You finish a chapter and it feels clear. Three weeks later a friend asks what it said, and little comes back. Your memory did not fail. The feeling at the end of the chapter was measuring something else.
This guide answers one question: how to know you have learned something, and not just read it. EdProcto's page for self-learners covers two of its five checks: questions you did not write, made from your own material, and a second sitting listed as due days later. The rest need only paper.
Why does reading something feel like learning it?
Because the feeling is built from how easily the material goes in, which says little about what you will get back out. Psychologists call such feelings judgements of learning: forecasts, made during study, of later recall. Three errors bend them.
Fluency: easy to read is taken for well learned
Kornell et al. (2011) showed single words once or several times, in large or small type. Type size changed how much people expected to remember, not how much they did. Extra chances to study did the reverse: they changed memory and scarcely touched the forecasts. The authors call it the ease-of-processing heuristic: what is easy to take in is judged well learned. In our reading, that is how familiarity passes for knowledge.
Foresight bias: the answer is in view when you judge
Koriat and Bjork (2005) pointed to a mismatch built into study: you judge with the answer on the page, and are tested with it gone. With word pairs, they argue, forecasts follow how well the two words seem to belong together when both are visible, while recall follows whether the first alone brings the second to mind. This is the illusion of competence. They call it the foresight bias, and later found that being tested, and delaying the judgement, both reduced it (Koriat and Bjork, 2006).
Stability bias: memory is assumed to stay as it is
Kornell and Bjork (2009) told people they would study word pairs between one and four times before a test. Across 12 experiments, people predicted little or no gain from the extra study, though the real gain was large. Koriat et al. (2004) found the same blindness to forgetting: forecasts did not fall as the promised wait before the test lengthened, unless people compared several waits side by side.
That answers "why do I forget what I read": you forgot what anyone would, and your forecast had left forgetting out.
When should you judge whether you know it?
Judge it later, with the answer out of sight, never at the end of the sitting with the page open.
Dunlosky and Nelson (1992) studied the delayed judgement-of-learning effect: forecasts are more accurate when made a short while after study than straight after it. When the prompt was the first word of a pair alone, every one of their 45 participants showed the effect. When it was the whole pair, answer included, there was little effect or none. Rhodes and Tauber (2011) pooled 112 comparisons covering 4,554 participants and found a large gain in accuracy from delaying (g = 0.93). A second pooling found only a small gain in memory itself (g = 0.08).
The 1992 study used word pairs, and its abstract says "shortly after study" with no figure in hours, so waiting until tomorrow is our working rule, not a finding.
Can you explain it, or do you only recognise it?
Write the explanation out step by step: if you only recognise the idea, the explanation stops part-way, and where it stops is what you do not know.
Rozenblit and Keil (2002) asked 16 graduate students to rate, from 1 to 7, how well they understood devices such as a zip, a flush toilet and a cylinder lock. Each then wrote a step-by-step causal account of four and rated again. The ratings dropped, and many said they were surprised at how much less they knew than they had thought. Across twelve studies the drop was large for explanations of how things work, smaller for facts and the plots of films, and absent for procedures. The authors named it the illusion of explanatory depth. One reason they offer: people seldom give explanations, so they have little record of how theirs turn out.
Who is worst at judging what they know?
The popular answer is beginners, and it is disputed.
Kruger and Dunning (1999) gave tests of humour, grammar and logic in four studies. People in the bottom quarter scored at about the 12th percentile and put themselves at about the 62nd. The authors' account: without the skill, you lack the means to see that you lack it.
Splitting a group into quarters, the usual method, has been charged with producing the pattern out of regression towards the mean and the common habit of rating oneself above average. Gignac and Zajenkowski (2020) used two statistical checks free of that problem on data from 929 adults, did not find the pattern, and judged that the effect may be much smaller than reported. Burson et al. (2006) found that on moderately hard tasks the best and worst performers judged themselves about equally well, and on harder ones the best judged worse.
Jansen et al. (2021), though, repeated the grammar and logic studies with about 4,000 people each and found support for low performers being less able to tell whether an answer was right.
So the question is open, and self-estimates look noisy at every level. Calibration, the match between how sure you are and how often you are right, is something to measure.
Is using it on something new the real test?
Yes, close to home: using an idea on a case you have not met is the strongest sign that you hold it, but do not expect it to travel far.
Near transfer carries a skill to a closely related task. Far transfer carries it to a loosely related one, as from Latin to mathematics. Sala and Gobet (2017) pooled studies of whether chess lessons, music lessons and working-memory training lift children's other abilities and school results. The effects were small to moderate and shrank as study design improved, and the authors concluded that far transfer rarely occurs.
Pan and Rickard (2018) drew together 192 effects from 122 experiments on whether practice testing transfers. On average it did (d = 0.40). Transfer was among the strongest to a different test format and to application and inference questions, and among the weakest to material studied and never tested. So a test should make you apply the idea, and passing a quiz says little about the pages it left out.
How do you know you have learned something? Five checks
Run five checks on each chapter over about ten days.
- Recall it with the book shut, a day later. Write what the chapter said, then mark your page against it. The answer is out of view and the judgement is delayed.
- Explain it, to someone or on paper. Write the central "why" or "how" step by step, rating your understanding from 1 to 7 before and after, as Rozenblit and Keil (2002) did.
- Answer questions you have not seen. Not ones you wrote or have answered before: being tested is what reduced the foresight bias. EdProcto writes them from the chapter you paste into the quiz maker on its home page. Before checking each answer, note how sure you are.
- Apply it to a new case. A news story, a problem from another book, your own work. Pan and Rickard (2018) found transfer strong to application questions.
- Do it again in a week. The same questions and explanation, book shut. By the stability bias, your forecast left out a week of forgetting.
The day, the week and the number five are our working rules, not findings. How to test yourself when you are learning on your own covers setting and scheduling the tests.
What do the five checks look like on one chapter?
They look like five dated notes and one small table. This illustration uses made-up figures: an adult teaching themselves economics reads the chapter on supply and demand on a Monday and finds it clear.
| Check | What was done | What it showed |
|---|---|---|
| 1. Recall, Tuesday | Wrote the chapter from memory, marked it against its ten summary points | 6 of 10. Missing: shifts of a curve against moves along it |
| 2. Explain, Tuesday | Wrote why a price ceiling below the market price causes a shortage | Self-rating 6 before, 3 after. Missing: at the capped price buyers want more and sellers offer less |
| 3. Unseen questions, Wednesday | Two ten-question quizzes written from the chapter by EdProcto, each answer rated 1 to 5 on paper first | Forecast 17 of 20, scored 12 |
| 4. A new case, Wednesday | A cap on rents, then a minimum wage as a price floor on labour | Rents right, being the chapter's own example. Wages wrong: employers taken for the suppliers |
| 5. Again, a week on | The same questions and explanation | Forecast 14, scored 15. Explanation complete |
The calibration tally
Check 3 gives this tally, computed for this guide.
| Rating before checking | Right | Share right |
|---|---|---|
| 5, certain | 5 of 8 | 63% |
| 4, fairly sure | 4 of 6 | 67% |
| 1 to 3, unsure | 3 of 6 | 50% |
| All | 12 of 20 | 60% |
The forecast was 85% and the result 60%: a gap of 25 points, the size of the illusion that day. A 5 was right no more often than a 4.
For a yardstick, take EdProcto's marking rule: with its Confidence bonus on, a 5 pays only from four chances in five of being right, and a 4 from two in three. By those lines the 4s were honest and the 5s were not, and the three answers rated 5 and wrong go first on the restudy list. Our guide to confidence-based marking has the arithmetic.
A week later the forecast was 70% and the result 75%: the learner knows more, and knows better what they know.
What can these checks not tell you?
They cannot show what you are unable to see, and they do not prove mastery.
- Marking your own answer is lenient. Lipko et al. (2009) had middle school pupils recall definitions and score themselves. They gave full or part credit to 73% of their plainly wrong answers with no model answer in view, and to 44% with the correct definition beside them. Mark checks 1 and 2 against the source, idea by idea.
- A test you wrote shares your blind spots. It can ask only about what you noticed. That is reasoning, not a cited study, and it is why check 3 wants outside questions.
- The evidence has edges. Much of the research cited here used word pairs and short definitions, some way from a whole field or a manual skill.
- A score is a reading on one day. No check, and no quiz, certifies that a subject is mastered.
How does EdProcto help with the checks?
EdProcto is the best tool for the two checks a lone learner cannot supply, unseen questions and a planned second sitting, because it writes the questions from the learner's own material: transfer was weak to material studied and never tested, so the questions must cover your own chapter.
- Unseen questions, checked at once. Paste a chapter, a transcript or notes into the quiz maker, choose "Test yourself", and Procto AI writes the quiz from that text. A study quiz checks every answer as you give it and ends with a retry of the misses.
- A rating with each answer. Multiple choice with confidence asks how sure you are, from 1 to 5. With the Confidence bonus off the rating changes no mark, and the results do not show your ratings back, so keep the tally on paper.
- The reason, and a new case. In a study quiz, choose and justify marks your choice at once and shows a model reason to set beside yours. The case study shape, in the builder now, unfolds one scenario across several questions.
- Again in a week. Your dashboard lists a quiz as due two days after a sitting with any misses, and seven days after one with none. No reminder is sent.
The Free plan carries 20 questions a month written by Procto AI, enough for the example's two quizzes. The Learner plan gives self-tests without limit and 100 written questions a month. The pricing page has each plan.
Quick questions
How do I know if I have really learned something?
When you can recall it with the book shut a day later, explain it step by step, answer unseen questions, use it on a new case, and still do so a week on.
What is the illusion of competence?
The sense during study of knowing something that later proves out of reach. Koriat and Bjork (2005) traced it to judging with the answer in view, so judge with the answer hidden.
How can I check my understanding without a teacher?
Write an explanation and mark it against the source idea by idea, then answer questions you did not write. EdProcto's page for self-learners says how it makes those from your own material.
Is the Dunning-Kruger effect real?
The pattern has been found again, and its cause is disputed. Gignac and Zajenkowski (2020) put most of it down to statistics, while Jansen et al. (2021) found support for low performers judging their own answers less well.
References
- Burson, K. A., Larrick, R. P., & Klayman, J. (2006). Skilled or unskilled, but still unaware of it: How perceptions of difficulty drive miscalibration in relative comparisons. Journal of Personality and Social Psychology, 90(1), 60–77. https://doi.org/10.1037/0022-3514.90.1.60
- Dunlosky, J., & Nelson, T. O. (1992). Importance of the kind of cue for judgments of learning (JOL) and the delayed-JOL effect. Memory & Cognition, 20(4), 374–380. https://doi.org/10.3758/BF03210921
- Gignac, G. E., & Zajenkowski, M. (2020). The Dunning-Kruger effect is (mostly) a statistical artefact: Valid approaches to testing the hypothesis with individual differences data. Intelligence, 80, Article 101449. https://doi.org/10.1016/j.intell.2020.101449
- Jansen, R. A., Rafferty, A. N., & Griffiths, T. L. (2021). A rational model of the Dunning–Kruger effect supports insensitivity to evidence in low performers. Nature Human Behaviour, 5(6), 756–763. https://doi.org/10.1038/s41562-021-01057-0
- Koriat, A., & Bjork, R. A. (2005). Illusions of competence in monitoring one's knowledge during study. Journal of Experimental Psychology: Learning, Memory, and Cognition, 31(2), 187–194. https://doi.org/10.1037/0278-7393.31.2.187
- Koriat, A., & Bjork, R. A. (2006). Illusions of competence during study can be remedied by manipulations that enhance learners' sensitivity to retrieval conditions at test. Memory & Cognition, 34(5), 959–972. https://doi.org/10.3758/BF03193244
- Koriat, A., Bjork, R. A., Sheffer, L., & Bar, S. K. (2004). Predicting one's own forgetting: The role of experience-based and theory-based processes. Journal of Experimental Psychology: General, 133(4), 643–656. https://doi.org/10.1037/0096-3445.133.4.643
- Kornell, N., & Bjork, R. A. (2009). A stability bias in human memory: Overestimating remembering and underestimating learning. Journal of Experimental Psychology: General, 138(4), 449–468. https://doi.org/10.1037/a0017350
- Kornell, N., Rhodes, M. G., Castel, A. D., & Tauber, S. K. (2011). The ease-of-processing heuristic and the stability bias: Dissociating memory, memory beliefs, and memory judgments. Psychological Science, 22(6), 787–794. https://doi.org/10.1177/0956797611407929
- Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: How difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of Personality and Social Psychology, 77(6), 1121–1134. https://doi.org/10.1037/0022-3514.77.6.1121
- Lipko, A. R., Dunlosky, J., Hartwig, M. K., Rawson, K. A., Swan, K., & Cook, D. (2009). Using standards to improve middle school students' accuracy at evaluating the quality of their recall. Journal of Experimental Psychology: Applied, 15(4), 307–318. https://doi.org/10.1037/a0017599
- Pan, S. C., & Rickard, T. C. (2018). Transfer of test-enhanced learning: Meta-analytic review and synthesis. Psychological Bulletin, 144(7), 710–756. https://doi.org/10.1037/bul0000151
- Rhodes, M. G., & Tauber, S. K. (2011). The influence of delaying judgments of learning on metacognitive accuracy: A meta-analytic review. Psychological Bulletin, 137(1), 131–148. https://doi.org/10.1037/a0021705
- Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: An illusion of explanatory depth. Cognitive Science, 26(5), 521–562. https://doi.org/10.1207/s15516709cog2605_1
- Sala, G., & Gobet, F. (2017). Does far transfer exist? Negative evidence from chess, music, and working memory training. Current Directions in Psychological Science, 26(6), 515–520. https://doi.org/10.1177/0963721417712760