Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster A in Poster Session A: Tuesday, August 4, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

PyDDMBayes: A Multilevel Bayesian Framework for Generalized Drift-Diffusion Modeling

Covert Geary¹, Maxwell Shinn², John D Murray¹; ¹Dartmouth College, ²University College London

Presenter: Covert Geary

Generalized drift-diffusion models (GDDMs) extend the classical DDM to support richer mechanistic hypotheses about decision-making. Standard approaches force a trade-off: fitting participants independently is noisy and data-hungry, while pooling them obscures individual variation. Multilevel modeling can resolve this trade-off through partial pooling, but existing multilevel implementations are restricted to classical DDM forms or require computationally expensive pre-training. More fundamentally, multilevel GDDMs pose statistical challenges that standard off-the-shelf methods fail to overcome. Here we present PyDDMBayes, a framework and Python toolbox for multilevel Bayesian analysis of arbitrary GDDMs. In simulation, multilevel estimation reduces participant-level parameter recovery error compared to unpooled maximum likelihood, particularly with few trials. We show how a single multilevel GDDM yields both population-level inference and participant-level estimates of individual differences, and how the framework supports model comparison between competing mechanisms. This framework unifies inference across levels of analysis for arbitrary GDDMs.

Topic Area: Methods, Tools, Theory & Neural Coding