Simulation vs. actual data

From: "Gordi, Toufigh" Toufigh.Gordi@cvt.com Subject: [NMusers] Simulation vs. actual data Date: Tue, June 14, 2005 2:27 pm Dear all, I have a trivial question I hope I can get help with. I have data from different doses of a compound and want to show a plot of concentration mean (+/-SD) values vs. time and a fit of the model curve that, hopefully, goes through the values. To get the model fitting curve, I plan to simulate a typical subject per dose group receiving doses used in the study. My question is what simulated values to use for the curve. A population curve does not seem to be a good choice: the observed data contains known and unknown errors, whereas a population curve is the true value of the population and thus the fit might not include all the observed values. The simulated IPRED or DV for a single subject per dose group is certainly not the curve either. The only way I can think of is to simulate a large number of subjects per dose group and generate a mean curve from the simulations. If this is the correct way, I would use the simulated DV values to get the mean values, since DV values include all the errors associated with the real observations and should give the best curve of the observed data. Any comments? Toufigh Gordi