Dissociating sensory, decisional, and metacognitive noise in perceptual decision making

Perceptual decisions are corrupted by sensory, decisional, and metacognitive noise. However, these noise types are typically confounded and dissociating them remains a central challenge for perceptual decision-making research. Here we introduce a simple approach in which a computational model applied to data from two different tasks can quantify these sources of noise in units of the physical stimulus. We first used the performance on a 2AFC task – a task thought to minimize decisional noise – to estimate an upper bound for the sensory noise associated with the visual stimuli. We then used this estimate in a separate discrimination task to compute a lower bound for decisional noise, defined as the trial-to-trial variability of the internal decision criterion. Finally, we isolated metacognitive noise by measuring the additional placement variability of confidence criteria after accounting for sensory and decisional noise. Across two experiments using different stimuli and tasks, we found that sensory and decisional noise were of comparable magnitude, whereas metacognitive noise was markedly lower. Importantly, the magnitude of metacognitive noise increased for confidence criteria further from the decision criterion, confirming a critical prediction of the lognormal meta noise model of metacognition. Overall, our results demonstrate the possibility of dissociating different sources of noise in perceptual decision making and provide a quantitative comparison on how strongly each noise source corrupts perceptual decisions.