Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

By default, PROC LOGISTIC assumes the highest level of the outcome variable is the
reference group. If we wish to make SUBTYPE¼0 (i.e., Adenosquamous) the refer-
ence group we can use the DESCENDING option in a similar manner as we did when
we ran a standard logistic regression using PROC LOGISTIC. The code follows:

PROC LOGISTIC DATA¼REF.CANCER DESCENDING;

MODEL SUBTYPE¼AGE ESTROGEN SMOKING/LINK¼GLOGIT;

RUN;

The key difference in the syntax for specifying a polytomous rather than a standard
logistic regression using PROC LOGISTIC is the LINK¼GLOGIT option in the
MODEL statement. LINK¼GLOGIT requests a generalized logit link function for
the model. If a three (or more) level outcome is specified in the model statement
without using the LINK¼option, the default analysis is an ordinal logistic regression
which uses a cumulative logit link function (see next section).

The output using PROC LOGISTIC for the polytomous analysis follows:

The LOGISTIC Procedure

Model Information

Data Set REF.CANCER

Response Variable SUBTYPE

Number of Response Levels 3

Model generalized logit

Optimization Technique Newton-Raphson

Number of Observations Read 288

Number of Observations Used 286

Response Profile

Ordered

Value SUBTYPE

Total

Frequency

1257

2145

3 0 184

Logits modeled use SUBTYPE¼0 as the reference category.

Model Fit Statistics

Criterion

Intercept

Only

Intercept and

Covariates

AIC 516.623 510.405

SC 523.935 539.653

2 Log L 512.623 494.405

618 Appendix: Computer Programs for Logistic Regression