I believe that the approach presented by Baogong is nearly, but not quite, the correct approach. I assume that there is correlation among the responses within each observation, and that this correlation should be accounted for. If we were to compare just treatments i and j, then the appropriate analysis would be a paired t-test. Is that correct?

If so, then I would recommend the MIXED procedure to account for these correlations. The procedure GLM could be employed, but is not as flexible as the procedure MIXED. Thus, I would code

proc mixed data=final; class obs treatment; model score = treatment; repeated treatment / subject=obs type=un; lsmeans treatment / pdiff adjust=tukey; run;

or

proc mixed data=final; class obs treatment; model score = treatment; repeated treatment / subject=obs type=cs; lsmeans treatment / pdiff adjust=tukey; run;

Executing each of these indicates that there are no differences between any of the treaments, even without employing the Tukey correction for multiple post-hoc comparisons. Unless there is more data than has been presented, I don't see any need to employ a Tukey test.

I would note that the response takes only three values in the data which are shown. Thus, assuming the response to be normally distributed may not be appropriate. It may be more appropriate to assume that the response values are ordered levels and fit a cumulative logits model accounting for the correlation across treatments within observations. You could employ the procedure NLMIXED to do this. I have posted on this topic to SAS-L several times. One such post can be found at

http://listserv.uga.edu/cgi-bin/wa?A2=ind0203A&L=sas-l&P=R21334

I don't immediately know how to perform a Tukey test with the output from the procedure NLMIXED. But, if the results assuming normality are any indication, there should be no need to perform a Tukey test to account for multiple post-hoc comparisons. You may not have any significant differences to begin with.

Dale

--- baogong jiang <bgjiang@GMAIL.COM> wrote: > hi adriano, try following code, > > > hope this help! > > > > data tep; > input obs p1-p6; > cards; > 1 5 5 5 4 4 5 > 2 4 3 4 5 4 4 > 3 4 4 4 5 4 5 > 4 4 4 . 3 4 4 > 5 5 5 5 5 5 5 > 6 4 . . 4 3 4 > 7 4 5 4 4 4 4 > 8 4 3 4 4 4 4 > 9 3 4 3 4 4 4 > 10 5 4 4 4 4 4 > 11 3 4 4 4 4 4 > 12 4 . . 4 3 4 > 13 4 . . 5 5 4 > 14 4 4 4 4 4 4 > 15 3 3 3 4 4 4 > 16 4 4 4 3 4 3 > 17 4 4 4 4 4 4 > 18 5 5 5 5 5 5 > 19 4 5 4 5 5 5 > 20 . 3 4 4 4 3 > ;run; > proc transpose out=final (rename= ( _name_=treatment COL1=score)); > var p1-p6; > by obs; > run; > proc glm data=final; > class treatment; > model score=treatment; > means treatment/tukey; > run; >

===== --------------------------------------- Dale McLerran Fred Hutchinson Cancer Research Center mailto: dmclerra@fhcrc.org Ph: (206) 667-2926 Fax: (206) 667-5977 ---------------------------------------

__________________________________ Do you Yahoo!? New and Improved Yahoo! Mail - 100MB free storage! http://promotions.yahoo.com/new_mail