Quantifying spatial and environmental effects in spatial autoregressive model with the spatialreg.hp R package

Abstract Ecological data frequently exhibit spatial autocorrelation, whereby geographically proximate sampling units are more similar in species distributions, community composition, or environmental attributes than expected by chance. Spatial Autoregressive models (SARs) address this issue by explicitly incorporating spatial dependence. However, disentangling the relative contributions of spatial structure and ecological predictors remains challenging, particularly in the presence of multicollinearity. The ‘spatialreg.hp’ R package extended the concept of average shared variance (ASV) to SARs, enabling the decomposition of total model R2 into unique and shared contributions of spatial and non-spatial predictors. The package calculated individual R2 values for spatial effects and environmental variables that summed exactly to the total model R2, thereby providing a new measure of predictor importance. We illustrated the package using case studies based on classic spatial datasets. The ‘spatialreg.hp’ package offers ecologists and geographers a new framework for quantifying the interplay between spatial processes and environmental drivers in ecological modeling.