Becquet and Przeworski look at both of these programs (and MIMAR is one of their own) and ask the simple question of how they perform when some of their underlying assumptions are violated. Their answer: not as well as one might like. When ancestral populations are structured, both IM and MIMAR can lead to rejections of a true allopatry (isolation only) model, even when there is no post-separation gene-flow. Moreover, parameter estimates – including divergence time and ancestral population sizes – become unreliable when there is ancestral population structure or when gene-flow rates change through time. Real population histories likely involve both time-varying levels of gene flow and complex current and historical population structure and it is not clear how to develop methods that are robust to a broad range of potential splitting scenarios. Approximate Bayesian methods, anyone?
Reporting on the Reptiles of Redonda
18 hours ago