Like deer caught in the headlights: Human evaluation of the intensity of emotions in brachycephalic and normocephalic dogs – A pilot study

Like deer caught in the headlights: Human evaluation of the intensity of emotions in brachycephalic and normocephalic dogs – A pilot study Petra Eretov´ a a,* , Quanxiao Liu a , Tereza Nekov´ aˇ rov´ a b,c , Helena Chaloupkov´ a a , Lucie Pˇ ribylov´ a a , ˇ S´ arka Krtiˇ ckov´ a a , P´ eter Pongr´ acz d a Department of Ethology and Companion Animal Science, Faculty of Agrobiology, Food and Natural Resources, Czech University of Life Sciences Prague, Kamýck´ a 129, Prague 165 00, Czech Republic b Department of Zoology, Faculty of Science, Charles University, Viniˇ cn´ a 7, Prague 128 00, Czech Republic c National Institute of Mental Health, Topolov´ a 748, Klecany 250 67, Czech Republic d Department of Ethology, ELTE E¨ otv¨ os Lor´ and University, P´ azm´ any P´ eter s´ et´ any 1/C, Budapest 1117, Hungary ARTICLE INFO Keywords: Dog Brachycephaly Emotional cues Baby schema Dog-human relationship ABSTRACT While the acoustic communication of dogs across a wide variety of contexts has been studied intensively, visual communication is, to date, less understood. Current research has focused on human assessment of canine signals or factors affecting the ability of humans to understand them. In this continuation of our previous pioneering study (Eretov´ a et al., 2024), we have examined the human-perceived intensity of basic emotions (Happiness, Anger, Fear, Sadness, Curiosity) in two dog breeds (brachycephalic - Boston Terrier, normocephalic – Jack Russell Terrier) via short, soundless videos showing canine behaviours in four simulated situations. Participants of the study submitted their assessment via online survey, using a 7-point Likert scale and also provided information about their demographics, attitude towards dogs, and dog ownership history. Statistical analysis was conducted in R 4.2.2. using one full cumulative link mixed model per each inspected emotion. Distributions of positive and negative emotions differed across situations in both breeds, suggesting good recognition of the presented materials of contexts even without any auditory stimuli present. Results showed that owners of normocephalic dogs viewed Boston Terriers as having lower intensities of positive emotions (Happiness, Curiosity; P <0.05) than brachycephalic dog owners, while also assessing Jack Russell Terriers with higher scores of positive and neutral emotions (P <0.05). Older participants also tended to assign lesser intensity to positive emotions than their younger counterparts (P <0.001). Despite predicting that the Boston Terriers would be rated with an overall higher intensity of positive emotions, this was not the case. This pilot study points out that context and the valence of the emotion are critical in its perceived intensity, and that the experience and attitude of assessors are important factors. More research which employs the scalability of emotions needs to be conducted on a wider variety of canine phenotypes to properly understand human-perceived dog emotions and the factors affecting them. 1. Introduction Compared to its closest wild-living relative, the grey wolf (Canis lupus), domestic dogs (Canis familiaris) (Krofel et al., 2022) show a remarkable variability in skull size and length, body size, and limb and tail length or shape. The shortening and broadening of the skull have been proposed as one of the paedomorphic changes that emerged in early dog domestication (S´ anchez-Villagra et al., 2016). These are among the various phenotypic traits have been intensified by humans during the artificial directional selection for morphological extremes in various dog breeds (Wayne, 1986). The skull length and shape can be evaluated by the cephalic index (CI), which is calculated as skull width divided by skull length, thus a higher CI number indicates shorter heads. Several small to mid-sized dog breeds with an extremely high CI (e.g. French Bulldogs) have become extremely popular with dog fanciers. According to the “baby schema effect” (Lorenz, 1943), brachycephalic characteristics such as large eyes, a pronounced forehead, a small nose, or a disproportionately large, round head elicit nurturing * Corresponding author. E-mail address: [email protected] (P. Eretov´ a). Contents lists available at ScienceDirect Applied Animal Behaviour Science journal homepage: www.elsevier.com/locate/applanim https://doi.org/10.1016/j.applanim.2025.106767 Received 22 April 2025; Received in revised form 23 July 2025; Accepted 24 July 2025 Applied Animal Behaviour Science 292 (2025) 106767 Available online 24 July 2025 0168-1591/© 2025 Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies. tendencies in observing humans, as they mimic the look of a human child. Observing photographs of infant faces causes physiological and emotional reactions in humans, particularly in women (Esposito et al., 2015; Hildebrandt and Fitzgerald, 1978). Similar results in the perception of cuteness have been achieved when evaluating juvenile animal faces, particularly those of immature dogs and cats (Borgi et al., 2014; Lehmann et al., 2013). The faces of animals whose features have been altered to look more infantile (such as enlarging the eyes and softening the features) captivate human viewers much more than the original faces (Borgi et al., 2014; Hecht and Horowitz, 2015). It was found that breedor age-related brachycephaly in dogs (such as the short and round heads of puppies) increased a perception of cuteness and/or adorability in humans. In adult dogs, brachycephaly did not necessarily elicit comments like ‘cute’ from observers (Alley, 1983; Paul et al., 2023a). To understand the factors that further enhance the popularity of brachycephalic dogs, some researchers have suggested that these dogs may express more dependent behaviour on humans (Ujfalussy et al., 2023). Dependency has been measured by the likelihood of them looking at humans during an ‘unsolvable task’. This suggests that brachycephalic dogs can communicate differently with humans than normocephalic breeds do, and one could hypothesise that if these breeds were able to express themselves in a more appealing or effective way, this could be to their advantage in the world of humans. Acoustic and visual modalities in canine communication are those most affected by breed-specific changes in canine morphology. It is known that canine vocalisation is affected by body size (B´ alint et al., 2013; Farag´ o et al., 2010a; Morton, 1977) and the inner state of the signaller (Farag´ o et al., 2017, 2010b; Silva et al., 2021). Human listeners are fairly accurate at classifying the contextual and affective content of dog barks or growls (B´ alint et al., 2013; Eretov´ a et al., 2020; Farag´ o et al., 2017; Pongr´ acz et al., 2011, 2005; Silva et al., 2021). On one hand, while canine acoustic signalling has attracted a lot of scientific attention, the interspecific visual communication of dogs had been, until quite recently, surprisingly underrepresented in ethological research, despite sight being the dominant sensory input for humans (Spence et al., 2012). There is evidence of positive selection for expressive facial signalling in dogs is the dog-specific development of the inner eyebrow elevator muscle (responsible for the ‘sad puppy’ face) – the corresponding muscle is almost entirely missing in wolves (Kaminski et al., 2019). Recent studies (Bloom et al., 2021; Bloom and Friedman, 2013) have showed that adult viewers are capable of identifying emotions from photographs of normocephalic canine faces. A canine version of FACS (Facial Action Coding System), a coding system designed to describe facial expression in humans (Ekman and Friesen, 1978), called DogFACS (Bremhorst et al., 2021; Waller et al., 2013), has become available for researchers. Additionally, artificial intelligence has been recently used for detecting short-term micro expressions in cats (Feighelstein et al., 2023, 2022) and dogs (Boneh-Shitrit et al., 2022). However, these tools are often still limited in their success when evaluating signals of dog faces with extreme morphological modifications, such as those of brachycephalic breeds. It has been revealed that the altered facial anatomy of these dogs (Schatz et al., 2021) causes certain facial expressions, particularly those of aversive inner states, to be barely, if at all recognisable (Goodwin et al., 1997). It has also been revealed that contextual visual cues of brachycephalic dogs are not very well understood by human receivers, particularly in negative (such as fear and aggression) and stimulating situations (such as playful behaviour) (Martvel et al., 2025). It was also found that human viewers are, on some level, aware of the decreased signalling ability (Goodwin et al., 1997; Martvel et al., 2025; Schatz et al., 2021) of brachycephalic dog faces. Consequently, they tend to ignore these dogs’ facial expressions when asked about theinner state of the dogs (Eretov´ a et al., 2024). Increasing age has also been noted affect human viewers to report exaggeratedly high or low levels of stress in dogs (Mariti et al., 2012) or to have a slight negative effect at recognising positive contexts of human-oriented dog behaviours (Eretov´ a et al., 2024). Last, but not least, dog ownership has been proven to have little effect on human ability to correctly interpret canine signals (Eretov´ a et al., 2024; Lakestani et al., 2014), including signals of brachycephalic dogs by people who have owned such dogs (Eretov´ a et al., 2024). However, people who expressed affinity towards brachycephalic breed tend to view them in a more positive light than people with a neutral or negative approach towards canine brachycephaly (Bogn´ ar and Kubinyi, 2023; Eretov´ a et al., 2024). Human viewers can recognise most contextual cues when evaluating photos and soundless videos of canine behaviour, with signals of normocephalic dogs being usually better recognised than brachycephalic dogs (Eretov´ a et al., 2024). However, any difference at identifying visually-only signalled emotions by people in the two groups of dogs with varying skull length has not yet been examined. While it has been shown that humans varied in success rate of identifying signals of dogs from visual or mixed visual-auditory stimuli (Eretov´ a et al., 2024; Meints et al., 2018; Tami and Gallagher, 2009), the perception of intensity of individual emotions is not well explored in the case of visual expressions. Most research focusing on identifying emotions in dogs by people operates in absolutes (i.e. whether one particular emotion is present or not in a sample stimulus), while, in fact, emotions often vary in their intensity. While there are multiple ways to define emotions (Mulligan and Scherer, 2012), here we understand them as defined by Paul and Mendl (2018): “an emotion is a multicomponent (subjective, physiological, behavioural and cognitive) response to a stimulus or event that is typically of importance to the individual, it is always valenced (pleasant or unpleasant) and can vary in activation/arousal and duration/persistence.” Anderson and Adolphs (2014) proposed four basic emotional descriptors: valence (positive or negative), scalability (intensity expressed by a scale), persistence (lingering presence) and generalisation (ability to respond based on previous experience). While these descriptors are key aspects of human emotions, they are also manifest in various forms across a wide range of animal species, which provides a robust basis for comparative studies (i.e. Murphy et al., 2014 for a review on measuring emotions in pigs, Boissy et al., 2007 for measuring emotions in farm animals; Lundblad, 2018 for inspecting facial expressions as stress indicators in horses; or Paul et al., 2005 for measuring emotional valence in animals). Emotional states and their alteration may be demonstrated by behavioural and physiological reactions, such as changes in locomotion, facial expressions, heart and respiratory rates, hormone and neurotransmitter levels, body temperature, and inner eye corner temperature (Neethirajan et al., 2021). In this paper we demonstrate a novel approach towards measuring the human perception of visually expressed emotions in dogs by employing five basic affective states (emotions) that were scored individually by their perceived intensity. We used this method of emotional scaling for comparing the efficacy of visual signalling in two dog breeds that differed in their cephalic index: a normocephalic and a brachycephalic breed. 2. Objectives and hypotheses The presented study is part of a larger project focused on the visual cues of brachycephalic dogs. In the first part (Eretov´ a et al., 2024), which shares the same background, material collection method and experiment of the present study, we examined the human ability to recognise contextual signals of brachycephalic and normocephalic dogs and specific body and facial areas of human interest. Now, we have investigated directly the human-perceived intensities of emotions attributed to brachycephalic and normocephalic dogs, regardless of their factual correctness or valence. In this study, five categories of emotions were used to assess humanperceived emotional intensity in brachycephalic and normocephalic dogs. Four of these represent basic emotions that are considered universal, while the fifth — ‘Curiosity‘— was included as a neutral affective P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 2 state often associated with exploratory behaviour and active interest in the environment, due to its wide prevalence in human-oriented canine behaviour. These affective states were selected based on the assumption that they can be reliably elicited and demonstrated in controlled experimental situations (for inducing emotions in dogs under experimental settings, see Bremhorst et al., 2022). We focused on investigating (1) how human viewers perceived the intensities of the five emotions from short, soundless videos capturing the behaviours of two dog breeds in four simulated situations. The two breeds (Boston Terrier – brachycephalic breed, Jack Russell Terrier – normocephalic breed) were chosen based on their overall similarities (size, body proportions, temperament), with the main phenotypical difference between them being their skull lengths (and resulting cephalic indexes). Additionally, we examined factors on the side of human observers regarding their ability to assess dog emotions. These factors included (2) their age and (3) their factual experience with dogs in general and brachycephalic dogs in particular, as well as (4) their subjective self-assessment regarding experience with brachycephalic dogs. We hypothesised that (i) the intensities of positive and neutral emotions (Happiness, Curiosity) would be overall significantly higher in brachycephalic dogs than in normocephalic dogs, while the intensities of negative emotions (Anger, Fear, Sandess) would not significantly differ between the two breeds. We hypothesised that (ii) owners of brachycephalic dogs and participants who did not own any dogs would rate these dogs with higher intensities of positive and neutral emotions than they would rate the normocephalic dogs, while owners of normocephalic dogs would evaluate them with lower intensities of positive or neutral emotions. We hypothesised that (iii) people who expressed their willingness for owning a brachycephalic dog in the future would rate these dogs with higher intensity of positive emotions and lower intensity of negative emotions than people who would not consider getting a brachycephalic dog. Last, but not least, we hypothesised that (iv) participants’ age would be in a negative association with the intensity of their emotional ratings. 3. Methods The materials used in the study, along with the experiment, is shared with our previous study (Eretov´ a et al., 2024). The previous study analysed the human ability to understand contextual cues of dogs and areas of the dogs’ bodies and faces that were most informative to human viewers. The presented study examined the human perception of emotional cues of the dogs. 3.1. Dogs featured of the study Altogether, the behaviours of 23 dogs were recorded. We recorded 16 Boston Terriers as representatives of brachycephalic dogs (10 females, 6 males; age range 1.25–6 years, mean age 3.17 years, σ =1.64) and 7 Jack Russell Terriers for normocephalic dogs (2 males, 5 females; age range 2–4 years, mean age 2.75 years, σ =0.96). The owners of the dogs volunteered for the study on the basis of public calls via Facebook. The research was conducted in accordance with the guidelines for the use of live animals in research, as outlined by the Association for the Study of Animal Behaviour (ASAB). The experiment was fully noninvasive to the canine and human participants. The methodology was reviewed by the Animal Welfare Committee of E¨ otv¨ os Lor´ and University (Certificate number PEI/001/1056–4/2015). Written informed consent was obtained from every dog owner participating in the study. As the videotaping of the dogs did not involve human experimentation or sensitive data collection from the dog owners, it was not necessary to obtain human ethical approval. The videos featured in the survey were edited so that participating dog owners were not identifiable. The dog owners were informed that they could withdraw from the experiment at any point. Material collection was carried out at the ethological laboratory (a camera-equipped tiled room with a false wall, dimensions 4.45 ×3.68 m) of the Department of Ethology at E¨ otv¨ os Lor´ and University, Budapest, Hungary, with material processing and survey creation being carried out mostly at Department of Ethology and Companion Animal Science, Czech University of Life Sciences Prague, Czechia. 3.2. Participants in the online survey In total, 350 participants fully completed and submitted the questionnaire. Informed consent was obtained from every participant at the very beginning of the survey. Proceeding with the survey was impossible without completing the informed consent. The informed consent form was approved by the Ethical Committee of the Czech University of Life Sciences in Prague, no. 07/2023. The survey included questions about the participants’ demographics. Participants were asked to disclose their age (in full years) and gender (male/female/do not wish to disclose). Another question aimed to investigate the number of children the participants have raised at the time of the survey – answering this question was voluntary. Participants could leave their e-mail addresses or other contact information to receive information about the study or the validity of their answers but were not demanded to do so. All personal information provided by the participants was anonymised and handled in accordance with the General Data Protection Regulation (Regulation (EU) 2016/679) (GDPR). The survey was distributed in three language versions (Czech – 246 participants, Hungarian – 93 participants, English – 14 participants). Participants had to be 18 years old or older, regardless of their dog ownership status or level of dog-related experience. The participants ranged between 18 and 73 years old (mean age 33.33 years, σ =11.74 years; median 30 years). Of all participants, 90 % reported having owned dogs in the past or present, and 25.5 % reported having owned at least one brachycephalic dog in their lives. 3.3. Experiment design 3.3.1. Obtaining the stimuli In order to demonstrate to human viewers the body cues and facial expression of the featured dogs, we video-recorded the dogs in four simulated situations (‘Called by name’, ‘Play’, ‘Separation from owner’, ‘Threatened by a stranger’). All dogs were recorded in individual sessions by the same two experimenters (two women in their 20 s). The experiment used two cameras (Sony FDR-AX33). One was mounted on a tripod and recorded continuously, operated by Experimenter 1 during the first two situations (‘Called by name’, ‘Play’), while running freely in the last two (‘Separation’, ‘Stranger’). The other camera was handoperated by Experimenter 2 during the first two situations and by Experimenter 1 during the last two. The recording session was preceded by a 5-minute period of free lab exploration and habituation for each dog. During this period, the owner was informed about the experiment and her/his role in it. No dogs present for filming failed to habituate within the 5-minute period. Each dog was recorded only once and individually, even in cases when one owner arrived with multiple dogs. In such cases, the dogs were habituated together and only separated for the individual testing. During the habituation phase, Experimenter 1 was free to engage with the dogs while Experimenter 2 avoided contact to maintain unfamiliarity. It was explained to the owners not to engage with the dogs physically or verbally during the experiment. After the 5-minute free exploration phase, the experiment began. It consisted of the following eight steps, each lasting 1 min. The order of the situations was fixed. Just before the first situation occurred, the dog was leashed to a handle secured to the floor near the owner. The leash allowed the dog to sit or stand in front of the owner but prevented physical contact with the tripod camera. If the step design permitted, the owner sat on a fold-out chair at the back of the room. The step-by-step sequence of the experiment and additional information are presented in Table 1. The interludes served to allow the dogs P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 3 to peacefully transition between emotions before the next situation occurred. We have not encountered a case in which a dog would be unable to participate in the experiment. 3.3.2. Materials The obtained video recordings showing dogs’ behaviours in the four situations were categorised by the contexts (situations), and all videos were muted. The videos were then included in an online survey distributed to the general public. For this study, assessment of emotional intensity was collected for each combination of breed/situation. The video sequences obtained through the experiment were further edited using Avidemux 2.7 (coded by ‘Mean’, ‘Gruntster’, and ‘Fahr’, 2019), a free video-editing software. Soundless video clips lasting 5–10 s were created from the source materials. Only materials showing the dogs from a frontal or semi-frontal angle where both eyes and the entire face were visible were included in pre-selection. The main criterion for the videos was sufficiently present behaviours typical for dogs in each featured situation. Out of the resulting pool of pre-selected videos, 16 videos of behaviours of Boston Terriers and 16 videos of behaviours of Jack Russell Terriers (four videos per each situation in each breed) were included in the final survey. The videos were selected based on a discussion with animal behaviour and communication experts affiliated with either of the participating institutions. The experts participating in the video selection process were handed the entire pool of videos and asked to select those they perceived as the most accurate for each situation and breed combination. Using this method, the final batch of 4 videos per breed and situation was obtained. These were then assigned at random to one of the four versions of the survey to achieve necessary randomisation. Each version of the survey contained one video for each combination of breed/situation. 3.3.3. Questionnaire The study was conducted using an online survey provided by Google Docs. This platform was chosen for its easy availability and uncomplicated programming, which allowed us to create multiple versions of the survey within one document. Three language copies (Czech, Hungarian, and English) identical in their contents and order of material presentation were created (for the English version, see Supplementary Material I). Data collection was open from June 2020 to November 2021. Each participant could only fill out the survey once. The survey included the demographic parameters of the participants as described above. Furthermore, we enquired into every participant’s dog-ownership history (‘realised ownership’). When a participant reported having owned at least one dog in their life, they proceeded to specify what kind of dog they owned. We asked them whether they had ever owned at least one brachycephalic dog (such as a Boston Terrier, French Bulldog, English Bulldog, Pug or Shih-Tzu – ‘brachycephalic’) or had ever owned dolichocephalic or mesocephalic (‘normocephalic’) dogs (such as a Jack Russell Terrier). Moreover, participants were asked to disclose their attitudes towards dog owning. We asked them whether they would ever consider owning a brachycephalic dog, or only ever a normocephalic dog, or whether they would prefer never owning a dog at all (‘hypothetical ownership’). Participants had to choose one of three possible answers – they would never consider owning any dogs (‘never’), they would consider owning only dolichocephalic or mesocephalic dogs (‘normocephalic’), or that they would consider owning brachycephalic dogs (‘brachycephalic’). All participants answered questions regarding both types of ownerships (realised and hypothetical) and both values were inspected separately. Finally, participants reported their experience with brachycephalic dogs on a 7-point Likert scale (‘subjective experience’). On this scale, 1 stands for the most inexperienced while 7 for most experienced. Following this, participants were asked to select one of the four random versions of the survey and could proceed. In each version of the survey, participants were presented a set of 8 videos (4 per each breed and the 4 presented situations). The order of situations was randomised in all survey versions. All four situations for both breeds were represented in each version of the survey by one video, totalling eight videos seen by every study participant. Participants also determined the contexts (situations) of the presented materials as well as evaluating the emotional intensity (see Eretov´ a et al., 2024), thus their emotional intensity assessments could have been somewhat influenced Table 1 Step-by-step sequence overview of the experimental situations and roles of featured personnel. Phase Dog Experimenter 1 Experimenter 2 Owner Duration Expected behaviours Pre-phase Free exploration Free interaction with the dog No interaction with the dog Filling out background information form 5 min  Situation 1 – Dog called by name (‘Called’) Being called by name by Experimenter 1, leashed Calling the dog by name while operating tripod camera Operating hand-held camera Sitting on the chair behind the dog 1 min Eye contact with the experimenter, head tilt, ear twitch Interlude 1 – Free play Playing freely with a tennis ball, unleashed Playing with the dog Filming on the hand-held camera Sitting on the chair behind the dog. 1 min  Situation 2 – Play (‘Play’) Playing and reacting to a tennis ball, leashed Playing with the dog while operating tripod camera Operating hand-held camera Sitting on the chair behind the dog 1 min Eye contact at the tennis ball, jumping, barking, attempted chasing of the ball Interlude 2 – Owner leaving Remained leashed in place Operating the handheld camera, no interaction with the dog Left the room Left the room <1 min  Situation 3 – Separation from the owner (‘Separation’) Remained leashed in place Operating the handheld camera, no interaction with the dog Absent from the room Absent from the room 1 min Eye contact in the direction of the owner, pacing, barking or whining Interlude 3 – Owner returning Remained leashed in place Operating the handheld camera, no interaction with the dog Entered the room following Situation 3, went to stand behind the dog Entered the room after the owner and closed the door <1 min  Situation 4 – Threatened by a stranger (based on Vas et al., 2005) (‘Stranger’) Remained leashed in place Operating the handheld camera, no interaction with the dog Approached the dog with a hunched-over posture and wide, sliding steps while maintaining eye contact with the dog (staring into its eyes), blinking as little as possible Standing behind the dog while not initiating contact with it 1 min Eye contact with the experimenter followed by looking away, lip licking, head down, looking back at the owner, backing away P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 4 by which situation they matched to each material. For analysis, we chose the participant-reported intensities of five emotions (Happiness, Anger, Fear, Sadness, Curiosity) displayed by the dogs. However, these emotions may not have been experienced by the dogs (or were not verified) but could have been attributed to them by human observers. We did not inspect the true affective states in the dogs via DogFACS or other means, instead we concentrated merely on the human perception of them. For each video, a set of five emotions (Happiness, Anger, Fear, Sadness, Curiosity) was provided on a 7point Likert scale, where 1 was the lowest intensity and 7 the highest intensity. The intensity of each reported emotion was assessed separately and independently of the other emotions. Participants were asked to rate the perceived intensity of all reported emotions in each featured video (‘emotional intensity’). These emotions were chosen as they were, in general, reasonable to expect in the dogs in the chosen situations and because they should be perceivable by human viewers. The situations ‘Called by name’ and ‘Play’ were designed to elicit mostly positive (Happiness) or neutral (Curiosity) emotions in the dogs, while ‘Separation’ and ‘Stranger’ were designed to elicit mostly negative emotions in the dogs (Anger, Fear, Sadness). Participants could not proceed with the survey until they had marked the intensity of all reported emotions for all presented videos. Once the participants had rated the entire video set, they were allowed to submit the survey. 3.4. Statistical analysis Analysis was performed in R 4.2.2 (R core team, 2022) To investigate how breed, realised ownership, hypothetical ownership and age affect the evaluation of emotional intensities from the videos, for each emotion, we created one full mixed model with cumulative link (‘clmm’ function in the ‘ordinal’ package in R) to avoid high computational costs. In total, we created five full models. The score of emotion was the ordered, factorial response variable. The interaction between hypothetical ownership (would own a brachycephalic dog – ‘brachy’/would only consider owning a normocephalic dog – ‘normo’/ would never consider owning a dog – ‘never’), breed (Boston terrier – ‘boston’/Jack Russell Terrier – ‘jack’) and situation (‘Play’/ ‘Called’/ ‘Separation’/‘Stranger’), the interaction between realised ownership (owns or owned at least one brachycephalic dog – ‘brachy’/only owned normocephalic dogs – ‘normo’/never owned a dog – ‘never’), breed and situation, and the age of the participants were included as fixed factors and the covariate. The IDs of participants and dogs were treated as random intercepts. We then compared this full model to all of its derivative models (‘dredge’ function in the package ‘MuMIn’) and selected the most parsimonious model within 2ΔAICc of the top model as the best model. If any interaction was included in the best model, we performed post hoc analysis (‘emmeans’ function in package ‘emmeans’) to examine the differences between interacted factors. We also bootstrapped the mean and 95 % CIs of the emotion score per emotion (‘boot’ and ‘boot.ci’ functions in the ‘boots’ package in R). 4. Results Detailed data characteristics, including intensity distribution per each emotion and descriptive statistics in Boston Terriers, can be found in Supplementary Materials 2 Table I-1 Description of the Boston Terrier emotional intensity values. For intensity distribution per each emotion and descriptive statistics in Jack Russell Terriers, see Supplementary Materials 2 Table I-2 Description of the Jack Russell Terrier emotional intensity values. The best model for each emotion had different combinations of the interactions of fixed factors and covariate (See Table 2). Only effects proven as significant were included in the final models. The intensity of each featured emotion was inspected independently. The interaction of breed and situation was included in all best models except for the model for Curiosity. The interaction between realised ownership and breed was included in the best models for Happiness and Anger. The interaction between hypothetical ownership and breed was included in the best model for Curiosity, and the interaction between hypothetical ownership and situation was included in the best model for Sadness. Overall, the age of participants had a weakly negative to no effect on their assessment of canine emotions. No inspected emotion showed an effect of subjective experience (results not shown). 4.1. Happiness The descriptive histogram showing the total raw scores (before statistical analysis was done) of Happiness ratings for all situations combined and divided by breed can be seen in Supplementary Materials 2 Figure II-1 - Total distribution of Happiness intensity in all situations in Boston Terriers and Jack Russell Terriers. For analysis, we chose the best fitting model based on ΔAICc. For five best fitting models for Happiness, see Supplementary Materials 2 Table II-1 Five best fitting models for analysing Happiness based on ΔAICc. The interaction between situations and breed, the interaction between realised ownership and breed, and the age of the participants significantly affected ratings of Happiness (see Supplementary Materials 2 Table II-2 Full model analysis details for Happiness). For Boston Terriers, ‘Play’ (mean [bootstrapped 95 % CIs] =3.31 [3.12, 3.51], the same for all values listed below) had a higher Happiness rating than other situations. ‘Called’ (2.58 [2.41, 2.79]) was rated higher in Happiness than ‘Separation’ (1.47 [1.38, 1.56]). The situation of ‘Stranger’ (2.2 [2.04, 2.37]) did not differ from either ‘Separation’ or ‘Called’ in the post hoc test. In Jack Russell Terriers, the intensity rating of Happiness was highest in the situation ‘Called’ (5.30 [5.11, 5.49]); ‘Called’ and ‘Play’ (3.90 [3.63, 4.16]) were rated higher than ‘Stranger’ (2.06 [1.90, 2.23]) and ‘Separation’ (1.79 [1.67, 1.92]), and the latter two did not differ from each other. Between the two breeds, Jack Russell Terriers were rated higher for Happiness than Boston Terriers for the ‘Called’ situation. The effect of the interaction of breed and situation in the two breeds, including differences in the rating of Happiness, can be seen in Fig. 1. A detailed post-hoc analysis of the breed and situation interaction on the rating of Happiness can be seen in Supplementary Materials 2 Table II-3 Emmeans of breed/situation interaction for Happiness. The interaction between realised ownership and breed significantly affected the rating of Happiness as including this interaction significantly improved the fitting of the model. Normocephalic dog owners gave Boston Terriers lower Happiness scores (2.31 [2.21, 2.43]) than brachycephalic dog owners (2.46 [2.28, 2.65]). Additionally, when rating happiness of Boston Terriers and Jack Russell Terriers, normocephalic dog owners rated Boston Terriers lower than they to Jack Russell Terriers (3.33 [3.18, 3.48]). The effect of interactions of breed, a) realised ownership and b) hypothetical ownership, including differences on the rating of Happiness, can be seen in Fig. 2. A detailed postTable 2 Summary of fixed factors and interactions in the best models. The √ marks a significant complex effect, while the ↓ marks a negative effect. Emotion/Effect Happiness Sadness Anger Fear Curiosity Age ↓   ↓ Breed * Situation √ √ √ √  Breed * Realised ownership √√  Breed * Hypothetical ownership     √ Situation * Hypothetical ownership √   Situation √ √ √ √ √ Breed √ √ √ √ √ Realised ownership √√  Hypothetical ownership √  √ P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 5 hoc analysis of the breed and realised ownership interaction on the rating of Happiness can be seen in Supplementary Materials 2 Table II-4 Emmeans of breed/realised ownership interaction for Happiness. There was a slight negative effect of participants’ age (OR =0.97; CI =0.96 – 0.98; p <0.001; for details see Supplementary Materials 2 Table II-1 Full model analysis details for Happiness) on the intensity rating of Happiness in dogs. 4.2. Anger The descriptive histogram showing the total raw scores (before statistical analysis was done) of Anger ratings for all situations combined and divided by breed can be seen in Supplementary Materials 2 Figure III-1 – Total distribution of Anger intensity in all situations in Boston Terriers and Jack Russell Terriers. For five best fitting models for Fig. 1. caption: The effect of breed and situation interaction on the distribution of the intensity rating of Happiness in Boston Terriers and Jack Russell Terriers. The dot and whiskers denote emmeans and 95 % CIs. The width of the violin represents the distribution of the scores, and the dashed lines indicate quartiles. Same for Figs. 2–9. †p<0.1; * p <0.05; *** p <0.001. Fig. 2. The effect of a) realised ownership and b) hypothetical ownership on participants’ distribution of the intensity rating of Happiness in Boston Terriers and Jack Russell Terriers. Emmeans, 95 % CI, †p<0.1; * p <0.05. P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 6 Anger, see Supplementary Materials 2 Table III-1 Five best fitting models for analysing Anger based on ΔAICc. The two interactions (interaction between situations and breed, interaction between realised ownership and breed) affected ratings of Anger (see Supplementary Materials 2 Table III-2 Full model analysis details for Anger). Although there were no differences in the ratings of Anger in all situations in Boston Terriers (‘Called’ =1.22 [1.15, 1.31]; ‘Play’ =1.28 [1.20, 1.37]; ‘Separation’ =1.33 [1.27, 1.41]; ‘Stranger’ = 1.22 [1.16, 1.29]), in Jack Russell Terriers, the Anger rating of the situation ‘Play’ (2.37 [2.18, 2.57]) was higher than ‘Called’ (1.34 [1.09, 1.19]) and ‘Stranger’ (1.7 [1.57, 1.82]), but was not statistically different from ‘Separation’ (1.47 [1.38, 1.57]). Between the two breeds, Boston Terriers were rated with less Anger in ‘Play’ and ‘Stranger’ situations than Jack Russell Terriers but were rated with more Anger in the ‘Called’ situation. The effect of the interaction of breed and situation in the two breeds, including differences in the rating of Anger, can be seen in Fig. 3. A detailed post-hoc analysis of the breed and situation interaction on the rating of Anger can be seen in Supplementary Materials 2 Table III-3 Emmeans of breed/situation interaction for Anger. Normocephalic dog owners gave Boston Terriers lower Anger scores (1.26 [1.21, 1.30]) than they gave Jack Russell Terriers (1.74 [1.65, 1.83]). Interestingly, participants without any dogs rated Boston Terriers as angrier (1.52 [1.37, 1.72]) than did participants owning at least one brachycephalic dog (1.18 [1.13, 1.24]). The effect of interactions of breed, a) realised ownership and b) hypothetical ownership, including differences on the rating of Anger, can be seen in Fig. 4. A detailed posthoc analysis of the breed and realised ownership interaction on the rating of Anger can be seen in Supplementary Materials 2 Table IIII-4 Emmeans of breed/realised ownership interaction for Anger. 4.3. Fear The descriptive histogram showing the total raw scores (before statistical analysis was done) of Fear ratings for all situations combined and divided by breed can be seen in Supplementary Materials 2 Figure IV-1 – Total distribution of Fear intensity in all situations in Boston Terriers and Jack Russell Terriers. For five best fitting models for Fear, see Supplementary Materials 2 Table IV-1 Five best fitting models for analysing Fear based on ΔAICc. The ratings of Fear were only affected by the interaction between situation and breed (see Supplementary Materials 2 Table IV-2 Full model analysis details for Fear). In Boston Terriers, videos of the ‘Play’ situation were rated with the least Fear (1.67 [1.54, 1.81]) compared to ‘Called’ (2.24 [2.07, 2.42]), ‘Stranger’ (2.72 [2.55, 2.91]) and ‘Separation’ (3.54 [3.35, 3.75]). The ‘Called’ situation also got a lower Fear score than ‘Separation.’ In Jack Russell Terriers, the ‘Called’ situation was rated with the lowest Fear (1.38 [1.29, 1.47]), followed by ‘Play’ (2.07 [1.92, 2.24]), ‘Separation’ (2.90 [2.72, 3.09]) and ‘Stranger’ (4.48 [4.27, 4.61]). Between the two breeds, Boston Terriers were rated with more Fear than Jack Russell Terriers in ‘Called’ and ‘Separation’ but were rated as having less Fear in the ‘Stranger’ situation. The effect of the interaction of breed and situation in the two breeds, including differences in the rating of Fear, can be seen in Fig. 5. A detailed post-hoc analysis of the breed and situation interaction on the rating of Fear can be seen in Supplementary Materials 2 Table IV-3 Emmeans of breed/ situation interaction for Fear. 4.4. Sadness The descriptive histogram showing the total raw scores (before statistical analysis was done) of Sadness ratings for all situations combined and divided by breed can be seen in Supplementary Materials 2 Figure V1 – Total distribution of Sadness intensity in all situations in Boston Terriers and Jack Russell Terriers. For five best fitting models for Sadness, see Supplementary Materials 2 Table V-1 Five best fitting models for analysing Sadness based on ΔAICc. Sadness ratings were affected by the interaction between situation and breed, as well as the interaction between the hypothetical Fig. 3. caption: The effect of breed and situation interaction on the distribution of the intensity rating of Anger in Boston Terriers and Jack Russell Terriers. †p<0.1; * p <0.05; *** p <0.001. P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 7 ownership and situation (see Supplementary Materials 2 Table V-2 Full model analysis details for Sadness). In Boston Terriers, videos of the ‘Play’ situation were rated with the lowest intensity of Sadness (1.59 [1.47, 1.72]) compared to ‘Called’ (2.01 [1.83, 2.18]), ‘Stranger’ (2.37 [2.18, 2.55]), and ‘Separation’ (3.02 [2.81, 3.23]). In Jack Russell Terriers, the ‘Play’ (1.49 [1.38, 1.63]) and ‘Called’ (1.19 [1.13, 1.26]) situations were rated with less intensity of Sadness than ‘Stranger’ (2.37 [2.19, 2.57]) and ‘Separation’ (2.91 [2.72, 3.11]). Between the two breeds, Boston Terriers were rated with a higher intensity of Sadness than Jack Russell Terriers in the ‘Called’ situation. The effect of the interaction of breed and situation in the two breeds, including differences in the rating of Sadness, can be seen in Fig. 6. A detailed post-hoc Fig. 4. caption: The effect of a) realised ownership and b) hypothetical ownership on participants’ distribution of the intensity rating of Anger in Boston Terriers and Jack Russell Terriers. Emmeans, 95 % CI, †p<0.1; * p <0.05. Fig. 5. caption: The effect of breed and situation interaction on the distribution of intensity rating of Fear in Boston Terriers and Jack Russell Terriers. †p<0.1; * p<0.05; *** p <0.001. P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 8 Fig. 6. caption: The effect of breed and situation interaction on the distribution of intensity rating of Sadness in Boston Terriers and Jack Russell Terriers. †p<0.1; * p<0.05; *** p <0.001. Fig. 7. caption: The effect of hypothetical ownership and situation interaction on the distribution of the intensity rating of Sadness in both breeds. Emmeans, 95 % CI, †p<0.1; * p <0.05; ** p <0.01; *** p <0.001. P. Eretov´ a et al. Applied Animal Behaviour Science 292 (2025) 106767 9