Confronting collinearity: comparing methods for disentangling the effects of habitat loss and fragmentation

Estimating the relative importance of habitat loss and fragmentation is necessary to estimate the potential benefits of specific management actions and to ensure that limited conservation resources are used efficiently. However, estimating relative effects is complicated because the two processes are highly correlated. Previous studies have used a wide variety of statistical methods to separate their effects and we speculated that the published results may have been influenced by the methods used. We used simulations to determine whether, under identical conditions, the following 7 methods generate different estimates of relative importance for realistically correlated landscape predictors: residual regression, model or variable selection, averaged coefficients from all supported models, summed Akaike weights, classical variance partitioning, hierarchical variance partitioning, and a multiple regression model with no adjustments for collinearity. We found that different methods generated different rankings of the predictors and that some metrics were strongly biased. Residual regression and variance partitioning were highly biased by correlations among predictors and the bias depended on the direction of a predictor's effect (positive vs. negative). Our results suggest that many efforts to deal with the correlation between amount and fragmentation may have done more harm than good. If confounding effects are controlled and adequate thought is given to the ecological mechanisms behind modeled predictors, then standardized partial regression coefficients are unbiased estimates of the relative importance of amount and fragmentation, even when predictors are highly correlated.