I'm running R version 3.1.1 but loaded a version of bayesm built under
version 3.1.3.  I couldn't figure out why the results for Arm A vs Arm
B looked to be back to front and why they didn't seem to change much
when the conjectured difference between the arms was increased
dramatically.  The problem seems to be in the post.update function.
The rdirichlet function returns an rx1 matrix while the rgamma
function returns a 1xr vector, so, when ma and mb are formed using the
%*% operator, the result is picked from the element in the top left
hand corner of the rxr matrix and not the scalar that should have
resulted from the multiplication of a 1xr with an rx1.  Easily fixed
by transposing the matrix returned by rdirichlet and a reminder to me
to always look carefully at warning messages.

Sincerely

John Reynolds