[R-sig-ME] advice with lmer syntax

[Christian Ritz]

Dear Giacomo,

I would prefer to include whichever random effects that could be relevant in the initial
model. Instead of checking afterwards by adding more terms and possibly even carry out a
significance tests!

For each random effects term I would ask a question such as:

Is it reasonable to assume that there could be random variation at this level? For
instance it seems sensible to me that there could be some random variation between season
in sites as (I guess) this would reflect variation between different time periods. So the
term (1|Season:Site) makes sense to me.

As you only have 3 sites there won't be that much variation to pick up be the random
effects you include (variation across 3 sites). However, I would still prefer to keep site
as a random effect in model as it is actually randomly chosen sites, and thus the model
reflects the design of the experiment.


Christian