In article <8oguil$pnu$1 at mercury.hgmp.mrc.ac.uk>,
Nicole Dubilier <ndubilie at postgate.mpi-bremen.de> wrote:
>I'm a bit worried about bringing down the wrath of the entire ML
>community on me for asking such a stupid question, but here goes:
Nick Goldmans 1993 paper in JME (there are two and I can
never remember which one it is in) where he suggests you
can test how bad your likelihood is by comparing it with
the 'unconstrained likelihood' (the probability of observing
the data if all patterns of nucleotides at a site are equally
likely). You have to simulate to obtain the null distribution
of the ratio of these two likelihoods (see paper for details).
I think this test is likely to always reject your model in
favour of the unconstrained likelihood, but then nothing
ruins good theory like data.