Why You Forget What You Read: The Forgetting Curve, Replicated

You finish a great article about the Roman Republic or how vaccines work, and a week later you can recall the general vibe and maybe one detail. That is not a personal failing. It is one of the oldest and most reliably measured findings in psychology, and the data on it is public.

This post walks through two studies: a modern replication of the original forgetting curve, and a large experiment on when to review. Then it does the arithmetic on what they mean for anyone who reads to learn.

The original experiment, run again

In the 1880s Hermann Ebbinghaus memorized lists of nonsense syllables and measured how much faster he could relearn them after different delays. The time saved on relearning, as a share of the original learning time, is called savings. If learning a list took 1,000 seconds and relearning it took 600, the savings is 40 percent.

In 2015 Jaap Murre and Joeri Dros published a careful replication in PLOS ONE. One subject, Dros, spent about 70 hours learning 69 lists of 104 syllables each, then relearned them after 20 minutes, 1 hour, 9 hours, 1 day, 2 days, 6 days or 31 days. The paper's Table 3 puts his savings next to Ebbinghaus's original figures:

• 20 minutes: Ebbinghaus 58.2 percent, Dros 47.2 percent

• 1 hour: Ebbinghaus 44.2 percent, Dros 37.3 percent

• 9 hours: Ebbinghaus 35.8 percent, Dros 27.6 percent

• 1 day: Ebbinghaus 33.7 percent, Dros 31.7 percent

• 2 days: Ebbinghaus 27.8 percent, Dros 23.0 percent

• 6 days: Ebbinghaus 25.4 percent, Dros 16.8 percent

• 31 days: Ebbinghaus 21.1 percent, Dros 4.1 percent

The authors conclude the curve "has indeed been replicated," with one twist: it is not perfectly smooth and probably jumps upward at the 24 hour point. They note that current sleep research would predict such a jump, but say a link to sleep "remains to be established." They also note Dros's learning times crept up over the experiment, possibly from interference or fatigue, and that correcting for this would put his 31 day savings at about 13.7 percent.

Running the numbers

Savings at 20 minutes is already well below 100 percent, so the clearest way to see the shape is to ask how much of the 20 minute level is left at each later point. We divided each value by the subject's own 20 minute savings.

For Dros:

• At 1 day: 31.7 divided by 47.2 is about 67 percent of the 20 minute level

• At 6 days: 16.8 divided by 47.2 is about 36 percent

• At 31 days: 4.1 divided by 47.2 is about 9 percent

For Ebbinghaus:

• At 1 day: 33.7 divided by 58.2 is about 58 percent

• At 6 days: 25.4 divided by 58.2 is about 44 percent

• At 31 days: 21.1 divided by 58.2 is about 36 percent

Two patterns hold for both men. First, the steepest drop happens in the first hours: Ebbinghaus lost 14 points of savings between 20 minutes and 1 hour, more than he lost in the entire stretch from 1 day to 31 days (12.6 points). Second, after a week, most of the advantage from the first learning session is gone.

Nonsense syllables are harder than a meaningful article, so your own curve will be gentler. The shape, fast loss early and a long slow tail, is the part that generalizes.

The fix is a well-timed second look

If forgetting is the problem, review is the obvious answer. The useful question is when. In 2008 Nicholas Cepeda and colleagues published Spacing Effects in Learning in Psychological Science. 1,354 people learned 32 obscure but true facts (for example, which European nation eats the most spicy Mexican food), reviewed them after a gap of up to 105 days, and took a final test up to 350 days later.

The results:

• For a fixed amount of study, the best gap gave a 64 percent increase in final recall compared with reviewing immediately

• To remember for a week, the best gap was about 3 days

• To remember for 35 days, about 8 days

• To remember for 70 days, about 12 days

• To remember for 350 days, about 27 days, which improved recall by 77 percent over an immediate review

Expressed as a share of how long you want to remember, the best gap fell from 43 percent for a one week test to 8 percent for a one year test. The authors' own summary is blunt: the interaction of gap and test delay "implies that many educational practices are highly inefficient."

What this means for everyday reading

Put the two studies together and a simple rule emerges. Reading something once and moving on loses most of it within days. A short, effortful return to the material, a day or a few days later, protects it far better than rereading on the spot.

That is hard to arrange with articles and books, because nothing prompts you to come back. It is easier with small units. A single fact is short enough to revisit, and a question about it ("what did you learn about black holes this week?") is a natural review.

Where Study Textapp fits

Study Textapp sends you one short fact a day by text, on topics you choose in your own words. You can ask follow-up questions when a fact makes you curious, and whenever you like, ask it to quiz you on your recent facts; it grades your answers. Asking for a quiz a few days after a batch of facts is a cheap way to add the delayed review the Cepeda data rewards. It does not schedule reviews for you automatically, so the timing is yours.

It is an AI, not a person or a teacher, and it can make mistakes. Facts never repeat, you can have up to three topics, and you can check your progress, including your quiz score, any time.

Try it by text

Text Hi to (424) 999-9536 and say what you want to learn: Roman history, the human brain, investing, anything. Your first fact arrives right away, then one every day at 10am or a time you pick. It works on any phone that can text, with no app and no account. The first 20 messages are free, then $9.99/month covers every Textapp. Text STOP whenever you want to end it. Message and data rates may apply. Details at textapp.co/apps/study.