Objective 1: determining whether predator movements indicated active search for parturient ungulates
We determined whether predators were actively searching for parturient female ungulates by assessing whether their actual movements (hereafter “steps,” or the Euclidean distance between subsequent GPS relocations) led to encounters with parturient females more often than hypothetical steps that they could have taken but did not. To do so, we fit step-selection functions (Fortin et al. 2005) for each carnivore species to estimate the relative probability of selecting a location on the landscape, given its previous location and a suite of covariates at the ending locations of both real and hypothetical (hereafter, “random”) steps. We excluded data from individuals whose home ranges did not overlap deer and elk relocations in Starkey Experimental Forest and Range. The covariate of primary interest was whether the endpoint of each observed or random step was within a 200-meter proximity of a parturient female ungulate. By comparing whether the observed steps were more often within close proximity of a parturient ungulate than were random steps, we could assess whether predators detected neonates more often than expected by incidental encounter, which is evidence of active search behavior. We restricted the analysis to include location data of each female ungulate in the 30 days after a parturition event was predicted to occur and fit separate models for deer and elk for each of the four carnivore species (8 models total). We compared each real step (coded 1) with 20 random steps (coded 0; Latombe et al . 2014) and fit models using conditional logistic regression. The random locations were generated by taking random draws from the fitted distributions of step lengths and turning angles (Gamma and von Mises distributions, respectively; Avgar et al. 2016) constructed from GPS data for each predator species and projecting the random locations for each GPS position onto the landscape given the previous location. To control for the possibility that observed steps landed in close proximity to an ungulate simply because of the predator’s preference for certain landscape or vegetative features, we included additional covariates known to influence carnivore movements (Ruprecht et al. 2021b): canopy cover (derived from LEMMA’s generalized nearest neighbor model; Ohmann et al. 2011), potential vegetation type (a factor variable with classes for open forest, closed forest, grassland, and other), ruggedness (using the vector ruggedness measure, a composite index of terrain encompassing both slope and aspect; Sappington, Longshore & Thompson 2007), the distance to nearest open road (natural log transformed), and distance to nearest perennial water source (natural log transformed). All continuous variables were centered to have a mean of 0 and scaled to have a standard deviation of 1. The step-selection function took the form\(w(x)\ \sim\ exp(\beta_{1}\times\text{parturient\ deer\ or\ elk\ presence}+\ \beta_{2}\times\text{canopy\ cover}+\ \beta_{3}\times p\text{otential\ vegetation\ type}\ +\ \beta_{4}\times\text{ruggedness}\ +\ \beta_{5}\times ln(\text{distance\ to\ road})\ +\ \beta_{6}\times ln(\text{distance\ to\ perennial\ water\ source})\ +\ \beta_{7}\times ln(\text{step\ length})\ +\ \beta_{8}\times cosine(\text{turning\ angle}))\). Because the response to neonates can differ by sex of bears (Raylet al. 2015), we fit additional models for male and female bears separately. We did not fit additional sex-specific models for cougars, coyotes, or bobcats, however, because we had no evidence a priorithat predation of neonates would differ between the sexes for those species.