Dual strategies in human confidence judgments

Abstract

Although confidence is commonly believed to be an essential element in decision-making, it remains unclear what gives rise to one's sense of confidence. Recent Bayesian theories propose that confidence is computed, in part, from the degree of uncertainty in sensory evidence. Alternatively, observers can use physical properties of the stimulus as a heuristic to confidence. In the current study, we developed ideal observer models for either hypothesis and compared their predictions against human data obtained from psychophysical experiments. Participants reported the orientation of a stimulus, and their confidence in this estimate, under varying levels of internal and external noise. As predicted by the Bayesian model, we found a consistent link between confidence and behavioral variability for a given stimulus orientation. Confidence was higher when orientation estimates were more precise, for both internal and external sources of noise. However, we observed the inverse pattern when comparing between stimulus orientations: although observers gave more precise orientation estimates for cardinal orientations (a phenomenon known as the oblique effect), they were more confident about oblique orientations. We show that these results are well explained by a strategy to confidence that is based on the perceived amount of noise in the stimulus. Altogether, our results suggest that confidence is not always computed from the degree of uncertainty in one's perceptual evidence but can instead be based on visual cues that function as simple Heuristics to confidence.

Suprathreshold perceptual decisions constrain models of confidence

“…All four possibilities (i.e, a probability metric, SNR-scaling, use of a centred-prior, heuristic cue use; see Fig 1) were supported by at least one observer. Taken together, these results suggest that the computation of confidence is highly idiosyncratic in environments where decision uncertainty is influenced by multiple factors, further supporting the hypothesis of a highly individual nature to perceptual confidence [42][43][44], which may also depend on the task at hand [32,35]. However, within the metric types, there was a preference…”

“…Heuristic observers tended to give more weight to the number of dots displayed than the other observers, with 5/7 observers giving almost equal or more weight to this predictor than the position of the dot cloud (i.e., stimulus strength). It is unclear why the number of dots was particularly salient to these observers in this task, given the amount of previous evidence suggesting that variability, here dot-cloud spread, is a strong heuristic cue [ 14 , 33 – 35 ]. However, to our knowledge, no other perceptual confidence task has jointly manipulated both the quantity and quality of sensory information.…”

“…While in some cases, the heuristic label has been applied to variants of the Evidence-Strength metric [ 12 , 19 , 32 ], the majority of “heuristic” confidence models have focused on the use of stimulus cues to infer decision uncertainty. Support for this method of computing confidence comes from a series of studies demonstrating that observers over- or under-weight external noise in the stimulus when reporting confidence [ 14 , 33 – 35 ]. That is, they display a dissociation between Type 1 and Type 2 performance such that, if perceptual performance is matched for two stimuli with different levels of external noise, confidence is not equated.…”

“…One approach has been to compare the behavioural signatures of competing models. Studies using this approach have tended to compare a Probability metric with a Heuristic metric, finding support for the probability metric [ 10 , 36 ] or differing support for each metric depending on the perceptual task [ 35 ]. Formal model comparisons, on the other hand, have been focused on Probability versus Evidence-Strength metrics with mixed results (note that this is according to our metric definitions not those of the authors who used the term “heuristic” more liberally).…”

Visual metacognition: Measures, models, and neural correlates.

“…Ideally, confidence should be based on a Bayesian computation of the posterior probability of being correct and multiple models assume that such computations underlie confidence judgments (Aitchison et al, 2015; Fleming & Daw, 2017; Meyniel et al, 2015). Nevertheless, it is still unclear whether people can actually perform the complex computations required to estimate the posterior probability of being correct and this is a topic of active research (Bertana et al, 2021).…”