The same hours, spread out: how far apart to space revision

One of the oldest results in psychology

Take a fixed amount of study and split it across sessions with a gap between them, or do it all at once. On any test taken more than a day later, the split version wins. That sentence has been true in the literature since the 1880s and it has never stopped being replicated, which makes it unusual among findings about learning.

The definitive count is Cepeda and colleagues’ 2006 synthesis: across a century of experiments, 271 comparisons drawn from 254 studies, spaced study beat massed study on the final test, 47 per cent recalled against 37, and only 12 of the 271 comparisons failed to show the benefit.¹ The scope is worth saying out loud, because it is narrower than the headline: the materials are overwhelmingly word lists and paired associates rather than school subjects. The direction is not in doubt. The size, on A-level chemistry rather than on Swahili vocabulary, is an extrapolation.

Dunlosky and colleagues, reviewing ten study techniques for how well each generalises across learners, materials and tests, gave their highest rating to only two, and this was one of them. The other was testing yourself.⁵

How far apart, for an exam on this date

“Spread it out” is advice. “How far apart” is a number, and it was measured directly rather than guessed at. In 2008 Cepeda’s group had more than 1,350 people learn facts, come back to review them after a gap ranging up to three and a half months, and then sit a test up to a year later. For every distance to the test, there was a best gap, and it moved.²

For a test a week away, the best gap was about a day. At five weeks it was eleven days; at ten weeks, twenty-one; and at a year, still about twenty-one. As a fraction of the time you have left, the optimum falls from roughly a fifth at short delays to nearer a twentieth at a year.² So there is no fixed multiplier to memorise, and the practical reading is simpler than the graph: the further away the exam, the longer you should leave it before you come back, and the answer is a proportion of the horizon, not a number of days.

The shape matters more than the values. The peak is a broad ridge rather than a spike, and it falls off gently on the long side and steeply on the short side. Coming back too soon wastes the session; coming back a little late costs almost nothing. If you are unsure, wait.

Why forgetting a little helps

Repetition inside one evening lands on a memory that is still fully available. There is nothing to rebuild, so the second pass adds very little, and the sensation of the second pass, everything arriving instantly, is pleasant precisely because it is doing so little. Leave a day, and some of the access has decayed. Bringing the thing back now costs effort, and a memory retrieved from a partly decayed state gains far more in how well it is stored than one that arrived without being asked for.

Sessions on different days also attach different cues to the same material, because you are in a different room, a different mood and a different part of the week. One long evening attaches one set of cues. Four short sessions attach four, and any of them might be the one available in the exam hall.

This is the same machinery that makes testing yourself beat rereading. In Roediger and Karpicke’s 2006 experiment the rereaders were ahead five minutes after studying and behind a week later, and they were the more confident group throughout.³ Spacing and retrieval are two ways of arranging for the material to be slightly hard to reach when you reach for it. Both feel worse. Both work.

Why nobody believes it about themselves

Massed practice feels like fluency right up until the exam, and the feeling is the problem. Kornell and Bjork let learners set aside the flashcards they judged they had already learned, which is the natural thing to do with a pile that feels finished, and found small but consistent decreases in later recall against keeping the whole set in rotation. They put the failure in the judgement itself: dropping a card has a compelling logic, and it is only ever as good as the learner’s sense of what they know, which is the unreliable part.⁴

So when a topic feels done after one evening, that is the evening talking. The instrument you would use to decide whether to come back is reading how easily the material comes to mind right now, and right now is the one moment that number is guaranteed to be high.

What to do with a timetable

• Divide, do not add. Whatever hours a subject was going to get this week, split them across three or four days instead of one. The workload is the same; only the calendar changes.
• Scale the gap to the horizon. A test on Friday wants a return tomorrow. A mock in five weeks wants one in about ten days. Exams in June want the first return in a few weeks, not a few days.
• Err long. The ridge is gentle on the far side. A review a week late is nearly as good as one on time; a review the same evening is nearly worthless.
• Return by testing, not by reading. The return session is where the gain is made, and it is made by producing the material, not by looking at it again.
• Do not trust “done”. A topic that feels finished is a topic due back on the calendar, not a topic to cross off.

None of this has been tested on Kerfox, or on any app; the studies above were done on other people’s materials with other people’s learners, and the size of the effect for a given student on a given subject is not a number anyone can honestly quote. What the app does is run this calendar for you, bringing each topic back when the ridge says it is due, because the finding is old, replicated and the opposite of what a timetable drawn by hand usually does.