Non-decision time: the Higg’s boson of decision

Abstract

Generative models of decision now permeate all subfields of psychology, cognitive and clinical neuroscience. To successfully represent decision mechanisms, it is necessary to also assume the presence of delays for sensory and motor information to travel through the brain; but like the Higg's boson in particle physics, directly observing this "non-decision time" from behaviour long appeared beyond reach. Here, we describe and apply a set of methods to empirically measure and characterise the properties of non-decision time in fast visually guided decisions (without requiring modelling assumptions). We gather 11 datasets from humans and monkeys from multiple labs and validate the method by showing that visual properties (brightness, colour, size) consistently affect empirically measured non-decision time, as predicted by neurophysiology. We then show that endogenous factors (pro-active slowing, attention) consistently do not affect non-decision time, in contrast to widespread reports based on model fits. Last, contrasting empirically observed non-decision time with estimates from the EZ, DDM and LBA models, we conclude that models cannot be generally trusted to provide valid estimates, either at a group level or for individual differences, and propose a hybrid approach that combines our empirical method with standard modelling.

Visual Feature Tuning Properties of Short-Latency Stimulus-Driven Ocular Position Drift Responses during Gaze Fixation

“…We es?mated saccade rate using the method described in (Khademi et al, 2023): briefly, we calculated saccade onset likelihood within 50 ms moving windows that were stepped in ?me by 1 ms steps, and we did this on a per-trial basis; across-trial average rates were then obtained in order to calculate L50 from the global saccade rate. While we acknowledge that there might be other means to es?mate the latency of saccadic inhibi?on (Bompas et al, 2023), we used L50 because of its consistent use in other studies Stampe, 2002, 2004;Rolfs et al, 2008;Khademi et al, 2023), and also because it does a good job in capturing the drop in saccade likelihood across condi?ons (see, for example, Fig. 7 later in Results).…”

Neural Signatures of Evidence Accumulation Encode Subjective Perceptual Confidence Independent of Performance

“…As a result there will be partial independence between RT and evidence-accumulation rate, so that controlling for RT does not necessarily fully control for decision time. The best approach for modeling nondecision time (e.g., as fixed or variable across trials) remains an issue of debate (Bompas et al, 2023;Weindel et al, 2021). Moreover, whether the PEB results in RT changes or not is still a matter of debate, as some studies do observe the RT effect alongside confidence changes in the PEB (Ko et al, 2022;Koizumi et al, 2015), whereas others do not (Maniscalco et al, 2021;Rollwage et al, 2020;Samaha & Denison, 2022).…”

Neural Signatures of Evidence Accumulation Encode Subjective Perceptual Confidence

“…As a result, there will be partial independence between RT and evidence accumulation rate such that controlling for RT does not necessarily fully control for decision time. The best approach for modeling non-decision time (e.g., as fixed or variable across trials) remains an issue of debate (Bompas et al, 2023;Weindel et al, 2021) A second possibility to explain the persistent link between CPP slope and confidence after controlling for RT is that post-decisional factors influence confidence as well. Many recent models propose that evidence continues to accumulate after a decision is made and that confidence is determined by the amount of choice-consistent or choice-inconsistent post-decision evidence (Desender et al, 2019;Desender, Ridderinkhof, et al, 2021;Navajas et al, 2016;Pleskac & Busemeyer, 2010;Yu et al, 2015).…”