Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: mi impute chained and 'imputing everything'

From (Yulia Marchenko, StataCorp LP)
Subject   Re: st: mi impute chained and 'imputing everything'
Date   Tue, 03 Jan 2012 12:31:38 -0600

Jonathan Bartlett <> asks if there is a
particular reason why the -mi impute chained- command fills in missing values
for observations containing missing values in all variables specified within
the command:

> When performing multiple imputation with the user-written ice command, if an
> observation had missing values for all of the variables passed to ice, ice's
> default behaviour was to ignore the observation. That is, the observation's
> missing values were not imputed. 
> ...
> Stata 12 includes the command mi impute chained, which does virtually
> everything ice does. However, mi impute chained differs in that it seems to
> 'impute everything' for observations who are missing for all variables
> passed to the command. Furthermore, there doesn't appear to be an option
> that requests that such observations not be imputed, although presumably
> this could be achieved through use of 'if'.
> I was wondering if there is a particular reason why Stata have chosen to
> make mi impute chained's default behaviour different to ice's in this
> respect?

The -mi impute chained- command, or more generally the -mi impute- command,
follows a general rule for imputing missing values.  All soft missing values
(system missing values, or .) in specified imputation variables are imputed,
and all hard missing values (extended missing values, or .a, .b, .c, etc.) are
not imputed and observations containing hard missings are not used during

As Jonathan pointed out, you can use an 'if' condition with -mi impute- to
restrict the imputation sample to the observations containing at least one
nonmissing variable among the variables specified within -mi impute-.
Alternatively, you can replace soft missing values nonapplicable for
imputation with hard missing values (.a, .b, etc.) and -mi impute- will
automatically omit the corresponding observations from imputation.

-- Yulia
*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index