st: multivariate imputation using chained equations in Stata 12?

[Masahiko Aida <maida@gqrr.com>]

Hello all

I am wondering the implementation of ICE in Stata 12.  Can it run imputation sequentially?

Back in when I used to work as RA in U of Michigan, I used multiple imputation software IVEware.   (I am happy that I do not have to deal with SAS anymore btw).  One nice feature of IVEware was it implemented Sequential Regression Imputation Method.  Currently Stata 11 does not have this capacity.    When Stata 11's MI IMPUTE command create multivariate predictive model like below --- predicting y using x1 and x2, it use complete dataset ( of pre-imputed data, or in other words running regress command when missing data indicator _mi_m ==0),  to estimate model  y = b1 x1 + b2x2 + e  (assuming there is missing values in x1 or x2) and while I do not think that is a huge issue, ideally Stata should not be using complete data of pre-imputed case size, but it should include cases whose values in independent variables are imputed.

#  mi impute reg y x1 x2 , replace

Ideally, Stata's new imputation command can run following process iteratively

#1 mi impute reg x1 z1 z2 z3  x2 y
#2 mi impute reg x2 z1 z2 z3  x1 y
#3 mi impute reg y x1 x2 z1 z2 z3

Repeat step #1 to #3 for set number of iterations - or until changes between y across different iteration becomes sufficiently small.


I would like to see if Stata 12's imputation command can handle sequential imputation - or if someone knows how to run it with Stata 11 and I am not aware - please let me know.

Best
Masahiko

*
*   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/