Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: multiple imputation and propensity score


From   Stas Kolenikov <skolenik@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: multiple imputation and propensity score
Date   Wed, 24 Aug 2011 09:13:34 -0500

Can you follow through your analysis in the multiple imputation
framework? The propensity score will probably go into some regression
or matching exercise; can you perform these with -mim-? That would be
the approach most closely consistent with MI framework.

On Wed, Aug 24, 2011 at 5:28 AM, Stefano Di Bartolomeo
<stefano.dibartolomeo@uniud.it> wrote:
> Dear Statalist members
>
> I am doing a study that compares survival after 2 types  of cardiological treatments angioplasty or by-pass. To limit confounding I use a propensity score, calculated as usual by a logistic model. A few covariates had missing values, which I imputed with ICE (5 imputations), separately for each group of treatment. Then I joined again the records in 1 file with 'append' and so have a file with N*6 observations. Then I calculate the propensity score :
> mim: logistic angioplasty_vs._bypass  + other_covariates
> Finally I obtain the propensity score with
> mim: predict pscore
> As expected, I have 6 sets of propensity scores, one for each set of imputed data (_mj = 1-5) plus the one (_mj = 0)  resulting from the combination of imputed estimators  according to Rubin's rules. Unfortunately, the propensity score of the set _mj = 0 (which is the one I would think correct to use for further analyses) makes no sense, being virtually the same in patients treated with angioplasty or by-pass. The propensity scores of the imputed sets _mj 1-5 instead are ok and distributed as expected in the two treatment groups. I could easily pick up one of this well-working propensity scores for further use, but I know it is not correct. Has anybody ever encountered such a problem? Is it normal that the application of Rubin's rules results in a virtually useless propensity score? If so, how can one properly calculate propensity scores with multiply imputed data-sets?

-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index