Quantification and relative severity of inflated branch-support values generated by alternative methods: An empirical example

► Clades uniquely supported by parametric methods are less robust. ► Methods that can create artifacts provide greater resolution and support. ► Partitioning matrices cannot be relied upon to obviate dubious likelihood support. ► Resampling artifacts should be accounted for in likelihood analyses. ► PhyML is more conservative than RAxML and GARLI for fully resolved trees.