Why corrections do not work

The obvious response to a false story is a true one. Publish the correction, show the evidence, and the belief should go away. Anybody who has tried this knows that it often does not, and the usual explanation offered is that people are stubborn or that they do not care about the truth. Neither explanation survives contact with the research.

The first difficulty is arithmetic. A false story that travels widely is seen by a very large number of people; the correction is seen by a fraction of them, usually days later, in a quieter place. Even a correction that works perfectly on everyone who reads it leaves most of the original audience untouched. Much of what looks like resistance is simply absence.

The second difficulty is that remembering a correction means remembering two things: the claim and the fact that it was wrong. These decay at different rates. The claim is a simple, vivid statement and the tag saying it is false is an extra piece of bookkeeping attached to it, so as the weeks pass the tag is the part that fades. The result is a person who recalls the claim clearly, recalls hearing something about it, and cannot reconstruct which way round the something went.

This points to what does work. Corrections that repeat the false claim prominently in order to deny it tend to strengthen the memory of the claim. Corrections that lead with the true account, and that explain what actually happened rather than only announcing that something did not, perform considerably better. The rule of thumb among researchers is to give people a replacement rather than a deletion, because a story with a hole in it is unstable and the hole tends to be filled by the thing that was removed.

None of this is an argument against correcting things. It is an argument about where the effort should sit. A correction is a late intervention competing against a head start, and the people best placed to limit the damage are not the fact-checkers but whoever decides what gets amplified in the first hours. That is an uncomfortable conclusion, because it moves the problem from a technical question that specialists can work on to an editorial one that somebody has to own.