Backfire Effect: When Correcting a Belief Makes It Stronger

The backfire effect was proposed as the extreme end of motivated reasoning. When a correction threatens a belief that is bound up with identity, the argument went, people do not weigh it; they argue against it, and the act of generating counterarguments leaves the original belief more rehearsed and more strongly held than before. Brendan Nyhan and Jason Reifler offered this account in 2010 after a correction embedded in a news article increased, rather than reduced, a misperception among the readers most inclined to hold it.

The idea spread because it explained something people wanted explained: why facts seemed not to work on political opponents. It was repeated in journalism, in guides for fact-checkers, and in advice to avoid stating a myth at all, and within a few years it was common knowledge that correcting people backfires.

The evidence did not keep up with the reputation. Thomas Wood and Ethan Porter tested 52 corrections on more than 10,100 people across the most polarizing issues in American politics and found none that backfired; corrections moved people toward accuracy, including people whose politics the correction challenged. Briony Swire-Thompson, Joseph DeGutis and David Lazer reviewed the literature in 2020 and concluded that reports of backfire are rare, often rest on single items and small subgroups, and can arise from measurement noise, since a belief measured once with error and once again will appear to move in some people by chance. Nyhan himself wrote in 2021 that the backfire effect does not explain why misperceptions persist, and pointed instead to the supply of misleading claims from trusted elites and the limited reach of corrections.

What survives is narrower and better supported. Corrections usually work, but partially; a corrected belief tends to drift back over time; and people who heard the misinformation continue to use it in reasoning even after accepting the correction, which Stephan Lewandowsky and colleagues described in 2012 as the continued influence effect. Backfire, in the strong sense of a correction producing a net increase in the false belief, is a possibility to design against rather than a rule to expect.

Nyhan and Reifler published "When corrections fail" in Political Behavior in 2010. Their first study, run in fall 2005 with 130 undergraduates at a Catholic university in the Midwest, asked participants to read a mock news article, attributed to the Associated Press, about a 2004 campaign speech in which President Bush defended the Iraq war and spoke of the risk that Saddam Hussein would pass weapons to terrorists. Half of the participants read a version that added a correction: the Duelfer Report had found no stockpiles of weapons of mass destruction and no active program before the invasion. Everyone then rated agreement that Iraq had such weapons before the invasion.

For most participants the correction did what corrections are supposed to do. Among those who described themselves as conservative, it did the opposite. In the raw data, the share of conservatives agreeing that Iraq had the weapons rose from 32 percent in the control condition to 64 percent in the correction condition, in a group of 33; among the other 97 participants, agreement fell from 22 to 13 percent. The authors called this a backfire effect.

Their second study, in spring 2006 with a new sample, tested the generality of the result. A correction of the claim that President Bush's tax cuts had increased federal revenue backfired among conservatives in the same way: agreement rose from 36 to 67 percent in a group of 60, against a small fall among others. But the Iraq correction, repeated with slightly different wording, now reduced the misperception among conservatives too, and a correction aimed at a misperception held more by liberals, that Bush had banned stem cell research, reduced it among centrists and conservatives while failing to move liberals, with no backfire. The paper reported two backfire effects in four experiments, one of which had already failed to repeat within it.

Wood and Porter's 2019 study, also in Political Behavior, is the replication that changed the field's view. Across experiments with more than 10,100 participants and 52 polarizing issues, they found no correction that produced backfire, and concluded that by and large people heed factual information even when it challenges their ideological commitments.

Because the strong backfire effect is rare, the useful examples are the situations where a correction fails in the ordinary ways, and the situations where the fear of backfire does more harm than the risk. The rows below are illustrations, not findings from the studies cited on this page.

The practical lesson of the replication record is to correct well rather than to avoid correcting. The steps below come from the research on effective corrections.

• Correct. The large studies find that corrections move people toward the facts, including people whose politics the facts challenge. Silence out of fear of backfire leaves the misinformation standing.
• Lead with the fact, state the myth once, and explain why it is wrong. Lewandowsky and colleagues' 2012 review found that a correction works best when it supplies an alternative account, since a gap in a story is filled by the myth that used to occupy it.
• Expect partial movement and repeat. Beliefs drift back toward the misinformation over time, so one correction is rarely enough.
• Do not take a single survey item as evidence of backfire. Swire-Thompson and colleagues showed how measurement error and small subgroups produce apparent backfire where none exists.
• Keep identity out of the message where you can. The 2010 backfire appeared where a correction was read as an attack on a group; a correction that does not carry that charge, like the knuckle-cracking one above, is simply heard.

The backfire effect is one of several ways a correction can fail, and the others are far more common.

• It is not the continued influence effect, in which people accept a correction and still reason as though the original claim were true. Continued influence is robust and well replicated; backfire, a net increase in the false belief, is not.
• It is not confirmation bias, the tendency to seek and weigh evidence that supports what one already believes. Confirmation bias explains why a correction may be discounted; backfire would require the correction to push the belief the wrong way.
• It is not reactance, the push against a message that feels like an attempt to control. Reactance is a response to pressure; the backfire effect was proposed as a response to information, and the best evidence is that information is usually taken on board.