Wrong on Purpose: Errors as Information
An error made while attempting retrieval, followed by the correct answer, strengthens memory — and the most confident errors, once corrected, are the best remembered of all. · 10 min
For most of the twentieth century, instruction treated errors as contamination. The behaviorists warned that a wrong answer, once produced, might be stamped in and repeated forever, so lessons were engineered to keep mistakes from happening at all. The modern memory laboratory has run that experiment, many times, and the verdict goes the other way. An error made while genuinely trying to remember — and then met with the correct answer — helps the very memory it seemed to threaten. This folio lays out the two findings behind that verdict, and why this Archive treats your most confident miss as its most teachable moment.
Guess before you learn
Before you study a chapter on a topic you know nothing about, you are made to answer a pretest on it — and you get nearly every question wrong. Compared with spending those same minutes on extra reading, what does the failed pretest do to your memory of the chapter?
It helps. In studies by Kornell, Hays, and Bjork (2009) and by Richland and colleagues (2009), people who guessed wrong and then saw the right answers remembered more than people who spent the same time studying without guessing. A failed attempt is not a wasted attempt — it prepares the ground the answer lands on. If you chose 'hurts,' you reasoned exactly like half a century of educators, and like most people meeting this result for the first time.
9–12
3–5
Guess first, even when you do not know. Suppose you guess that whales breathe water, then read that they breathe air. Your wrong guess does not stick — the correction does, and it sticks better because you guessed. Scientists have tested this again and again: a wrong guess followed by the right answer beats reading the right answer alone.
One rule keeps it safe: always check. A guess that never meets the real answer teaches you nothing — and cannot be corrected.
6–8
Pretesting means attempting questions before you study the material. It feels pointless — you miss most of them — yet in experiment after experiment the failed attempts improve memory for the answers once they arrive. Kornell, Hays, and Bjork (2009) had people guess the endings of word pairs they could not possibly know, then see the answer; those pairs were later remembered better than pairs simply studied for longer. The attempt prepares the memory. The error does not stick — provided the correction follows.
The same holds during practice: a wrong answer met with feedback leaves you better off than a cautious blank. The only error that teaches nothing is the one never risked, or never checked.
9–12
The stranger finding is Butterfield and Metcalfe's (2001). People answered general-knowledge questions and rated their confidence in each answer. After feedback, the errors most likely to be fixed on a retest were the ones held with the highest confidence — the hypercorrection effect. A high-confidence error means two things at once: this is territory you know well, and one belief in it is wrong. The correction arrives as a surprise; surprise commands attention; and attention is what encoding runs on (folio 2).
This Archive is built to that finding. Answer SURE and miss, and you receive an errata slip — a correction worth reading twice — set louder than an ordinary correction and never attached to a penalty. The moment you were surest and wrong is the moment memory is most ready to be rewritten.
K–2
Try to guess what sound the new letter makes before your teacher says it. If your guess is wrong, listen close to the real sound. Trying first helps the real sound stay.
Wrong guesses are not bad. A guess, then the answer, sticks better than the answer alone. Guess first. Then check.
Undergrad
Why do failed attempts help? The search itself does encoding work: attempting retrieval activates knowledge related to the question, and the answer, when it arrives, is bound into that activated network rather than stored in isolation — the same elaborative machinery behind the testing effect (folio 5). Grimaldi and Karpicke (2012) sharpened the claim: pretesting pays when guess and answer are semantically related, which is why an informed wrong guess helps and a random one does not.
Hypercorrection has two supported accounts: high-confidence errors occur where prior knowledge is rich, so the feedback connects to much that is already stored; and the mismatch between expectation and outcome drives attention to the correction. Surprise and prior knowledge each appear to carry part of the effect.
Postgrad
Metcalfe's 2017 Annual Review synthesis states the reversal plainly: errorful generation with corrective feedback outperforms errorless study, undoing a Skinnerian presumption that shaped instruction for decades. The boundary conditions are narrow. The benefit requires feedback; it survives — and in some designs grows with — delayed feedback; and it holds across ages, though hypercorrection attenuates in older adults, consistent with a dulled novelty response.
Mechanism weighting stays open: semantic-activation accounts of pretesting (Grimaldi and Karpicke, 2012) sit beside attention-to-surprise accounts of hypercorrection — Fazio and Marsh (2009) found enhanced memory even for the incidental context of surprising feedback. The instructional prescription converges regardless: solicit attempts before teaching, collect confidence ratings, and spend the richest correction where confidence was highest.
pretesting
Attempting questions on material before studying it. The attempts mostly fail; memory for the studied answers improves anyway.
Why is this true?
Why does a wrong guess, corrected, beat reading the right answer from the start?
The attempt activates what you already know around the question, so the correction is woven into that prepared network instead of arriving cold — more routes lead back to it later.
Read the slope again, because it reverses common sense. The errors you were surest about are — once corrected — the ones most likely to be gone by next week, provided the correction gets your full attention when it arrives. That is the design brief behind this Archive's errata slip — a correction worth reading twice. Mark an answer SURE and miss, and the correction comes back set louder and asks to be read twice, because the evidence says this moment is the best memory-forming chance you will get all day. Nothing is deducted. The attempt already did its work.
To collect these effects on purpose: before a chapter, write three questions from its headings and answer them cold — wrong answers welcome. During practice, commit to an answer before checking, because only a committed guess can be corrected. And attach an honest confidence to what you produce — SURE, THINK SO, or GUESSING — since confidence is what turns an ordinary miss into a hypercorrection. The one error that teaches nothing is the one you never let yourself make.
Note
The confidence stops — SURE / THINK SO / GUESSING — exist for this folio's reason: an answer with a confidence attached is an answer that can be hypercorrected.
Errors, then, are not the opposite of learning; uncorrected errors are. The next folio leaves the single session behind and asks a scheduling question: holding total hours fixed, does it matter when they happen? The measured answer — the spacing effect — may be the cheapest improvement this course has to offer.
Practice — new ink and old, interleaved
1.Which group predicted they would remember more?
2.A practice answer you marked GUESSING comes back wrong. Compared with a SURE miss, what does the evidence predict?
3.You are about to read a section titled 'Why the Moon always shows one face.' Write the pretest question you would attempt cold, before reading.
4.From memory: why did errorless instruction lose to errorful practice with feedback?
Attempts — even failed ones — prepare encoding by activating related knowledge, and corrected errors, especially confident ones, are remembered better than answers received without an attempt. Errors only harm when they go uncorrected.
How close were you? Grade yourself honestly — it sets your review date.
5.Which of these is retrieval practice?
6.From folio 1: attempting an answer and failing, then reading the correction — which act did the attempt itself exercise?
7.On the week-delayed test, the rereaders recalled about 40% of the passage's ideas. About what percentage did the repeated-recall group keep?
8.From folio 5: after a week, Roediger and Karpicke's rereaders kept about 40% of the passage's ideas. About what percent did the repeated recallers keep?
9.From folio 4: pretesting feels useless while rereading feels productive. What is that feeling actually tracking?
10.Which change most improves the accuracy of your judgments of learning?