Model selection statistics for alternative mortality models.

Models describe mortality processes for small (D<20 cm) and large (D≥20 cm) trees during three census periods. Models were fitted in a Bayesian framework using MCMC (Markov chain Monte Carlo) simulations. DIC (Deviance Information Criterion) was used to identify the best-fitting model [51], the model with the lowest DIC having strongest support. ΔDIC is the difference in DIC between the best-fitting ‘main effects’ model (either the size-symmetric or size-asymmetric model) and the two other alternatives, with the best initial model having ΔDIC = 0. Negative ΔDIC values for full models (which include interactions) indicate that the incorporation of the interaction improved the model. AUC provides a measure of overall accuracy of the model at all probability thresholds [53].