Ecological Archives E096-124-A4
Maya C. Pfaff, George M. Branch, Jennifer L. Fisher, Vera Hoffmann, Allan G. Ellis, and John L. Largier . 2015. Delivery of marine larvae to shore requires multiple sequential transport mechanisms. Ecology 96:1399–1410. http://dx.doi.org/10.1890/14-0229.1
Appendix D. Background on the application of Model Selection sensu Burnham and Anderson (2002).
The information-theoretic approach of model selection is favored over the better-known significance-based hypothesis testing if several competing a priori hypotheses exist for processes underlying observed patterns. In particular, it is preferred for situations where an experimental approach is not possible, such as in historical or observational studies (Burnham and Anderson 2002, Johnson and Omland 2004).
Model selection requires formulating a candidate model for each explanatory hypothesis, and a global model incorporating all parameters, which is assessed for goodness-of fit. If the global model fits the data, then the AIC-selected model will also fit. Each model is then assigned an AIC value, which quantifies model fit (by log-likelihood) while penalizing for additional parameters. Adjustments can be made to the AIC for small sample size (AICc), overdispersed count data (QAIC) or both (QAICc). Candidate models are then ranked according to the (Q)AIC(c) scores (the lower the value, the better the fit), and Akaike weights are calculated to quantify the evidence in favor of each model relative to the other candidate models (Burnham and Anderson 2002).
Literature cited
Burnham, K., and D. Anderson. 2002. Model Selection and Multimodel Inference: a Practical Information-Theoretic Approach. Second Edition. Springer, New York, New York, USA.
Johnson, J., and K. Omland. 2004. Model selection in ecology and evolution. Trends in Ecology and Evolution 19:101–108.