How is Emotion Knowledge Represented Across Development

• In Event: The development of emotion organization from infancy to adolescence

Abstract

Children’s emotion representations are hypothesized to be driven by the dimensions of valence and arousal (Russell 2003), and to become more discrete over time (Widen 2013). However, our knowledge of children’s representations of facial cues is often limited by the methods used. Much of this work relies on children’s ability to produce or use emotion labels, which may not capture all that they know about emotion categories (Barrett et al., 2019).
In the present study, we used a more open-ended task that does not prescribe the use of specific dimensions or emotion labels (Goldstone 1994; Unger et al., 2016) . We aimed to learn how children believe different facial cues relate to one another, and what factors underlie the dimensions that children use when evaluating similarity. To do so, we asked participants to sort facial cues of emotion on a grid based on whether people were “feeling the same kind of thing”. We tested 107 children (3;0 to 6;11) and 40 adults. Participants completed three sorts: a practice phase with images from different superordinate categories (objects and animals), a facial sort with 18 images of one individual, and a facial sort with 18 different individuals. The facial stimuli included open and closed mouth images of anger, calm, disgust, excitement, fear, happiness, neutral, sadness, and surprise. All facial stimuli were rated on valence and arousal by 50 separate adults.
To examine how different faces were grouped together we used hierarchical clustering on the distances between sorted images. To analyze whether sorts differed across development, we compared children’s and adult’s clustering structures using the Adjusted Rand index – a measure of the similarity of two clusters – and the correlation between dendrograms of the clustering solutions. Overall, we found that children became increasing adult-like across development, as indicated by increasing Adjusted Rand values and correlations (see Table 1). These changes appear to be emotion specific and not improvement on the task generally, as there were much smaller differences between age groups for practice trials (except for 3-year-olds). Children closer in age to one another also had more similar clustering solutions. For example, 5-year-olds had a stronger correlation with 6-year-olds (cor = 0.82) than with adults (cor = 0.38).
Next, we explored what features related to the different clustering solutions by examining ratings of valence and arousal for the images in each cluster. Figure 1 plots these clusters on top of a 2-dimensional representation of an “average sort”. We found that cluster group predicted ratings of images’ valence in children, χ2(2)=42.92, p<.001, and adults, χ2(2)=41.30, p<.001. Cluster group only predicted arousal ratings in children, χ2(2)=5.82, p=.01. For both adults and children, mean valence ratings were different for each cluster. One cluster had more positive mean valence, one had more negative mean valence, and one was in between. Overall, children’s knowledge of facial cues of emotion gradually becomes more adult-like. Images in each cluster also had differences in mean valence ratings – suggesting this feature is particularly important in adult’s and children’s sorting behaviors.

Authors

• Kristina Woodard, University of Wisconsin - Madison
  Presenting Author

• Martin Zettersten, Princeton University
  Non-Presenting Author

• Seth David Pollak, University of Wisconsin - Madison
  Non-Presenting Author