Drift rate increases as a continuous exponential function of cumulative trial number across four days of perceptual training¶

Claim · All claims
Evidence · 1 study · associational r2 · q3
Currency · newest study 2023 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? In the best-fitting hierarchical Bayesian DDM, drift rate increased with training as an exponential function of overall trial number (trials 1 through 2800), with every participant's drift rate estimated as increasing. → Cochrane 2023

Evidence¶

Cochrane 2023¶

Cochrane, A., Sims, C. R., Bejjanki, V. R., Green, C. S., & Bavelier, D. (2023). Multiple timescales of learning indicated by changes in evidence-accumulation processes during perceptual decision-making. npj Science of Learning 8:19. https://doi.org/10.1038/s41539-023-00168-9

q3 · i? · associational · r2

Fully Bayesian nonlinear mixed-effects DDM fits to joint RT and accuracy data from 21 participants completing 700 trials per day across 4 days of dot-motion discrimination. Fixed effects show "DR increased with training" with start mean 0.44 and asymptote mean 0.96, time-to-half-of-learning 1261 trials; no standardized effect size is printed.

"Model fixed effects indicated that DR increased with training (start mean = 0.44, se = 0.05; asymptote mean = 0.96, se = 0.11), with a sample-level time- to-half-of-learning of 1261 trials. Every participant 's DR was estimated as increasing"

Discussion¶

Related Claims¶

• By-participant model fits replicate the continuous DR, flexible RB result for most individual learners — related
• Model comparison favors continuous drift rate with flexible response boundary over constant, fully continuous, and day-resetting alternatives — related
• The exponential functional form of the BKT Markov chain calls into question the popular practice of fitting that model form to student data — related
• The HMM form of BKT, solved analytically, is a three-parameter exponential in opportunity number — related
• Accuracy-only logistic psychometric fits support a two-timescale model but cannot identify which decision component changed — related