Preschoolers' Chicken game choices are best explained by reward-based learning with a higher learning rate for the competitive opponent¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Among hierarchical Bayesian RL models, a simple reward-based model with separate learning rates per opponent best fit preschoolers' trial-by-trial choices. → Zhang 2026 q2 i? Preschoolers updated more strongly from outcomes when facing the high-competitive opponent. → Zhang 2026 (2)

Evidence¶

Zhang 2026¶

Zhang, R., Zhang, L., Hao, X.-M., Dreher, J.-C., & Qu, C. (2026). Learning to adaptively cooperate through social interactions during childhood and adolescence. npj Science of Learning. https://doi.org/10.1038/s41539-026-00423-9

q2 · i? · causal · r2

Hierarchical Bayesian computational modeling of Study 1 choices; LOOIC comparison favored the reward-based two-learning-rate model with preference parameter, and posterior predictive checks showed high trial-level correlations (rLCO = 0.86, rHCO = 0.85).

"Further model comparison revealed that M1 -3 was the best-fitting model (LOOIC = 4564.48), which posited that children's cooperative behaviors were based on the current reward, following a simple RL rule, and that children had different learning rates for different opponents."

Zhang 2026 (2)¶

Zhang, R., Zhang, L., Hao, X.-M., Dreher, J.-C., & Qu, C. (2026). Learning to adaptively cooperate through social interactions during childhood and adolescence. npj Science of Learning. https://doi.org/10.1038/s41539-026-00423-9

q2 · i? · causal · r2

Group-level hyperparameters of the winning model: mean learning rate for HCO was 0.32 versus 0.20 for LCO; a Bayesian Wilcoxon Signed-Rank Test found a significant difference (BF₁₀ > 100).

"Preschoolers showed higher learning rates when interacting with HCO (M = 0.32, SD = 0.11, 95% HDI = [0.105, 0.561]) than with LCO (M = 0.20, SD = 0.15, 95% HDI = [0.2×10-4, 0.519])."

Discussion¶

Related Claims¶

• Higher reward sensitivity (BAS) is associated with more competitive Go straight choices in children — related
• Learning-rate asymmetries between opponent types reverse across grade groups, with a non-monotonic age trend — related
• Model support shifts from reward-based learning in early primary school to belief-based opponent tracking from grades 4-5 onward — related
• Preschoolers gradually learn to cooperate more with high-competitive than low-competitive opponents in the Chicken game — related
• An instructive reinforcement learning model with a free constant learning rate best explains participants' contingency learning — related
• School-age children and adolescents distinguish competitive from cooperative opponents, with cooperation peaking in late childhood — related
• A reinforcement learning model with separate surprise and uncertainty learning-rate parameters per condition best fits gambling behavior, though individual parameter estimates did not reach significance — related
• Children's reinforcement learning is best fit by a one-learning-rate model, whereas adolescents and adults are best fit by Bayesian models incorporating hidden-agent beliefs — related