Study area and animals
The Peninsular Ranges of southern California extend approximately 150 km north from the United States-Mexico border. Lower elevations of the Peninsular Ranges are in the Colorado subdivision of the Sonoran Desert, and hot dry summers and mild winters characterize the climate [4]. Bighorn sheep in the Peninsular Ranges are a California protected species and have been federally listed as endangered since 1998 [5]. They are typically found below 1,400-m elevations in the eastern portion of the Peninsular Ranges, and are sympatric with mule deer (Odocoileus hemionus) at the upper elevations of bighorn sheep range. Our study from 2002-2004 included radiocollared and uncollared bighorn sheep from 5 of the 8 recognized subpopulations [6], and we use the term "radiocollar" to include either VHF (Telonics, Inc., Tempe, Arizona) or GPS (Televilt Simplex P-1D, Telemetry Solutions, Concord, California) collars on bighorn sheep. We placed GPS radiocollars on 3 pumas that each killed multiple bighorn sheep in the Peninsular Ranges: female F7 and her offspring males M5 and M6. During the entire time that M5 and M6 were GPS radiocollared for this study, they were independent of F7, and each hunted alone to feed only themselves. Puma GPS collars were programmed to acquire locations 4 times within a 24-h period: at night (0000 h), crepuscular periods (0600 and 1900 h), and midday (1200 h).
Bighorn Sheep Population Estimates
The total number of bighorn sheep in each of the 5 subpopulations was estimated based on biennial helicopter surveys conducted by the California Department of Fish and Game. Using radiocollared bighorn sheep in each subpopulation as "marked" animals (each was also ear-tagged), estimates were calculated with capture-recapture methods using Chapman's [7] derivation of the Lincoln-Petersen estimator as described in Rubin et al. [6]. We subtracted the numbers of radiocollared sheep from the total subpopulation estimates to determine the numbers of uncollared bighorn sheep at risk in each subpopulation.
Predation Events
Kill sites of radiocollared bighorn sheep were identified by field investigation of all radiocollars detected in mortality mode. For uncollared bighorn sheep, we identified potential kill sites by visually examining GPS data of radiocollared pumas for locations that suggested a particular puma was returning to or remaining at a kill site. When a dead bighorn sheep was found in the field, puma predation was ruled in or out as the cause of death following the criteria of Hayes et al. [8]. We subsequently developed an algorithm to identify potential kill sites with puma GPS data using a method similar to that of Anderson and Lindzey [9]. GPS data clusters representing potential kill sites were defined as 2 or more consecutive GPS locations at night or crepuscular times (1900, 0000, 0600 h) that occurred within 24 h and within 200 m of each other. The algorithm was then applied retrospectively to the GPS data from the 3 pumas to see if it delineated the 23 known bighorn sheep kill sites that we investigated.
We assigned puma GPS locations to 1 of 3 timeframes to facilitate statistical comparisons of how far pumas were located from kill sites in the periods before, during, and after a kill. The "during kill" timeframe was defined as the mean plus 1 standard deviation of the total time in hours that pumas spent at kills (1st arrival to final departure). The "before kill" and "after kill" timeframes were defined by adding this same length of time to the period before or after the "during kill" timeframe, respectively. We calculated the 2-dimensional Euclidean distances between puma locations and bighorn sheep kill sites using Universal Transverse Mercator (UTM) coordinates for GPS positions.
Statistical Analysis
We tested the null hypothesis that there was no difference in predation by the 3 pumas on radiocollared versus uncollared bighorn sheep using the 2-sided Fisher's Exact Test (JMP Version 8.0, SAS Institute Inc., 2008). Our comparison was stratified by individual puma and geographical area because individual pumas may differ in their predilection for attacking radiocollared versus uncollared sheep, and geographical areas may influence the effect of radiocollars on risk of predation due to differences in cover, forage quality, or behaviour of sheep subpopulations, etc. We deemed results significant when P < 0.05. Assuming equal expected kill frequencies (0.25) at each of the 4 times of day, we compared estimated and expected times of death using an exact multinomial test [10].
We tested the null hypotheses that pumas were not found at different distances from kills across the 3 timeframes, and that no differences existed in distances among the 4 different times of day using mixed-models repeated-measures analysis of variance (ANOVA)[11]. When significant differences were detected, pairwise comparisons (randomized by puma) were made using the Tukey-Kramer HSD (honestly significant difference) test. We compared overall mean distances to kill sites between timeframes, and within timeframes we compared overall mean distances both across and among times of day. Finally, within the during-kill timeframe, distances were evaluated by time of day and sequential day; mean daily distances to kills sites were compared using mixed-models repeated-measures ANOVA with the Tukey-Kramer HSD test. All statistical tests except the exact multinomial test were conducted using JMP software (Version 8.0, SAS Institute Inc., 2008).