Results of spatial regression models relating environmental and historical factors to compositional axes.

We used spatial error models to identify the most important factors (environment or history) for species composition taking spatial autocorrelation into account. To select explanatory variables sequentially we used a simple forward approach. At each step from the null model, we used the Likelihood Ratio test (LR test) to assess the significance of adding a new explanatory variable in the model and the Bayesian Information Criteria (BIC) to select the most important variable. Best spatial models (lowest BIC) are given for the three compositional axes. The value and significance of the spatial autoregression coefficient (λ) is also given for the best spatial models.
*For convenience, the spatial term (λWu) and coefficients (β) have been omitted in the model description (see Material and Methods for details).