A sample MrBayes analysis
=========================

The data file is "data.nex", and the MrBayes commands are in the file
"commands.nex".  I have already chosen the model.  To start it at the
command line, I said

    mb commands.nex

This does 2 separate runs, the default.  I tell it to run for 100000
generations, and collect 2001 samples from each run.  I tell it to
discard the first 1001 samples as burn-in.  At the end, a consensus
tree is made from post burnin samples, showing only those splits with
more than 50% support.

The mbout.* files were all produced during the analysis.  We can look
at some diagnostics to see how well the analysis went.

Probably what we are interested in most is the topology, and we want
to know if the topology has converged.  Perhaps the best diagnostic
for that is the "Average standard deviation of split frequencies".  We
can see that it went to a low level (less than 0.01) during the burnin
period.  This gives us confidence that we have reached topological
convergence.  Other data that are less decisive might not show such
low values.

We can look at sump results in the mbout.log file.  We can see that
the acceptance probabilities are all within the suggested range of
10-70%.  A couple of the swap frequencies are higher than this; if we
were having trouble with convergence then this might be something to
adjust.  The two runs were very similar to each other in their
likelihood values.  In all, this looks like a well-behaved analysis.

The results show that we can have more or less confidence in different
parts of the consensus tree.  Little confidence should be given to
supports less than about 90%.

