You present the data. Photographs, measurements, a study that seems to answer the disputed question. The other person does not pause or reconsider. They find a reason to dismiss the source, challenge the method or move to another claim.
It is a familiar frustration for anyone trying to explain something carefully. If the difficulty were a lack of information, more information should help. When it does not, the available explanations can become unkind very quickly: they are stupid, dishonest, incapable of reason.
I want to understand what happens between receiving evidence and deciding whether to accept it. What does the person think the evidence shows? Why do they trust or distrust its source? Would changing their mind affect their relationships with people who share the belief?
Asking about relationships can lead to a premature conclusion of its own: that someone needs a belief to keep their friends or feel secure, so no evidence could persuade them to revise it.
The research makes that conclusion difficult to sustain.
Do we demand more proof when we dislike a conclusion?
In an experiment, Charles Lord, Lee Ross and Mark Lepper gave supporters and opponents of capital punishment descriptions of two invented studies. One suggested that the death penalty reduced murders; the other challenged that conclusion. Participants tended to judge the study supporting their existing position more favourably and examine the opposing study more critically.1
Both sides encountered the same conflicting findings. One group did not have information the other lacked. Which study they found convincing differed with the position they already held.
A person can examine evidence closely while demanding more proof from a study they disagree with. Rejecting its conclusion does not necessarily mean they failed to understand it.
Being better at interpreting numbers does not guarantee that someone applies the same standards to every claim. Dan Kahan and colleagues gave participants numerical results presented either as evidence about a skin treatment or about a gun-control policy. People with stronger numerical skills did better on the skin-treatment question. On the political question, answers were more likely to fit participants’ political views, and this division was greater among those with stronger numerical skills.2
That experiment tested numerical reasoning on a politically charged question. It does not show that intelligence generally makes people less accurate. It gives us a reason to ask how someone uses their reasoning skills, alongside asking whether they have the skills to understand the evidence.
There is a question here for the person presenting the evidence too. Do I demand the same standard of proof from a conclusion I welcome and one I would rather reject? What would I accept as evidence against my own account?
What could changing a belief cost someone?
Someone may have friends who share a belief or belong to a group that formed around it. They may take pride in having understood something others missed. They may also have been misled by the institution now asking them to trust its account.
These are possibilities to ask the person about. Knowing what they believe does not tell us whether any of these circumstances apply.
Someone who has argued a position publicly may have to acknowledge an error to people they persuaded. In a group that treats doubt as disloyalty, questioning a shared belief may risk rejection. Those consequences could make reconsidering difficult; they do not establish whether the belief itself is true.
An institution may have concealed a mistake or failed to investigate a complaint. Someone describing that experience deserves to be heard. But evidence of one failure does not establish every later allegation against the institution, or make a finding from another source false.
“They have lied before” and “this particular finding is wrong” require different evidence. Keeping those questions separate allows us to take the experience seriously while still examining the claim.
In my work on TEG-Blue, I want to ask whether a belief helps someone feel less uncertain, gives them people to turn to, or explains an experience they could not previously understand. These questions concern what the belief offers the person. They do not establish its truth, reveal why the person adopted it, or show that they cannot reconsider it.
Do corrections make people more certain of a false belief?
When every fact produces another objection, it can seem that the correction is strengthening the belief. Yet someone becoming angrier during an argument is different from becoming more certain that their factual claim is true.
Thomas Wood and Ethan Porter tested factual corrections on 52 issues in five experiments involving more than 10,100 participants. None of the corrections increased belief in the false claim being corrected—the result known as a factual backfire effect—even on politically divisive subjects.3 This does not guarantee that every correction will succeed. It challenges the assumption that accurate information usually makes false beliefs stronger.
Thomas Costello, Gordon Pennycook and David Rand tested conversations with an AI system across experiments involving 2,190 conspiracy believers. Participants rated how certain they were that their chosen theory was true, on a scale from 0 to 100. Those who received evidence addressing their own reasons for belief reported an average reduction of about 20% from their starting confidence. Participants who discussed unrelated topics showed much smaller changes. In the first experiment, the reduction remained at a two-month follow-up.4
The 20% describes a change in confidence ratings, not the proportion of people who abandoned their belief. Nor does the result guarantee what will happen in a conversation with a relative. The researchers tested conversations that combined evidence and responses to the person’s specific concerns; they did not isolate which part produced the change.
Those participants were able to reconsider strongly held beliefs. One possibility is that earlier attempts to persuade people had not answered their specific reasons for believing a claim.
Before concluding that someone cannot be persuaded, I would want to know what they believe and why. Are we answering their objection to the evidence? If they distrust the organisation that collected the data, are we addressing that distrust or simply repeating the figures? Have they rejected the evidence, or are they unwilling to admit a change of mind in front of us?
Asking “What would change your mind?” gives the other person a chance to name the evidence they would accept. I would need to listen to the answer and explain what could change my own view too.
Listening does not require agreeing with a false claim or continuing indefinitely. A person may refuse to examine evidence, change the claim whenever it is answered, or use the exchange to insult or threaten someone. You can decide to end that conversation without resolving the disagreement.
An immediate agreement is not the only possible sign that someone has reconsidered. They may become less certain or begin asking a question they previously dismissed. We can keep examining the factual claim while asking what evidence each person would accept—and whether we are willing to apply that standard to our own beliefs.
Sources
Footnotes
-
Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098–2109. DOI. ↩
-
Kahan, D. M., Peters, E., Dawson, E. C., & Slovic, P. (2017). Motivated numeracy and enlightened self-government. Behavioural Public Policy, 1(1), 54–86. DOI. ↩
-
Wood, T., & Porter, E. (2019). The elusive backfire effect: Mass attitudes’ steadfast factual adherence. Political Behavior, 41, 135–163. DOI. ↩
-
Costello, T. H., Pennycook, G., & Rand, D. G. (2024). Durably reducing conspiracy beliefs through dialogues with AI. Science, 385(6714), eadq1814. DOI. ↩