Caudal autotomy is a dramatic antipredator adaptation where prey shed their tail in order to escape capture by a predator. The mechanism underlying the effectiveness of caudal autotomy as a pre-capture defense has not been thoroughly investigated. We tested two non-exclusive hypotheses, that caudal autotomy works by providing the predator with a ‘consolation prize’ that makes it break off the hunt to consume the shed tail, and the deflection hypothesis, where the autotomy event directs predator attacks to the autotomized tail enabling prey escape. Our experiment utilized domestic dogs Canis familiaris as model predator engaged to chase a snake-like stimulus with a detachable tail. The tail was manipulated to vary in length (long vs. short) and conspicuousness (green vs. blue), with the prediction that dog attacks on the tail should increase with length under the consolation prize hypothesis, and conspicuous color under the deflection hypothesis. The tail was attacked on 35% of trials, supporting the potential for pre-capture autotomy to offer anti-predator benefits. Dogs were attracted to the tail when it was conspicuously colored, but not when it was longer. This supports the idea that deflection of predator attacks through visual effects is the prime antipredator mechanism underlying the effectiveness of caudal autotomy as opposed to provision of a consolation prize meal.

Procedure

The experiment took place in two urban parks popular with dog walkers (Singleton and Brynmill Park, Swansea, UK) between 9am and 5pm on 8 days between 25th September and 21st October 2019. Dog owners were informed about the experimental objectives, the procedure, their right to withdraw at any point, and then made full written consent for their dog to participate. The procedure started with the experimenter getting the dog interested in the stimulus by playing with the dog, waving the snake and letting them see and smell it, sometimes with help or input from the owner(s). When the dog was engaged with the stimulus, for example by maintaining eye contact with it and attempting to grasp it, the experimental trials began. The dog owner maneuvered the dog to approximately 3m away from the tail detachment point. At this point the experimenter started moving the snake to get the dog’s attention and initiate a chase. When the dog began chasing the snake the experimenter moved the snake away from the dog in a non-linear path slightly slower than the dog’s speed, aiming for the tail to detach when the dog was approximately 1m away from the snake. After the body and tail separated, the experimenter continued to move the body away from the dog at the same speed until either the tail or body was touched by the dog. The experimenter recorded whether the dog continued to chase and attack the body, attacked the tail, or lost interest in the chase. If a dog lost interest in the chase, the experiment was halted. Trials where the dog lost interest were not included in analyses. A trial would also have been excluded if the dog caught the stimulus before the tail detached but this did not occur. If the dog continued to engage in the experiment, the experimenter changed the tail to the next treatment and the next trial began. Treatment order was counterbalanced using a balanced Latin square (ABDC, BCAD, CDBA, DACB; A = long green, B = short green, C = long blue, D = short blue).

Thirty-four dogs participated in this experiment. Each dog participated in a maximum of four trials, one for each experimental condition. 61.8% (n=21) of participants completed all 4 trials, 5.9% (n=2) completed 3 trials, 14.7% (n=5) completed 2 trials, and 17.6% (n=6) completed 1 trial. In total 106 trials were completed.

Analysis

The binary response variable, whether the dog attacked the body or tail, was modelled using a generalized linear mixed model with binomial link function, implemented using the R package lme4 (Bates et al., 2015; R Core Team, 2018). The model included tail length (short vs. long), tail color (green vs. blue) and trial order as categorical fixed effects, and number of trials completed (1-4) as a continuous fixed effect. The interaction between tail color and length was also included. Individual dog identity was incorporated as a categorical random effect. The model was checked using the DHARMa package function testResiduals (Hartig, 2020) which identified no issues with the distribution of residuals, outliers, or dispersion of data.

tail_attack: 1 = subject attacked tail, 0 = subject attacked body, NA = subject did not complete trial.

order: order of presentation of four experimental treatments 1 = ADBC, 2 = BCAD, 3 = CADB, 4 = DBCA; A = long green, B = short green, C = long blue, D = short blue).