whitehatstoicAll articles

Study Shot

Hypercorrection Effect: Why Confident Errors Get Fixed

The hypercorrection effect: a wrong answer you were sure of is fixed by feedback more often than a guess. How it works, where it fails and one example.

By Miguel de Guzman · · 8 min read

Study Shot

Turns a photo of a page from a textbook, notes or a worksheet into multiple-choice practice questions, from the basics to advanced, with every answer explained.

Website

The hypercorrection effect is the finding that a wrong answer you were sure of is more likely to be fixed by feedback than a wrong answer you doubted. It was first reported in November 2001, and it comes with a catch found ten years later: after a week, the confident error is also the one most likely to come back.

The finding matters to anyone who studies with practice questions. Study Shot turns a photo of a page from a textbook, notes or a worksheet into multiple-choice questions and explains every answer, so each miss is followed at once by its correction. The research says which misses that correction repairs with little effort, and which ones need a second look.

How the hypercorrection effect is measured

The basic experiment has four steps.

  1. A person answers a question from memory.
  2. The person says how sure they are of that answer.
  3. The correct answer is shown.
  4. Later, the same question is asked again.

The measure looks only at the answers that were wrong at step 1. It sorts them by the confidence given at step 2, then counts how many are right at step 4. Hypercorrection is the name for one pattern in that count: the surer the person was of the wrong answer, the more often the second answer is right.

In one version, by Butler, Fazio and Marsh at Duke University, the questions were general knowledge about science, and half of the people answered again at once while the other half waited a week. Notice what the measure leaves out. It says nothing about whether confident people are right more often. It compares two kinds of error, after both have been corrected.

A prediction that failed in 2001

The effect was found by people who expected the opposite. Butterfield and Metcalfe, at Columbia University, started from what most models of memory imply. High confidence means the answer that came to mind is strongly tied to the question. A strong tie gets in the way of rival answers more than a weak one does. So an error held with high confidence should be the hardest to replace.

Their paper, in the Journal of Experimental Psychology: Learning, Memory, and Cognition in November 2001, reports that the result ran the other way. Highly confident errors were the most likely to be corrected when the question came round again. A second result sits beside it. When people produced both the right answer and their old error at the second attempt, they nearly always picked the right one. The paper's title gave the pattern its name: errors committed with high confidence are hypercorrected.

Why a confident error is easier to fix

Two explanations have evidence behind them, and they work at different moments.

The first is surprise. Being wrong when you felt sure is a jolt, and a jolt draws attention to what comes next. Fazio and Marsh ran two experiments on this in a paper of February 2009. People remembered what the feedback looked like more often when it did not match what they expected, and they remembered both its surface details and its content better. The correction is the same for everyone. How closely it gets read is not.

The second is that a confident error often sits close to the right answer. Metcalfe and Finn, in March 2011, asked whether people who say "I knew it all along" after a correction are fooling themselves. They were not. After a confident error, people more often picked the right answer on a second guess, and needed fewer letters of it as a hint, than after an unsure error. The authors conclude that the knowledge was really there, and that it helps explain why those errors are corrected so easily.

A brain imaging study of July 2012 by Metcalfe, Butterfield, Habeck and Stern supports both at once. Any mismatch between confidence and outcome activated the anterior cingulate and the medial frontal gyrus, which fits the account from attention. Only confident errors also activated the right temporoparietal junction and the right dorsolateral prefrontal cortex, which the authors read as holding the false belief beside the true one and pushing the false one down.

A worked example: ten questions from one page

Say you photograph a biology page on photosynthesis and take a quiz of 10 questions, each with four choices. Keep a slip of paper beside you. Before you tap each answer, write S if you are sure and G if you are guessing.

You get 6 right and 4 wrong. Two of the misses are marked S and two are marked G. One of the S misses is the question "Where in a plant cell does photosynthesis take place?" You tapped the mitochondria without a doubt. The answer is the chloroplasts, and the explanation appears straight away.

The research sorts those four misses into two piles.

  • The two S misses. These are the ones most likely to be right next time. You already knew that a plant cell has both mitochondria and chloroplasts, and the surprise made you read the explanation closely. Today they need nothing more.
  • The two G misses. No surprise came with these, so the explanation got less attention, and there is less nearby knowledge for the answer to attach to. These need work today: go back to the page and find the sentence each answer comes from.

Now follow it forward a week. The quiz is kept with its page photo, so you take it again. Check the two S questions first. If the correction has faded, the answer that returns may well be your old one, the mitochondria. If you get an S question wrong twice, treat it as the most stubborn item on the page, not the easiest.

The 6 right answers get the least time. In a study of 258 people learning Luganda and English word pairs, Pashler and colleagues report that showing the right answer after a wrong response raised what was remembered a week later by 494%, while feedback after a right response made little difference.

The limit: confident errors come back

The strongest objection comes from the one-week version of the experiment. Butler, Fazio and Marsh, in December 2011, report the effect on both the immediate and the delayed second attempt. But fewer errors stayed corrected after a week. When a correction was lost, people sometimes gave their first wrong answer again, and confident errors were more likely to be given again than unsure ones.

Both results hold together. A confident error is the most likely to be corrected and, once the correction is forgotten, the most likely to return. The correction does not erase the old answer. A week later it can still be the one that comes to mind.

This is also where people most often misread the effect. It does not say that strong beliefs are easy to change. It covers a factual error that is followed by the correct answer, and it only compares that error with a less confident one. Take the correct answer away and nothing in these experiments applies.

What is still unsettled

Three questions are open.

How far it reaches outside the laboratory. Carpenter and colleagues at Iowa State University noted in 2018 that the effect had mostly been studied with trivia questions. They ran the four steps in an introductory horticulture class and saw a significant effect, with a tendency for students who knew more of the material to be surer of their errors and to correct them more. That is one course. How the effect behaves with questions written from a learner's own page has not been reported in these studies.

How much of it is confidence at all. In a study of kanji word reading with 15 and then 40 participants, Iwaki and colleagues report that once they took account of how useful each person judged a word to be, the tie between confidence and later correction shrank. Part of the effect may come from how much a person values getting that item right.

What kind of correction works. In a two-year study with two groups of about 88 algebra students each, Metcalfe and colleagues report that sessions built around students' own errors produced more learning for the teacher's time than direct teaching did. The gain varied from teacher to teacher. Getting students to work out why the error happened did more than telling them the right solution.

Frequently asked questions

Does the hypercorrection effect mean you should answer with confidence on purpose?

No. The experiments recorded the confidence people actually reported. They do not show that acting sure changes anything.

How long does a hypercorrected answer last?

In the one-week experiment, the effect was still there after seven days, but fewer errors stayed corrected than on the same day. Errors that were not corrected were sometimes given again, confident ones more often than unsure ones.

Does feedback help after a right answer?

Much less. In the word-pair study of 258 people, feedback after a correct response made little difference, whether or not the person was confident.

Has the effect been seen in a real class?

Yes, in at least one: an introductory horticulture class in 2018, where college students answered questions on course content and corrected their confident errors more often than their unsure ones.

What the hypercorrection effect changes about a wrong answer

Before 2001, the models of memory implied that a confident wrong answer was the hardest kind to fix. The hypercorrection effect splits the wrong answer in two. A sure miss is quick to repair on the day and fragile a week later. An unsure miss draws less attention when it is corrected and has less to hold on to. How sure the person was at the moment of answering turns out to be information about what the error needs, and it is lost unless it is recorded before the answer is revealed.

Study Shot

Turn any page into a practice quiz. Visit studyshot.vlvd.net

Study Shot on desktop
Study Shot on tablet
Study Shot on phone