Metacognition: ideas and insights from neuro- and educational sciences

In this article, we reviewed the literature on metacognition in educational sciences and cognitive neuroscience with the aim to investigate gaps in current research and propose ways to address them through the exchange of insights between the two disciplines and interdisciplinary approaches. The main aspects analysed were operational definitions of metacognition and metacognitive training, through the lens of metacognitive knowledge and metacognitive control. Our review also highlighted an additional construct in the form of the distinction between online metacognition (on the fly and largely automatic) and offline metacognition (slower, reflective and requiring meta-representations). In cognitive neuroscience, research has focused on metacognitive judgements (mainly offline) and EF (mainly online). Metacognition is operationalised with tasks carried out in the lab and are mapped onto brain functions. In contrast, research in educational sciences typically measures metacognition in the context of learning activities, mostly in schools and universities. More recently, EF has been studied in educational settings to investigate its role in academic achievement and whether training it may benefit learning. Evidence on the latter is however mixed. Regarding metacognitive training in general, evidence from both disciplines suggests that interventions fostering learners’ self-reflection and knowledge of their learning behaviour (i.e., offline meta-knowledge) may best benefit them and increase academic achievement.

We focused on four aspects of research that could benefit from an interdisciplinary approach between the two areas: (i) validity and reliability of research protocols, (ii) under-researched dimensions of metacognition, (iii) metacognitive training, and (iv) domain-specificity vs. domain generality of metacognitive abilities. To tackle these issue, we propose four avenues for integrated research: (i) investigate the degree to which different protocols relate to similar or different metacognitive constructs, (ii) implement designs and perform experiments to identify neural substrates necessary for offline meta-control by for example borrowing protocols used in educational sciences, (iii) study the effects of (offline) meta-knowledge training on the brain, and (iv) perform developmental research in the metacognitive brain and compare it with the existing developmental literature in educational sciences regarding the domain-generality of metacognitive processes and metacognitive abilities.

First, neurocognitive research on metacognitive judgements has developed robust operationalisations of offline meta-knowledge. However, these operationalisations often consist of specific tasks (e.g., 2-AFC) carried out in the lab. These tasks are often very narrow and do not resemble the challenges and complexities of behaviours associated with learning in schools and universities. Thus, one may question to what extent they reflect real-life metacognition, and to what extent protocols developed in educational sciences and cognitive neuroscience actually operationalise the same components of metacognition. We propose that comparing different protocols from both disciplines that are, a priori, operationalising the same types of metacognitive processes can help evaluate the ecological validity of protocols used in cognitive neuroscience, and allow for more holistic assessments of metacognition, provided that it is clear which protocol assesses which construct. Degrees of correlation between different protocols, within and between disciplines, may allow researchers to assess to what extent they reflect the same metacognitive constructs and also identify what protocols are most appropriate to study a specific construct. For example, a relation between meta-d′ metacognitive sensitivity in a 2-AFC task and the meta-knowledge subscale of the MAI, would provide external validity to the former. Moreover, educational scientists would be provided with bias-free tools to assess metacognition. These tools may enable researchers to further investigate to what extent metacognitive bias, sensitivity and efficiency each play a role in education settings. In contrast, a low correlation may highlight a difference in domain between the two measures of metacognition. For instance, metacognitive judgements in brain research are made in isolated behaviour, and meta-d’ can thus be viewed to reflect “local” metacognitive sensitivity. It is also unclear to what extent processes involved in these decision-making tasks cover those taking place in a learning environment. When answering self-reported questionnaires, however, subjects make metacognitive judgements on a large set of (learning) activities, and the measures may thus resemble more “global” or domain-general metacognitive sensitivity. In addition, learners in educational settings tend to receive feedback — immediate or delayed — on their learning activities and performance, which is generally not the case for cognitive neuroscience protocols. Therefore, investigating metacognitive judgements in the presence of performance or social feedback may allow researchers to better understand the metacognitive processes at play in educational settings. Devising a global measure of metacognition in the lab by aggregating subjects’ metacognitive abilities in different domains or investigating to what extent local metacognition may affect global metacognition could improve ecological validity significantly. By investigating the neural correlates of educational measures of metacognition, researchers may be able to better understand to what extent the constructs studied in the two disciplines are related. It is indeed possible that, though weakly correlated, the meta-knowledge scale of the MAI and meta-d’ share a common neural basis.

Second, our review highlights gaps in the literature of both disciplines regarding the research of certain types of metacognitive processes. There is a lack of research in offline meta-control (or strategic regulation of cognition) in neuroscience, whereas this construct is widely studied in educational sciences. More specifically, while there exists research on EF related to planning, common experimental designs make it hard to disentangle online from offline metacognitive processes. A few studies have implemented subject reports (e.g., awareness of error or desire for reminders) to pin-point the neural substrates specifically involved in offline meta-control and the current evidence points at a role of the lPFC. More research implementing similar designs may clarify this construct. Alternatively, researchers may exploit educational sciences protocols, such as self-report questionnaires, learning journals, metacognitive prompts and feedback to investigate offline meta-control processes in the brain and their relation to academic proficiency and achievement.

Third, there is only one study known to us on the training of meta-knowledge in the lab. In contrast, meta-knowledge training in educational sciences have been widely studied, in particular with metacognitive prompts and learning journals, although a systematic review would be needed to identify the benefits for learning. Relative to cognitive neuroscience, studies suggest that offline meta-knowledge trained in and outside the lab (i.e., metacognitive judgements and meditation, respectively) transfer to meta-knowledge in other lab tasks. The case of meditation is particularly interesting since meditation has been demonstrated to beneficiate varied aspects of everyday life. Given its importance for efficient regulation of cognition, training (offline) meta-knowledge may present the largest benefits to academic achievement. Hence, it is important to investigate development in the brain relative to meta-knowledge training. Evidence on metacognitive training in educational sciences tends to suggest that offline metacognition is more “plastic” and may therefore benefit learning more than online metacognition. Furthermore, it is important to have a good understanding of the developmental trajectory of metacognitive abilities — not only on a behavioural level but also on a neural level — to identify critical periods for successful training. Doing so would also allow researchers to investigate the potential differences in terms of plasticity that we mention above. Currently, the developmental trajectory of metacognition is under-studied in cognitive neuroscience with only one study that found an overlap between the neural correlates of metacognition in adults and children. On a side note, future research could explore the potential role of genetic factors in metacognitive abilities to better understand to what extent and under what constraints they can be trained.

Fourth, domain-specific and domain-general aspects of metacognitive processes should be further investigated. Educational scientists have studied the development of metacognition in learners and have concluded that metacognitive abilities are domain-specific at the beginning (meaning that their quality depends on the type of learning activity, like mathematics vs. writing) and progressively evolve towards domain-general abilities as knowledge and expertise increase. Similarly, neurocognitive evidence points towards a common network for (offline) metacognitive knowledge which engages the different regions at varying degrees depending on the domain of the activity (i.e., perception, memory, etc.). Investigating this network from a developmental perspective and comparing findings with the existing behavioural literature may improve our understanding of the metacognitive brain and link the two bodies of evidence. It may also enable researchers to identify stages of life more suitable for certain types of metacognitive intervention.