Here's a follow up response to my earlier question.
----- Forwarded message from Michelle Cheuk <[email protected]> -----
Date: Fri, 01 Dec 2006 00:07:18 -0500
From: Michelle Cheuk <[email protected]>
Reply-To: Michelle Cheuk <[email protected]>
Subject: ice and constructed variables
To: [email protected]
Hi,
I saw your post on the Stata list on ice and constructed variables.
I
am new to reading the list, so I don't know how to post. I had a
similar question, and I had e-mailed the author of the ice program.
Here is his response. My question to him is below that.
Could you post this to the Stata list so that others can read about
it
too? Thanks!
Michelle
----- Forwarded message from [email protected] -----
Date: Wed, 29 Nov 2006 10:27:59 -0000
From: Patrick Royston <[email protected]>
Reply-To: Patrick Royston <[email protected]>
Subject: RE: creating variables after imputation
To: Michelle Cheuk <[email protected]>
Dear Michelle,
In principle, if you anticipate needing an interaction in the
substantive model, you ought to include it in the imputation model
fit
by ice. Otherwise the strength of the interaction in the substantive
model will be attenuated (i.e. biased towards 0).
You need to check out the passive() option of ice to see how to
recalculate interaction terms within the imputation process.
There should be no problems creating an ordinal categorical variable
from a continuous one post-imputation.
Best wishes
Patrick Royston
-----Original Message-----
From: Michelle Cheuk [mailto:[email protected]]
Sent: Tue 28/11/2006 19:17
To: Patrick Royston
Cc: [email protected]
Subject: creating variables after imputation
Dear Patrick,
I am a graduate student who is new to multiple imputation. I am so
excited to use your ICE Stata program! Thank you so much for writing
it.
I have a question for you. Is there any problem with doing the
following:
- Use ice to create a multiply imputed dataset (imputed.dta)
- Then opening this dataset (imputed.dta) to create some interaction
variables (var1*var2) from imputed variables (var1 and var2) and to
create an ordinal categorical variable from an imputed continuous
variable
- Then running micombine
?
Thank you very much! I'm looking forward to hearing from you.
Michelle
*****************************************
Michelle Cheuk
Department of Sociology
Carolina Population Center
University of North Carolina, Chapel Hill
[email protected]
*****************************************
----- End forwarded message -----
----- End forwarded message -----
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
