Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

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

Re: st: MI: Multiple Imputation

From (Wesley D. Eddings, StataCorp)
Subject   Re: st: MI: Multiple Imputation
Date   Mon, 20 Dec 2010 15:17:27 -0600

Shikha Sinha <> asked about using -mi impute mvn- to
impute a categorical variable:

> Hi,
> I am using -mi for imputing the missing values. The variable is "edu" has
> five categories (1=primary, 2=secondary, and so on...). I used the following
> command:
> ...
> mi impute mvn edu = i.gender age country, add (5)
> mi estimate, or: logistic Y X1 X2 edu
> ...
> (a) Why some observations have negative values for  _1_edu, _2_edu, _3_edu,

The -mi impute mvn- command uses a multivariate normal model and treats "edu"
as continuous, so the imputed values are not guaranteed to be positive.
Shikha may instead use -mi impute ologit- or -mi impute mlogit-, depending on
whether or not the categories of "edu" should be considered ordered.

Shikha also asked,

> (b) How do we use the five new generated variable? Can I just run
> logistic Y X1 X2 _1_edu
> or
> logistic Y X1 X2 _2_edu
> (c) Can I create a dummy for primary or secondary schooling from the imputed
> values _1_edu?

The five new variables _1_edu, _2_edu, ... , _5_edu are the imputed values of
"edu" in the -wide- style for each of the five imputations.  It is not
appropriate to include the new variables directly as predictors; the
appropriate command for a multiple-imputation analysis is -mi estimate-, as
illustrated in Shikha's code.  -mi estimate- will analyze the five imputations
and combine them into a single multiple-imputation result.

Shikha may add the i. factor-variable prefix to "edu" to include dummies for
the imputed "edu" variable.  It is not necessary to create dummies manually
from a -mi- variable like _1_edu.

We also recommend that Shikha include the variables Y, X1, and X2 in the
imputation model.  More model-building guidelines appear in the "Model
building" section of the documentation entry -[MI] mi impute-.



*   For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index