Subjects
Dogs were recruited through the Laboratory of Applied Ethology database of volunteers. The criteria for recruitment were that dogs were food motivated, in good health and adults (1 to 7 years old). Overall, 29 pet dogs (18 females, 11 males) were recruited, of which 14 did not complete the entire procedure. The 15 dogs which completed the entire test phase were 12 females and 3 males (mean age ± SD of 3.6 ± 1.8 years). Eight dogs were mixed breeds and seven were purebred from various dog breeds.
Experimental setting
Dogs were tested in a quiet room measuring 4.7 × 5.8 m (Fig. 1). The room was equipped with a table behind a 1.3 m tall barrier. Two wooden stands were placed 1 m apart in front of the barrier to hold the panels during the presentation. One meter from the middle point of the stands, a sign was marked on the floor to indicate the position of the dogs’ head during the presentation, and a chair for the owner behind it. Two camcorders (Xacti VPC-WH1, Sanyo, Moriguchi, Japan) were used to record the experiment, one facing the dog and the other facing the panels.
Stimuli
The stimuli consisted of two panels, with different arrangements of food and non-food elements placed upon them. The panels were square shaped (30 cm x 30 cm x 0.5 cm thickness) and made of black polycarbonate. The food elements were discs cut out from turkey ham slices (1 cm diameter x 1 mm thickness). The non-food elements were thin lines made of white plastic masking tape (2.5 cm x 2.0 mm).
A panel containing 30 discs and 30 lines served as the positive stimulus in the training phase. The simultaneously presented negative stimulus was a panel with 10 discs and 10 lines. On all stimuli the dots were arranged semi-randomly with at least 1 cm distance between individual discs. Approximately half of the lines were in contiguity with the discs on both ends, while the rest were not. All elements were homogeneously spread across the panel and the convex hull for both stimuli was approximately 550 cm2. Several panels with different arrangements were used both as positive and negative stimulus.
In the first trials of the training phase (see below) dogs were presented with a negative stimulus where the food elements were replaced with discs made of polymer clay with an identical colour and size to the real food elements. To control for odour cues, a piece of turkey ham with a corresponding size was placed behind this negative stimulus.
Two panels, both containing 30 food elements and 30 lines, served as the test stimuli. Discs on both stimuli had an identical spatial arrangement; however, on one stimulus discs were connected by the lines forming three clusters of ten discs each (Fig. 2a) and on the other stimulus, the lines were not in contact with the discs (Fig. 2b). Each dog was presented with four different arrangements of test stimuli.
Procedure
The experiment was carried out in two different days. The first day started with the dog being brought into the experimental room and let off leash for about 5 minutes to become acquainted with the area. Meanwhile, the owner was instructed to sit on the chair and keep the dog in between her/his legs in a marked location; owners were instructed to gently hold their dog from the shoulders or from the harness and to look at a fixed position in the wall in front of them.
When owner and dog were in position, an experimenter, who stood between the two stands, presented a single central panel with three food pieces on it and told the owner to release the dog, who was free to reach the panel and eat the food. This preliminary presentation was meant to ascertain that dogs were food-motivated and comfortable in approaching the experimenter to retrieve food. Following this preliminary trial, the actual training phase began. Each training trial started with the experimenter standing between the stands while a second experimenter behind the barrier passed two panels into her hands to limit as much as possible unnecessary movements. The presenting experimenter placed the panels simultaneously on the stands, then tapped shortly on both panels to attract the dogs’ attention to them. After 5 seconds the experimenter said “OK” and the dog was released. If the dog approached the positive stimulus, it was allowed to eat the food off the panel. If the dog approached the negative stimulus, the panel was quickly removed before the dog could eat from it. The dog was then taken back to the initial starting position for the next trial. If the dog had approached the negative stimulus, it was presented with a correction trial (i.e. positive and negative stimuli in the same position as the previous trial) until it chose the positive stimulus. The position of the stimuli in each trial was pre-determined according to semi-random sequence of 15 trials. Within the sequence, the positive stimulus was presented 7 times on one side, and 8 times on the opposite, and never presented on the same side for more than two trials in a row.
In the training phase, the negative stimulus used was the one with food imitation until dogs chose the positive stimulus twice in a row. In these trials, if the dog chose the negative stimulus, it was allowed to explore it (look at, sniff and lick) before the panel was eventually removed. When the dog chose the positive stimulus for five consecutive trials, the training phase was considered successfully completed. If the dog did not reach the learning criterion within 15 trials the dog went out for a 10-minute break, before undergoing another training session. The training phase continued until either the dog reached the learning criterion or four training sessions of 15 trials were completed without reaching the learning criterion or the experiment exceeded the time limit of two hours. In the case the dog did not reach the learning criterion in the time limit, it was released from further testing.
Dogs that reached the learning criterion underwent a second training session two to six days after the first day. This second training meant to ensure the dog was still able to perform the task, before being tested for susceptibility to the illusion. If the dog did not successfully complete this second training phase within the same limits used in the first day, testing was interrupted, and the dog’s data were eliminated from the study. Dogs reaching the learning criterion were presented with the test phase on the same day, after a break of 30 minutes.
The test phase consisted of four test trials and eight training trials. The procedure was the same as described above, but correction trials were not allowed and only panels with real food elements were used. In test trials, test stimuli were used and dogs were allowed to eat off from the panels regardless of their choice. The trial sequence was semi-randomised, with the constraints that test trials could never being presented as the first trial, two test trials could never being presented one after another and no more than three training trials in a row were presented. The side of the connected and the unconnected stimulus were randomised and counterbalanced across the whole test phase. In case the dog made more than one mistake in the training trials of the test phase, the dog went back to the training phase. If the dog was unable to reach the learning criterion or made two mistakes in the test phase for the second time, it was released from further testing, and its data were not used for analysis.
Data collection and analyses
The number of trials in the training phase and the dogs’ choices of either stimulus both in the training phase and in the test phase were collected during testing as a binary variable. The same data was later collected from the videos, resulting in 100% agreement.
Behavioural data were collected from videos using the Observer XT software (version 12.5, Noldus, Grœningen, The Netherlands). A continuous sampling method was used to collect data about dogs’ head orientation (looking either at the panel on the right, the panel on the left, the presenting experimenter or elsewhere) during the test trials from the moment the panels became visible to the dog, until the moment in which the dog started moving to make its choice. Dogs’ head orientation data collected by a second independent observer (30% of videos) resulted in a high inter-observer reliability (Pearson’s correlation; looking at the panel on the right: r = 0.97, p < 0.001, looking at the panel on the left: r = 0.97, p < 0.001, looking at the experimenter: r = 0.97, p = 0.004).
A binomial logistic regression model was run to determine if the sex of the dog explained its’ success in completing all the test procedure as hypothesised in previous studies 67,68. The dependent variable in the model was the dogs’ success in completing the experiment (i.e. dog was presented with all four test trails) and the independent variable was the dog’s sex.
To determine dogs´ susceptibility to the illusion, an intercept-only generalized linear mixed model with the subject being a random effect was run to test the null hypothesis (H0) that dogs’ choices in the test phase were not different from chance level.
A generalized linear mixed model was run to determine if dogs’ choices during the test trials was affected by the dog’s sex, the number of training trials to reach the criterion combined for both days, the percentage of time spent looking at the unconnected or the connected panel during the presentation. The dogs’ name was included in the model as a random factor.
All statistical analyses were conducted using R, with the statistical significance level set at 0.05.