Hi Jeanette,
thanks for your input, I understand the bit with going back to an easier model in the presence of an interaction. Sorry if I haven't been clear... From your previous message I thought that I could add a single column with the nuisance variable to the interaction model, plus 2 columns with age split. But that would only work if I had only one group (I attach this for better clarity). The other option is that I have 2 group interaction model and I separate the nuisance variable as well, so I would have 2 columns for nuisance, and 2 for age. Since I want to get rid off the nuisance effect, I have the feeling it needs to be the model I've attached (or could I combine the 2 nuisance columns using a contrast in the 2nd model??)
Chossing the model may be irrelevant though as I think, please forgive me if I am wrong, using OLS or randomise won't make a difference (as opposed to FLAME), no matter which of the models I have described. Wouldn't both models give the same results?
Thanks again,
Torsten
|