Hi, I have been doing population modeling on data using an extravascular 3-compartment model and have experienced some phenomenon I find odd. I have two models, only difference is the initial parameters used: - one where the parameters obtained are extremely close to the intial estimates and give extremely low parameter CV% (lower than 1%) and give an AIC of -979. - the other has parameters that are far from the initial estimates but then gives very high CV% such as 60% up to 200% for the volume. The 2.5% confidence interval now includes negative values! The AIC is of -931. This model also shows better predictive ability on new data. I am wondering what the reason behind all of this is? Does the first model mentioned encounter a local minimum or does the engine simply fail because of the numerous parameters? I find especially odd that despite such high CV% values the AIC doesn’t seem so bad compared to the first one. I am using multiplicative weighting and the naïve-pooled engine. No covariates are included in the study. I hope you can help. Best regards, Yassine