The oft-mentioned glmm wiki FAQ recommends a minimum of 5 or 6
levels when using a mixed model. I am not sure how to interpret
this statement. I have 4 experimental treatments, each with 3
replicates, for a total of 12 plots. Measurements were made over 4
years and there are thousands of individual data points within each
of the 12 plots. I was planning to use nested random effects of
plot within treatment within year. So I would have 3 replicate
plots within each of 4 treatments within each of 4 years.
For nested effects, at what level does the recommendation pertain?
Is it only the lowest level? Would 5 or 6 plots within 3 treatments
within 3 years be acceptable? Or should one have 5 or 6 plots
within 5 or 6 treatments within 5 or 6 years? Is the amount of data
within plot irrelevant to this issue?
If 4 treatments are insufficient for a mixed model, is treatment as
fixed effect the only alternative? I need an approach for which
there is a method for multiple comparison of means. I believe lm,
lme, lme4 are the only options since the aov methods do not accept
an Error term.