Relative Variable Importance

[Bret Collier]

I was just wondering if there is a standard for interpreting relative variable importance using summed Akaike weights across candidate model set. Burnham and Anderson (2002: pages 167-169) compare the summed weights to the Akaike weight of the "best fitting" model.

For example, if you have a best model with a Akaike weight of 0.40, with 3 predictors(A-C) across the model set with summed weights A=0.97, B=0.88, and C=0.27, you interpret them as A and B are substantially more important than C in the candidate model set.

But, if you have the same best fitting model (Akaike weight=0.40) with A=0.55, B=0.46, and C=0.35, is the interpretation the same? Or is the actual difference between the summed weights important? If so, what is the correct interpretation?

Any comments would be welcome.

Thanks in advance,