Hi Helen,
If I understand your situation correctly, it makes sense that the deltas do
not change but the weights do. The minimum AIC will always be the minimum,
regardless of how you change the delta threshold. In comparison, the
weights are calculated off of the models compared. So, if you have 100
models, but you only compare 10, the weight refers only to the 10 involved
in the comparison.
Hope this helps!
Cheers,
Sherry
{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}
{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}
{}:{}
{}:{} Sherry L. Martin
{}:{} Michigan State University
{}:{} Department of Fisheries and Wildlife
{}:{} Program in Ecology, Evolutionary Biology and Behavior
{}:{} 13 Natural Resources
{}:{} East Lansing, MI 48824
{}:{} marti686@ msu.edu
{}:{} (517) 432-8048
{}:{}
{}:{} http://michiganstate.academia.edu/SherryMartin
{}:{}
{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}
{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}:{}
At 09:00 PM 4/20/2010 -0400, Helen Bothwell wrote:
In working with AICc model averaging and selection, I am not sure what the
precedence is for choosing a subset of models to compare. We are
investigating genetic loci under supposed adaptive selection pressure from
environmental variables. We have no prior expectations of which specific
variables may be influencing certain loci and can not a priori reduce
models for comparison. Our data set includes a large number of plausible
explanatory environmental variables. We initially compared a subset of
models with delta AIC values ranging from 0-10. I have found information
saying that delta AIC values from 0-2 indicate that a given model should
be considered within the range of plausible models for the set under
investigation. Coding in R for a given locus, when I change the input
call for the subset of deltas to compare (I've tried 10, 5, and 3) the
Deviance, AICc, and Delta of the various models do not change, however the
weights given to the various models do change as do the values of Relative
Variable Importance. For example, with input delta <10, the best model
weight is 0.03, whereas it is 0.074 with input delta <3. Is there a
precedence for how many models to compare, a subset Delta AICc cut-off
line for model comparison? Any feedback is much appreciated!
Sincerely,
Helen Bothwell
