Subset selection of the best treatment via many-one tests

Abstract

The method of Gupta (1956, 1965) was developed to select a subset from k normal populations that contains the best populations with given probability. This paper shows a duality between the general goal of selecting a subset for the best population and many-one tests. A population should be regarded as ‘candidate’ for the best population and thus retained in the subset if the samples from the other populations are not significantly better. Based on this ‘idea’ a general selection procedure is proposed using many-one tests for the comparison of each population against the remaining ones.

Keywords:

• Subset selection procedure
• best population
• multiple comparisons
• many-one test
• unequal sample sizes
• nonparametric methods