The reason an anecdote survives a refutation

The statistic is withdrawn, the study is retracted, and the man who lost everything after the policy change is still the thing everybody remembers. He was an illustration of the retracted finding. He outlived it by years.

What everyone sees

A refutation is expected to remove whatever it refutes, on the model of correcting an entry in a ledger. Memory does not work as a ledger. An episode is encoded with imagery, sequence and causal texture, and a correction is a proposition — the two are not stored in a form where one can overwrite the other.

What is actually happening

Nisbett and Ross documented the disproportionate weight of vivid, concrete case information relative to statistically superior aggregate information, with the vividness advantage surviving explicit warnings about its unrepresentativeness. Johnson and Seifert examined the continued influence effect directly, finding that people who accepted a correction and could recall it still used the retracted information in later inferences, and that corrections were most effective when they supplied a replacement account rather than merely negating the original. That is the operative point. A retraction without an alternative explanation leaves the episode intact and the causal gap open, so the discredited story continues to be the only available account of what happened.

Why it stays hidden

The persistence hides because people can report the correction accurately. Ask directly and they will state that the claim was withdrawn. The retracted material shows up somewhere else — in what they consider plausible, in which risks feel live, in what they mention when the topic recurs — and none of those are places anyone checks for consistency.