Projections in Model Space: Multi-model Inference Beyond Model Averaging

A reviewer of the manuscript challenged us to do two things. First, to move beyond simple likelihood ratio examples and show how evidential ideas are used in the practice of science. And, second to solve some deep problem in ecology using this framework. To answer the first challenge, we discuss information criteria differences as a natural extension of the likelihood ratio that overcomes many of the complexities of real data analysis. To answer the second more substantive challenge, we then extend the information criterion model comparison framework to much more effectively utilize the information in multiple models, and contrast this approach with model averaging, the currently dominant method of incorporating information from multiple models (Burnham and Anderson 2002). Model averaging is a confirmation-based-approach. Because of limitations in both time and allowable word count, this will be a sketch of a solution. We deeply appreciate the reviewer’s challenge because the work it has forced us to do has been very rewarding.