🤖 AI Summary
This study addresses the limited cognitive interpretability of reaction time (RT) modeling in choice tasks. Methodologically, it proposes a conditional modeling framework that integrates the first-passage-time distribution of diffusion models with generalized linear mixed models (GLMMs): RTs are conditionally distributed as inverse Gaussian or Gamma variates, while key diffusion parameters—such as drift rate and boundary threshold—are embedded into the GLMM’s linear predictor, enabling joint estimation of population- and subject-level cognitive parameters. Its primary contribution is the first construction of an identifiable, cognitively interpretable, and computationally tractable “cognitive–statistical” bridge: it preserves the theoretical foundations of diffusion modeling while remaining fully compatible with standard mixed-model software. Simulation and empirical analyses demonstrate robust recovery of canonical cognitive effects (e.g., speed–accuracy trade-offs) and yield substantial improvements in RT distribution fit and mechanistic inference reliability.
📝 Abstract
This study connects two methods for modeling reaction times (RTs) in choice tasks: (1) the first-hitting time of a simple diffusion model with a single barrier, representing the cognitive process leading to a response, and (2) Generalized Linear Mixed Models (GLMMs). We achieve this by analyzing RT distributions conditioned on each response alternative. Because certain diffusion model variants yield Inverse Gaussian (IG) and Gamma distributions for first-hitting times, we can justify using these distributions in RT models. Conversely, employing IG and Gamma distributions within GLMMs allows us to infer the underlying cognitive processes. We demonstrate this concept through simulations and apply it to previously published real-world data. Finally, we discuss the scope and potential extensions of our approach.