Hi, Constantine.
Thanks for your note and example. My post was too hung-up on the
idea of modeling the missingness mechanism.
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
----------------------------------------------------------------------------
------------
-----Original Message-----
From: Constantine Daskalakis [mailto:[email protected]]
Sent: Thursday, June 10, 2004 1:12 PM
To: Gregorich, Steve
Subject: RE: [Impute] Re: Interactions
At 03:20 PM 6/10/2004, Gregorich, Steve wrote:
>Hi, Everyone.
>
>I'd like to comment on the inclusion of X1Y and X2Y interactions in the
>imputation model. For now, I'll assume that such interaction effects will
>not be tested in any substantive model, and so the efficiency of such
>tests within the multiple imputation framework are not a concern. Given
>that, one might include X1Y and X2Y in the imputation model because those
>effects are believed to represent mechanisms of missingness--and thus help
>to reduce bias. However, given a strict causal interpretation of the
>substantive regression model, can this ever be true?
>That is, if X1, X2, and X1X2 all cause Y, then is it possible for, say,
>X1Y to inform missingness on X2? Strictly speaking, no.
Steve:
I am not sure what you mean by this.
Example -- case-control study of asbestos, smoking and lung cancer.
Asbestos (X1) and smoking (X2) and their interaction (X1X2) are all causal
for cancer (Y).
Smoking status is missing (randomly, for say 20%) of subjects, if either
unexposed to asbestos (X1=0) or control (non-cancer, Y=0).
However, smoking status is missing for no exposed cases (X1Y=1), perhaps
because records are more complete.
X1Y does inform missingness on X2.
> From a practical standpoint, however, Y may be a proxy for some variable
> that is contemporaneous with X2 and therefore inclusion of X1Y in the
> imputation model may be beneficial.
I do not understand this point. What does timing matter?
>I trust someone will gently stomp on me if I'm missing an important point
>here.
>
>Steve
The documents accompanying this transmission may contain confidential
health or business information. This information is intended for the use of
the individual or entity named above. If you have received this information
in error, please notify the sender immediately and arrange for the return
or destruction of these documents.
________________________________________________________________
Constantine Daskalakis, ScD
Assistant Professor,
Biostatistics Section, Thomas Jefferson University,
211 S. 9th St. #602, Philadelphia, PA 19107
Tel: 215-955-5695
Fax: 215-503-3804
Email: [email protected]
Webpage: http://www.jefferson.edu/medicine/pharmacology/bio/
