2.2.2. Likelihood Analyses
Gap characters are binary, so a two-state Markov model (the Cavender-Farris-Neyman [CFN] model [47±49]) is appropriate for their analyses, at least in principle. However, all observed gap characters are by definition variable²their occurrence differs among taxa, otherwise they would not EHGLVFHUQLEOH7KXVJDSFKDUDFWHUVH[KLELWDQ³DFTXLVLWLRQELDV´VLPLODUWRWKDWIRXQGLQWSLFDO discrete morphological character matrices [50]. The acquisition bias for morphological data reflects the fact that most researchers only score parsimony informative characters; the failure to score uninformative characters is analogous to the inability to recover invariant gap characters (Felsenstein [51] referred to DVLPLODUSKHQRPHQRQIRUUHVWULFWLRQVLWHGDWDDV³DVFHUWDLQPHQWELDV´%HFDXVHRIWKLVLVVXHZH employed a corrected CFN model that accommodates acquisition bias (we call this the CFNv model). The CFNv model is a special case of the more general Mkv model proposed by Lewis [50]; readers are referred to that publication for details. ML analyses using the CFNv model were conducted in PAUP* and GARLI v0.951 [52] after we converted the binary (01) gap characters to RY codes ĺ5ĺ<