Skip to the page
Kerfox

Edge

Join the waitlist

EDGE

The evidence on how to study.

Everything here is something that was measured, on people, and written down. Each claim carries how big it was and how sure anyone is, including the ones that turned out smaller than you have been told and the ones that did not survive being tried again.

5 topics 60 findings 111 studies

THE FORGETTING CURVE

The same hours, spread out, are worth more.

This is not a claim that you should study more. It is a claim that four hours split across four days beats four hours in one evening, at no extra cost, on every test taken more than a day later. It is one of the oldest results in psychology and one of the most reliably replicated, and almost nobody believes it about themselves.

or drag the 3 reviews along the month

all of itnoneexama month laterlearn itday 0day 30day 60left alone
7
14
21
allnoneexama month laterlearn itday 0day 30day 60left alone
7
14
21

ON EXAM DAY

Most of it

A MONTH LATER

Over half of it

3 reviews, a week apart. All 3 the night before instead: nearly all of it on exam day, and about a third of it a month later.

How this is drawn. A model of the mechanism, not a measurement. Each review either brings the memory back or relearns it from the answer, and one that lands after some forgetting makes it last longer than one that lands while it is still fresh. The axis has no numbers because the real ones depend on what you learn and how; the measured ones are under this card.

Well established Spaced practice recalled 47.3% against massed practice’s 36.7%, a gap of 10.6 percentage points, across 254 studies and more than 14,000 participants (Cepeda, Pashler, Vul, Wixted and Rohrer, 2006).

THE EVIDENCE

5 topics, 60 findings.

FIRST, WHAT A NUMBER MEANS

Effects are reported in standard deviations.

Written d or g. Drag it and watch what the size actually buys. Almost everything below sits between 0.3 and 0.8.

everyone who did noteveryone who didd = 0.5, the whole effectexam scorehow many people
everyone who did noteveryone who didd = 0.5, the whole effectexam scorehow many people

d = 0.5 is about 0.79 of an A-level grade. The average person who did it lands at the 69th percentile of those who did not, and picking one of each at random, the one who did it wins 64% of the time.

01 SLEEP

2 weeks

at six hours a night leaves you as impaired as one night with none.

The problem with sleep loss is not that it makes you tired. It is that it stops making you feel tired long before it stops damaging you, so the instrument you would use to decide whether you need more sleep goes flat while the damage keeps accumulating. Everything else in this topic follows from that one result, including the parts that are more permissive than you expect.

14 findings 24 studies

Well established

You stop noticing long before you stop declining

Feeling levels off in days. Performance does not.

Forty-eight people were kept in a laboratory for two weeks on four, six or eight hours in bed, and tested every two hours. The four-hour and six-hour groups got steadily worse the whole time. How sleepy they said they felt rose for a few days and then flattened, and the four-hour group did not report feeling any sleepier than the six-hour group despite making roughly twice as many attention errors. This is the single most important thing in this section, because it means your own judgement of whether you are getting enough sleep is not evidence.

how tired you feelhow much you have lost50%100%0481438% of the damage74% of the tirednessnights at six hours. 100% = the whole fortnight.
how tired you feelhow much you have lost50%100%0481438% of the damage74%nights at six hours. 100% = the whole fortnight.
Good evidence

Below six hours, each hour is worth about 0.07 of a grade point

Real, and smaller than you would guess

The best-measured study of this found a threshold rather than a slope. Above roughly six hours a night, sleeping more was not associated with better marks. Below it, every hour lost tracked about 0.07 of a grade point across the term. Both halves of that sentence matter: there is a floor worth defending, and there is no bonus for lying in bed longer than you need.

0-0.07-0.144h5h6h7h8h9htheir average, 6.5hgrade pointsevery hour below six
0-0.07-0.144h5h6h7h8h9htheir average, 6.5hgrade pointsevery hour below six

At 6.5 hours a night you are on the flat part. More sleep than this was not associated with better marks in this study, and less was.

12 more findings

Open one for the studies, their sizes and their limits.

  • Good evidence

    Once it is routine, you stop being able to tell at all

    Self-prediction accuracy fell from 0.80 to 0.20 over a week

  • Well established

    Six hours a night for a fortnight is worth about one lost night

    Four hours a night is worth two

  • Good evidence

    The sleep that shows up in your exam was banked weeks earlier

    The night before predicted nothing. The month did.

  • Good evidence

    When you sleep counts, not only how much

    Same total hours, 0.30 of a grade point apart

  • Well established

    One hour lost is not one hour back

    No study anywhere establishes an hours-per-hour rate

  • Well established

    One night without sleep breaks specific things, not everything

    Attention collapses. Reasoning survives.

  • Good evidence

    Sleep before you learn matters more than sleep after

    g = 0.62 before learning, g = 0.28 after it

  • Good evidence

    The best nap is much shorter than you think

    Ten minutes beat twenty, thirty and five

  • Good evidence

    Every hour later in the day costs a little

    About 0.9% of a standard deviation per hour

  • Good evidence

    You are almost certainly not a natural short sleeper

    Well under 1% of people. About 50 families known worldwide.

  • Good evidence

    You are probably wrong about how much you slept

    Teenagers overstate it by about 85 minutes

  • Mixed evidence

    More sleep is not indefinitely better

    Optimum around 7.4 hours, and it explains 0.7% of anything

02 HOW TO STUDY

84%

of students reread. It is close to the least useful option.

Nearly everything students do when they revise is reading the material again, and reading it again is close to the least useful thing available. The methods that beat it are not harder or longer. They are less pleasant, because every one of them works by making you struggle to produce the answer instead of letting you recognise it, and struggle feels like failure while recognition feels like knowing. That feeling is the trap. Here is what was measured, and by how much.

14 findings 30 studies

Good evidence

Cramming is not useless. It is a loan.

Wins by 6 points at five minutes. Loses by 14 at a week.

Worth saying plainly, because the opposite is usually said and the opposite is not true. In the experiment that founded this whole literature, the students who simply read the passage again beat the students who tested themselves when the test came five minutes later, 81% against 75%. Then the lines cross. Two days later the rereaders are behind, and after a week they have lost more than half of what they had while the other group has lost a seventh. Cramming is not a myth. It is a loan at a punitive rate.

read it againtest yourself40%60%80%5 minutes2 days1 weekhow long after studying the real test came
read it againtest yourself40%60%80%5 minutes2 days1 weekhow long after studying the real test came

Tested after 5 minutes, rereading wins: 81% against 75%. Cramming is not a myth, it is a short-dated one.

Good evidence

How far apart depends on how far away the exam is

Roughly a tenth to a third of the time you have left

There is no universally correct gap. The longer you need to hold on to something, the longer the gap that produces the best recall, and the relationship was measured directly rather than guessed at. Drag the exam date below and the band shows where the answer sits. It is a broad ridge, so being out by a factor of two costs almost nothing.

close enough to optimalyour test7d30d90d1yr1d7d30dDots: the four gaps Cepeda measured. Axes are log.
close enough to optimalyour test7d30d90d1yr1d7d30dDots: the four gaps Cepeda measured. Axes arelog.

Test in 29 days: revisit each thing roughly every 3 to 9 days, which is about 5 passes over it rather than one long one.

12 more findings

Open one for the studies, their sizes and their limits.

  • Well established

    Testing yourself beats reading it again

    g = 0.50 over 159 comparisons

  • Well established

    The same hours, spread out, are worth more

    g = 0.74 against doing them in one block

  • Good evidence

    Mixing problem types up beats twenty of the same

    g = 0.42 pooled; 61% against 38% in one classroom trial

  • Good evidence

    Asking yourself why beats highlighting the sentence

    g = 0.55 across 69 effect sizes

  • Well established

    You are a poor judge of your own learning

    84% reread. 1% name self-testing as their main method.

  • Good evidence

    Explaining it to someone works, if you do it from memory

    g = 0.48 when you knew you would have to. g = -0.02 when you did not.

  • Mixed evidence

    An AI tutor helps while you have it and can leave you worse without it

    Up to +127% with it. 17% worse on the unaided exam after it went.

  • Good evidence

    Paper beats a screen by a little, and the gap is widening

    g = -0.21, and growing by about 0.01 a year

  • Good evidence

    A memory palace genuinely works, for the narrow thing it does

    From 26 words to 62 in six weeks

  • Not supported

    Learning styles are not a real thing

    No adequate evidence, after decades of looking

  • Not supported

    Paper, tablet or laptop does not decide anything

    g = 0.04 on learning. The medium is not the lever.

  • Not supported

    Writing in blue pen does nothing

    The studies people cite for it do not exist

03 CRUNCH TIME

8 minutes

of sleep is what an extra hour of revision usually costs. Usually.

The choice between another hour of revision and another hour of sleep gets discussed as though it were a moral question. It is an arithmetic one, and the arithmetic is more interesting than the lecture: the trade is usually much smaller than it feels, the damage lives in the tail rather than the average, and some of the things people do to salvage a bad night work while others are worth nothing at all.

11 findings 14 studies

All 11 findings

Open one for the studies, their sizes and their limits.

  • Good evidence

    An extra hour of revision usually costs eight minutes of sleep

    0.14 hours of sleep per extra hour studied, at sixteen to eighteen

  • Good evidence

    When it does cost sleep, the sleep is what does the damage

    Sleep loss accounted for 90% of the next-day cost

  • Well established

    More practice of what you have already practised buys nothing

    Three problems and nine problems performed identically

  • Good evidence

    Cramming and spacing are level at a week, and not at a month

    Massing wins during practice. Spacing doubles it at four weeks.

  • Good evidence

    Sleep-deprived students score lower and rate themselves higher

    The judgement fails in the same direction as the performance

  • Good evidence

    The first minutes after waking are worse than not sleeping at all

    Briefly comparable to up to 40 hours awake

  • Mixed evidence

    Ten minutes of writing about your nerves is worth a grade, for some people

    B minus to B plus, but only if you are the anxious kind

  • Good evidence

    Practice testing protects your memory from your own nerves

    Stress impaired the studiers and not the testers

  • Good evidence

    Change the answer if you think you should

    Changes go wrong-to-right twice as often as right-to-wrong

  • Well established

    What actually predicts your grade, in order

    Turning up beats intelligence. Rote repetition predicts nothing.

  • Mixed evidence

    The mindset intervention is real, and it is small

    +0.10 grade points, in lower-achieving students only

04 FUEL

38%

of a four o’clock coffee is still in you at eleven at night.

Nutrition research is where the gap between what is sold and what was measured is widest, so this topic is organised around one rule that survives almost every study in it: correcting a deficiency works, and enhancing a well-fed person mostly does not. That clears away a great deal, including several things sold specifically to students, and leaves a short list that is worth knowing.

11 findings 23 studies

Well established

Your last coffee is still there when you go to bed

400 mg six hours before bed costs about 41 minutes of sleep

Caffeine leaves at a half-life of about five hours, so a four o’clock coffee is still 38% present at eleven. The consequence was measured directly: a dose taken six hours before bed removed about 41 minutes of sleep, and the people it happened to could not tell. Only the monitor knew.

average, 5h half-lifefast and slow50100mgdrink4h8h12h16hbedtimehours after the drink
average, 5h half-lifefast and slow50100mgdrink4h8h12h16hbedtimehours after the drink

Drink it 6 hours before bed and roughly 41 mg is still circulating when you switch the light off, out of 95 mg. Somewhere between 24 and 56 mg, depending on how fast you clear it.

10 more findings

Open one for the studies, their sizes and their limits.

  • Good evidence

    Caffeine mostly returns you to normal rather than taking you past it

    A regular drinker gains about 17ms. Withdrawal costs 52ms.

  • Good evidence

    Caffeine buys attention, not learning

    200 mg is the whole dose. 300 mg adds nothing.

  • Not supported

    There is no sugar rush

    More tired, not less, in the first hour

  • Good evidence

    A big lunch before an afternoon exam is a real mistake

    Worst about an hour after eating

  • Mixed evidence

    Water helps if you are actually short of water

    Effects appear past about 2% of body mass, which is 1.4 kg

  • Mixed evidence

    Breakfast matters a lot if you are undernourished and little if you are not

    Maths 0.14 in school feeding trials, reading zero

  • Good evidence

    Iron is the one worth checking, and mostly if you menstruate

    Attention improved by 0.59 standard deviations when deficient

  • Not supported

    The supplements sold to students do not work

    Four large trials, four nulls

  • Good evidence

    Study drugs raise effort and lower the quality of it

    g = 0.20 on inhibition, 0.13 on working memory, no grade benefit

  • Good evidence

    What you did last night is still in the exam hall

    Hangover costs about 0.6 SD of memory. Cannabis fades in 3 days.

05 FOCUS

17 points

is what sitting near someone on a laptop costs you in a lecture.

More confident advice is given about concentration than about anything else in studying, and less of it survives being checked. The famous phone study failed to replicate twice. The Mozart effect turned out to be any music, and then to be nothing. The Pomodoro technique has no trial behind it. What is left after that clear-out is short, and it includes the single cheapest intervention anywhere on this page.

10 findings 20 studies

All 10 findings

Open one for the studies, their sizes and their limits.

  • Well established

    Deciding when and where beats deciding to try harder

    d = 0.65 across 94 tests and 8,461 people

  • Well established

    How long you study barely predicts anything

    Hours 0.13. Habits 0.36.

  • Good evidence

    A notification you cannot even answer still costs you

    Errors up 23 to 28%

  • Mixed evidence

    The phone-on-the-desk study did not survive replication

    Pooled g = -0.14, and mostly on working memory alone

  • Well established

    Multitasking in a lecture costs the person next to you more than it costs you

    11 percentage points for you, 17 for them

  • Good evidence

    Lyrics cost you. Instrumental is close to a wash.

    Reading r = -0.11. Liking the song does not help.

  • Thin evidence

    The Pomodoro timer is a brand, not a finding

    No trial of 25 and 5. Breaks help mood more than output.

  • Good evidence

    Study in more than one place

    Two rooms produced about 50% more recall than one

  • Mixed evidence

    Exercise helps the memory you make that day, not the grades you get that year

    Memory 0.52 acutely. Academic achievement, nothing reliable.

  • Good evidence

    Heat and noise cost real marks, and they are not your fault

    An exam at 32 degrees costs about 0.19 SD

HARM REDUCTION

When the good option is already gone.

Students sit exams tomorrow and sleep 4 hours tonight, and a page that only says sleep well has nothing to offer them at the moment they most need it. Where the evidence says a worse choice can be made less bad, it says so, with numbers.

  • SLEEP

    If tonight is going to be short anyway

    Sometimes the choice is already made. This is what the evidence supports when it is, ranked by how well supported it is.

    7 points

  • HOW TO STUDY

    If you have one evening and a whole topic

    The realistic case, and the advice above assumes weeks you do not have.

    5 points

  • CRUNCH TIME

    If tonight is the all-nighter

    Nobody plans these. Here is what the evidence supports once you are in one, and where it runs out.

    8 points

  • FUEL

    If you are running on nothing tomorrow

    Caffeine is the tool almost everyone reaches for. This is how to get the most out of it.

    7 points

  • FOCUS

    If you cannot concentrate at all

    The evening is already going badly and the usual advice is not landing.

    6 points

HOW KERFOX USES IT

With or without us.

Everything on this page is how to study, for anyone who does. Which of these findings Kerfox is built on, where in the app you meet each one, and what we check before a build ships are on the Science page, which opens by saying that nobody has studied Kerfox itself yet.

How it works