Fact-Checking Doesn't Backfire. 52 Issues, 10,100 People, Zero Boomerangs

For more than a decade, the reigning advice about misinformation has been a counsel of despair: do not correct people, because facts backfire. The idea entered circulation in 2010, when Brendan Nyhan and Jason Reifler reported that conservatives shown corrections about the absence of Iraqi weapons of mass destruction ended up more convinced that such weapons had been found, and science communicators absorbed the lesson. Guides warned that arguing with the misinformed would entrench the falsehood, and fact-checking playbooks were written around the fear of the boomerang, with a whole industry of consultants learning to open every talk by warning that the audience's good intentions would only make things worse.

The boomerang was mostly a phantom, as Thomas Wood of Ohio State and Ethan Porter of George Washington University discovered when they set out to find the effect under the conditions where it should thrive. Their paper, "The Elusive Backfire Effect," published in Political Behavior in 2019, reports five experiments enrolling more than 10,100 subjects, pitting corrections against 52 separate issues chosen precisely because they were the polarized kind where backfire was supposed to appear, yet not one correction triggered it, and beliefs moved toward the facts even when the facts cut against people's partisan attachments.

The design was almost adversarial toward its own hypothesis: subjects included committed partisans, and the corrections targeted false claims made by politicians those subjects supported, which is exactly the configuration where the original theory predicted the strongest boomerang. In an earlier conference version of the project, four experiments with 8,100 people across 36 issues managed to produce exactly one backfire, on the original Iraq WMD question. Even that collapsed under inspection: when the same question was phrased more simply, the backfire vanished, revealing a wording artifact rather than a psychological law.

The most striking development is that the original finding's own author signed the retraction of the doctrine. Nyhan later collaborated with Wood and Porter on a follow-up that fact-checked Donald Trump during the 2016 campaign and found no backfire. Then, in the Proceedings of the National Academy of Sciences in 2021, he wrote that the 2010 study "has often been interpreted to mean that these effects are widespread" when "subsequent research suggests that backfire effects are extremely rare in practice," a walk-back that carries unusual weight because it comes from the man whose name the effect bears.

A 2022 study in the Journal of Experimental Psychology: General explains why the phantom kept appearing. Briony Swire-Thompson and colleagues ran two nearly identical experiments, with 388 and 532 participants, testing corrections against 21 misinformation items in a longitudinal pre/post design. No item backfired more than a no-correction control. What predicted apparent backfire was not how much people cared about the topic, and not how strongly they believed the falsehood; it was how unreliable the survey question was. Backfire rates correlated strongly and negatively with test-retest reliability, at −0.61 and −0.73 across the two experiments, which means the less trustworthy the question, the more the data resembled a boomerang, while worldview-related attributes like perceived importance and strength of prior belief showed no such relationship. Reviewing the literature, the authors found that 81 percent of reported backfire effects came from single-item measures, the noisiest kind in the survey toolbox. Random wobble in a flaky question reads as a boomerang if you squint hard enough. That is the whole story.

The corrections themselves worked, and not subtly, since in the 2022 experiments the correction condition updated beliefs dramatically, with an effect size (ηp² = 0.46) that towers over typical persuasion research. The control group, exposed to the claims without any correction, drifted slightly in the wrong direction, an illusory-truth effect of about d = 0.22, the small cost of merely repeating a falsehood. Across the extended Wood-and-Porter program, as summarized in a 2023 review, corrections improved belief accuracy on more than 96 percent of 63 tested issues and backfired on none. Here is arithmetic no one has run against these numbers: that is at least 60 corrections that worked for every zero that backfired. Read that again. Meta-analyses point the same way, since a 2017 synthesis of 20 studies found that misinformation degrades accuracy while fact-checks restore it.

The strongest case against this cheerful picture starts with a distinction the data insists on. Facts move; attitudes mostly don't. When Nyhan, Porter, Reifler and Wood fact-checked Trump during the 2016 campaign, his supporters accepted the corrections and updated their factual beliefs, while their feelings about Trump did not move at all. A 2023 review concluded that corrections and misinformation both have, at most, very small downstream effects on political attitudes. If the goal is changing someone's vote or loyalty, a fact-check is a weak instrument, which is why the honest headline for this whole literature is narrower than the cheerful version that travels on social media: corrections fix beliefs, not allegiances, and anyone promising more is selling something the data does not support.

Second, corrections fade. In recontact studies, roughly 40 percent of the accuracy gain was still detectable two weeks later, first across three countries and then across ten. In a smaller two-country study of COVID-19 claims, the gains did not persist at all, which means a correction is a treatment with a half-life rather than a cure. Third, the continued-influence effect survives intact. Even corrected falsehoods keep shaping people's reasoning afterward, as a 2022 review in Nature Reviews Psychology documents. And fourth, context is doing quiet work in the background. In a 2018 experiment, exposing partisans to the other side's views on social media for a month made Republicans more conservative rather than less, which suggests that a correction delivered as partisan combat is a different animal from a correction in a survey, and that the real world mostly serves the combative kind.

One more finding deserves attention, because it refines how to correct rather than whether to. Repeating a myth before debunking it raises the myth's familiarity, the illusory-truth drift the 2022 study measured at d ≈ 0.22 in its control group. That does not mean corrections backfire, because the same study found that corrections presented without any prior exposure to the falsehood still reduced false beliefs. It means format matters: lead with the fact, keep it short, explain why the falsehood is wrong, and do not open by quoting the lie at length.