Fred S. Guthery

Introduction * Information and Likelihood Theory: A Basis for Model Selection and Inference * Basic Use of the Information-Theoretic Approach * Formal Inference From More Than One Model: Multi-Model Inference (MMI) * Monte Carlo Insights and Extended Examples * Statistical Theory and Numerical Results * Summary

Northern bobwhites (Colinus virginianus) have received considerable research attention since the 1920s. I evaluated published results for patterns that might be meaningful in developing a general philoso- phy of habitat management for this species. Bobwhite populations show similar mean demographics (sur- vival, productivity) as climates, landscapes, and predator populations vary about them; this suggests some operational constancy in habitat quality wherever populations persist. Neither food abundance nor habitat type interspersion are satisfactory general predictors of population density on a management area, although interspersion provides a limiting condition (after min. interspersion requirements are met, further intersper- sion has, at best, neutral effects on density). Long-term, mean density on an area may vary in proportion to the quantity of space (amt permanent cover) that fits the physical, behavioral, and physiological adaptations of bobwhites through time. The goal of habitat management on an area should be to provide bobwhites the opportunity for unconstrained use of space through time (space-time saturation). This common sense out- look seems to have been obscured by unjustified concerns over food and interspersion and lack of a general understanding of successional affiliation. J. WILDL. MANAGE. 61(2):291-301

We question whether the growing popularity of model selection based on information theory (IT) and using the Akaike's Information Criterion (AIC) represent a useful paradigm shift in data analysis or a substitution of 1 statistical ritual for another, which leaves in place long-standing problems in wildlife science. We discuss the relevance of model selection in science, problems in the IT-AIC algorithm, errors of commission and omission in IT-AIC-based studies, and the role of IT-AIC in knowledge accrual. Model selection is just another minor tool in the grand panorama of science. The human mind, not statistical methods, produces scientific breakthroughs. Although IT-AIC might include elements of hypothetico-deductive science, it is arguably a form of sensitivity analysis, magnitude of effects estimation, or simple description as currently applied. Accordingly, it is largely an inductive approach to knowledge accrual and, therefore, subject to the pitfalls of induction. The algorithm tends to over fit data (i.e, use too many variables), resulting in models that contain useless variables and that generalize poorly. Errors of commission in IT-AIC-based papers include hopelessly uninformative lists of encrypted models and imposition of the model-selection approach on studies better executed in a simple, descriptive format. The major error of omission is an almost universal failure to test selected models on independent data. From our perspective, IT-AIC is a harmless human construct that is being ritualistically applied and therefore cannot be expected to correct long-standing problems in the conduct of wildlife science, such as failure to apply the hypothetico-deductive method. We view the growing application of IT-AIC as problematic because that growth might discourage use of the full panoply of available methods of inquiry. Accordingly, we urge colleagues to avail themselves of the rich pageant of available analytical techniques that can be applied in wildlife research under the hypothetico-deductive method and to keep ecology, rather than statistics, in the forefront of wildlife science.

Steven D. Kopp, Fred S. Guthery, N. David Forrester, Will E. Cohen, Habitat Selection Modeling for Northern Bobwhites on Subtropical Rangeland, The Journal of Wildlife Management, Vol. 62, No. 3 (Jul., 1998), pp. 884-895