Computational protocol: Human behaviour can trigger large carnivore attacks in developed countries

Similar protocols

Protocol publication

[…] Considering the total dataset on large carnivore attacks since 1955, we first assessed whether the number of attacks varied over time, on a yearly basis, and among species by fitting a Generalized Linear Model (GLM) with the number of attacks against year and species (). We also included the interaction term between year and species to account for the fact that the number of attacks may vary over time heterogeneously across species. Because our data were overdispersed, we fitted the GLM using a Negative Binomial distribution instead of a Poisson distribution. Next, to assess a potential change in the behavioural temperament of large carnivores over time, we tested whether the log-transformed age of the victim and party size (three levels) varied over time and among species by fitting a linear model with a Gaussian distribution and a GLM with a multinomial distribution (three levels), respectively (). Party size was classified into three categories, which allows differentiating between attacks on lone individuals and groups, as well as if the victim in a group was a young person: i) the victim was alone; ii) the victim was a young person (<16 years old) in a group of adults (2 or more people); and iii) the victim was an adult (>16 years old) in a group of adults (2 or more people). We also considered the interaction term in these models to account for the fact that the surrogates of the changes in the temperament of large carnivores used may vary over time differently across species. Finally, we analysed a subset of the dataset considering only those attacks occurring in the US and, together with information on human influx in natural areas, we tested if the number of attacks was related to the number of people involved in outdoor activities by building a GLM with a Gamma distribution, considering year, the number of visitors and their interaction term as factors in the model ().For each analysis, we used an information theoretic framework to rank a set of competing models based on AIC (Akaike’s Information Criterion [AIC]). We used a stepwise selection procedure to create a candidate set of a priori competing models starting from the simplest null model (intercept only model) to the full model (). To select the best candidate model, we used AIC value corrected for small sample sizes (AICc) and Weighted AIC, which indicates the probability that the model selected is the best among the candidates. Models within ΔAIC <2 were considered to have substantial empirical support. All statistical analyses were performed using R 3.0.2 statistical software. GLMs were run with the “lme4” and “nlme” package. […]

Pipeline specifications

Software tools lme4, nlme
Application Mathematical modeling
Organisms Homo sapiens