Metacognition & Control

Metacognition refers to the ability to accurately judge one’s own cognitive processes (monitoring), and to adjust one’s cognitive processes based on monitoring (control). Metacognition research has extensively investigated metacognitive monitoring through confidence judgments and its impact on performance. Recent investigations however show that performance is not only affected by the accuracy of monitoring but more importantly by the way monitoring judgments are translated into metacognitive control decisions. Building on recent findings about the variability in metacognitive control and the strong relation between metacognitive control and academic achievement, a preregistered longitudinal experiment investigated how participants differ in translating monitoring judgments into metacognitive control decisions, and whether time affects this translation. Participants completed a metacognition task at 3 measurement points over 3 weeks. Metacognitive skills including monitoring (confidence judgments), control (restudy selections), and task performance were examined through a series of Bayesian generalized linear mixed models (Bayesian GLMMs). Results showed that restudying task items for which low monitoring judgments were made improved performance. However, individual differences emerged: some participants selected items for restudy for which their monitoring judgments were low, while others selected items for restudy for which their monitoring judgments were high. These differences decreased over time, with participants more consistently selecting items with lower monitoring judgments in later sessions. Also, the probability of restudying decreased over time. The findings shed light on the importance of metacognitive control when investigating metacognition. Future research could further examine why people differ in how they use their monitoring judgments to make control decisions.

Planning is a crucial capacity for making sequential decisions in everyday life. However, people's planning strategies often deviate from the optimal approach. One possible explanation for suboptimal planning is the constraint imposed by metacognitive ability, which is closely linked to the accuracy of evaluating uncertainty in the task environment. In five experiments, we explored the relationship between metacognition and planning by asking participants to make plans in a multi-armed bandit task and rate their confidence in the final decisions. We characterized planning ability using the metalevel Markov Decision Process (MDP) framework and the Softmax choice model (Callaway et al., 2022), and quantified metacognitive ability as the correlation between actual confidence and the optimal confidence generated by an ideal observer, who is assumed to rely on the Bayesian updating rule to construct the posterior distribution of reward probability for each bandit based on the planning process. Results from our experiments showed that metacognitive ability reliably predicted planning ability, particularly the ability to optimally terminate the planning process. This relationship persisted when the working memory constraint in the task was removed. Additionally, direct training in planning strategy did not improve metacognitive evaluation, suggesting that metacognitive ability is stable and not merely a consequence of enhanced planning skills. Further computational modeling analysis indicated that individual differences in planning depth (the extent to which future rewards were discounted) could partly explain the relationship between metacognition and planning. These findings highlight the critical role of metacognition as a predictor of efficient planning.

Cognitive control allows us to overcome automatic processing to enable goal-directed behavior. Doing so requires time for control processes to overcome the influence of faster, automatic processing. However, most research on conflict resolution processes measures free response times (RTs), making it difficult to disentangle how the contributions of automatic and goal-directed processing evolve over time to produce behavior. To overcome this limitation, the forced-response method can be employed to trace out the time course of conflict resolution processes. Here, we apply a new response preparation model (RPM) to data from forced-response conflict tasks to disentangle the component processes of cognitive control. This model estimates the latency of habitual and goal-directed response preparation. We show that conceptualizing the conflict resolution process in terms of the underlying competing responses can account for congruency effects in the Simon and flanker tasks. We also demonstrate that the RPM can successfully model control adjustments that take place over long and short timescales (e.g., the proportion congruency and the congruency sequence effect). Furthermore, we evaluate RPM’s performance through parameter recovery. Finally, we show how evidence accumulation models (EAMs) for conflict tasks can account for data obtained using the forced-response method and how parameters from conflict EAMs map onto RPM through a cross-fitting study. These results suggest that the RPM may be fruitful for examining control adjustments that regulate the competition between automatic and goal-directed processes, and that data obtained using the forced-response method can be used to distinguish between models of conflict resolution.

Many tasks require sustained attention and result in unwanted consequences when vigilance suffers. Vigilance is usually assumed to suffer due to mental fatigue, resulting in performance decrements, such as increased errors and longer reaction time (Warm, Finomore, Vidulich, & Funke, 2015). A common theory holds that mental fatigue is the result of depletion of resources (Grier, et al., 2003). However, some propose that the performance decrement is due to mind wandering (Zanesco, Denkova, Barry, & Jha, 2024), a shift in strategy (Jongman & Taatgen, 2020), or a proactive shift in behavior to conserve resources (Rubinstein, 2020). Analysis of a data set including 67 participants performing three naturalistic tasks related to target verification, strategic decision making, and image recognition in three consecutive days found evidence of a vigilance decrement in the hit rate of perceptual tasks. However, the expected decrement was accompanied by unexpected performance improvements in other measures, such as reaction time and correct rejections. In addition, the image recognition task showed evidence of performance decrement across days, which would be inconsistent with fatigue. We propose that this surprising combination of performance decrement and performance improvement is due to a complex combination of fatigue, interference, and strategic learning. To investigate this proposal, we use hierarchical modeling of performance in each task, using metrics from task performance data and from physiological data. Understanding which cognitive mechanisms influence performance can facilitate effective interventions for improving vigilance in real-world tasks.