Ebbinghaus's forgetting curve, plus what the spacing research adds about when to review.
About 18 min · 7 pages
Learning outcomes
Describe how Ebbinghaus measured memory and what happened as the delay before relearning grew.
Explain how the best gap between study sessions relates to how far away the test is.
Compare rereading with trying to recall material when memory is tested after a delay.
Page 1 of 7Learn
How Ebbinghaus measured memory
Key idea Savings show how much of the original learning effort a second session still saves. (How much of the original learning effort is saved when the same material is relearned.)
In the 1880s the German psychologist Hermann Ebbinghaus studied memory with himself as the only participant. To keep prior knowledge out of it he learned lists of 13 nonsense syllables, meaningless three-letter items such as "WID" or "ZOF", until he could recite a list without error, then relearned the same list. (Meaningless three-letter items used to keep prior knowledge out of learning.) His measure was savings: how much of the original effort was saved the second time. If relearning took half as many repetitions, savings were 50%, and savings of 0% would mean the delay had wiped out everything (Ebbinghaus, 1885/1913).
Page 2 of 7Predict
Predict what happens as the delay grows
Ebbinghaus relearned each list after a delay, from 20 minutes up to 31 days. What happens to savings across that range?
Predict, then reveal
Savings fall as the delay grows. The fall is steep early: savings dropped 24 points in the first day, from 58% to 34%, and only 13 more over the next 30 days.
Takeaway Most of the forgetting happens early, and later delays add only a little more loss.
Page 3 of 7Explore
Step through the forgetting curve
Same lists, longer waits, less saved.
1 Predict
Predict which delay leaves the least saved.
2 Explore
Savings58%
20 minutes
9 hours
2 days
6 days
31 days
Delay before relearning
Stays fixed the same list, relearned to errorless recall
You change the delay before relearning
20 minutesExample values
Delay before relearning of 20 minutes; savings 58%.
Try two cases, or the key case, to see the result.
Try each option again and compare what changed with your prediction.
Takeaway Savings fall as the delay before relearning grows.
All cases
Delay before relearning
Savings
20 minutes
58%
9 hours
36%
2 days
28%
6 days
25%
31 days
21%
Page 4 of 7Explore
Most forgetting happens in the first day
Between 20 minutes and 31 days, savings dropped 37 percentage points (58% to 21%). When did most of that drop happen?
The fall in savings between the 20-minute delay and the 31-day delay
Lost in the first day (58% to 34%)−24 points · about two-thirds of the totalMost of the loss happens early.
Lost over the next 30 days (34% to 21%)−13 points · about a third of the total
More detail
Widths are each part's share of the total shown.
Lost over the next 30 days (34% to 21%) Between the sixth and thirty-first day the figure fell by only four points.
percentage points (The plain difference between two percentages, not a share of them.)
What this shows
About two-thirds of the drop happened in the first day. The next month added only a third more.
Takeaway Most of what is lost goes in the first day; later delays cost much less.
Page 5 of 7Predict
Spaced or all at once?
Two students cover the same material. One does it in a single session, and the other splits it across separate sessions. Tested after a delay, which one remembers more?
Predict, then reveal
The student who split the sessions usually remembers more. This is the spacing effect, also called distributed practice: spaced study generally beat massed study when memory was tested after a delay. (Better later memory when study is split across separate sessions.)(Studying the same material in separate sessions rather than all at once.)(Learning material all at once in a single session.) Cepeda and colleagues (2006) combined the verbal-recall literature: 839 comparisons from 317 experiments in 184 articles.
Takeaway Learning the same material in separate sessions beats learning it all at once.
Page 6 of 7Check
How long should the gap be?
Page 7 of 7Predict
Reread or try to recall?
Roediger and Karpicke (2006) had students study short prose passages and then either reread them or try to recall them. Tested five minutes later, which did better? And what happened when students were tested two days and a week later?
Predict, then reveal
Rereading did better five minutes later; trying to recall did better after two days and a week. Testing yourself is not just a way of checking memory, it strengthens it.
Takeaway Practising recall beats rereading when memory is tested after a delay.
Optional
Go deeper
Optional topics that go further. Open any one.
Check How far should you trust the curve?
Answer each question before revealing the source's answer.
The curve comes from one participant learning nonsense syllables. Does that make the numbers easier or harder to generalise than everyday learning?
It describes a hard case, not everyday learning. Meaningful material is usually kept better.
Murre and Dros (2015) repeated the method with a single participant over about 70 hours of learning. Did their results follow a smooth curve?
No: savings after one day were higher than a smooth curve predicts. They suggested a boost around the 24-hour mark, possibly linked to sleep. Otherwise the results were similar to Ebbinghaus's.
Is there one settled equation for the forgetting curve? (The pattern of savings falling quickly at first and then levelling off.)
No. In a review of 210 published data sets, Rubin and Wenzel (1996) found that four different two-parameter functions fitted most of them, and the data could not easily tell those four apart.
Explore What the teaching equation says
What does each part of this model stand for?
The exponential model
= e^(−/)
Select a highlighted part to see what it means.
Nothing selected yet.
How retrievable something is.
Time.
The stability of the memory.
What this shows
Treat it as a convenient model, not a law.
Takeaway The exponential is a handy model of the curve, not a settled law.
Practise Set the review gap
Using the source's ranges, when should the review session fall in each case?
Example A. A learner will be tested one week after first learning the material.
Example B. A learner will be tested one year after first learning the material.
See the model answer
For Example A the best gap is about 20% to 40% of the delay: about 1.4 to 2.8 days, so a review one to three days after first learning.
For Example B the best gap is about 5% to 10% of the delay: about 18 to 37 days, since 5% of 365 days is about 18 and 10% is about 37.
At any test delay, performance first rose and then fell as the gap grew, so a gap that is too short or too long is worse than one in the middle.
Check Which techniques earn their place?
Practise What the sources do not show
Learn Where the numbers come from
Key idea The whole-number table is a rounding of Ebbinghaus's own values, listed to one decimal place.
Ebbinghaus's table is rounded to whole numbers. To one decimal place his seven savings values are 58.2, 44.2, 35.8, 33.7, 27.8, 25.4 and 21.1 per cent, as tabulated in Table 3 of Murre and Dros (2015), who take them from Ebbinghaus's own manuscript.
