***********************************************************************
                               PAN
    Multiple imputation for multivariate panel or clustered data
                  Version (6/97) for Splus/Unix
***********************************************************************
Description:

This package contains a total of eight files, three of which are
contained in a subdirectory called "help". The files "pan.s" and
"pan.f" are an Splus and Fortran implementation of two algorithms
described by Schafer (1997):

pan() - an Splus function for multiple imputation of missing
   covariates under a multivariate linear mixed model; and

pan.bd() - an alternative version of pan() that restricts the
   covariance matrix for the random effects (psi) to have
   a block-diagonal form; and

ecme() - an Splus function for maximum-likelihood estimation
   under a univariate linear mixed model.

The files "panex.s" and "ecmeex.s" contain Splus code and extensive
comments showing how these functions can be used.  The three files
in "help" are help files for the Splus functions.
***********************************************************************
Installation:

(1) Before you can use these functions, you will need to compile
the Fortran source code. At the Unix prompt, type:

   f77 -c -O pan.f

This will create object code "pan.o".

(2) Copy the two help files, "pan" and "ecme", from the "help"
directory into a subdirectory named ".Help" of the ".Data" directory
where the Splus functions will reside. For example, at the Unix
prompt, you could type

   cp help/* .Data/.Help

(if .Data/.Help does not exist, you will need to create it).

(3) Enter Splus and type

   source("pan.s")

to load the functions pan() and ecme(). Then load the Fortran object
code by typing:

   dyn.load("pan.o")

This last step, the dynamic loading of the Fortran code, needs to be
carried out each time you enter Splus and want to use pan(), pan.bd()
or ecme().  You may want to store this last command in the ".First"
function, so that the Fortran is automatically loaded and ready for
you each time you enter Splus. Consult the Splus documentation about
this.
***********************************************************************
Using the program:

Assuming you have correctly installed the program, you now have at
your disposal the Splus functions pan(), pan.bd() and
ecme(). Descriptions of these functions are found in the help
files. For example, type "help(pan)" from within Splus to see a full
description of pan().  Detailed examples are provided in the files
"panex.s" and "ecmeex.s".
***********************************************************************
Author:

This code was written by Joe Schafer, Dept. of Statistics, Penn State
University, and may be used by anyone if credit is given. The author
assumes no liabilities for its use or misuse.
Please make suggestions and report bugs to jls@stat.psu.edu.
***********************************************************************
References:

Schafer, J.L. (1997), Multiple imputation of missing covariates under 
a multivariate linear mixed model. Technical report, Department of
Statistics, The Pensylvania State University.
***********************************************************************
