Hello Everyone,
I have a question about what kinds of covariates to include in an MI analysis. Specifically, is it possible to include covariates that ultimately wind up introducing biases into your results? For example, I'm trying to impute start and stop dates for HIV antiretroviral drugs. I'm thinking that the timing of increases in viral load to a clinical cutoff and the timing of decreases in CD4 count to a clinical cutoff will be good predictors of the start date for a new drug regimen (especially the first one). But suppose I then also want to include subsequent measures of viral load and CD4 in my imputation dataset to be used as outcome variables in an analysis? Will including the earlier covariates likely bias the associations between my drug data and the subsequent measures of viral load and CD4? Or might including the earlier covariates actually help to more accurately preserve these associations? I'm thinking it might be possible to do some simulations but I'd like to know that the theory says about this beforehand. Thanks, Paul Paul J. Miller, Ph.D. Research Scientist and Statistician Ontario HIV Treatment Network 1300 Yonge St., Suite 308 Toronto, Ontario M4T 1X3 Phone: (416) 642-6486 ext 232 Fax: (416) 640-4245 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://lists.utsouthwestern.edu/pipermail/impute/attachments/20060926/8a782d9d/attachment.htm From von-hippel.1 <@t> osu.edu Tue Sep 26 13:29:24 2006 From: von-hippel.1 <@t> osu.edu (Paul von Hippel) Date: Tue Sep 26 13:29:27 2006 Subject: [Impute] Re: Covariates in MI and the introduction of biases? In-Reply-To: <[email protected]> References: <[email protected]> Message-ID: <[email protected]> My understanding is that you can get bias from having too few variables but not from having too many.
