Dedication:To the memory of Rossiter H. Crozier , an evolutionary biologist, who with his generosity and inquisitiveness inspired many students and scientists, in Australia and abroad. doi: bioRxiv preprint CONFIRMATION BIAS IN PHYLOGENETICS 4 distributed processes (the IID assumption). If model-based molecular phylogenetic methods are chosen, it is also assumed that the evolutionary processes operating at the variable sites can be modelled accurately by time-reversible Markov models, and that the processes were stationary, reversible and homogeneous 56-58 over time (the assumption of SRH conditions).In practice, the choice is one between methods assuming that the underlying evolutionary process can be modelled using a Markov model of nucleotide or amino-acid substitutions (i.e., distance methods 59-64 , likelihood methods 59, 61, 63-69 , Bayesian methods 70-75 ), or using non-parametric phylogenetic methods (i.e., parsimony methods 59, 61, 63, 64,[76][77][78] ). In reality, most researchers analyze their data by using different model-based phylogenetic methods, and reports that only use parsimony methods are increasingly rare. Depending on the chosen phylogenetic method, researchers may have to select a suitable model of sequence evolution (i.e., a substitution model and a rate-across-sites model) to apply to the sub-MSA.This choice is often made by using model-selection methods 6, 79-90 .Having chosen the phylogenetic method and, in relevant cases, a suitable model of sequence evolution, the next step involves obtaining accurate estimates of the tree and evolutionary processes that led to the data. There is a plethora of programs that implement phylogenetic methods 59-78 . Depending on the methods chosen, users often also obtain the nonparametric bootstrap probability 91 or clade credibility 92 to measure support for individual divergence events in the phylogeny. These estimates are often thought of as measures of the accuracy of the phylogenetic estimate or the confidence we might have in the inferred divergence events. Doing so might be unwise because they are only measures of consistency 93 (e.g., a phylogenetic estimate may consistently point to an incorrect tree).Having inferred the phylogeny, with or without bootstrap probability or clade credibility for all of the internal branches (edges), the final step in the protocol is to interpret the result.Under some conditions-most commonly the inclusion of out-group sequences-the tree can be drawn and interpreted as a rooted phylogeny, in which case the order of divergence events and the lengths of the individual edges may be used to infer, for example, the tempo . CC-BY-NC-ND 4.0