Re: st: ICE: Combining proportions using Rubin's rules via mim

[Stas Kolenikov <skolenik@gmail.com>]

On Sat, Jan 1, 2011 at 12:53 PM, Leslie Phillips
<leslie.phillips@gmail.com> wrote:
> I am using Patrick Royston's ICE to multiply impute data that will be
> used to calculate proportions.  I have calculated my proportions on
> the m imputed datasets which ICE appends into one dataset.  I now want
> to combine my proportions using Rubin's rule.  It is my first
> experience with MICE/ICE and I am a basic to less-than-basic Stata
> user so I am stumbling through this a bit but my read of it is that
> the overall estimate is the average of the individual estimates (in my
> case proportions).  It seems ICE will do this for me as follows:
> 1) I use ice to carry out m imputations
> 2) I calculate my proportion ("myprop") in the appended dataset which
> calculates m proportions for each _mi
> 3) Here's where it gets tricky for me, is it correct to use the mim
> combine command as below to get the appropriately combined proportions
> and a standard error that accounts for both the within and between
> imputation variance a la Rubin?  And if so, I should just repeat this
> for each _mi?
>
> mim, cat(combine)  est(_b[myprop]) se(_se[myprop]) : mean myprop if _mi==1

-mim- is aware of the structure created by -ice-, so the -if-
condition is redundant (and, in fact, inappropriate). Also, since
-mim- knows that -mean- is an estimation command, you don't need to
specify the -category()- option. So your very slim syntax will be

mim : mean myprop

and it will invoke Rubin's combination rules. (Of course you still
need to have a good imputation model and a reasonable number of
imputations to get result you can trust.)

-- 
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/