
 0.1   First "release" -- appears to work and do bernoulli regression with
           normal random effects.  Big W is really slow.

 0.2   Redesign access to weight vector, don't always return to R.
       Speed up big W using method of batch means.
       Big V now as given by (6a) in mcla.tex (was off by factor of nobs)
       mcla.tex in inst/doc

 0.3   remove O(nmiss) memory requirement for bnmarg when weights not wanted
       now all routines use O(dim(x) + dim(y) + nparm^2)
       allows unlimited sample sizes (if you have the patience to wait)

 0.3.1 added "booth" toy data set
       fixed tests/*.R so they give o. k. results with R-1.9.1

 0.3.2 changed "salam" toy data set back to parameterization used in Booth
       and Hobert (fixed effects R/R, R/W, etc.)

 0.3.3 changed "toy" toy data set to have evidence of creation

       added directory inst/makedata that shows where the data sets
       came from

 0.3-4 removed restriction that components of sigma be nonnegative

 0.3-5 added weigh argument to bnlogl and flu data set and flu.R in tests
       note: this is rush job.  bnbigw doesn't yet handle weights, so can't
       do a complete analysis.

 0.3-6 bnbigw now also handles weights.

 0.3-7 added bigw.hat.exact and info.hat.exact to booth dataset

 0.3-8 (Ben Bolker) adaptations for R >=3.0.0; NAMESPACE, useDynLib, etc.
