Lingo
← Back to Neuroscience

Hierarchical Drift Diffusion Model (HDDM)

Overview

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.

Shrinkage and Statistical Power

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.

  • Subjects with lots of consistent data strongly influence their own parameter estimates.
  • Subjects with very few trials rely heavily on the group prior, preventing wildly inaccurate estimates.

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.

Clinical and Cognitive Applications

The HDDM is widely used in clinical neuroscience and psychology to detect subtle differences between groups:

  • Parkinson's Disease: HDDM analyses have shown that deep brain stimulation alters the decision threshold (boundary) rather than the rate of information processing.
  • ADHD: HDDM reveals differences in drift rate and non-decision time compared to neurotypical controls.
  • Pharmacological Interventions: Tracking how specific drugs (like L-DOPA) affect distinct cognitive sub-processes.

References

  1. Wiecki, Sofer, & Frank (2013) — Frontiers in Neuroinformatics— HDDM: Hierarchical Bayesian estimation of the Drift-Diffusion Model in Python.

Examples