Karen Masse <KMasse@PREVISIONMARKETING.COM> wrote [in part]: > I am looking to take a two part reporting process (SAS processing then Excel > formatting) and put it into one process. Step 1 performs a proc means by a > class variable and determines whether the class mean for each variable is > statistically significant compared to the population mean. Thus, the final > dataset contains the following variables: > > _Label_ > _Freq_ > X01 - Xn... > P01 - Pn...

You have already received two expert replies on the question you asked. But I have a comment on the question you DIDN'T ask.

You are apparently making multiple comparisons here. This is a well-known way to get into trouble if you are not being careful about how you define your "p-value". Are you defining an experiment- wise error rate, or are you just testing a bunch of differences? After all, just due to variability, someone has to come out at the maximum, and someone has to come out at the minimum. Are you dealing with these outcomes in a careful, statistical manner, or are you just assumin that the minimum and maximum are distributed in the same way as everything else?

We on the list cannot tell what is actually being done in the code. But if you would like to discuss the potential statistical issues involved, you can write back to the list and clarify how your differences are being computed, and how your p-values are being calculated.

HTH, David -- David Cassell, CSC Cassell.David@epa.gov Senior computing specialist mathematical statistician