Aggressive signaling among competing species of birds

Signal data

We compiled published videos and photographs of aggressive signaling between different species of birds in the context of competition or defense of resources. The majority of the videos and photographs (N = 259) that we used were available from WikiAves (https://www.wikiaves.com.br/), a Brazil-based website for birdwatching and citizen science, where users can publish photos and recordings in a searchable database focused on Brazil and nearby countries (since 2008). Our goal was to find material documenting the interspecific signals used by diverse groups of birds, and thus we prioritized material representing diverse taxonomic families; the avian biodiversity of South America represented in the WikiAves database provided an excellent foundation for this dataset.

We relied on descriptions of the context of the videos or photographs to ensure that we included only media capturing aggressive encounters in our dataset. In WikiAves, we used the “Advanced Search” tool, filtering under “Main Action” to include only “fighting” interactions, and then used only the resulting images involving interactions among multiple species (database searched July 2019). Similarly, we searched the Internet Bird Collection database (HBW.com, searched July 2019, now available in The Cornell Lab of Ornithology Macaulay Library: macaulaylibrary.org) for media with captions including the word “fight*” and included only results depicting interspecific interactions in our study. We note that the Macaulay Library – an excellent source for photographs and video – did not have the advanced search options to identify fighting birds, and thus we were unable to search this data source for our study. We included additional images of interspecific aggressive interactions from other sources (e.g., Martin & Briskie, 2021; YouTube.com: searched July 2019 for specific taxa and the search term “fight”) for certain groups that were underrepresented in our dataset to increase the taxonomic breadth of our study. Sources and credit for each item are listed in our data set.

We further refined our dataset to focus only on signals used in aggressive interactions between heterospecifics. We did not include photographs or video segments that depicted direct fights (i.e., physical contact between competitors) or chases, and instead focused on displays that did not involve contact between focal individuals. We note that signals used during direct fights and chases were thus omitted from our study, and could be different from those used in aggressive signaling without contact. In addition, we did not include birds that were retreating from a resource or interaction in our dataset because we wished to characterize aggressive signals; retreating from a resource or interaction meant that the focal bird was no longer aggressively challenging other species. Retreating birds may signal submission to aggressive heterospecifics (see also Gill, 1971; Sætre, Král & Bičík, 1993); however, our study was not designed to identify such signals. We considered a bird as retreating from a resource or interaction when it appeared to be actively moving away from the resource or aggressively signaling heterospecific, respectively (e.g., beginning to fly or move away). For photographs, we used only images that appeared to capture a full display or signal, although we could not always be sure that the image captured the point of peak intensity.

Our dataset is comprised of photographs and videos (N = 337 interactions) of 164 different bird species from 120 genera, 50 families, and 24 orders, following the IOC World Bird List taxonomy (Gill, Donsker & Rasmussen, 2020). It includes a diversity of aggressive interactions, with birds signaling from a perch (53%), the ground (16%), the air (22%), and the water (9%). While interactions from South America comprise the majority of our dataset (N = 259), we also include interactions from Africa (N = 17), Europe (N = 15), North America (N = 14), Oceania (N = 12), Asia (N = 9), and Antarctica (N = 7). For each aggressive encounter, we documented details of the signaler’s posture, and the body regions and colors that the signaler highlighted.

Body regions highlighted by signaling birds

We identified body regions that focal individuals featured most prominently (i.e., highlighted) in aggressive displays, grouping regions by location on the bird and their likelihood of being collectively visible. Some regions that are typically examined separately in studies of coloration (e.g., lores, forehead, auriculars, chin, throat) tended to be visible together, and thus we grouped them for our analyses (e.g., face/throat). We illustrate the regions included in our study in Fig. 2. Distinguishing between these highlighted regions is somewhat subjective; thus, each photograph or video was examined separately by three naïve human observers, who each recorded up to three regions that they perceived as most highlighted by the signaler from the positional perspective of the heterospecific receiver (Appendix II). For our analyses, we considered a body region to be highlighted in an interaction if it was indicated as highlighted by at least two of the three observers.

Statistical tests

We conducted all of our statistical analyses and plotting in R (R Core Team, 2020). We provide the R code that we used for our analyses and figures, along with our dataset, as Supporting Information.

We used Bayesian phylogenetic mixed-effects models using the MCMCglmm function in the MCMCglmm R package (version 2.32; Hadfield, 2010) to test whether birds were more likely to use specific postures, or highlight specific body regions or colors in aggressive encounters with heterospecifics, as well as to compare whether birds use the same signals in interspecific aggressive interactions as in intraspecific aggressive encounters. Bayesian phylogenetic mixed-effects models allowed us to incorporate the effects of phylogeny in our analyses. We obtained a phylogeny for the signaling species in our dataset from Jetz et al. (2012) (birdtree.org; maximum clade credibility tree of 1,000 Hackett all-species trees; Fig. S1) and included the phylogeny, species, and focal.interaction (i.e., a numeric identifier for the documented interaction) as random effects in each model. We did not specify priors for fixed effects so that all variance parameters were estimated (Hadfield, 2018), but specified priors where V = 1 and nu = 1 for both R and G structures (Hadfield, 2010, 2018). We used a categorical (i.e., binomial) distribution and ran simulations for 2,000,000 iterations (burnin = 10,000 iterations, thinning interval = 100) for all analyses, except for two posture analyses examining tail position and the part of the signaler’s body positioned closest to the receiver, both of which had position categories that were rarely assumed and were thus run for more iterations (5,000,000 iterations) to achieve sufficient effective sample sizes. For each analysis, we tested whether the model with the focal predictor performed better than a null model using DIC values (better performing models were identified as those with lower DIC values; Spiegelhalter et al., 2002), thus indicating that our focal birds differentially used postures, or differentially highlighted body regions or colors, in aggressive encounters with heterospecifics. We provide details about model diagnostics in the Supporting Information.

We used a separate model for each component of posture (i.e., body orientation, head position, etc.) to test whether different positions of each component (e.g., forward.lowered, forward.upright, etc.) were equally likely to be assumed in interspecific aggressive interactions. We used a binary response variable indicating whether each position was assumed (1 = position assumed; 0 = position not assumed) as the response variable, and position as a categorical predictor variable. To describe the body regions most commonly highlighted in aggressive signals, we calculated the proportion of interactions in our dataset which highlighted each body region (as indicated by at least 2/3 naïve observers). To test whether certain body regions were most likely to be highlighted, we used a binary response variable indicating whether the focal body region was highlighted (1 = highlighted; 0 = not highlighted) as the response variable and body region as a categorical predictor variable. We excluded three body regions from our analysis (uppertail coverts, undertail coverts, and tarsal feathers) because they were never identified among the most highlighted body regions in our dataset.

Similarly, to describe the color or color groups most commonly highlighted in aggressive signals, we calculated the proportion of interactions in our dataset in which the signaler highlighted each color or color group (as indicated by at least 2/3 naïve observers). To test whether certain colors or color groups were most likely to be highlighted, we used a binary response variable indicating whether the signaler highlighted the focal color (1 = highlighted; 0 = not highlighted) as the response variable and color as a categorical predictor variable. Colors or color groups that were not present in the signaling species in each focal interaction were excluded from this analysis.

To understand whether birds use the same signals in aggressive encounters with heterospecifics as in aggressive encounters with members of their own species, we calculated the proportion of interspecific interactions in our dataset in which the signal used had each degree of similarity to intraspecific signals (same, similar, similar to congener, or different). We used a binary response variable indicating whether each degree of similarity was assumed (1 = yes; 0 = no) as the response variable and degree of similarity as a categorical predictor variable.

To ensure that source format (photo vs. video) did not have a large influence on our results, we repeated all MCMCglmm models with the subset of the dataset captured in photographs (N = 289 interactions; see Appendices VI, VIII, and IX in Supporting Information).

We used binomial tests using the binom.test R function to examine the consistency of signals used in aggressive encounters within a species. For species for which we had sufficient data (>6 videos or images) we tested whether the majority of birds in a species (>50%) used certain postures or highlighted certain body regions during aggressive encounters with heterospecifics.

Finally, to examine how the signals used in interspecific aggressive interactions vary among families, we calculated the proportions of each family that used each posture, and highlighted each body region and color. We included only families that were sufficiently represented (>6 videos or images) in our dataset in this analysis (N = 14 families).

Postures assumed, and body regions and colors highlighted during aggressive encounters with heterospecifics

The model that included the focal predictor performed better than the null model in all of our analyses (Tables S3, S14), indicating that our focal birds differentially used postures, and differentially highlighted body regions and colors in aggressive encounters with heterospecifics.

Each body region position was not equally likely to be used in aggressive signals by birds in our dataset (Fig. 3; Table S3). During aggressive interactions with other species, birds in our dataset more commonly assumed a forward-facing, lowered body position (42% of aggressive interactions in our dataset; Fig. 3A; Table S4), with a forward-facing, lowered head (52%; Fig. 3B; Table S5), open wings (57%; Fig. 3C; Table S6), open, forward-facing bills (68%; Fig. 3E; Table S8), and trailing, unfanned tails (39%; Fig. 3F; Table S9). Birds were more likely to hold their bill closest to the heterospecific receiver than other body regions (87%; Fig. 3H; Table S11). Birds in our dataset were more likely to plant their feet on a substrate during aggressive signals than to assume other foot positions (82%; Fig. 3G; Table S10), and they were equally likely to assume a shoulder position with their underwing forward (38%; Fig. 3D), as a shoulder position with their wing closed with their shoulder visible (34%; Fig. 3D), but other shoulder positions were less common (Fig. 3D; Table S7).

Birds in our study were more likely to highlight their face and throat area than other body regions during aggressive signals directed towards heterospecifics (77% of aggressive interactions in our dataset; Fig. 4A; Table S15). Birds also more commonly highlighted their mouth (37%; Fig. 4A; Table S15), underwings (27%; Fig. 4A; Table S15), and bill (27%; Fig. 4A; Table S15) than other body regions, including the breast, nape/back, legs/feet, crown, undertail, upperwings, belly, uppertail, shoulders, and sides. No birds in our dataset highlighted their uppertail coverts, undertail coverts, or tarsal feathers in aggressive interactions with members of other species (Fig. 4A). Birds were more likely to highlight carotenoid colors, including red, pink, orange, and yellow, than other colors in the aggressive signals captured in our study (55% of aggressive interactions in our dataset; Fig. 4B; Table S16).

Model results for postures, highlighted body regions, and highlighted colors remained similar when run using only the subset of the data obtained from photographs (Appendices VI, VIII), indicating that the source format had little influence on our results.

Similarity to signals used in aggressive encounters with conspecifics

We scored the similarity of 307 of the interactions in our dataset to signals used during aggressive encounters with conspecifics. The model that included the degree of similarity as a predictor performed better than the null model (Table S20). The majority of birds in our dataset used the same signal (body position and highlighted body regions) during aggressive encounters with heterospecifics that they use in aggressive encounters with conspecifics (87% of aggressive interactions in our dataset; Table S21). Model results remained similar when run using only the subset of the data obtained from photographs (Appendix X), indicating that the source format had little influence on our results.

Within-species similarity in signals used in aggressive encounters with heterospecifics

Four species were sufficiently represented in our study to be examined for consistency in the signals that they use during aggressive encounters with heterospecifics. Each of these species had several posture categories that remained consistent across all images or videos. Columbina talpacoti (Columbidae) always had their wings raised upward and their feet planted on the substrate (N = 7, binomial test: P = 0.008). Eupetomena macroura (Trochilidae) always had a forward, upright body position, a fanned tail pointed down, and held their bill closest to the heterospecific competitor (N = 7, binomial test: P = 0.008). Similarly, the bill of Pitangus sulphuratus (Tyrannidae) was always one of the body parts closest to the other bird (N = 13, binomial test: P < 0.001). Thraupis sayaca (Thraupidae) always had an open, forward-facing bill position, planted feet, and again, held their bill closest to the heterospecific competitor (N = 24, binomial test: P < 0.001).

All four species highlighted at least one body region in the majority of signals in our dataset. All Columbina talpacoti individuals highlighted their underwings (N = 7, binomial test: P = 0.008). All Eupetomena macroura highlighted their face/throat (N = 7, binomial test: P = 0.008). The majority of Pitangus sulphuratus individuals in our dataset highlighted their face/throat region (77%; N = 13, binomial test: P = 0.046). Similarly, Thraupis sayaca individuals in our dataset tended to highlight their face/throat (96%; N = 24, binomial test: P < 0.001) and mouth (75%; N = 24, binomial test: P = 0.01).

Within- and among-family variation in signals used in aggressive encounters with heterospecifics

Many birds in our study showed some aspects of their posture that were fairly stereotyped across interspecific aggressive signals by members of their family, and some postures were fairly unique to a specific family (Fig. 5). Members of Falconidae always assumed an underwing.forward shoulder position and typically extended their feet (92%; Table S24). Anatidae and Spheniscidae always positioned their bill forward and open, and had a trailing, unfanned tail (Table S24). Fringillidae and Turdidae also always held their bill open and forward (Table S24). Columbidae was the only family in which most birds in our dataset did not have a forward-oriented body orientation (75% side oriented; Table S24) or head position (58% side oriented; Table S24). Columbidae, Falconidae, and Trochilidae always had open wing positions (i.e., flapping, spread.outward, partially.spread, raised.upward, soaring.gliding; Fig. 5; Table S24). While most families in our study positioned their bill closest to their competitor, the bill was not involved in the majority of interspecific aggressive signals by members of Accipitridae (50% feet; Table S24), Columbidae (50% wing; Table S24) or Falconidae (86% feet; Table S24).

Most families in our dataset showed intra-family variation in which body regions they were most likely to highlight in interspecific aggressive signals (Table S25). Members of Anatidae, however, always highlighted the mouth, while members of Ardeidae, Columbidae, Falconidae, and Icteridae never did so (Fig. 6; Table S25). Falconidae was the only family in which most members were most likely to highlight their legs and feet (Figs. 5, 6; Table S25).

Many colors or color groups were highlighted by many different families included in our study, but we observed some differences among families in the signals commonly used in aggressive interactions with heterospecifics (Fig. 7; Table S26). In our dataset members of Diomedeidae always highlighted white and members of Spheniscidae always highlighted contrasting dark and white in aggressive signals. Despite carotenoid colors being commonly highlighted by most families in our dataset, Columbidae never highlighted red, pink, orange, or yellow, and instead were most likely to highlight gray in aggressive interspecific interactions (Fig. 7; Table S26). We note that no families that were widely represented in our study most commonly highlighted rufous or chestnut, while all families had some representatives that highlighted black (Fig. 7; Table S26).