Date:         Tue, 14 May 1996 15:43:20 GMT
Reply-To:     David Nichols <nichols@SPSS.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@UGA.CC.UGA.EDU>
From:         David Nichols <nichols@SPSS.COM>
Organization: SPSS, Inc.
Subject:      Re: Stepwise regression with categorical variables

In article <4n8k5p$f66@seagoon.newcastle.edu.au>,
Dobbins T <sttd@alinga.newcastle.edu.au> wrote:
>I have a client who wants to mimic some analyses done in a paper for her
>thesis. In particular, the paper (*) has a stepwise regression using some
>variables that my client has defined as categorical (marital status,
>education level, family income). Now the two level variables are fine -
>just create a single dummy variable, but the multi-level ones are more
>problematic.
>
>For example, education level has 4 categories. Now I can create three
>dummy variables to replace the original variable, but my question is :
>       how do I ensure that all levels of the dummy variables enter the
>       stepwise regression.
>
>For example, I may find that one level of education is
>significant but the others are not - this will not help, we want overall
>significance of education.
>
>I can't use the middle value of 'years studied' as the levels are highest
>education level attained (eg high school, first degree, masters etc).
>
>Also, is this approach even valid? I personally have my doubts about
>stepwise regression. But as this is just an exploratory exercise, it may
>not matter. The next step (in the original paper) is to use PCA which may
>be more appropriate.
>
>Any comments are greatly appreciated.
>
>Tim
>
>(*) The original paper is 'Predictors of a General Attitude Toward Gifted
>Children' - Jean Begin and Francoys Gagne, Journal for the Education of
>the Gifted, Vol 18, No 1, 1994, pp 74-86
 
The REGRESSION procedure in SPSS does not explicitly handle categorical
variables, so it's not possible to just let fly with a stepwise with
categorical predictors with more than two categories. Newer nonlinear
regression procedures such as LOGISTIC REGRESSION and COX REGRESSION
have automatic handling of categorical variables, but REGRESSION does
not at the current time have this capability. You can, of course, do
what a stepwise analysis does "by hand," using multiple runs to get the
same results. It's tedious, but it's the only way.
 
--
-----------------------------------------------------------------------------
David Nichols             Senior Support Statistician              SPSS, Inc.
Phone: (312) 329-3684     Internet:  nichols@spss.com     Fax: (312) 329-3668
-----------------------------------------------------------------------------