Search for a command to run...
Treating all available characters simultaneously in a single data matrix (i.e. combined or simultaneous analysis) is frequently called the ‘total evidence’ (TE) approach, following Kluge's introduction of the term in 1989, quoting Carnap (1950) . However, the general principle and one of the possible procedures involved in its application are often confused. The principle, first enunciated within the context of inductive logic by Carnap in 1950, did not refer to a particular procedure, and TE meant using all relevant knowledge, rather than a combined analysis of all available data. Using TE, all relevant knowledge should be taken into account, including the fact that some data are probably misleading as indicators of species phylogeny and should be discarded. Based on the assumption that molecular partitions have some biological significance (process partitions obtained from nonrandom homoplasy or from ‘processes of discord’), we suggest that separate analyses constitute an important exploratory investigation, while the phylogenetic tree itself should be produced by a final combined analysis of all relevant data. Given that the concept of process partitions is justified and that reliability cannot be evaluated using any robustness measure from a single combined analysis, the analysis of multiple data sets involves five steps: (1) perform separate analyses without consensus trees in order to assess reliability of clades through their recurrence and improve the detection of artifacts; (2) test significance of character incongruence, using, for example, pairwise ILD tests in order to identify the sets responsible for incongruence; (3) replace likely misleading data with question marks in the combined data matrix; (4) perform simultaneous analysis of this matrix without the misleading data; (5) assess the reliability of clades found by the combined analysis by computing their recurrence within the previous separate analyses, giving priority to repeatability.