Readwyn 0.1.0

Why you forget most of what you read, and what helps

One day after learning a list, Hermann Ebbinghaus needed about two thirds of his original effort to learn it again. After a month he still saved about a fifth. He published those numbers in 1885, from years of tests on himself, and the shape of his curve is the one that matters for a book you've just finished: the steepest loss comes in the first hours, and what survives fades slowly after that. Reading a book once leaves you on that curve, and reading it again helps less than it feels. Pulling the content back out of your own head, and doing it again a few days later, helps far more.

What Ebbinghaus measured

He learned series of 13 nonsense syllables until he could recite them twice without a mistake, then learned the same series again after a gap of 20 minutes to 31 days and timed how much work the second learning saved. Chapter 7 of the 1913 English translation of Memory puts the result in one sentence:

After 24 hours about one third was always remembered; after 6 days about one fourth, and after a whole month fully one fifth of the first work persisted in effect.

"Remembered" there means a saving in relearning time, not the share of the syllables he could still repeat, so a claim such as "you forget two thirds of a book in a day" reads more into the curve than it holds.

The curve has held up. In 2015 Jaap Murre and Joeri Dros of the University of Amsterdam repeated the experiment, with Dros as the only subject, spending about 70 hours on lists relearned at the same intervals. Their results came out close to the original. Comparing Ebbinghaus's data with theirs and with an earlier German replication, they conclude that the curve most probably jumps upwards at the 24 hour point. They note that research on sleep would predict such a jump, though they don't claim sleep as its cause.

Nonsense syllables are a harsh test, and Ebbinghaus knew it. He also memorised stanzas of Byron's Don Juan, 80 syllables each, and, as he reports in chapter 5, needed hardly nine readings per stanza, about a tenth of what he expected a nonsense series of that length to take. Meaning, rhythm and rhyme made learning far cheaper. A novel is full of meaning too, but one pass through it with no return is the condition the curve describes.

Why rereading feels better than it works

The test case is from Washington University in 2006: 180 undergraduates, two short passages from a TOEFL preparation book, one on the sun and one on sea otters, and two ways of studying them (Henry Roediger and Jeffrey Karpicke, Psychological Science). In the second experiment of the paper, one group read a passage in four study periods of 5 minutes. Another read it once, then spent three periods writing down all they could recall, with no feedback. The rereaders got through the text about 14 times in all, the recall group about 3 times.

Five minutes later, the rereaders recalled more: 83% of the ideas against 71%. A week later the order had flipped, with the recall group at 61% and the rereaders at 40%. The rereaders had also predicted that they'd remember the passage better. The authors suggest that students base such predictions on what pays off quickly.

When John Dunlosky and four co-authors rated ten study methods in 2013, they put practice testing and spaced practice in its high-utility group and rereading and highlighting in the low one (abstract on Europe PMC). The evidence on highlighting has a post of its own: does highlighting help you remember?

How far apart to review

Ebbinghaus saw the spacing effect too. In chapter 8 he reports that 38 repetitions spread over three days did as much for a series of 12 syllables as 68 repetitions made on the day before the test.

How long to wait depends on how long you want to remember. More than 1,350 people each learned a set of facts, got one review after a gap of up to 3.5 months, and took a final test up to a year later (Cepeda and colleagues, Psychological Science, 2008). The best gap grew as the test moved further away, but it shrank as a share of that delay: about 20 to 40% of a week, and about 5 to 10% of a year. For something you want to remember next week, that works out at a review after one and a half to three days; for next year, after roughly 18 to 36 days.

Applying it to a book

None of these studies used whole books, so carrying them over to a novel or a history is an inference, though not a long one. At the end of a chapter, close the book and write or say what happened and what the author argued before you look at anything. A list of highlights grouped by chapter makes a fair test sheet: the chapter heading is the prompt, and the highlights under it are the answer. Readwyn's Notebook lists them that way, in reading order. Do the recall again two or three days later, and only then check the highlights for what you missed.