PROC REG ....; BY GROUP; RUN;

WenSui Liu (http://statcompute.blogspot.com) Senior Decision Support Analyst Cincinnati Children Hospital Medical Center

The easy AND smart way to do this is by using the BY-processing features of SAS. I'm putting a PROC SORT in here, in case the data are not sorted on GROUP first. BY-processing needs that. proc sort data=a;

by group;

run;

proc reg data=a; by group; model y= x; output out=c2 P=PRED ; Now you have the regressions done separately, by group. BUT you don't have any regression diagnostics or anything, and you have 200 (or so) regressions to grapple with. Are you sure that you are ending up with a good solution from a statistical point of view? Are you sure that your results are meaningful, if you have no checks for outliers or leverage points or non-linearity? Are you sure that your output will be helpful, when 200 separate tests, all at alpha=0.05, would have about 10 apparently significant results just by chance even if the null hypotheses were true? How will you separate out the specious significant value from the real ones? I think you have a host of potential statistical problems awaiting you. You may want to write back to SAS-L and explain more fully what you are trying to achieve in the big picture here. HTH, David -- David L. Cassell mathematical statistician Design Pathways 3115 NW Norwood Pl. Corvallis OR 97330