BioGeobears
The probability of ancestral ranges was also estimated using BioGeoBEARS (Matzke, 2013) in R (R Core Team, 2016). Pigeon species were assigned to the same eight regions as the Markov Maximum likelihood inference. The maximum range size was set to eight regions as some species are widespread across all regions. Other settings were left as default. We used the standard six different biogeographic models tested by BioGeoBears: BAYAREA, which does not allow dispersal at cladogenesis; DIVA, which allows vicariance [i.e. if parent lineage occurs in (x,y), then one daughter lineage can occur in only (x) whereas the other can occur in only (y)] but disallows subset speciation, in which one daughter lineage inherits a subset of the distribution of the parent [i.e. if parent lineage occurs in (x,y), then one daughter lineage can occur in only (x) or only (y) whereas the other can occur in (x,y)]; DEC, which allows both vicariance and subset speciation; and BAYAREA+J, DIVA+J and DEC+J, which are equivalent to the previous three models but also allow jump dispersal through founder effects at cladogenesis(Matzke 2014). The BioGeoBEARS analysis was run using two separate phylogenetic inputs: 1) using the BEAST MCC tree and 2) a subset of 100 BEAST trees. Results of the BioGeoBEARS run across multiple trees were averaged. Repeated analyses were run to account for possible different topologies and poorly supported nodes in the phylogeny. The best-fit model for the pigeon data was DEC+J, followed by DIVALIKE+J (LnL −444.3 and −455.1, respectively; Table S7).