RESAMPLING METHODS FOR COMPUTATION-INTENSIVE DATA ANALYSIS IN ECOLOGY AND EVOLUTION

The advent of fast, relatively inexpensive (thus, widely available) microcomputers is transforming the way we analyze data in ecological and evolutionary research. Even more profound, however, are the associated changes in questions asked, empirical methods used, studies conducted, and interpretations offered. Now that an array of computation-intensive statistical methods is newly available for general use, it seems particularly important to assess their advantages and limitations, to note how they are currently being used, and then to consider implications for the future. I focus in this review on four related techniques known in the statistical and biological literature as randomization (or permutation) tests, Monte Carlo methods, bootstrapping, and the jackknife. I refer to them collectively as resampling methods, because each involves taking several-to-many samples from the original data set (randomization, bootstrap, jackknife) or from a stochastic process like the one believed to have generated the data set (Monte Carlo). Each of these methods is actually an extensive family of techniques and specific applications that cannot be thoroughly examined here; instead, I briefly characterize the focal methods and then survey the recent literature in ecology and evolution to identify the issues most frequently associated with these techniques. It emerges that resampling methods are well represented in