Table 2.

Pairwise Bayes factor matrix computed for models with different rate shift configurations simulated over the Velloziaceae dataset (k = number of rate shifts)

k	0	1	2	3	4	5	6	7	p.p.
0	1.00	0.12	0.001	0.0004	0.0003	0.0002	0.0003	0.0003	0.001
1	8	1.00	0.01	0.003	0.002	0.002	0.003	0.003	0.004
2	696	87	1.00	0.27	0.20	0.21	0.26	0.24	0.190
3*	2576	322	3.7	1.00	0.74	0.76	0.98	0.87	0.360*
4	3504	438	5.03	1.36	1.00	1.03	1.33	1.19	0.240
5	3392	424	4.87	1.32	0.97	1.00	1.29	1.15	0.120
6	2624	328	3.77	1.02	0.75	0.77	1.00	0.89	0.046
7	2944	368	4.23	1.14	0.84	0.87	1.12	1.00	0.026
8	2816	352	4.05	1.09	0.80	0.83	1.07	0.96	0.012

Values above the diagonal represent the prior probability of the corresponding model with k shifts. Values below the diagonal represent the Bayes factor (BF) evidence in favour of the model with k shifts (indicated by the list number) relative to the simpler model with k shifts (indicated in the column number). Last column indicates the relative posterior distribution (p.p.) of the models with k shifts. Usually, the overall best model from a BAMM analysis is the model with the highest BF relative to the null model, M₀ (k = 0). However, model probabilities for rarely sampled models are probably inaccurate (http://bamm-project.org/postprocess.html#bayesfactors), which seems the case of the null model for the Velloziaceae dataset, and the posterior probabilities indicate that the model with three shifts (marked by *) is the most frequent among the 95 % credible set of rate shift configurations sampled with BAMM (see text).

BF > 20: evidence for one model over another; BF > 50: very strong evidence in favour of the numerator model; BF < 5: weak evidence in favour of the numerator model.