Front cover

Winner, National Media Award For Distinguished Contribution
From the American Psychological Foundation

As pertinent today as when it was first published more than a decade ago, this engaging and highly poised study makes the psychological case against the reliability of the eyewitness.

By shedding light on the many factors that can intervene and create inaccurate testimony, Elizabeth Loftus illustrates how memory can be radically altered by the way an eyewitness is questioned, and how new memories can be implanted and old ones altered in subtle ways. She thus calls into question today's widely held assumption of eyewitness authority over the details of a crime or other events.

Eyewitness Testimony provides a sobering counterpoint to today's theatrical reliance on eyewitness accounts in the media, and should be required reading for trial lawyers, psychologists, jurors, and anyone who considers the chilling prospect of confronting an eyewitness accusation in a court of law.

"An important book about a critical question."

-- New York Times Book Review

"Loftus is a quiet, sincere, and reliable guide to this important research and its applications."

-- Times Higher Education Supplement

"Provide[s] sorely needed information and guidance . . . [Loftus's] purpose is to provide a theoretical framework for the psychological research that has been done in the areas of perception and memory and to offer specific suggestions on how the results can inform legal processes. She succeeds in what is an ambitious and difficult task."

-- Psychology Today

Elizabeth F. Loftus is Professor of Psychology and Adjunct Professor of Law at the University of Washington, Seattle


With a New Preface By The Author

Elizabeth F. Loftus

Cambridge, Massachusetts
London, England

For Geoffrey R. Loftus,
my closest friend

Copyright © 1979,1996 by the President
and Fellows of Harvard College
All rights reserved
Printed in the United States of America
Seventh printing, 1996

Library of Congress Cataloging in Publication Data

Loftus, Elizabeth F., 1944 -
Eyewitness testimony / Elizabeth F. Loftus;
with a new preface by the author.
p. cm.
Originally published: Cambridge, Mass.:
Harvard University Press, 1979.
Includes bibliographical references and index.
ISBN 0-674-28777-0
1. Eyewitness identification --- United States. 2. Eyewitness
identification --- Psychological aspects. 3. Forensic psychology.
I. Title.
KF9672.L63 1996
345.73'066 --- dc20 95-51176
[347.30666] CIP

Retaining Information
in Memory

O N APRIL 6, 1975, a man whom I will refer to as Aaron Lewis was arrested after leaving a grocery store carrying bottles of wine and beer for which he had not paid. The arrest came about because a clerk in the store called the police and stated that the man leaving was the same man who robbed him at knifepoint on February 15, 1975. The police picked up Lewis a few blocks from the store, took him to police headquarters, and booked him for the armed robbery that had occurred seven weeks before.

Although Lewis admitted to having shoplifted the beer and wine (he was caught red-handed), he denied having committed the armed robbery, a much more serious offense. He protested his arrest, arguing that the clerk had made a mistaken identification. The clerk, on the other hand, said that even though it had been seven weeks, he was sure that it was Lewis who had robbed him. Thus, the only piece of evidence against Lewis at his trial for armed robbery was the word of a single eyewitness. Incidents such as these cause us to wonder what happens to a person's memory over a period of time such as seven weeks. What about periods that are shorter, or those that are even longer?

When a witness perceives a complex event, a number of factors, such as the exposure time, or the salience of the event, or the witness's prior expectations, will affect the accuracy of what is perceived and stored in memory. But to compound the problem, once the material has already been encoded, further changes can take place. The time between a complex experience and a witness's recollection of that experience is a crucial period. Both the length of this retention interval and the events that take place during it affect a witness's testimony.

It is by now a well-established fact that people are less accurate and complete in their eyewitness accounts after a long retention interval than after a short one. The classic research, conducted by Ebbinghaus in 1885, is probably the most often cited study dealing with the loss of retention with the elapse of time. Ebbinghaus used only a single subject in his experiments -- himself. Typically, he learned a list of nonsense syllables, put them away for a certain interval of time, and then relearned them. He recorded the saving in time or the saving in number of readings necessary for relearning, assuming that the better his memory, the less time it would take to relearn the material. His results (which he plotted on the now-famous "forgetting curve") showed that we forget very rapidly immediately after an event, but that forgetting becomes mote and more gradual as time passes.

Ebbinghaus's findings have been confirmed again and again in subsequent research, using different types of people and different types of learning materials. Shepard (1967) tested thirty-four clerical workers for recognition of pictures after intervals of two hours, three days, one week, or about four months. Shepard found that the retention of the picture material dropped from 100 percent correct recognition after a two-hour delay to only 57 percent correct after four months. While 57 percent may seem high, it actually represents mere guessing on the part of the subjects: the test consisted of showing individual pictures to subjects and asking them to say yes if they recognized the picture as one they had seen before and no if they did not; a person who had never seen the pictures could guess correctly half of the time (obtaining a score of 50 percent) simply by chance.

An even more realistic event was shown to the subjects who participated in Marshall's study (1966), which was mentioned briefly in the previous chapter. Subjects looked at a forty-two-second film which Marshall describes as follows (pp. 53-54):

The picture which we showed had background music through most of the sequence. It opened with a boy lowering a mosquito net on a baby carriage. The boy was in his late teens or early twenties, of average build, wore a dark jacket, lighter baggy slacks, and a white shirt open at one neck. There were white buttons on the jacket and around his neck there were two strings or chains with a flute or whistle on one and a looking glass on the other. He had sideburns, curly hair. The baby was crying. At the beginning, the boy was smiling at the baby carriage. He appeared uncertain or nervous. He faced the baby carriage and touched the handle and started rocking it, then he removed his hands from the handle. The baby cried louder. The boy rocked the baby carriage back and forth, shifting his weight from foot to foot. The rocking became more violent and he pulled the carriage backward off the grass on which it had been to the driveway A woman called out or shouted, "Mrs. Gerard, Mrs. Gerard, quick! Someone's running away with your baby." The boy turned toward the fence of the yard and then toward the house A woman came running from the house toward the boy and the carriage. The woman was young. She wore a white smock and a darker skirt. She shouted, "You bad boy! You bad boy!", and waved her left arm as she ran. The boy hesitated, looked startled, ran through the gate and crouched in a corner by a white picket fence near a bush growing through the fence.
It is obvious that the movie contained quite a few details and the subjects in this experiment -- 167 law students, 102 police trainees, and 22 low-income people -- answered questions about them either immediately after the movie was shown or after a one-week interval. The subjects were consistently more accurate when they answered items immediately than when they answered them after a one-week interval. Referring to this decline as a "slippage of memory," Marshall indicated that he would have liked to see what would have happened to his subjects' memories after an interval of a month, or even a year. The answer seems clear. The accumulation of research tells us that after a year, memory will be less accurate than after a month; after a month it will be less accurate than after a week. This is, of course, another way of stating the result that Ebbinghaus discovered nearly one hundred years ago.

Postevent Information

Time alone does not cause the slippage of memory. It is caused in part by what goes on during the passage of time. Often after witnessing an important event, one is exposed to new information about it. For example, a person sees an automobile accident and then learns from the newspaper that the driver of the car had been drinking before the accident. Or one witnesses an argument between two people, and then overhears a friend tell a third person all the gory details of the argument. Evidence has recently been accumulating to indicate that postevent experiences such as exposure to newly released information can dramatically affect our memory of the original event. Bird (1927) provides an early example. During the course of a routine classroom lecture, the instructor was discussing the results of a series of experiments. A well-meaning but not very thoughtful reporter on the local newspaper printed an account of the lecture that was riddled with errors. Many students read the newspaper account, and nearly all of these thought it was an accurate report. The instructor gave an exam at the end of the week and after the usual set of exam questions he asked each student to indicate whether or not the student had read the press account. Those who had read the article made many more errors on the exam; they remembered the erroneous information that was in the newspaper, assuming that they had learned it from the instructor's original lecture.

The same sort of thing happens when witnesses to an event later read or hear something about it and are subsequently asked to recall the event. Postevent information can not only enhance existing memories but also change a witness's memory and even cause nonexistent details to become incorporated into a previously acquired memory.

Enhancing Memory

It is quite common for witnesses to a serious event to discuss the event after it is over. For example, the robbers leave the grocery store and one checker says to another, "Did you get a look at the guy with the green hat?" For a moment, let us suppose that one of the robbers actually was wearing a green hat. What are the consequences of the first checker's remark for the second checker's memory? As it turns out, this remark can increase the likelihood that the second checker will also claim to have seen the green hat.

This was demonstrated in the following experiment (Loftus 1975). Subject-witnesses were shown a film of a multiple-car accident in which one car, after failing to stop at a stop sign, makes a right-hand turn into the main stream of traffic. In an attempt to avoid a collision, the car in the oncoming traffic stop suddenly and a five-car bumper to bumper collision results. The film lasted less than one minute and the accident itself occupied a four-second period. At the end of the film, the subjects were given a diagram of the accident, in which the letter A represented the car that turned right and ran the stop sign, while B through F represented the cars involved in the collision. All subjects were asked a series of ten questions. The first question in the series asked about the speed of the car that caused the accident in one of two ways:

(1) How fast was car A going when it ran the stop sign?
(2) How fast was car A going when it turned right?

Seventy-five subjects were asked the first question, and seventy-five different subjects were asked the second question. The last question in the series, question 10, was identical for all subjects; it asked whether the subject had actually seen a stop sign for car A. If the earlier question had mentioned a stop sign, 53 percent of the subjects reported later on that they had seen a stop sign. However, if the earlier question had not mentioned a stop sign, only 35 percent of the subjects claimed to have seen the stop sign when asked later on. Thus, by simply mentioning an existing object, it is possible to increase the likelihood that it will be recalled later on.

Compromise Memories

What happens when a witness sees some event and later learns a piece of new information which conflicts with some aspect of what was previously seen? It appears that when possible many witnesses will compromise between what they have seen and what they have been told later on.

In one experiment (Loftus 1975) forty subject-witnesses were shown a three-minute videotape taken from the film Diary of a Student Revolution. The incident involved the disruption of a class by eight demonstrators. The confrontation, which was relatively noisy, caused an interruption in a professor's lecture and finally ended when the demonstrators left the classroom. At the end of the videotape the subjects received one of two questionnaires containing one key question and nineteen filler questions. Half of the subjects were asked, "Was the leader of the four demonstrators who entered the classroom a male?" whereas the other half were asked, "Was the leader of the twelve demonstrators who entered the classroom a male?" The subjects responded to all of their questions by circling yes or no. One week later, all subjects returned to answer a new set of questions. The critical question at this time was, "How many demonstrators did you see entering the classroom?" Those subjects who had previously been asked the "twelve" question reported having seen an average of 8.9 people when questioned one week later, whereas the subjects interrogated with the "four" question recalled an average of 6.4 people. When the individual responses were examined, it was noted that most subjects tended to compromise between what they had actually seen, namely eight, and what they had been told later on -- four in one case, twelve in the other.

In a second study showing the compromise response (Loftus 1977), a series of thirty color slides depicting an auto-pedestrian accident was shown for three seconds each to one hundred subjects. In this series a red Datsun is seen traveling along a side street toward an intersection. The car turns right and knocks down a pedestrian who is crossing at the crosswalk. A green car drives past the accident but does not stop (fig. 4.1). A police car arrives, the officer attempts to help the victim, while a passenger who had been in the Datsun runs for help.

Immediately after viewing the slides, the subjects answered a series of twelve questions. For half of the subjects, question ten falsely informed them that the car that drove past the accident was blue rather than green.

Figure 4.1. Black and white reproduction of one of thirty color slides depicting an automobile-pedestrian accident. Subjects' memories of the critical item -- the green car seen at the center of the picture -- tended to be a compromise between the green they saw and the blue they were told they saw. (Loftus 1977.)

The other half of the subjects (the controls) received no color information. After a twenty-minute filler activity, a color recognition test was administered. All subjects were shown a color wheel containing thirty color strips and were given a list of ten objects. For each object their task was to pick the color that best represented their recollection of the object.

The results showed that the subjects who had been given the blue information tended to pick a blue or bluish-green as the color that they remembered for the car that passed the accident. Those not given any color information tended to choose a color near the true green. Thusly the introduction of the false color information significantly affected the ability of subjects to correctly identify a color that they had seen before.

Introducing Nonexistent Objects

When estimating numbers of people, or when recalling colors, witnesses can readily compromise between what they actually saw and what they were told. The compromise could be conscious and deliberate, or it could be unconscious. However, with other kinds of objects such compromise is not easy. For example, suppose a witness saw a car speed through a stop sign and later learned that the traffic sign was actually a yield sign. It would probably be the unusual witness who could come up with some compromise sign; most would stick to the stop sign that they actually saw, or decide upon the yield sign that they learned about later on. In fact, this is what people tend to do.

In an experiment by Loftus and colleagues (1978) nearly two hundred subjects viewed a series of thirty color slides depicting successive stages in an auto-pedestrian accident. The auto was a red Datsun shown traveling along a side street toward an intersection with a stop sign for half of the subjects and a yield sign for the remaining subjects (fig. 4.2). The Datsun became involved in an accident with a pedestrian, as described in the last section. Immediately after viewing the slides, the subjects were asked some questions, one of which was critical. For about half of the subjects the critical question asked was, "Did another car pass the red Datsun while it was stopped at the stop sign?" The remaining subjects were asked the same question with the words "stop sign" replaced by "yield sign." For some of the subjects, the sign mentioned in the question was the sign that had actually been seen; in other words, the question gave them consistent information. For she remaining subjects, the question contained misleading information.

Figure 4.2. Critical slides used in an experiment which showed that eyewitnesses can "remember" seeing nonexistent objects (From Loftus et al. 1978.)

After completing the questionnaire, the subjects participated in a twenty-minute filler activity, which required them to read an unrelated short story and answer some questions about it. Finally, a recognition test was administered. Pairs of slides were presented to the subjects and they had to indicate which member of each pair they had seen before. The critical pair was a slide depicting the Datsun stopped at a stop sign and a nearly identical side depicting the Datsun at a yield sign.

The results indicated that when the intervening question contained consistent information, 75 percent of the subjects accurately responded. When the question contained misleading information, only 41 percent of the subjects accurately responded. If the subjects had been simply guessing, they would have been correct about half the time, or 50 percent, so the misleading question reduced their accuracy below that which would have been expected from a person who was merely guessing.

I have conducted numerous demonstrations showing how nonexistent objects can be introduced into people's recollections. For example, college students were presented with a film of an accident, followed by a misleading question (Loftus 1975). Some subjects were asked, "How fast was the white sports car going when it passed the barn while traveling along the country road?" A control question asked other students, "How fast was the white sports car going while traveling along the country road?" One week later, all of the students were asked whether they had seen a barn. In fact, no barn existed, but over 17 percent of the students whose question mentioned the nonexistent barn claimed to have seen it later on. In contrast, less than 3 percent of control subjects recalled a barn. Thus, casually mentioning a nonexistent object during the course of questioning can increase the likelihood that a person will later report having seen that nonexistent object.

The basic phenomenon noted here, namely, that nonexistent objects can become incorporated into people's memories, has been observed by other investigators. Two psychologists, Lesgold and Petrush (1977), conducted a study in which ninety-nine subjects saw a series of slides depicting a bank robbery. Every slide was accompanied by a narration of approximately sixty words which tied the slides together so that the subjects definitely felt as if they were experiencing a unified event. In each slide there was one detail (such as an alarm button) that was either present or not present (fig. 4.3). Following the slides, a series of questions was asked, some of which mentioned the existence of the key objects and others of which did not. Finally, all subjects were given the names of the key items, and some filler items, and were asked to indicate whether each item had actually been seen. The two psychologists found that simply mentioning a nonexistent object after the bank robbery slides had long since been viewed was sufficient to increase the likelihood that subjects would think they had seen the object.

Figure 4.3. Slide from bank robbery series. A companion slide, seen by half of the subjects, did not contain the key item -- an alarm button. (From Lesgold and Petrush 1977.)

In addition to the laboratory studies, demonstrations outside the laboratory have uncovered the same phenomenon at work. For example, some years ago during a course on cognitive Psychology I gave my students the following assignment: I told them to go out and create in someone's mind a "memory" for something that did not exist. My hope was that they would discover how relatively easy this can be, and, further, that they would see that a memory so acquired can be as real to a person as a memory that is the result of one's own ordinary perceptual sensations. One group of students conducted their study in train stations, bus depots, and shopping centers, proceeding as follows: Two female students entered a train station, one of them leaving her large bag on a bench while both walked away to check the train schedules. While they were gone, a male student lurked over to the bag, reached in, and pretended to pull out an object and stuff it under his coat. He then walked away quickly. When the women returned, the older one noticed that her bag had been tampered with, and began to cry, "Oh my God, my tape recorder is missing!" She went on to lament that her boss had loaned it to her for a special reason, that it was very expensive, and so on. The two women began to talk to the real eyewitnesses who were in the vicinity. Most were extremely cooperative in offering sympathy and whatever details could be recalled. The older woman asked these witnesses for their telephone numbers "in case I need it for insurance purposes." Most people gladly gave their number.

One week later an "insurance agent" called the eyewitnesses as part of a routine investigation of the theft. All were asked for whatever details they could remember, and finally, they were asked, "Did you see the tape recorder?" Although there was in fact no tape recorder, over half of the eyewitnesses "remembered" seeing it, and nearly all of these could describe it in reasonably good detail. Their descriptions were quite different from one another: some said it was gray and others said black; some said it was in a case, others said it was not; some said it had an antenna, others claimed it did not. Their descriptions indicated a rather vivid "memory" for a tape recorder that was never seen.

In real life, as well as in experiments, people can come to believe things that never really happened. One of the nicest examples of this can be found in the reminiscences of the psychologist Jean Piaget (1952):

There is also the question of memories which depend on other people. For instance, one of my first memories would date, if it were true, from my second year. I can still see, most clearly, the following scene, in which I believed until I was about fifteen. I was sitting in my pram, which my nurse was pushing in the Champs Elysees, when a man tried to kidnap me. I was held in by the strap fastened round me while my nurse bravely tried to stand between me and the thief. She received various scratches and I can still see vaguely those on her face. Then a crowd gathered, a policeman with a short cloak and a white baton came up, and the man took to his heels. I can still see the whole scene, and can even place it near the tube station. When I was about fifteen, my parents received a letter from my former nurse saying that she had been converted to the Salvation Army. She wanted to confess her past faults, and in particular to return the watch she had been given as a reward on this occasion. She had made up the whole story, faking the scratches. I, therefore, must have heard, as a child, the account of this story, which my parents believed, and projected into the past in the form of a visual memory. (pp. 187-188)

Central Versus Peripheral Details

Dritsas and Hamilton (1977) noted that the misleading bits of information in much of the previous research dealt with items that were not central to the actions being observed by witnesses. They became interested in comparing the ease with which one could modify memory for a salient detail -- one that is perceived easily and has a high chance of being recalled accurately. The investigators hypothesized that peripheral items would be easier to modify than central, salient items, and they set out to test their hypothesis.

Seventy-two subjects looked at a videotape of three industrial accidents. In one, a male job trainee is hit by a metal chip in the cornea of the right eye. In a second, a maintenance man is struck in the back by a spinning metal rod and thrown to the ground, injuring his head and neck. In the third, a woman's right hand is caught in a punch press. Thus, the subjects viewed relatively stressful events which -- if they had happened in real life -- would probably lead to a fairly rigorous questioning of witnesses by investigators. It took eleven minutes to view all three tapes.

After watching the videotape, each subject completed a questionnaire which included thirty questions on both central and peripheral details. (The details had been rated as either central or peripheral on the basis of the responses of an independent group of twenty-five students.) Six of the thirty questions -- three central and three peripheral -- were misleading in that the contained false information intended to alter the subjects' later recollection of the event.

The results indicated that salient or central items were recalled with significantly greater accuracy and were much more difficult to alter with misleading information than were peripheral items. Eighty-one percent of the central questions were accurately answered, while only 47 percent of the peripheral questions were correctly recalled. Misleading information about central items altered subsequent recall 47 percent of the time, whereas misleading information about peripheral items altered subsequent recall 69 percent of the time. Thus we can conclude from this study that it is harder to mislead a witness about important, salient, or central aspects of an event than about peripheral ones.

Timing of Postevent Information

Consider the following dilemma: A witness to an accident has actually seen a vehicle run through a red light. The witness will be leaving town for a week and will then return to testify about the accident. An unscrupulous lawyer for the driver of the vehicle has been granted a few minutes with the witness and wishes to use that time to suggest to the witness that the light was actually green. Which time would be best for him to query the witness -- right after the accident or one week later, just prior to the time the witness must testify? (For the moment, set aside the ethics of this demonstration and consider only the effects of the lawyer's behavior on the witness's final testimony.)

When people are asked to predict when the misleading information will have the greatest impact, those who vote for the earlier input often say something like, "You have to put the new information into the person's memory close in time to the actual event so that they will be stored in memory near each other." Those who vote for the later input often remark, "If you put the new information in too early it will conflict with what the witness already has in memory; if you wait until later, the memory will be weaker and will not conflict as much." Who is right? An experiment my colleagues and I conducted suggests an answer (Loftus et al. 1978).

The question asked in this study is: Does information introduced subsequent to an event have a different impact depending upon whether it is introduced immediately after the event or just prior to the time the event is to be recalled? Over six hundred participants saw the series of thirty color slides described previously, which included one slide showing a red Datsun stopped at either a stop sign or a yield sign. A questionnaire was administered, followed by a final forced-choice test; this test occurred after a retention interval of either twenty minutes, one day, two days, or one week. Half of the subjects answered the questionnaire immediately after viewing the accident slides (the "immediate" questionnaire), and the other half answered it just before the final forced-choice test ("delayed" questionnaire). In addition, another group was both questioned and tested immediately after seeing the accident. In each group of subjects, some received information on their questionnaires that was consistent with what they had seen (that is, they actually saw a stop sign and were told it was a stop sign); others received information on their questionnaires that was misleading (they actually saw a stop sign but were told it was a yield sign), and still others received no information about the signs. Our major interest was performance on the final test, during which the subjects were shown the two critical slides, one containing a stop sign and the other containing a yield sign. What proportion of them correctly chose the sign they had actually seen before?

Figure 4.4. Proportion of subjects who correctly selected the slide they had seen before. The subjects were tested after an interval that ranged from immediate to one week. They were given either consistent information, misleading information, or no information about the critical sign. Their information came either immediately after the accident slides or else it came just prior to the final test. (From Loftus et al. 1978.)

The results are shown in figure 4.4. The curves marked by triangles show what happens to performance when no relevant information is introduced during the retention interval. Performance on the test is quite high immediately after the slide sequence -- almost 90 percent accuracy. It drops gradually as the interval increases, and by the time two days pass, subjects are performing at about 50 percent correct, a figure which of course means that the subjects are simply guessing. These data constitute additional support for the claim that people are less accurate after a long interval than after a short one.

Now, look at the curves marked with circles. These show what happens to performance when information given during the retention interval is consistent with the film. Here a subject was shown a stop sign, was given a questionnaire which mentioned the existence of this stop sign, and then was finally tested. The circle curves are generally higher than the triangle curves, indicating that consistent information boosts performance above that of a group of subjects who were given no relevant information. For example, after a two-day interval, the subjects given no information are performing at about a chance level. However, the subjects given consistent information are correct over 70 percent of the time. This result lends additional support to the idea that by mentioning an object which did in fact exist, one may enhance a witness's memory for that object.

Finally, let us examine what happens when misleading information is presented to a subject between the initial experience and the final test. The data in the curves marked with squares correspond to this situation. When misleading information is given immediately, the subjects answer correctly just over 50 percent of the time after a retention interval of one week. When misleading information is given just before the test, subjects are correct a mere 20 percent of the time after a retention interval of one week. Thus, misleading information more severely retards performance when it is delayed. Why is this? If misleading information is given immediately after the slides, one week later both the information from the slides and the new misleading information have faded. When tested, the subject can only guess, and he is correct about half of the time. However, when the misleading information is delayed for a week, the information from the slides has faded but the misleading information which has just been introduced is quite strong. Thus, the subjects tended to "recall" the misleading information and thus to perform incorrectly on their final test.

This experiment shows that, in general, longer retention intervals lead to worse performance; consistent information improves performance and misleading information hinders it; and misleading information that is given immediately after an event has less of an impact on the memory than misleading information that is delayed until just prior to the test. Apparently, giving the event information a chance to fade in memory makes it easier to introduce misleading information.

However, if one desires to get the witness to completely reorganize the full content of the critical event, rather than simply to change memory for a detail (as was the aim of the experiment just described), different timing may be called for. In research by James Dooling and his colleagues, subjects were presented with written passages; one, for example, was about a ruthless dictator (Sulin and Dooling 1974):

Gerald Martin strove to undermine the existing government to satisfy his political ambitions. Many of the people of his country supported his efforts. Current political problems made it relatively easy for Martin to take over. Certain groups remained loyal to the old government and caused Martin trouble. He confronted these groups directly and so silenced them. He became a ruthless, uncontrollable dictator. The ultimate effect of his rule was the downfall of his country.
Subjects in one group believed that the dictator was the fictitious Gerald Martin. Subjects in another group were told after reading the passage that the main character was really Adolph Hitler; these subjects could then understand the passage with respect to their prior knowledge about Hitler. In the final test given to all subjects, seven sentences from the passage were randomly mixed together with seven false sentences. Of special interest was the subjects' performance on critical false sentences such as, "He hated the Jews particularly and so persecuted them." This sentence did not occur in the original passage and thus would not apply to the fictitious Gerald Martin but would be true of Hitler. When the test was delayed for one week, subjects in the group which had been told that the passage was about Hitler were much more likely than the other group to answer that they had read this critical sentence.

But what about the timing of the Hitler information? Would final test performance depend upon whether the Hitler information was presented just after the reading of the passage or just before the final test? It does matter. Manipulation of the main character immediately after the reading of the passage led to more errors on the final test (Dooling and Christiaansen 1977). The explanation offered for this finding is quite speculative: the investigators argue that those subjects who were given the Hitler information immediately had to perform a difficult cognitive manipulation to comprehend the material. They integrated passage information with new information, and the new information showed up in their memories when tested one week later. On the other hand, subjects who were given the Hitler information one week later, just before the test, did not have much specific memory for the passage, and thus they did not have as much passage information to integrate with the new information. Why under these conditions the subjects do not rely more heavily on their general knowledge store is still something of a mystery.

One Year Later

Most psychologists have studied memory using retention intervals of a day, a week, or occasionally a month or so. Shepard (1967) looked at memory for pictures after a four-month interval, and that was quite rare; it requires a good deal of effort to round up a group of subjects four months after their initial learning. But Davis and Sinha (1950) were sufficiently enterprising to get some of their subjects back after a full year had elapsed since an initial learning experience, in order to investigate the influence of postevent information.

The subjects in this experiment, university students, were presented with a 750-word story about a feud between two families and their reconciliation in the betrothal of the son of one with the daughter of the other. The last part of the story described a wedding feast. The entire story was written in a rather pompous style, as this excerpt shows:

The wedding feast, which was held at the Weyden farm, | was honoured even by the sheriff, whose efforts had often been necessary to curb the violence of their quarrels, | and to arrange a truce between the families. | Sitting in a high-backed chair | at the head of the broad and laden table, | a dark Holbein cap, | almost hiding his white hair, | and the velvet of his nobility being somewhat incongruous | in the humble interior of the farmhouse, | he gazed anxiously at Hans Loon, | the uncle of the bridegroom and the brother of the murdered man. | His solemn expression was unsuited | to the rustic boisterousness of the village folk. | But he had reasons to be alarmed, | for several unhappy incidents | made him doubt the permanence of the sudden friendship between the two families | . . . But nothing seemed to daunt the vivacious gossiping of the women, | and as dish after dish was brought | for the guests' pleasure, | and the coloured earthenware pitchers were filled | again and again | with sweet, white wine, | the danger that rancour would flare up again | seemed to grow less. |
Three or four days after reading the story, one group of subjects was shown seven postcards and asked to identify the one that depicted a scene described in the story. One card was picked by all of the subjects: it was a reproduction of Pieter Bruegel the Elder's painting Peasant Wedding, which, although it is famous, was not familiar to these subjects (fig. 4.5). Some of the details in the picture were consistent with the story, but there were discrepancies between picture and story as well.

A major interest in this experiment was how the picture influenced memory for the story. Subjects were asked to recall the story on one occasion only, at intervals varying from immediately to more than three weeks. For purposes of analyzing the students' recollections, the investigators divided up the original story arbitrarily into about 120 items as shown in vertical lines in the excerpt. Similarly, the items in each subject's written recollection were enumerated. Of those subjects whose initial recall occurred within about a month, those who had seen the picture listed more items than did those who had not seen the picture. In addition, those who had seen the picture tended to incorporate more totally false items into their story. For example, they claimed to have read about two long tables -- an item from the picture but not from the story. Further comparisons revealed a difference in the way the two groups tended to recall the story. The recall of the subjects who saw the picture was fuller and more graphic (see table 4.1).

Figure 4.5. Pieter Bruegel the Elder's Peasant Wedding (c. 1565). Details from this painting were incorporated into subjects' memories of a story they had read one year earlier. (By permission of the Bettmann Archive.)

One year after reading the story, nine subjects who saw the picture and seven who did not were found and asked to come back and recall the story again. The two groups tended to recall different kinds of items. Those who had seen the picture tended to structure their fragmentary recollection around the wedding feast. Those who had not tended to recall the concept of a feud in a peasant society. Most of the former subjects said they could remember the picture better than the story, but in general they could not decide whether a detail which had been recalled came from the picture or the story. Mistakes were numerous, with picture details appearing as intrusions in the story. Thus, their recollection of the story included a memory for "laborers with colored hats" and for "servants carrying plates on a tray," items which were never part of the story at all. It appears that one year later there was a total merging of the ideas from the two separate sources.

Subjective Recollections Can Change

We have seen how postevent information can cause objects to be added to people's memories under some conditions (for example, a barn introduced) and can also cause objects to be altered in memory (for example, a stop sign becomes a yield sign). It also turns out that postevent information can have a fairly significant impact on people's subjective recollections; people's feelings about how noisy an event was, or how violent it was, can also be changed.

                         Retention        Mean number of    Median number
Subjects                 interval         items recalled    of importations
Saw picture            Less than 8 days         50                4
                       21-28 days               43                7
Did not see picture    Less than 8 days         45                3
                       21-28 days               24                4
Table 4.l. Performance on memory test by subjects who saw a misleading picture during the retention interval, compared to the performance of subjects who did not. (Data from Davis and Sinha 1950.)

In a study I conducted in collaboration with Altman and Geballe (Loftus et al. 1975), over fifty, subjects were shown a three-minute videotape taken from the film Diary of a Student Revolution. The sequence depicted the disruption of a class by eight demonstrators; the confrontation was relatively noisy but basically nonviolent, and ended with the demonstrators leaving the classroom.

At the end of the videotape the subjects received one of two questionnaires. The neutral/passive questionnaire contained twenty-five questions, all phrased in a relatively mild way. For example, two of the questions were: Did you notice the demonstrators gesturing at any of the students? Did the professor say anything to the demonstrators? The other questionnaire was more emotional, with questions worded in a more aggressive way. It contained twenty-five analogous questions, for example: Did you notice the militants threatening any of the students? Did the professor shout something at the activists? No instructions were given as to possible future questions that would be asked.

One week later all subjects returned and answered a new series of questions about the disruption. As part of this session, the subjects were asked to indicate their recollection by checking the appropriate place on a five-point scale:

(1) The incident could be described as: quiet __ __ __ __ __ noisy
(2) Would you describe the incident as: peaceful __ __ __ __ __ violent
(3) Do you recall that the demonstrations were: pacifistic __ __ __ __ __ belligerent
(4) In general, the students' reaction to the demonstration was: sympathetic __ __ __ __ __ antagonistic

The subjects responded to these questions by placing an X or a check mark on the line that best reflected their recollection of the incident.

The results are shown in table 4.2. For each of the four critical questions, a single number was obtained for the subjects answering the emotional/aggressive questionnaire and for those answering the neutral-passive one in this way: if a subject checked the first spot on the scale, he received a score of one; if he checked the second spot, he received a score of two, and so on. We then took the average score of the emotional/aggressive subjects and the average score of the neutral/passive subjects. As the table shows, the subjects who were interrogated with questions worded in an emotional, aggressive manner reported that the incident was noisier and more violent, that the perpetrators of the incident were more belligerent, and that the students in the classroom were more antagonistic than did those subjects whose questionnaires were neutral and passive. Thus, this simple postevent interrogation in which just a few words were changed produced marked changes in people's subjective recollections about the incident.

response                   Emotional/aggressive        Neutral/passive
Quiet/noisy                       3.71                      3.04
Peaceful/violent                  2.50                      2.04
Pacifistic/belligerent            3.33                      2.64
Sympathetic/antagonistic          4.50                      3.64
Table 4.2. Average responses of four critical questions asked in an emotional/aggressive or in a neutral/passive manner. (Data from Loftus et al. 1975.)

Nonverbal Influences

The fact that a picture can influence someone's memory for a previously heard story is one indication that postevent information need not be in the form of words or language spoken to a witness after an event is over. Words, in fact, make up only a small part of the information that people convey to one another. The tone of voice, the movements of heads and eyes, the gaze, the posture, and other -- sometimes very subtle -- behaviors can be used to convey ideas to others.

Several psychologists at Harvard University have been studying the power of nonverbal communication (Hall et al. 1978). Some very interesting studies on nonverbal messages came about as a result of research done by one member of the group on the effect of teachers' expectations on the performance of their pupils. In that research, elementary school children took a standard intelligence test which the teachers were told would distinguish those children who were intellectually gifted. The teachers were then told that certain children who had taken the test (about one-fifth of the class) were expected to be intellectual bloomers. Unknown to the teachers, these children had been chosen at random. After eight months had passed, all of the children in the class were given a second intelligence test, and the random fifth that the teachers believed to be special did in fact outperform the rest of the children. What had apparently happened is that the teachers behaved differently toward these "special" children and unintentionally influenced them to behave as if they were intellectually advantaged. The teachers could have communicated high expectations for the "special" children through some nonverbal channel, perhaps tone of voice used when speaking to the student, facial expressions, or whatever. These positive expectations are known to create a favorable atmosphere for children to develop in.

Another important study (Hall et al. 1978) has indicated that if a scientific experimenter expects research subjects to behave in a certain way, the behavior can be unintentionally influenced through nonverbal communication. Students who were studying psychology were told either that the rats in their experiment were specially bred for high intelligence or that they had been bred to be stupid. The students then studied them as the rats attempted to learn how to navigate a maze. Actually the rats were all from the same source -- one was not really smarter than another. Yet the "intelligent" rats learned to run mazes far more quickly and accurately than their "stupid" counterparts. How could this happen? The researchers argue that "the students who had been told their rats were bright said that they handled their animals gently, that they liked them, and that they were enthusiastic about the experiment. The students whose rats supposedly were dull said they handled them less but talked to them more than did students with 'bright' rats" (p. 70). It is thus apparent that nonverbal communication played a part in fulfilling the expectations that the students held about their particular animals. Performance changed in the direction of the student-experimenter's expectations.

People vary in their ability to send and perceive nonverbal messages, but most people have some ability to be swayed by the nonverbal communication that comes their way. In the legal world this can be potentially very important. A police officer may tell a witness that a suspect has been caught and the witness should look at some photographs or come to view a lineup and make an identification. Even if the policeman does not explicitly mention a suspect, it is likely that the witness will believe he is being asked to identify a good suspect who will be one of the members of the lineup or set of photos. It is here that nonverbal as well as verbal suggestions can easily be communicated. If the officer should unintentionally stare a bit longer at the suspect, or change his tone of voice when he says, "Tell us whether you think it is number one, two, THREE, four, five, or six," the witness's opinion might be swayed.

Another area for potential danger has to do with the possible communication from one witness to another. In a lineup situation this has been recognized and has led Wall (1965) to note that "the most important rule here is that each witness must view the line-up separately, unaccompanied by another witness" (p. 49). This eliminates the problem of witnesses either verbally or non-verbally, intentionally or unintentionally, communicating with one another. Of course, no law enforcement procedures can prevent witnesses at the scene of a crime or accident from interacting with each other before the police arrive. Memories may be distorted long before anyone arrives to investigate. This problem became apparent to me several years ago when I arrived at the scene of a serious armed robbery moments after the robbers had fled from the grocery store. Somewhere between four and six intruders had held up the store, made the employees lie on the floor, taken the money, and run. When I arrived the manager was up off the ground, the police had been called, and the witnesses were all agitated and communicating with one another. "Did you get a look at the weird one with the blue cap?" "Boy, that short one was really frightening!" Words were used, but so was nonverbal information through the voice, gaze, and body movements. In this way one witness indicated a great deal of confidence in what she had seen, a confidence that was likely to enhance the probability that her views would be accepted by some of the others who were less confident of what they had seen. In short, there is a problem when witnesses to a crime are friends or fellow employees and have time to compare mental notes with each other before an "official" investigation begins. Nonverbal influences are as worrisome as verbal ones.

Investigations by Police and Attorneys

Recently in a town in Idaho a man raped a young woman, Jane, while forcing her friend Susan to lie with them on the bed. The two roommates had been asleep in their apartment when Susan heard Jane screaming. Since Jane often had nightmares and occasionally talked in her sleep, Susan began to get up from her bed to go into Jane's room and awaken her. Then she heard a male voice. Before she could even get up, her own bedroom door opened and revealed the intruder holding Jane. The stranger was standing directly behind Jane with his left arm across her upper chest, holding her close to him. He held a long knife -- possibly a kitchen steak knife -- in his right hand. Jane was forced to lie down on the bed with Susan and both women were told to cover their faces. But Susan looked up periodically. What happened at this point, from Susan's point of view, could be inferred from what she told the police during an interview several hours later (Lewiston, Idaho, December 14, 1977):

What was he doing when you looked up?
I don't, he was trying to bother Jane and I just took the pillow away from my face and started to say something to him and he saw me and told me to get my face back down.
What do you mean bother?
He was starting to -- you know.
Have intercourse with her?
Yea, he was just kind of fondling her I guess, I don't know, I didn't see anything he was doing that, you know, she was telling him to stop and please don't do that.
All right, let's go back, you heard the scream?
And you heard a man's voice?
The next thing you knew your door opened?
Right, it was partway open to begin with.
And then it was just pushed open?
And then he brought her into your room, into your bedroom, with his hand around her throat, or did he have her in a head lock or . . . ?
This is a portion of an actual interview conducted after an actual rape had occurred. It may appear as if the purpose of the interview is to get information from the witness, and indeed this is the major purpose. However, such interviews can have side effects which may be unintentional. They can impart information to the witness at the same time that they are apparently obtaining information from the witness. This is particularly dangerous when the police have a suspect in mind, or a theory about the case, for their ideas can be transmitted to the witness and can affect the witness's memory. But it is also a danger when there is no theory and the investigating officer or attorney is simply trying to get information from the witness.

Vidmar (1978) studied the effects of various investigative procedures on the ultimate memory and testimony of an eyewitness. Subjects came into the laboratory and were shown a ten-minute incident involving a discussion in a pub that resulted in a fight in which the defendant, Zemp, struck the victim, Adams, on the head with a bottle. The events in the incident were presented via slide projector and tape recorder and were designed to be ambiguous in places, although in one version the incident was clearly biased against the defendant, Zemp. Only after the incident was over were the subjects informed that they were to be called as witnesses in a civil trial to take place in one week. Adams was suing Zemp for damages incurred in the incident.

Between the incident and the trial, a lawyer came to interview each of the student-witnesses. The lawyers were students recruited to play this role. They were given a summary of the suit, statements given by both Zemp and Adams, and some basic information about relevant law, and they were told they would have to prepare and argue a case in front of a judge in a trial to be held in one week. To prepare for that trial each lawyer was given one witness to interview -- this was to be the only source of evidence. Some lawyers were told they represented the victim Adams, while others represented the defendant Zemp. Each lawyer was given a "retainer" fee of $2.00 and told that if the judge hearing the case was to decide in favor of his client, he would win an additional $2.00 as his contingency fee; if he lost he would receive nothing.

One week from the incident, after each witness had been interviewed by either the plaintiffs or the defendant's attorney, the witness appeared for the trial. Each was met by a judge, an older student who was very formal in both dress and decorum. The judge did not know what type of lawyer had interviewed the witness. The judge informed the witness that his testimony was being recorded so that the judge could have it for further reference in deciding the case. The witness was then asked to state his or her name and to swear to tell the truth, the whole truth, and nothing but the truth. The witness was asked to recall what had happened while the judge listened passively except for a few prompting cues such as "Is that all?" or "Go on." When the witness completed testimony the judge asked the witness some questions designed to probe the witness's memory for the incident. The judge then made a decision. Was the witness's testimony biased toward the plaintiff or toward the defendants The results were clear-cut: Even though the case itself favored the plaintiff, witnesses who had been interviewed by attorneys for the defendant gave testimony that was definitely biased in favor of the defendant; their testimony was clearly contrary to the facts of the case. Why did this happen? Vidmar argues that a lawyer for the defendant, Zemp, is confronted with a witness who has seen a version of the incident contrary to the defendant's position. In an attempt to build a case, he must exert substantial effort to find favorable facts in order to increase has chances of winning. Ultimately, this extra effort affected the witness, possibly by changing his memory of the actual incident.

Why Postevent Information Works

How and why do people come to believe that they have seen nonexistent stop signs, tape recorders, alarm buttons, barns? The answer to this question is exceedingly important to understanding how the human mind works. The preliminary ideas I will offer here are best understood in the context of a specific experiment (Loftus and Palmer 1974) in which subjects viewed a film of a traffic accident and then answered questions about the accident. Some subjects were asked, "About how fast were the cars going when they smashed into each other?" whereas others were asked, "About how fast were the cars going when they hit each other?" The former question elicited a much higher estimate of speed. One week later the subjects returned and, without viewing the film again, they answered a series of questions about the accident. The critical question was, "Did you see any broken glass?" There was no broken glass in the accident, but because broken glass usually results from accidents occurring at high speed, it seemed likely that the subjects who had been asked the question with the word "smashed" might more often say yes to this critical question. And that is what we found:

"Smashed"          "Hit"
  16 yes             7 yes
  34 no             43 no

In discussing our results, we proposed that two kinds of information go into one's memory for some complex occurrence. The first is information gleaned during the perception of the original event; the second is "external" information supplied after the fact. Over time, information from these two sources may be integrated in such a way that we are unable to tell from which source some specific detail is recalled. All we have is one "memory." In the smashed-hit experiment, the subject first forms some representation of the accident he has witnessed. Some bits and pieces of information get into memory. The experimenter, then, while asking, "About how fast were the cars going when they smashed into each other?" supplies a piece of external information, namely, that the cars have indeed smashed into one another. When these two pieces of information are integrated, a subject has a "memory" of an accident that is more severe than the accident in fact has been. This situation is depicted in figure 4.6. Because broken glass is typically associated with a severe accident, the subject is more likely to think that broken glass has occurred.

Figure 4.6. Two types of information in memory. (From Loftus and Loftus 1976.)

A more general statement can actually be made. Anytime after a witness experiences a complex event, he may be exposed to new information about that event. The new information may come in the form of questions -- a powerful way to introduce it -- or in the form of a conversation, a newspaper story, and so on. The implication of these results for courtroom examination, police interrogations, and accident investigation is fairly obvious: interrogators should do whatever possible to avoid the introduction of "external" information into the witness's memory.

Intervening Thoughts of a Witness

At the Twenty-ninth Annual Advocacy Institute at the University of Michigan, over one thousand lawyers received materials regarding the hypothetical case of Duncan v. The Americraft Industries, Inc. (Stein 1978). Although this case was hypothetical it was derived from a couple of actual cases that had occurred in Michigan. The Duncan case arose after David Allen, who worked as a salesman for Americraft, was driving southbound on Ridge Road at about thirty-five miles per hour and struck a four-year-old girl, Mary Lou Duncan. The little girl suffered a broken arm and skull fracture; but perhaps worst of all, she later began having grand mal seizures.

Mary Lou's mother, Louise Duncan, who was driving alone in her own car, had been a witness to the accident. Officers attempted to interview her at the scene, but she was hysterical and kept repeating, "My God, there was nothing he could do." Later, at the hospital, she was a bit calmer. She said that she saw her daughter run across the street and then hesitate, as if she wasn't sure she could make it. The car seemed to Mrs. Duncan as if it was going "a mite fast." Yet, she conceded that because the car was coming at her and she was concerned about her daughter, it was difficult for her to tell the exact speed. She restated that there was nothing the driver could do.

Over a year later, a deposition was taken from Mrs. Duncan. She seemed to remember things differently:

Where were the other children when the accident happened?
They made it to the other side. I don't know why they weren't holding her hand, but they weren't.
Did you sense that Mary Lou was going to be struck?
I sure did. That car was coming like a bat out of hell. I remember hitting my brakes and screaming out, "God, Mary Lou, watch out!" Then she got run down and I don't remember very much of what happened for the next few minutes.
The lawyers at the institute were asked how a person could remember that the driver of a car was going "a mite fast" at one point in time and a year later recall that he was going "like a bat out of hell." To examine this question for ourselves, let us first set aside the possibility that the witness is deliberately lying on one of these two occasions. Next, we might ask whether the witness was exposed to any postevent information during the interval between her two recollections. If no evidence for this sort of external influence can be found, a reasonable alternative is that the witness's own internal thoughts, wishes, and desires intruded during the interval. The witness's thoughts bend in a direction that would be advantageous for her purposes. The strong influences that one's wishes and desires have can be quite unconscious. The witness in this case was suing the driver defendant and the company that employed him. Thus, it was in her best interests to "remember" that the driver was speeding. It would then be far more likely that the jury would find him negligent. The way a person thinks about things clearly affects how they are remembered.


The classic study showing that the way a person labels a given object or situation can dramatically affect the way that situation is remembered was conducted in the early 1930s (Carmichael et al. 1932). The subjects in their research were shown nonsense shapes such as those depicted in figure 4.7. Some subjects saw only the shapes listed in the column called "original stimuli." Other subjects were given a label that corresponded to each shape. Thus, the first nonsense shape shown in figure 4.7 might be called "curtains in a window" or it might be called "diamond in a rectangle."

Later on, subjects were asked to reproduce the figures they had seen. Two things happened. First, the subjects who had been given verbal labels reproduced more of the figures. Second, the drawings by subjects who were given verbal labels looked very different from the drawings by subjects who were given no labels. The labels caused the drawings to be distorted in a very specific direction. Thus, a subject who had been given the label "curtain in a window" drew a figure that looked very much like curtains.

Figure 4.7. Stimuli and verbal label given to subjects, and the reproductions that the subjects produced. (Adapted from Carmichael et al. 1932.)

Of course in this study the experimenters themselves provided the labels to the subjects. Thus, in a sense the labels acted as a piece of postevent information. But labels thought of spontaneously by the subject also affect later memory. This result was shown in a study of memory for colors (Thomas and DeCapito 1966). The subjects in this study were seated in the experimental room and were told that they would be exposed to a color which they should label. Later they were tested in the following way: Each time they recognized the color that had been initially shown, they should respond by lifting a finger off of a telegraph key where it was resting. The critical trials concerned a bluish-green color stimulus. On the later test, those subjects who had initially called the color green tended to respond later to greener colors; those who had initially labeled it blue tended to respond to bluer colors. Other investigators have found this same effect, namely, that when a subject labels a color in a particular way, he tends to remember it differently than does a subject who initially labels it another way (Bornstein 1974). These investigators have also found, as Carmichael and colleagues did, that when the experimenter provides the label, the same distortion in memory occur (Thomas et al. 1968; Bornstein 1976). Thus, it appears that a subject's linguistic responses to an object can cause distortions similar to those that have been observed when others provide the information externally.


It is common practice for witnesses to a robbery to attend a five- or six-person lineup and to be asked to indicate whether they recognize the person who committed the robbery or whether anyone in the lineup resembles that person. When a witness responds with a statement such as, "I'm not really sure, but number five sort of looks like him," the witness is giving a response that is obviously closer to a guess than to a confident report. Unfortunately, this guess has all too often graduated over time to the standing of a confident observation.

Guessing can be dangerous. When a witness is uncertain, guessing can fill gaps in memory. Later, when searching his memory, a witness may incorrectly "recall" something that had earlier been merely a guess but is now a part of memory. Furthermore, while an initial guess may be offered with low confidence, later, when the witness mistakes the guess for a real memory, the confidence level can rise. This seems to occur because a witness is now "seeing" an item that he himself has constructed in memory.

In the first experiment in a set of studies designed to explore these phenomena (Hastie et al. 1978), undergraduates were shown a slide show depicting a street scene which involved four actors in a simulated mugging: a thief, a victim, and two bystanders. The show included twenty-five slides and was displayed at a two-second rate so that the entire series lasted about one minute. After a short filler activity, subjects were presented with a booklet containing a number of questions about the slide show. They were asked about particular objects, as well as things like: What was happening in the street scene? What were the motives of each of the characters? What do you think happened in the scene after the slide show was finished? One group of subjects was told that if they could not remember a particular detail, they should answer the question anyway, guessing if they had to. Another short filler activity was presented, and finally the subjects' memories were tested. They were asked questions about details from the slide show, such as: "Was the male bystander wearing glasses? Yes or no?" "What color was the victim's coat? Black, green, or brown?" For each question they were encouraged to say "I don't know" if they had no memory. Guessing was discouraged.

Subjects who had guessed on the earlier test were significantly less accurate on this final test than were subjects who had not guessed on the earlier test. The subjects who had guessed earlier were less likely to use the "I don't know" response category and more likely to produce an error by giving an answer that was simply wrong.

The second experiment in this series gave essentially the same result. Subjects viewed a short film depicting an automobile-pedestrian accident. After a short filler activity, they answered some questions about the accident. A critical question was posed to a portion of these subjects. It asked, "What color was the stationwagon that passed the accident?" when in fact there was no stationwagon in the film. The subjects were urged to guess when uncertain. After another filler activity, the subjects took a test. The critical question on this test was. "Did you see the stationwagon that passed the accident?" Subjects answered the questions and then gave their confidence in their answer by responding on a five-point scale, "one" indicating that the subject was guessing and "five" indicating high confidence. Two days later the subjects returned and took the identical test again.

The results of this experiment were clear. The subject who had initially been asked to guess about the color of the stationwagon were more likely to think they had seen such a vehicle when asked later on. Furthermore, their confidence continued to rise between their first test and their second one. Control subjects who had not guessed about the stationwagon were less likely to think they had seen one, and their confidence was lower on the second test than the first.

There are several explanation for why these guessing effects may occur. One is that guessing causes a change in the witness's underlying memory representation. A witness to a fast-moving event forms a skeletal representation of that event. Gaps in that representation are filled by whatever means are available. If the witness guesses the color of a supposed stationwagon, this information may be used to fill out the schematic representation. A stationwagon of some given color is essentially added to the representation. Later on, when queried about the stationwagon, the witness may retrieve from memory the stationwagon that he himself has constructed. It appears to be more vivid each time it is recalled, and this accounts for the witness's greater confidence. In addition, there are undoubtedly great pressures upon witnesses to be complete and accurate. Furthermore, there are social pressures to be sure of one's views. It may be that once a witness has given his response, he has a need to indicate that he is confident of his response, not wishy-washy. It is likely that both memory and social factors operate in this sort of eyewitness situation.

Freezing Effects

When a person is asked to recall some previously learned material, statements that appear in an early recollection tend to reappear later on. Thus, if a witness to an accident reports early on that the driver of the damaged vehicle ran a red light, this detail would be likely to appear in later recollections, whether it was true or not. This high degree of persistence in the contents of one's recollection has been called the freezing effect, and is closely related to the guessing problem. In essence, early comments are frozen into place in one's memory and pop up frequently when the witness recalls his experiences at later times. The problem here lies in the fact that, although the early comment may often be true reflections of reality, sometimes they may not. It is not uncommon for false comments to persist as well.

Nowhere have freezing effects been shown as effectively as in the study by Kay (1955), whose work was substantially influenced by the work of Bartlett during the 1920s and 1930s. Bartlett (1932) presented subjects with a story or drawing and later asked people to repeat or reproduce this information several times. Among other things, Bartlett reported great inaccuracy in people's reproduction; but of real interest here is the great persistence he found in the content of a reproduction once the first version had been given. Kay has also noted that the content of people's recollections, though often far from accurate, is stubbornly maintained. Kay put the question in an interesting way: "Why should the initial reproduction be so stable when that version itself is an inaccurate reproduction of the original? or alternatively, why can accuracy of reproduction be achieved between one inaccurate version and another and not between the original and its first reproduction?"

Kay's experimental procedure was straightforward. Prose passages were read to subjects, who were asked to reproduce them five minutes later. The passages were read again, and the subjects reproduced them again. One week later the subjects attempted another reproduction. followed by the originals being read to them again, and this routine was followed the next week. The experiment was conducted on these lines until six reproductions in all had been made. Four months after this they attempted one last reproduction. The following example gives some idea of the passages that Kay used:

Wanted stenographic position -- I take shorthand notes quicker than a new band of Oklahoma outlaws can spring up. I lay a smoke screen with a typewriter, and I do not stay long with one firm because I work so fast I burn the office down.

References: several large firms in Oklahoma City had complete new furniture of steel installed after I had worked there a few days, and the police raiding squad stationed me with a typewriter, and copying the encyclopedia, front of the door of a speakeasy to smoke out the bootleggers instead of throwing tear gas bombs into the building. Warning -- I advise the person who contemplates hiring me to have his office heavily insured, because if I work for him he will have a sure case of conflagration on hand.
When Kay examined the various reproductions, he found that any one version bore a much greater resemblance to its immediate predecessor than either one of them bore to the original material. If only correct information persisted, this would be a fine result. The problem is that errors persist as well. In fact, as many as 95 percent of the errors that existed in one version might persist in the next version, despite the fact that the original material had been presented again between the two versions. One subject used the expression "the entire stock" in all seven of his reproductions when the original passage he heard had said "the whole stock." Another kept using "a vague urge" when the original had said "a vague impulse." Another kept saying "a half forgotten geography lesson" when the correct phrase was "subconscious memory of a geography lesson." These verbal changes might perhaps be considered minor. But others are major. One subject persistently remembered something about "the grandmother in Australia" when he had actually heard about "the geography lesson on Australia" and "the recollection of his grandmother" -- an interesting amalgamation. Errors persisted in people's memories of the passage about the stenographic position, quoted above. For example, one subject insisted on interpreting "several large firms" as "the last three firms" and in two versions she also "set fire within three days" to the offices, while another subject kept repeating "three weeks" for "a few days." One subject maintained throughout that it was the "Oklahoma City Police Vice Squad," while another made it "F.B.I. officials"; a third described it as "the police riot squad," while a fourth related how "the State Police hired me to sit on the sidewalk."

In the typical eyewitness situation, or even the typical memory experiment, a subject is presented with information once; his memory or recollection is tested at a later time, perhaps on several occasions. Kay's experiment is unusual -- and his results more striking -- because the person's memory for the original material is refreshed by presenting the original again, and yet inaccurate recollections persist.

The important role that the intervening thoughts, remarks and recollections of a witness can have upon a final account -- say, in court -- were well known long before Kay did his work. Whipple (1909) wrote, "When a given reporter is called upon to make his report several times, the effect of the repetition is complex, for (1) it tends in part to establish in mind the items reported, whether they be true or false, and (2) it tends also to induce some departure in later reports, because these are based more upon the memory of the verbal statements of the earlier reports than upon the original experience itself" (pp. 166-167). Whipple felt the judicial system should act on this knowledge and argued that "most writers. . . believe that legal procedure should, insofar as possible, be arranged to reduce the number of times that witnesses are called upon to testify."

During the time between an event and a witness's recollection of that event -- a period often called the "retention interval" -- the bits and pieces of information that were acquired through perception do not passively reside in memory waiting to be pulled out like fish from water. Rather, they are subject to numerous influences. External information provided from the outside can intrude into the witness's memory, as can his own thoughts, and both can cause dramatic changes in his recollection.

People's memories are fragile things. It is important to realize how easily information can be introduced into memory, to understand why this happens, and to avoid it when it is undesirable.