st: stata 11 MI

[Stavrakakis Nikolaos <nikolaos.stavrakakis@gmail.com>]

Dear Statilist,

I ve been trying to impute multivariate variables (combination of
continuous and ordinal variables) using mvn in stata 11. My research
design is longitudinal so I am using as independent (predictors)
variables that I want to impute.  Some variables (mainly from the last
measurement wave, T3) have a missing rate from 22 to 32%.  The problem
I get is this, for every variable that I want to impute if I use
variables that have a large proportion of missingness I get a small
percentage of variables that are finally imputed (eg. i have 43
missing values for x variable and manage to impute only 3 of them)
but if I take out  the variables (mainly T3) that have a large number
of missing as my predictors then I get for the same variable a larger
number imputed (eg for x variable with 43 missing stata imputes 23 of
them).  Can I trust these imputations? and if yes how can I impute a
larger number of missing values but stil being certain that these
imputations are reliable?

Nik

*
*   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/