Randomization tests in ecology

It is the norm more than the exception in ecology to have small data sets that are not normally distributed. Hence, with the increasing availability of personal computers and swift diffusion of randomization tests, ecologists are rapidly adhering to these resampling techniques based on fewer assumptions. However, fewer assumptions does not mean no assumptions. Indeed, depending on the type of resampling tests used, ecologists may or may not be testing correctly their ecological hypotheses. In fact, randomization tests require the understanding of the "user" as to what are the defined null and alternative hypotheses of each test and make sure that it corresponds well to the biological or ecological context under study. Hence, caution must be exercised because the permutation procedures assume data independence, and this assumption is invalid when the data are spatially or temporally autocorrelated. The main goal of this symposium is to show how randomization tests should be with ecological data that are spatially and temporally structured.