Friday, May 2, 2008

Are we there yet?

Selecting an appropriate burn-in point is critical for Bayesian analyses. Many people continue to do this arbitrarily, by excluding, for example, the first 10% of trees sampled. This seems silly and wrong. Others visualize their posterior scores in programs like Tracer and or Microsoft EXCEL and eliminate the set of initial trees possessing likelihood scores that are obvious outliers. This is also a bit arbitrary. An even more sophisticated approach involves visualization of split posteriors using the on-line application AWTY (Are We There Yet). Seems like a cool idea, leading me to wonder why it isn't more widely used. Probably because its more or less impossible to figure out what its doing!

No comments: