ParAcT-DDM: A diffusion-based framework for modelling systematic, time-varying cognitive processes

As people engage in tasks over extended periods, their psychological states systematically change due to factors, such as practice, learning, and/or boredom. However, the dominant frameworks for modelling cognitive processes – such as evidence accumulation models – only consider a single estimate of a process across the duration of an experiment, with the justification often being that these changes are not of interest, are likely minimal, and/or can be eliminated by experimental procedures such as practice blocks. Our study proposes, develops, and assesses the ParAcT-DDM framework: the parameters across time diffusion decision model – which unifies previous modelling efforts from practice and decision-making research. Specifically, our framework models time-varying changes to diffusion decision model parameters by assuming that rather than being constant across time, their estimates follow theoretically informed trial-varying or block-varying functions across time. Focusing on two diffusion model parameters – drift-rate (task efficiency) and threshold (caution) – and a large number of candidate time-varying functions, our simulation results suggest that most variants of the ParAcT-DDM framework display robust inference properties, both in terms of the key parameters being well identified, and the different variants being distinguishable from one another. Our empirical results show that ParAcT-DDM variants vastly outperform the standard diffusion model in three existing data sets, including one where participants completed a practice block before data began recording, suggesting that time-varying cognitive processes often occur in typical cognitive experiments, even when the experimental design explicitly tries to remove practice effects. Finally, we find that the existence of time-varying processes cause systematic biases in the parameter estimates of the standard diffusion model, suggesting that our ParAcT-DDM framework can be crucial to ensuring the robustness of inferences against time-varying changes, regardless of whether these changes are of direct interest.