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

The LOGISTIC Procedure

Analysis of Maximum Likelihood Estimates

Parameter DF Estimate

Standard

Error

Wald

Chi-Square Pr>ChiSq

Intercept 1 2.6163 0.2123 151.8266 <.0001

CAT 1 0.6223 0.3193 3.7978 0.0513

AGEGRP 1 0.6157 0.2838 4.7050 0.0301

ECG 1 0.3620 0.2904 1.5539 0.2126

Contrast Test Results

Contrast DF Wald Chi-Square Pr>ChiSq

1 DEGREE OF FREEDOM TEST 1 13.2132 0.0003

2 DEGREE OF FREEDOM TEST 2 13.4142 0.0012

A difference between the CONTRAST statements in PROC LOGISTIC and PROC
GENMOD is that with PROC LOGISTIC the default test is the WALD test while
with PROC GENMOD the default test is the likelihood ratio test.

D. Using Frequency Weights

Individual level data can also be summarized using frequency counts if the variables
of interest are categorical variables. The dataset EVANS3 contains the same infor-
mation as EVANS2 except that each observation represents cell counts in a four-way
frequency table for the variables CHD, CAT, AGEGRP, and ECG. The variable
COUNT contains the frequency counts. The code that creates EVANS3 follows:

DATA EVANS3;

INPUT CHD CAT AGEGRP ECG COUNT;

CARDS;

1000 17

0000257

1010 15

0010107

1001 7

0001 52

1011 5

0011 27

1100 1

0100 7

1110 9

0110 30

1101 3

0101 14

1111 14

0111 44

;

610 Appendix: Computer Programs for Logistic Regression