The Decisive Role of Non-Decision Time for Interpreting the Parameters of Decision Making Models

Abstract

Computational models of decision making account for decision performance by postulating latent cognitive stages, typically distinguishing the durations of decision and non-decision processes. Decision times result from the progressive accumulation of the evidence needed to make a choice. Non-decision times result from the need to encode the stimulus early on, and to generate a response after a decision is reached. The properties of this non-decision time have seldom been investigated. Here we used the onsets of electro-myographical activity in the responding muscles to decompose each reaction time into a pre-motor and a motor time. Drift diffusion models were then fitted on reaction times, and on pre-motor times. This comparison, combined with experimental manipulations of stimulus contrast and response difficulty, provided a window on non-decision processes. The results are best explained by assuming that stimulus encoding does not stop when evidence accumulation starts, and that motor response onset does not always mark the end of the deliberation process, in stark contrast with standard assumptions. This interpretation applies to conditions where participants are asked to favor accuracy over speed, but not vice-versa. Accordingly, the non-decision parameter has to be interpreted differently depending on the participants’ adjustments to task instructions. In conclusion, non-decision times reflect cognitively richer processes than is usually assumed, and they can play a decisive constraining role for parameter interpretation in decision making models.