Coefficient shifts in geographical ecology: an empirical evaluation of spatial and non-spatial regression

摘要: A major focus of geographical ecology and macroecology is to understand the causes spatially structured ecological patterns. However, achieving this understanding can be complicated when using multiple regression, because relative importance explanatory variables, as measured by regression coefficients, shift depending on whether explicit or non-spatial modeling used. extent which coefficients may why shifts occur are unclear. Here, we analyze relationship between environmental predictors distribution species richness, body size, range size abundance in 97 multi-factorial data sets. Our goal was compare standardized partial ordinary least squares regressions (i.e. models fitted without taking autocorrelation into account; ‘‘OLS models’’ hereafter) eight spatial methods evaluate frequency coefficient identify characteristics that might predict likely. We generated three metrics sets shifts. Typical data, residuals OLS found most The varied they minimized residual autocorrelation. Patterns also among datasets, although magnitudes tended small all cases. were unable strong shifts, including levels either variables model residuals. Thus, changes depend method used largely idiosyncratic, making it difficult occur. conclude cannot evaluated with confidence irrespective modelling not. Researchers have little choice but more about uncertainty cautious their interpretation.