weitzman, R. A.The twenty-seven percent rule. The Annals of Mathematical Statistics, 1964, 35, 214-221.
