Is the Mandela Effect Real?

The name “Mandela effect”, for a false memory shared by many people, is credited to Fiona Broome, a self-described paranormal researcher. She found that other people shared her false memory of Nelson Mandela dying in prison in the 1980s. Britannica dates her finding to 2009 and Dictionary.com to 2010; Wordorigins.org, an etymology site, says there is no evidence that she originated the phrase.

Online communities have since collected other examples: the Berenstain Bears spelled “Berenstein”, the Monopoly man wearing a monocle, Darth Vader saying “Luke, I am your father”, and a cornucopia in the Fruit of the Loom logo. Some people in those communities take the errors as signs of parallel worlds or other realities.

Memory errors are not new to science. Roediger and McDermott, in their 1995 word-list study, note that Bartlett’s experiments, published in 1932, are usually credited as the first on false memories. The question here is narrower. Do many unrelated people really share the same false memory, and if they do, does ordinary memory explain it?

In August 2022 YouGov asked 1,000 US adult citizens about nine well-known examples, offering a common false version beside the accurate version. For seven of the nine, more people chose the false version than the accurate version. The effect was weakest for its namesake: only 13% chose Mandela dying in prison in the 1980s, against 57% who chose his death in 2013. YouGov notes that its questions did not ask about memories directly, though many people presumably answered from memory.

Laboratory studies show three things about memory errors. A single word in a question can change what people report seeing. Parts of a childhood event can be planted by suggestion. And people can recall, with confidence, a word they never heard. These studies use films, childhood stories and word lists, not famous images, but they show the ordinary processes that any explanation of the Mandela effect starts from.

In 1974 Elizabeth Loftus and John Palmer showed students films of car crashes. In their first experiment, 45 students estimated how fast the cars were going. Those asked about the cars that “smashed into” each other gave a mean of 40.8 mph, against 34.0 mph when the verb was “hit”. In a second experiment, 150 students were asked a week later whether they had seen broken glass, which the film did not show. Of the 50 asked earlier about “smashed”, 16 said yes. Of the 50 asked about “hit”, 7 did.

In the “lost in the mall” study of 1995, Loftus and Jacqueline Pickrell gave 24 adults four short stories from their childhood, supplied by an older relative. Three were true; one, about being lost in a mall or large department store at about the age of 5, was invented. At both follow-up interviews, 6 of the 24 claimed to remember it, fully or partially, and they rated those memories as less clear than their true ones.

How often a whole false event takes root is disputed. A replication in Ireland, published in 2023 by Murphy and colleagues, was preregistered: its methods and analysis plans were posted publicly before the data were collected or analysed. It coded 35% of its 123 participants as reporting a false memory of getting lost.

Bernice Andrews and Chris Brewin reanalysed the same transcripts. They estimated that only 5 participants could reasonably be said to have a false memory, rather than the 43 the replication counted. None of the 43 had recalled an entirely false event, and half had been lost before and were likely describing a real event from another time or place.

Wade, Loftus and colleagues replied in 2025 that memory is reconstructive, so reports should be expected to mix remnants of true memories with knowledge and speculation. They added that even low rates of false memory matter.

A pooled analysis gives a figure between the replication’s 35% and Andrews and Brewin’s 4%. In 2017 Scoboria and colleagues coded reports from eight implantation studies, with 423 participants. They classified 30.4% of cases as false memories.

Source monitoring is how people judge where a memory came from. A 1993 review by Johnson, Hashtroudi and Lindsay argued that these judgments rest on qualities of the remembered experience and “are subject to error and disruption”. The review applies this to eyewitness testimony and to the incorporation of fiction into fact.

Fuzzy-trace theory is set out in a 2016 review by Valerie Reyna, Charles Brainerd and colleagues. It holds that memory keeps two independent records: a verbatim trace of exact details and a gist trace of meaning. The review describes gist as a relatively stable form of memory, which it says is why false memories can outlast true ones.

Errors also spread between people. In a 2001 experiment by Roediger, Meade and Bergman, a participant and a confederate looked at household scenes and then took turns recalling items. The confederate acted as a second participant but was part of the experiment, and named some items that were not there. Participants later recalled more of those items than in a control condition with no suggestion. This extra recall happened especially with suggested items that fitted the scene, such as a toaster in a kitchen.

A 2025 study in Cognitive Science by Valle and colleagues had laboratory networks of participants recall the same kind of word lists together by computer. Clustered networks were split into two tight groups joined through one member of each; in the others, members were linked across the whole network. Networks that were not clustered produced more false memories, while clustered networks reinforced true ones. The study tested word lists, not online communities, so its bearing on Mandela effect forums is an inference it did not test.

Download this conclusion: the laboratory evidence

This section as one image. It shows the word-list experiment, with the missing word recalled 40% and 55% of the time. It charts the “lost in the mall” studies, at 25%, 35% and 4%, and the pooled rate of 30.4% across eight studies. The image is 1,080 by 1,350 pixels.

Download the card (PNG, 1,080 × 1,350)

Ordinary explanations have been offered for several of the examples. The ending “-stein” is far more common than “-stain”. The Census Bureau lists every US surname held by at least 100 people in 2010. Of those surnames, 246 end in -stein and 11 in -stain. “Berenstein” is on that list; “Berenstain” is not. Wordorigins.org suggests that people read the familiar ending when they see the rarer one.

Schema theory, the idea that memory fills gaps with what usually goes together, would explain the monocle, since people associate it with wealth. Mr. Peanut, the Planters mascot first drawn in 1916, was given one, along with a top hat and cane (Hagley Museum and Library). For Mandela, Wordorigins.org suggests that people may be conflating him with Steve Biko. Biko, an anti-apartheid activist, died in police custody in 1977, the 46th person to die in detention (South African History Online).

Shared exposure can also spread an error. The fact-checking site Snopes found a monocled Monopoly man on the banknotes of at least one Monopoly Junior edition and in a 2016 post on the official Monopoly Facebook account. The misquote “Luke, I am your father!” is spoken in the 1995 film Tommy Boy. In the laboratory, people adopt each other’s errors, as Roediger, Meade and Bergman showed.

These explanations do not account for everything. In 2022 Deepasri Prasad and Wilma Bainbridge of the University of Chicago showed 100 adults 40 famous logos and characters. Each appeared in its correct version beside two altered ones. For 7 of the 40, including Pikachu, the Monopoly man and the Fruit of the Loom logo, people consistently chose the same wrong version, and were confident in it.

Tracking where people looked found no differences in attention that could explain the errors. The authors conclude that the errors occur “despite the majority of visual experience being the canonical image”, meaning the correct version.

The finding has been repeated. In 2024 Transparent Replications, a Clearer Thinking project that runs rapid replications of recently published psychology studies, repeated the 40-image test with 389 people. The replication found the effect for the same 7 images and one more. It also found an error in one of the original analyses; once corrected, the original data showed the effect for 6 images.

Schema theory does not fit every case. Offered the Fruit of the Loom logo with a cornucopia or with a plate, people chose the cornucopia, although plates are more often associated with fruit. The researchers ruled out schema theory as a universal explanation. When the University of Chicago reported the study in June 2022, the researchers had not yet found a single reason why the errors happen.

Corrections often fail to undo a false belief. A 2012 review by Lewandowsky and colleagues set out why retractions of misinformation are “so ineffective in memory updating”. It also set out why efforts to retract misinformation can even backfire and increase misbelief. The same review describes techniques that do reduce the effect of misinformation.

Neither an exceptional memory nor a night’s sleep reliably protects against false memories. In 2013 Patihis and colleagues tested people with highly superior autobiographical memory, who retrieve abundant and accurate memories of their own lives, against controls matched for age and sex. Both groups falsely recognised the missing words in the word-list task. In a misinformation task, people saw an event in photographs and later read an account of it with misleading details. The people with superior memory reported more false details than the controls.

Sleep, and the lack of it, can both work in favour of false memories. In 2014 Frenda and colleagues found that 24 hours without sleep increased false memories in a misinformation task when people were deprived of sleep while they took in the event. When the deprivation came afterwards, it had no significant effect. In 2009 Payne and colleagues found that a night of sleep increased both true and false recall of word lists, compared with a day awake. They also found that sleep preserved the false memories preferentially.

But a 2019 meta-analysis of word-list studies by Newbury and Monaghan found no overall effect of sleep on true or false memories. Whether sleep raised or lowered them depended on the task.

Yes, as a shared memory error. For seven of nine popular examples in YouGov’s 2022 poll, more US adult citizens chose the false version than the accurate version. In a controlled study of 40 famous images, people made the same confident, specific error for 7 of them, and a larger replication found it again.

The Nelson Mandela Foundation, the publisher of the Berenstain Bears, Fruit of the Loom and Snopes document the accurate versions, and none of their records shows a change. The parallel-world explanation offers no testable prediction. Laboratory research shows how ordinary memory produces errors and spreads them between people. It does not yet explain why a few images draw the same specific error from so many people, when most of what people see is the correct version.

Download the conclusion

The verdict above as one image, with the YouGov poll shares as its chart: 62%, 61%, 58% and 13% choosing the false version. It is 1,200 by 630 pixels, the size social networks use for link previews, and anyone may download it and post it.

Download the card (PNG, 1,200 × 630)