Re: st: summary statistics with mi multiple imputation

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

--- On Tue, 20/7/10, David Bell wrote:
> One imagines that you do not have much missing data for
> your demographic variables.  I would in general be
> inclined to give descriptive statistics on non-missing data
> only.  This avoids any question from readers (and
> reviewers) about whether the imputation method introduced
> any biases.  The non-missing data are are, of course,
> the sample from which imputations are to be made.  If
> you include Ns, then readers can see how much data were
> imputed.

The purpose of showing describtive statistics is to give
the reader an idea of what the data is that is being used
in your analysis. Since what you use in your analysis is
the imputed dataset, it would make sense to describe that. 
On the other hand the empricial information comes from 
your raw/unimputed data. 

So I would present both, as it also gives a bit of insight 
how the imputation process changes your data. If there are 
some big differences then you will need to justify it, but 
that is in those extreme cases usually quite easy (e.g. 
the missing values are mostely young people who...). This 
way you can try to avoid that the reader regards the 
imputation as some form of white/black magic. 

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/