Introspection and decision-making: value-based decision-making and confidence computations

National Institute Of Mental Health

Investigators

Abstract

During the fiscal year, we have been working on 5 main aspects of this project, totaling over 855 online datasets: 1) we have completed a study aimed at the development of a novel computational model for metacognition in value-based decisions. We tested this model in 149 participants who performed computerized tasks with trial-by-trial confidence reports, two with perceptual choices (visual orientation discrimination with high and low contrast volatility) and two with value-based choices (risky decision-making and delay discounting). Our model provides a good fit to the data and reveals that our computational measure of metacognitive ability is strongly conserved within different perceptual tasks and within value-based tasks, and moderately correlated across perceptual and value-based tasks. This study has resulted in a manuscript that is currently under review (Plate CR, Govil D, Zheng CY, Boundy-Singer ZM, Ziemba CM, Lopez-Guzman S. Computational characterization of metacognitive ability in subjective decision-making. bioRxiv [Preprint]. 2025). 2) we have collected data on 41 adult participants who have completed a computerized task with confidence reports on risky decisions at home and in the lab. We have found that our task measures show very good test-retest reliability and remain stable across the two contexts in which the tasks were performed. We have now adapted this task for a youth sample and so far, have collected data on 41 participants with and without diagnoses of anxiety and/or depression. Our goal is to characterize metacognition and risky decision-making as it relates to psychopathology. 3) we have collected and analyzed data from 182 adult online participants in two studies designed to investigate the effect of monetary incentives on confidence reports. We have found when monetary incentives bias trial-by-trial confidence reports in a perceptual decision-making task. We also have measured risk preferences in these participants and found that individual’s confidence reports track the estimated expected utility and not the expected value of the incentive, suggesting risk preferences may interact with self-reported confidence. We are now preparing a manuscript with these results (Raymond C, Govel D, Lopez-Guzman S, Ziemba C, 2025, in preparation) 4) we have analyzed two datasets (n=43 and n=460) in which adult participants completed a risky decision-making task and a delay discounting task and found a correlation in the computational model’s estimate of choice randomness. This means that the level of choice randomness is conserved across different decisions and suggests it may be a trait-like measure. Further computational modeling indicates this may be related to cognitive imprecision, proposing a new interpretation for this randomness trait. This corresponding manuscript is now in preparation (Haeffner C, Barretto-Garcia M, Hartley C, Lopez-Guzman S, 2025 In preparation). 5) In an online study, we have explored the effect of engaging in metacognition on performance in an attentional blink task. Our preliminary results suggest that engaging in metacognition makes choices more conservative in the visual attention task, increasing the decision criterion for detecting the target. Moreover, we find that individual differences in metacognitive ability correlate with decision sensitivity in the baseline attentional blink task. This work indicates that metacognition and visual attention may be related. Taken together these preliminary results give support to our novel computational methods for studying decision-making and metacognition, and for the use of our model-derived parameters as meaningful individual difference metrics, with high internal validity, and good reliability. With these promising validation studies, we are launching future studies that will utilize these tasks and computational approaches in clinical samples.