FIG. 4.

Rate with which substitution models are not rejected by (a) GC tests and (b) PPSs for ten replicates generated by simulating stochastic evolution using a partitioned GTR + I + Γ model constrained to the ML topology identified for a shrew mitochondrial data set. GC tests typically did not reject models that incorporated both unequal base frequencies and transition/transversion substitution rates, especially when the models also included Γ. Conversely, PPSs often failed to reject even the simplest models as long as they incorporated Γ.