Permutation Tests and the Bootstrap

Abstract

In this chapter we discuss two techniques, permutation testing and the bootstrap, of immense practical value. Both techniques involve random resampling of observed data and liberate statisticians from dubious model assumptions and large sample requirements. Both techniques initially met with considerable intellectual resistance. The notion that one can conduct hypothesis tests or learn something useful about the properties of estimators and confidence intervals by resampling data was alien to most statisticians of the past. The computational demands of permutation testing and bootstrapping alone made them unthinkable. These philosophical and practical objections began to crumble with the advent of modern computing.