Psychological Science, 1999
Giving eyewitnesses confirming feedback after they make a lineup identification (e.g., "Good, you identified the actual suspect") not only inflates their recollections of how confident they were at the time of the identification, but also how good their view was, how much attention they paid during witnessing, how quickly they identified suspect, and other testimony-relevant judgments (Wells & Bradfield, 1998). We replicated the post-identification feedback effect with eyewitnesses who had made false identifications (N = 156). Critical conditions were added in which, after their identification but prior to the feedback, eyewitnesses were given instructions to privately think about their confidence, their view, and other matters. Other eyewitnesses were given the thought instructions subsequent to the feedback manipulation. Prior thought served to mitigate the effects of feedback but subsequent thought did not. In addition, even without feedback, privately thinking about confidence had some confidence-inflating properties of its own.
There is increasing evidence that mistaken eyewitness identifications from lineups and photospreads is the largest cause of jury convictions of innocent persons. Analyses of cases in which people were mistakenly convicted by juries have shown eyewitness misidentification to account for more than all other causes of wrongful conviction combined (see Wells, Small, Penrod, Malpass, Fulero, & Brimacombe, 1998). The problem of mistaken identification is exacerbated when eyewitnesses are mistaken yet highly confident in their testimony (Wells, 1998).
Eyewitness scientists have been exploring ways that the justice system could minimize the problem. Among the recommendations is blind testing; the person who administers the lineup should not know which person is the suspect (Wells, 1993; Wells & Seelau, 1995). Lineup procedures in the U.S., however, do not involve blind testing and, instead, are conducted by the case agent, the very person who developed the suspect in the first place. Hence, the opportunity for the lineup agent to influence the eyewitness’s identification or influence the way that the eyewitness feels about the identification is substantial. The emphasis of the work we report here is how the lineup agent can influence the testimony that eyewitnesses give about their identifications.
How confident an eyewitness feels about his or her identification is a primary determinant of whether jurors will accept the identification evidence as being accurate (e.g., Cutler, Penrod, & Dexter, 1990; Fox & Walters, 1986; Lindsay, Wells & Rumpel, 1981; Wells, Lindsay & Ferguson, 1979). Therefore, particularly problematic is the phenomenon of false confidence; the tendency of an eyewitness to be highly confident in the accuracy of a false identification. Some instances of false confidence are to be expected from the fact that eyewitness identification confidence and accuracy are only modestly related (see meta-analysis by Sporer, Penrod, Read, D. & Cutler, 1995). Other instances of false confidence, however, could be a product of the way the eyewitness identification evidence is collected. We are concerned with the creation of false confidence by external influences, in particular giving feedback to eyewitnesses after they make their identification. Telling an eyewitness who has made a false identification that s/he identified the same person that another witness identified or that the eyewitness identified the actual suspect leads to robust inflation in eyewitnesses' confidence in their identifications (Luus and Wells, 1994; Wells & Bradfield, 1998).
The Post-identification Feedback Effect
The idea that confirming feedback would lead to confidence inflation is not surprising. What is surprising, however, is that confirming feedback that is given after the identification leads eyewitnesses to misremember how confident they were at the time of the identification (Wells & Bradfield, 1998). Because the feedback manipulation occurs after the identification has already been made, random assignment to conditions assures that the confirming feedback and no feedback (control) eyewitnesses are in fact equally confident at the time of the identification. Hence, confirming feedback does not merely inflate how confident the witness feels after the feedback; it distorts eyewitnesses’ recollections of how confident they recall having been at the time of the identification. Even more surprising is that confirming feedback also distorts the eyewitnesses' recollections of their witnessing conditions (e.g., how good of a view did you have of the perpetrator?) as well as eyewitnesses' recollections of their identification behaviors (e.g., how long did it take you to make the identification?). Simply put, feedback has effects on eyewitnesses' recollections of a wide array of variables. We refer to this as the post-identification feedback effect.
According to Wells and Bradfield (1998), the post-identification feedback effect occurs because eyewitnesses are not forming "on line" evaluations of these variables (e.g., confidence, view, speed to identification) prior to the feedback manipulation. The first time eyewitnesses consider these variables is when they are asked to estimate them; by that time, the feedback manipulation has already happened. Having no significant memorial record of these judgments prior to the feedback, the eyewitnesses can only infer values for these variables in the context of the feedback. Consistent with this interpretation, Wells and Bradfield found that most eyewitnesses thought that the feedback did not influence their judgments. Furthermore, those who said that the feedback did not influence them were as influenced as were those who said the feedback did influence them. Apparently, the absence of a clear memorial trace for these variables prior to the feedback prevented the eyewitnesses from accurately determining whether the feedback influenced their judgments.
We propose a close parallel between the basic tenets of self-perception theory (Bem, 1967) and the processes governing the post-identification feedback effect. According to self-perception theory, to the extent that internal cues are weak or ambiguous, people infer their beliefs from their behavior according to the context in which it occurs. Because eyewitnesses do not have clear memorial traces on which to rely, their internal cues (regarding prior certainty, attention, view, and so on) are weak, making them rely on their behavior (an identification) and the context (feedback) to infer these variables.
If this interpretation is correct, it should be possible to "inoculate" eyewitnesses against the post-identification feedback effect by asking them to privately think about critical variables prior to the feedback manipulation. This pre-feedback thought manipulation should produce a retrievable memory trace that would allow the eyewitnesses to more accurately recall their pre-feedback confidence, their view, how long they took to make the identification, and so on. Hence, with pre-feedback thought, eyewitnesses should be able to rely on internal cues to resist the influence of the feedback manipulation.
We used the post-identification paradigm developed by Wells and Bradfield in which all eyewitnesses made false identifications followed by random assignment to either confirming feedback or no feedback conditions. In some conditions, eyewitnesses were instructed to think privately about several variables (e.g., think about how certain you were, how good your view was, how long you took to make an identification) prior to the feedback manipulation. Others were instructed to think about these variables after the feedback manipulation. Still others were not instructed to think about these variables at all.
Participants and Pre-manipulation Procedure
Participants were 156 undergraduates who received partial course credit. Participants were told that the study involved making personality and occupation judgments about people. Participants were then placed in their own cubicle in which they watched a video recording from an actual in-store security camera that captured the image of a person who shot and killed a security guard shortly after leaving the view of the camera. The video was edited to cue them to the gunman, but participants did not see the shooting. The experimenter then explained that the true purpose of the study was to see if they could identify the gunman. The experimenter gave the eyewitness a five-person photospread that did not contain the actual gunman along with an identification sheet. Because the procedure implies that the actual gunman is in the lineup, eyewitnesses uniformly select someone. As with Wells and Bradfield’s (1998) studies, all of the participants made identifications.
Experimental Manipulations and Dependent Measures
Once the eyewitness selected someone, participants were assigned randomly to one of the five conditions. In all conditions, 6 min elapsed before the dependent measures were taken. In the no thought/no feedback condition, the experimenter waited 6 min, re-entered the cubicle, and gave the eyewitness the dependent variable sheets. In the no thought/confirming condition, the experimenter waited 3 min and re-entered the cubicle saying "Oh, good. I noticed on your identification sheet that you identified the actual murder suspect." The experimenter then left and returned 3 min later with the dependent variable sheets. In the prior thought/no feedback condition, the experimenter re-entered the cubicle and gave the eyewitness written "think instructions." The think instructions sheet asked the participants to spend time thinking privately about how clearly they could see the gunman’s face in the video, how much they focussed on the gunman’s face, how easy it was for them to select someone from the photospread, how good they are at remembering faces, and how sure they were that they identified the right person in the photospread. The experimenter re-entered after 6 min with the dependent variable sheets. In the prior thought/confirming feedback condition, the experimenter re-entered the cubicle and gave the eyewitness the written "think instructions". After 3 min, the experimenter re-entered the cubicle, gave the confirming feedback, and exited. Three min later, the experimenter re-entered with the dependent variable sheets. In the subsequent thought/confirming condition, the experimenter re-entered the cubicle, gave confirming feedback, and exited. Three min later, the experimenter re-entered the cubicle and gave the eyewitness the written think instructions. Three min later the experimenter re-entered the cubicle with the dependent variable sheets. Participants were then left alone to answer the 13 questions shown in Table 1.
As expected, most of the measures were correlated with each other (see Table 2). Nevertheless, it is conceptually important to know whether the feedback manipulation has effects on each of these measures. Clearly, it is possible for feedback to affect confidence without affecting the eyewitnesses' recollections of how good their view was, and so on. Hence, each measure was analyzed with separate 2 (Feedback) X 2 (Prior Thought or Not) ANOVAs. This allowed us to test whether the feedback effect was present and also for each measure and also whether the prior thought manipulation moderated the feedback effect. When the interaction was significant, simple t-tests were used to examine the feedback effect for the prior thought conditions and for the no thought conditions. Then, the subsequent thought/confirming feedback condition was compared to the no thought/confirming feedback condition to see if subsequent thought moderated the effects of confirming feedback. All tests, except otherwise noted, were two-tailed with a criterion alpha level of .05 for ascribing statistical significance.
Confidence. Prior thought served to moderate the effect of feedback on eyewitnesses' recollections of their confidence as evidenced by a significant 2 (Feedback) X 2 (Prior Thought or Not) interaction, F (1, 146) = 7.12, p< .01. Mean confidence by condition is shown in Figure 1 along with 95% confidence intervals around the means. The moderation pattern was as predicted, namely a large effect of feedback in the no-thought conditions, t (59) = 4.7, p < .001, d = .95, was eliminated in the prior thought conditions, t (59) = 0.62, ns, d = .12. The subsequent thought/ confirming feedback condition, however, did not differ from the no thought/ confirming feedback condition, indicating that subsequent thought did not moderate the effects of confirming feedback, t (59) = .4, ns. In addition, the prior thought/no feedback condition yielded significantly more confidence than did the no thought/no feedback condition, indicating that prior thought alone produces confidence inflation, t (59) = 2.05, p < .05.
Other measures. Nine of the other 12 measures yielded significant 2 (Feedback) X 2 (Prior Thought or Not) interactions [all Fs (1, 116) > 4.0, ps < .05)]. The only three measures that failed to yield the interaction were clarity of image, estimated distance of the gunman, and length of time the gunman was in view. Note, however, that the latter two measures also failed to show simple effects of feedback for no-thought conditions, ts (59) < 1.5, ps > .05, making them poor tests of the moderation hypothesis. Further, these two measures have consistently failed to show the feedback effect in prior studies (Wells & Bradfield, 1998). Nine measures that were affected by the feedback manipulation are shown in Figure 2 as difference scores between the confirming and no-feedback conditions for the prior thought and no-thought conditions. [Scoring was reversed in Figure 2 for the ease of identification and speed of identification measures.] As predicted, moderation consistently took the form of lessening the effect of feedback in the think conditions, with the average effect size in the no-thought conditions of d = .60 and the average effect size in the prior-thought conditions of d = .20. Three of the measures, namely view, ease of identification, and speed of identification also showed moderation effects in the subsequent thought condition, as evidenced by significantly lower means in the subsequent thought/confirming feedback condition than in the no thought/confirming feedback condition, ts (59) = 2.17, 2.29, 3.42, respectively, ps < .05). Unlike the confidence measure, there were no differences between the no thought/no feedback condition and the prior thought/no feedback condition for any of these 12 measures, all ts (59) < 1.6), indicating that prior thought alone did affect witnesses’ answers to these questions.
One measure not shown in Figure 2 (because it is not on a similar measurement scale) is the percentage of eyewitnesses who reported that the culprit’s face just "popped out" in the lineup. Overall, only 17% reported the pop out process of identification and the expected frequencies were too small to analyze each condition separately. However, contrasting the three confirming feedback conditions to the two no-feedback conditions indicated a significant one-tailed difference in which "pop out" was reported by only 9% of the no feedback eyewitnesses versus 23% of the confirming feedback eyewitnesses, C 2 (N=150) = 2.97, p < .01, one tailed.
Confirming feedback following eyewitnesses’ false identifications had very strong inflation effects on their recollections of how confident they were at the time of their identification. Feedback also distorted eyewitnesses’ recollections of other testimony-relevant characteristics of the witnessed event and their recollections of the identification experience. Consistent with the contention that such effects occur because of the relative absence of memorial access to pre-manipulation thoughts on these variables, eyewitnesses who were instructed to think about these variables prior to the feedback manipulation were largely unaffected by the manipulation whereas eyewitnesses who were not instructed to think about these variables were strongly affected by the manipulation. In addition, for most of the measures, instructions to think about these variables after the feedback manipulation did not moderate the feedback effect. The latter finding indicates that thought about these variables per se does not inoculate against the feedback effect.
It is tempting to interpret the post-identification feedback effect as a type of hindsight bias in which, having been told the "correct" answer, eyewitnesses assume that they "knew it all along" (Fischhoff, 1975). We are reluctant, however, to consider the feedback effect to be a traditional hindsight effect. Although confidence inflation from feedback seems consistent with the hindsight bias, it is unclear how the other measures (e.g., attention, view) can be construed as hindsight effects in the traditional sense. Why, for instance, should eyewitnesses think that they paid more attention to the culprit’s face after receiving confirming feedback? Why should they recall their view as having been better following confirming feedback? Why should they be more likely to recall that the culprit’s face just "popped out" while viewing the lineup? Although we suspect that these effects are related to the hindsight bias, it appears that feedback effects on these other measures involve an inference process of some type (e.g., I must have had been paying attention; I must have had a good view; it must have just "popped out"). Having no pre-manipulation recollection of these qualities of their witnessing experience, eyewitnesses seem to infer these qualities from the feedback.
We believe that these effects are better construed as a self-perception effect. According to self-perception theory (Bem, 1967), people infer their beliefs or feelings from their behavior in the context in which the behavior occurs, much like an outside observer would. The "outside observer" test is one of the critical features of self-perception theory and it seems to fit the behaviors of eyewitnesses in this work. For instance, an outside observer who believed that an eyewitness made an accurate identification would likely infer that the eyewitness was more confident, had a better view, made the identification more quickly, and so on, than if the outside observer believed that the eyewitness had made a false identification. Furthermore, self-perception theory states that this inference-as-though-observer process occurs only to the extent that internal cues are weak. The prior thought manipulation was designed to strengthen internal cues (creation of a pre-feedback memory) and, as predicted, inferences from the feedback were muted when prior thought was introduced.
Somewhat surprisingly, merely thinking about one’s confidence, view, and so on, itself seems to produce confidence inflation. We call this the "thought alone" effect because the thought instruction itself does not suggest anything about the accuracy of the identification. The size of this effect was about half the size of the feedback effect. The "thought alone" effect, however, seems qualitatively different from the feedback effect because, unlike the feedback effect, it appears to be restricted to the confidence judgments rather than spreading to the other judgments. Although we did not explicitly predict confidence inflation for thought alone, others have found that eyewitness confidence increases when eyewitnesses are instructed to prepare themselves for cross examination (Wells, Ferguson, & Lindsay, 1981) or when eyewitnesses repeatedly answer the same question (Shaw, & McClure, 1996; Shaw, 1996).
The function of a lineup is to find out what the eyewitness knows from his or her own memory, independently of any influences from the agent administering the lineup. There are no legal prohibitions in the U.S. against lineup agents giving feedback to eyewitnesses. To the extent that feedback is a relatively normal practice in real cases, significant percentages of eyewitnesses who make false identifications are being "shifted up" in the confidence distribution in ways that confound the meaning of high confidence. It is difficult enough for triers of fact to distinguish between accurate and false identification testimony without confounding matters with post-identification feedback (see Wells et al. 1979). Feedback contributes additional muddle to an already-confusing set of cues that jurors use in trying to discriminate between accurate and false identification testimony.
The solution to preventing feedback is apparent from Wells and Luus’s (1991) argument that a properly conducted lineup is like a properly conducted scientific experiment. The legal system should require blind testing (the lineup agent should not know which person in the lineup is the suspect) and collect the primary dependent measures (confidence information and other statements) from the eyewitnesses prior to debriefing them regarding the "status" of the identified person.
Bem, D. J. (1967). Self-perception: An alternative interpretation of the cognitive dissonance phenomena. Psychological Review, 74, 183-200.
Cutler, B. R., Penrod, S. & Dexter, H. R. (1990). Juror sensitivity to eyewitness identification evidence. Law and Human Behavior, 14, 185-191.
Fischhoff, B. (1975). Hindsight ¹ foresight: The effect of outcome knowledge on judgment under uncertainty. Journal of Experimental Psychology: Human Perception and Performance, 1, 288-299.
Fox, S. G., & Walters, H. A. (1986). The impact of general versus specific expert testimony and eyewitness confidence upon mock juror judgment. Law and Human Behavior, 10, 215-228.
Luus, C. A. E., & Wells, G. L. (1994). The malleability of eyewitness confidence: Co-witness and perseverance effects. Journal of Applied Psychology, 79, 714-724.
Lindsay, R. C. L., Wells, G. L., & Rumpel, C. (1981). Can people detect eyewitness identification accuracy within and between situations? Journal of Applied Psychology, 66, 79-89.
Shaw, J. S., III, & McClure K. A. (1996). Repeated postevent questioning can lead to elevated levels of eyewitness confidence. Law and Human Behavior, 20, 629- 654.
Shaw, J. S., III. (1996). Increases in eyewitness confidence resulting from postevent questioning. Journal of Experimental Psychology: Applied, 2, 126-146.
Sporer, S., Penrod, S., Read, D. & Cutler, B. L. (1995). Choosing, confidence, and accuracy: A meta-analysis of the confidence-accuracy relation in eyewitness identification studies. Psychological Bulletin, 118, 315-327.
Wells, G. L. (1993). What do we know about eyewitness identification? American Psychologist, 48, 553-571.
Wells, G. L. (1997). Eyewitness identification. In D. Faigman, D. Kaye, M. Saks, & J. Sanders (Eds.) , Modern scientific evidence: The law and science of expert testimony (pp. 451-479). St. Paul: West Publishing Co.
Wells, G. L. & Bradfield, A. L. (1998). "Good, you identified the suspect:" Feedback to eyewitnesses distorts their reports of the witnessing experience. Journal of Applied Psychology, in press.
Wells, G. L., Ferguson, T. J., & Lindsay, R. C. L. (1981). The tractability of eyewitness confidence and its implication for triers of fact. Journal of Applied Psychology, 66, 688-696.
Wells, G. L., Lindsay, R. C. L., & Ferguson, T. J. (1979). Accuracy, confidence, and juror perceptions in eyewitness identification. Journal of Applied Psychology 64, 440-448.
Wells, G. L., & Luus, E. (1990).
Police lineups as experiments: Social methodology as a
framework for properly-conducted lineups.
Personality and Social Psychology Bulletin,
16, 106-117.
Wells, G. L. & Seelau, E. P. (1995). Eyewitness identification: Psychological research and legal policy on lineups. Psychology, Public Policy, and Law, 1, 765-791.
Wells, G. L., Small, M., Penrod, S., Malpass, R. S., Fulero, S. M. & Brimacombe, C. A. E. (1998), Eyewitness Identification Procedures: Recommendations for Lineups and Photospreads, APLS Scientific Review Paper.
"At the time that you identified the person in the photospread, how certain were you that the person you identified from the photos was the gunman that you saw in the video?" from 0% (not at all certain) to 100% (totally certain) in 10% intervals
"How good of a view did you get of the gunman?" from 1 (very poor) to 7 (very good)
"How long would you estimate that the gunman’s face was in view during the video?" from 1 (very little time) to 7 (quite a bit of time)
"How well were you able to make out specific features of the gunman’s face from the video?" from 1 (not at all) to 7 (very well)
"How far away was the gunman?" from 1 (not far) to 7 (very far)
"How much attention were you paying to the gunman’s face while viewing the video?" from 1 (none) to 7 (my total attention)
To what extent do you feel that you had a good basis (enough information) to make an identification?" from 1 (no basis at all) to 7 (a very good basis)
"How easy or difficult was it for you to figure out which person in the photos was the gunman?" from 1 (extremely easy) to 7 (extremely difficult)
"After you were first shown the photos, how long do you estimate it took you to make an identification?" from 1 (I needed almost no time to pick him out) to 7 (I had to look at the photos for a long time to pick him out)
"On the basis of your memory of the gunman, how willing would you be to testify in court that the person you identified was the person in the video?" from 1 (not at all willing) to 7 (totally willing)
Generally, how good is your recognition memory for the faces of strangers you have encountered on only one prior occasion?" from 1 (very poor) to 7 (excellent)
"How clear is the image you have in your memory of the gunman you saw in the video?" from 1 (not at all clear) to 7 (very clear)
"Which of the following statements
best describes how you went about trying to identify the gunman from the
five photos? (circle one)" with choices "the gunman’s photo just ‘popped
out’ at me and I recognized it immediately" versus "I used a process of
elimination, deciding which photos were not of the gunman before
deciding which photo must be that of the gunman"
|
Table 2
|
|||||||||||||||||||||
|
|
|||||||||||||||||||||
|
|
|
|
|
|
|
|
|
|
|
|
|
||||||||||
| Certainty |
0.66
|
|
|
|
|
|
|
|
|
|
|
|
|||||||||
| Goodness of view |
|
|
|
|
|
|
|
|
|
|
|
||||||||||
| Time face was in view |
|
|
|
|
|
|
|
|
|
|
|||||||||||
| Details of face |
|
|
|
|
|
|
|
|
|
||||||||||||
| Distance from gunman |
|
|
|
|
|
|
|
|
|||||||||||||
| Amount of attention |
|
|
|
|
|
|
|
||||||||||||||
| Basis for identification |
|
|
|
|
|
|
|||||||||||||||
| Ease of identification |
|
|
|
|
|
||||||||||||||||
| Speed of identification |
|
|
|
|
|||||||||||||||||
| Willingness to testify |
|
|
|
||||||||||||||||||
| Ability to identify strangers |
|
|
|||||||||||||||||||
| Clarity of memory image |
|
||||||||||||||||||||