Learning Science · Metacognition: knowing what you know and what needs work

Comparing a prediction with later performance

Make a prediction specific enough to compare with a later result and interpret the discrepancy.

Use the examples to connect this lesson with the earlier topics. No specialist equipment is needed.

After reading, you can

Calculate a prediction–performance difference.
Distinguish overall calibration from identifying weaker topics.

Make the prediction testable

Before a ten-question map quiz, Esra predicts eight correct answers without notes. She later answers six correctly. “I am bad at judging myself” is too broad a conclusion. For this quiz, in these conditions, she overpredicted by two questions. Recording the prediction before seeing answers prevents memory of the prediction from drifting towards the result.

Specify the target

A useful prediction names the task, timing, support and scoring rule. Predicting performance on an immediate multiple-choice quiz does not predict explaining a mechanism next week. For a longer response, use a rubric: the learner may predict that an answer will identify the claim, evidence and a limitation.

On a percentage scale, predicted score minus observed score is a simple signed discrepancy. Here 80 percent minus 60 percent is 20 percentage points. The unit is percentage points, not “20 percent better”. A positive discrepancy means overprediction under this convention. This arithmetic describes one result; it does not establish a stable trait.

Signed discrepancy in two authored prediction examples
Predicted score	Observed score	Prediction minus result	Interpretation
80 percent	60 percent	+20 percentage points	Overprediction in this check
70 percent	85 percent	−15 percentage points	Underprediction in this check

Two kinds of useful accuracy

Overall calibration concerns whether predicted success agrees with success over suitable repeated observations. Discrimination concerns whether the learner can distinguish better-learned items from less-learned ones. Someone can predict the total accurately while identifying the wrong topics for revision.

Keep a small record of item-level judgements and results when decisions require that detail. Look for patterns across comparable checks instead of reacting to one miss. Easy guessing, vague scoring, a tiny number of questions and differences in content can all distort a comparison. Improve the measure before drawing a strong conclusion about self-knowledge.

Authored counts from the worked example; they do not estimate a stable trait or calibrate one probability from one outcome.
Topic	Predicted correct	Observed correct	Question for revision
Geography	4	5	Which strengths remain usable?
Chronology	4	1	Sequence/duration confusion to inspect
Total	8	6	Aggregate overprediction hides the uneven pattern

Worked example

A total that hides a decision problem

What does Esra’s topic-level pattern reveal beyond her total prediction?

Esra predicts that she will answer four geography questions and four chronology questions correctly. She actually answers five geography questions and one chronology question correctly. Her total prediction was too high, but the practical finding is the uneven pattern. Revisiting both topics equally would ignore it. She checks the chronology errors for a repeated confusion between sequence and duration, repairs that confusion, and makes a fresh prediction for a comparable set.
A predicted 80 percent chance of answering one item correctly cannot be judged accurate or inaccurate from that one response. The outcome will be correct or incorrect. Probability judgements need repeated suitable observations to compare predicted frequencies with observed frequencies. This differs from predicting eight correct answers on a ten-item set, where the stated total can be directly compared.

The total prediction 8 exceeds the observed 6, but geography 5 versus chronology 1 identifies an uneven gap. Inspect that gap instead of revisiting both topics identically.

Try it yourself

Think through each question before opening the answer.

A learner predicts 70 percent and scores 85 percent. Find the signed discrepancy.
Show the answer and reasoning
70 minus 85 is minus 15 percentage points.
The learner underpredicted this result under the stated convention.
Can an accurate total prediction show that a learner knows which items need work?
Show the answer and reasoning
No.
Opposing item-level errors can cancel. Topic or item judgements are needed if the decision is which material to revisit.

What to carry forward

Define the prediction before checking.
Use repeated comparable results and inspect the pattern that matters for the next decision.

References

Sources supporting the explanations and learning checks in this lesson.