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

response is not specified. Rather a GEE model requires that the mean–variance
relationship of the response be specified. What the DIST¼BINOMIAL option does
is to define the mean–variance relationship of the response to be the same as if the
response followed a binomial distribution (i.e.,V(Y)¼m(1m)).

The REPEATED statement indicates that a GEE model rather than a GLM is requested.
SUBJECT¼IDNO in the REPEATED statement defines the cluster variable as IDNO.
There are many options (following a forward slash) that can be used in the REPEATED
statement. We use three of them in this example. The TYPE¼AR(1) option specifies
the AR1 working correlation structure, the CORRW option requests the printing of the
working correlation matrix in the output window, and the WITHIN¼MONTH option
defines the variable (MONTH) that gives the order of measurements within a cluster.
For this example, the WITHIN¼MONTH option is unnecessary since the default
order within a cluster is the order of observations in the data (i.e., the monthly
measurements for each infant are ordered in the data from month 1 to month 9).

The ESTIMATE statement with the EXP option is used to request the odds ratio
estimate for the variable DIARRHEA. The quoted text in the ESTIMATE statement is
a label defined by the user for the printed output. The CONTRAST statement requests
that the Score test be performed to simultaneously test the joint effects of the variable
BIRTHWGT and DIARRHEA. If the REPEATED statement was omitted (i.e., defining
a GLM rather than a GEE model), the same CONTRAST statement would produce a
likelihood ratio test rather than a Score test. Recall the likelihood ratio test is not valid
for a GEE model. A forward slash followed by the word WALD in the CONTRAST
statement of PROC GENMOD requests results from a generalized Wald test rather
than a Score test. The CONTRAST statement also requires a user-defined label.

The output produced by PROC GENMOD follows:

The GENMOD Procedure

Model Information

Data Set REF.INFANT

Distribution Binomial

Link Function Logit

Dependent Variable outcome

Observations Used 1203

Missing Values 255

Class Level Information

Class Levels Values

IDNO 136 00001 00002 00005 00008 00009 00010 00011 00012
00017 00018 00020 00022 00024 00027 00028 00030
00031 00032 00033 00034 00035 00038 00040 00044
00045 00047 00051 00053 00054 00056 00060 00061
00063 00067 00071 00072 00077 00078 00086 00089
00090 00092 ...
MONTH 9 1 2 3 4 5 6 7 8 9

SAS 623