Designating a reference class in logistic regression

[Java Joe]

Hi,

 

I'm using SPSS's regression module to run a binomial logistic regression.  It seems that instead of dummy coding, the program allows you to designate the reference class (e.g., last, first, custom) for any categorical variable used in the model.  I'm unclear , though, as to the advantages/disadvantages of these options.  Is it "better" to designate the largest category as the last reference category (i.e., doing a "custom" designation) or should I use the last (or first) category regardless of the size?

Thanks very much for advice.

 

Joesy

[Bruce Weaver]

Java Joe wrote

Hi,

I'm using SPSS's regression module to run a binomial logistic regression.
It seems that instead of dummy coding, the program allows you to designate
the reference class (e.g., last, first, custom) for any categorical variable
used in the model.  I'm unclear , though, as to the advantages/disadvantages
of these options.  Is it "better" to designate the largest category as the
last reference category (i.e., doing a "custom" designation) or should I use
the last (or first) category regardless of the size?

Thanks very much for advice.

Joesy

"Better" in what sense?  If you take the simple case of a binary variable, choosing one category rather than the other as referent will just change the sign of the coefficient (on the log scale).  I'm not sure what you're getting at, I guess.