I have been reading the Stata 11 imputation manual. .
The manual states (page 107), "In practice, multiple variables usually
must be imputed simultaneously, and that requires using a multivariate
imputation method. The choice of an imputation method in this case
depends on the pattern of missing values." In my instance this means
using -mi impute mvn-
Using Royston's multivariate -ice-, it was possible to specify
mulivariate matching, oligit, mlogit, and logit. This is not possible
with -mi impute mvn -. From a users point of this means out of usual
(expected) range values (e.g., ages <0, non-integer categorical
values). The manual suggests (page 109), "For multiple categorical
variables with only two categories (binary or dummy variables), a
multivariate normal approach ([MI] mi impute mvn) can be used to
impute missing values and then round the imputed values to 0 if the
value is smaller than 0.5, or 1 otherwise. For categorical variables
with more than two categories, Allison (2001) describes how to use the
normal model to impute missing values."
I wonder if it might be possible in a revision of the manual to
actually describe how to impute categorical values without having to
purchase Allison's book (available on Amazon.com at a reasonable
cost). There are a lot of "simple" examples in the manual. but no
complex examples - somethings that would be helpful.
Would it be possible for StataCorp people to indicate on the list the
advantages of their multivariate method compared with Royston's.
--
Fred Wolfe
National Data Bank for Rheumatic Diseases
Wichita, Kansas
NDB Office +1 316 263 2125 Ext 0
Research Office +1 316 686 9195
[email protected]
*
* 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/
