missing

["Frank Lawrence" <Cougar@psu.edu>]

In response to my question [at end of this message] on multiple
imputation, I received two replies.  I thank Jim Schimert and Tim
Hesterberg for their very helpful comments.  For those interested, I
copied Tim Hesterberg's reply below.


#-------- Tim Hesterberg wrote -------------------------

#Generate list of starting values using a bootstrap
start <- list()
for(i in 1:5)
  start[[i]] <- paramIter(emGauss(cholesterol,
                          subset=sample(1:28,14,T), prior="ml")) # draw
5 imputations from 5 parallel chains, each run for 50 iterations
cholesterol.imp <- impGauss(cholesterol, prior="non",
                            start=start, control=list(niter=50))


Second, help(impGauss.methods) gives specifications for the `start'
argument, including this text:
          For several chains, start may be a list of such lists, a class
          "Gauss" object, or a list of "Gauss" objects.
          For a list of lists, each interior list must contain the two
          components mu and sigma. The number of imputations equals the
          length of the outermost list.
          A class "Gauss" object is the paramIter component of a class
          "missmodel" object, produced by routines such as mdGauss,
          daGauss, and emGauss. This is a matrix with as many rows as
          there are imputations.

Third, this code:
for(i in 1:5)
        start[i,] <- paramIter(emGauss(twin.d0, subset = sample(x =
                1:nrow(twin.d0), size = 0.5*nrow(twin.d0), replace = T),
        prior = 'ml'))[1,]
tries to shoehorn paramIter objects into columns of an ordinary matrix;
some information is lost.

Tim Hesterberg

#--------my question----------------------

I am trying to run multiple imputations through DA using the missing
        data library [Splus 6R2, Win2K].  I was trying to write a start
function
        but I am having trouble understanding how many columns need to
be in the
        start matrix [code below].  In other words, what is the relation
between
        columns in the start matrix and the number of variables
subsequently
        used in the impGauss command?

Respectfully, 
Frank R. Lawrence