model averaging phi p constrained models

[dhewitt]

I'd reiterate that the question of whether model-averaging will help depends on model selection uncertainty and precision of the estimates of interest. Can you tell us how much model selection uncertainty there is without adding these constrained models to the mix? And how precise the reach-specific estimates are in the well-supported models? You say that in the past you have used the fully reach-specific model... so was it most supported (with high weight) and had great estimate precision on Phi?

I think the above is an important starting point. Beyond that, to your question of whether a "spurious model" could get a high weight and cause estimate bias in your approach... I don't see how this could occur. I don't really know what you mean by "spurious model", but given that the models you propose are rather simple constrained versions of other models in the set (the full one), I wouldn't worry much about making the situation worse. It just seems to me that it might be unnecessary.