Fig. 3

a Chronogram based on a Bayesian approach using a coalescent prior under an uncorrelated lognormal relaxed clock model and assuming constant population size of Psittacanthus schiedeanus sequences in BEAST. Pink bars indicate 95 % highest posterior density (HPD) intervals for nodes of particular interest. These nodes all have posterior probabilities above 0.9. b *BEAST model with simulation run with all samples as two lineages separated by the Isthmus of Tehuantepec (WEST and EAST), (c) samples as six lineages corresponding to geography and mountain ranges (nSMO, cSMO, sSMO, CHIS, OAX, BREE) or (d) three separate lineages according to habitat (SCHI cloud forests from San Luis Potosí to Oaxaca and Chiapas, CALY xeric vegetation in central Oaxaca, BREE tropical deciduous forests in Chiapas). Maximum-likelihood scores are indicated for *BEAST analyses (b–d). The likelihood scores under the three species delimitation hypothesis (SCHI, BREE, CALY) are compared with ln Bayes factors (BF) tests. The three-species hypothesis produced a higher likelihood score than those for the alternative hypotheses although the difference was not very strong (2 × ln BF; six-species hypothesis versus three-species hypothesis = 11.228; two-species hypothesis versus three-species hypothesis = 4.046; six-species hypothesis versus two-species hypothesis = 7.182)