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¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2020 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? Children (ages 7–12) were best fit by the one learning rate model (PXP = 0.98), indicating they updated values from experienced outcomes alone without rationally discounting agent-attributable outcomes. → Alexandra O. Cohen 2020 q2 i? Adults were best fit by the empirical Bayesian model (PXP = 0.75) and adolescents by the adaptive Bayesian model (PXP = 0.89), indicating belief-guided learning. → Alexandra O. Cohen 2020 (2)

Evidence¶

Alexandra O. Cohen 2020¶

Alexandra O. Cohen, Kate Nussenbaum, Hayley M. Dorfman, Samuel J. Gershman and Catherine A. Hartley. (2020). The rational use of causal inference to guide reinforcement learning strengthens with age. npj Science of Learning 5:16. https://doi.org/10.1038/s41539-020-00075-3

q3 · i? · causal · r2

Random-effects Bayesian model comparison of seven reinforcement learning models fit to choice data within three age groups. Children's protected exceedance probability for the one learning rate model was 0.98; all other models' PXPs were below 0.01.

"children were best fit by the one learning rate model (PXP = 0.98) relative to the other models (all PXPs < 0.01)"

Alexandra O. Cohen 2020 (2)¶

Alexandra O. Cohen, Kate Nussenbaum, Hayley M. Dorfman, Samuel J. Gershman and Catherine A. Hartley. (2020). The rational use of causal inference to guide reinforcement learning strengthens with age. npj Science of Learning 5:16. https://doi.org/10.1038/s41539-020-00075-3

q3 · i? · causal · r2

Model comparison within age groups: adolescents were best fit by the adaptive Bayesian model (PXP = 0.89; all others < 0.10), adults by the empirical Bayesian model (PXP = 0.75). Simulation recovery confirmed all best-fitting models were recoverable (PXP = 1).

"adult choices were better captured by the empirical Bayesian model (PXP = 0.75) over the other models (all PXPs < 0.08)"

Discussion¶

Related Claims¶

• Adolescents flexibly estimate hidden-agent intervention probability from experience, less tied to their explicitly reported beliefs than adults — related
• Participants aged 7–25 attribute positive and negative outcomes to hidden agents in line with the true causal structure of each environment — related
• Preschoolers' Chicken game choices are best explained by reward-based learning with a higher learning rate for the competitive opponent — related
• Model support shifts from reward-based learning in early primary school to belief-based opponent tracking from grades 4-5 onward — a broader claim this one bears on
• Adults with ADHD show reduced reliance on Bayesian inference strategies (lower alpha) than controls in complex learning settings — related