RE: Subject: RE: st: Situation where multiple imputation may be of no use?

[Cameron McIntosh <cnm100@hotmail.com>]

Hi Clyde,

Thanks for your reply. So, FIML (MAR) and listwise deletion (MCAR) performed identically in simulations (and with your real data)? However, I'm not sure if that proves conclusively that the missing data did not bias the intervention effect in the actual study (and I don't know how you specified your simulations).  You may want to look at the following before proceeding:

Jo, B., Ginexi, E.M., & Ialongo, N.S. (2010). Handling missing data in randomized experiments with noncompliance. Prevention Science, 11(4), 384-396.

Puma, M.J., Olsen, R.B., Bell, S.H., & Price, C. (October 2009). What to Do When Data Are Missing in Group Randomized Controlled Trials (NCEE 2009-0049). Washington, DC: National Center for Education Evaluation and Regional Assistance, Institute of Education Sciences.http://ies.ed.gov/ncee/pdf/20090049.pdf

Mealli, F., & Rubin, D.B. (2002). Assumptions when analyzing randomized experiments with noncompliance and missing outcomes. Health Services Outcomes Research Methodology, 3, 225-232.

Horowitz, J.L., & Manski, C.F. (2000). Nonparametric Analysis of Randomized Experiments with Missing Covariate and Outcome Data. Journal of the American Statistical Association, 95(449), 77-84. 

Hope this helps,

Cam
----------------------------------------
> From: clyde.schechter@einstein.yu.edu
> To: statalist@hsphsun2.harvard.edu
> Subject: Subject: RE: st: Situation where multiple imputation may be of no use?
> Date: Sat, 11 Feb 2012 20:25:33 +0000
>
> In response to my original query about whether MI is of any use in a situation where only the dependent variable will have missing values, Cameron McIntosh writes:
>
> "So why not try FIML? What analytical technique are you using?
> Cam"
>
> The simplified situation I described involves a single continuous outcome variable measured in subjects randomly assigned to two groups. So it's a regression of the outcome against an indicator for treatment group. FIML can be applied to this, but I ran some simulations and in this situation it doesn't perform any differently from complete case analysis. And, from a theoretical perspective, I don't think it should. I don't see how in this situation there is any information in the data set that is not already found in the complete cases. But I'd be delighted to learn that I'm wrong about that.
>
> Clyde Schechter
> Department of Family & Social Medicine
> Albert Einstein College of Medicine
> Bronx, NY, USA
>
>
> *
> * For searches and help try:
> * http://www.stata.com/help.cgi?search
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/

 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/