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

Testing Global Null Hypothesis: BETA¼ 0

Test Chi-Square DF Pr>ChiSq

Likelihood Ratio 18.2184 6 0.0057

Score 15.9442 6 0.0141

Wald 13.9422 6 0.0303

Type 3 Analysis of Effects

Effect DF

Wald

Chi-Square Pr>ChiSq

AGE 2 5.9689 0.0506

ESTROGEN 2 3.5145 0.1725

SMOKING 2 6.5403 0.0380

Analysis of Maximum Likelihood Estimates

Parameter SUBTYPE Estimate

Standard

Error

Wald

Chi-Square Pr>ChiSq

Intercept 2 1.2032 0.3190 14.2290 0.0002

Intercept 1 1.8822 0.4025 21.8691 <.0001

AGE 2 0.2823 0.3280 0.7408 0.3894

AGE 1 0.9871 0.4118 5.7456 0.0165

ESTROGEN 2 0.1071 0.3067 0.1219 0.7270

ESTROGEN 1 0.6439 0.3436 3.5126 0.0609

SMOKING 2 1.7910 1.0463 2.9299 0.0870

SMOKING 1 0.8895 0.5253 2.8666 0.0904

Odds Ratio Estimates

Effect SUBTYPE

Point

Estimate

95% Wald

Confidence Limits

AGE 2 1.326 0.697 2.522

AGE 1 2.683 1.197 6.014

ESTROGEN 2 0.898 0.492 1.639

ESTROGEN 1 0.525 0.268 1.030

SMOKING 2 0.167 0.021 1.297

SMOKING 1 2.434 0.869 6.815

In the above output, there are two parameter estimates for each independent vari-
able, as there should be for this model. Since the response variable is in descending
order (see the response profile in the output), the first parameter estimate compares
SUBTYPE¼2 vs. SUBTYPE¼0 and the second compares SUBTYPE ¼1 vs.
SUBTYPE¼0. The odds ratio for AGE¼1 vs. AGE¼0 comparing SUBTYPE¼ 2
vs. SUBTYPE¼0isexp(0.2823)¼1.326.

PROC GENMOD does not have a generalized logit link (link¼glogit), and cannot run
a generalized polytomous logistic regression.

SAS 619