Dear imputers, I originally started just to reply to Paul, but thinking about my reply raised a question that I thought it could be interesting to put round. VIZ: How big should an imputation model be? and what are the costs of including too many variables? ------------------------------------------------------------------------------------------------------------------------------------------------------- I am a bit puzzled by your query. Some questions A) Do you assume Z is completely observed? B) You say that you will assume (X,Y) are ignorable missing. What do you mean by this? Ignorable means that missingness does not depend on any unobserved values. Do you mean that missingness is ignorable given all the observations of X,Y and anything else that might have been observed including Z ( of course this is an assumption, since you can't test for it)? If B is assumed then there are two possibilities. 1) Missingness would have been ignorable even without information on Z. That is the distribution of Y|X would be the same for observed and missing Ys integrating over Z. 2) Missingness is only ignorable given Z I would guess that if 1) is true, then including Z in the imputation process will have no obvious benefit and I would have thought that it might hurt. But this is really an academic question since we can never know if MAR is valid and we always put Zs in to try get as close to MAR as possible. It would however have a bearing on how big an imputation model should be, since if it hurts to have fairly useless stuff in your imputation then it would be good to make it smaller. Even if the theory says it doesn't hurt there is a good practical case for making the model small since the programs will have problems and asymptotics for things like estimation of variance covariance matrices is likely to mess up. See the example on the web site that I circulated to the list yesterday http://www.napier.ac.uk/depts/fhls/peas
and the section on problems with imputation software in praticular http://www.napier.ac.uk/depts/fhls/peas/imputation.asp#features This message is intended for the addressee(s) only and should not be read, copied or disclosed to anyone else outwith the University without the permission of the sender. It is your responsibility to ensure that this message and any attachments are scanned for viruses or other defects. Napier University does not accept liability for any loss or damage which may result from this email or any attachment, or for errors or omissions arising after it was sent. Email is not a secure medium. Email entering the University's system is subject to routine monitoring and filtering by the University. -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20050818/e62c9784/attachment.htm From wilmar.igl <@t> mail.uni-wuerzburg.de Thu Aug 18 07:07:01 2005 From: wilmar.igl <@t> mail.uni-wuerzburg.de (Wilmar Igl) Date: Thu Aug 18 07:07:13 2005 Subject: [Impute] auxiliary variables In-Reply-To: <3ccae204656c9d4d91c118cd01ef63d001541...@evs2.napier-mail.napier.ac.uk> References: <3ccae204656c9d4d91c118cd01ef63d001541...@evs2.napier-mail.napier.ac.uk> Message-ID: <[email protected]> Dear Mrs. Raab, I'd like to point out the following publication to you, where you can find information about the consequences of including too many or too little variables in an imputation model: Collins, LM, Schafer, JL, Kam, C-M (2001). A Comparison of Inclusive and Restrictive Strategies in Modern Missing Data Procedures. Psychological Methods, 6(4), 330-351. Best wishes, Wilmar Igl Raab, Gillian wrote: > Dear imputers, > > I originally started just to reply to Paul, but thinking about my reply > raised a question that I thought it could be interesting to put round. VIZ: > How big should an imputation model be? and what are the costs of > including too many variables? > > ------------------------------------------------------------------------------------------------------------------------------------------------------- > > I am a bit puzzled by your query. Some questions > A) Do you assume Z is completely observed? > > B) You say that you will assume (X,Y) are ignorable missing. What do you > mean by this? > Ignorable means that missingness does not depend on any unobserved values. > Do you mean that missingness is ignorable given all the observations of > X,Y and anything else that might have been observed including Z ( of > course this is an assumption, since you can't test for it)? > > If B is assumed then there are two possibilities. > 1) Missingness would have been ignorable even without information on Z. > That is the distribution of Y|X would be > the same for observed and missing Ys integrating over Z. > 2) Missingness is only ignorable given Z > > I would guess that if 1) is true, then including Z in the imputation > process will have no obvious benefit and I would have thought that it > might hurt. > > But this is really an academic question since we can never know if MAR > is valid and we always put Zs in to try get as close to MAR as possible. > > It would however have a bearing on how big an imputation model should > be, since if it hurts to have fairly useless stuff in your imputation > then it would be good to make it smaller. Even if the theory says it > doesn't hurt there is a good practical case for making the model small > since the programs will have problems and asymptotics for things like > estimation of variance covariance matrices is likely to mess up. > > See the example on the web site that I circulated to the list yesterday > > http://www.napier.ac.uk/depts/fhls/peas > > > and the section on problems with imputation software in praticular > > http://www.napier.ac.uk/depts/fhls/peas/imputation.asp#features > This message is intended for the addressee(s) only and should not be > read, copied or disclosed to anyone else outwith the University without > the permission of the sender. It is your responsibility to ensure that > this message and any attachments are scanned for viruses or other > defects. Napier University does not accept liability for any loss or > damage which may result from this email or any attachment, or for errors > or omissions arising after it was sent. Email is not a secure medium. > Email entering the University's system is subject to routine monitoring > and filtering by the University. > > > ------------------------------------------------------------------------ > > _______________________________________________ > Impute mailing list > [email protected] > http://lists.utsouthwestern.edu/mailman/listinfo/impute -- Dipl.-Psych. Wilmar Igl c/o Institut f?r Psychotherapie u. Medizinische Psychologie Arbeitsbereich Rehabilitationswissenschaften Marcusstrasse 9-11 (R. 409), 97070 W?rzburg Telefon: 0931/31-2573, FAX: 0931/31-2078 E-mail: [email protected] URL: http://www.uni-wuerzburg.de/psychotherapie/mitarbeiter/igl.html
