Model selection and the cult of AIC

Speaker: Mark Brewer

Abstract: Model selection is difficult. Even in the apparently straightforward case of choosing between linear regression models, there does not yet appear to be consensus in the statistical ecology literature as to what to do. A series of Forum articles in the journal Ecology last year dealt with part of the issue: the distinction between AIC and p-values. There are, however, more fundamental concerns with regard to types of model comparison and to why models are being compared. In this talk, works seeking to evangelise AIC in the context of model selection will be criticised, and it will be argued that AIC and p-values are just different points on the same curve. Differences from using AIC, BIC and p-values (the last via either Likelihood Ratio Tests or stepwise regression) will be explored, and a strategy proposed which differentiates between explanation and prediction, a key distinction in model selection.

This presentation was part of the Methods in Ecology and Evolution 5th Anniversary Symposium. To see other videos from the Symposium, please visit our website: bit.ly/1FRpIMV