Date: Mar 17, 2013 12:06 PM
Author: Luis A. Afonso
Subject: Re: An experiment . . . at random  in intra-Permutations

Kruskal-Wallis test with Rank Intra-Permutations

The common test statistics is

___T = 12/N*(N+1)* [Ri^2/ni] - 3*(N+1)

where N=total values of the Table, Ri = sum of runks sample i <= i <= k, [ ] summation (Gauss notation),
ni= size sample i.


I dare what happens if we change, with E(mmRi)= mean(Ri),

___T(mm) = 12/N*(N+1)* [ni*(mmRi)^2] - 3*(N+1)
(mm = modifyied mean)

Can we achieve a set of T(mm), calculate the respective 0.025, 0.975 quantiles and consequently to find out a
confidence interval without to base the decision on the
T Distribution which assumes normal data?

Luis A. Afonso