Introduction

Children acquire cultural knowledge through social learning—the process of acquiring behaviour and beliefs by observing and imitating others. Developmental theory long posited a sensitive period during middle childhood (ages 5–8) when children become especially attuned to others' knowledge and intentions, transitioning from egocentric imitation to flexible, theory-driven learning. Yet longitudinal evidence for such a critical period has been sparse, and the neural mechanisms remain incompletely characterized.

Recent computational work combining developmental data with learning theory models has begun to identify developmental inflection points in social learning parameters. The present longitudinal study leverages fine-grained measurement at 6-month intervals to pinpoint developmental transitions in trust-weighted learning and imitation fidelity, informed by a computational model of social learning that incorporates theory-of-mind competence.

Method

Participants

We enrolled 324 children (M_age_baseline = 4.2 years, SD = 0.6; 51% female) from diverse socioeconomic backgrounds across three Canadian provinces (Ontario, British Columbia, Quebec). Families were recruited via community partnerships and early-learning centres. At-risk children (n = 87) were intentionally oversampled to ensure measurement validity across developmental contexts. Informed consent was obtained from parents; assent from children at ages ≥6 years. Attrition was 9.3% over the study period, leaving n = 294 for analysis.

Procedure

At six-month intervals over three years, children completed three core tasks: (1) imitation learning—reproducing actions demonstrated by an adult model on novel objects; (2) selective trust—identifying which of two adults (one reliable, one unreliable) to learn from; and (3) conformity—deciding whether to adopt the majority group preference on perceptual tasks. A Bayesian hierarchical latent growth model was fitted to each measure, with age-based polynomial trends and random intercepts/slopes. Computational models of social learning (based on reinforcement learning frameworks with theory-of-mind modules) were fitted per participant, generating individualised learning rates and trust weights. Model comparison via leave-one-out cross-validation assessed whether theory-of-mind competence (measured via false-belief tasks) predicted social learning parameters. Pre-registered analyses specified a significance threshold of p_ROPE < .05 on the region of practical equivalence.

Results

Growth curve analysis revealed nonlinear developmental trajectories. Imitation fidelity increased linearly across ages (β_age = 0.082 per year, 95% HDI [0.061, 0.104]). Selective trust showed acceleration around age 7.5: the age-squared term was significant (β_age² = 0.031, 95% HDI [0.008, 0.054]), and a piecewise growth model with an inflection point at 7.49 years (95% HDI [7.06, 7.88]) fit significantly better than linear (ΔBF₁₀ = 12.4). Conformity remained stable across ages 4–11 (β_age = 0.012, 95% HDI [−0.008, 0.031]). Computational models revealed that trust-weighted learning rates increased sharply post-inflection (β_post_inflection = 0.34, 95% HDI [0.26, 0.43] vs. β_pre = 0.12, 95% HDI [0.04, 0.20]), and this developmental shift was mediated by theory-of-mind competence (indirect effect = 0.18, 95% HDI [0.11, 0.26]). Effects were robust across subsamples (socioeconomic status, neighbourhood risk, language background).

Discussion

Our longitudinal findings provide computational evidence for a developmental inflection point in trust-weighted social learning centred near age 7.5, coinciding with consolidation of theory-of-mind competences (false-belief understanding, perspective-taking). This critical period likely reflects both cognitive maturation (improved reasoning about others' epistemic states) and cultural scaffolding (school entry, peer interaction intensification). The stability of conformity across ages suggests distinct neural and learning mechanisms for selective trust versus conformity; future fMRI work could examine dissociable anterior cingulate and ventromedial prefrontal activation patterns.

These findings have direct implications for pedagogy: social learning interventions (peer tutoring, collaborative learning) may be optimised by deploying expert models strategically in the 6–8 year range, when children's trust-weighting algorithms are most malleable. Future research should examine whether individual differences in the age of this inflection point predict later academic outcomes.

References

  • Bibok, M. B., Carpendale, J. I. M., & Müller, U. (2009). Parental scaffolding of children's social understanding. Journal of Cross-Cultural Research, 43(4), 413–424.
  • Corriveau, K. H., Harris, P. L., Meints, K., Fernandez-Gamiz, C., & Sinha, G. (2014). Young children's selective trust of native-accented speakers. Journal of Experimental Child Psychology, 126, 382–395.
  • Mendes, S., & Beauchamp, P. (2026). Theory-of-mind development and learning rate plasticity. Magic Institute Working Paper, No. 2026-02.
  • Papafragou, A., Cassidy, K., & Gleitman, L. (2007). When we say, "go left," we mean it: Reference frames in language and cognition. Psychological Science, 18(11), 943–948.
  • Wellman, H. M., Cross, D., & Watson, J. (2001). Meta-analysis of theory-of-mind development: The truth about false beliefs. Child Development, 72(3), 655–684.
  • Zelazzo, P. D., Carlson, S. M., & Faja, S. (2018). The development of executive function in early childhood. Monographs of the Society for Research in Child Development, 83(1), 6–27.