Hidden Markov Models of Evidence Accumulation in Speeded Decision Tasks

The evidence accumulation models (EAMs) are useful to study cognitive processes and their effects on response times and accuracy, capturing dependencies between the two. One phenomenon of interest is the speed-accuracy trade-off, where individuals sacrifice one for the other. Classical EAMs assume a continuous trade-off between speed and accuracy, thereby allowing performance to vary between guessing and (in principle) almost perfect responding. However, alternative tradition of thinking suggests that participants may switch between distinct states rather than control the trade-off on a continuum. Hidden Markov Models (HMMs) are typically used to describe such behaviour, assuming two states - random guessing and stimulus-controlled states. Typical HMM applications assume that speed and accuracy are independent of each other, conditioned on the states. However, evidence accumulation presumably takes place at the least under the controlled state, inducing a speed-accuracy trade-off within that state. In this talk, we introduce a model that combines a HMM with an EAM that contains a discontinuous speed-accuracy trade-off on a larger scale (between states) and a continuous speed-accuracy trade-off on a smaller scale (within states), and show some applications on empirical data. We’ll also discuss our experiences with a robust Bayesian workflow employed to validate the implementation of the model, and potential extensions to the model and its applications.