Ecological and phylogenetic predictors of mobbing behavior in a tropical dry forest

Mobbing represents a well‐known anti‐predatory behavior, where potential prey display aggressively against a predator. Despite considerable experimental and descriptive work, no models predict species participation in mobbing assemblages. Here, we aimed to understand why some bird species engage in this behavior, while others do not, and what factors can be used to predict mobbing engagement within an avian community. We investigated whether certain functional traits, such as body size, foraging guild, foraging mode, and strata, as well species abundance and evolutionary relatedness, are important mobbing predictors. To address these goals, we simulated the presence of the Ferruginous Pygmy‐Owl (Glaucidium brasilianum) by broadcasting its voice in 230 experiments conducted in 115 points, systematically distributed in a dry forest of northeastern Brazil. We compared these results to 162 avian surveys (point counts) conducted in the same area. Our avian surveys detected 108 bird species (local avian community), whereas our playback experiments attracted 72 species (mobbing assemblage). In general, small, canopy insectivorous or frugivorous birds dominated the mobs. The best mobbing predictors were body mass and guild, whereas species abundance, foraging mode, and strata were not retained in the best models. We found a strong phylogenetic component in body mass and mobbing propensity (almost 90% of the species and individuals participating in the mobs were passerines). At the community level, we found significant differences in the functional and phylogenetic structure of the mobbing assemblage in relation to the avian community. Our results suggest that mobbing behavior is tightly associated with predation risk and the capacity of individual species to find and detect predators, and that functional and phylogenetic features can predict species participation in this complex animal behavior.


| INTRODUC TI ON
In the animal world, the risk of being killed triggers anti-predatory behaviors to avoid potential predators. These behaviors may affect different aspects of animal decision-making, such as time and place for feeding and handling food, sociality, as well as activity and vigilance patterns (Lima & Dill, 1990). Once a potential predator is detected, however, rather than escaping and hiding, some species engage in risky mobbing attacks that can put their own lives in peril.
Mobbing represents a well-known anti-predatory behavior, where members of one or more species display aggressive behaviors aiming to disclose the presence of a predator and eventually chase it away (Hartley, 1950). Although mobbing is widespread in the animal world, studies have barely evaluated which ecological variables drive the expression of this defensive behavior against predators.
Untangling those drivers is a necessary step to understand mobbing function and origin, two aspects that have been elusive to behavioral ecologists and remain poorly understood. One way of moving the field forward is by evaluating which species in a local community engage in this behavior and assess whether particular traits can predict species participation in the mobs. It also remains untested, whether close relatives share this behavior or whether is widespread throughout the community.
Several hypotheses were invoked to explain the origins of mobbing and understand why potential preys may face and fight a predator, putting their lives in peril, rather than hide or scape.
In the Neotropics, the Ferruginous Pygmy-Owl (Glaucidium brasilianum) is one of the most frequently mobbed species (Cunha & Vasconcelos, 2009;Cunha et al., 2013;Motta-Junior, 2007). The vocalization of this small and relatively common owl results in an immediate reaction from mobbing birds (Cunha & Vasconcelos, 2009). The imitation of the vocalization of the Ferruginous Pygmy-Owl in the field often attracts many otherwise silent or hidden birds, and represents a tool often used by ornithologists and birdwatchers alike to disclose the presence of many different bird species. In fact, mobbing on the Ferruginous Pygmy-Owl is possibly one of the most well-studied cases in the Neotropics (Cunha & Vasconcelos, 2009;Reudink, Nocera, & Curry, 2007;Sandoval & Wilson, 2012;Tilgar & Moks, 2015). Nearly, 250 bird species from 27 different avian families display mobbing behaviors against this species throughout its range (Data available in the Dryad Digital Repository, see Data Accessibility for more information). Despite this massive community-wide behavioral response, predicting whether those species responding to predators share similar traits and a common evolutionary history is a new step forward to understand the origin and maintenance of mobbing behavior.
While considerable descriptive and experimental work on mobbing has already been conducted, one particular aspect of this behavior remains largely unexplored: why some species are frequent mobbers, while others never or rarely engage in such behavior? To answer this question, it is necessary to know not only which species engage in mobbing in a given community, but also which ones are present in the area and do not participate in the mobs. This approach will allow us to assess how predictable is the participation of mobbing and test whether particular functional traits can predict species participation in the mobs.
Here, we evaluate how size, trophic guild, foraging strata, foraging mode, abundance, and phylogenetic relatedness are related to mobbing participation. We hypothesize that smaller species will be more likely to participate in the mobs because they are more likely to be predated upon, although under the parental care hypothesis larger birds could have an advantage and have a better chance in protecting their offspring. We also postulate that the kind of food consumed by birds (feeding guild) and the way they look for it (foraging mode) may influence their rate of encounter with a potential predator. Foraging mode determines the birds' general activity patterns and it may define vigilance rates . More active feeding foragers, such as gleaners, or more observant species, such as salliers, may be more common in the mobs than other species. Canopy bird species will more likely encounter Pygmy-Owls and may therefore dominate the mobbing assemblages. Finally, we argue that species abundance is an important variable to take into consideration. More common species may be more prone to join the mobs, simply by being close to a vocalizing owl. By testing these traits, we can build predictions on which part of the community will attend the mobs and create models of mobbing attendance irrespectively of the geographic location of the study area.
In this study, we conducted playback experiments simulating the presence of the Ferruginous Pygmy-Owl to stimulate mobbing behavior in co-occurring birds, in a dry forest of northeastern Brazil. Our main goals were to first characterize the assembly of birds attracted to the voice of the Pygmy-Owl and compare this assemblage to the entire avian community found at the study site. By doing this, we aimed to understand whether some functional traits are better represented in the mobbing assemblages that it would be expected by chance alone. Finally, we evaluated whether evolutionary relatedness predicts the ability of certain birds to participate in the mobbing behavior. We can precisely test this prediction by comparing the phylogenetic patterns between mobbers and co-occurring (non-participant) species. This represents a novel approach to studying mobbing that aims to enlighten our understanding on an important and widespread agonistic behavior in the animal world.

| Study area
This study was conducted at the Fazenda Pau D'Arco, a ~2,125 ha privately owned farm, located within the Chapada do Araripe National with small sub-arboreal and arboreal species, and largely dominated by woody plants (Araújo, Martins, & Shepherd, 1999;Giulietti et al., 2004). The study area is being used under a legal management program that includes areas for logging. The effect of logging on the mobbing experiments will be dealt elsewhere. In the current study, we consider every point sampled as a replicate encompassing the entire variation found in the area (both natural and human-induced).

| Mobbing experiments
Mobbing experiments were conducted in 115 sampling points, systematically distributed every 250 m along 10 different dirt roads with lengths ranging between 4 and 5 km, which cover a total of 1,670 ha ( Figure 1). The distance of 250 m between sampling points was established to ensure the independence of the data (playback broadcast was inaudible beyond 100 m), avoiding individuals to respond to successive playback experiments along the road. At each sampling point, we conducted playback experiments simulating the presence of a Ferruginous Pygmy-Owl for 5 min. We considered as participants in the mobbing, birds that after the beginning of the playback stimulus, began to vocalize and/or approach the vocalization source, exhibiting signs of agitation (raising feathers, tail swings), producing alarm calls to other species, or moving quickly between nearby perches and visually searching for the source of the acoustic signal. We noted the number of individuals, species, and sex and age, whenever possible.
We used a recording of the natural vocalization of the Ferruginous Pygmy-Owl obtained from the Sound Guide of the Birds of Brazil (Vielliard, 1995). The recording was edited using the software Audacity 2.0.4 (Mazzoni & Dannenberg, 2000) to remove background noises (such as background mobbing calls) that could affect the experiments. We broadcasted the vocalization using an amplifier loudspeaker Bluetooth Bright® (model 0374), connected to an iPod player, remotely controlled. The loudspeaker was hidden in the vegetation at ~2 m above the ground. All points were sampled by two observers (HSL and FMGLC), who remained silent at ~3 m from the loudspeaker, a distance that we considered adequate to visualize and listen to the birds responding to the sound stimulus, interfering minimally in their response to our own pilot tests.
All experiments were conducted at the onset of the dry season, between March 24 and April 2, 2015, end of the breeding season. At each sampling point, we conducted a morning (dawn to ~9:30) and F I G U R E 1 Geographic location of the Fazenda Pau D'Arco, at the Chapada do Araripe in the state of Ceará, northeastern Brazil. In detail, a map of the Fazenda and location of the 115 sampling points where we conducted playback experiments (black dots) and 162 sampling point where we conducted the acoustic census (white and black dots). Numbered lines (2 to 11) represent dirt roads that give access to management plots (1 to 22) an afternoon (from 15:30 to dusk) 5-min playback experiment aiming to cover both peaks of avian activity along the day. Road sampling was chosen at random, and within the road, sampling points were sampled consecutively from NW to SE. We avoided sampling the same road more than once in the same day.

| Avian surveys
The community-wide quantitative abundance data from the study area was obtained from an independent sampling of the Fazenda's avifauna carried out a few months before, between 15-25 October and 11-20 December 2014. Sampling was conducted by JRR and FMGLC. Point counts and mobbing experiments were conducted at the same sites, but we had 53 more point counts than experiments, totaling 162 sampling points ( Figure 1). Point counts lasted for 5 min, when all birds seen or heard were recorded. Although community sampling was conducted a few months before the mobbing experiments, it did not involve any major seasonal shift in terms of migratory species or activity patterns, and we certainly did not notice any consistent temporal variation in the Fazenda's avifauna.

| Functional traits
To determine whether some functional traits of the mobbing bird assemblage can predict the response of species to the voice of the Pygmy-Owl, we characterized the avifauna in terms of (a) body mass, (b) trophic guild, (c) foraging strata, and (d) foraging mode. These traits were determined for all species recorded in the study area during our community point counts, including those species that were not attracted to our playback experiments.
Avian body masses were obtained mainly from bird specimens held at the Ornithological Collection of the Universidade Federal de Pernambuco (UFPE), complemented by data from the literature (del Hoyo, Elliott, Sargatal, Christie, & deJuana, 2017) for those species not available at the collection. For each species, we used the mean weigh value of up to 10 specimens selected with no a priori reason from the collection. When fewer specimens were available, we used all available individuals to obtain mean mass values.
Trophic guilds were established based on two sources: an extensive literature search (del Hoyo et al., 2017), complemented by stomach contents available at the collection. Birds were separated in seven guilds, which included: (a) insectivores (species that feed mainly on arthropods); (b) frugivores (species that feed mainly on fruits); (c) insectivore/frugivores (species that are mainly insectivorous, but complement their diet with fruits); (d) granivores (seed eaters); (e) nectarivores (species that feed mainly on nectar); (f) omnivores (species that feed on both plant and animal material, in somewhat similar proportions); and (g) carnivores (species that feed predominantly on other vertebrates).
Foraging strata and foraging mode were classified based on our own field observations at the study area, complemented when necessary from the literature (del Hoyo et al., 2017). Species were divided into four foraging strata, including (a) ground (species that live and feed predominantly on the ground); (b) understory (species that use bushes or the lower parts of trees, up to ~2.5 m); (c) canopy (species that live predominantly on the upper part of trees, reminding that the canopy at our study site is rather low, rarely above 10 m); and (d) open air (species that forage flying above the trees). In terms of foraging mode, species were classified in 11 categories, adapted with modifications from Remsen and Robinson (1990), including (a) scavengers (species that feed on dead animals); (b) bark probers (species that search for food on trees trunks); (c) drillers (species that open holes in trunks for food); (d) flower probers (species that visit flowers); (e) fruit perchers (species that actively peak fruits perching on them); (f) gleaners (species that hunt arthropods along branches); (g) seed predators (species that feed on seeds); (h) ground probers (species that seek for food on the ground); (i) hunters (species that actively seek for moving preys); (j) salliers (species that use a sit and wait strategy to capture their prey on the wing); and (k) aerial (species that feed on the air).

| Mobbing propensity
Based on the frequency of each species at our mobbing sampling points, which was defined as the proportion of playback experiments when a given species was detected. We have established mobbing propensity to each species that participates of mobbing behavior, and this variable was used as a dependent or response variable in the functional and phylogenetic analyses.

| Functional structure
We adapted the method proposed by Duarte, Debastiani, Freitas, and Pillar (2016) to compare whether those species participating in the mobs differ in terms of their functional structure from the avian community at the study site. To do this, we compared the results from Similarly, we used a Principal Coordinate Analysis (PCoA) over Bray-Curtis (Krebs, 1999) dissimilarity matrices to compare species composition among predictor variables (e.g., mobbing and census).
We combined the species occurrences matrix (i.e., abundance of each species on each point of mobbing or census) with the species trait matrix to obtain a matrix where each sampled point is described by the proportion of each functional trait. Thus, we can count the frequency of each trait within guild, foraging mode, and strata. Then, we used an effect size statistic to compare how much each trait varies between mobbing and census. This comparison was made with Hedges' d statistics, that is the difference between the mean of both groups, expressed as the number of standard deviations that they differ (Koricheva et al., 2013). We selected the d statistics because it is not affected by unequal variances (Koricheva et al., 2013).
Although there are other methods for comparing group means, the d statistics performs well (with some bias corrected: see discussion in Koricheva et al., 2013)

| Functional dissimilarities: mobbing vs. censuses
We repeated the PERMANOVA over distance matrices (obtained from the PCFS) on different subsets of trait data (see Pillar, 1999).
First, we run a global model including all traits to calculate Gower's distance. Then, we created four trait subsets by removing each trait (body mass, guild, foraging mode, and foraging strata). We recalculated Gower's distance of each trait subset (for example, trait distance without considering species body mass), re-run the PCFS, and compared the functional structure between mobbing and census species. By doing this, we obtained five R 2 values (from the PERMANOVA model) representing the global model and the four removed traits. Finally, we compared the effect of each trait removal with the global R 2 . If a trait contributes to functional dissimilarities between mobbing and census, the R 2 value will decrease in comparison to the global model. Conversely, the R 2 of the subset with traits too similar between bird communities will increase.

| Avian phylogeny
To verify if the species' propensity to participate in the mobs is influenced by the phylogenetic structure of the community, or in other words, whether phylogenetically closer species are more likely to behave similarly in terms of mobbing participation, we constructed a phylogenetic tree including all bird species recorded in the study area. The tree was constructed from 1,000 phylogenetic trees obtained using the Phylogeny Subset tool from the Global Phylogeny of Birds website (https://birdtree.org), which builds trees combining relaxed clock molecular trees of well-supported avian clades with a fossil calibrated backbone with representatives from each clade (Jetz, Thomas, Joy, Hartmann, & Mooers, 2012).
The 1,000 trees were imported into the software Mesquite 3.21 (Maddison & Maddison, 2018) where a Majority-Rule Consensus Tree was built.
We organized the phylogenetic analysis in species and community levels (see below). We investigated in the species level whether closely related species share similar traits and how those traits influence mobbing propensity. Conversely, at the community level, we evaluated whether clade (or trait) distribution varies across mobbing and non-mobbing species. By doing this, at this scale, we will be able to understand which clades (or traits) are likely to participate in mobbing behavior.

| Phylogenetic signal and drivers of mobbing propensity
We investigated whether there is phylogenetic signal in mobbing propensity and body size using Pagel's lambda (Blomberg, Garland, & Ives, 2003). We also evaluated whether different bird traits (body size, abundance, guild, foraging mode, and foraging strata) affect mobbing propensity with a Phylogenetic Generalized Least Squares We compared this global model with simpler models (removing insignificant predictors) to improve model fit. We removed foraging mode (which were not different from ~1) and recoded levels from categorical variables with less than 10 records in 108 bird species to avoid estimating wrong parameters with few replicates within certain unusual levels.

| Phylogenetic structure at the community scale
We tested whether the phylogenetic structure of the avian assemblage had a predictable effect on mobbing by using a Principal Coordinates of Phylogenetic Structure analysis (PCPS, Duarte et al., 2016). This analysis uses pairwise dissimilarities (based on phylogenetic patristic distances) between sampling points to explore clade distribution across meta-communities weighted by species abundances. As an eigenvector analysis, PCPS produces orthogonal axes that can be further used in multivariate analyses, whereas the first PCPS axis with higher eigenvalues describe basal nodes contributing to phylogenetic structure, lower values can be interpreted as representing the structure of terminal nodes (Duarte et al., 2016).
Differences in phylogenetic and functional structure between sampling units (and their predictor variables, such as mobbing vs. census) were tested with a PERMANOVA. To avoid misinterpreting location and dispersion effects in comparisons of community structure, we used the PERMDISP method that evaluates the homogeneity of multivariate variances (Anderson & Walsh, 2013). The visual inspection with PCoA allows a full comprehension of differences in location and dispersion.
In addition, we controlled the effect of phylogenetic dependence in bird functional structure by using a recently proposed method, which decouples trait dissimilarity (which represents functional variation) from the phylogeny (de Bello, 2017). Then, we obtained a distance matrix (called dcFdist: decoupled trait dissimilarity) that was used in the same way we performed a PCPS with PERMANOVA (see above). Methods such as the PCPS is used for describing the spatial distribution of phylogenetic lineages across sites which, in turn, can be combined with distance-based method to check whether some potential drivers (here, the mobbing behavior) can explain this phylogenetic structure at the community level (Duarte et al., 2016).
We found that species abundance in the study area is not a good predictor of mobbing participation. The most abundant species recorded in the study area were not the most common species displaying mobbing behavior (Figure 2). There is a weak correlation between species abundance and mobbing propensity (r = 0.207, p = 0.03).

| Functional and phylogenetic structure of mobbing at the species and community levels
Body masses at the study area varied from 2.75 g in the Glitteringbellied Emerald (Chlorostilbon lucidus) to 1,520 g in the Black Vulture (Coragyps atratus). The largest species attracted to the mobbing experiments was the Blond-crested Woodpecker (Celeus flavescens), a medium-sized (137.5 g) bird species (Data available in the Dryad Digital Repository, see Data Accessibility for more information). Comparing the assembly of mobbers with the avian community recorded during the censuses, we found that only smaller birds participated in the mobs (Figure 3a). In fact, body size appears as the best predictor for mobbing propensity (F = 6.21, p = 0.014); smaller birds are the most common in the mobs (Figure 3b).
All trophic guilds, except carnivores, were attracted to our mobbing experiments (Data available in the Dryad Digital Repository, see Data Accessibility for more information). More than half (53%) of the species (38) recorded in the experiments were insectivores, followed by insectivore/frugivores (28%, 20 species), and granivores (8%, 6 species), which together accounted for more than 90% of the species. In relation to foraging mode, almost half (46%) of the species (33)  At the community level, we found that the phylogenetic structure of the mobbing assemblage is significantly different to that of the entire avian community (R 2 = 0.536, p = 0.001, Figure 5a). This difference is strong throughout the first axis of the Principal Coordinates of Phylogenetic Structure and allows to predict which portion of the community engages in the mobbing behavior ( Figure 5). Considering the traits analyzed, we found a strong difference in the functional structure between the bird assemblages resulting from the mobbing experiments and that obtained in our community-wide avian surveys (R 2 = 0.508, p = 0.001, Figures 4 and 5b). Both location and dispersion differ between the mobbing assemblage and the community detected during point counts ( Figure 5). Confirming our previous analyses, global models also show guild and body mass as the main traits contributing to functional dissimilarity between the two groups, whereas foraging mode and strata had no significant effect on the functional structure of the assemblages (Figures 4 and 5b).
Importantly, this functional structure is still evident even after controlling the effect of phylogenetic relatedness on functional traits ( Figure 5c).
Despite the major presence of passerines in the mobs, we found a stronger phylogenetic structure that suggests that mobbing behavior is not a passerine-exclusive trait ( Figure 6). However, even within the Passeriformes (the most common order found in our mobbing experiments), there is a strong phylogenetic structure suggesting that the participation in the mobs represents a non-random selection of closely related passerine families (R 2 = 0.619, p = 0.001, Figure 5d). By mapping mobbing propensity in the community phylogeny, we found that even within the same clade, not all species participate in the mobs ( Figure 6).

| D ISCUSS I ON
Our study offers a novel approach to investigate mobbing behavior by untangling which functional traits are more likely to predict which birds participate in the mobs. Furthermore, we demonstrated for the first time that although this behavior is widespread in the community, there is a significant phylogenetic relatedness in this anti-predatory behavior. Hence, displaying mobbing behavior may benefit coexisting individuals from different species, and most importantly, it seems to be transmitted among close relatives.
Three key conclusions emerge from this study: (a) the pervasive-

| Mobbing behavior in bird communities: the case of the Ferruginous Pygmy-Owl
Mobbing on the Ferruginous Pygmy-Owl is a widespread behavior throughout the Neotropics. The perused literature yielded a total of 247 bird species mobbing upon this owl throughout its range (Data available in the Dryad Digital Repository, see Data Accessibility for more information), representing nearly 6% of the entire Neotropical avifauna (Stotz, 1996). We have shown the pervasiveness of the avian response to the vocalization of the Ferruginous Pygmy-Owl in a tropical dry forest of northeastern Brazil, where two-thirds (66%) of the species recorded at our site responded to the podcasted voice of the predator. Despite representing one of the most well-studied cases of mobbing behavior, half of the species recorded by us represent novel records of mobbing behavior against the Ferruginous Pygmy-Owl. In addition, we added an entire order (Trogoniformes) and two families (Trogonidae and Scleruridae), previously unknown to mob on the Pygmy-Owl.
The number of bird species responding to the vocalization of the Ferruginous Pygmy-Owl may increase with further sampling. One of the leading hypotheses behind mobbing participation is the association of mobbing with offspring defense during the reproductive season Ficken & Popp, 1996;Ostreiher, 2003). Therefore, it is possible that additional species participate in the mobs during the breeding season. Taken together, these results F I G U R E 2 Cross-correlation between species abundance and mobbing propensity. Each dot represents a different species, which is connected by a line to the same species; left column represents the individual log abundance observed in the censuses, whereas the right column represents the log of the mobbing propensity. This figure shows that the most abundant species found in the study area were not the most common species displaying mobbing behavior suggest mobbing behavior may be even more widespread in bird communities than previously thought.
Differently from other owl species, the Ferruginous Pygmy-Owl is also active during daytime. Peak periods of Ferruginous Pygmy-Owl prey delivery to their nest concentrate in the early mornings, middays, and evenings (Holt et al., 2017). This activity pattern may explain why we found as many species participating in our mobbing experiments during the early morning and late afternoons. On the other hand, we found a marginally significant effect on the period of the day in terms of the number of individuals detected during the experiments. The general higher activity of birds at dawn is likely responsible for the relatively higher participation in the experiments during the early mornings. At the species level, the two traits that best predict species participation in the mobs are body mass and foraging guild; our explanatory power decreased significantly when these traits are excluded from the analyses. Smaller species tend to engage more in mobbing than larger ones, possibly because they represent those species more likely to be predated upon. In fact, the most frequent species engaging in owl mobbing are those most often eaten by them (Gehlbach & Leverett, 1995). The predominance of smaller individuals (97% of the mobbing species were birds smaller than 100 g) could be explained by the fact that small birds are more commonly attacked by Ferruginous Pygmy-Owls. Body mass is one of the most important functional traits determining prey vulnerability to predation risk . It may define whether a species can be caught, subdued, and consumed by a predator . Potential preys are known to adjust the strength of their mobbing behavior according to their risk of predation (Motta-Junior & Santos-Filho, 2012;Sandoval & Wilson, 2012;Tilgar & Moks, 2015). Therefore, mobbing may vary in response to predation pressure (Dutour, Lena, & Lengagne, 2016;Sandoval & Wilson, 2012). In fact, mobbing intensity by potential preys has been associated with their prevalence in the diet of the Eurasian Pygmy-Owl (Glaucidium passerinum); the more a species was preyed upon, more likely it was to exhibit a mobbing response (Dutour et al., 2016). Passerine birds represented 40% of the diet of the Ferruginous Pygmy-Owl in northern Bahia (Lima & Lima-Neto, 2008) and represented the most common prey item in the diet of the same species in the dry Chaco of Argentina (Carrera, Fernández, Kacoliris, Pagano, & Berkunsky, 2008), and hummingbirds also were documented to represent one of the top main feeding items in southwestern Brazil (Sazima, 2015).

| Phylogenetic and ecological predictors of mobbing behavior
Unfortunately no data are available from the Brazilian Caatinga and future studies on the Ferruginous Pygmy-Owl's diet will enlighten the association between food preference, predation risk, that presence or absence of some species in the mobs could not be correlated with species' taxonomic position, we found that some clades within the passerines are formed both by species that participate in mobbing behavior and by species that do not participate.
Indeed, even within passerines, there are some clades that do not engage in mobbing behavior, which reinforces that phylogenetic relatedness interacts with ecological predictors to affect this behavior.
Based in our results, we argue that there is a predictable difference between the phylogenetic and the functional structure of mobbers compared to the entire avian community. The implication of these results is two-folded; first, it emphasizes that mobbing behavior, although widespread, could be phylogenetically conserved (e.g., Randler, 2012), but that ecological predictors (such as guild) may also help explain this anti-predatory behavior.

| CON CLUS IONS
Our results indicate that the mobbing behavior is widespread in the avian community studied, and by being shared by different avian orders it probably appeared early in the evolutionary history of birds. The maintenance of the mobbing behavior through evolutionary time seems to be transmitted among close relatives, especially within the passerines. However, whether this behavior originated within this lineage and spread to other smaller non-passerine birds, or simply developed within preyed species remains to be clarified.
Our results also provide empirical evidence that guild and body mass are essential traits explaining the propensity of bird species to participate in mobbing behavior. Accordingly, we argue that two key insights can stimulate studies in other systems and organisms: on the one hand, prey species sharing similar food sources (same guild) may facilitate encounter with predators which, in turn, may favor the development of defensive behaviors. On the other hand, small-bodied organisms tend to respond more aggressively to the presence of a natural enemy. Data on the Owl's diet and the social networks (i.e., order of mobbing attendance) related to mobbing behavior are two central topics that will allow us advance on this elaborated animal behavior.

ACK N OWLED G M ENTS
We thank the owners and managers of the

CO N FLI C T O F I NTE R E S T
The authors have no conflict of interests.

AUTH O R S CO NTR I B UTI O N S
HSL, FMGL-C, and LNN designed the study; HSL, FMGL-C, and JRR collected field data; HSL, LNN, and TG-S analyzed the data; HSL wrote the manuscript with guidance from LNN and TG-S; all the authors edited and revised the manuscript.

DATA ACCE SS I B I LIT Y
All data used in this publication, including bird species who display mobbing behavior against the Ferruginous Pygmy-Owl (Glaucidium