MODEL WEIGHTS AND THE FOUNDATIONS OF MULTIMODEL INFERENCE

摘要: Statistical thinking in wildlife biology and ecology has been profoundly influenced by the introduction of AIC (Akaike's information criterion) as a tool for model selection basis averaging. In this paper, we advocate Bayesian paradigm broader framework multimodel inference, one which averaging are naturally linked, performance AIC-based tools is evaluated. Prior weights implicitly associated with use seen to highly favor complex models: some cases, all but most parameterized models set virtually ignored priori. We suggest usefulness weighted BIC (Bayesian computationally simple alternative AIC, based on explicit prior probabilities rather than acceptance default priors AIC. note, however, that both procedures only approximate exact Bayes factors. discuss illustrate technical difficulties factors, approaches avoiding these context logistic regression. Our example highlights predisposition weighting suggests need caution using computing posterior weights.