Why we remember some facts and forget others

Try this before you read on. Without looking, draw a coin you've handled every week of your adult life. Which way does the head face? What's written round the edge? Where does the date sit?

Most people can't do it. Not badly — at all. In the late 1970s the psychologists Raymond Nickerson and Marilyn Adams put this to Americans using the one-cent coin, and the results were dismal: people who'd handled pennies daily for decades couldn't reproduce one, and struggled even to pick the correct version out of a line-up of near-misses. It's been repeated with other currencies since. You've looked at that coin thousands of times. You never once looked at it.

That isn't a broken memory. That's a memory doing its job.

Three jobs, and they all fail identically

Stop treating memory as one action. Getting something into your head and back out takes three separate stages, and when it goes wrong at any of them the experience from the inside is the same: you just don't know the thing.

Encoding turns an experience into something storable. Storage keeps it there. Retrieval fetches it back. When people say they've got a terrible memory they nearly always mean retrieval — the answer's in there somewhere, sitting just out of reach, which is maddening. But an enormous share of ordinary failures are encoding failures. You didn't forget where you put your keys. You never registered putting them down, because you were thinking about something else at the time, and there's nothing in there to retrieve.

The tip-of-the-tongue state is the clean example of the other kind. You can often produce the first letter, the number of syllables, a word it rhymes with — everything except the word. Whatever's stored clearly hasn't gone anywhere. The route to it has.

Ebbinghaus and his nonsense syllables

In the 1880s a German psychologist called Hermann Ebbinghaus decided to measure forgetting, which nobody had seriously attempted, and he did it on the only subject he could fully control: himself. He built lists of nonsense syllables — WUX, ZOF, that sort of thing — precisely so that meaning couldn't come to his rescue. Then he memorised list after list, waited, and measured how much relearning each one needed later on.

It took him years and it can't have been much fun. But he produced the first real curve of forgetting, and its shape has held up through more than a century of refinement: a steep collapse in the first day, then a long slow flattening out. Most of what you lose, you lose fast. What survives the first week tends to stick around.

He found something else in those lists that's proved just as durable. Items at the start of a list came back reliably, items at the end came back reliably, and the middle sank without trace. That's the serial position effect, and you can watch it operate on any shopping list you didn't write down.

His practical conclusion was about timing, and it's the most useful sentence in this article. Reviewing something while it's still fresh does very little. Reviewing it just as it begins to slip does a great deal, because the effort of reaching for it is what does the work.

Nothing in there is a file

Your brain doesn't store memories the way a drive stores files. Recall is reconstruction — fragments held in different parts of the cortex, assembled on demand into something that feels like a recording and isn't one.

Which explains a lot. It explains why two people remember the same argument differently and both of them mean it. It explains why memories drift towards the version you've told most often, since every retelling is a fresh act of building rather than a replay. And it explains the misinformation effect, which Elizabeth Loftus spent a career documenting: ask people a leading question after an event and the wording of the question quietly edits what they'll later report having seen. The rebuilt memory doesn't arrive with a label marking which parts were added afterwards.

Confidence isn't much of a guide either. Ulric Neisser had students write down where they were when they heard the Challenger shuttle had broken up, the morning after it happened, then asked them again years later. Plenty of the later accounts were wrong — not slightly wrong, wrong about the room and the people and the time of day — and the students giving them were certain. Shown their own handwriting from the day after the event, some of them didn't believe it.

Vivid doesn't mean accurate. It never has.

Why re-reading feels like it works

If retrieval builds memory and re-reading doesn't, why does almost everybody re-read?

Because it's pleasant. The third pass through a chapter is smooth and quick and faintly satisfying, and that smoothness gets mistaken for knowledge. Researchers call it the illusion of fluency: you're familiar with the text, and you read that familiarity as command of the content. Testing yourself feels considerably worse, because it shows you precisely what you can't do yet.

Robert Bjork gave the underlying principle a name that's stuck — desirable difficulties. Certain kinds of friction, in the right places, make learning slower now and better later. Hardly anyone chooses them voluntarily, because the feedback you get in the moment says you're doing badly at it. That mismatch, between what feels productive and what is, is the single most useful idea in the study of studying, and it's why so much revision is wasted effort rather than insufficient effort.

Spacing beats cramming, and it isn't close

Four reviews in one evening are worth far less than one review a day for four days. Same total time. Different result entirely.

The spacing effect is one of the most reliable findings in the whole of psychology, which counts for something in a field that's spent the past decade discovering how many of its famous results won't replicate. Spacing has survived that reckoning comfortably. It shows up across ages, subjects, formats and languages, in laboratories and in real classrooms, and reviews pulling together well over a hundred separate experiments keep landing in the same place.

The rough rule that falls out of the work: the longer you want to hold onto something, the wider the gaps should be. For a test next week, days. For something you'd like to still have in ten years, months.

Cramming isn't useless, and pretending otherwise is dishonest. It genuinely works for tomorrow morning, and if tomorrow morning is the whole of the goal then crack on. What it won't do is leave anything behind by the following term. Harry Bahrick tracked down people who'd studied Spanish at school and then never used it, some of them decades earlier, and found that whatever survived the first few years afterwards then survived more or less indefinitely — a plateau he called permastore. Getting material into that state takes spacing. Cramming doesn't come close.

The physical version of the idea is a box of index cards divided into compartments: cards you get right move further back, cards you get wrong return to the front. Sebastian Leitner popularised it in Germany in the 1970s. Every spaced-repetition app since is that box with a scheduler bolted on.

Testing is studying

The testing effect deserves its own heading, because most people file testing under assessment, and that's where the damage happens.

Retrieving a fact strengthens it more than seeing it again does. Not marginally — substantially, and the gap widens the longer you leave it before the final test. Henry Roediger and Jeffrey Karpicke ran the comparison in a form that's been repeated many times since: one group re-reads a passage, another reads it once and then tries to recall it from memory. On a test five minutes later, the re-readers look better. On a test a week later they're well behind.

There's a related effect that costs nothing to use. Anything you produce yourself sticks better than the identical thing handed to you, which is called the generation effect. Guessing at an answer and getting it wrong, then being told the right one, beats being told from the start. The failed attempt isn't wasted time — it's most of the work.

So a quiz isn't a way of finding out what you know. It's one of the better ways of coming to know it.

Meaning is the glue

Ebbinghaus used nonsense syllables to strip the meaning out, and he then had to grind at them relentlessly to make them stick. That was the design working as intended, and it's also the lesson.

Fergus Craik and Robert Lockhart set this out in 1972 as levels of processing. Attend to the surface of something — the shape of a word, the sound of it — and it fades quickly. Attend to what it means, how it connects to what you already have, whether it's even true, and it lasts. What you were asked to do at the moment of reading turns out to matter more than how long you spent reading.

A list of capital cities is miserable to learn. The reason Canberra exists — a compromise capital built out in the bush because Sydney and Melbourne each refused to let the other have it — sticks after one telling, because it's got something to attach itself to.

Oddity helps as well. Hedwig von Restorff showed in 1933 that an item standing out from its neighbours gets remembered at the expense of the rest. It's why the one strange fact in a list is the one you'll still have next month, and why a page of uniform yellow highlighting is worth almost nothing — if everything's marked as important, nothing is.

Working memory is smaller than you'd like

All of the above concerns long-term storage. The bottleneck sits earlier than that.

George Miller's 1956 paper on the magical number seven, plus or minus two, is among the most cited things in psychology and among the most misquoted. Miller's actual point was subtler than the number he's remembered for: what's limited isn't items, it's chunks, and a chunk is however much you can treat as a single unit. Later work has argued the true capacity is nearer four, which is worse news than seven.

Either way, chunking is the whole game. Eleven separate digits blow straight past the limit. The same eleven digits in three groups don't, and nothing about the information has changed — only the packaging. Chess players show the extreme version of this. A strong player can rebuild a board after a few seconds' glance, but only when it's a position from a real game; scatter the pieces at random and they're no better than anybody else. They aren't storing thirty-two pieces. They're storing half a dozen familiar shapes.

Where you learn it matters, a bit

In 1975 Duncan Godden and Alan Baddeley talked a diving club into memorising word lists, some on dry land and some underwater, then tested them in both places. People recalled more when the setting matched the one they'd learned in.

It's a lovely study and it's routinely oversold. The effect turns up strongly for free recall and largely disappears for recognition, which means it matters rather less for exams than the advice columns imply. Internal state does something similar to external setting — mood, caffeine, time of day all act as cues in the same way.

The sensible reading isn't "revise in the exam hall". It's that a memory formed in one narrow setting comes with that setting stitched onto it, so varying where and when you practise builds more routes back in. Which is, once again, a difficulty that makes practice feel worse and results better.

What sleep is doing with it

A good deal of the filing happens after you've stopped paying attention.

During sleep the brain replays recent activity, strengthening some connections and letting others weaken, and the evidence that this matters for learning is about as solid as anything in the field. Deep slow-wave sleep, concentrated early in the night, appears to do much of the work on facts and events. Cut sleep short after learning something and the material is measurably worse the next day, even when a full night follows.

This is the least glamorous advice going and among the best supported. Given a choice between one more hour of revision and an earlier night, take the earlier night. It isn't a consolation prize. It's the more effective of the two study sessions.

Interleaving, and why nobody does it

The intuitive way to study is to finish one topic before starting the next. It's also the weaker way.

Mixing related material together — interleaving — reliably produces better long-term performance, and reliably feels worse while you're doing it. Studying in blocks lets you stop choosing. Twenty problems of the same type in a row means picking your method once and then repeating it nineteen times, which trains execution and never trains selection. Mixed problems force you to work out what sort of problem you're looking at before you can start, and that's the part you'll actually need later on, when nothing turns up helpfully labelled.

Performance during practice drops when you interleave. That's what puts people off it, and it's also the sign that it's doing something.

Memory athletes are ordinary people with a system

When researchers have scanned people who compete at memorising shuffled decks of cards, they haven't found unusual brains. What they've found is unusual strategy — competitors convert flat material into vivid, spatial, story-shaped material, because that's the kind memory handles well.

The method of loci is the oldest example still in daily use. You walk a route you know intimately, your flat or the way to work, and you leave the things you're memorising at fixed points along it; then you walk it again to collect them. The Greeks credited the technique to the poet Simonides, and Roman students were still being drilled in it centuries later. You aren't adding capacity. You're borrowing the enormous capacity you already have for places and routes, which is far older and far sturdier than anything you have for arbitrary lists.

And an extraordinary memory isn't obviously a prize. Alexander Luria spent about thirty years studying a Russian journalist, Solomon Shereshevsky, who seemed barely able to forget anything at all. He struggled with abstraction, with faces that had changed since he last saw them, with reading a passage while every word set off a cascade of images and tastes. He ended up inventing techniques for forgetting.

What to do on Tuesday

Test yourself instead of re-reading. Shut the book and try to say what was in it. The uncomfortable bit, where you can't, is the bit that's working.

Space the reviews out across days rather than hours, and let a little forgetting happen in the gaps. Reaching for something half-gone is the mechanism, not a sign you've left it too long.

Ask why an answer's true, every time. A fact with a reason attached has somewhere to live and several routes back to it. A fact on its own is a nonsense syllable with better public relations.

Mix the topics up, accept that it'll feel worse, and judge it a week later rather than in the moment.

Then sleep on it. None of this requires a gift for memory. It requires practising in the way memory works rather than the way it feels like it ought to.