Coding quantitative character data for phylogenetic analysis: A comparison of five methods

Five character construction methods were compared for a data set of 25 quantitative morphological characters sampled from the Guvenia superba complex. The methods were simple-gap, gap coding, gap-weighting, analysis of variance followed by a multiple range test, and an "arbitrary" coding method. Three indices, data decisiveness, consistency index, and skewness (g(1)), were used as estimators of phylogenetic signal. Performance of coding methods was evaluated by the following criteria: (1) number of informative characters, (2) number of equally parsimonious trees, (3) clade support measured by bootstrapping, and (4) phylogenetic signal, compared using randomization tests that established critical values for the three indices and allowed them to be compared between coding methods. This study demonstrates that a large number of characters are accessible for phylogenetic analysis when continuous characters are included. The quantitative character data for the Govenia complex contain significant phylogenetic information when the gap-weighting method is used to construct character states. This indicates that methods that divide variation into small segments and allow overlapping of measurements recover a strong phylogenetic signal, and yield small number of well-resolved trees with strong bootstrap support and statistically significant phylogenetic signal. In contrast, methods such as gap coding or analysis of variance-multiple range test lead to data matrices with few informative characters and many equally parsimonious trees and weak phylogenetic signal. Moreover, the simple-gap coding method data matrix failed to recover significant phylogenetic signal.