Bayesian Regression Analysis with the Drift-Diffusion Model

2025-07-01

AI Summary

Traditional drift-diffusion model (DDM)–phenotype association studies commonly employ a two-step approach—first estimating trial-level DDM parameters, then regressing subject-level phenotypes onto these estimates—introducing substantial estimation bias. To address this, we propose and implement a unified Bayesian hierarchical regression framework that jointly models trial-level diffusion processes and subject-level phenotype associations within a single coherent model, thereby eliminating two-step bias. The framework leverages Markov chain Monte Carlo (MCMC) sampling, enabling flexible covariate specification and cross-subject parameter sharing. We develop RegDDM, an open-source R package providing standardized fitting interfaces and comprehensive diagnostic tools. Empirical analyses and simulation studies demonstrate that our method substantially improves parameter estimation accuracy and statistical inference reliability compared to the two-step approach—particularly under small-sample or weak-effect conditions. This work establishes a robust, scalable, and integrated analytical paradigm for cognitive modeling and individual-differences research.

Technology Category

• Cognitive Modeling & Cognitive Systems: Simulating Human Behavior
• Reasoning under Uncertainty: Relational Probabilistic Models
• Computer Vision: Diffusion Models for Vision

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Content-based information diffusion

Abstract

The Drift-Diffusion Model (DDM) is widely used in neuropsychological studies to understand the decision process by incorporating both reaction times and subjects' responses. Various models have been developed to estimate DDM parameters, with some employing Bayesian inference. However, when examining associations between phenotypes of interest and DDM parameters, most studies adopt a two-step approach: first estimating DDM parameters, then applying a separate statistical model to the estimated values. Despite the potential for bias, this practice remains common, primarily due to researchers' unfamiliarity with Bayesian modeling. To address this issue, this tutorial presents the implementations and advantages of fitting a unified Bayesian hierarchical regression model that integrates trial-level drift-diffusion modeling and subject-level regression between DDM parameters and other variables. The R package RegDDM, developed and demonstrated in this tutorial, facilitates this integrated modeling approach.

Problem

• Estimating DDM parameters with Bayesian hierarchical regression
• Avoiding bias in phenotype-DDM parameter association studies
• Integrating trial-level DDM and subject-level regression modeling

Innovation

• Bayesian hierarchical regression model integration
• Trial-level drift-diffusion modeling combined
• R package RegDDM facilitates implementation