Appendix A. Supplemental methods.
Following Burnham and Anderson (2002), model-averaged estimates of effects () for each confidence set were calculated as:
where was the maximum likelihood estimate of an effect from model i (or zero if the associated effect was excluded from the model variation) and was the Akaike weight of model i. Model-averaged variance estimates () also were calculated:
where was the Akaike weight of model i, was the bootstrap-derived estimate of variance about the maximum-likelihood estimate of from model , and was the model-averaged estimate of the parameter of interest. If an effect was excluded from model , and were fixed to zero. This formulation for estimating model-averaged variance included a component reflecting model selection uncertainty and thus was not conditioned on any single model in the confidence set.