BARRY@VM.UTCS.UTORONTO.CA (Barry Lay) (02/13/90)
The PROC TABULATE question reminds me of a problem I had a while ago. I wanted to analyse a set of data which contained frequencies of occurances of certain actions within various classes. In addition, the total number of occurances for each action was given, which was usually greater than the sum of the classified data. The discrepancy was due to the fact that the classified data were sampled, whereas the totals were obtained by an accurate count. I wished to spread the difference (total - sum) over the various classes by weight; that is, to each classified frequency I wished to add the difference multiplied by the frequency in the class divided by the sum of the classes. I accomplished this by a couple of PROC SUMMARYs and a bunch of data steps, but I would be interested in a simpler solution. This is a case where the output of PROC TABULATE might have been helpful, as it quite readily figures out percentages of classes within groups. In case anybody is interested, this comes up in reducing RMF data for MVS performance analysis. Barry