Confidence and knowledge are supposed to travel together. In one experiment published in Science Advances in 2022, they pulled apart in a way that had a price. When 1,173 US adults were offered the chance to bet that they would beat the average on a science knowledge test, the people most opposed to the scientific consensus on the issue they were asked about tended to be the most overconfident, and they earned less as a result.
The experiment was the third of five studies in a paper led by Nicholas Light with Philip Fernbach, Nathaniel Rabb, Mugur Geana and Steven Sloman. Across the studies, the team compared three things: how strongly people opposed the expert consensus on a contested scientific issue, how much they actually knew, and how much they believed they knew.
This is one research group’s work, not settled consensus. But it sits on top of a decade of related research, some of which held up and some of which did not, and that history says as much about forming opinions from evidence as the headline finding does.
What the bet measured
The first two studies set up the pattern. According to PsyPost’s report on the paper, 3,249 US adults recruited through online survey platforms were each asked where they stood on one of seven issues on which scientists broadly agree: climate change, genetically modified foods, nuclear power, vaccination, evolution, the Big Bang and homeopathic medicine. Those who fully agreed with the scientific consensus were excluded, leaving 2,133 participants who rated their own understanding of the issue on a seven-point scale running from “vague” to “thorough.” Then they answered 34 true or false questions on general science, items such as whether the center of the earth is very hot.
People who were more opposed to the consensus were more likely to claim a thorough understanding. They also tended to score worse on the test.
A self-rating costs nothing, so the third study added stakes. Participants could bet on their ability to score above average on the knowledge test. Those with greater opposition to the consensus tended to earn less, and the authors attribute the shortfall to knowledge overconfidence: they bet on a performance they did not deliver.
Two further studies, run in July and September 2020, found similar patterns for willingness to take a COVID-19 vaccine, before one was publicly available, and for attitudes toward mitigation measures such as mask wearing and social distancing. In the September study the pattern held after the researchers accounted for political identity. In all, the five studies recruited more than 5,600 participants, before exclusions.
The same gap appeared in an earlier study of GM foods
The 2022 paper extended work the same group had published in Nature Human Behaviour in 2019. In that study, led by Fernbach, more than 2,000 adults in the United States and Europe rated how well they understood genetically modified foods and then took a true or false test on science and genetics. More than 90 percent expressed at least some opposition to GM foods. As opposition became more extreme, test scores fell and self-rated understanding rose.
The paper’s abstract puts it plainly: “Extreme opponents know the least, but think they know the most.” The pattern repeated in representative samples from the United States, France and Germany, and in a separate study of attitudes toward gene therapy.
It did not repeat for climate change.
Light’s team found the same exception in 2022. “The main caveat is that although this pattern of effects appears to be fairly general, we don’t find it for all issues,” Light told PsyPost. “One notable example is climate change.” The researchers also saw some evidence that on more politically polarized issues, the link between opposition and low knowledge was weaker. One plausible reading, which the authors of the 2019 paper raised, is that on polarized topics people’s positions follow their group more than their grasp of the details.
What the studies do not show
These are correlations. The design cannot say whether knowing less leads people to oppose a consensus, whether opposing it leads them to stop learning, or whether something else drives both. The samples were recruited online, and the knowledge test measured general science literacy rather than expertise on each specific issue.
The findings also describe tendencies across many people. They do not mean that any particular person who disagrees with a scientific consensus is uninformed, and plenty of people who accept the consensus could not explain the science behind it either.
The authors draw a practical conclusion of their own. Because the strongest opponents tend to be the most overconfident, they wrote, “fact-based educational interventions are less likely to be effective for this audience.” Someone who already feels they know all there is to know has little reason to listen.
A neat fix for overconfidence that failed to replicate
If overconfidence props up strong opinions, puncturing it might soften them. In 2013, Fernbach and colleagues reported exactly that in Psychological Science. In three small online experiments, Americans who were asked to explain step by step how policies such as a flat tax or cap-and-trade would work rated their own understanding lower afterward, and their positions became modestly less extreme. Simply listing reasons for a view did not have the same effect.
It was an appealing result. Then other researchers tested it.
In 2021 Jarret Crawford and John Ruscio published three preregistered replications, meaning the methods and analyses were fixed in advance, with 306, 405 and 343 participants. People again reported understanding the policies less after explaining them. Their positions were no less extreme. The authors concluded that explaining “increases people’s recognition of their ignorance but is unlikely to increase their political moderation, at least under these conditions.” A 2024 study in Political Psychology by Richard Walker and colleagues, with 5,139 people in Hong Kong, went further and found extremity was higher when people set out their understanding of a policy.
The research on overconfident opinion was, in other words, corrected by the same process it was studying. A plausible idea was put to a test, the test was repeated, and the results proved mixed: a 2022 study found reduced extremity for policies framed around consequences but not for those rooted in protected values.
Expert opinion gets tested too
That process applies to professionals as well as the public. In 2013 Vinay Prasad, then at the US National Cancer Institute, and colleagues reviewed 2,044 original articles published in the New England Journal of Medicine between 2001 and 2010, of which 1,344 concerned a medical practice, reporting in Mayo Clinic Proceedings. Of the 363 that tested an established medical practice, 146, or about 40 percent, found the practice was no better than, or worse than, the alternative it was compared with. Another 138 reaffirmed the practice, and 79 were inconclusive.
The reversals included stenting for stable coronary artery disease, which for most patients did no better than medication, and hormone therapy for postmenopausal women intended to protect the heart, which did worse than no treatment. The figure describes one journal over one decade, and Prasad noted that the reversals were not the final word on each practice. It is not a rate for medicine as a whole.
What it does show is that confident professional opinion, held by people with real expertise, was overturned again and again when someone ran the trial.
What evidence asks of an opinion
Read together, these studies point in two directions at once. The gap between feeling informed and being informed was widest among the people most sure they were right, at least on several of the issues tested. And science itself kept revising its claims, about medicine and about the psychology of opinion, whenever a careful test disagreed with what people expected.
Basing an opinion on science, on this evidence, means more than citing a study that agrees with it. It means checking how much is actually known, how strong that evidence is, and being prepared to hold a view a little more loosely when the next test comes in.
