Taxonomic Congruence Re-Examined

Abstract

An attempt was made to repeat computations reported in 2 previous studies of taxonomic congruence, which concluded that the Wagner method yielded much more stable classifications than did phenetic clustering methods. Two of the 9 data sets used in the previous studies could not be reconstructed from the cited sources. None of the values for the consensus index could be confirmed, 1 reason being that an inappropriate method for computing consensus trees was used which was dependent upon an arbitrary and unspecified system of labeling the internal nodes of the trees being compared (i.e., Adams-1 consensus trees were used rather than Adams-2 trees). In the majority of the data sets, trees were shorter than those reported in the previous studies. For many of the data sets, a number of distinct tree topologies had the same minimum length. In such cases the congruence between 2 data sets was estimated both by comparing all possible pairs of trees between two data sets and by computing the strict consensuses of trees found within a data set. Strict consensus trees were used in the comparisons, since Adams-2 trees too often contain sets that are not consistent with either of the 2 original trees. Since the CIM [maximum] consensus index has been shown to have the undesirable property of being very sensitive to the symmetry of the trees being compared, other indices (CIC and CIRI) were computed. Contrary to the results of the previous studies, classifications based on minimum length trees were about as stable as those based upon the UPGMA [unweighted pair-group method of cluster analysis using arithmetic averages] clustering method. The results varied from data set to data set, but the averages for the 2 methods were very similar (and both were very poor). The implications of these results are discussed.