Scat-based vs. Kill-site Predation Risk
We compared the predictions of total predation risk from all predators (i.e. , sum of standardized predictions for each predator) based on scat locations and contents to predation risk to predictions derived at kill sites analysis at the known kill sites of collared and uncollared elk killed by predators between 2002 and 2016 (Appendix S3, Supporting Information). We derived kill-site models by comparing features at 104 locations of elk kill sites (1) to 20 random points (0) each using conditional logistic regression. Random points were generated within 13.2 km of each kill site, to account for the largest average movement per day of the predators (i.e. , cougar; Dickson et al., 2005; Laundré, 2005). Landscape variables were the percentage or mean value within a 250-m buffer around the kill sites or random location. We used inverse frequency weighting to account for differences in number of kills by each predator species. We evaluated models with AICc (Appendix S4, Supporting Information) and used the best-supported model to map predation risk to elk (Appendix S5, Supporting Information), assuming that the sample of elk mortalities reflected the relative predator-specific kill rates. We conducted Spearman rank correlations between predictions from scat-based and kill-site models at 1,000 random points, but graphically presented smoothed graphs based on the mean risk value of 10 equal-area bins.