Split support and split conflict randomization tests in phylogenetic inference.

Randomization tests allow the formulation and statistical testing of null hypotheses about the quality of entire data sets or the quality of fit between the data and particular phylogenetic hypotheses. Randomization tests of phylogenetic hypotheses based on the concepts of split support and split conflict are described here, as are tests where splits, rather than the data, are randomly permuted. These tree-independent randomization tests are explored through their application to phylogenetic data for caecilian amphibians. Of these tests, split support randomization tests appear to be the most promising tools for phylogeneticists. These tests seem quite conservative, are applicable to nonpolar data and unordered multistate characters, and do not have the problems of nonindependence that affect split conflict and hierarchy tests. Unlike split conflict tests, their power does not appear to be correlated with split size. However, all tests are sensitive to taxonomic scope. Split support tests may help discern data that are likely to be affected by the problems of long-branches effects. Comparison of test results for mutually incompatible splits may help identify the presence of strong misleading signals in phylogenetic data. Significant split support could be a prerequisite for considering phylogenetic hypotheses to be well supported by the data, and split support randomization tests might be usefully applied prior to or as part of tree construction.