Table 5. Statistical models for tree persistence. For each model, the table shows the predictor variables entering the model (remaining variables did not enter), respective direction of association (positive + or negative -) with the response variable (presence of cork oak persistence), the AICc value, AICc differences (Δi) and Akaike weights (wi), and model fit (full model χ²) and area under curve (AUC±s.e.). Models are ordered by increasing Δi, with the best model having Δi=0 (model 1) and the worst model having Δi=29.0 (model 12). The type of model is also coded according to the method used: (a) separate models for each variable, (b) models for each set of biophysical and management variables, (c) stepwise models for each set of biophysical and management variables, (d) stepwise models with the whole set of variables. Codes for management variables: underst= understory management pre 1975; cuttree= tree cutting post 1975; afforest= tree planting post 1985.

Variables
Model aspect slope fire underst cuttree afforest AICc Δi wi χ² AUC
1 d   + - -   - 22.618 0.000 0.759 34.45 0.98±0.019
2 d     - -   - 24.922 2.304 0.240 29.41 0.95±0.035
3 c       -   - 37.062 14.444 0.001 14.71 0.83±0.069
4 b       - - - 38.583 15.965 0.000 15.75 0.85±0.066
5 c   + -       39.933 17.315 0.000 11.84 0.80±0.084
6 a       -     40.265 17.647 0.000 9.11 0.69±0.087
7 a     -       41.612 18.994 0.000 7.76 0.73±0.097
8 a           - 43.947 21.329 0.000 5.43 0.70±0.096
9 b - (S) + -       44.481 21.863 0.000 12.59 0.81±0.080
10 a   +         45.902 23.284 0.000 3.47 0.66±0.100
11 a         -   48.278 25.660 0.000 1.01 0.56±0.100
12 a - (S)           51.661 29.043 0.000 0.12 0.52±0.103

less probability of tree persistence in southern exposures