After eliminating the invariant characters, we were left with simulated datasets of 25 to 30 characters to compare with the 100 characters that experienced no horizontal transfer, which is comparable to our empirical dataset of 24 pile-weave design characters and 98 non-pile-weave design characters.
Nearly identical results were obtained when invariant characters were excluded, as shown in simulation studies [27]; since significance probabilities for this test tend to indicate improved phylogenetic signal when p>0.01, and the test in general is overly conservative [28], [29], nEF1a was retained.
The data set contained 1512 invariant characters and 306 parsimony-informative characters.
This concatenated dataset contained 938 parsimony informative characters, 239 uninformative variable characters, and 1496 constant (invariant) characters.
Parameters were modelled using empirically derived nucleotide frequencies, proportion of invariant characters, gamma shape parameter, and six-parameter rate matrix model (see above).
Even with invariant characters (62% of total) removed, only Zygaenoidea were significantly non-homogeneous, due in part to the highly divergent taxon Epipyropidae.
Potential conflict between N- and C-termini was also evaluated with the partition homogeneity test (PHT), implemented in PAUP and excluding invariant characters.
To explore the effect of excluding characters, as we had done, on phylogenetic reconstructions, we also subjected the complete data set to reconstructions (but excluding invariant characters); this data set included 99 characters.
The parameter gamma (G) was favored for all data sets but not invariant characters (I) in the case of nuclear ribosomal genes, although this parameter was included in the second best fitting model.
