Dear all,
I've seen several posts in the archives about gross errors in
ancestral state reconstructions, and I believe I have located one of
the sources in MrBayes' output. Apparently if you specify multiple
nodes, the output for the ancestral states gets garbled. For
example, suppose you have constrained two nodes. In the .p file, the
state probabilities for the first character for node 1 should be
reported on lines starting with p(X){1@...}, the probs for the second
character of node one should be p(X){2@...}, the third character should
be p(X){3@...}, etc. But that is not what is happening -- the way
MrBayes currently handles it, the probs for the second character for
node 1 are found in p(X){1@...}, the third character for node 1 in p(X)
{2@...}, the fourth character for node 1 in p(X){2@...}, the fifth
character in p(X){3@...} etc.
The problem is evidently in the source of mcmc.c, where the character
and node labels printed in lines 30677-30699 are not synched properly
with the state probabilities printed in lines 30908-30937. The probs
appear to be correct internally, but just get printed out wrong.
AFAICT, it should however work fine if you only specify one node for
ancestral state reconstruction.
If anyone else can verify this I'll submit it as a bug report.
Cheers,
Douglas
|