ABSTRACT
In th is chapter we give a brief description of m ethods th a t require extensive computing. Three classes of procedures th a t are of in terest in the context of nonparam etric sta tis tics are (1) perm utation tests (2) random ization tests and (3) boo tstrap techniques. It should be noted th a t the tools used for the perm utation tests and the random ization tests are the same, b u t the con tex ts in which they are used are diffferent. W hen the d a ta are collected from a designed experiment, trea tm en ts usually are assigned random ly to the ex perim ental units, and we use random ization tests, which are distribution-free procedures. If the d a ta are random samples and under the null hypothesis, all the relevant perm utations of the d a ta are equally likely, and we use permu ta tion tests. The procedures based on ranks, which have been discussed thus far, are a class of perm utation tests. The discussion is restric ted to the analysis of continuous responses and is
related to the procedures we have stud ied in the previous chapters. All these m ethods to be discussed can be called resampling methods. As this name implies, the methods use samples drawn from the d a ta a t hand. Since we are sampling the d a ta (the sample) a t hand, not the populations, the sampling process is ap tly called the resampling process. We discuss perm utation tests and random ization tests in Section 7.2. The final section deals w ith boo tstrap methods.
