Reading a claim about your memory
95 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.
- Sort a claim about memory into measurable, unmeasurable as stated, or false, and say what would settle it
- Identify what a memory claim was measured on, and say what a small or specific sample licenses
- Decide what to do about one remembered event of your own, with what each option costs
Every figure in the other six lessons will date. This one is about what is left when they have.
This subject has an unusually confident popular literature sitting on top of unusually careful research, and the gap between the two is where you will meet almost everything anybody tells you about memory.
Habit one: sort the claim
The institute's standing sort is three categories, and Digital Literacy lesson 8, Using AI Effectively lesson 11, Time Management lesson 8, Habits and Self-Discipline lesson 7 and Focus and Deep Work lesson 7 all use it. Measurable: a study could check it. Unmeasurable as stated: it uses a term nobody has defined. False: it has been checked and didn't hold.
It is used here rather than re-taught. The four claims worked below are specimens written for this lesson rather than quotations from anywhere, and the sorting of each is the course's reasoning.6
Work one. "Most people forget half of what they learn within an hour." Measurable, and something close to it has been measured: lesson 2's replication plotted savings on relearning for one subject across six delays.1 So the sort is easy and the sentence still misleads, for a reason habit two is about.
Work two. "Memory is reconstructive, so your memories are not recordings." Measurable in the second half and almost content-free in the first. Nothing in the sentence says what a reconstruction would have to do for the claim to fail, and a claim that nothing could falsify isn't a finding. As it stands it is a framing, and framings are useful without being results.
Work three. "Eyewitness testimony is notoriously unreliable." Measurable, and the answer turns out to be conditional, which is the most interesting sort in the course. Lesson 5's synthesis opens by saying the pessimistic view was entirely reasonable for the procedures that produced it, and then reports that under five named conditions an initial high-confidence identification by an adult is remarkably accurate.4 So the sentence is false of a properly run lineup, and still true where the five conditions have not been adopted, which the paper says is many jurisdictions and says in the present tense. A sorter who stopped at "false" would have dropped the sentence the synthesis is most often quoted without.
Work four, and this is the commonest shape. "Don't trust your memory of an argument, because questions change what people remember." Two claims stuck together. The second is lesson 4's, at one remove and on a hundred and fifty undergraduates with a week between viewing and questioning.3 The first is advice about your own life, and the study underneath it wasn't about your life. The two halves need separate verdicts, and a reader who accepts or rejects both together has done neither.
Habit two: ask how many people, and who
This is the course's own addition to the sort, and it's here because of something the research turned up about this subject in particular.6
The samples behind this course are the smallest you will meet on the Core path, and the claims made about memory in general are among the largest. That comparison is this course's own rather than a finding, and it is a comparison with the courses on the path rather than with every study anybody has run.6 Lesson 2's founding measurement of the whole field is one subject.1 Lesson 4's famous demonstration is a hundred and fifty undergraduates in 1974.3 Lesson 6's training study is about fifty people.5 Set those against what gets said about memory in general, which is usually about everybody who has ever lived.
The question to put to a measurable claim, then, isn't only whether somebody checked but how many people they checked and who those people were.
Before the next section. One subject, a hundred and fifty students, fifty trainees. Does a finding from a sample that small tell you anything at all?
Show the answer
Yes, and saying exactly what takes a sentence rather than a shrug.
A small sample is bad at one specific job: telling you how big an effect is in a population. It is much better at a different one, which is showing that something happens at all. One subject relearning lists over thirty-one days cannot tell you how fast you forget. It can show that savings on relearning falls steeply and then flattens, and that the shape is regular enough to fit an equation.1
So the sample question isn't a way of dismissing things. It's a way of reading a finding for what it is: a demonstration, an estimate, or a claim about everyone, and those three want very different amounts of trust.
The trouble comes when a demonstration is reported as an estimate. "You forget half of what you learn within an hour" takes a curve measured on one person relearning nonsense syllables and prints it as a fact about you learning anything.1 The study is sound and the sentence isn't its finding.
Before the table. Five sources, five literatures. Guess which one rests on the most data, and guess which one supports the biggest claim.
Show the answer
Most people put the same source in both answers, and in this course they come apart.
The most data is the pooled review in lesson 3, which turns many studies into a single estimate across four hundred and seventy-two samples.2 And it supports the narrowest claim of the five: a cost of a few percentage points on items sitting beside something you retrieved, under laboratory conditions, over minutes.
That is not a coincidence. Pooling is what turns a demonstration into an estimate, and estimates are usually smaller than the demonstrations that made people interested in the first place.6
The biggest claim in the table belongs to the smallest samples. Read the two columns together and you will see it in every row.
The five sources, side by side
What this course read, what each rests on, and what it supports. The first three columns are the sources'; the last is this course's own reading of what each one licenses.6
| Source | Sample | Read level | What it licenses |
|---|---|---|---|
| Forgetting curve replication[1] | One subject, 70 hours, six delays | Abstract and summary statements | The shape of forgetting on relearning, not a rate for you |
| Retrieval-induced forgetting review[2] | Pooled across many studies | Abstract, effect sizes, one analysis and its discussion | A small cost to neighbouring items, and an unsettled mechanism |
| Questions and memory[3] | 150 participants, 1974 | A published teaching summary, not the paper | That a verb's wording moved later answers, on those people |
| Eyewitness confidence synthesis[4] | A literature of mock-crime and field studies | Summary, one box, the five-recommendations section, one passage | Initial high-confidence identifications by adults under five conditions |
| Method of loci training[5] | About fifty people, three arms | Abstract and participant counts | That the technique produced durable memories against an active control |
Read the sample column and the last column together. In every row the licence is narrower than the source's reputation, and in every row something real survives. Row four points the other way from the claim you will meet in the wild, which is what makes it the interesting one.
One row is different from the others and it's worth naming. The pooled review is the only source here that turns many studies into a single estimate, and it supports the narrowest claim of the five.26 That isn't a coincidence. Pooling is what turns a demonstration into an estimate, and estimates are usually smaller than the demonstrations that made people interested.
The books you will be handed
You will meet this subject mostly through three very widely read books: Schacter's The Seven Sins of Memory, Loftus's own writing for general readers, and Foer's Moonwalking with Einstein.
This course has opened none of them.6 They sit on the unread line of its research file, the build fails on a citation to any of them, and no claim in these seven lessons rests on any of the three. Naming them is all that's honest, and it's worth doing, because you will be handed one.
What to do when that happens is the two habits above, and the second especially. What this course can tell you is what to do with any book that builds a chapter on one famous demonstration, which is habit two: how many people was it run on, and who were they.
What this course will not do
Here's the refusal, and it's sharper than the ones ending the two courses before this.
This course cannot tell you whether a particular thing you remember happened the way you remember it. Not the argument, not the childhood scene, not what somebody said to you last year. Nothing in seven lessons is a test you can run on one recollection.
What it can tell you is what memory does in general, under conditions somebody arranged and measured. That is a different question, and the difference is the last thing the course teaches.
Both directions of overreach are the same mistake. A reader who takes lesson 4 and concludes that a person contradicting them has been led by a question has gone past the evidence.3 So has a reader who finishes the course and decides to stop trusting anything they remember: that is a verdict on particular memories too, reached from general findings, and it costs them something real in exchange for a confidence that isn't more justified than the one it replaced.
This course has spent seven lessons being careful about what the evidence supports. So apply that to the course. Which of the things you have been told here should you hold most loosely, and which most firmly?
Show the answer
Worth doing, and the order isn't the obvious one.
Hold most firmly: the scope question. Not a finding, a way of reading, and the evidence for its usefulness is the table above, where every row's licence turned out narrower than the claim you will meet in the wild. Nothing that happens to these figures touches it.
Hold nearly as firmly: lesson 5's shape. A belief that was reasonable, procedures that changed, a claim that now holds under named conditions and fails outside them.4 That is the most careful thing this course read, and its usefulness outlives the particular numbers.
Hold loosely: the specific figures. Lesson 2's curve is one subject.1 Lesson 4's numbers come to you at one remove through a teaching summary, which the lesson says at the point of use.3 Lesson 6 gives no figure at all, for a reason it states twice.5
Hold most loosely: the mechanisms. Lesson 3's review found the inhibition account largely supported and, in the authors' own words, also facing "some challenging evidence".2 A mechanism is the thing likeliest to be different in ten years, and the effect will still be there when it is.
And hold this against the course itself. Five of its seven lessons draw on failure literatures, which is a selection. It was made because those are the literatures with recent syntheses and it is still a selection, and a reader who finishes here with a bleaker view of memory than the evidence supports has been affected by the shape of the reading list rather than by the findings.
Three things people believe about this literature
"Somebody must have checked." Lesson 4's study is the most famous demonstration in the subject and this course could not open it, reporting its figures through a teaching summary and saying so.3 If that is the state of the most famous one, assume nothing about the rest.
"The science says." There is no single body here that says anything. There is a replication of an 1885 result on one subject, a pooled review with an unsettled mechanism, a demonstration from 1974, a synthesis that overturned a consensus by making it conditional, and a training study on fifty people. They don't speak with one voice and they were never asked the same question.
"A famous study is a big study." Lesson 4's is famous and had a hundred and fifty participants.3 The study lesson 2 is built on replicates the one that founded the field, and both had one subject.1 Fame in this subject tracks how good the demonstration was, which is a different property from how much it licenses.
Practice
Take 20 minutes, and put it down at twenty whether or not the trail has ended. Lesson 4 traced a study; this one traces a number.
Find one claim about memory with a number in it. An app's onboarding, a headline, a book jacket, something somebody told you at work.
Write it down word for word, with where you found it. Then follow it back, writing each step.
- What does this source cite? If nothing, write "nothing", and you are finished.
- Open what it cites. Is the number there? Not something like it.
- How many people, and who? Students, adults, a single subject, somebody's customers.
- When was it measured, and how long after what? This subject's findings are mostly about a delay, and a figure without one is usually missing its conditions.
Write your finding in one sentence. Five endings are all results, and the list is Habits and Self-Discipline lesson 7's and Focus and Deep Work lesson 7's, recast here for a number rather than a study: the number is there with its sample and delay; the number is there stripped of them; the trail ends in nothing; the trail leads somewhere you cannot get to; or twenty minutes ran out.
Take 30 minutes. This is the last thing the course asks, and it needs the file you wrote in lesson 1.
Read your first account again, now, for the first time since you wrote it. Then decide what to do with that memory. Three endings, and all three are legitimate. Write whichever you choose in four lines.
One: treat it as it stands. You are going to go on holding it as an account of what happened. Say why, in terms of something other than how vivid it feels: a document, another person, a consequence that followed and is still checkable. That is a real check and most accounts do not have one. And say what you are accepting, which is that a detailed account written years afterwards has been built more than once.
Two: treat it as an account you cannot check. Not false, not unreliable, just unverified, which is the status of most of what anybody remembers. Say what changes: probably how you would report it to somebody else rather than whether you believe it. The cost is real, and it is that you give up a certainty you had this morning.
Three: find out. Name one external thing that could corroborate or contradict part of it, and what you will do to look. A photograph, a message, a receipt, a person who was there. Then write what you will do if it contradicts you, before you look, because deciding that afterwards is the thing this whole course is about.
Then one line, whichever you chose: what would have to be true for this to have been the wrong call?
The course doesn't recommend one of the three. That is decision 6 and it holds here: the evidence covers what memory does in general and doesn't reach your particular event, so the choice is yours and the reasoning is what is being asked for.
Connections
Back. This lesson uses the whole course. Lesson 1's scope table is habit two's ancestor and the first account is the exercise above. Lesson 2 supplies the one-subject row and the worked claim about forgetting half an hour's learning. Lesson 3's unsettled mechanism is the checkpoint's loosest item. Lesson 4 is the provenance case and the two-claims sort. Lesson 5 is the false-as-stated sort and the shape worth keeping. Lesson 6 is the row with no figure in it. From earlier Core courses: the sort is Digital Literacy lesson 8's, in Using AI Effectively lesson 11's refinement of it, and Focus and Deep Work lesson 7 is the nearest relative of this one, with the instrument question where this lesson puts the sample question.
Forward. Note-Taking and Personal Knowledge is next in this term and is unwritten. It will need the reconstructive account from lesson 1, because the case for writing things down rests on it.
Go deeper
- The Relationship Between Eyewitness Confidence and Identification Accuracy: A New Synthesis (Psychological Science in the Public Interest, 2017). Read in part by this course, as lesson 5 says. If you read one thing from this reading list, read its Summary: it is the clearest example this course met of a field revising a consensus without pretending the old one was foolish. - Forgetting as a Consequence of Retrieval: A Meta-Analytic Review (Psychological Bulletin, 2014). Read in part, as lesson 3 says. Worth opening for how carefully its authors describe a mechanism they haven't settled.
Sources
- Jaap M. J. Murre and Joeri Dros, "Replication and Analysis of Ebbinghaus' Forgetting Curve", PLOS ONE 10(7), 2015. Read in part: the abstract verbatim and the summary statements of method and result. Supports: one subject, seventy hours, six delays, and savings on relearning as the measure. Lesson 2 carries the full account.
- Kou Murayama, Toshiya Miyatsu, Dorothy Buchli and Benjamin Storm, "Forgetting as a Consequence of Retrieval: A Meta-Analytic Review of Retrieval-Induced Forgetting", Psychological Bulletin 140(5), 2014. Read in part. Supports: that it pools many studies, and that its authors describe the inhibition account as largely supported with challenging evidence. Lesson 3 carries the full account.
- Elizabeth Loftus and John Palmer, "Reconstruction of automobile destruction", Journal of Verbal Learning and Verbal Behavior 13, 1974. The original was not opened. The figures reach this course through a published teaching summary, which lesson 4 says at the point of use and this lesson repeats. Supports: a hundred and fifty participants in the second experiment, and a week between viewing and the later question.
- John Wixted and Gary Wells, "The Relationship Between Eyewitness Confidence and Identification Accuracy: A New Synthesis", Psychological Science in the Public Interest 18(1), 2017. Read in part: the Summary verbatim, one box, the section introducing the five recommendations, and one field-study passage. Supports: that the pessimistic view was reasonable for the procedures that produced it, and that the claim holds for adults, for initial identifications, under five named conditions. Lesson 5 carries the full account and the qualifications.
- Isabella C. Wagner and colleagues, "Durable memories and efficient neural coding through mnemonic training using the method of loci", preprint. Read in part: the abstract verbatim and the participant counts. Supports: about fifty people, three arms, and that the abstract reports no figure for how much recall improved.
- The sample question is this course's own addition to the institute's standing sort, and the reading of what each source licenses in the table's last column is the course's reasoning rather than any source's. The four worked claims are specimens written for this lesson rather than quotations from anywhere, and the body says so where they are introduced. The three books named are on the research file's unread line: none was opened, the build fails on a citation to any of them, and nothing here assesses their contents.
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.