why is K10=K01 a special scenario in Compartment modeling, what is banana and funnel plot in residual plots and what does it mean with regard to model misspecification and residual error misspecification? What is weighting? When would I use I/Yhat?
Please help me understand these concepts?
if you choose k10=k01, then you reduce the model from 3 parameters to two, so it is a simpler model that maybe advantageous in terms of parameter precision etc if it is justified by the data you are trying to fit.
‘Banana’ and ‘funnel’ describe the shape of the distribution seen in residual plots, the former you have under then over prediction then under prediction again (or the reverse) and often suggests model misspecification.
When you see fanning of your data particularly at the lower concs on pred conc vs obs conc plot (i.e. funnelling to towards the right) then you may want to use 1/Yhat^2 to give more weight to those lower concentrations although if you use the Phoenix model engine a multiplicative error model might be a good equivalent but you can also apply more sophisticated residual error models. This is generally the better engine to use in compartmental modelling as it supports many more approaches than the classic WNL engine.
you could also look to at the on demand or in person courses avaialbe through http://www.certarauniversity.com/lms/
Simon
Thank you so much sir!!