Subjects
Pet dogs from two breed groups (herding dogs and terriers) between the ages of 1 and 10 years were included in the study (mean age: 5.4 ± 2.7 years; 29 females, of whom 15 neutered; 27 males, of whom 18 neutered; Table S1). Dogs were included if either purebred or crosses between two breeds from within the same target breed group; crossbreeds between different breed groups were excluded. Both experimenters and all included owners were female, since a study indicated that dogs prefer to gaze at women compared to men (Carballo et al. 2020).
A between-subjects design was used. Twenty-nine dogs, including 15 herding dogs and 14 terriers, were included in Group E (experimenter-responsible group). For these dogs, the experimenter performed the unsolvable task procedure, including rendering one of the rewards inaccessible in a box and placing the other reward accessibly on the floor, while the owners were holding their dogs. The dogs of Group E were a subsample of dogs tested in the framework of a play motivation test by A.M. for a different study. The behaviour test included several sequences of play of the dog with the owner (four subtests), free access to a toy while being ignored, as well as two subtests in which the experimenter played with the dog. The unsolvable task was the last interactive test in the sequence. Except for the treatment (identity of the person responsible for rendering the reward inaccessible) and the prior play sequences, the procedures relevant for the current study were identical for Group O, as described below.
Dogs from Group O (owner-responsible group) were recruited to additionally assess the effect of familiarity and/or the dog–human relationship on dogs’ gazing behaviour at the two people during the unsolvable task. In this group, the owners were responsible for handling the rewards and placing them unattainably into the box, while the experimenter (A.H.) was handling the dogs. Twenty-seven dogs were included, including 14 herding dogs and 13 terriers.
The majority of subjects participated in one or several dog sports, including nosework or agility, which have been shown to be associated with opposite effect on dogs’ gazing behaviour. Eleven dogs from Group O and 12 dogs from Group E practised agility or Hoopers agility. Twelve dogs from group O and 13 dogs from group E were trained in a type of nosework (including search and rescue, (man)trailing, and nosework). Of these, nine and eight dogs, respectively, participated in both agility and nosework. Other types of dog sports included obedience, dog dance, disc-dog, and herding. No dog sports were reported for three dogs in Group O and six dogs in group E. For four dogs in group E, these data were not available (Online Resource 1.1).
Test room
Experiments took place in an experimental room (Fig. 1), measuring 5.22 m × 3.36 m, on the campus of the Vetsuisse Faculty, University of Bern (CH). A wooden partition wall divided the room into two parts, so that the effective testing space was 3.60 m × 3.36 m. The room was furnished with two chairs and several shelves on the wall. One of the chairs was placed in front of the wooden partition wall (facing the entrance door); the other chair was placed at a 90° angle against the wall to the left. The person responsible for handling the rewards (Group E: experimenter; Group O: owner) was always seated on the left; the person handling the dog prior to the beginning of the experiments (Group E: owner; Group O: experimenter) was seated on the right, in front of the wooden partition wall (Fig. 1).
The room further contained a water bowl, the unsolvable task box, and the food puzzle. The unsolvable task box was a modified commercial cat carrier (65 × 37 × 31 cm), screwed onto a heavy wooden board. The handle was removed, and the gaps at the top were covered with masking tape to minimise the risk of injury. During all test trials, the box was locked using the secure looking mechanism at the front of the box, which the dogs were unable to open. The food puzzle was a commercially available food puzzle for cats (‘Trixie Cat Activity Fun Board’), a square plastic plate with several pegs, slots, and dents from which food treats could be extracted (c.f. Riemer et al. 2018). To reduce the size of the puzzle to 30 × 30 cm, so that it would fit into the unsolvable task box, an 8 cm-wide part of the puzzle featuring small removable plastic bowls had been cut off. The unsolvable task box and the food puzzle, placed into predetermined positions for the ‘toy in box’ subtest at the front of the room, 40 cm apart, can be seen in Fig. 1. For the ‘food in box’ subtest, the box was placed in the same position, and the toy was placed 40 cm to the right.
Experiments were filmed from four different angles, using Fixed Dome IP cameras. Testing took place between March 2019 and December 2020.
Procedure
Habituation and training phase
After the owner and the dog had entered the room, the dog was unleashed. A habituation phase of 3 min commenced, during which the dog was free to explore the room and the unsolvable task box, which was initially placed against the left side wall. Next, a brief preference test was performed to ensure that the dog was motivated for the toy to be used in the subsequent experiment. The experimenter retrieved a box full of dog toys of various types from the adjacent storage room and asked the owner to select three toys (one ball, one tug toy, and one plush toy) which she thought her dog would like. After removing the remaining toys from the room, the experimenter placed the three selected toys on the floor, approximately 40 cm apart, while the owner was sitting on her chair, holding the dog by the collar or harness. Once the experimenter had returned to her chair, the dog was released and had 30 s to interact with the toys. The toy the dog spent the most time interacting with was subsequently used for the unsolvable task experiment.
After the preference test, the owner was asked to play with the dog with the chosen toy, as they usually would when playing at home, for 1 min. While dogs from Group O then proceeded directly to food puzzle familiarisation, dogs from Group E were tested in further subtests involving toy play due to participating in a different study. These subtests included tug-of-war with the owner, social play without toys with the owner, tug-of-war with the experimenter, and free access to the toy while the owner and experimenter were either present but passive or briefly left the room, as well as playing fetch with both the owner and the experimenter.
All the procedures from then on were identical for the two groups, except that the owner’s and experimenter’s roles of handling the dog and the rewards, respectively, were reversed between the two groups. Before the start of the unsolvable task experiment, all dogs were familiarised with the food puzzle. The puzzle was filled with ten pieces of semi-dry dog food (which is generally highly palatable to dogs), and the dogs had 3 min to extract the food. If needed, the owners were allowed to help and encourage the dog.
Unsolvable task procedure
Two subtests were performed directly after each other. In the first subtest, ‘toy in box’, the toy was enclosed in the box, while the food puzzle, filled with 5 pieces of food, was freely available (Fig. 1). In the second subtest, ‘food in box’, the food puzzle, filled with 5 pieces of food, was placed in the box; meanwhile, the toy was freely accessible. Thus, the current study is—to our knowledge—the first study to perform an unsolvable task paradigm with dogs when a different type of reward is concurrently available. By consecutively carrying out the unsolvable test with both reward types, we catered both for dogs with higher toy motivation and for dogs with higher food motivation. The order of tests was fixed for all subjects (first the ‘toy in box’ subtest and then the ‘food in box’ subtest). This fixed order was necessary, because the data from Group E were also used to characterise individual differences in toy motivation for a different study.
In line with Rao et al. (2018), the dogs were never exposed to a ‘solvable’ version of the task, thus preventing possible effects of the previously reinforced manipulative behaviours. The box and the respective alternative reward were placed into predetermined positions at the front of the room, 40 cm apart (see Fig. 1). In Group E (experimenter-responsible group), the owner was sitting on the chair opposite the front of the room where the box and the alternative reward were placed. She held the dog by the collar, harness, or lead, while the experimenter placed the rewards on the floor and into the box, respectively, in full sight of the dog. In Group O (owner-responsible group), roles were reversed. The experimenter was handling the dog, while the owner performed the handling of the rewards and enclosed one of the rewards in the box.
After placing one reward on the floor and enclosing the other one in the box, the person responsible for handling the rewards returned to her chair on the left side of the room, and the person handling the dog released the dog. Each subtest (‘toy in box’ subtest, followed by the ‘food in box’ subtest) lasted 3 min, during which the owner and the experimenter were observing, but did not interact with the dog.
Coding
Behaviour coding was performed using Solomon Coder (© 2006–2019 by Andràs Péter) at a time resolution of 0.2 s. Durations were coded by the first author. The frequency of gaze alternations was coded by a second, uninvolved coder.
The following variables were coded as durations (full definitions in Table 1): interacting with the box (low effort and high effort; both subtests), interacting with the food or food puzzle (‘toy in box’ subtest), interacting with the toy (‘food in box’ subtest), gazing at the experimenter and gazing at the owner (both subtests). Gazing was only coded when the dog’s gaze appeared to be directed at the person’s face, but not when it was directed at other body parts (c.f. Smith and Litchfield 2013). Finally, drinking was coded as a duration. We subsequently calculated the effective evaluation period for each dog by subtracting the drinking time from the total duration of the subtest.
Table 1 Definitions of coded duration variables Since the toy was concurrently freely available in the ‘food in box’ subtest, the looks at the owner and the experimenter in this subtest were subsequently classified as box-related or toy-related. If the dog had the toy between the paws or if s/he interacted with or looked at the toy prior to or after gazing at the person, this was classified as a toy-related look. If the dog was interacting with or looking at the box prior to or after gazing at the person, this was classified as a box-related look. No such differentiation was made for the ‘toy in box’ subtest, because although the food puzzle was freely available during this subtest, as soon as the five pieces of food were eaten from the puzzle, the puzzle became another “unsolvable task”, and it was less likely that dogs might want to initiate social interaction with one of the persons in relation to the food puzzle than in relation to the toy.
Low effort and high effort interactions with the box were summed up to yield a total duration of interaction with the box. In nine cases, a subtest was terminated up to 3 s early and lasted between 177 and 179 s instead of 180 s. For one dog, the ‘toy in box’ subtest was terminated after 153.6 s, as the dog managed to obtain the toy. For these dogs and dogs that spent part of the time drinking, all durations were extrapolated to 180 s. Analyses are based on extrapolated values.
Gaze alternations, or ‘referential looks’, were coded following the definition after Miklósi et al. (2000), as proposed by Mendes et al. (2021): a gaze alternation was coded when the dog looked at the box and then at a person, or vice versa, within 2 s. Gaze alternations were coded separately as owner-box, box-owner, experimenter-box, and box-experimenter. For the analysis, we used the sums of both types of gaze alternations for the owner and the experimenter, respectively.
Reliability
Ten dogs each were coded for reliability by a second, uninvolved coder (one coder: durations of interacting with the box and gaze durations; one coder: gaze alternations). Intra-class correlation coefficients (two-way, random, absolute consistency, single measures) were computed in IBM SPSS Statistics Version 23 (IBM Corporation and its licensors 1989, 2015) and indicated excellent reliability (ICC > 0.9) for all durations and very good reliability (ICC > 0.8) for frequencies (see Tables S2 and S3 for full results).
Analysis
Statistics were performed in Statistica 6.1. (Statsoft Inc. 1984–2004) or IBM SPSS Statistics Version 23 (IBM Corporation and its licensors 1989, 2015). R version 4.1.0 (The R Foundation for Statistical Computing, 2021) was used to create boxplots. As the data were non-normally distributed, non-parametric statistics were used.
Assessment of group differences in food or toy motivation
To assess whether differences in motivation or persistence existed between the owner-responsible and the experimenter-responsible group that might influence the results independently of treatment, we performed Mann–Whitney U tests to compare the two groups in time interacting with the box during both subtests, time interacting with the food puzzle during the ‘toy in box’ subtest, and time interacting with the toy during the ‘food in box’ subtest.
There were no differences between the two treatment groups in persistence in either the ‘toy in box’ subtest or the ‘food in box’ subtest. Moreover, there was no effect of group on the duration of interacting with the food puzzle during the ‘toy in box’ subtest, nor with the toy during the ‘food in box’ subtest (Tables S4 and S5).
Effect of face masks on gazing behaviour
From July 2020, the experimenters and the owners wore face masks during the test procedure as a protective measure against COVID-19. As testing of dogs from Group O commenced later than testing of dogs from Group E, all dogs from group O were tested using masks, whereas 17 dogs of Group E were tested without and 12 dogs with masks. To assess possible effects of mask-wearing on dogs’ gazing behaviour, we calculated Mann–Whitney U tests comparing gazing behaviour between the dogs from Group E that were tested with vs without masks. The results showed no significant difference in any gazing variables between dogs tested with masks and those tested without masks (Table S6).
Changes in gazing behaviour over time (‘food in box’ subtest)
Since our subtests were longer than in most other unsolvable task studies, we aimed to assess possible changes in (box-related) human-directed gazing over time. To this end, we calculated box-related gaze duration for each of the 3 min of the ‘food in box’ subtest separately. A Friedman test was performed to detect possible differences between the three minutes, and a Dunn’s test was used for post hoc pairwise comparisons. For the latter, we report adjusted p values, which are calculated according to the Dunn–Bonferroni approach to correct for multiple comparisons.
Changes over time were not assessed for the ‘toy in box’ subtest, as it was very variable between individuals at what point the food was eaten; therefore, using time only as a predictor would not be as informative. Instead, as described under “Correlations between persistence and human-directed gazing” (below), we analysed gazing behaviour during the ‘toy in box’ subtest both for the whole test period and for the time until the food was consumed.
Hypothesis testing
Correlations between persistence and human-directed gazing
Spearman rank correlation tests were used for correlational analyses. In the ‘toy in box’ subtest, we correlated (1) total interaction time with the box and total time gazing at the two people for the entire subtest (180 s), and (2) the same variables but capping total time when the dog had consumed all the food (individually for each dog), which occurred on average after 81.5 s (SD: 62.8 s).
For the ‘food in box’ subtest, we analysed correlations between interaction time with the box and total time of box-related looks and, separately, total time of toy-related looks.
Effect of the responsibility of the owner vs the experimenter
Latency to gaze at the owner vs the experimenter
We used Mann–Whitney U tests to assess differences between Group O and Group E in the latencies to gaze at the owner and the experimenter, respectively. Within groups, we performed Wilcoxon tests to test whether latencies to first gaze at the two people differed from each other.
If a dog showed no human-directed gazing during a subtest, this was classified as NA and was excluded from the analysis. If a dog gazed only at one of the two people present, we used the maximum duration of the test (300 s) as the latency value for the other person. For the ‘food in box’ subtest, where we differentiated box-related gazing and toy-related gazing, latencies were calculated for box-related gazing only.
Proportion of gaze duration directed at the owner
We calculated the proportion of gazing at the owner out of the total time gazing at the two people by dividing the duration of gazing at the owner by the sum of gazing at the owner and the experimenter. Mann–Whitney U tests were used to assess whether the two treatment groups differed in proportion of gazing at the owner (1) for the total duration of the ‘toy in box’ subtest, (2) for the total duration of the ‘food in box’ subtest, (3) for the duration of box-related gazing in the ‘food in box’ subtest, and (4) for the duration of toy-related gazing during the ‘food in box’ subtest.
Frequency of gaze alternations involving the owner vs the experimenter
The effect of treatment group on the frequency of gaze alternations between the box and the owner and the experimenter, respectively, was also analysed with Mann Whitney U tests.
Effect of breed group
Data from Group O and Group E were combined to assess possible effects of the breed group, since the treatment groups were balanced for breed group. The effect of breed group on total time interacting with the box (both subtests), total time looking at the two people (both subtests), and total time interacting with the food or food puzzle (‘toy in box’ subtest) and the toy (‘food in box’ subtest) was analysed using Mann–Whitney U tests.
Correction for multiple testing
Sequential Bonferroni was applied separately for each of the six families (Voelkl 2019) of tests (1—correlations of persistence with gazing; 2—group differences in latency to gaze at the owner vs the experimenter; 3—within-group differences in latency to gaze at the owner vs the experimenter; 4—group differences in the proportion of gazing at the owner; 5—group differences in gaze alternations; 6—differences between the breed groups). With two exceptions, significant results remained significant after correction. In the Results, we only mention the correction when the significance of results changed when using the corrected alpha level. The full results, including the corrected alpha levels, are presented in Table S7.