Prev: sas procedure
Next: Test to see if a SAS dataset exists in a library
From: Forstero on 7 Feb 2007 12:42
I am working with with a dataset that has categorical variables and
some of these categorical variables has missing observations. Th
eproblem I am having is imputing for these missing observations. Can
anyone please help me with how to impute for categorical data?
I know about the proc MI data step but it only seems to work for
continous variables with two or more variables having missing
observations.
I will appreciate any help I can get on this subject.

Thanks,

Forster

From: Paige Miller on 7 Feb 2007 13:11
On Feb 7, 12:42 pm, "Forstero" <Forst...(a)gmail.com> wrote:
> I am working with with a dataset that has categorical variables and
> some of these categorical variables has missing observations. Th
> eproblem I am having is imputing for these missing observations. Can
> anyone please help me with how to impute for categorical data?
> I know about the proc MI data step but it only seems to work for
> continous variables with two or more variables having missing
> observations.
> I will appreciate any help I can get on this subject.
>
> Thanks,
>
> Forster

The problem with imputing values is that all imputation that I am
familiar with is set up to minimize (or optimize) some quantity. And
this minimization can only be done with continuous variables.

Now, I suppose you could define a "distance metric" among your
categorical levels and then write your own imputation program to
minimize the distance in this newly defined "distance metric", but I
have never seen such a thing done.

Good luck.

--
Paige Miller
paige.miller(a)kodak.com

From: tanwan on 7 Feb 2007 14:31
You might try IVEWARE software, available from the University of
Michigan at http://www.isr.umich.edu/src/smp/ive/.
It is free. It is callable in SAS (think of it as a big macro). You
will need to install a small file on your PC. You will need to devote
time to study the documentation. But after that, it becomes a cinch.
And you should NOT run it in the SAS's Enhanced Editor. It will not
work.

But it can impute categorical variables.

 | 
Pages: 1
Prev: sas procedure
Next: Test to see if a SAS dataset exists in a library