Misinformation Effect: How a Question Rewrites What You Saw

Memory for an event is not a recording that is played back; it is reconstructed each time from fragments, and the reconstruction draws on everything relevant that has been stored since, including things said about the event by other people. A question that presupposes a detail, such as the yield sign in the demo, supplies a fragment with the same subject matter as the original memory, and at the time of recall the two are hard to tell apart. Loftus's account was that the new information is integrated into the memory of the event, so that a single, altered memory results.

That account was challenged early. McCloskey and Zaragoza argued in 1985 that the standard test could not distinguish an altered memory from several other things: a participant who never encoded the sign in the first place and simply follows the question, a participant who remembers both and defers to the experimenter, or one who guesses. When they used a modified test that did not offer the misleading item as an option, misled and control participants performed the same, which suggested that the original memory was still available. The debate that followed produced a more careful picture. Misinformation reliably changes what people report, and in some conditions people report the false detail with high confidence and even claim to remember seeing it; whether the original trace is overwritten, blocked or merely outcompeted is still argued, and probably differs across cases.

Several conditions make the effect larger. Time matters: misinformation does more damage when the original memory has had a chance to fade. Source matters: a detail from someone who seems to have been there or to be in authority is accepted more readily than one from an obviously unreliable source. Attention matters: a detail that was never attended to at the time is easy to supply afterward because there is nothing to compete with it.

Loftus's 2005 review of thirty years of this work found the effect with children and adults, and with slides, films and staged events.

Elizabeth Loftus, David Miller and Helen Burns published "Semantic integration of verbal information into a visual memory" in the Journal of Experimental Psychology: Human Learning and Memory in 1978. Across five experiments and a pilot study, 1,242 participants saw a series of slides showing a single accident in which a red Datsun turned at an intersection and knocked down a pedestrian. For half of the participants the intersection had a stop sign; for the other half, a yield sign.

In the first experiment, 195 students saw the slides and then answered a questionnaire about them. One question asked whether another car passed the Datsun while it was stopped at the sign, and the sign named in that question either matched the slide the participant had seen or named the other sign. After a filler activity, participants were shown pairs of slides and asked which one they had seen. Those whose question had named the sign correctly chose the right slide 75 percent of the time. Those whose question had named the wrong sign chose the right slide 41 percent of the time; the rest chose the slide showing the sign named in the question.

The later experiments varied when the misleading question arrived relative to the slides and the test, and found that the misinformation did the most damage when it came long after the event and shortly before the test, when the original memory had weakened and the planted detail was fresh.

The best-known forerunner is Loftus and Palmer's 1974 paper. In its first experiment, 45 students watched films of traffic accidents and estimated the cars' speed; the question used one of five verbs, and the mean estimate was 40.8 miles per hour for "smashed" and 31.8 for "contacted," with "collided," "bumped" and "hit" in between. In the second, 150 students watched a film of a collision, and a week later were asked whether they had seen any broken glass, of which there was none: 16 of the 50 who had been asked about cars that "smashed" said yes, against 7 of 50 asked about cars that "hit" and 6 of 50 who had not been asked about speed at all.

The pattern shows up wherever an account is shaped by what arrived after the event. The rows below are illustrations, not findings from the studies cited on this page.

The effect works on everyone, so the counters are procedures rather than willpower, and most of them have been written into guidance for interviewing witnesses.

• Record first, discuss second. The earliest account, written before any conversation about the event, is the one least contaminated by other sources. Keep it even when later information seems to improve on it.
• Ask open questions before specific ones. A question that assumes a detail supplies it. Begin with a free narrative and move to specifics only afterward, phrased so that no answer is suggested.
• Separate what you saw from what you heard. When recalling an event, ask of each detail where it came from; details that arrived from another person or a report should be labeled as such in your own account.
• Prevent the contamination rather than hoping to undo it. Warnings after the fact have been studied as a counter with mixed results, which is why interviewer training concentrates on not supplying details in the first place.

The misinformation effect is one specific route to a distorted memory. Its neighbors differ in where the distortion comes from. This page is educational: nothing on it can establish whether a particular memory is accurate.

• It is not false memory in general, which covers any recollection of something that did not happen, including the word-list illusion in which people recall a related word that was never presented. The misinformation effect requires misleading input from outside; the word-list illusion arises with none.
• It is not source confusion, though it depends on it. Source confusion is misremembering where a piece of information came from; the misinformation effect is what happens when information from a question or another person is mistaken for information from the event.
• It is not suggestibility as a trait, the degree to which a given person tends to accept suggestions. The misinformation effect is a property of the situation that affects most people; individual differences in suggestibility change its size.
• It is not the continued influence effect, in which a false claim keeps shaping reasoning even after it has been corrected. That effect concerns beliefs about facts; the misinformation effect concerns memory for a witnessed event.