Spatial autocorrelation in country-level models of species imperilment

Modeling the determinants of species' ecological fragility using country-specific data may be complicated by the fact that factors that influence species imperilment may extend or operate beyond arbitrary political boundaries. Following McPherson and Nieswiadomy [McPherson, M.A. and Nieswiadomy, M.L., 2005. Environmental Kuznets curve: threatened species and spatial effects. Ecol. Econ. 55: 395–407.], we confirm the advisability of controlling for spatial autocorrelation in models focusing on imperilment of birds, mammals, reptiles, amphibians, and vascular plants. We also compare the performance of different definitions of the spatial dependency. Although our a priori expectation was that measures that more accurately reflect the degree of spatial interaction between countries, such as the percentage of shared border, would be superior to a measure of simple adjacency, in fact we find that the simple adjacency measure outperforms the other measures in most cases.