Table 1. Fixed effects selection using the Bayesian Information Criterion (BIC).
| Fixed effects selection | df | BIC | Delta BIC | R2-m | R2-c |
|---|---|---|---|---|---|
| Full-model (round #1) | 33 | 610912.3 | 0 | ||
| SIZE × SBA × POFT (removed) | 32 | 610889.8 | −12.5 | ||
| SIZE × POFT × PET (removed) | 31 | 610937.1 | 24.8 | ||
| POFT × SBA × PET (removed) | 31 | 610900.8 | −11.5 | ||
| SIZE × SBA × PET (removed) | 31 | 610914.1 | 1.8 | ||
| Full-model (round #2)* | 28 | 610891.0 | 0 | ||
| SIZE × SBA (removed) | 27 | 614430.1 | 3539.1 | ||
| POFT × SBA (removed) | 27 | 610885.3 | −5.7 | ||
| PET × SBA (removed) | 26 | 611043.7 | 152.7 | ||
| Best-supported model** | 23 | 610885.3 | 0 | 0.367 | 0.705 |
| null model (intercept only) | 6 | 719146.1 | 108260.8 |
We used a hierarchical backward selection of fixed terms starting with a full model that included all the possible three-variable interactions between the proportion of other functional types (POFT), tree basal area (SIZE, m2), stand basal area (SBA, m2 ha−1), and potential evapotranspiration (PET, mm). We tested the contribution of each interaction by removing them one at a time. R2-m is the marginal pseudo R2 (fixed effects only) and R2-c is the conditional pseudo R2 (including both fixed and conditional effects).
*The round#2 of the backward selection started with the best supported model obtained in the first round.
**Best supported model: SIZE × POFT × PET+SIZE × SBA + PET × SBA.