Discrete-Choice Modeling in Wildlife Studies Exemplified by Northern Spotted Owl Nighttime Habitat Selection

Abstract

Discrete‐choice models are a powerful and flexible method for studying habitat selection, in part because they allow resource availability to change at every choice. Here, we consider application of discrete‐choice models to data typically collected in wildlife science because different discrete‐choice data are usually collected in other disciplines. We generalize the classic discrete‐choice model to the situation in which multiple choices are made from 1 or more choice sets, and only 1 random sample from each choice set is available. We discuss analysis using 1) logistic regression, 2) maximum likelihood when choices are made with replacement, 3) maximum likelihood when the temporal order of selection is known, and 4) maximum likelihood when the order of selection is unknown. We show that 1) provides a good approximation to discrete choice models if the expected number of uses is much <1 for all units. We show that 2) and 3) can be fit using stratified Cox proportional hazards software. Analysis 4) must be fit using special purpose maximization routine such as Newton‐Raphson. Finally, we demonstrate 2) on a case study of nightime habitat selection by 28 northern spotted owls (Strix occidentalis caurina), and conclude that these owls selected for locations low on the slope in stands >41 years old with high levels of hardwoods adjacent to stands 6–20 yrs old.

Greater sage‐grouse winter habitat use on the eastern edge of their range

“…Because weather conditions fluctuated on a daily basis (i.e., extreme cold and high winds) during our study, conditional logistic regression models accurately reflected variation in sage-grouse habitat use patterns at true presence-absence points (Duchesne et al 2010, Dzialak et al 2011) that could have otherwise been lost with a pooled unpaired logistic regression design. Although we lacked a detailed spatial map of sagebrush vegetation for this region, future research that uses resource selection functions (RSF; Boyce et al 2002, Duchesne et al 2010) to link spatial patterns of habitat and sage-grouse distribution based on conditional logistic regression models (e.g., Compton et al 2002, Boyce et al 2003, McDonald et al 2006) could be valuable to resource managers in the region.…”

Comparation of logistic regression methods and discrete choice model in the selection of habitats

“…McDonald et al (2006) used back-pack harness mounted radio-transmitters, and in this way it was possible to verify that five owls resided in Klamath and twenty-three in Korbel. Forty-six explanatory variables were simultaneously observed (Table 1), resulting in a total of 8,739 observations (Ryan, 2004;McDonald et al, 2006).…”

“…According to McDonald et al (2006), applications of discrete choice models generally assume that animals make a series of choices based on a finite set of discrete habitat units, known as choice sets. Other resource selection analyses include logistic regression that is applied to a sample of used and not used resource units and assumes that choices are made from a set of available resource units.…”

Incorporating within- and between-patch resource selection in identification of critical habitat for brood-rearing greater sage-grouse

“…We matched each used location with a set of five non-used locations drawn from within the circular buffer and considered each used location with its associated five non-used locations as a single stratum. This discrete-choice design (Cooper and Millspaugh 1999;Compton et al 2002;McDonald et al 2006) quantifies a choice made by an individual female sage-grouse (i.e., used location) relative to five alternative choices that also were available temporally and spatially but were not chosen (i.e., non-used locations). One benefit of discrete-choice models is that inference is conditional on individual strata (e.g., a single used location and the paired non-used locations), thus accounting for potential spatiotemporal or within-individual autocorrelation among used locations (Pendergast et al 1996;Johnson et al 2004;Baasch et al 2010;Cushman and Lewis 2010).…”