Hello. I am trying to figure out how (if?) I can adapt a test of planned means (Dunnett's test?) to a loess fit. Context: The experiment consists of growing plants in hydroponic solution of a period of time (t=0 to crit), after which the nutrient solution concentration is decreased to a new (lower) level. Plants are harvested daily during the course of the experiment and the foliar nitrogen concentration is measured each day from these destructive harvests. The statistical goal is to determine the shortest time following the change in nutrient solution concentration at which one can detect a significant (p=0.05) decrease in the foliar nitrogen concentration. Therefore, I wish to compare the predicted foliar N concentration at t=crit to the foliar N concentrations at time t=crit+1, +2, .. Because the time course of N does not follow any simple parametric function, I am using smoothing splines via loess. Is there any way to adapt a test of planned means (say, Dunnett's test) to this problem? If not, can anyone suggest how this problem can be tackled via loess? Any suggestions are gratefully welcomed. Bill Shipley [[alternative HTML version deleted]]