Model selection criteria?

From: "Catherine Sherwin" catherine.sherwin@stonebow.otago.ac.nz Subject: [NMusers] Model selection criteria? Date: Fri, July 1, 2005 4:34 am Hi NMusers, I would like your help and advice in regards to correct model selection when using NONMEM. I have looked at using the Akaike Information Criterion. The equation I used was the following: AIC= N x log (WRSS) + 2P. I have also looked at using the Schwartz Criterion. The equation was: SC = N x log(WRSS)+ log(N) x P. In both equations, N is the number of observations, WRSS is the weighted residual sum of squares and P is the number of parameters. Can you tell me if these are appropriate guides to model selection for NONMEM? Also can you comment on the use of the following in determining selection of the correct model? 1) Precision (root mean squared prediction error)? 2) Accuracy or bias (mean prediction error)? 3) And reduction in objective function of more than 5.02 (p<0.01)? Regards Catherine Sherwin