Author(s): Boyce MS, Vernier PR, Nielsen SE, Schmiegelow FKA
A resource selection function (RSF) is any model that yields v alues proportional to the probability of use of a resource unit. RSF models often are fitted using generalized linear models (GLMs) although a v ariety of statistical models might be used. Information criteria such as the Akaike Information Criteria (AIC) or Bayesian Information Criteria (BIC) are tools that can be useful for selecting a model from a set of biologically plausible candidates. Statistical inference procedures, such as the likelihood-ratio test, can be used to assess whether models de v iate from random null models. But for most applications of RSF models, usefulness is e v aluated by how well the model predicts the location of organisms on a landscape. Predictions from RSF models constructed using presence/absence (used/ unused) data can be e v aluated using procedures de v eloped for logistic regression, such as confusion matrices, Kappa statistics, and Recei v er Operating Characteristic (ROC) cur v es. Howe v er, RSF models estimated from presence/ a v ailable data create unique problems for e v aluating model predictions. For presence/a v ailable models we propose a form of k -fold cross v alidation for e v aluating prediction success. This in v ol v es calculating the correlation between RSF ranks and area-adjusted frequencies for a withheld sub-sample of data. A similar approach can be applied to e v aluate predicti v e success for out-of-sample data. Not all RSF models are robust for application in different times or different places due to ecological and beha v ioral v ariation of the target organisms.