"firstname.lastname@example.org" <suffyan.koroma@FAO.ORG> wrote:
> Does anyone have a SAS code to perform Markov Chain Monte Carlo
> Please respond. The experimental M1 procedure is not satisfactory.
I assume you mean the MI procedure. [The M1 is a rifle. :-]
Yes, it is 'experimental' in V8. Yes, it is not complete in
all senses of multiple imputation.
Version 9 PROC MI will have MCMC [if I remember correctly what
Maura Stokes said]. But do you have the underlying structure
that makes MCMC the right choice? Is your data MCAR? MAR?
None of the above? Is multiple imputation appropriate for your
situation? I only ask because I regularly work with samples for
which any imputation choices are generally in error - the missing
data are typically not definable as coming from the exact same
population as the non-missing data, so even assuming that the
same population of values can be used is problematic.
For now, if MCMC is the right choice for your imputation problem,
you will have to code it up yourself. Or talk to SAS and try to get
an Early Adopter copy so that you can use version 9 ASAP.
David Cassell, CSC
Senior computing specialist