Ethics

Ethical approval was obtained by the National Animal Experimentation Ethics Committee (PE/EA/3741-4/2016). The experiment was performed in accordance with the EU Directive 2010/63/EU. Owners provided a written consent form to voluntarily permit their dogs to participate in the study.

Subjects

We tested 25 dogs from multiple-dog households. We could not finish testing one dog because the dog showed distress in the room. Thus we had 24 dogs in the final analysis (11 different breeds and 14 mongrels; 14 females; mean age (year) ± SD 4.9 ± 2.71, see details in the Supplementary Information).

Questionnaire

Owners filled in an online jealousy questionnaire about the subject dogs prior to the test, the invitation of dogs depended on owners’ report (see responses in the Supplementary Information). The invitation to the test depended on the following questions of the jealousy questionnaire: (1) How jealous do you think your dog is compare to the average dog? (scale from 1 to 10), (2) Who does the dog usually gets jealous of? (3) Where does your dog get jealous? (at home, at unfamiliar places; on a scale from 1 to 5). We only invited dogs the owner of which indicated that the dog shows jealous behaviour toward another dog in the household, and/or other dogs in general. Owners that filled in the questionnaire (overall 631 dogs) gave a mean (±SD) 5.68 (±2.67) jealousy score to their dogs; a mean (±SD) 3.19 (±1.43) score at home, and 2.36 (±1.33) score at unfamiliar places. The tested dogs had a mean (±SD) 7.08 (±1.81) jealousy score given by the owner; a mean (±SD) 4.04 (±1.11) score at home, and 2.65 (±1.34) score at unfamiliar places. In case of one subject the owner filled in the questionnaire for the other dog in the household (i.e. we had information about the familiar dog (test partner), not the subject). Based on the data provided by the owner we could decide whether the subject dog was the dominant or subordinate one in the household (as information about one dog provides information about the other; see below), but we do not have the other information (see Supplementary Information).

In addition we asked the owners four questions to decide whether the dog has a dominant or subordinate rank among dogs in the household (based on19,20): (1) When a stranger comes to the house, which dog starts to bark first (or if they start to bark together, which dog barks more or longer)?, (2) Which dog licks more often the other dog’s mouth?, (3) If the dogs get food at the same time and at the same spot, which dog starts to eat first or eats the other dog’s food?, and (4) If the dogs start to fight, which dog wins usually?. We considered the dog dominant if the owner named the subject dog in the answer to the fourth question, or at least twice in the other three questions. We considered the dog as subordinate if the owner indicated another dog from the household in the fourth question, or at least twice in the other three questions. This is a slight change compared to the original criteria used by Pongrácz et al.19,20 who based the decision on the response to the fourth question, or when the owner uniformly indicated the same dog in response to the other three questions. Pongrácz et al.19,20 invited dogs based on owners’ response to these questions, thus they could categorise the dogs as dominant or subordinate prior to the test. Considering that in the present study the effect of dominance rank among dogs on their behaviour was not the main question, we did not base the invitation on this. However, we suggest that this change in criteria does not weaken the argument (although these results should be treated with caution).

In case of 7 dogs the owner’s response did not allow for determining whether the dog was a dominant or subordinate individual. In the case of one dog the social status was decided only on the basis of the first three questions because the owner claimed that the subject dog wins in fights only due to the size difference between the dogs (Sheltie vs. Belgian shepherd). In the analysis we had 11 dogs labelled as dominant and 6 labelled as subordinate (see details in the Supplementary Information).

Test partners

We used four different types of test partners: familiar dog, unfamiliar dog, unfamiliar object and familiar object. The familiar dog was (one of) the other dog(s) from the household (see details in the Supplementary Information). The unfamiliar dog was a middle-sized, neutered female mongrel dog with therapy dog training. Before the test the subject dog and the unfamiliar dog were introduced to each other for about 5 min to see whether any of them shows distress in the presence of the other (in case the owners indicated distress in either the subject or the unfamiliar test partner dog, the introduction was interrupted immediately). We could not test two dogs in the unfamiliar dog condition; however, we tested them in the all other trials. The owner of the unfamiliar dog stood in the room, next to the door she entered during the Unfamiliar dog condition. This allowed her to intervene if needed (e.g. dogs start to fight). She avoided any eye contact with the subject dog and did not talk during the trial.

The unfamiliar object was a remote control car (#32710 RTR Switch Abarth 500, 28 cm × 16 cm × 13 cm) that did not move during the trial. In case of dogs that have already seen the remote control car moving in previous studies (e.g.21,22,23), we used a thermos (25 cm × 14 cm × 14 cm) that had similar colour and size as the car. The familiar object was a newspaper. All subjects encountered all test partners (see above the exception).

Procedure

Dogs were tested in a 5.2 m × 3 m test room at the Department of Ethology, Eötvös Loránd University. All tests were recorded by four cameras attached to the ceiling.

We had five trials separated by ca. 1–2 min breaks. All trials consisted of two phases: familiarization phase (30 s) and test phase (90 s) that followed each other without a pause. In Trial 1 and 5 the test partner was the familiar dog (condition Familiar dog I and Familiar dog II) to be able to examine the consistency of the behaviour of subjects and the effect of time spent in the room (e.g. fatigue). In Trial 2, 3 and 4 the test partners were the unfamiliar dog, unfamiliar object, and familiar object (conditions named accordingly); we counterbalanced the order of these among subjects.

In Trial 1, the owner and the two dogs entered the room. Dogs could explore the room while the Experimenter (E) informed the owner about the procedure. In the other trials E placed the test partner objects in the room after the owner and subject entered. The subject dog and the unfamiliar dog entered the room at the same time. The familiarization phase started when E left the room. During all familiarization phases the owner ignored both the subject and the test partner; he/she measured the 30 s on a stopwatch.

After the time elapsed, the test phase started during which the owner focused his/her attention on the test partner while continued to ignore the subject. In Trial 1, E told the owner to behave in a way that usually elicits jealousy in the subject dog. Owners mostly choose to pet and talk to the test partner. After Trial 1 ended E told the owners to behave in the same way as with the familiar dog in Trial 1 in the following trials, in order to make the conditions as similar as possible (e.g. in case of the familiar object the owner should read aloud only, if he/she was talking to the familiar dog in Trial 1, and had to repeat at least the most often used words that he/she used before).

Behavioural and data analyses

Tests were analysed with Solomon Coder 16.06.26. (by András Péter: http://solomoncoder.com). We excluded two dogs from the Unfamiliar dog condition (see above), and we could not code the behaviour of one dog in Trial 1 (Familiar dog I condition) because the owner’s positioning blocked the view of the cameras to the subject.

We measured subjects’ behaviour only in the test phase. Coded behavioural variables were: looking duration at the owner, test partner or owner-test partner interaction (s), duration of body positioned toward the owner, test partner or owner-test partner interaction (s), duration of touching the owner, test partner or owner-test partner interaction (s), duration of moving toward and in parallel with the owner or test partner (owner-, and test partner-related motion) (s), duration of moving toward the owner-test partner interaction (s), and time spent within 0.5 m of the owner. We also coded how many times the subjects tried to interrupt the owner-test partner interaction (move between them). Inter-coder reliability for all variables were tested on a random subsample of the recordings (20% of the subjects) (IBM SPSS 22, Cronbach’s alpha; see results in parenthesis). For the statistical analysis we kept the looking duration (0.749), duration of body position (0.719), time spent next to the owner (0.884), and attempts of interruption data (0.862). However, we excluded the duration of touch (0.649) and motion (0.592) from the analysis due to the low alpha values.

We used IBM SPSS 22 for statistical analyses. Principal component analysis (PCA) with Varimax rotation, Eigenvalue >1 was used for data reduction. We decided the number of factors (three) after the visual inspection of the Scree test. Factor scores were calculated by SPSS automatically, using Regression method.

We used Box-Cox transformation in PC I (Lambda = 0.5), PC II (Lambda = −0.2) and PC III (Lambda = −1.5) as well. We used linear GLMMs to analyse the effect of dominance rank, trial, condition and order of condition on the principal components; the random variable was the ID number assigned to all dogs (within-subject design). Backwards model selection was based on AIC values; the model with the lowest AIC value was kept, we considered a model better when delta AIC was ≥2. For significant explanatory variables in the final models, we provide contrast estimates (B ± SE) and t values.

Frequency of attempts to interrupt the owner-test partner interaction was not normally distributed (Kolmogorov-Smirnov test, Familiar dog I: D22 = 0.156, p = 0.172; Unfamiliar dog: D22 = 0.249, p = 0.001; Unfamiliar object: D22 = 0.340, p < 0.001; Familiar object: D22 = 0.400, p < 0.001; Familiar dog II: D22 = 0.187, p = 0.044). We used related-samples Friedman test to compare the frequency of interruption of the owner-test partner interaction between conditions. Pairwise comparison by SPSS relied on Dunn’s pairwise post hoc tests followed by Bonferroni correction for multiple testing.

Data availability

Measurement data of subjects are uploaded as Supplementary Information.