This is a horribly inadequate analysis, and fails to account for correlated responses over time within a subject and to the different levels of nitrate applied. If it is submitted for publication to any of the Agronomy journals that I am familiar with, it will be rejected. Reformat the data to include nitrate level, day and subject. Try the following then: proc mixed data=newdata; class day nitrate expression subjectid; /*This assumes that root length was measured on the same plant, or on similar plants in a plot */ model count=day|nitrate|expression; repeated day/subject=subjectid; lsmeans day|nitrate|expression; slice day*nitrate*expression/sliceby=day*nitrate diff; run; The slice statement gives the extension of the t-test you propose, testing differences on each day at each nitrate level concentration. There are still multiple testing, variance homogeneity, and distributional questions to be addressed, but this PROC MIXED approach will reflect the experimental design, and is a good start. Steve Denham
... View more
