The forgetting curve

Hermann Ebbinghaus learned rows of 13 nonsense syllables and timed how long relearning took. After 20 minutes he saved 58.2% of the work; after 31 days, 21.1%.

Most people know the feeling of reading a chapter on Sunday and finding on Wednesday that most of it has slipped away. The first person to put numbers on that slide was the German psychologist Hermann Ebbinghaus. He finished his experiments in 1880 and published them in 1885, and his curve still appears in every introductory textbook. This unit shows what he measured, why his measure was clever, and what a careful modern replication found.

Measuring memory that you cannot recall

His measure was savings: the share of the original learning time that was saved when relearning. Suppose a row first takes 25 repetitions and a day later takes 20. Five repetitions were saved, so savings are 5 out of 25, or 20 percent. The beauty of savings is that it detects memory even when nothing can be recalled. You may not be able to say a single syllable of the row, yet you relearn it faster than a fresh one, so something was kept.

Steep at first, then nearly flat

The curve drops hardest in the first hour: from 58.2 percent savings at 20 minutes to 44.2 percent at one hour. After that the fall slows a great deal. Between day 6 and day 31, a gap of 25 days, savings fell by only about four points. Ebbinghaus put it this way: after 24 hours about one third was always remembered. Notice what the curve does not say. It does not say you forget 70 percent of everything in a day. The numbers are savings for meaningless syllables, learned by one careful man. Meaningful material that you connect to what you already know is forgotten more slowly, a point OpenStax makes about encoding.

Did it replicate?

Psychology has had a replication crisis, so it is fair to ask whether a result from one man in 1880 holds up. In 2015 Jaap Murre and Joeri Dros at the University of Amsterdam repeated the experiment as closely as they could. One subject spent 70 hours learning and relearning rows of 13 syllables at the same intervals. The curve came out very similar to the original. Two differences are worth knowing. Their 31-day value was much lower than Ebbinghaus's, perhaps because many more lists were learned in between, which adds interference. And the curve was not perfectly smooth: savings jumped up at the 24-hour point, which the authors suggest may reflect the night of sleep in between.

Why do we forget at all? Fading over time is one answer, which Daniel Schacter calls transience. Interference is another. Proactive interference is old learning blocking new, such as writing last year's date in January. Retroactive interference runs the other way: new learning blocks old, such as forgetting your old phone number after you get a new one. The next unit turns the curve into a plan: if forgetting is fast at first and slow later, reviews should come soon at first and then further and further apart.