Tree-Length Distribution Skewness: An Indicator of Phylogenetic Information

Computer simulations in which phylogenies were generated under various conditions were used to examine the relationship between the phylogenetic signal of a character data set, the skewness of the tree-length distribution, and the position of the real tree relative to the most parsimonious tree for a four-character-state system. Character data that are consistent with one phylogenetic hypothesis produce tree-length distributions that are highly skewed to the left, whereas character data consistent with many phylogenetic hypotheses produce more symmetrical tree-length distributions that cannot be distinguished from tree-length distributions produced by random character data. The relationship between the skewness of the tree-length distribution, as measured by the g1 statistic, and the amount of phylogenetic signal varies depending on tree topology, the proportion of the sequence that is informative (e.g., informative change at only the first codon position as opposed to informative change at the first and second codon positions), and the number of sequence data. However, in all simulations that produced tree-length distributions that were significant at the 1% level under Hillis's (1991, in Phylogenetic analysis of DNA sequences M. M. Miyamoto and J. Cracraft, eds., Oxford Univ. Press, Oxford, England in press) statistical test, the real tree was very close to or was the most parsimonious tree. Conversely, in trees with nonsignificantly skewed tree-length distributions, the real tree could be almost anywhere in the distribution. However, the real tree tended to be closer to the most parsimonious tree than would be expected at random. This study suggests that tree-length distribution skewness provides a useful criterion for judging the phylogenetic information content of character data. Tree-length distribution; skewness; phylogenetic information.