Animals’ fear of people is widespread across taxa and can mitigate the risk of human-induced mortality, facilitating coexistence in human-dominated landscapes. However, humans can be unpredictable predators and anthropogenic cues that animals perceive may not be reliable indicators of the risk of being killed. In these cases, animal fear responses may be ineffective and may even exacerbate the risk of anthropogenic mortality. Here, we explore these questions using a 10-year dataset of movement and mortality events for the puma (Puma concolor) population in the fragmented Santa Cruz Mountains of California, for whom the leading cause of death was retaliatory killings by people following livestock loss. We modeled retaliatory killing risk and puma habitat selection relative to residential housing density to evaluate whether puma avoidance of human cues reflected their risk of being killed. We documented a mismatch between human cues, fear responses, and actual risk. Rather than scaling directly with housing density, retaliatory killings occurred at intermediate levels of human development and at night. Pumas avoided these areas during the day but selected for these high-risk areas at night, resulting in a mismatch between cue and risk impacting 17% of the study area. These results are unlikely to be driven by puma hunting behavior: livestock constitute a very small proportion of puma diets, and we found no evidence for the alternative hypothesis that state-dependent foraging drove depredation of livestock and subsequent retaliatory killings. Our findings indicate that puma responses to human cues are not sufficient to enable human-carnivore coexistence in this area and suggest that reducing risk from humans in places with few perceptible human cues would facilitate carnivore conservation in human-dominated landscapes. Furthermore, a mismatch between human cues and responses by carnivores can lead to selection rather than avoidance of risky areas, which could result in an ecological trap.

We captured adult and subadult pumas from 2009-2019 and fit pumas with GPS collars set to record a GPS location every 4 hours. For animals that died during the study, we recorded date, location, and cause of death. We 1) modeled overall and cause-specific mortality rates using the Kaplan-Meier procedure and non-parametric cumulative incidence functions, 2) modeled habitat selection using step selection functions, 3) modeled spatial predictors of the distribution of sites where retaliatory killings occurred, and 4) compared time since last predicted deer kill and body weights between animals killed after depredating livestock and the overall puma population. Included in this data product are the time-to-event data (tte_cause.csv) for analysis 1, covariate information for 4-hour puma locations (ssf_data.csv) for analysis 2, covariate information for locations of retaliatory killings (rk_rsf_data.csv) and other sources of mortality (other_mort_rsf_data.csv) for analysis 3, body weight data (body_weight_data.csv) and kill rate data (times_since_last_kill.csv and kill_rates.csv) for analysis 4.

Please see Nisi AC, Benson JF, Wilmers CC. 2022. Puma responses to unreliable human cues suggest an ecological trap in a fragmented landscape. Oikos: 10.1111/oik.09051 for a full description of data collection methods, and the README file for detailed descriptions of each dataset.