A review of the Incongruence Length Difference randomisation test: uses and abuses

Abstract

The incongruence length difference (ILD) randomisation test of Farris et al. (1995a, 1995b) is often applied to systematic datasets comprising qualitatively or logically distinct data partitions (e.g., molecules and morphology, different loci). It has variously been used to assess phylogenetic accuracy (1), data combinability (2), difference in evolutionary history (3) and difference in evolutionary rate (4). Several authors have noted that the ILD test is not designed to address all of these issues, particularly the first two, and that the inferences drawn from test results are often questionable. Here, I quantitatively review the usage of the ILD test in more than 300 papers published between 2009 and 2010 (an almost exhaustive sampling). RESULTS: There is very little consensus on how best to implement the ILD test. There is often no justification for the number of replications used, or for the search parameters specified. In many cases, the settings used are not reported at all, making it impossible to reproduce the results. Where p-values are not reported, results cannot be re-validated. Finally, there is circumstantial evidence that some authors appear to be unaware that this is a parsimony-based test. A review of the proper use and reporting of the ILD test is therefore overdue