Dan L. Warren, Jamie M. Kass, Evan Economo

In the field of niche modeling, data are often subject to multiple interacting sources of uncertainty, bias, and autocorrelation that make them difficult to analyze using traditional statistical approaches. Randomization is often used in statistical tests in order to estimate distributions that are difficult to specify analytically. Decades of development in the niche modeling literature have resulted in randomization tests that allow us to study phenomena as disparate as variable importance, methodological bias, and patterns of niche evolution. Here we present a novel conceptual framework that allows us to both take a synthetic view of existing tests and highlight potentially fruitful avenues for future methodological exploration. We argue that further development of randomization tests and rigorous exploration of their performance will be essential to the development of the field going forward.