Combinability of phylogenies and bootstrap confidence envelopes.

Recently, Lutzoni (1997) used a test described by Rodrigo et al. (1993; hereinafter referred to as the RKB3 test) to compare phylogenies constructed by using ribosomal RNA gene sequences (rDNA) and intergenic transcribed spacer sequences (ITS) from a group of related basidiomycete species. Lut-zoni discussed the relationships amongst the species, but he also commented extensively on the RKB3 tests, particularly on a presumptive bias in the method and a proposed workaround. Here, I conne myself to a discussion of Lutzoni's assessment of the RKB3 test. In particular, I suggest that (1) the bias described by Lutzoni is an arti-fact of inadequate bootstrap sampling and (2) Lutzoni's proposed workaround may result in misleading inferences. The RKB3 procedure is designed to test whether two (or more) cladograms, each derived from different data, are sample estimates of the same parametric tree. In other words, the RKB3 test is a test of whether the true (or parametric) phylogenies are identical , not just similar. Lutzoni (1997) and also Cunningham (1997) have evaluated the RKB3 procedure on the basis of whether it will allow us to decide whether data can be combined to produce a better estimate of phylogeny. Although there is nothing objectionable about this, that is not what the RKB3 procedure was designed to do. In the same way, for instance, the classical Student's t test was not designed to test whether and how data should be pooled to obtain an overall estimate of the population mean. With the RKB3 test, the sample phylogenies may be constructed from different types of data (e.g., sequences, morphological characters , restriction fragment length polymor-phisms, behavioral and quantitative traits) and using very different tree-reconstruction methods. Pooling different data types can easily present problems, but for each type of data one should be able to compare the estimated (or sample) trees to test the hypothesis of parametric tree identity. The RKB3 test does this by constructing a set of boot-strapped trees from each data set and comparing these bootstrap prooles with each other. The bootstrap prooles are equivalent to condence limits or envelopes (Felsen-stein, 1985), in that they map in tree-space the variability of our phylogenetic sample estimates. Rodrigo et al. (1993) argued that if the condence envelopes of two data sets do not overlap (i.e., no trees are common to both bootstrap prooles), then the data fail to support the hypothesis that the different sample trees are …