Example 23.3 ("Cluster Analysis of Fisher Iris Data") in the SAS/STAT User's Guide (for both Version 8 and SAS 9) shows one way to handle large data sets by first running PROC FASTCLUS, then PROC CLUSTER, and finally combining the results. To combine the results, a macro called CLUS is defined. The macro contains a MERGE statement which will cause incorrect results:
merge prelim out;
If this MERGE statement is used, then some of the data values will be overwritten by the means of the variables. There is no obvious way to detect that this happened.
The correct MERGE statement to use is:
merge out prelim;
Note that the correction also needs to be made to the sample library program corresponding to this example: "clusex3.sas"