Date: Wed, 12 Nov 2003 08:29:06 -0500
Reply-To: Mark Davenport <madavenp@OFFICE.UNCG.EDU>
Sender: "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From: Mark Davenport <madavenp@OFFICE.UNCG.EDU>
Subject: Re: Missing value analysis
Content-Type: text/plain; charset=US-ASCII
The estimates that MVA produces are (I believe) based on an multiiterative EM procedure. Every time you run it you will get different estimates for your missing values. My guess is that several things should factor into how close your estimates will be among test sets (original sample size, randomness of the 'missingness' of the data, etc.). Also, keep in mind that the MVA procedure available in SPSS won't recreate the distribution of error expected in an analysis with complete data. This requires an extra step. See some of Joe Schafer's and John Graham's work on data augmentation.
Mark A. Davenport Ph.D.
Asst to the Vice Chancellor for Student Affairs/Research and Evaluation
The University of North Carolina at Greensboro
149 Mossman Bldg.
Greensboro, NC 27402-6170
'An approximate answer to the right problem is worth a good deal more than an exact
answer to an approximate problem' -- J. W. Tukey
>>> Matthew Neale <firstname.lastname@example.org> 11/12/2003 1:44:05 AM >>>
I have been using the MVA procedure in SPSS 11.5 and have been conferring
with a colleague who is using the same procedure on the same data in SPSS
10.0.7. We have been obtaining different estimated means, variances and
covariances when we apply EM estimation, and the results for Little's MCAR
test have also been substantially different across the two versions.
Can anyone advise me on which set of results are likely to be correct or
most accurate? Or whether we both need to upgrade to later versions? I
tried searching the SPSS website but they seem to be in the middle of an
update, and are returning nothing but 404s :)