ELSEVIER Statistics & Probability Letters 36 (1997) 9-21
STATISTII~ &
Studentized permutation tests for non-i.i.d, hypotheses
and the generalized Behrens-Fisher problem
Arnold Janssen*
Mathematical lnstitut, University of Dfisseldorf Universitiitsstrasse1, D-40225 Diisseldorf Germany
Received 1 December 1995; receivedin revised form 1 December 1996
Abstract
It is shown that permutation tests based on studentized statistics are asymptotically exact of size ~ also under certain
extended non-i.i.d, null hypotheses. To demonstrate the principle the results are applied to the generalized two-sample
Behrens-Fisher problem for testing equality of the means under general non-parametric heterogeneous error distribu-
tions. Within this setting we propose a permutation version of the Welch test which is an extension of Pitman's
two-sample permutation test. These results are special cases of a conditional central limit theorem for studentized
permutation statistics which also applies to asymptotic power functions. © 1997 Elsevier Science B.V.
AMS classification: 62G 10; 62G09
Keywords: Behrens-Fisher problem; Permutation test; Permutation statistic; Conditional central limit theorem; Survival
test
1. Introduction and examples
Throughout, we are concerned with a two-sample non-parametric testing problem given by an extended
null hypothesis Ho which is strictly larger than the restricted null hypothesis of i.i.d, random variables. The
restriction of Ho to the i.i.d, case is denoted by iZlo, i.e. 1-7Io ~ Ho. A famous example is the extended
Behrens-Fisher problem where equality of two means is tested for a model with different unknown variances
of the error variables. For this type of examples the choice of critical values (under Ho) is a serious problem.
As message of this paper we suggest to carry out the underlying tests as permutation tests for studentized
statistics which includes a variance correction of the permutation distribution. It is shown that in this case the
conditional critical values of the permutation distribution work well at least in the asymptotic setting. We
follow the approach of Neuhaus (1993), who successfully applied studentized permutation tests for survival
problems, see Example 1.2 below.
* Tel.: + 49 21181 12165; fax: + 49 211 81 13117; e-mail:janssena@clio.rz.uni-duesserdorf.de.
0167-7152/97/$17.00 © 1997 Elsevier Science B.V. All rights reserved
PII S0 1 67-7 1 52(97)00043-6