[R-sig-ME] Properly dealing with study design

[Tom Bishop]

Hi list, I hope you are all well.

I have hierarchical dataset and am having difficulty deciding whether I 
have specified my random terms correctly.

I am trying to determine which environmental factors are influencing 
species richness. My data are structured like so:

I have a dataset of insect communities sampled from permanent plots 
along an altitudinal gradient for 7 years and for 2 seasons in each 
year. There are 8 altitudinal bands at which samples have been taken and 
4 independent replicate communities sampled within each band. This gives 
me 4 replicates * 8 altitudinal bands * 2 seasons * 7 years = 448 rows 
in my dataframe. So essentially, the same 32 plots have been sample 
twice yearly for a number of years.

I also have variable describing available area of the altitudinal bands 
(hence, there are only 8 unique values of area) and the temperature of 
the sites at the sampling time (at the replicate level).
I am not interested in the effect of year on species richness and 
understand that a mixed effects modelling approach may be appropriate to 
deal with this. I am, however, interested in how season, altitude, 
temperature and available area influence my species richness values.

I have been reading Zuur et al and scouring the web but have come up 
against a bit of a wall with specifying my model correctly. I think I 
should be starting with a full model as follows:

model <- lme(species.richness ~ 1 + altitude * season * area * 
temperature, random = ~ 1 | year)

I am unsure if I am fully accounting for the structure in my dataset 
here. Do I need to somehow specify the structure more explicitly? For 
example, if season was dropped from the model following simplification 
then my data will surely be temporally pseudoreplicated within year. I 
suspect that I should add in the "replicate" variable in the model 
somehow. Each of the 32 replicates is uniquely coded (1a, 1b, 1c, 1d, 
2a, 2b etc).

Any thoughts, or links to similar examples, on how to correctly specify 
my full model would be much appreciated.

Thanks in advance,

Tom