Landscape-level Modeling and Decision Support for Wildlife Management

Use of Logistic Regression in Wildlife Habitat-selection Modeling
Logistic regression is one of the most widely used statistical tools for modeling wildlife-habitat relationships. However, frequent misapplication of this method by wildlife scientists reflects an inadequate understanding of the logistic model, its interpretation, and the influence of sampling design. To encourage proper application, the correct use and interpretation of logistic regression was reviewed for 3 common sampling designs, and guidelines for applying logistic regression were offered for each. A particularly controversial finding was that logistic regression is generally inappropriate for modeling habitat selection in studies that employ a use–availability design, whereby 2 random samples are drawn independently, one from habitats available to the animal, another from habitats used by the animal. Because this is perhaps the most common sampling design in wildlife-habitat research, results of this study underscore the need for further work to develop credible statistical methods for modeling habitat selection in the use-availability setting.

Robust Methods for Modeling Probability of Use
Habitat selection models have become an important wildlife conservation tool that, ideally, allow one to estimate expected relative densities of animal use over a landscape, and to forecast likely effects of habitat change. Unfortunately, no generally robust method for building such models has yet been proposed. Working from first principles, a simple conceptual model of the habitat-selection process suggests that nonparametric methods may offer important advantages for modeling probability of use. In this study, the accuracy and precision of selected nonparametric approaches is being examined to test this prediction.

Return to Big Game Species