Table 1.

Model comparisons using Bayes factors calculated from marginal likelihoods in BEAST

Dataset	Preferred model combination	Non-preferred model combination	Bayes factor with interpretation
Monocots, cpDNA
	UCLN, BD, p	UCLN, BD, np	19510.4***
	UCLN, Yule, p	UCLN, Yule, np	19510.8***
	UCLN, Yule, p (best overall)	UCLN, BD, p	19.31***
Xanthorrhoeaceae, cpDNA
	UCLN, BD, np	UCLN, BD, p	26.53***
	UCLN, Yule, np	UCLN, Yule, p	29.66***
	RLC, BD, np	RLC, BD, p	38.85***
	RLC, Yule, np	RLC, Yule, p	59.98***
	UCLN, Yule, np	UCLN, BD, np	8.39**
	RLC, Yule, np	RLC, BD, np	16.26***
	RLC, Yule, np (best overall)	UCLN, BD, np	7.36**
Xanthorrhoeaceae, rpb2
	UCLN, BD, p	UCLN, BD, np	238.44***
	UCLN, Yule, p	UCLN, Yule, np	60.42***
	RLC, BD, p	RLC, BD, np	219.8***
	RLC, Yule, p	RLC, Yule, np	267.27***
	UCLN, BD, p	UCLN, Yule, np	16.08***
	RLC, Yule, p	RLC, BD, p	204.75***
	RLC, Yule, p (best overall)	UCLN, BD, p	249.98***

Within each dataset, model comparisons follow the nested sequence: partitioned (p) vs non-partitioned (np) loci; Birth-death (BD) vs Yule phylogenetic tree models; and uncorrelated lognormal (UCLN) vs random local clocks (RLC). Each row compares the bolded models, with other models held constant. For each dataset, the best model combination overall (highest marginal likelihood) is indicated, and in each case is (very) strongly preferred to the second best. Asterisks after the Bayes factors indicate their interpretation according to [27]: ** = “strong” (6 ≤ BF < 10) and *** = “very strong” (BF ≥ 10) evidence favouring the model with the higher lnL. Results are shown for path sampling only because those from stepping stone sampling were identical.