Leveling and Sharpening: What a Retold Story Loses and Keeps

Leveling is the loss of detail, and it is the largest and fastest of the changes. Each person who passes a story on keeps fewer items than the person before them, and the drop is steepest at the start: the first retelling throws away more than any later one. Names, numbers, times, qualifications and the small facts that made the account specific go first, because the teller is working from what the story was about rather than from a recording of it. The result is shorter, tidier and easier to repeat, which is part of why it travels.

Sharpening is what happens to the few details that are left. Allport and Postman defined it as the selective perception, retention and reporting of a limited number of details from a larger context, and they pointed out that it is the unavoidable other side of leveling: when most of a story is gone, whatever remains is by contrast featured. A detail that was one of twenty becomes one of four. It gets a bigger share of the telling, it attracts emphasis, and the words used for it drift toward the more vivid end: said becomes shouted, several becomes a crowd, a dent becomes a wreck.

Their third process, assimilation, decides which details are kept and sharpened. It is the pull of the teller's own habits, interests and expectations: items shift so that the story fits a familiar shape, numbers move to round figures, and things that belong together in the teller's mind end up together in the telling. The three are not separate mechanisms working in turn. They are three descriptions of one process that reduces an unclear event to something simple enough to carry.

The underlying reason is that memory for a story is not a copy of it. Bartlett's account, which the rumor work built on, is that people hold the gist of material against what they already know and rebuild a version of it at recall, filling gaps with what would make sense. That is why the loss is not neutral: a dropped qualification cannot be recovered by the next person in the chain, and a sharpened verb is passed on as though it had been in the original.

Gordon Allport and Leo Postman reported their rumor experiments to the New York Academy of Sciences in November 1945 and published the full program as The Psychology of Rumor in 1947. The method was a chain. A slide of a semi-dramatic scene with a large number of related details was projected on a screen. Six or seven people who had not seen it waited next door. The first came in, stood where the screen was out of view, and heard someone describe the picture in about twenty details. A second person then came in and heard the story from the first, a third heard it from the second, and so on down the line, with every teller instructed to report as accurately as possible what they had heard.

Plotting the percentage of the original details that survived each reproduction produced a curve much like Ebbinghaus's forgetting curve: steep at the start, then flattening. Averaged over 11 experiments, about 70 percent of the details were eliminated over five or six retellings, even though almost no time passed between them. The procedure was run with more than forty groups, from college students to police officials in training. Leveling never ran all the way to nothing: the last few items stabilized and were passed on intact.

Allport and Postman listed the limits of their own setup, and they matter when the figure is quoted. Nearly all the runs happened in front of an audience, which made tellers cautious: without an audience, subjects gave on average twice as many details. The instruction to be accurate is not the motive people have in real gossip, listeners could not ask questions, and almost no time passed between hearing and telling. The experiments are also widely miscited, which Treadway and McCloskey documented in 1987.

The predecessor is Frederic Bartlett's Remembering, published in 1932. Bartlett had readers reproduce an unfamiliar Native American folk tale, The War of the Ghosts, either in a chain of people or repeatedly by the same person over weeks and months, and described the same shortening and reshaping toward what his English readers found plausible. Bergman and Roediger repeated his procedure under controlled conditions in 1999, testing people 15 minutes, one week and six months after reading the story, and found what he described: less came back as time passed, and more of what did come back was substantially distorted.

The rows below are places where an account is passed by people rather than written down, not findings from the studies cited here.

Leveling cannot be switched off, because no retelling carries everything. What can be controlled is how many hands the account passes through and whether the surviving details get sharpened on the way.

• Shorten the chain. Send people to the original account rather than to your version of it. One link removed is the single largest saving available.
• Write it down early. A written record stops leveling at the point it was made; everything after that is a reader's problem rather than a teller's.
• Pass on the numbers and the qualifications first. These are what the process drops soonest, so they need the protection of being stated plainly and early.
• Watch the verbs when you retell something. Said, asked and walked are the words that sharpening turns into shouted, demanded and stormed. If you cannot remember which word was used, say that instead of choosing one.
• When an account matters, ask the person who told you where they got it, and keep asking until you reach someone who was there.

Several neighboring entries describe memory going wrong, and they differ in what causes the change.

• It is not false memory, a recollection of a whole event or detail that never happened. Leveling and sharpening describe what repeated telling does to an account that is real; a false memory needs no chain of tellers at all.
• It is not the misinformation effect, where information encountered after an event alters what a person remembers of it. There the distortion comes from outside, in a leading question or a later report; here it comes from the reduction that telling performs on its own.
• It is not suggestibility, the tendency to take up another person's version when it is offered. Suggestibility is about accepting an outside account; sharpening happens in the teller's own.