Personality composition affects group cohesion of homing pigeons in response to novelty and predation threat

Understanding how and why animal groups behave collectively is a central question in behavioural and social sciences. Variation in the phenotypic composition of the individuals within a group can lead to differences in group attributes and performance. However, whether and how individual personalities translate into group performance is not yet fully understood because experiments that test such hypotheses in realistic set-ups are still scarce. We explored how between-group variation in personality composition affected flock cohesion during homing flights of homing pigeons, Columba livia. Based on consistent individual differences, we established flocks of either ‘more reactive’ (MR flocks) or ‘less reactive’ (LR flocks) pigeons naïve to homing. Cohesion of flocks was tested in three distinct challenges: (1) first-ever collective homing experience (novelty); (2) release from a novel site (novel site homing); and (3) hunt by a robotic peregrine falcon (predation threat), with the latter two challenges performed with flocks trained for homing. MR flocks were more cohesive than LR flocks in the novelty challenge, but showed similar levels of cohesion during the novel site homing challenge. Predation threat decreased cohesion in both flock types, with a stronger effect in LR flocks. These results indicate that differences in the composition of personalities of group members can produce detectable differences in collective performance, and highlight the importance of accounting for individual-level behavioural variation when studying collective patterns in nature.

Methods
Personality

Each bird was subjected to three behavioural tasks to define its personality traits: Flight Initiation Distance (FID), Escape Reaction Time (ERT) and Exploration Propensity (EXP). To assess the consistency of individual responses over time, we performed two replicas of each test. The second replica took place 1 week after the completion of the first one. The sequence of tests was randomized for each individual in each replica. The repeatability of the response of the birds to the first and second replicas of each test was assessed using mixed effect models through package rptR 0.9.22 (Stoffel et al., 2017) with 1000 bootstraps. The repeatability was adjusted taking into account the level of experience of tested birds, which could affect the response.

The personality identified for each pigeon was used to assort monotypic flocks constituted of four individuals (only more reactive or less reactive). To assess the bird personality, we firstly ranked individual responses to each test at their first replica based on their reactivity, assigning the lowest rank to the least reactive individual. For each test, the ranks were then transformed into a three-level score based on the tertiles of the rank distribution. The first, the second and the third tertiles were scored as 0, 1 and 2, respectively. The scores of the three tests were then summed to obtain an Overall Personality Score (OPS). Subjects with OPS ≥ 4 were classified as ‘more reactive’ (MR pigeons), provided that the FID score was greater than 0 (i.e. took off at greater distances from the experimenter). Subjects with OPS ≤ 2 were classified as ‘less reactive’ (LR pigeons), provided that the FID score was ≤ 1 (i.e., individuals seemed less intimidated by humans and took off only when the experimenter was close to them). Where these criteria were not fulfilled the birds were excluded from homing trials.

Homing flights

All birds of the flock were equipped with GPS data loggers to record their homing route. Monotypic flocks were exposed to three functional challenges: novelty (first release of the newly established monotypic flock), novel site homing (established flock released from a new unknown release site) and predation threat (established flock released from familiar release site and attacked by a robotic falcon - RobotFalcon). Each flock was released 7 times from the R1 site and one time from a non familiar release site (R2 or R3). At the 7th release from R1 pigeons were attacked by the RobotFalcon. The tracks have been preprocessed in order to remove locations collected between the time the pigeon had reached the home loft after completing the homing flight (the first time crosses a distance < 100 m from the home loft) and the time the GPS was retrieved from the bird.

Assess degree of flock cohesion and identify first flock split event

“Flock cohesion was assessed only for flocks with complete GPS recording for all four birds until at least one pigeon entered a 100 m buffer around the home loft. For each flock in each homing trial, the number of GPS locations used for assessing cohesion was the number of data points of the track of the bird that homed first. A cohesion index was computed for each homing trial as follows. The distance of each pigeon from all the others in the group was computed for each sampling step. If the distance between two individuals was equal to or higher than 500 m, we set a cutoff value of 500 m. These distances were summed, and this sum was divided by the sum of the theoretical maximum distances if the pigeons were flying alone (3000 m; i.e. all six pairwise distances were at the cutoff value of 500 m). The resulting value was then subtracted from 1. Therefore, the cohesion index varies from 0, when pigeons fly separately, to 1, when pigeons fly in group. To examine potential differences in flock splitting between flock types, we measured the time elapsed since a flock's release (TR) and the distance to home (DH) when the first split occurred (if any). A split was defined as at least one pigeon leaving the flock, with a separation threshold of 20 m (Biro et al. 2006).” (extracted from paper, Cerritelli et al. 2025).

“We considered two types of splits: (1) to stop and rest before completing the journey (‘split-to-rest’); (2) to follow a different homing path from the rest of the flock (‘split-to-fly’). In the novelty challenge we performed the analysis considering both types of split behaviour in order to detect potential biases due to stress or physical fatigue potentially affecting flight performances. In the other two challenges, we considered only ‘split-to-fly’ individuals, given that an individual showed ‘split-to-rest’ behaviour in only two cases (one for the novel site challenge and one for the predation threat challenge).” (extracted from paper, Cerritelli et al. 2025)

Statistical Analyses

“For analysing cohesion responses to the novelty challenge and the novel site homing challenge in relation to flock type, we fitted a generalized linear model (GLM) with beta error distribution, with the cohesion index as the response variable and flock type (two-level factor: MR and LR) as predictor. The effect of flock type on the time since release (TR, log-transformed) and distance to home (DH) at the location of the first split was tested using linear models (LM). For the novelty challenge, the analyses were performed both including and excluding those groups containing ‘split-to-rest’ individuals. Flock cohesion of MR and LR flocks during the predation threat challenge was assessed using a generalized linear mixed model (GLMM) with beta error distribution. The cohesion index of each flock during the homing flight after exposure to the RobotFalcon (RobotFalcon flight) was compared to the cohesion index of the last available training release without predator exposure (control flight). Predictors included predator treatment (control versus RobotFalcon flight), flock type (MR versus LR) and their interaction as fixed effects, and flock identity as random intercept. Similar GLMMs were fitted to test the effect of the RobotFalcon on first flock split after release (TR and DH), using a negative binomial and a Gaussian distribution, respectively, and flock identity as random intercept.” (extracted from paper, Cerritelli et al. 2025)

References

Biro, D., Sumpter, D. J. T., Meade, J., & Guilford, T. (2006). From compromise to leadership in pigeon homing. Current Biology, 16(21), 2123e2128. https:// doi.org/10.1016/j.cub.2006.08.087

Stoffel, M. A., Nakagawa, S., & Schielzeth, H. (2017). rptR: repeatability estimation and variance decomposition by generalized linear mixed-effects models. Methods in Ecology and Evolution, 8(11), 1639e1644. https://doi.org/10.1111/ 2041-210X.12797