Concepts · Action and Control

Metacognition

Metacognition allows people to evaluate and regulate aspects of their cognition. It can reveal uncertainty, misunderstanding and possible error, but it does not provide complete self-knowledge or guaranteed control.

By Yona Ole Lobulu ·

Concept9 min readD8.6

Topic
Action and Control
Read first
2 pieces should be read before this one
Reading time
About 9 minutes of reading
Difficulty
Late reading: this page sits at the far end of the Library, after many other pieces

The question

How does cognition become an object of knowledge, monitoring and regulation, and why does evaluating one's own thinking support, without guaranteeing, its correction?

Definition

Metacognition is cognition concerned with monitoring, evaluating or regulating cognition: the knowledge, judgments and regulatory processes through which thinking itself becomes an object of evaluation.

You reach the end of a paragraph and realise that you did not understand it.

Your eyes moved across the words. You recognised the sentences. You may even have continued reading without interruption. But something then changed: you evaluated what had just happened and concluded that the reading had not produced understanding.

Reading the paragraph was one cognitive activity. Judging whether you understood it was another. The activities are functionally different, but both are forms of cognition.

That second activity is an example of metacognition.

Metacognition is often described as "thinking about thinking." The phrase is memorable, but incomplete. More precisely, metacognition includes the knowledge, judgments and regulatory processes through which cognition becomes an object of evaluation. It allows you to assess whether you understand, remember, know or are performing effectively—and it can influence what you do next.

You might reread the paragraph, slow down, seek another explanation or continue despite the confusion. But recognising that you did not understand does not explain exactly why comprehension failed. Nor does it guarantee that your next attempt will work.

Metacognition makes aspects of cognition available for evaluation. It does not place an infallible observer outside the mind.

Thinking—and evaluating the thinking

Cognition includes processes such as perceiving, remembering, attending, deciding and understanding. Metacognition occurs when aspects of those processes become the subject of knowledge, monitoring or regulation.

The distinction appears in many ordinary experiences:

  • answering a question and judging how confident you are;
  • trying to remember a name and feeling that you would recognise it;
  • reading a page and evaluating whether you understood it;
  • working through a problem and noticing that your approach is not succeeding;
  • losing focus and recognising that your attention has drifted.

Foundational accounts described metacognition broadly in terms of knowledge about cognition, experiences that occur during cognitive activity, goals and the strategies used to pursue or evaluate them. Flavell, 1979

Later frameworks distinguished between the cognition being performed and the processes that monitor or regulate it. Monitoring provides information about how cognition appears to be going. Control changes how cognition proceeds—for example, by allocating more time, checking an answer or changing strategy. Nelson and Narens, 199060053-5)

These are functional distinctions, not evidence for a second mind watching the first. Metacognitive judgments are themselves cognitive products, so they remain subject to limited information, bias and error.

That is why metacognition can be both useful and wrong.

Three parts of metacognition

Metacognition can be divided into three related components: knowledge, monitoring and control.

Metacognitive knowledge

Metacognitive knowledge includes beliefs about cognitive tasks, strategies and your own performance—even when those beliefs are incomplete or mistaken.

You might believe that one study method helps you remember information or that a particular kind of task requires more concentration. Such beliefs can guide action. But a strategy can feel effective because it creates familiarity while producing little durable learning. Another can feel difficult even when it improves later performance.

Metacognitive knowledge is therefore knowledge or belief about cognition, not guaranteed insight into how cognition actually works.

Monitoring

Metacognitive monitoring involves assessing cognition or performance.

It appears in judgments such as:

  • Do I understand this?
  • Will I remember it later?
  • How certain am I?
  • Am I still paying attention?
  • Is my current approach working?
  • Was that response probably correct?

These judgments can occur before, during or after a cognitive activity. Predicting whether something will be easy to learn is different from judging confidence in an answer already given, but both involve evaluating cognition.

Control

Metacognitive control concerns what happens in response to an evaluation.

You may:

  • allocate more time;
  • check an answer;
  • reread an explanation;
  • seek additional information;
  • change strategy;
  • continue;
  • stop.

Monitoring evaluates cognition. Control changes what happens next.

The first can inform the second without determining it. You can recognise that you do not understand something and still continue without changing your approach. You can feel uncertain but lack a better strategy. You can also regulate cognition without first forming an explicit judgment about it.

Research supports treating monitoring and control as distinguishable processes that can sometimes operate independently. O'Leary and Sloutsky, 2019

Confidence is not the same as accuracy

Confidence is one of the clearest forms of metacognitive judgment.

When you answer a question, the answer is your response to the task. Your confidence expresses a judgment about the likelihood that the response is correct.

Those are different outcomes.

Two people can give the same correct answer while feeling very different levels of confidence. One may be certain; the other may believe they guessed. Two people can also feel equally confident even when one is correct and the other is not.

Researchers therefore distinguish several questions.

First, how well did the person perform?

Second, how confident were they?

Third, did their stated confidence correspond to their observed accuracy? If answers given with 80 per cent confidence are correct roughly 80 per cent of the time, confidence is well calibrated at that level.

A related but distinct question is whether a person gives higher confidence to correct answers than to errors. Someone can be generally cautious while still distinguishing their successes from their mistakes. Another person can feel highly confident almost all the time while distinguishing them poorly.

Modern measurement frameworks separate task performance, general confidence and the ability of confidence to track correctness because none can stand in for the others. Fleming and Lau, 2014

Feeling certain is an experience, not proof. But confidence is not meaningless. It can contain useful information about performance without being identical to accuracy.

How the mind estimates how it is doing

If confidence is not a direct reading of correctness, where does it come from?

Metacognitive judgments can draw on cues generated during cognition. Information may feel:

  • familiar;
  • easy to process;
  • easy to retrieve;
  • effortful;
  • conflicting.

These experiences can provide real information. An answer that comes to mind quickly may often be more likely to be correct than one that cannot be retrieved. A difficult paragraph may genuinely require more attention. Conflict between possible responses may signal uncertainty.

But no single cue proves what it appears to indicate.

Familiarity can arise because a claim has been repeated, not because it is true. Smooth processing can result from clear presentation, not accurate content. Effort may signal confusion, but it can also accompany productive learning. Easy retrieval can reflect strong memory, while inaccessible information may still be recognised later.

Research on judgments of learning shows that people estimate future memory using multiple sources of information, including properties of the material, study conditions and subjective experiences such as fluency. These cues do not always affect judgments and actual performance in the same way. Koriat, 1997

Metacognitive judgment is therefore better understood as an estimate based on available information than as direct access to an underlying cognitive process.

Its value depends on the quality of that information, how the cues are interpreted and what the judgment is expected to reveal. A cue can be genuinely useful in one context and misleading in another.

Useful access, not complete access

The fallibility of metacognition creates two opposing mistakes.

The first is to assume that people can inspect their cognition transparently. A feeling or reported reason is then treated as direct evidence of how the relevant mental process works.

The second is to conclude that people cannot know anything meaningful about their cognition.

Neither position is adequate.

People can often detect:

  • uncertainty;
  • failed retrieval;
  • weak comprehension;
  • attentional drift;
  • conflict between responses;
  • a mismatch between an expected and actual result.

These signals can be informative even when their causes remain unclear.

You may know that you have lost the thread of an argument without knowing whether the cause was fatigue, unfamiliar terminology, distraction or a missing assumption. You may recognise that an answer feels uncertain without knowing which cues produced that feeling. You may correctly detect that a strategy is failing while choosing an ineffective replacement.

Knowing that something may be wrong is different from knowing exactly why it is wrong.

Metacognition overlaps with introspection and self-awareness, but the concepts are not interchangeable. Introspection can concern emotions, bodily experiences or imagery. Self-awareness can include identity, behaviour and social position. Metacognition is more specific: it concerns cognition becoming an object of knowledge, evaluation or regulation.

Metacognitive accuracy also varies across tasks. Accurately monitoring one kind of decision does not guarantee equally accurate monitoring of memory, learning or every other cognitive domain. Evidence indicates that metacognition includes both shared and domain-specific components. Rouault et al., 2018

Metacognition offers partial and context-sensitive access—not an unrestricted capacity for knowing the mind.

Why metacognition matters for change

Change can depend not only on what a person thinks, understands or does, but also on whether that cognition becomes available for evaluation.

A person may need to detect:

  • that attention has drifted;
  • that understanding is weaker than assumed;
  • that familiarity is being mistaken for truth;
  • that confidence exceeds the available evidence;
  • that a strategy is producing the same failed result;
  • that external feedback is needed.

Metacognition can make these conditions available for deliberate response.

Recognising uncertainty can prompt checking. Detecting confusion can prompt a search for another explanation. Noticing unjustified confidence can make reconsideration possible. Realising that a strategy is not working can create an opportunity to change it.

But opportunity is the correct level of claim.

Metacognition does not supply every missing strategy. It does not create unlimited attention, emotional capacity or behavioural control. It does not remove the influence of habit, context, reward or prior belief. Even an accurate judgment may fail to change what happens next if a better response is unavailable or cannot be performed.

Its contribution is more precise:

Metacognition can make cognition available for possible revision while cognition is shaping understanding, confidence and action.

Detection is not correction. But without detection, correction may never become a deliberate possibility.

You reach the end of the paragraph and realise that you did not understand it.

That recognition does not explain what failed. It does not ensure that rereading will help. It does not place you outside your own cognition with a complete view of how it works.

But it changes the situation.

The paragraph can now be read again, approached differently or checked against another source. The failure has become something that can be responded to.

That is the value of metacognition: not perfect self-knowledge, but the possibility of looking again.

Behind this page

The claims this concept makes, the evidence behind them, and the limits it accepts.

Evidence status

High confidence

Strongly supported, though resting on synthesis or principle rather than a single decisive body of evidence.

Claims

  1. Performing a cognitive activity and evaluating that activity are functionally different, though both are forms of cognition

    Established

    What this does not assert: The founding distinction.

  2. Metacognition occurs when aspects of those processes become the subject of knowledge, monitoring or regulation

    Established

    What this does not assert: Level distinction.

  3. Judging confidence in an answer is a metacognitive act distinct from producing the answer

    Established

    What this does not assert: Everyday instance.

  4. Feeling that an unretrieved name would be recognised is a metacognitive experience

    Established

    What this does not assert: Feeling-of-knowing.

  5. Noticing that an approach to a problem is not succeeding is a metacognitive judgment

    Established

    What this does not assert: Performance monitoring.

  6. Recognising that attention has drifted is a metacognitive judgment about an ongoing process

    Established

    What this does not assert: Depends on attention as selection (D6.3).

  7. Foundational accounts described metacognition in terms of knowledge about cognition, cognitive experiences, goals and strategies

    Established

    What this does not assert: Flavell, 1979.

  8. Later frameworks distinguish the cognition being performed from the processes that monitor or regulate it

    Established

    What this does not assert: Nelson and Narens, 1990.

  9. Monitoring provides information about how cognition appears to be going

    Established

    What this does not assert: Definition of monitoring.

  10. Control changes how cognition proceeds, for example by allocating more time, checking or changing strategy

    Established

    What this does not assert: Definition of control.

  11. Monitoring and control are functional distinctions, not evidence of a second mind observing the first

    Canonical inference

    What this does not assert: Guards against homuncular reading.

  12. Metacognition includes the knowledge, judgments and regulatory processes through which cognition becomes an object of evaluation

    Established

    What this does not assert: Working definition.

  13. Metacognitive judgments are themselves cognitive products and remain subject to limited information, bias and error

    High confidence

    What this does not assert: Central limitation.

  14. Metacognition can be both useful and wrong

    Canonical inference

    What this does not assert: Both halves are load-bearing.

  15. Metacognition can be divided into knowledge, monitoring and control

    Established

    What this does not assert: Standard componential division.

  16. Metacognitive knowledge includes beliefs about cognitive tasks, strategies and one's own performance

    Established

    What this does not assert: Beliefs may be incomplete or mistaken.

  17. Beliefs about how learning works can guide action

    High confidence

    What this does not assert: Knowledge feeds control.

  18. A strategy can feel effective because it creates familiarity while producing little durable learning

    Established

    What this does not assert: Fluency illusion.

  19. A strategy can feel difficult while improving later performance

    Established

    What this does not assert: Difficulty is not a reliable failure signal.

  20. Metacognitive knowledge is knowledge or belief about cognition, not guaranteed insight into how cognition works

    Canonical inference

    What this does not assert: Boundary on the knowledge component.

  21. Metacognitive monitoring involves assessing cognition or performance

    Established

    What this does not assert: Definition.

  22. Monitoring appears in judgments of understanding, future memory, certainty, attention and correctness

    Established

    What this does not assert: Judgment classes.

  23. 'Thinking about thinking' is memorable but incomplete as a definition

    Canonical inference

    What this does not assert: It omits knowledge and control.

  24. Metacognitive judgments can occur before, during or after a cognitive activity

    Established

    What this does not assert: Temporal positions.

  25. Predicting ease of learning differs from judging confidence in an answer already given

    Established

    What this does not assert: Distinct judgment types.

  26. Both prospective and retrospective judgments involve evaluating cognition

    High confidence

    What this does not assert: Shared functional class.

  27. Metacognitive control concerns what happens in response to an evaluation

    Established

    What this does not assert: Definition.

  28. Control responses include allocating time, checking, rereading, seeking information, changing strategy, continuing or stopping

    Established

    What this does not assert: Response repertoire.

  29. Monitoring evaluates cognition while control changes what happens next

    Established

    What this does not assert: The functional contrast.

  30. Monitoring can inform control without determining it

    High confidence

    What this does not assert: Key boundary.

  31. A person can recognise a failure of understanding and continue without changing approach

    High confidence

    What this does not assert: Monitoring without control.

  32. A person can feel uncertain while lacking a better strategy

    Canonical inference

    What this does not assert: Control depends on an available alternative.

  33. Cognition can be regulated without an explicit prior judgment about it

    High confidence

    What this does not assert: Control without explicit monitoring.

  34. Metacognition allows a person to assess whether they understand, remember, know or are performing effectively

    Established

    What this does not assert: Functional scope.

  35. Research supports treating monitoring and control as distinguishable processes that can sometimes operate independently

    Established

    What this does not assert: O'Leary and Sloutsky, 2019.

  36. Confidence is one of the clearest forms of metacognitive judgment

    Established

    What this does not assert: Most studied measure.

  37. An answer is a response to the task while confidence is a judgment about the likelihood that the response is correct

    Established

    What this does not assert: Two different outcomes.

  38. Two people can give the same correct answer with very different confidence

    Established

    What this does not assert: Confidence is not entailed by correctness.

  39. Two people can feel equally confident when only one is correct

    Established

    What this does not assert: Correctness is not entailed by confidence.

  40. Task performance, stated confidence and the correspondence between them are separate questions

    Established

    What this does not assert: Measurement separation.

  41. Confidence is well calibrated when answers given with a stated confidence are correct at approximately that rate

    Established

    What this does not assert: Definition of calibration.

  42. Whether a person assigns higher confidence to correct than incorrect answers is a distinct question from overall confidence level

    Established

    What this does not assert: Discrimination ≠ bias.

  43. A person can be generally cautious while still distinguishing successes from mistakes

    Established

    What this does not assert: Low bias, good discrimination.

  44. A person can be highly confident throughout while discriminating poorly

    Established

    What this does not assert: High bias, poor discrimination.

  45. Metacognitive evaluation can influence what a person does next

    High confidence

    What this does not assert: Influence, not determination.

  46. Modern measurement frameworks separate performance, confidence level and confidence's ability to track correctness

    Established

    What this does not assert: Fleming and Lau, 2014.

  47. Feeling certain is an experience rather than proof of correctness

    Canonical inference

    What this does not assert: Core corrective.

  48. Confidence is not meaningless and can carry useful information about performance

    High confidence

    What this does not assert: Guards against the opposite error.

  49. Metacognitive judgments can draw on cues generated during cognition

    Established

    What this does not assert: Cue-utilisation account.

  50. Cues include familiarity, ease of processing, ease of retrieval, effort and conflict

    Established

    What this does not assert: Principal cue classes.

  51. An answer retrieved quickly may often be more likely to be correct than one that cannot be retrieved

    High confidence

    What this does not assert: Cues carry genuine information.

  52. Conflict between possible responses can signal uncertainty

    High confidence

    What this does not assert: Conflict as an informative cue.

  53. No single cue proves what it appears to indicate

    High confidence

    What this does not assert: Cue diagnosticity is conditional.

  54. Familiarity can arise from repetition rather than truth

    Established

    What this does not assert: Repetition can inflate belief.

  55. Smooth processing can result from clear presentation rather than accurate content

    Established

    What this does not assert: Fluency ≠ truth.

  56. Recognising that comprehension failed does not explain why it failed

    Canonical inference

    What this does not assert: Detection ≠ diagnosis.

  57. Effort may signal confusion but can also accompany productive learning

    Established

    What this does not assert: Effort is ambiguous.

  58. Easy retrieval can reflect strong memory while inaccessible information may still be recognised later

    Established

    What this does not assert: Accessibility ≠ availability.

  59. People estimate future memory using multiple sources including material properties, study conditions and subjective experience

    Established

    What this does not assert: Koriat, 1997.

  60. Cues do not always affect judgments and actual performance in the same way

    Established

    What this does not assert: Source of systematic miscalibration.

  61. Metacognitive judgment is better understood as an estimate from available information than as direct access to a cognitive process

    High confidence

    What this does not assert: Canonical framing.

  62. The value of a metacognitive judgment depends on cue quality, cue interpretation and what the judgment is expected to reveal

    High confidence

    What this does not assert: Three dependencies.

  63. A cue can be genuinely useful in one context and misleading in another

    High confidence

    What this does not assert: Context sensitivity.

  64. Assuming transparent inspection of one's own cognition is an error

    High confidence

    What this does not assert: First opposing mistake.

  65. Treating a reported reason as direct evidence of the underlying process is unwarranted

    Established

    What this does not assert: Introspective reports can be reconstructive.

  66. Concluding that people can know nothing meaningful about their cognition is also an error

    High confidence

    What this does not assert: Second opposing mistake.

  67. Recognising a failure does not guarantee that the next attempt will succeed

    Canonical inference

    What this does not assert: Detection ≠ correction.

  68. People can often detect uncertainty, failed retrieval, weak comprehension, attentional drift, response conflict and mismatched expectations

    Established

    What this does not assert: Detectable signals.

  69. These signals can be informative even when their causes remain unclear

    High confidence

    What this does not assert: Partial access is still access.

  70. A person may know they have lost the thread of an argument without knowing why

    Canonical inference

    What this does not assert: Detection without diagnosis.

  71. A person may correctly detect that a strategy is failing while choosing an ineffective replacement

    High confidence

    What this does not assert: Accurate monitoring, poor control.

  72. Knowing that something may be wrong differs from knowing exactly why it is wrong

    Canonical inference

    What this does not assert: Canonical distinction.

  73. Metacognition, introspection and self-awareness overlap but are not interchangeable

    High confidence

    What this does not assert: Concept boundaries.

  74. Introspection can concern emotions, bodily experiences or imagery

    High confidence

    What this does not assert: Wider than cognition.

  75. Self-awareness can include identity, behaviour and social position

    High confidence

    What this does not assert: Different object.

  76. Metacognition specifically concerns cognition becoming an object of knowledge, evaluation or regulation

    High confidence

    What this does not assert: Narrowest of the three.

  77. Metacognitive accuracy varies across tasks

    Established

    What this does not assert: No single global faculty.

  78. Metacognition makes aspects of cognition available for evaluation without placing an infallible observer outside the mind

    Canonical inference

    What this does not assert: Canonical thesis.

  79. Accurate monitoring in one domain does not guarantee accurate monitoring in another

    Established

    What this does not assert: Limits generalisation.

  80. Evidence indicates metacognition includes both shared and domain-specific components

    Established

    What this does not assert: Rouault et al., 2018.

  81. The relative weight of domain-general and domain-specific metacognitive components is not settled

    Contested

    What this does not assert: Active debate.

  82. Metacognition offers partial and context-sensitive access rather than unrestricted knowledge of the mind

    Canonical inference

    What this does not assert: Summary boundary.

  83. Change can depend on whether cognition becomes available for evaluation, not only on its content

    Canonical inference

    What this does not assert: Relevance to change.

  84. Detecting attentional drift, weak understanding, misplaced familiarity, excess confidence, repeated failure or the need for feedback can each matter for change

    High confidence

    What this does not assert: Conditions metacognition can reveal.

  85. Recognising uncertainty can prompt checking

    High confidence

    What this does not assert: Monitoring enabling control.

  86. Noticing unjustified confidence can make reconsideration possible

    High confidence

    What this does not assert: Possibility, not guarantee.

  87. Metacognition does not supply missing strategies

    Canonical inference

    What this does not assert: Limit on control.

  88. Metacognition does not create unlimited attention, emotional capacity or behavioural control

    High confidence

    What this does not assert: Capacity limits remain; see D8.1.

  89. Cognition includes perceiving, remembering, attending, deciding and understanding

    Established

    What this does not assert: Scope of the object level.

  90. Metacognition does not remove the influence of habit, context, reward or prior belief

    High confidence

    What this does not assert: Other determinants persist.

  91. An accurate judgment may fail to change behaviour when no better response is available or performable

    High confidence

    What this does not assert: Why detection underdetermines change.

  92. Metacognition can make cognition available for possible revision while cognition is shaping understanding, confidence and action

    Canonical inference

    What this does not assert: Canonical statement of contribution.

  93. Detection is not correction, but without detection correction may never become a deliberate possibility

    Canonical inference

    What this does not assert: Closing formulation.

Where to go from here

Next published piece

Error Monitoring and Behavioural Adjustment

Detecting a mistake and improving because of it are not the same event. Error monitoring makes a mismatch available to influence what happens next; whether that influence becomes useful adjustment depends on the standard applied, the information available and the actions the person can reach.

Continue through the Library →See where this sits in the graph →

Back to the Library →