2.1 Sample selection
We performed a search on YouTube (www.youtube.com) on the Google chrome incognito window during November 2020, filtering by relevance using the following key terms: “vultures”, “carnivores” or “predators” combined with “livestock”, “cow”, and “lamb”, both in English and Spanish. The aim of this search was to find videos showing obligate scavenging birds (vultures) or mammalian predators (e.g., wolves, bears, lions) from different parts of the world interacting with live or dead livestock (e.g., eating, flying around, hunting, injuring, walking around, etc.). We evaluated up to 200 videos for each search we performed, excluding videos unrelated to the issue studied, videos without comments, and videos not showing any species of interest. We obtained a subset of 56 videos involving vultures and mammalian predators. For each video we recorded the following data: the title, URL, number of comments, number of views and days of permanence online since publication.
Based on the title, description and content of the videos, we classified each one as positively or negatively framed. Negatively framed videos of vultures and mammalian predators had titles or descriptions that included words such as "predation" or “killing” (Fig. 1a, b). In this category we also placed videos suggesting that vultures eat live animals (e.g., calves, lambs) instead of carrion, or videos edited to emphasize vultures injuring or intimidating livestock when still alive, in some cases without considering the entire story (e.g., not providing information on the previous events, the place, or the health status of the animal affected). Positively framed videos did not use words such as "predation" or “killing” in their titles or descriptions, they referred to vultures eating carrion and not live animals, or were edited to emphasize the role of these birds in the ecosystem (e.g., cleaning carcasses) (Fig. 1c, d).
In addition, we classified videos according to their psychological distance framing, as close or distant (Fig. 1e, f) 16. Close psychological distance videos included contents showing farmers’ testimonies about wildlife interaction with their livestock (e.g., farmers describing vultures or mammalian predators killing their livestock; Fig. 1e). Distant psychological videos showed wildlife interacting with livestock but did not include farmer testimonies (Fig. 1f). We excluded videos without a clear frame (not classifiable in the categories described above).
2.2 Comment codification
One author (FB) codified the comments posted on each selected video, and this was then repeated by the other authors to verify the result or discuss any dissidence. We evaluated up to the first 500 comments in each video, ordering them from the most recent to the oldest. Comments not related to the issue under study (e.g., political opinions, racist comments, etc.) and responses to the main comments were excluded. To standardize the data, we included only one comment from each user.
Comments were grouped into the following categories: 1) empathetic comments, with the following non-exclusive subcategories: a) toward wildlife (vulture or predators), b) toward the livestock attacked or consumed, c) neutral (comments non-classifiable in any previous category); 2) comments consonant with or dissonant to the opinions or images presented in the videos, with the following non-exclusive subcategories: a) consonant comments: comments posted by people who believed that the videos were showing the natural behavior of the species involved, b) dissonant comments: comments posted by people who believed that the video was showing uncommon or exceptional behavior for the species, c) neutral (non-classifiable under any of the previous categories); 3) comments proposing strategies to deal with species involved (vultures-mammalian predators), with the following non-exclusive subcategories: a) lethal strategies (e.g., poisoning or shooting), b) non-lethal strategies (e.g., livestock guardian dogs, surveillance), c) neutral (non-classifiable under any of the previous categories) (Table 1). We downloaded the comments of each video from the YouTube platform, and considered only videos with at least 10 comments.
After all exclusion criteria had been met, we obtained a definitive sample of 25 videos (11 negatively framed on vultures, 7 positively framed on vultures, and 7 negatively framed on predators). These videos received 1179 classifiable comments (494 belonging to negative frames on vultures, 132 comments belonging to positive frames on vultures, and 553 comments belonging to negative frames on predators) (Table S1, S2).
2.3 Statistical analysis
We first computed descriptive statistics to show the percentage of comments made in response to the videos analyzed, classifying them according to (i) comment category (empathetic comment, consonant-dissonant comment and strategy proposed to deal with species involved), (ii) video frame (negative-positive) and (iii) species involved (vultures-carnivorous predators) (Fig. 1; Table S2).
A set of four Generalized Linear Mixed Models (binomial distribution with Logit function) were then performed, with the video ID as a random effect. The first two models were used to evaluate a subset of videos, positively and negatively framed, showing vultures interacting with livestock. They were assessed for: i) the influence of the predictor video framing (negative or positive) on the probability of a video receiving comments proposing a lethal strategy (e.g., shooting); ii) the influence of the predictor video framing (negative or positive) on the probability of videos receiving empathetic comments toward vultures. We then performed two models with a subset of negatively framed videos of vultures and mammalian predators interacting with livestock, to assess the following: i) the influence of predictor species (vulture or mammalian predators) and psychological distance (close-distant) on the probability of videos receiving comments proposing lethal strategies to deal with these species; ii) the influence of predictor species and psychological distance on the probability of videos receiving empathetic comments toward the species involved. For all these models we assigned (1) to comments proposing a lethal strategy toward vultures or mammalian predators and for empathetic comments toward these species, and (0) for comments proposing non-lethal strategies toward vultures or mammalian predators and for empathetic comments toward the species consumed (e.g., sheep or cows).
Finally, a Generalized Linear Model with a Poisson distribution (Log function) was used to evaluate whether the framing of the video, number of days available online, or interactions between these two factors influenced the number of views of videos on vultures (negative and positive frame). All statistical analyses were performed with R core team (2015)25 , and we considered P values <0.05 as significant. The lme4 package 26 was used for the Generalized Linear Mixed Models.