Preschoolers use Minimal Statistical Information About Social Groups to Infer Individuals’ Preferences and Group Membership

• In Event: The Statistics of Group Cognition: New Evidence for Children’s Statistical Learning of Social Groups

Abstract

Social groups give order to an otherwise messy social world by highlighting similarities and relationships between people (Macrae & Bodenhausen, 2000; Rhodes, 2012). One important property of social groups is their preferences (i.e., what its members like and dislike). Past work has primarily focused on how children learn which preferences are important and generalizable secondhand through generic language (Gelman & Roberts, 2017; Roberts et al., 2017). The present work leverages computational and behavioral methods to provide evidence for a second, complementary learning mechanism: Children can also build representations of social groups firsthand by harnessing the statistics of the social environment.

We examined whether preschool-aged children use statistical information about how preferences are distributed across social groups to infer the preferences and group membership of new individuals. In Study 1, children (N=97, ages 4–5) watched agents in two teams choose which of two novel fruits they’d like to eat at snack time, and they were then introduced to a new agent (the “target”; Figure 1a). In the Infer Group task, children saw which fruit the target likes and predicted which team it belongs to. In the Infer Preference task, children saw which team the target belongs to and predicted what it likes. Within each task, we also manipulated whether a preference was unique to a team and unanimously shared by its members, in order to manipulate the statistical evidence supporting each of these inferences (Figure 1b).

We find, first, that children use agents’ preferences to infer group membership, even in the more challenging condition where the preference is not uniquely associated with a particular team (Figure 1c). These findings stand in contrast to past work, which has proposed that preschoolers have difficulty inferring group membership (Gelman, Collman & Maccoby, 1986; Vélez, Wu & Gweon, 2018). Our results suggest that, when given appropriate evidence, children can make flexible and sophisticated inferences about group membership. Second, children generalized preferences to new individuals when both teams’ preferences were unanimous. However, somewhat puzzlingly, they did not do so when the target’s team was unanimous, but the opposing team’s preferences were heterogeneous (Figure 1c, grey square).

Ongoing work seeks to explain how children make these inferences using probabilistic cognitive models. We hypothesize that children’s inferences about social groups can be characterized by a hierarchical, generative model of how individual agents’ preferences relate to the group’s, and of how different groups relate to one another. We are currently testing the predictions of such a model by asking whether children’s inferences are also sensitive to the size of each group (Study 2; N=64, ages 4–5) and to information about how groups were generated (Study 3; pilot sample of N = 282 adults). Developing formal theories of how children reason about social groups may yield new insights into a fundamental problem in social cognitive development: how children put together a patchwork of firsthand experiences and secondhand testimony to form a coherent picture of the social world.

Authors

• Natalia Velez, Stanford University
  Presenting Author

• Hyowon Gweon, Stanford University
  Non-Presenting Author