www.cmm.bristol.ac.uk/forum

Hi all

I have a large number of variables for potential inclusion in a 2-level model. If this were a single level model I would apply a sequential modelling strategy: identify a number of blocks of variables and enter each block in order of block priority.

I'm not sure if or how the existence of the hierarchical structure changes this. I could do the above for blocks comprising only level 1 variables; then repeat for level 2 blocks - but if I went down this route, I'm not sure whether L-1 or L-2 blocks should be higher priority. Or I could enter blocks in order of increasing priority regardless of whether they comprised L-1 or L-2 variables (but which each block containing only L-1 variables or L-2 variable), but not both). Or I could create blocks that include both L-1 and L-2 variables - i.e. completely ignore the ML structure.

Any advice gratefully received.
Thanks
John

Hi John,
Good question. I think probably the answer is to not worry overly about the levels when choosing your blocking though often your blocks might naturally group into levels anyway. The bigger challenge is when you move on to random slopes and deciding which variables to allow to vary at higher levels.
Best wishes,
Bill.

Thanks Bill - good to know that the structure may not be the constraint I thought it might be. I take your point about the random slopes - I will have to give that some careful consideration.
John