Tuesday, November 26, 2013

Bug fix for anc.Bayes

phytools has a function, anc.Bayes, for ancestral state estimation of continuous traits using Bayesian MCMC. In preparing a tutorial on ancestral character estimation for my phylogeny methods graduate class, I uncovered a serious error in how the prior probability was calculated. I have now fixed it and posted the code, along with a new phytools build (phytools 0.3-79) that contains this fix.

If you used anc.Bayes with an uninformative prior, this bug probably did not affect you. However, if you tried a strong informative prior in your analysis, then prior versions of this function will not have worked properly.

I also (re-) discovered that the way this function handles specification of the prior and other control parameters of the MCMC is pretty annoying. The function is probably due an overhaul! Hopefully, I can get to that at some point.

No comments:

Post a Comment