Hello all,
I've recently been examining some data and, with the help of an expert in the field, have determined that some of our biological assays obtained invalid responses. It is appropriate to omit these data and impute them? If so, does this make the data MNAR? If it can't be done, are there alternatives to simply omitting the data from analyses (as these are single instances in a before and after repeated measures)? Many thanks, Jason ************************************************************** Jason C. Cole, PhD Statistician Department of Psychiatry and Biobehavioral Sciences Cousins Center for Psychoneuroimmunology 300 UCLA Medical Plaza, Room 3148 Los Angeles, CA 90095-7057 Tel: 310 267 4390 FAX: 310 794 9247 E-mail: <mailto:[email protected]> [email protected] <http://www.cousinspni.org> http://www.cousinspni.org ************************************************************** -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20030910/3515d7fa/attachment.htm From zaslavsk <@t> hcp.med.harvard.edu Mon Sep 15 16:06:23 2003 From: zaslavsk <@t> hcp.med.harvard.edu (Alan Zaslavsky) Date: Sun Jun 26 08:25:01 2005 Subject: IMPUTE: Re: treatment of invalid data In-Reply-To: <[email protected]> Message-ID: <pine.gso.4.05.10309151658400.28377-100...@hcp> > From: "Cole, Jason Ph.D." <[email protected]> > Subject: IMPUTE: Treatment of invalid data > Date: Wed, 10 Sep 2003 10:19:15 -0700 > > I've recently been examining some data and, with the help of an expert in > the field, have determined that some of our biological assays obtained > invalid responses. It is appropriate to omit these data and impute them? > If so, does this make the data MNAR? If it can't be done, are there > alternatives to simply omitting the data from analyses (as these are single > instances in a before and after repeated measures)? MAR vs MNAR is not something that can be determined within the data at hand. You have to bring in some assumptions about the reasons for missingness. For example if the reasons for the errors in the assays have nothing to do with the true values but only with unrelated and independent errors in the instruments, you might expect the data to be MCAR (an even stronger assumption) -- this would be a practical example of the illustrative hypothetical I often use in teaching of data that are missing because somebody spilled coffee on the datasheets. If the errors in the assays are related to the true values than the data might indeed not be even MAR. However it seems that you have not many alternatives to doing the imputation. Imputation under MAR is more likely to give you valid answers than is casewise exclusion (which requires assuming MCAR, in general), but you might want to do some MNAR sensitivity analyses to see if postulating some residual effects makes a difference.
