It depends on your Hypothesis. You can use both.

Bruno

Have you used Delta AIC as a criterion for model selection? What kind of improvement in R-squared are you seeing when you add all the variables? Does adding the variables really make ecological sense to you or were they just potential predictors you added as after thoughts to your model? I'm not second guessing you because you know your data. I don't. I simply make this suggestion because I believe everyone speculates to infinity and beyond - to quote a cartoon character. Anyway, seems to me that what you want is the simplest out the eight possible models that fit your data. For example, if your R squared is say .68 with 4 variables and .74 with 8 is that really and improvement or is that just noise? If the AIC for doesn't change much (rule of thumbs vary I've seen 2, 3, or 10) maybe you might reconsider which model you are interested in. And if that doesn't help well, plotting the significant ones is probably the best approach to help you figure out what the heck is going on. I've played a little bit with this stuff but I'm no expert by any means. Just my two cents. Hope it helps.

Hi Mike, thanks for the answer. I agree that plotting the significant predictors is a good way to show the results of the averaged model.