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¶

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? The best-fitting RL model separated learning rates for experiential and observational trials, with significantly lower AIC than all alternatives. → Maria A. Woitow 2026 q2 i? Median surprise and uncertainty estimates trended above zero but none reached statistical significance and did not differ between conditions. → Maria A. Woitow 2026 (2)

Evidence¶

Maria A. Woitow 2026¶

Maria A. Woitow, Anthony I. Jang, Ben Eppinger, Matthew R. Nassar, Marcel Brass & Julia M. Rodriguez Buritica. (2026). Shared effects of one's own and others' experiences during reinforcement learning on episodic memory. npj Science of Learning 11:16. https://doi.org/10.1038/s41539-026-00409-7

q2 · i? · causal · r2

RL models fitted to choices of 79 participants; the model with separate surprise and uncertainty parameters for EL and OL trials had "significantly lower AIC" than all other models (p < 0.001 across pairwise comparisons), replicating Jang et al.'s modeling results.

"The model that best accounted for participants’behavior (Fig.2b, c) sepa- rated the learning rate for experiential (EL) and observational learning (OL) trials"

Maria A. Woitow 2026 (2)¶

Maria A. Woitow, Anthony I. Jang, Ben Eppinger, Matthew R. Nassar, Marcel Brass & Julia M. Rodriguez Buritica. (2026). Shared effects of one's own and others' experiences during reinforcement learning on episodic memory. npj Science of Learning 11:16. https://doi.org/10.1038/s41539-026-00409-7

q2 · i? · causal · r2

Wilcoxon signed-rank tests of maximum likelihood parameter estimates against zero: Surp EL median 0.11 (p = 0.87), Surp OL 0.30 (p = 0.19), Unc EL 0.49 (p = 0.06), Unc OL 0.29 (p = 0.06); estimates did not differ between conditions (Surp p = 0.34, Unc p = 0.97).

"While all four parameters trended above zero, none reached statistical significance (Surp EL: median = 0.11, p = 0.87; Surp OL: median = 0.30, p = 0.19; Unc EL: median = 0.49, p = 0.06; Unc OL: median = 0.29, p = 0.06; Wilcoxon signed-rank t-tests against 0)"

Discussion¶

Related Claims¶

• Preschoolers' Chicken game choices are best explained by reward-based learning with a higher learning rate for the competitive opponent — related
• A risk-sensitive reinforcement learning model with anticorrelated valuation fit reversal learning behavior better than standard models — related
• Adolescents learn faster from negative reward prediction errors than adults in probabilistic reversal learning — related
• Choice behavior is more value-driven in the observational than individual learning condition, and adults show higher choice specificity than children — related
• Observational learning yields high performance early in learning while individual learning improves gradually across trials — related