What forgetting looks like when somebody plots it
80 min
Two hosts talk the lesson through. The voices are synthetic; the script was written from this lesson and checked against it, and asserts nothing the lesson does not.
- State what the method of savings measures and how it differs from a percentage recalled
- State what the 2015 replication found, with its sample and its relearning intervals
- Explain what the jump at 24 hours is, and why it matters that the curve is not smooth
Every poster about studying has the same curve on it. This lesson is about where it came from, what it measures, and the two things that fall off before it reaches you.
The measure comes first
Before any curve makes sense you have to know what is on the vertical axis, and in this case it isn't what almost anybody assumes.
The measure is savings. You learn a list until you can produce it. Some time later you learn it again, and the question is how much of the original effort you are spared. If learning took twenty minutes and relearning takes ten, you saved half.
That isn't a percentage of items recalled, and the difference isn't a technicality.
Somebody can be unable to produce a single item and still show substantial savings, because relearning is easier than learning even when nothing comes back on demand. On a savings measure that material is partly there. On a recall measure it is entirely gone. That follows from the definition of the two measures rather than from any comparison this course read, and no source here puts the two instruments side by side.2
Which is right? Both, and they are answering different questions. Focus and Deep Work lesson 1 called this the instrument question, and this is the same question in a subject a century older: before asking whether a claim is true, ask what would have had to be measured.
And here it bites unusually hard, because the figure that travels from this curve is reported as though it were recall. "You forget half of what you learn in an hour" is a sentence about production. The curve behind it is about relearning.
The study
Murre and Dros published a replication of the classic forgetting curve in 2015.1 This course has read its abstract and the summary statements of method and result; the curve-fitting analysis was not opened. The 1885 monograph it replicates was not opened either, and everything this lesson says about the original reaches you at one remove, through the 2015 paper.
The design, in the paper's own words: "One subject spent 70 hours learning lists and relearning them after 20 min, 1 hour, 9 hours, 1 day, 2 days, or 31 days."1
Read that twice, because the sample is the second thing this lesson is about.
One subject. Not a small sample. One person, doing seventy hours of it, across six relearning intervals from twenty minutes to a month, on nonsense syllables.
Before you read on: is a study with one participant worthless? Write down your answer and one sentence of reasoning, and then read what this lesson says.
Show the answer
Most people say yes, some say it depends, and the useful answer is narrower than either.
What it can't do is tell you about people. One person isn't a sample of a population, and no amount of data from that person makes it one. Any sentence beginning "people forget" is out of reach.
What it can do is test whether an older result reproduces, which is exactly what it was built for. The study it replicates was also a study of one subject, from the 1880s. Replicating a one- subject study with one subject is the right shape of test for the question "does this hold up?", even though it is the wrong shape for "is this true of everybody?"
So the honest sentence is: in one modern participant, the savings curve reproduced the shape reported in the 1880s. That is narrow, it is real, and it is very much more than nothing given how much of this subject rests on the original.
And it is worth noticing what a century of citation did to it. The founding measurement of an entire field is a study of one person, and the presentations of it this course has met do not say so. That last sentence is the course's own observation about the presentations it looked at, not a survey of the literature.3
What it found
The authors' own conclusion, verbatim: "We conclude that the Ebbinghaus forgetting curve has indeed been replicated and that it is not completely smooth but most probably shows a jump upwards starting at the 24 hour data point."1
Two claims there and they pull in different directions. The curve replicated. And the shape isn't the one everybody draws.
A jump upwards in savings means more of the original learning survives at that point than a smooth decay would predict. Not that forgetting stops. The measured value at a day is higher than the smooth curve through the other points would put it.
This course can't tell you why, because the analysis that would bear on it is in the parts of the paper that were not read.1 The authors offer it as their conclusion and this lesson reports it as their conclusion.
What it does establish is smaller and more useful: the curve on the poster is smoother than the curve in the data. A curve drawn from an equation is a model of the measurements, and the measurements did not lie on it.
Worked: the same material, two instruments
This case is constructed, because the paper reports one participant's curve rather than a comparison of instruments.3
Priya learns twenty Welsh words on a Monday. She can produce all twenty at the end of the session, which took her eighteen minutes.
On Friday, somebody asks her to write down as many as she can. She gets four. On a recall measure she has retained twenty percent, and if somebody put that on a slide, that is the figure that would travel.
On the same Friday, she relearns the list to the same standard. It takes her seven minutes against the original eighteen. On a savings measure she has retained about sixty percent.
Same person, same material, same day, and two answers that differ by a factor of three.3
Neither number is wrong. They are answers to different questions: what can you produce right now, and how much of the work is still done. A course that gave you one of them and called it "how much you forgot" would be misleading you, and every presentation of the curve this course has met gives one number and no instrument.3
And notice which direction the gap runs in Priya's case: the savings figure is the larger of the two. Whether that holds generally is not something this course can tell you, because no source it read compares the two measures on the same material.2 What the case shows is that the two can differ by a lot, which is enough for the point being made.
So is the forgetting curve useless? A poster with a smoothed curve, a measure nobody explains, and one participant.
Show the answer
No, and the reason is worth more than the curve.
What it establishes is the shape of the problem, and the shape is what you act on. Forgetting is fast at first and slows down. That isn't in doubt, it is what the replication reproduced, and it is why How to Learn Anything lesson 4's spacing advice is built the way it is: the first review matters more than the fourth.
What it doesn't establish is any number you should plan with. Not seventy percent, not a day, not any of it. The interval schedule that lesson gives you comes from studies designed for the purpose, not from this curve, and this lesson isn't undermining it.
And there is a third thing, which is why the lesson exists here rather than in the earlier course. A measurement whose instrument nobody explains, on a sample nobody mentions, smoothed into a curve nobody drew, is the single best worked example this subject has of how a real finding travels badly. The curve isn't the problem. What happened to it on the way to the poster is.
So the useful version is: forgetting is rapid then slow, the shape has been reproduced, the numbers on the posters are not measurements you can use, and the measure is savings rather than recall.
Three things people get wrong about the curve
"You forget half of what you learn in an hour." The measure it would have to come from is savings rather than recall, and the number is repeated without either its instrument or its sample.
"The curve is a law." It is one subject, twice, a century apart, and the 2015 authors say the shape is not smooth.
"A replication means the original was right." It means the result reproduced. This replication reproduced the broad result and corrected the shape, which is what a good replication usually does.
Practice
Take 25 minutes across a week, in three short sittings.
Sitting one, about 15 minutes. Pick twenty arbitrary items: foreign words, capital cities, names paired with numbers. Something you don't already know and will not meet in between.
Learn them until you can produce all twenty once, and time it. Write the number of minutes down. That is your baseline.
Sitting two, five days later, about 5 minutes. Before relearning anything, write down as many as you can. Count them. That is your recall figure.
Sitting three, immediately after, about 5 minutes. Relearn the list to the same standard, and time it. Your savings figure is the original time minus the relearning time, over the original time.
Then two lines.
- The two percentages, side by side.
- Which one you would have quoted if somebody asked how much you had forgotten, and why.
One person for one week establishes nothing about anybody, which you now have an unusually good reason to believe.
Take 20 minutes.
Find the forgetting curve being used somewhere: a study app, a training deck, a blog post, a poster in a school.
Write down four things.
- What the vertical axis says, word for word, or that it is unlabelled.
- Whether the measure is named anywhere on or near it.
- Whether a sample is given, and what.
- What number the piece wants you to take away, and where that number appears in anything you can check.
Then one line: which of the four did the source give you?
If the answer is none, that is the usual result and it is what this lesson is for. A curve with no axis label, no instrument, no sample and a confident headline number is the standard presentation of the best-known finding in this subject.
Connections
Back. Lesson 1's reconstructive account is why a savings measure makes sense at all: material that can't be produced on demand can still be partly there. Focus and Deep Work lesson 1 is the instrument question, and this lesson is it applied to a measure from the 1880s. How to Learn Anything lesson 4 is spacing, which this lesson supports and doesn't replace.
Forward. Lesson 3 is a second reason things become hard to reach: retrieving one thing can make its neighbours harder to get at. Whether that is part of what shapes this curve is not something this course read the analysis to answer. Lesson 7 comes back to what happened to this curve on its way to the poster.
Go deeper
- Replication and Analysis of Ebbinghaus' Forgetting Curve (PLOS ONE, 2015). Open access, read in part by this course: the abstract and the summary statements of method and result. Read the second sentence of the abstract, which is where the sample is, and then look at how the curve is presented anywhere else you have met it.
- Forgetting as a Consequence of Retrieval: A Meta-Analytic Review of Retrieval-Induced Forgetting (Psychological Bulletin, 2014), open in the University of Reading repository and read in part by this course. Lesson 3's subject. Read it beside this one: a one-subject study and a pooling of 472 samples, on the same question of why things become hard to reach.
Sources
- Jaap M. J. Murre and Joeri Dros, "Replication and Analysis of Ebbinghaus' Forgetting Curve", PLOS ONE 10(7), 2015, e0120644. Read in part: the abstract verbatim and the summary statements of method and result. The curve fits and the serial-position analysis were not opened, and the body says so where the shape is discussed. Supports: the quoted design sentence with its one subject, seventy hours and six relearning intervals; the quoted conclusion about replication and the jump at the 24-hour point; and that the measure is the method of savings.
- The description of what savings measures, and of how it differs from a recall measure, is the standard definition in this literature and is stated in the source above as the method the study used. That somebody can show savings while recalling nothing follows from the two definitions and not from a measurement this course read, and the body says so at the point of use. No source read for this course sets the two instruments side by side on the same material, which is why the worked case stops short of a general claim about which measure runs higher.
- Priya is constructed, and so is every figure attached to her; the lesson says so where she appears. Her two percentages are invented to show the gap between the instruments and are not a reported result. The observation that a century of citation has detached the curve from its sample is also this course's own, said as such in the predict block and again in the worked case. It describes the presentations this course has met and is not a survey of how the curve is cited.
Check your understanding
This lesson has a 6-question quiz. Pass it and the questions come back on a schedule in Review, so what you learned stays learned. Your progress is saved in your browser; no account needed.