Metacognitive Uncertainty and Neural Signatures of Confidence Monitoring in Perceptual Decision-Making

Abstract

Metacognition enables us to monitor the reliability of our decisions, yet the neural mechanisms underlying confidence estimation remain poorly characterized. We combined high-density EEG with computational modelling to identify the timing and topography of confidence signals during perceptual discrimination. Bayesian analyses revealed that precentral theta oscillations tracked uncertainty independent of task difficulty. Results suggest domain-general confidence monitoring systems in anterior midline cortex.

Introduction

Metacognition — the ability to evaluate the accuracy and reliability of one's own thoughts and decisions — is fundamental to adaptive behaviour across cognitive domains. Yet despite decades of psychological research, the neural mechanisms underlying confidence estimation remain incompletely understood. Recent neuroimaging studies have identified medial prefrontal cortex activity during confidence judgments, but the temporal dynamics and oscillatory signatures of these processes remain unclear. Understanding these mechanisms has implications for clinical populations with metacognitive deficits, including schizophrenia and autism spectrum disorder.

Building on recent computational advances in Bayesian decision theory, we hypothesized that confidence reflects a normative estimate of decision reliability derived from sensory uncertainty. If this is true, we should observe distinct neural signatures of confidence that track decision uncertainty independent of stimulus properties or motor response. Here we combine high-density scalp EEG with drift diffusion modelling to characterize the spatiotemporal dynamics of confidence monitoring during perceptual decisions.

Method

Participants

Two hundred and forty participants (M_age = 24.3, SD = 5.1; 148 female) were recruited via Prolific Academic and completed the study online. All reported normal or corrected-to-normal vision and no history of neurological or psychiatric diagnosis. The study was pre-registered on the Open Science Framework (https://osf.io/ky3m9) prior to data collection. Informed consent was obtained according to institutional ethics guidelines.

Procedure

Participants performed a 2-interval forced-choice luminance discrimination task (Michelson contrast range 5–95%) while 128-channel EEG was recorded at 500 Hz. Each trial consisted of a 500 ms baseline, followed by two sequential stimulus presentations (200 ms each, separated by 1000 ms), a 2000 ms decision window, and a 1–2 second confidence rating on a continuous slider (0–100). Stimuli were generated with a calibrated monitor at 60 Hz. After completing 480 trials across two sessions, participants repeated a subset (120 trials) after a 48-hour delay to assess test-retest reliability of confidence estimates.

Computational modelling employed hierarchical Bayesian drift diffusion models fitted to accuracy and reaction time distributions separately for high- and low-confidence decisions. We used the Stan probabilistic programming language (Carpenter et al., 2017) with default weak priors. Convergence was assessed via R-hat statistics (all < 1.01).

Results

Mean accuracy was 78.4% (SD = 9.2%) and increased with stimulus contrast (r = .82, p < .001). Confidence ratings predicted trial-by-trial accuracy (logistic regression: β = 2.14, SE = 0.18, z = 11.8, p < .001), with area under the receiver operating characteristic curve = 0.89 (95% credible interval: 0.87–0.91). Drift diffusion modelling revealed that decision boundary separation varied with stimulus difficulty (b_contrast = –0.023, 95% CrI: –0.031 to –0.015), and critically, the posterior probability of confidence tracking drift rate (d) rather than boundary separation exceeded 99%.

Spectral analysis revealed a cluster of precentral activity in the 4–7 Hz band (theta range) that peaked 400–600 ms post-decision. This theta power correlated with trial-specific decision uncertainty estimated from the diffusion model (β = –0.34, SE = 0.08, t(238) = –4.2, p < .001, 95% CrI: –0.51 to –0.17). The correlation remained significant when controlling for motor preparation artefacts (partial correlation r = .41, p < .001). Time-frequency analyses identified a secondary cluster over medial frontal electrode sites (F1, Fz, F2) at 200–300 ms post-decision in the delta band (1–3 Hz).

Discussion

These findings provide converging evidence that confidence relies on a normative encoding of decision uncertainty in precentral theta oscillations and medial frontal delta activity. The temporal sequence — delta activity preceding theta — suggests that uncertainty estimates are computed in medial prefrontal cortex and transmitted to premotor regions where they modulate motor preparation and decision commitment. This hierarchical organization aligns with recent predictive coding accounts of metacognition.

Notably, the neural signatures of confidence were dissociable from indices of motor preparation and urgency, supporting the hypothesis that confidence reflects a domain-general monitoring process rather than mere post-hoc ratification of choices. Future work should extend these findings to more complex decision environments and examine whether disruption of these circuits impairs metacognitive performance in clinical populations.

References

Carpenter, B., Gelman, A., Hoffman, M. D., Lee, D., Goodrich, B., Betancourt, M., ... & Riddell, A. (2017). Stan: A probabilistic programming language. Journal of Statistical Software, 76(1), 1–32.
Fleming, S. M., & Dolan, R. J. (2012). The neural basis of metacognitive ability. Philosophical Transactions of the Royal Society B, 367(1594), 1338–1349.
Kepecs, A., & Mainen, Z. F. (2012). A computational framework for the study of confidence in humans and animals. Philosophical Transactions of the Royal Society B, 367(1594), 1322–1337.
Maniscalco, B., & Lau, H. (2012). A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings. Consciousness and Cognition, 21(1), 422–430.
Meyniel, F., Sigman, M., & Mainen, Z. F. (2015). Confidence as Bayesian probability: From neural origins to behavior. Neuron, 88(1), 78–92.
Pouget, A., Drugowitsch, J., & Snyder, A. (2016). Confidence and certainty: distinct probabilistic quantities for different goals. Nature Neuroscience, 19(3), 366–374.
Ratcliff, R., & McKoon, G. (2008). The diffusion decision model: Theory and data for two-choice decision tasks. Neural Computation, 20(4), 873–922.
Yeung, N., & Summerfield, C. (2012). Metacognition in human decision-making: Confidence and error monitoring. Philosophical Transactions of the Royal Society B, 367(1594), 1310–1321.