we used PC
scores of the PCA on the original song data. PC scores were made
positive by adding the minimum score on each PC to the individual
values on that PC, since the method does not allow negative values
(Mooers et al. 1999). We calculated the fit of song to a polyphyletic
cladogram (Fig. 1b), with the same topology as the tree in Fig. 1a
(except that it was unrooted, which is a requirement of the
method), and to a monophyletic cladogram (Fig. 1c) obtained by
segregating Serinus and Carduelis but otherwise maintaining the
relationships between individual species. These topologies differ
only on whether the traditional genera are monophyletic, but not in
finer-scale details.We used the ‘free model’ of Mooers et al. (1999),
which does not make assumptions about branch lengths, but rather
adjusts them optimally to fit the data. Therefore, by comparing the
fit to these two topologies we tested for an effect of monophyly
only, but not of finer-scale differences in species relationships or of
evolutionary rates (i.e. preset branch lengths). Since the two
models are equally parameterized (same number of adjustable
branch lengths), their fit was compared with a simple likelihood
ratio test, with significance at a ¼ 0.05 indicated by a log-likelihood
difference of 2 (Edwards 1992; Mooers et al. 1999). Calculations
were made with the version of the CONTML program of PHYLIP
(Phylogeny Inference Package) version 3.0 (J. Felsenstein, Department
of Genome Sciences, University of Washington, Seattle, WA,
U.S.A.) modified by Mooers et al. (1999).