The Hierarchical Drift Diffusion Model (HDDM), popularized by Wiecki, Sofer, and Frank (2013), is a statistical framework for estimating the parameters of the Drift Diffusion Model (DDM) across a group of subjects or conditions.
Rather than fitting the DDM to each subject independently (which is noisy and requires hundreds of trials per subject) or pooling all data together (which ignores individual differences), the HDDM uses Bayesian hierarchical modeling. It assumes that individual subject parameters are drawn from an overarching population distribution.
The defining mathematical feature of the HDDM is shrinkage. Because the model knows that subjects belong to a group, it "shrinks" extreme, noisy individual estimates toward the group mean.
This allows researchers to reliably estimate DDM parameters (like drift rate and decision boundary) even when they only have 20–40 trials per participant—something completely impossible with traditional Maximum Likelihood fitting methods.
The HDDM is widely used in clinical neuroscience and psychology to detect subtle differences between groups: