In article <8hpas2$jlo$1 at mercury.hgmp.mrc.ac.uk>, Edna Huelsenbeck
<huelsen at brahms.biology.rochester.edu> wrote:
> I am not certain that a relative rates test is what you want
> to do, as that tests the molecular clock assumption. It seems
> to me that you want to test the null hypothesis that:
I am not convinced that your choice of null hypothesis is correct:
Given that Dmitri has found that the data fit an A-B-C-D type
tree then H1 under your test will be very similar to your H0
(branches leading to B and C close or at zero). So this will
test whether they are actually zero.
Given that he has observed a tree close to H0, picking H0
to be A-B-C-D, it is unlikely to be rejected. If for example,
he had observed that the branches leading to B and C were
exactly 0, then you wouldn't be testing anything at all
(what is the chance that my observed hypothesis could
be got by chance if it is also the true hypothesis!).
Given that he has found that these branch lengths are zero
or nearly zero, surely he should be testing whether you
would expect to get such small branch lengths if they
were actually non-zero. The obvious null hypothesis is
that the branch lengths are proportional to time (i.e.