Shorter headed dogs, visually cooperative breeds, younger and playful dogs form eye contact faster with an unfamiliar human

Authored by nature.com and submitted by mvea

The Animal Welfare Committee of Eötvös Loránd University approved and accepted the experimental protocol (Ref. no.: PE/EA/2019-5/2017) and the tests were performed in accordance with the Hungarian regulations on animal experimentation and the Guidelines for the use of animals in research described by the Association for the Study Animal Behaviour (ASAB) and ARRIVE.

In this study 130 pet dogs were tested, from which five had to be excluded: (1) because of problems with the video (N = 1), (2) visibility of dog’s eyes (due to coat N = 1), (3) problems with eating the food from the ground because of mouth morphology (N = 2) and (4) insufficient food motivation (N = 1). Wallis et al.27 found that the peak of dogs’ performance in eye contact forming with humans is at middle age, and we were interested in aging, not maturation, hence we only tested dogs older than 2.5 years. Thus, 125 dogs (male = 62) were included in the analysis (cephalic index value: 43.5–74.7 (median = 53.2); age: 31.4–174.5 months (median = 106.5 months).

The grouping of the dogs into breed functions was based on Gácsi et al.13 and the dogs’ breed history. The Cooperative breed group (N = 42) contained breeds which have been selected to work in continuous visual contact and interaction with a human partner (e.g. sheepdogs, gundogs), in contrast to the Non-cooperative breed group (N = 27; e.g. hounds, sled dogs, guard dogs, earthdogs). The Mixed breed group (N = 56) consisted of non-purebred dogs with unknown ancestors. Owners provided information about how did they get their dog; 68% of mixed breed dogs in our sample (38/56) were adopted from an animal shelter or found on the street, while only 7% of them (4/56) were adopted from a previous owner, and 16% of them (9/56) were gifted to the actual owner (with no information on whether they were found on the street or rescued from a shelter). The origin of 9% of mixed breeds (5/56) was totally unknown. It is unlikely, that the mixed breed dogs in our sample were first-generation mixtures, but testing the mixed breed dogs' lineages goes beyond the aims of the present study.

Neither the distribution of cephalic index nor the distribution of age differed between the breed groups (see Table 1). Cephalic index value and age did not correlate (R = 0.137; p = 0.129). All subjects’ demographic data are presented in Supplementary Table S1.

Table 1 Cephalic index and age distribution of the sample among the different breed function groups. Full size table

Dogs can be taught to form eye contact with the owner. In several dog schools, one of the first tasks is to teach the dog to make eye contact and thereby increase attention to the owner. We have data on the previous dog school experience (yes/no) from only 113 dogs. 73 dogs attended dog school, while 40 dogs did not. Dog school attendance had no connection to dogs’ performance in our test (hazard ratio = 1.049; p = 0.847).

As mentioned in the introduction, the typical classification of dogs’ head shape based on cut-off values (Fig. 1) has been criticized by Georgevsky et al.15 due the arbitrary nature of the grouping, thus we studied the effect of head shape with actual cephalic index values. We measured each dog’s cephalic index from photographs. The method of measuring cephalic index from photographs was suggested by previous studies11,16,39. The cephalic index value was measured from photographs with the GIMP image editing program 2.2.13. (http://www.gimp.org/). The index was calculated as the ratio of the maximum width of the head multiplied by 100 divided by the head’s maximum length (Fig. 1). Skull width was measured from one zygomatic arch to the other and skull length was measured from the nose to the occipital protuberance. Each picture was taken from the same angle (perpendicular to the top of the skull; see examples in Fig. 1). The distance of the camera (Samsung T710 Galaxy Tab S2) to the top of the dogs’ head was not uniform (as each dog was a different height, and the camera was not fixed), however, this did not affect the measurement, as cephalic index is a ratio. To check the reliability of measuring the cephalic index from photographs, a second coder, naïve to the hypotheses of the study, measured a random sample of subjects (~ 20% of dogs; ICC: 0.91, p < 0.001), and in addition, the heads of ~ 20% of the dogs were also measured with a calliper (ICC: 0.98, p < 0.001). In the case of a Puli, measuring the cephalic index from a photograph was not possible because of its hair, thus it was measured only with a calliper (the dog’s hair was tied up during the test, so its eyes were visible).

All dogs participated in the “Canine Cognitive Battery” (Kubinyi et al., in prep.), which consists of 12 subtests. The prerequisite of participation was to meet the requirements of a sensory examination40. The same experimenter performed all subtests for an individual dog; however, the experimenter identity could differ between dogs. The first test where the dog met with the unfamiliar experimenter was the Greeting test, which was immediately followed by the Human-directed play test. The Eye contact establishment test was the tenth subtest. Thus, before the Eye contact establishment subtest, all dogs had prior experience with the experimenter, who positively interacted with the dog (stroking, playing, speaking, and feeding them). As the participating dogs were enrolled in our longitudinal research project and it was important for the experimenter to be unfamiliar to the dog at the beginning of the Greeting subtest, 8 experimenters took part in testing to comply with this requirement (the experimenter’s ID can be found in Supplementary Table S1). All experimenters were young women (age: 20–27 years). All three tests were carried out during one test session in the same laboratory room (6.27 m * 5.4 m).

In this test, which was based on Wallis et al.27 and Chapagain et al.24, dogs were rewarded for repeatedly forming eye contact with the experimenter. During the test, the experimenter stood in the centre of the laboratory room, while the owner sat on a chair (Fig. 2A). The experimenter held a clicker-like device (which made a “boing” sound, different from the usual clicker sound) in one of her hands, while the other hand was free. During the test, both hands were in a relaxed position by her side. She also had a food pouch at her back on her belt containing pieces of sausage as a food reward.

Figure 2 Test setup. Eye contact establishment test (A), Greeting test (B) and Human-directed play test (C). Full size image

At the beginning of the test, the experimenter called the dog to her, and threw a piece of sausage from her pouch on the ground. Then she remained motionless until the dog formed eye contact with her. She marked the correct behaviour using the clicker-like device before again throwing a piece of sausage. Unlike Wallis et al.27 and Chapagain et al.24, the experimenter did not rustle her pouch when the dog no longer showed interest. The test ended after 15 eye contacts or after 300 s elapsed.

Before this test, the dog had the opportunity to explore the laboratory room for 2 min in the presence of the owner. At the beginning of the Greeting test, the owner stood in the middle of the room and leashed the dog. The experimenter entered the room for the first time, approached the dog-owner pair and said “Hello” to the owner and the dog. She stopped for 1 s in front of the dog, outside the reach of the leash. If the dog approached the experimenter and showed “friendly” or neutral behaviour, she stepped towards the dog, and petted it while continuously speaking to the dog in a friendly tone (Fig. 2B). If the dog showed fearful behaviour, she ignored the dog and talked to the owner for approx. 30 s. After this test the owner unleashed the dog, and the experimenter left the room.

To measure the playfulness of dogs, we tested them in a situation where they could freely interact with the experimenter. At the beginning of the test, the owner and the dog (off-leash) were in the room, and the experimenter entered with toys (ball and rope) in her hand (Fig. 2C). She offered the toys to the dog, and the dog was free to choose between them. Then, they played with the chosen toy until the first minute elapsed. If the dog was not interested in toys, she tried to initiate social play for 1 min.

We recorded the tests with video cameras, which were connected to computers outside of the testing room. The Eye contact establishment test was coded from videos by using Solomon Coder beta 19.08.02 (copyright 2006–2019 by András Péter). The latency to form each eye contact with the experimenter was measured from the moment the dog took the food into its mouth until it formed eye contact with the experimenter again. We defined eye contact as the situation in which the dog oriented towards the front of the experimenter and looked up with both eyes into the experimenter's eyes. Eye contact occurrences were indicated by the experimenter with an auditory marker (clicker-like device), thus eye contact was coded from the videos’ audio spectrograms. The videos were coded in 0.1 s time frames. We analysed the first 15 eye contacts of each dog. If a dog went over the allotted 300 s, but formed less than 15 eye contacts, we gave the maximum latency to each remaining trial (e.g. if the dog formed eye contact 13-times within 300 s, in the 14th and 15th trials the latency to form eye contact was set to 300 s and marked as a censored event). In this way, we did not have to exclude those dogs from the analysis which did not form eye contact 15 times within the allotted time period.

The Greeting test and Human-directed play test were live coded using a binary variable, based on the following definitions: (A) Greeting test: (1) greet immediately in a friendly way (score = 1; N = 73): the dog approached the experimenter immediately when she entered the room and she could pet it; (2) no greeting behaviour (score = 0; N = 52): the dog did not approach the experimenter without calling or she could not pet it; (B) Human-directed play test: (1) high playfulness (score = 1; N = 60): the dog played enthusiastically with the experimenter, it brought back the ball at least once to her or tugged the rope; (2) low playfulness (score = 0; N = 65): the dog did not touch the toys, or it ran after the ball, but did not bring it back to the experimenter, or it took the rope into its mouth a bit, but did not tug it.

A second coder, naïve to the hypotheses of the study, coded a random sample of subjects (~ 20% of dogs). This sample was analysed using intra-class correlations to check the interrater reliability. We found robust reliability for latency to form eye contact (ICC: 0.82–1.00, median: 1.00, p < 0.001), greeting behaviour (ICC: 0.81, p < 0.001) and playfulness with a human (ICC: 0.91, p < 0.001).

We analysed the results using R statistical software (version 3.6.3)41 in Rstudio42. We used survival analysis, as suggested for latency outcomes in behavioural experiments by Jahn-Eimermacher et al.43, as it can handle events which have not occurred within a specified time. We examined each latency per dog (i.e. 15 latencies belong to each dog), thus Mixed Effects Cox Regression Models (“coxme” function of “coxme”44 package) were used to analyse the effect of cephalic index value, breed function, age and sociability (greeting behaviour score and playfulness score) on the latency to form eye contact, with subject ID as a random factor. We also included trial numbers and experimenter identity in the analysis as confounding variables.

Binomial Generalized Linear Models with logit link (“glm” function of “stats”41 package) were used to check the possible relationship between the demographic and morphological factors (cephalic index value, breed function, age) and the sociability binary scores (greeting behaviour and playfulness with a human). We also included experimenter identity in the analysis as a confounding variable. To test the independence of the two subtests of sociability, we used a Pearson's Chi-squared test with Yates' continuity correction (“chisq.test” function of “stats”41 package).

For the Mixed Effects Cox Regression Models, bottom-up model selection was used (“anova” function of “stats”41 package), where the inclusion criteria were a significant likelihood ratio test for each tested variable. The most parsimonious model contained breed function, playfulness with a human and trial number as factors, and cephalic index value and age as covariates (for more details see Supplementary Table S3). A Tukey post-hoc test was used for comparisons between the three breed function groups (“emmeans” function of “emmeans”45 package). For the Binomial Generalized Linear Models, AIC based model selection was applied to find the most parsimonious model using “dredge” function of “MuMIn”46 package. According to the model selection, it contained only age as a covariate for both Sociability subtests, and no other factors (for more details see Supplementary Table S4–S5).

We used the “vif” function of the “car”47 package to check the possible multicollinearity among the independent variables. The variance inflation factor (VIF) measures how much the variance of any one of the variables is inflated due to multicollinearity in the overall model. If the VIF score is over 5, there is a problem with multicollinearity.

We used the “survfit” function of the “survival”48 package and the “ggsurvplot” function of the “survminer”49 package to create survival plots and the “ggplot” function of the “ggplot2”50 package to produce probability plots.

Busterlimes on May 2nd, 2021 at 14:18 UTC »

Weird how domestication did this because looking a wolf in the eye is a sign of agression.

pbrevis on May 2nd, 2021 at 12:56 UTC »

Somewhat related question: is there any condition in dogs equivalent to autism or social anxiety, where a dog would avoid eye contact?

sonofabutch on May 2nd, 2021 at 12:41 UTC »

Another interesting thing about dogs and gazes: dogs will look to where a human is pointing or looking. Wolves and chimps do not.

https://wagwalking.com/behavior/why-do-dogs-understand-pointing

Tests were carried out requiring them to find a treat hidden under a cup. They had to find the treat by following the eye movements and finger pointing action of the person testing them. Dogs responded well to the cues and were happily rewarded with the treat. Chimps did not see the reason for looking where someone was pointing and wolves were not able to follow the commands either. It was interesting to note that even young canine pups were able to follow the pointing finger and figure out there was a treat in store.