Interest in modeling habitat use by individual species has intensified in recent years as ecologists seek to better understand and predict consequences of habitat change. Logistic regression has been one of the most widely used statistical tools in these modeling efforts. Our paper is highly cited because it shows that logistic regression has often been used incorrectly in habitat modeling, and also because it provides practical guidelines regarding correct use and interpretation for the three types of sampling designs most often encountered in ecological studies.

Misapplications of logistic regression in habitat studies reflect an inadequate understanding among ecologists concerning the logistic model, its interpretation, and especially the influence of sampling design. This paper brings together information from the fields of statistics and econometrics, and combines that information with results from simple simulations to illustrate the correct use and interpretation of logistic regression in ecological settings. Although most of this information is not new, it is undoubtedly new to many ecologists.

Our paper promotes correct usage of a key statistical method used to construct habitat models. As such, it fosters development of more credible and effective tools for evaluating consequences of human activities affecting a wide range of species.

We began habitat modeling in 1997, initially using established methods to model ungulate habitat use in the greater Yellowstone area. Mathematical discrepancies in our early results caused us to question the validity of our findings and prompted detailed simulation studies that revealed deeper problems in the modeling methodology itself.

The chief problems encountered in this work have been bureaucratic and social: it is difficult to generate financial support for basic research into modeling methods, and scientists whose programs have been founded on erroneous applications of logistic regression have been reluctant to accept that such applications lack the necessary statistical foundation.

The social and political implications of this work are indirect. Species habitat models have many potential applications; for example, environmental impact assessment, habitat restoration planning, or establishing conservation priorities. In turn, these activities can have important social and political consequences, making it essential that the underlying science be grounded in sound statistical practice.