1 reply to this topic

[LLLi]

Hi all,

 

I fitted a 2 compartment model to a set of data (only one subject) using different residual error structures. Please see the phoenix project attached.

 

I compared the Ind DV vs IPRED; lnd DV, IPRED vs IVAR (y is logarithmic scale); lnd IWRES vs IPRED; and Ind IWRES vs IVAR. I also gave a score to each structure for each figure (score 1 means the best, score 3 means the worst). Please see the excel attached. Which residual error structure is the best one?

 

Thank you!

 

LLLi

Attached Files

•  Exercise 22 one subject.phxproj   927.1KB   504 downloads
•  score.xlsx   8.11KB   523 downloads

[Simon Davis]

I would personally go with mixed, it has the lowest AIC of the 3 models and generally the best residual plots.

 

I suggest you take another look at Ind IWRES vs IPRED, since for mixed there are no points outside the 2 sd lines and generally are closer to the centre line than additive or indeed multiplicatve alone

 

  Simon