An adapted Qualitative Behavioural Assessment of dogs’ facial expressions of fear and frustration

Abstract

Qualitative Behavioural Assessment (QBA) involves the quantification of people’s holistic interpretations of the style and quality of movement and has been used to assess the emotional states of many species, including dogs. Faces are a rich source of emotional expression, but there is enormous variability in dog facial morphology limiting the potential to use standardised anatomically-based descriptions across breeds. However, we hypothesise that style of movement in relation to emotion may be consistent and detectable by humans, by focusing on the face and irrespective of the morphological differences that occur in dogs. We aimed to see whether we could use an adapted form of Free Choice Profiling QBA (aFCP-QBA) where participants concentrated on the face of the dog, to reveal qualitative discriminations of fear and frustration. Twenty-two behaviourists were asked to describe using a aFCP-QBA methodology (i.e. using their own descriptors adjectives) the facial expressions of 5 fear and 5 frustration videos from a previously validated dataset of videos. They were blind to the video’s previous validation of their emotional content. Forty-nine adjectival descriptors were generated and Generalised Procrustes Analysis demonstrated significant inter-observer reliability (55 %, p < 0.01) suggesting congruence in the description of facial emotions using our aFCP-QBA. Three Principal Components explaining 61 % of data variance were extracted and appeared to represent expressions of arousal, anticipation and valence. Participants generally assigned the videos to two groups consistent with our a priori emotional classification, suggesting they discriminated the two different emotions. There was good consensus for the emotional content of the fear videos, with similar adjectives used by participants. For the frustration videos, participants agreed that these videos differed from the fear videos (as they were separated by arousal, anticipation and valence) but the consensus profile of the adjectives was weaker. Participants also rated the videos using four visual analogue scales (VAS) relating to the level of fear, frustration, happiness and positive anticipation expressed by the dogs. Fear videos were rated significantly higher on fear (Friedman Chi-square 184.67, p < 0.01), and frustration videos were rated significantly higher for frustration (Friedman Chi-square 161.80, p < 0.01). This provides convergent validity for each of the two emotional expressions revealed by the aFCP-QBA. These results indicate that humans may identify the emotional state of dogs using holistic evaluations of the style when asked to concentrate on the facial expressions of dogs.