Backfire effect

The sugar rush

A parent has told other parents for years that sugar makes children hyperactive, and runs sugar-free birthday parties because of it. A friend shares a summary of double-blind studies in which children’s behavior didn’t differ after sugar or a sugar-free placebo. The parent replies: “I’ve seen it with my own eyes. If anything, this makes me more sure.”

The studies are evidence against the belief, and nothing in them supports it, yet the parent’s confidence goes up. That is the move the worldview version of the backfire effect describes: a belief tied to someone’s identity or past choices gets defended harder when it’s challenged. This is the claimed pattern, not a typical one: in studies, people in this position more often stay put or move at least partly toward the correction (see Evidence). The claimed backfire is the increase.

The myth-versus-fact poster

A clinic waiting room has a poster: “MYTH: We only use 10% of our brains. FACT: Brain scans show activity throughout the brain.” A week later, a visitor who hadn’t heard the claim before remembers something about “only using 10% of your brain” and thinks it was one of the facts.

This is the claimed familiarity version: repeating a myth in order to correct it makes the myth more familiar, and familiar statements feel true. Nobody is defending an identity here. The failure would be a correction that leaves people more likely to believe the myth than if they had never seen the poster. Tests built to give it the best chance, with myths people hadn’t heard before and a week’s delay, have mostly found no such increase.

• Worldview backfire: a correction that threatens a belief central to someone’s values or group identity is said to strengthen it. This is the version from the famous 2010 study.
• Familiarity backfire: repeating the myth while correcting it is said to spread or strengthen it.
• Overkill backfire: giving too many counterarguments is said to make a correction less convincing. As of a 2022 review, the one study to test it found the opposite, as long as the counterarguments were relevant.

• When belief stays the same. Not being persuaded is not backfire. A correction that fails leaves belief where it was; backfire requires it to rise.
• When the correction comes from a source you have good reason to distrust. Discounting a claim from a source with a poor track record or a stake in the answer is reasonable, whichever way the claim points.
• When the correction turns out to be weak. If you examine a challenge carefully and find it mistaken, a belief that survived a serious test can fairly gain a little confidence.
• When the correction contains evidence for the belief. Some “debunkings” concede the main point. Updating on what they actually show is just updating.

The test: did the correction give you a reason to be more confident, or only a reason to defend what you already thought?

Unconvinced, not more convinced

Someone who believes goldfish have a three-second memory reads a short post saying that goldfish can learn and remember tasks for months. They say: “Maybe. It’s one post, though, and I’d want to see the actual research before I change my mind.”

Their belief hasn’t gone up; they’ve declined to update on a single unsourced post and said what would change their mind. That may be too cautious, but it isn’t backfire. This is the belief stays the same condition above. Research finds this outcome, and partial updating, far more often than backfire.

Discounting a source with a bad record

A cook has always been told not to wash a seasoned cast-iron pan with soap. An account that sells enamel cookware, and has been caught posting made-up cooking “facts” before, posts that the soap rule is a myth. The cook ignores it and keeps washing the pan with water only.

The soap rule may well be outdated, but the cook’s reason for not changing is the source, not a desire to defend the habit, and the belief didn’t grow stronger. When you would be accepting a claim on someone’s say-so, their reliability is relevant: the same testimony condition that makes some attacks on a source legitimate rather than an ad hominem. Researchers reviewing this literature note that discounting a correction from a low-credibility source can be quite rational.

The backfire effect was proposed as an extreme form of motivated reasoning. People tend to scrutinize unwelcome evidence more harshly than welcome evidence, a relative of Confirmation bias, and when a belief is tied to their identity they may generate counterarguments against a correction. The claimed step beyond that is that the counterarguing overshoots and leaves the belief stronger. The familiarity version rests on a different, well-established finding, the illusory truth effect: statements you’ve encountered before feel truer. But that finding concerns repetition without a correction; when a false claim is clearly labeled false, studies rarely find the familiarity winning.

Two things that are real can make backfire seem common. People often don’t update much, and when they do, the change can fade until they are back where they started; neither is backfire. And apparent backfire in studies can come from measurement: a belief measured with a single noisy question, especially within small subgroups, will sometimes go up after a correction by chance alone.

Research on corrections has tested several practices:

• Don’t hold back a correction for fear of backfire. Reviews of the evidence conclude that corrections rarely increase false beliefs, including in the groups once thought most at risk.
• Explain what is true, not just what is false. A plausible alternative explanation reduces reliance on a myth more than a bare “that’s not true”. In classic experiments on corrected explanations of an event, supplying another cause reduced reliance on the discredited one, where simply negating it did less.
• Give detail. Refutations that explain why a claim is wrong generally work better than plain retractions.
• Repeating the myth once is fine. Stating the myth, clearly flagged as false, alongside the correction does not appear to backfire and can help people notice the conflict. Needless repetition is still best avoided.
• Expect partial, fading effects. Corrections often reduce belief without erasing it, and the gain can decay over time or be outweighed by repeated exposure to the original claim.

For yourself: if a correction has made you more sure, look for the part of it that counted as evidence for your view. If there isn’t one, the extra confidence came from somewhere else.

Status: failed. The central claim, that corrections make people’s false beliefs stronger, has not held up in large and preregistered tests, and recent reviews, including one by the original study’s lead author, treat backfire as rare in practice and not a robust phenomenon. What is well supported is weaker and different: corrections often work only partly, and their effect can fade.

• Nyhan and Reifler (2010) gave undergraduates mock news articles containing a misleading claim by a politician, with or without a correction, on contested U.S. political questions (whether Iraq had weapons of mass destruction before the 2003 invasion, whether tax cuts increased government revenue, and whether there was a ban on stem cell research). The main finding was that corrections often failed to reduce misperceptions among the ideological group inclined to believe the claim. Backfire, meaning a correction increasing the misperception in that group, appeared in two of the experiments: the first weapons experiment and the tax experiment. When the authors re-ran the weapons experiment in simplified form, the correction reduced the misperception among conservatives, the group inclined to believe it; only a small post hoc subgroup, those who rated the issue most important, showed no benefit. The stem cell correction had no effect in the relevant group, but did not backfire. The paper describes four experiments and mentions a fifth, with no substantive results, only in a footnote; Nyhan later summarized the result as backfire in two of five studies.
• Wood and Porter (2019) ran five experiments with more than 10,100 participants, including an online panel matched to national demographics, on 52 polarizing issues chosen to favor backfire. They found no correction that triggered it: people across the political spectrum moved toward the facts, though sometimes less when the correction challenged their own side. Re-running the original weapons article, they found that the original, multi-part survey question left respondents unmoved by the correction rather than pushing them the wrong way, while a simpler question about the same fact showed more updating. They concluded the original backfire was partly a consequence of that question’s wording, and explicitly did not conclude that backfire is impossible.
• Porter and Wood (2021) ran preregistered experiments simultaneously in Argentina, Nigeria, South Africa and the United Kingdom, testing 22 fact-checks. Fact-checks reduced false beliefs in every country, most effects were still detectable more than two weeks later, and the authors found no instances of backfire.
• Swire-Thompson, DeGutis and Lazer (2020) reviewed both the worldview and familiarity literatures and concluded that backfire effects “are not a robust empirical phenomenon”. They pointed out that many reported backfires used a single survey question, which is noisy, and appeared in subgroups. With unreliable measures, regression to the mean (extreme answers drifting toward average on re-measurement) can produce spurious backfire, especially among people who started with low belief.
• Swire-Thompson and colleagues (2022) tested this directly. In two experiments (388 and 532 participants), people rated 21 myths before and three weeks after a correction, and a control group re-rated them with no correction. No myth backfired more in the correction group than in the control group; in the first experiment, belief increases were nearly twice as common among people who saw no correction. How often an item appeared to backfire was strongly associated with how unreliable its measurement was, and not with how important or strongly held the belief was. Features related to familiarity did correlate with it, but did not consistently explain anything beyond reliability.
• The familiarity backfire effect traces largely to an unpublished study of a “myths versus facts” flu vaccine flyer, which a later attempt failed to replicate directly. Ecker, Lewandowsky and Chadwick (2020) gave it a favorable test: corrections of entirely novel false claims, measured a week later. Their first experiment found some evidence of backfire, but two preregistered follow-ups did not replicate it. Across the experiments, the main reasoning measure favored no effect by roughly nine to one, while the belief ratings were inconclusive. They concluded it is safe to repeat misinformation when correcting it.
• Nyhan (2021) reassessed his own finding. He argues that it was often misread as showing that all corrections are counterproductive, or that backfire explains why false beliefs persist; that later research finds backfire “extremely rare in practice”; and that corrections typically improve accuracy at least somewhat, but their effects often decay or are overwhelmed by cues from political elites and the media promoting more congenial but less accurate claims.
• Ecker and colleagues (2022), a review in Nature Reviews Psychology, conclude that worldview backfire effects have proven difficult to replicate, partly because of unreliable methods, and should not generally discourage debunking; and that a myth’s familiarity contributes to its lingering influence but does not typically produce backfire.

Not a failure to replicate: the continued influence effect. In Johnson and Seifert’s (1994) experiments, people read an account in which information explaining an event was later corrected, and they went on using the discredited information in their inferences; supplying a plausible alternative cause reduced this. In the standard version of the task, people keep relying on the corrected detail even when they can recall the correction. This persistence is well established. It describes a correction that reduces a myth’s influence without erasing it. That is different from backfire, where the correction would leave belief stronger, and should not be confused with it in either direction.

The popular version goes further than the original research ever did: that facts make people dig in, so correcting people is pointless or harmful. The 2010 paper reported backfire only in specific groups on specific issues, and Nyhan has since written that its findings support neither of those broader readings.

What remains uncertain. Backfire hasn’t been shown to be impossible. Occasional findings in particular subgroups are still reported, and with noisy measures it is hard to tell a rare real effect from chance. Many studies measure belief within days or weeks of a single correction, so less is known about how corrections hold up over months of repeated exposure to the original claim.