SAS/STAT Software

LOGISTIC Procedure

You can now fit a generalized logit model (Agresti, 1990) when you have nominal response data. In this multiple-group logit model, each nonreference response category is contrasted against the reference category. This model reduces to the binary logit model when there are only two response categories. Unlike the cumulative logit model (for ordinal response data) that contains one set of slope parameters for the entire model, the generalized logit model contains a set of slope parameters for each response category other than the prespecified reference category.
The generalized logit model is specified by the MODEL statement option LINK= GLOGIT. You can choose a response category as the reference in forming the various logit functions. The default is to use the last response category as the reference.
Options specific to the response variable can be specified immediately after the response variable in the MODEL statement by enclosing them in a pair of parentheses. These response variable options include the EVENT= option for specifying the event category for the binary response model, the REFERENCE= option for specifying the reference category, the ORDER= option for specifying the order of the response categories, and the DESCENDING option for reversing the order of the response categories.
For an exact conditional analysis, the new EXACTOPTIONS option NOSORT processes your data without resorting. This enables you to input your data in an order that you think may speed up computations.

Chapter Contents
Previous
Next
Top of Page