I can't think of one.  The sorts of estimates you get with a declared 
subpopulation are the same that you would get if you made your subpopulation 
variable a dimension in a table and then dropped the table elements outside of 
the subpopulation of interest.  So it is like having imputed both a row 
variable and a response variable.  I think WesVar will only allow multiply 
imputed response variables but if SUDAAN and Stata also allow multiply table 
variables, I am not sure why there would be a problem.  Of course, if the two 
variables were not jointly imputed, then you may get bad point estimates, but 
that is a different issue.

--Dave

From: Impute -- Imputations in Data Analysis 
[mailto:[email protected]] On Behalf Of Lucy Bilaver
Sent: Wednesday, July 10, 2013 2:16 PM
To: [email protected]
Subject: Re: MI analysis with complex survey data

Thanks. SUDAAN does allow multiply imputed variables in the subgroup statement. 
 The regression will proceed without error or warning.  My concern is that 
there is a statistical issue that I am overlooking.  Stata allows users to 
specify a multiply imputed variable in a subpop statement also but requires an 
option  (esampvaryok) after a big warning.
-L

On Wed, Jul 10, 2013 at 12:37 PM, David Judkins 
<[email protected]<mailto:[email protected]>> wrote:
Users of SUDAAN need to be careful to always give it the full dataset and use 
the subgroup statement to get subpopulation estimates.  Prior subsetting will 
not give correct results.  I do not know though if SUDAAN allows you to have 
multiply imputed values for variables listed on the subgroup statement.  I 
suspect not.  This might be a case, where you need to write your own software.  
Easiest if you use replicated weights for the within-imputation variance.

Here is one version of the SUDAAN warning:
http://www.ats.ucla.edu/stat/sudaan/faq/subpopn.htm

--Dave Judkins

From: Impute -- Imputations in Data Analysis 
[mailto:[email protected]<mailto:[email protected]>]
 On Behalf Of Lucy Bilaver
Sent: Wednesday, July 10, 2013 12:33 PM
To: 
[email protected]<mailto:[email protected]>
Subject: MI analysis with complex survey data

Hi everyone,
I have fit an imputation model for complex survey data that I am working with.  
I ran the imputation in Stata and have been doing the analysis in Stata and 
Sudaan.

I am trying to understand what the issues are when analyzing a subsample of 
multiply imputed data that varies in size across imputations.  The subsample is 
based on an imputed variable thus the reason for the slight variation across 
imputations is clear.  Stata warns that this situation could "bias results" 
while Sudaan does not.  Does anyone have thoughts on the implications of having 
estimation samples that vary across imputations when using a complex survey 
design?  In both Stata and Sudaan, I am asking for SE to be computed using a 
Taylor linearization method.
Thanks!
-L

--
Assistant Professor
Northern Illinois University
School of Nursing and Health Studies
[email protected]<mailto:[email protected]>

Affiliated Scholar
Chapin Hall at the University of Chicago
[email protected]<mailto:[email protected]>
773-256-5237<tel:773-256-5237>

________________________________
This message may contain privileged and confidential information intended 
solely for the addressee. Please do not read, disseminate or copy it unless you 
are the intended recipient. If this message has been received in error, we 
kindly ask that you notify the sender immediately by return email and delete 
all copies of the message from your system.



--
Assistant Professor
Northern Illinois University
School of Nursing and Health Studies
[email protected]<mailto:[email protected]>

Affiliated Scholar
Chapin Hall at the University of Chicago
[email protected]<mailto:[email protected]>
773-256-5237


________________________________
This message may contain privileged and confidential information intended 
solely for the addressee. Please do not read, disseminate or copy it unless you 
are the intended recipient. If this message has been received in error, we 
kindly ask that you notify the sender immediately by return email and delete 
all copies of the message from your system.

Reply via email to