Hi, Pat.
WRT to (b), I would also consult the efficiency index described
by Rubin (1987) and discussed by Schafer in the Norm help file
under "How many imputations do I need?". I believe the
efficiency is output by Norm (but strangely not output by SAS
PROC MIANALYZE, though it can easily be computed from the
information provided).
In many longitudinal data sets I have found that 10 imputed
data sets did not result in sufficient efficiency. I typically
go with 20 imputed data sets.
Steve
-------------------------------------------------------------------
Steve Gregorich
University of California, San Francisco
Department of Medicine
3333 California Street, Suite 335, Box 0856
San Francisco, CA 94143-0856
(FedEx and UPS use zip code 94118)
[email protected]
http://mywebpage.netscape.com/segregorich/index.html
-------------------------------------------------------------------
> Subject: IMPUTE: Satisfying reviewers
> From: "Patrick S. Malone" <[email protected]>
> Date: Mon, 03 Mar 2003 09:05:10 -0500
>
>
> Greetings.
>
> I'm hoping to get some pointers to useful citations to satisfy an editor's
>
> concerns on our use of multiple imputation (using NORM). Here are the two
>
> issues:
>
> "However, I could be persuaded by citations (to publications by imputation
>
> experts) or evidence (e.g., from Monte Carlo studies) showing that: (a) it
>
> is acceptable to impute missing data on the outcome variable; (b) 40%
> falls
> within the acceptable range for data imputation."
>
> I understand that (a) is not only acceptable, but obligatory in a
> covariance analysis, because a covariance matrix makes no distinction
> between outcomes and anything else. However, this is so fundamental, I'm
> not finding explicit statements of it in my sources. For (b), I realize
> that it's fraction of missing information that's the issue. We used 10
> imputations, so we should be in good shape for the missingnes we have, but
>
> are there any good simulation studies varying the missing information and
> showing satisfactory results? Schafer (97) talks about rates up to 90%
> just increasing the number of iterations needed, but there's not much
> detail on performance.
>
> In other words, has anyone written, "Multiple imputation for content
> journal editors" yet?
>
> Thanks,
> Pat Malone
>
> --
> Patrick S. Malone, Ph.D., Research Scholar
> Duke University Center for Child and Family Policy
> Durham, North Carolina, USA
> e-mail: [email protected]
> http://www.duke.edu/~malone/
>
>
