Logistic Regression Polynomial Contrast Variable Selection

[martus]

I note there is the "polynomial" contrast option in logistic regression analysis
for categorical variables, the coefficients to create orthogonal polynomials
are given in the output.
However, til now I only can include the whole set of polynomial contrasts in
a model even if using variable selection methods. Thus, if I want e.g. only
to include the linear and quadratic contrast (if significant), do I have to code "manually"
or is there any elegant way to do it?
Thanks a lot for any help.
Peter Martus

[Bruce Weaver]

AFAIK, you'll have to include two manually coded variables.  

If I was doing it, I'd first run a model with the "manually" coded variables for all k-1 polynomial contrasts to verify that they give the same results as the built in polynomial contrast.  (But you've probably already thought of that.)

martus wrote

I note there is the "polynomial" contrast option in logistic regression analysis
for categorical variables, the coefficients to create orthogonal polynomials
are given in the output.
However, til now I only can include the whole set of polynomial contrasts in
a model even if using variable selection methods. Thus, if I want e.g. only
to include the linear and quadratic contrast (if significant), do I have to code "manually"
or is there any elegant way to do it?
Thanks a lot for any help.
Peter Martus