OF JOURNAL VERBAL LEARNING AND VERBAL BEHAVIOR
13, 585-589 (1974)
Reconstruction of Automobile Destruction : A n Example of the Interaction Between Language and Memory' ELIZABETH F. LOFTUSN D JOHN C. PALMER A
University of Washington
Two experiments are reported in which subjects viewed films of automobile accidents and then answered questions about events occurring in the films. The question, "About how fast were the cars going when they smashed into each other?" elicited higher estimates of speed than questions which used the verbs collided, bumped, contucted, or hit in place of smashed. On a retest one week later, those subjects who received the verb smashed were more likely to say "yes" to the question, "Did you see any broken glass?", even though broken glass was not present in the film. These results are consistent with the view that the questions asked subsequent t o an event can cause a reconstruction in one's memory of that event.
1
How accurately do we remember the as to how fast a vehicle was actually traveling details of a complex event, like a traffic (Gardner, 1933). In one test administered to accident, that has happened in our presence? Air Force personnel who knew in advance More specifically, how well do we do when that they would be questioned about the speed asked to estimate some numerical quantity of a moving automobile, estimates ranged such as how long the accident took, how fast from 10 to 50 mph. The car they watched was the cars were traveling, or how much time actually going only 12 mph (Marshall, 1969, elapsed between the sounding of a horn and p. 23). Given the inaccuracies in estimates of the moment of collision? It is well documented that most people are speed, it seems likely that there are variables markedly inaccurate in reporting such numeri- which are potentially powerful in terms of cal details as time, speed, and distance (Bird, influencing these estimates. The present 1927; Whipple, 1909). For example, most research was conducted to investigate one people have difficulty estimating the duration such variable, namely, the phrasing of the of an event, with some research indicating that question used to elicit the speed judgment. the tendency is to overestimate the duration of Some questions are clearly more suggestive events which are complex (Block, 1974; than others. This fact of life has resulted in Marshall, 1969; Ornstein, 1969). The judg- the legal concept of a leading question and in ment of speed is especially difficult, and legal rules indicating when leading questions practically every automobile accident results are allowed (Supreme Court Reporter, 1973). in huge variations from one witness to another A leading question is simply one that, either This research was supported by the Urban Mass by its form or content, suggests to the witness Transportation Administration, Department of Trans- what answer is desired or leads him to the portation, Grant No. WA-11-0004. Thanks go to desired answer. Geoffrey Loftus, Edward E. Smith, and Stephen In the present study, subjects were shown Woods for many important and helpful comments, films of traffic accidents and then they Reprint requests should be sent to Elizabeth F. Loftus. answered questions about the accident. The Department of Psychology, University of Washington, subjects were interrogated about the speed of Seattle, Washington 981 95.
Copyright 0 1974 by Academic Press, Inc. All rights of reproduction in any form reserved. Printed in Great Britain
585
586
LOFTUS AND PALMER
the vehicles in one of several ways. For example, some subjects were asked, "About how fast were the cars going when they hit each other?" while others were asked, "About how fast were the cars going when they smashed into each other?" As Fillmore (1971) and Bransford and McCarrell (in press) have noted, hit and si.r.lasl?ed may involve specification of differential rates of movement. Furthermore, the two verbs may also involve differential specification of the likely consequences of the events to which they are referring. The impact of the accident is apparently gentler for hit than for .n~ia.dtctI.
EXPERIMENT 1 Metliod Forty-five students participated in groups of various sizes. Seven films were shown, each depicting a traffic accident. These films were segments from longer driver's education f i l m borrowed from the Evergreen Safety Council and the Seattle Police Department. The length of the film segments ranged from 5 to 30 sec. Following each film, the subjects received a questionnaire asking then1 first to, "give an account of the accident you have just seen," and then to answer a series of specific questions about the accident. The critical question was the one that interrogated the subject about the speed of the vehicles involved in the collision. Nine subjects were asked, "About how fast were the cars going when they hit each other?" Equal numbers of the remaining subjects were interrogated with the verbs smashed, collided, buiiiped, and contacted in place of hit. The entire experiment lasted about an hour and a half. A different ordering of the films was presented to each group of subjects.
TABLE 1 S PEED ESTIMATESR THE VERBS FO USED IN E XPERIMENT 1 Verb
Sinas hed Collided Bumpcd Hit Contacted
1
1
II
Mean speed estimate 40.5 39.3 38.1 34.0 31.8
efTects, yielding a significant quasi F ratio, !"(5,55) = 4.65, < .005. p Some information about the accuracy of subjects' estimates can be obtained from our data. Four of the seven films were staged crashes; the original purpose of these films was to illustrate what can happen to human beings when cars collide at various speeds. One collision took place at 20 mph, one at 30, and two at 40. The mean estimates of speed for these four films were: 37.7, 36.2, 39.7, and 36.1 mph, respectively. In agreement with previous work, people are not very good at judging how fast a vehicle was actually traveling. Discussioii The results of this experiment indicate that the form of a question (in this case, changes in a single word) can markedly and systematically affect a witness's answer to that question. The actual speed of the vehicles controlled little variance in subject reporting, while the phrasing of the question controlled considerable variance. Two interpretations of this finding are possible. First, it is possible that the differential speed estimates result merely from response-bias factors. A subject is uncertain whether to say 30 mph or 40 mph, for example, and the verb siiiaslied biases his response towards the higher estimate. A second inter- P s pretation is that the question form causes a change in the subject's memory representation of the accident. The verb siiiashed may change a subject's memory such that he
R esiilt s Table 1 presents the mean speed estimates for the various verbs. Following the procedures outlined by Clark (1973), an analysis of variance was performed with verbs as a fixed effect, and subjects and films as random
LANGUAGE A N D MEMORY CHANGES
587
"sees" the accident as being more severe than it actually was. If this is the case, we might expect subjects to "remember" other details that did not actually occur, but are commensurate with an accident occurring at higher speeds. The second experiment was designed to provide additional insights into the origin of the differential speed estimates.
with hit the estimate was 8.00 mph. These means are significantly different, t (98) = 2.00, p < -05.
TABLE 2
D ISTRIBUTION
OF "YES" AND "NO" PONSES TO THE
A NY
QUESTION, "DID BROKEN GLASS?'
YOU
R ESSEE
E XPERIMENT I1
Verb condition
Method Response Smashed Hit Control ~One hundred and fifty students participated Yes 16 7 6 in this experiment, in groups of various sizes. No 34 43 44 A film depicting multiple a car accident was shown, followed by a questionnaire. The film Table 2 presents the distribution of `.yes" lasted less than 1 min; the accident in the film and "no" responses for the smashed, Itit, and lasted 4 sec. At the end of the film, the subjects received a questionnaire asking them first to control subjects. An independence chi-square describe the accident in their own words, and test on these responses was significant beyond 2 ) then to answer a series of questions about the the .025 level, ~ ~ ( = 7.76. The important result in Table 2 i s that the probability of accident. The critical question was the one that interrogated the subject about the speed saying "yes," P(Y), to the question about of the vehicles. Fifty subjects were asked, broken glass is .32 when the verb sn?a.died is "About how fast were the cars going when used, and .14 with hit. Thus smashed leads they smashed into each other?" Fifty subjects both to more "yes" responses and to higher were asked, "About how fast were the cars speed estimates. It appears to be the case that going when they hit each other?" Fifty the effect of the verb is mediated at least in subjects were not interrogated about vehicular part by the speed estimate. The question now arises : Is sniushed doing anything else besides speed. One week later, the subjects returned and increasing the estimate of speed? To answer without viewing the film again they answered this, the function relating P(Y) to speed a series of questions about the accident, The estimate was calculated separately for stiiashed critical question here was, "Did you see any and hit. If the speed estimate is the only way broken glass?" which the subjects answered in which effect of verb is mediated, then for a by checking "yes" or "no." This question was given speed estimate, P(Y) should be inembedded in a list totalling 10 questions, and dependent of verb. Table 3 shows that this is it appeared in a random position in the list. TABLE 3 There was no broken glass in the accident, PROBABILITY OF SAYING "YES" TO, "D ID Y O U SEE but, since broken glass is commensurate with N ON S PEED accidents occurring at high speed, we expected A NY B R O K ~GLASS?' CONDIT~ONALIZED ESTIMATES that the subjects who had been asked the smashed question might more often say "yes" Speed estimate (mph) k to this critical question. Verb condition 1-5
.09 .06
6-10 .27 .09
11-15 .41 .25
16-20 .62 .50
Results The mean estimate of speed for subjects interrogated with smaslted was 10.46 mph;
Smashed Hit
588
LOFTUS AP4D PALMER
not the case. P(Y) is lower for hit than for smashed; the difference between the two verbs ranges from .03 for estimates of 1-5 mph t o .I8 for estimates of 6-10 mph. The average difference between the two curves is about .12. Whereas the unconditional difference of . I8 between the smashed and hit conditions is attenuated, it is by no means eliminated when estimate of speed is controlled for. It thus appears that the verb smashed has other effects besides that of simply increasing the estimate of speed. One possibility will be discussed in the next section.
DISCUSSION
To reiterate, we have first of all provided a n additional demonstration of something that has been known for some time, namely, that the way a question is asked can enormously influence the answer that is given. In this instance, the question, "About how fast were the cars going when they smashed into each other?" led to higher estimates of speed than the same question asked with the verb smashed replaced by hit. Furthermore, this seemingly small change had consequences for how questions are answered a week after the original event occurred. As a framework for discussing these results, we would like to propose 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." Discussing the present experiments in these terms, we propose that the subject first forms some representation of the accident he has witnessed. 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 did indeed smash into each other.
When these two pieces of information are integrated, the subject has a memory of a...