Re: st: mi and ice commands

[ymarchenko@stata.com (Yulia Marchenko, StataCorp)]

Aggie Chidlow <mojamalarybka@googlemail.com> asks about the difference between
Stata's official command -mi- and the user-written command -ice-:

> As I am new to "mi" and "ice" commands I have been reading about them quite
> extensively. However I am still a bit confused,and as a result I would
> really appreciate if somebody clarified the following to me:
>
> a) what is the difference between  "mi" and "ice" ?

The -mi- command performs all steps of multiple-imputation
analysis: imputation, data manipulation, and final primary-data
analysis.

For the imputation step, -mi impute- supports several univariate
imputation methods and multivariate imputation based on
the multivariate normal distribution (Schafer 1997).  See

   http://www.stata.com/capabilities/mi.html

for the full list of capabilities of -mi-.

The -ice- command performs the first step of multiple-imputation
analysis: imputation.  -ice- performs multivariate imputation
using chained equations (van Buuren et al. 1999).  This method is
currently not available in -mi impute-.

After imputation using -ice-, the other steps of multiple-imputation
analysis can be performed by the user-written command -mim-, or
the imputed data from -ice- can be imported into the -mi- framework
using -mi import ice-.

The following FAQ provides full details:

    http://www.stata.com/support/faqs/stat/mi_ice.html


> b) what is the preffered "mi style" for a panel data?

There is no preferred mi style specific for panel data.  In general, I like
to work in -wide- because I often compare my multiple-imputation results to
the complete-cases analysis and this is the most convenient style for that.
Depending on how you impute your panel data (see, for example,
http://www.stata.com/support/faqs/stat/impute_cluster.html for some
strategies), style -mlong- may be more convenient.


References:

Schafer, J. L. 1997. Analysis of Incomplete Multivariate Data. Boca Raton, FL:
  Chapman & Hall/CRC.

van Buuren, S., H. C. Boshuizen, and D. L. Knook. 1999. Multiple imputation of
  missing blood pressure covariates in survival analysis. Statistics in
  Medicine 18: 681â??694.


-- Yulia
ymarchenko@stata.com
*
*   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/