Abstract
Metacognition, the capacity to monitor and control our own mental processes, is thought to automatically and effortlessly accompany decision-making. This assumption underlies all models of metacognition, which were developed using experiments involving hundreds of trials with no incentives to motivate metacognitive performance. However, evidence shows that many other cognitive functions are mentally demanding, and the costs of cognitive effort lead people to disengage in effortful tasks unless suitably motivated. Here, we propose that metacognition is mentally demanding, leading people to employ heuristics to avoid metacognitive effort. We develop a flexible, effort-based decision-making paradigm that allows participants to trade-off rewards for reduced metacognitive effort. We operationalise metacognitive effort as the precision of confidence judgments, participants can maintain more confidence criteria (higher precision) at the expense of mental exertion. Critically, our design incentivises metacognition while controlling for fatigue and the effortfulness of the primary task. Across three experiments, we demonstrate that individuals sacrifice rewards to avoid metacognitive effort. Then, an important question arises: is effort a practical concern for understanding metacognition? To illustrate the consequences of metacognitive effort we take a famous observation in metacognition research—confidence leaks (correlations in confidence ratings for independent tasks performed in close temporal proximity). Introducing incentives into a confidence leak paradigm, we show that confidence leaks are significantly reduced when metacognition is rewarded. This indicates that confidence leaks, previously considered intrinsic to metacognition, can emerge due to a lack of incentives for investing metacognitive effort. Our research shows that metacognition is sensitive to effort and value, challenging the assumption that it automatically accompanies decision-making. We demonstrate that metacognition is effortful and when effort is counteracted by reward, metacognitive inefficiencies are reduced. Our model proposes that these inefficiencies may be rational heuristics by a system that faces substantial computational demands with few incentives to exert metacognitive effort.