The forgetting curve, explained

In 1885, a German psychologist named Hermann Ebbinghaus did something nobody had thought to do: he measured forgetting.

He memorised lists of nonsense syllables, waited, and tested himself. Then he plotted how much survived against how long he had waited. The shape that came out of it has been replicated for well over a century, and it is bad news if you were planning to reread your notes the night before.

What the curve actually says

Most of what you learn today is gone within days. Not some of it. Most of it.

The decline is steepest immediately — a large share of new material is lost within the first twenty-four hours — and then it flattens out. Whatever survives the first few days tends to stick around much longer.

Two things follow from that shape, and they are the whole reason spaced repetition exists.

The first is that forgetting is not a character flaw. It is not carelessness, or bad note-taking, or not caring enough. A healthy brain discards information that does not appear to be needed. The curve is what that discarding looks like when you graph it. Every person who has ever studied hard for an exam and then blanked three weeks later was on the same curve as everyone else.

The second is more useful: the curve resets every time you successfully recall something. Each retrieval flattens it. The material decays more slowly the next time, and more slowly again after that. The interval you can leave before the next review stretches out — a day, then several days, then weeks.

That is the entire mechanism. Space your reviews to land near the moment you are about to forget, and each one buys you exponentially more time than the last.

Why rereading feels like it works

Here is the part that costs students the most marks.

In a well-known 2006 experiment, Roediger and Karpicke gave students a passage to learn. One group reread it repeatedly. Another read it once and then spent the remaining time testing themselves on it.

The rereaders were more confident. They reported feeling that they knew the material better. They were wrong: a week later, the self-testing group recalled roughly fifty percent more.

The confidence gap is the dangerous bit. Rereading produces a sensation psychologists call the fluency illusion — the text is familiar, it flows easily, and your brain reads that ease as evidence of knowledge. It is not. Ease of recognising something is nearly unrelated to your ability to produce it later, unprompted, under pressure.

This is why "I've been through my notes four times" and "I know this material" are different statements, and why only one of them predicts an exam result.

Difficulty is the point

The related finding is stranger still, and it is worth knowing because it makes the right study method feel worse than the wrong one.

Rohrer and Taylor, in 2007, gave two groups the same practice problems. One group drilled them in tidy blocks — all of topic A, then all of B, then all of C. The other got the same problems shuffled together.

The blocked group sailed through practice. The shuffled group struggled visibly. On a test a week later, the blocked group scored around twenty percent. The shuffled group scored around sixty-three.

Robert Bjork's term for this is desirable difficulty: effort that feels unproductive in the moment but produces deeper, more durable encoding. Mixing topics forces you to work out which approach applies before applying it, which is the skill an exam actually tests — and which blocked practice quietly does for you.

There is a third finding in the same family. Kornell, Hays and Bjork showed in 2009 that learners forced to guess before being told the answer remembered it better afterwards — even when every guess was wrong. A failed retrieval attempt seems to open the slot that the correct answer then fills.

Put together, these three results say something uncomfortable: the study methods that feel most productive are usually the least effective, and the ones that feel like failing are often working.

What to do about it

The practical version is short.

1. Test yourself instead of reviewing. Close the notes and try to produce the answer. If you can see the answer while you are studying, you are rereading.
2. Space the reviews out. Same total hours, spread across days, beat the same hours in one sitting — reliably and by a wide margin.
3. Mix topics rather than blocking them. Shuffled practice feels worse and scores better.
4. Guess before you check, even when you have no idea. The wrong guess is not wasted.
5. Distrust fluency. If it feels easy, that is information about the format, not about your memory.

None of this requires software. Paper flashcards and a calendar implement all five. What software does is handle the arithmetic — tracking when each individual item is due, so you review the twenty things that are fading today rather than the four hundred things in the deck.

Where this leaves the night before

Cramming is not useless. It genuinely works for a test tomorrow, and if the exam is tomorrow you should cram.

It simply does not produce durable knowledge, because durable knowledge is a function of spacing and retrieval, neither of which a single long session contains. Material crammed the night before is on the steepest part of the curve on exam day, which is fine if the exam is the finish line.

For a certification exam it usually is not. The written test is a checkpoint; the knowledge has a job to do afterwards, months later, without warning and without a search bar. That gap — between passing and knowing — is the one the forgetting curve describes, and spacing is the only known way across it.

Sources

• Ebbinghaus, H. (1885). Über das Gedächtnis. The original forgetting-curve measurements.
• Roediger, H. L., & Karpicke, J. D. (2006). Test-enhanced learning. Psychological Science, 17(3), 249–255.
• Rohrer, D., & Taylor, K. (2007). The shuffling of mathematics problems improves learning. Instructional Science, 35, 481–498.
• Kornell, N., Hays, M. J., & Bjork, R. A. (2009). Unsuccessful retrieval attempts enhance subsequent learning. Journal of Experimental Psychology, 35(4), 989–998.