[R-sig-ME] using AIC or AICc with lme to select random effects

[Rian Dickson]

hello all,

This is my first post to the list, and I hope that I am not asking questions that are too elementary.

I am using lme to run linear mixed models, and I want to use the AIC values of the models to guide my choice of terms to be included in the random effects. However, I'm not sure if I should be using the AIC or the AICc (corrected for small sample sizes).
I am dealing with repeated measures on individuals, and have 454 observations on 91 individuals. My most highly-parameterized model includes 11 fixed effects.

>From what I have read, I understand that while it may be preferable to use the AICc due to sample size, it is difficult (impossible?) to determine what value to use for the number of observations needed to calculate AICc. Is this correct? If so, is there a consensus on what should be done in situations like this?

My data structure is as follows, where scoterID is the subject ID and Log.minH is the response variable.

$ scoterID            : Factor w/ 91 levels "A","AA","AAA",..: 1 1 1 1 5 5 5 5 5 5 ...
$ site                : Factor w/ 2 levels "PSGB","SEAK": 1 1 1 1 1 1 1 1 1 1 ...
$ cohort              : Factor w/ 4 levels "FASY","FSY","MASY",..: 2 2 2 2 3 3 3 3 3 3 ...
$ emerg               : num  71.1 71.1 71.1 71.1 36.5 36.5 36.5 36.5 36.5 36.5 ...
$ pri                 : num  20.5 38.9 51.9 54.6 35.4 42.7 59.9 64.8 69.7 79.5 ...
$ priSQ               : num  422 1510 2698 2977 1252 ...
$ start.t             : num  14 13.1 11.9 10.6 8.7 ...
$ startSQ             : num  196 171.1 142.3 112.8 75.7 ...
$ sea                 : Factor w/ 3 levels "calm","mod","rough": 1 1 1 1 2 3 1 1 2 1 ...
$ tidem               : num  3.66 3.13 3.85 2.45 1.13 1.9 3.56 1.48 1.69 1.97 ...
$ tided               : Factor w/ 3 levels "fall","rise",..: 2 3 2 2 1 2 2 2 1 3 ...
$ year              : Factor w/ 2 levels "2008","2009": 1 1 1 1 1 1 1 1 1 1 ...
$ Log.minH            : num  0 2.878 0 0.322 0 ...

My most highly parameterized model is:
Log.minH ~ site + year + cohort + emerg + pri + priSQ + start.t + startSQ + sea + tidem + tided

And the random effects that I want to test are:
random = ~1|scoterID
random = ~site|scoterID
random = ~year|scoterID
random = ~emerg|scoterID
random = ~pri + priSQ|scoterID
random = ~start.t + startSQ|scoterID
random = ~sea|scoterID
random = ~tidem|scoterID
random = ~tided|scoterID

I hope that I have provided enough information. Any advice would be much appreciated.

thank you,

Rian Dickson

*************************************

Rian Dickson
M.Sc. candidate
Centre for Wildlife Ecology
Department of Biological Sciences
Simon Fraser University
8888 University Drive
Burnaby BC V5A 1S6
778.782.5618