Metacognition: knowing what you know, the judgement-of-learning trap, and the five-minute check

Metacognition is thinking about your own thinking: knowing which ideas you have and which you only recognise. It matters because the feeling of knowing is unreliable. Students who reread a chapter feel more confident than students who tested themselves, and remember less a week later. The fix is not more confidence or less; it is a check that replaces the feeling with a mark. Here is what the research found, the trap it names, and a five-minute check you can run at the end of tonight's session.

What metacognition is

The word was given its current meaning by John Flavell (1979), who defined it as knowledge about your own thinking, together with the monitoring and control of it. In plain words: knowing how your memory works, noticing what it is doing right now, and steering it.

For a student that is two parts:

• Monitoring: do I know this? Can I produce it, or do I only recognise it when I see it?
• Control: so what do I study next? Which idea gets tonight's time, and which can wait?

The second part depends on the first. If your monitoring says "I know this" about something you cannot produce, your control sends tonight's time to the wrong place. Most of what is called "learning how to learn" is getting both parts honest, and the methods that do it are the other pages on the study-methods hub. This page is about the honesty.

The judgement-of-learning trap

A judgement of learning is the prediction you make, while studying, about whether you will remember something later. Researchers write it JOL. You make dozens a night without noticing: every time you turn a page and think "got it", that was one.

The trap is that the judgement is made at the wrong moment and from the wrong evidence.

Koriat and Bjork (2005) called it an illusion of competence. They had students study pairs of words and judge, pair by pair, how likely they were to recall the second word when shown the first. While studying, the second word is right there, so the pair feels obvious, and the judgement is made with the answer in view. At test the answer is absent. The students' predictions tracked how the pair felt with both words present, and did not track what happened when one was taken away. A judgement made with the answer visible is a judgement about the wrong situation.

Rereading inflates the judgement and not the memory. Roediger and Karpicke (2006) had students read a passage and then either reread it or take a recall test on it. The rereaders predicted they would remember more. A week later they remembered less. The feeling went one way; the memory went the other.

The judgement improves when it is delayed. Nelson and Dunlosky (1991) asked students to predict their recall either right after studying each item or after a delay, with only the cue in view. The delayed judgements were far more accurate than the immediate ones. Waiting, and hiding the answer, turns a guess about fluency into an attempt at retrieval, and an attempt at retrieval is what the test will be. Bjork, Dunlosky and Kornell (2013) review these illusions and others in one place; the pattern throughout is that what feels fluent during study is a poor guide to what will come back later.

Why the least prepared feel the most ready

Kruger and Dunning (1999) gave participants tests of logic, grammar and humour and asked them to estimate how they had done. The participants who scored in the bottom quarter had, on average, scored at about the 12th percentile and placed themselves at about the 62nd. The people with the least skill were the furthest off in their estimates, and the authors put the gap down to the same ability doing both jobs: judging whether an answer is right needs the skill that produces right answers.

Two honest notes. The pattern is about judging your own skill in general, not about studying in particular, and later work has argued about how much of it is a statistical artefact of asking people to rank themselves. The useful lesson for a student is narrower than the famous version and stands either way: when you have not been tested on an idea, your estimate of how well you know it is a guess, and if the idea is new to you the guess is probably high.

The cost of the guess is measurable. Dunlosky and Rawson (2012) had students learn definitions of key terms and judge, after each attempt to recall one, whether they had got it. Students who judged a wrong or partial answer as right stopped studying that term too early and retained less of it later. Overconfidence did not only feel wrong; it cut the study short in exactly the places that needed more.

The five-minute check

At the end of a session, before you pack up.

1. List the ideas you studied tonight. One line each. One minute.
2. Beside each, predict: Sure, Maybe or No. Sure means "I could produce this now with nothing in front of me." One minute.
3. Answer one question on each idea, notes closed. From the end-of-section set, or your own cards. Short questions take about two minutes for the lot; a longer problem gets one each and the check runs long, which is fine.
4. Mark Right or Wrong beside each prediction. Thirty seconds.
5. Read the two columns together. Sure-and-Wrong rows go first tomorrow. No-and-Right rows mean you were underconfident and can move on. Thirty seconds.

Two rules make it work. Make the prediction before the question, never after; a prediction made after you have seen the answer is a memory of the answer, not a judgement. And for ideas you only read tonight and did not practise, make the prediction tomorrow, because the delayed judgement is the honest one. Chapter 1 of the free playbook has the check as a one-page table.

Worked example: ten derivative cards, predicted then tested

A Grade 12 calculus (MCV4U) student has ten cards on the product rule, the quotient rule and the chain rule, studied yesterday, mostly by reading the solutions.

Predictions: Sure 7, Maybe 2, No 1. Counting the Sures and half the Maybes, the student expects about 8 right (7 + 1 = 8), which is 80 per cent.

Cold test: 5 right. Of the 7 Sures, 3 were wrong, and all three were cards whose solution the student had read the day before and never redone.

One of the three:

Find d/dx [x² · eˣ].

The correct working:

1. This is a product of two functions, u = x² and v = eˣ, so the product rule applies: (uv)′ = u′v + uv′.
2. u′ = 2x and v′ = eˣ.
3. Put them in: 2x · eˣ + x² · eˣ.
4. Factor the common x · eˣ: x · eˣ (2 + x).

The common wrong line is 2x · eˣ alone: the product rule skipped because "eˣ is its own derivative", so the student differentiated the x² and left the eˣ as it was. That is the derivative of x² with eˣ carried along, not the derivative of the product.

The gap: predicted 80 per cent, actual 50 per cent. Thirty points of overconfidence, and all of it on the read-not-redone cards. Cards whose solution you read and never redid are the usual culprit, and the fix is the one on the maths page: redo the solved problem cold, the solution covered, before you call the card Sure. Tomorrow's first block is those three cards.

Calibration: how to get your predictions honest over time

Calibration is the match between what you predict and what you score. You improve it the way you improve anything else: by measuring it.

Keep the two columns for a week. Each night, count the Sure-and-Wrong rows. The count falls as you learn to trust only what you have produced, and it falls fastest when you adopt one rule: judge an idea only after you have tried to produce it, never after reading it. Producing before looking is that rule as a method, and the delay before the second visit is why a judgement made tomorrow is worth more than one made tonight.

After a week most students find that their Sures are reliable and their Maybes were mostly Nos. That is calibration, and it is the thing an exam cannot take away from you: knowing, before you walk in, which ideas are actually there.

The mistakes

• Predicting after you have seen the answer. A prediction made with the answer in view is the illusion the research names.
• Counting "I basically knew it" as Right. If you could not produce it, you did not have it.
• Rereading to feel ready before an exam. It raises the feeling and nothing else; the night before has the honest version.
• Never writing the prediction down. A prediction in your head is revised after the fact, every time.
• Testing only the ideas you like. The check is for the whole list, and the ideas you avoid are the Sure-and-Wrong rows waiting to be found.
• Mistaking nerves for the trap. Low confidence with high scores is not a metacognition problem; if it is nerves rather than overconfidence, the fix is different.

The method on one page

• Metacognition has two parts: monitoring (do I know this?) and control (what do I study next?).
• The trap in one sentence: a judgement made right after reading, with the answer in view, measures how fluent the reading felt and not whether the memory will last.
• The five-minute check: list the ideas, predict Sure/Maybe/No, one question each with the notes closed, mark Right/Wrong, read the two columns together.
• Predict before the question, never after. Judge an idea only after trying to produce it.
• Keep the two columns for a week and watch the Sure-and-Wrong count fall.

Where YSR Studies does the bookkeeping

On paper, the check is a table you draw each night. In YSR Studies a card's answer stays on the server until you have tried it, so the judgement is made after an attempt by construction, and the answers are checked by code, the maths compared rather than the spelling. The course map shows each idea as Learned, Shaky or Not shown yet, which is the two columns kept for you across the course; the schedule is FSRS at a desired retention of 0.9. Five free hours every month.