Error Can Improve or Disrupt Learning

You answer a question incorrectly. The correct answer appears, and the difference is so striking that you remember it long afterwards.

In another situation, you encounter an incorrect answer, learn that it is wrong and still repeat it later. The correction was available, but the error remained familiar enough to return.

Both outcomes are possible because memory does not preserve only what is true. It preserves what was attempted, attended to and retrieved. An error can make corrective information more distinctive, but it can also make the incorrect response more accessible.

The difference cannot be explained by the error alone. It depends on the learning sequence surrounding it: what the learner tried to retrieve, how the incorrect response related to the target, what corrective information followed and what remained available later.

An error is not automatically a learning opportunity

An error shows that current performance departed from a relevant standard. It does not, by itself, show that learning occurred.

The learner may not recognise the discrepancy. The correct alternative may never be supplied. Feedback may be misunderstood or forgotten. The original response may remain more familiar than its correction.

Even when someone immediately replaces a wrong answer with the right one, that correction does not demonstrate durable learning. The new answer may disappear quickly, apply only to the original question or remain difficult to retrieve without the feedback still visible.

Evidence of learning requires some relevant change to persist. That change might appear as better delayed recall, fewer repetitions of the original error, improved discrimination between similar alternatives or successful use of the corrected understanding in another situation.

Errors can contribute to such changes, but only as part of a larger learning event.

An unsuccessful attempt changes what follows

When people try to answer a question before seeing the correct response, they search memory, activate possible answers and establish an expectation about what the answer might be. This attempt can change how the correct information is processed.

Kornell, Hays and Bjork (2009) asked participants to attempt answers they were unlikely to retrieve successfully before being shown the correct responses. Later memory was better than when participants simply studied the questions and answers together. Richland, Kornell and Kao (2009) found a related pretesting effect: attempting questions before learning the material could improve subsequent learning even when the initial responses were unsuccessful.

The failed response was still a failure at the moment it occurred. Its potential value came from how the attempt altered the next encounter with the information.

The retrieval attempt may focus attention on the question, activate relevant knowledge or expose uncertainty about what belongs in the missing position. The correct answer then arrives in relation to an active search rather than as an isolated fact.

This is the principal error–correction sequence examined here, not a universal model of every way in which people learn from mistakes. The evidence shows that trying to retrieve an answer can sometimes prepare the learner to encode subsequent corrective information more effectively. It does not show that any guess will do so.

Useful errors are connected to correction

What the learner retrieves during an unsuccessful attempt matters.

In laboratory tasks using related verbal material, an incorrect response may activate information connected to the target. Corrective feedback can then distinguish between nearby alternatives. An unrelated response may provide less useful preparation and become another item that must later be separated from the correct answer.

Huelser and Metcalfe (2012) found that generating errors improved later retention when cues and targets were related. When they were unrelated, the same benefit did not appear. Grimaldi and Karpicke (2012) similarly found that retrieval attempts improved later cued recall for related word pairs but not unrelated pairs. When retrieval was constrained towards a particular incorrect candidate, it could impair later recall.

These findings suggest that an unsuccessful attempt can activate a set of possible answers from which the correct response later becomes easier to encode. They do not establish that every related error helps or that semantic relatedness explains the entire effect.

Metcalfe and Huelser (2020) found evidence that the benefit can also depend on remembering the original error–correction episode. The learner may retain not only the correct answer but the contrast between the attempted response and the feedback that followed it. Under some conditions, remembering that contrast can help the learner select the correct response later.

The error becomes useful when the later memory preserves its relation to accurate corrective information—not merely the incorrect response itself.

Surprise can strengthen correction

A strongly held error might seem especially difficult to change. Because the incorrect response is accessible and endorsed with confidence, it could be expected to interfere with correction.

Yet some laboratory findings show the opposite pattern.

Butterfield and Metcalfe (2001) found that errors made with high confidence were more likely to be corrected on a later test than errors made with low confidence. This became known as the hypercorrection effect.

One possible explanation is that feedback contradicting a high-confidence response is especially surprising or attention-grabbing, making the correction more distinctive. The result does not establish surprise as the sole mechanism.

Nor does it mean that strongly held misconceptions are generally easy to revise. A mistaken belief embedded in identity, social relationships or an explanatory worldview is not equivalent to a wrong answer on a general-knowledge test.

The more limited conclusion is that confidence does not always protect an error from correction. Under some conditions, a strong discrepancy between expectation and feedback may make the corrective event more memorable.

Incorrect information can also be learned

The processes that make an error memorable can preserve the wrong information as well as the correction.

Multiple-choice testing demonstrates the tension clearly. Retrieving information during a test can strengthen later memory. At the same time, incorrect alternatives expose the learner to plausible responses that may later feel familiar.

Roediger and Marsh (2005) found that multiple-choice testing improved later performance overall but also produced negative effects. Participants sometimes supplied earlier incorrect alternatives on a later test. Exposure to more lures reduced the positive testing effect and increased the production of false answers.

The same procedure therefore supported correct learning while also creating false knowledge.

An incorrect answer can influence later retrieval simply because it was considered. The learner may remember the information without accurately remembering whether it appeared as the correct answer or as an alternative that should have been rejected.

Remembering an answer and remembering its status are different achievements.

Feedback must do more than signal failure

Recognising that a response is wrong does not reveal what should be selected instead.

A learner who receives only the message "incorrect" has gained information about the current answer but may still lack an accurate alternative. The feedback has identified a discrepancy without resolving it.

Mullet and Marsh (2016) demonstrated this distinction in research on false memories. Conditions that made errors noticeable supported correction most effectively when the feedback also supplied the correct information. Knowing that an error had occurred was not sufficient on its own.

Feedback can mark an answer as wrong, provide the correct response and make the difference between them visible. An explanation may also help the learner understand why the alternatives should not be treated as equivalent.

Butler and Roediger (2008) found that feedback after multiple-choice testing increased later correct recall and reduced intrusions from incorrect alternatives compared with receiving no feedback. Both immediate and delayed feedback produced benefits in their study.

This should not be interpreted as literal deletion of the error. The incorrect response may remain available. Learning may instead strengthen the correct answer, improve discrimination between competing responses or make accurate selection more likely later.

Correction can therefore improve retrieval without erasing everything that was previously wrong.

The learner changes the effect

The same error-and-feedback sequence will not affect every learner in the same way.

Prior knowledge shapes which responses can be generated and whether a correction can be understood. A learner with relevant knowledge may produce an answer close to the target and use feedback to refine an existing distinction. Someone without that structure may receive the same answer as an isolated fact, with little basis for understanding why it is correct.

Memory capacity also matters.

Baddeley and Wilson (1994) found that errorless learning was especially beneficial for people with amnesia. They proposed that explicit memory helps eliminate previously generated errors. When that capacity is substantially impaired, incorrect responses may create interference that the learner has difficulty resolving.

This does not show that errorless learning is superior for everyone. It establishes a boundary on generalisation. Findings in which corrected errors benefit neurologically typical learners cannot be transferred automatically to people whose memory impairments make the error–correction relation difficult to retain.

Errors are not productive independently of the learner's capacity to process them.

Learning is what remains later

An immediate correction tells us that performance changed in the presence of feedback. It does not tell us what will be available tomorrow or in another context.

Evidence of learning becomes stronger when the learner can later retrieve the correct answer, avoid reproducing the original error, distinguish between similar alternatives or apply the corrected understanding under changed conditions.

These outcomes are not interchangeable. Delayed recall demonstrates retention across the tested interval. Reduced intrusions show that the incorrect response is less likely to dominate retrieval. Transfer requires the learner to use the corrected understanding beyond the original item.

Much of the evidence on errorful learning concerns memory for particular answers. Those findings matter, but they should not be extended automatically to conceptual understanding or broad behavioural change.

Errors can improve learning when an active attempt makes accurate corrective information more noticeable, distinctive or easier to integrate. They can disrupt learning when the incorrect response becomes familiar or retrievable without an adequately processed correction.

The difference lies in what the learning sequence leaves behind.

An error can expose a discrepancy. Feedback can supply accurate information. Attention, prior knowledge and memory can connect the attempted response with its correction. But only later performance reveals whether the correction became usable—or whether the error remained the stronger answer.