Table 2.

A comparison of data set difficulty and posterior shape parameters

Data	$\mu$Iter	$\mu$MaxErr	$\mu$RMSD	Radius	95CI	Cred	Peaks
DS1	850,200	0.0819	0.0375	4	41	95	Y
DS2	8200	0.0976	0.0272	2	5	95	N
DS3	12,800	0.0757	0.0225	4	16	95	N
DS4	160,800	0.1139	0.0332	6	210	95	Y
DS5	626,000	0.0864	0.0163	16 (8)	240,311	38.9	Y
DS6	397,000	0.1046	0.0244	12 (7)	157,435	39.1	Y
DS7	62,600	0.1616	0.0397	9	735	95	Y
DS8	283,400	0.0882	0.0205	8	3545	95	N
DS9	347,200	0.1063	0.0208	23	712,502	0.6	?
DS9-U	255,200	0.1019	0.0216
DS10	322,400	0.1087	0.0226	15 (12)	286,604	30	Y
DS11	338,200	0.0503	0.0119	24	712,502	0.6	?
DS11-U	167,000	0.0533	0.0143

Notes: The first three columns show the mean number of iterations required to reach ASDSF less than 0.01 ($\mu$Iter) using the MrBayes default parameters (4 runs, 2 chains) as well as the resulting mean maximum split frequency error ($\mu$MaxErr) and mean split frequency RMSD ($\mu$RMSD) as compared with the golden runs. From the golden runs, we considered properties of the top trees—the at most 4096 highest probability trees from the 95% credible set. We inferred the SPR radius (Radius) which we define as the maximum SPR distance from any top tree to the topology with highest posterior probability, the size of the 95% credible set (95CI), the cumulative posterior probability of the top trees (Cred), and the presence of peaks. Note that our credible set clearly underestimates the true credible set size when it exceeds the number of samples (e.g., DS9 and DS11). “-U” data sets include only one member from each set of identical sequences. Note that each golden run contained 750,000 samples.