I think you raised some very good question that many people in the forum will find interesting. When you have worked out your solutions, perhaps you can post it back to the forum (when all confidential data is stripped off). We will all benefit from that.
Bernd
Thank you for this informative demo.
A question to ask: any particular reason to choose power residual model as it is not often used? Or you have run in preliminary test to select from additive/multiplicative/mixed/power model and found out power residual model is the best?
Thanks
Jo
Hi, I think the poisson error model (power 0.5) is just less well known. I have found it a useful option sitting between additive and multiplcative.
Generally if my data is within one order of magnitude then I will start with additive, if several orders of magnitude then I start with multiplicatve. Ultimately the choice will be driven by the data and i use the residual plots, particularly IWRES vs IPRED to assess the one that enables the best fit.