Statistical learning and parochial norms

• In Event: Poster Session 03
  In Poster Session: PS 03 Topic - Social Cognition

Abstract

Many norms are parochial, meaning they apply to certain groups (e.g., the ingroup) and not to others (e.g., the outgroup). Developmental accounts of parochialism typically appeal to biases in social learning (e.g., Dunham, 2018, 781; Roberts et al., 2017, 593). In contrast, research in cognitive development (Xu, 2020, Kushnir et al., 2010) indicates that children make rational statistical inferences across domains. This suggests an alternative account of parochialism based on statistical learning: given the evidence, learners make statistically appropriate inferences of normative scope. In the present study, we put this account to the test.

In an ongoing online study, children ages 5 to 8-years-old (current N = 58; Mage = 7.08 years, SD = 0.86, planned N=100) learn about two groups (“Hibbles” and “Glerks”) who live on an island. In one of two randomly assigned conditions, the size of the target group (e.g. “Hibbles,” counterbalanced) is either 20% or 80% of the total population. Within each group, a fixed proportion of individuals (35%) are wearing a clothing item (e.g., ribbons). Children watch a video that highlights 4 members of the target group, all wearing the clothing item, and are told, “This is against the rule.” Afterwards, we ask children whether the rule applies to three individuals wearing the clothing item: a member of the target (sampled) group, member of the non-sampled group, and a novel agent not in the original video. We end with an open-ended question “What is the rule?” Responses that were clear (93/116, 80%) were coded as: ‘Inclusive’ = 1 (e.g., “Nobody on the island can wear ribbons”) or ‘Parochial’ = 0 (e.g., “Only Hibbles can wear ribbons”). The procedure is repeated in Trial 2 with a different clothing item (e.g., hats).

The statistical learning account predicts that inferences the 20% condition should be more parochial; meaning, children should be more likely to apply the rule restrictively to the target group alone. Indeed, though both groups of children applied the rule significantly more to the sampled group (matching what they saw, see Figure 1), children in the 20% condition are ~5.7 times less likely to generalize it to non-sampled group (OR20 = 5.65, 95% CI: 2.5, 13.5), while children in the 80% condition are ~2.4 times less likely to do so (OR80 = 2.41, 95% CI: 1.1, 5.3). Open responses (Figure 2) directionally support the statistical learning account (M20 = 0.39, M80 =0.55) – more children explicitly stated a parochial rule in the 20% condition – but this difference is not significant (β= 0.33, SE= 0.21, p = .12). Children’s application of the rule to the novel agent was not significantly different than from the target, suggesting avoidance of the non-sampled group was responsible for narrowing generalizations, rather than a wholesale bias towards parochiality. In sum, despite the typical emphasis on biases in social learning, early results suggest that children can also be appropriately sensitive to statistical evidence when learning normative scope.

Authors

• Scott Partington, Cornell University
  Presenting Author

• Shaun Nichols
  Non-Presenting Author

• Tamar Kushnir, Duke Univeristy
  Non-Presenting Author