You can fit a GEE model to data that is summarized down to the cluster level only. That is, if you have multiple subjects that match exactly on all predictor and response values for all their measurements, then only one copy of the subject needs to appear in the data and you specify the frequency of its occurrence in the FREQ statement variable. The count values in the FREQ variable apply to the entire cluster indicating the number of times the cluster should be duplicated, and that means that the FREQ variable values must be the same for all observations (measurements) within a cluster. If this is not the case, PROC GENMOD will display the following message in the SAS log:

   ERROR: A cluster has been detected with the frequency counts specified
          by the FREQ statement unequal within the cluster. The frequency counts
          must be equal within clusters.

The following example shows the raw, subject-level data containing all measurements from all subjects and the equivalent summarized data set. The PROC GENMOD statements below each data set would produce identical analyses. Note that the first two subjects in the raw data have identical values for all variables on all three measurements. Consequently, those two subjects can be represented in a summarized data set as a single set of the measurements all with a frequency variable value of 2. The third subject, which has no duplicates in the rest of the data, remains as in the raw data set but with frequency values of 1 for all measurements.

SUBJECT  TIME  RESPONSE
   1       1       0
           4       1
           8       0
   2       1       0
           4       1
           8       0
   3       1       1
           4       1
           8       0
    ... etc.

SUBJECT  TIME  RESPONSE COUNT
   1       1       0      2
           4       1      2
           8       0      2
   2       1       1      1
           4       1      1
           8       0      1
    ... etc.

proc genmod;
  class subject;
  model response = time / dist=bin;
  repeated subject=subject;
  run;

proc genmod;
  freq count;
  class subject;
  model response = time / dist=bin;
  repeated subject=subject;
  run;

Operating System and Release Information