The misinformation effect: how Loftus rewrote a stop sign as a yield sign

The misinformation effect is the change in a memory that happens when a person meets misleading information after an event. In Elizabeth Loftus's classic experiments, people watched a filmed or pictured car accident and then answered questions that carried a suggestion. Later, some of them reported details that were never in the event. One changed verb, or one mentioned road sign, was enough to move the answers.

The APA Dictionary defines it as a phenomenon in which a person mistakenly recalls misleading information that an experimenter provided, instead of the correct information presented earlier. It concerns memories of specific events, the kind covered in episodic versus semantic memory. This article covers the two studies behind the name, the standard design, the competing explanations, and what is known about warnings and eyewitnesses. It also separates the effect from four things it is often confused with.

• In 1974, Loftus and Palmer found that the verb "smashed" produced higher speed estimates than "hit" or "contacted". A week later, more of the "smashed" group said they had seen broken glass that was not in the film.
• In 1978, Loftus, Miller and Burns found that misleading information made people less accurate about what they had seen. It did more damage when it arrived just before the test than right after the event.
• Studies of the effect usually use three stages: an event, later information, and a memory test.
• Whether the original memory is overwritten, blocked or still intact but outcompeted is still debated.
• Warnings reduce the effect on average but do not remove it, and a confident, detailed memory is not proof of an accurate one.

The loaded verb: Loftus and Palmer (1974)

In the first experiment of Loftus and Palmer (1974), 45 students watched seven short films of traffic accidents. After each film they answered a questionnaire. The critical question asked how fast the cars were going, and the verb changed between groups: nine students got each of "smashed into", "collided with", "bumped into", "hit" and "contacted".

One word moved the estimates by about nine miles an hour. The authors noted that the real speed of the cars controlled little of the variation in answers, while the phrasing controlled a good deal. Four of the films were staged crashes at 20, 30, 40 and 40 mph, and the mean estimates for them ranged only from 36.1 to 39.7 mph.

The second experiment asked whether the verb changed what people remembered about the scene, not only the number they gave. A total of 150 students watched a film, under a minute long, of a multiple-car accident. Fifty were asked how fast the cars were going when they "smashed into" each other, fifty were asked with "hit", and fifty were not asked about speed. One week later they answered a list of questions that included "Did you see any broken glass?" There was no broken glass in the film.

In the "smashed" group, 16 of 50 said yes. In the "hit" group it was 7 of 50, and in the control group 6 of 50. Most people in every group answered correctly, so the verb shifted the odds rather than rewriting every memory. The authors also checked whether the verb only raised the speed estimate, which then made glass seem likelier. For a given speed estimate the verb still made a difference, though a smaller one.

Loftus and Palmer proposed that two kinds of information go into a memory of an event: what was perceived, and what was supplied afterwards. Over time the two can be integrated so that a person cannot tell from which source a detail comes. All the person has is one memory.

The stop sign that became a yield sign: Loftus, Miller and Burns (1978)

Loftus, Miller and Burns (1978) tested the idea across five experiments and a pilot study with 1,242 people in total. Participants saw slides of a single car-and-pedestrian accident. They then received information about it that was consistent, misleading or irrelevant, and finally took a recognition test.

The best-known detail was a traffic sign: people who had seen a stop sign were later given information that mentioned a yield sign instead. Misled participants were less accurate on both a yes-no test and a two-alternative forced-choice test. The damage was larger when the misleading information came just before the final test than when it came right after the event. The authors concluded that information supplied after an event, whether consistent or misleading, is integrated into the witness's memory of it.

The sign experiment has been run again. In Wiechert and colleagues (2026), whose authors include Loftus, participants watched a car accident video made in Grand Theft Auto and then read co-witness statements with consistent, misleading or no information about a traffic sign. Misled participants reported the correct sign less often than controls.

The three stages of a misinformation study

Studies of the effect usually follow the same three-stage procedure, which Loftus describes in her 2005 review. The effect is the gap between the misled group and the controls.

The 1978 tests were recognition tests, in which the answer is in front of the participant. Recall versus recognition explains how the two kinds of test differ.

Why it happens: four explanations

No single explanation has won. Each of the four below has some evidence behind it.

Blending

This is the account Loftus and Palmer gave. The original information and the later information merge into one memory, and the person cannot separate them. The title of the 1978 paper points the same way: "Semantic integration of verbal information into a visual memory".

What happens to the original memory?

Loftus's 2005 review asks what happens to the original memory once misinformation is accepted: is it impaired, perhaps because its trace is altered, or is it only made harder to retrieve? Three positions answer differently, and the 2026 replication discussed below uses their names.

• Overwriting, the single-trace view. The misinformation replaces the original detail, so only the false version is left in memory.
• Blocking, or retrieval impairment. The original detail is still stored, but the misinformation makes it harder to reach.
• Coexistence, the activation view. The original detail is untouched and both versions remain in memory, but the later one is more active, so it is the one that gets reported.

McCloskey and Zaragoza (1985) argued for the no-impairment position: misinformation does not change the original memory at all. In their view it affects people who never stored the detail, or who remember both versions and choose the misinformation after deciding it must be correct. Their evidence was a modified test that leaves the misinformation out as an answer. In Loftus's illustration, a person who saw a burglar pick up a hammer and was told it was a screwdriver chooses between the hammer and a wrench, a new item. On that test people were very good at picking the original item.

Loftus's review reports that later analyses still found small effects on the modified test. Her summary is that sometimes there is no original memory to compete with, sometimes the person deliberates and picks the misinformation, and sometimes the original memory seems to be impaired. The 2026 replication reversed the order of the original and misleading events and found no effect on a binary test. Its authors read this as favouring single-trace or blocking explanations over activation-based ones, but warn that the result may come from the order itself.

Source misattribution

Zaragoza and Lane asked people detailed questions about where suggested items came from. As Loftus's review reports, misled people sometimes came to remember seeing things that were only suggested to them. They called this the source misattribution effect, and noted that it varies in size and is not inevitable. The skill involved is source monitoring, the process of determining the origins of one's memories, knowledge or beliefs.

Guessing and demand

Another worry is that participants are guessing or trying to please the experimenter. Loftus and colleagues argued in 1978 that a simple demand-characteristics explanation could not account for their results. The 2026 replication used a four-option test with confidence-based betting. The effect shrank but remained, so the authors conclude that demand effects are not the sole driver. Loftus's review also reports effects in pigeons, which suggests the effect is not only a matter of pleasing an experimenter.

What makes the effect stronger or weaker

• Delay. Misinformation is more effective when time has weakened the original memory. Loftus's explanation is the discrepancy detection principle: a memory is more likely to change if the person does not immediately notice a conflict between the new information and what they remember. The forgetting curve describes how memory fades with time.
• The source. In Dodd and Bradshaw (1980), leading questions with no stated source led people to "remember" the presupposed facts, but attributing the questions to a lawyer representing the defendant eliminated the effect. In a second experiment the misleading details were adopted from a neutral bystander's account but not from the account of the driver who caused the accident.
• Warnings. According to Loftus's review, a warning given before the misinformation sometimes helps people resist it, while a warning given afterwards helped only in limited circumstances. The Blank and Launay (2014) meta-analysis of 25 post-warning studies found that, on average, post-warnings reduced the effect to less than half of its size. Some kinds of warning worked better than others.

Why it matters for eyewitness testimony

Loftus's review lists three ways misinformation reaches a witness: talking with other witnesses, leading questions, and media coverage. Each can put a detail into an account that the witness never saw. Small details such as a sign or an object are what these studies measure, and they are the kind of detail that can matter when an account is weighed.

A confident account is not a reliable guide. The same review notes that false memories, once embraced, can be expressed with confidence and detail, and that, as of 2005, there was no reliable way to tell from a single memory report whether it was real. This article describes how memory behaves in experiments. It is not a way to judge any particular witness, and it is not legal advice.

What the misinformation effect is not

• Retroactive interference. Retroactive interference is newer learning making older learning harder to retrieve. Loftus's review reports the argument that the misinformation effect is more than that, because interference only disrupts performance while misinformation pulls toward a specific wrong answer.
• False memory in general. The APA Dictionary defines a false memory as a distorted recollection of an event or, most severely, a recollection of an event that never happened. The misinformation effect changes details of an event the person did witness. Planting a whole event, as in the "lost in the mall" technique, is a different and stronger procedure. Across studies using it, an average of about 30% of participants produced a partial or complete false memory, according to Loftus's review.
• The Mandela effect. This is an internet label for shared and consistent false memories of specific icons in popular culture. In Prasad and Bainbridge (2022), 100 adults consistently made the same false-memory errors for certain images. Further experiments found no attention or viewing differences that explained the errors, and a separate sample of 50 adults made the same kind of errors spontaneously during recall. The procedure was not the controlled post-event design described above.
• False memory syndrome. The APA Dictionary notes that the label is controversial, as is the evidence about recovered memories of childhood abuse, and that false memory syndrome is not an accepted diagnostic term. A laboratory effect on ordinary details cannot show whether any one person's memory of a serious event is accurate. A psychotherapist or doctor is the right person to talk to about memories that are distressing.

Using these terms when you study

Misinformation effect, source misattribution, source monitoring, retroactive interference, false memory and eyewitness testimony sit close together and are easy to mix up. In Psychology World you can save them to one list and drill them as flashcards, as multiple-choice questions once the list has four or more words, or by typing the answer out. What each study mode does is on the home page. For the methods with the best evidence, read our guide to memorising psychology terminology.