The interpretation of habitat preference metrics under use-availability designs

Abstract

Models of habitat preference are widely used to quantify animal-habitat relationships, to describe and predict differential space use by animals, and to identify habitat that is important to an animal (i.e. that is assumed to influence fitness). Quantifying habitat preference involves the statistical comparison of samples of habitat use and availability. Preference is therefore contingent upon both of these samples. The inferences that can be made from use versus availability designs are influenced by subjectivity in defining what is available to the animal, the problem of quantifying the accessibility of available resources and the framework in which preference is modelled. Here, we describe these issues, document the conditional nature of preference and establish the limits of inferences that can be drawn from these analyses. We argue that preference is not interpretable as reflecting the intrinsic behavioural motivations of the animal, that estimates of preference are not directly comparable among different samples of availability and that preference is not necessarily correlated with the value of habitat to the animal. We also suggest that preference is context-dependent and that functional responses in preference resulting from changing availability are expected. We conclude by describing advances in analytical methods that begin to resolve these issues.

Figure 1.

This simulated example illustrates how preference is conditional upon the sample of available habitat. Fifty hypothetical use locations (black dots) were generated in each of two landscapes (a,b; see insets for magnified view of use locations). The background depicts the distribution of forest (white) and meadow (grey) vegetation types. The scale at which availability was sampled ranged from 200 m to 5 km in 200 m intervals (black lines). For clarity, only every fifth contour is shown, representing scales of 1, 2, 3, 4 and 5 km. Preference for meadow was estimated using logistic regression at each sampling scale. (c) The relationship between the regression coefficient for meadow and sampling scale in each landscape (black and white circles correspond to landscapes a and b, respectively). (d) The same coefficients plotted against the proportion of meadow available at each sampling scale. The inverse logit-shaped line is the prediction based on an analytical solution of this problem (see text).