RE: posthoc step

From: "Leonid Gibiansky" leonidg@metrumrg.com Subject: RE: [NMusers] posthoc step Date: Wed, December 8, 2004 8:27 am Jerry, Most likely, this is a problem with too extreme data and implementation of POSTHOC in NONMEM. I checked how solution depends on OMEGA. With OMEGA= 8.39 or higher, NONMEM gives k=0.942, pretty close to what you would expect. But for OMEGA=8.38, solution jumps to K=4.19; for OMEGA= 8.37, K= 9.04 and then goes up to 9.78 when you move OMEGA down to 4 (e.g, OMEGA=8, K= 9.27) I also checked the probability of K=1 with mean=10 and various variances: > pnorm(1, mean = 10, sd = sqrt(4)) [1] 3.397673e-006 > pnorm(1, mean = 10, sd = sqrt(8.4)) [1] 0.0009504466 Probability of the realization that you tried is about 3 / million. NONMEM starts to give correct results at about p = 0.001. It this can be generalized, then NONMEM POSTHOC fails at outliers with p < 0.001 Best regards, Leonid