Persistence of misinformation
A Socratic walk-through of persistence of misinformation — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does a correction so often fail to undo a false story?
A newspaper prints a claim, then prints a retraction. Ask readers afterwards whether the original claim was withdrawn and many will tell you correctly that it was — and then, asked to explain what happened, reach straight back for it.
So the question is sharper than "why do people believe false things?" It is: why does a piece of information keep doing work in someone's reasoning after they have accepted it is false?
Reasoning it through
REASONING #Start with what you are actually storing when you hear a story. Not a list of facts — a little model of an event, with things in it that explain other things. A warehouse burned; there were cans of paint and gas cylinders in a side closet; the fire spread ferociously. The claim about the closet is not a loose item. It is the joint holding the ferocity to the fire.
Now issue the correction: there were no volatile materials in the closet. What has it done to the model? It has removed a load-bearing part and put nothing in its place. And you are still going to be asked why the fire spread so fast.
Here is the step worth pausing on. Which does a mind prefer — a model it knows is wrong in one spot, or a model with a hole where the explanation should be? The experimental answer, from a line of work beginning with Johnson and Seifert in the early 1990s and replicated many times since, is uncomfortable: the flawed model wins. Retracted information keeps being used in inference, and it keeps being used by people who correctly report that it was retracted. Researchers call this the continued influence effect, and it is one of the sturdier findings in the area.
That immediately suggests a test. If the trouble is the hole, then filling it should help more than widening it. And it does: a correction that supplies an alternative causal account — it was arson, and here is what was found — reduces continued reliance substantially more than a correction that only negates. This is the most practically useful thing known about the whole problem. A retraction is subtraction; people need a replacement.
Why does bare negation do so little? Partly the gap. Partly something quieter: repeating a claim, even to deny it, makes it more fluent — easier to bring to mind — and fluency is one of the cues people use to judge truth. The illusory truth effect, that repeated statements are rated truer, has been demonstrated since Hasher and colleagues in 1977, and it does not politely exempt statements repeated inside a denial.
And now the thing you have probably heard, which I want to be careful about. The "backfire effect" — the idea that correcting someone makes their false belief stronger — became famous from a small number of studies around 2010. It has since largely failed to replicate. Large-sample attempts, notably Wood and Porter's work published in 2019, found that people overwhelmingly move toward accurate information when corrected, including on politically charged claims. So the honest picture is not "correction backfires". It is "correction works, but incompletely, and the residue is stubborn."
The analogy
THE ANALOGY #Think of a criminal trial where evidence is ruled inadmissible. The judge instructs the jury to disregard it, and the jurors sincerely try. But the evidence was not sitting in a drawer to be removed — it was the reason the story they had assembled hung together. Told to strike it, they still have to explain the case somehow, and the struck evidence is the only explanation they were given.
A juror is asked to ignore something they still believe happened, whereas a corrected reader has been told the thing did not happen at all — so the persistence here is not defiance or motivated reasoning, and can occur in someone with no stake in the outcome whatsoever.
Clarifying the model
THE MODEL #Three refinements hold the pieces together.
First, this is a claim about inference, not about memory failure. The effect shows up most clearly in the gap between what people can recall about the correction and what they use when reasoning. Those are separate operations, and the correction reliably reaches the first without fully reaching the second.
Second, "supply a replacement" is a real finding, not a slogan, but it is bounded: the alternative has to be plausible and roughly as complete as what it displaces. A vague substitute does not close the gap.
Third, the strength of all this varies with the material. Corrections do less well when the false claim fits a person's prior commitments, and better when it does not — but that is a gradient in how much correction achieves, which is different from the correction making things worse. Keep the two apart. The field's own emphasis has shifted this way over the past decade, and older popular accounts have not all caught up.
A picture of it
THE PICTURE #How to readStart at the top parallelogram, the original story, which becomes a causal model. The first diamond splits on what kind of correction arrives: a bare negation leaves a gap, a replacement account fills the slot, and restating the myth to deny it adds an exposure that makes the claim feel truer. The second diamond is the moment that matters — what the person actually reaches for when asked to explain — and both the gap route and the fluency route arrive at continued influence, which is the red node. The back-edge from there shows the only reliable exit: supply an alternative cause, and reliance drops.
What became clearer
WHAT CLEARED #The reason a correction underperforms is structural, not moral. People hold events as causal models, and a retraction removes a part without replacing it — leaving a person who accepts the correction, can repeat it, and still reasons with the thing it withdrew. That reframes the practical task: not "say it is false, loudly", but "say what happened instead", and avoid restating the claim more than the correction requires. And it is worth holding the popular version at arm's length — the evidence says corrections mostly work and work incompletely, not that they rebound.
Where to go next
ONWARD #- Inoculation, or "prebunking": warning people about a manipulation technique before they meet it, rather than correcting afterwards.
- Why source credibility and the format of a correction change how much of the residue is cleared.
Key terms
TERMS #| Term | What it means |
|---|---|
| Continued influence effect | the tendency for retracted information to keep affecting reasoning even in people who correctly recall the retraction. |
| Illusory truth effect | the finding that repeated statements are judged more likely to be true, partly because repetition makes them easier to process. |
| Backfire effect | the proposed strengthening of a false belief by correcting it; famous, but largely unreplicated in later large-sample work. |
| Prebunking (inoculation) | pre-emptively exposing people to a weakened form of a misleading argument so they resist the full version later. |
Every term the collection defines is gathered in the glossary.