Determining why some animals form groups while others remain solitary is a longstanding goal in behavioural ecology. Group formation can help mitigate predation risk through a variety of mechanisms, including risk dilution and group vigilance. The ‘selfish herd hypothesis’ proposes that prey can reduce their risk by minimizing the area around which all points in that area are closer to them than to another conspecific (i.e., by minimising their ‘domain of danger’). This hypothesis assumes that an individual’s predation risk is proportional to the size of its domain of danger, however, the relationship between risk and proximity to conspecifics may depend on additional factors. Specifically, approaching conspecifics may be costly for prey that rely on crypsis because group formation increases detectability. Using model prey, we experimentally manipulated prey colouration as well as the domain of danger, then tracked their ‘survival’ under natural field conditions. We found that an individual’s predation risk increased with their domain of danger for conspicuous (red) prey, but decreased with the domain of danger in cryptic (green) prey. Our results are consistent with patterns in natural systems and indicate that the relationship between predation risk and domain of danger depends on additional factors like prey colouration.

Study Site: Our experiment took place in Peterborough, Ontario, Canada, in a mixed deciduous forest. Potential predators of our models included several species of insectivorous birds observed during our experiment (e.g., Poecile atricapillus, Sitta carolinesis, Picoides pubescens; full list in Supplementary Material). Caterpillars similar to our models which are active when we conducted our study include members of Saturnidae (e.g., Actias luna, Antheraea polyphemus, Callosamia promethea), Sphingidae (e.g., Sphinx chersis, Eumorpha pandorus), Noctuidae (e.g., Acronicta clarescens, Noctua pronuba) among others. Air temperature during trials one and two and averaged 17^˚C and 14^˚C, respectively.

Field Methods: Model caterpillars made from plasticine (40 mm × 10 mm) were created by hand and deployed along tree branches to test our hypotheses. Our experiment followed a 4 × 2 factorial design. Specifically, each tree received a single group of four caterpillars, following one of four prey spacing treatments (2, 4, 8 or 16 cm between individuals) and one of two prey colour treatments (Green or Red). Differences in prey spacing allowed us to establish prey with a variety of DODs. The colour treatments were selected to examine the effect of prey conspicuousness (Green = cryptic, Red = conspicuous), as has been done previously (e.g., 23-25). Our characterization of green plasticine models as cryptic and red plasticine models as conspicuous from the viewpoint of insectivorous birds is supported by previous work [e.g., 22,26]. Further, human vision detects most of the relevant variation in colour within the visible range and is a valid proxy for avian colour perception [27,28].

While the DOD can be quantified in several ways, we arranged prey along tree branches so that the DOD could be manipulated and calculated along a single dimension. Specifically, we calculated the DOD as the linear distance within which the target prey is closer than any of the other prey items, up to a maximum of 2 prey body lengths on either side of the prey (i.e., a limited DOD). Each prey, therefore, had one of seven different DODs (6, 8, 12, 13, 14, 16 and 20 cm).

Prey were deployed as two 10-day trials (September 27 - October 7, 2022, and October 14 - October 24, 2022), each with 5 replicates of each treatment (i.e., n = 320 prey total). Between the two trials, we left a 10-day rest period to avoid possible acclimation effects in the predator community. Treatment was assigned randomly to each tree, and trees with groups of prey were separated by no less than 1 m. The tree species we used were Fagus grandifolia, Carpinus caroliniana, and Ulumus sp. Models were deployed at least 15 m away from hiking trails on branches that were at least 64 cm long and between 1-2 m from the ground. Models were checked daily between 12:00h and 14:00h and considered attacked when birds left beak imprints on the model. Attacked models were photographed and immediately replaced with another model of the same treatment to maintain a constant DOD for all prey throughout the experiment. 

Data Analysis: We used Cox proportional hazards regression to analyze patterns of survival. This approach uses a continuous measure of survival as the response variable and has been widely used to analyze survival of model prey deployed in the field [e.g., 23,29-30]. Separate models were created to evaluate ‘survival’ at the group level (i.e., time until a group member is attacked) and ‘survival’ of individual prey. Each analysis was stratified by trial to account for possible differences in baseline hazard. We included survival data from prey that were added to replace attacked prey. Prey not attacked on the final day of each trial were censored at that time. The group-level analysis tests whether prey colour, spacing, or their interaction, influenced the probability that a group was attacked (Hypothesis 1). The individual-level analysis tests whether the prey colour, DOD, or their interaction influenced the survival of individual prey (Hypothesis 2). Spacing and DOD were treated as continuous variables. All analyses were conducted in R version 4.0.1 [31], using the survival package [32]. The assumption that hazard functions are proportional over time was tested using the cox.zph function.

Code is provided in R. Data files are in *.csv format which can be opened with Excel, Notebook, and other equivalent programs.