Re: st: multiple imputation using a log-linear model

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Tue, 1/6/10, Sheena Sullivan wrote:
> I am trying to use missing value analysis for bias
> analysis, as done in Greenland 2009 IJE 38(6):1662-1673. 
> The technique requires imputing the missing (true,
> unmisclassified) data using log-linear/Poisson regression.
>   However, in looking through the documentation for ICE
> there doesn’t appear to be an option capable of doing this. 
> Has anyone been able to trick the software or find some
> other solution that permits imputation using a log-linear
> model?

Often you can represent simple loglinear models as -logits-,
in that case you can use -ice-, bit more complex models can 
often be represented as count models, which can be approximated
with the -cmd(nbreg)- option. This will be a bit more flexible
than the -poisson-, but extra flexibility is not a bad thing
in an imputation model. Some log-linear model (e.g. RCII) are
not so easily implemented. If you want to estimate such a model
you'll just have to relax the proportionality assumption they
often impose, and use the -cmd(nbreg) option using this more 
flexible model.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   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/