Suppose you were a juror in a rape trial where the complainant was alcohol-intoxicated during the alleged attack. The complainant positively identified (ID) the defendant as the attacker from a police lineup a week later, What would you make of this evidence? What if the complainant was highly certain that her ID was accurate? Chances are, the defense would argue that the ID evidence is not reliable because the complainant was intoxicated.
For decades, researchers have argued that witnesses' confidence in the likely accuracy of their memory is not related to their actual accuracy. Some psychologists in the US have testified to this 'fact' before juries in countless cases in an effort by the defence to discredit the complainant's memory evidence.
Eyewitness Memory and DNA Exonerations
The general view that ID evidence is inaccurate is no doubt underscored by DNA exonerations, the vast majority of which are rape cases. The defendant was implicated as the culprit by an eyewitness or victim in more than two-thirds of these cases. Rape is a traumatic experience (stress is often thought to negatively affect memory accuracy, according to memory experts) and victims are often under the influence of alcohol (also thought by experts to decrease memory accuracy). Everything considered, perhaps we should be especially skeptical of the rape complainant’s positive ID of the defendant even though the victim was highly confident.
Is there scientific justification for the view that victims who have positively identified someone as the culprit with high confidence are inaccurate if they were alcohol-intoxicated during the crime?
Research just published by Semmler, Dunn. Mickes, and Wixted (2018) sheds light on the issues (see below for some technical details). Their research (also see Palmer et al., 2013) shows that confidence increases with accuracy, even when learning conditions are less than optimal (e.g.,the witness got a brief look a the culprit's face). In other words, IDs made at the highest levels of confidence are more likely to be accurate compared to IDs made at low levels of confidence, regardless of learning conditions.
These are laboratory findings – what about the real world? Interestingly, in every DNA exoneration case for which data are available, the identifying witness/victim expressed low certainty in the accuracy of their ID (see John Wixted’s research, for example). Had investigators taken the witness’/victim’s confidence into account, the ID evidence would have been viewed as non diagnostic of the suspect’s guilt, and the case would not have been tried on the basis of the ID evidence. Further, confidence tracks accuracy, even in real world cases (see Wixted et al. 2016).
Alcohol and Lineup Identification Research
Does confidence track accuracy even if the victim was alcohol-intoxicated during the rape? We set out to test this question in a study in which we randomly assigned research participants to consume an alcoholic (dosed to achieve a BAC of .08% by drinking 3 vodka drinks within 15 minutes) or nonalcoholic beverage (tonic water) and then had them engage in an interactive dating scenario that ended in rape (Flowe et al., 2017). Up to a week later, we tested their ability to identify the rape perpetrator from a simultaneous lineup (i.e., all of the test faces were presented together). Half of our participants saw a lineup in which the perpetrator was present. The other half saw a lineup in which the perpetrator was absent. A correct response was identifying the perp when he was present, or not identifying anybody when the perp was absent.
We ran this study to test two competing hypotheses regarding the effects of alcohol on the confidence-accuracy relationship. One idea, the optimality hypothesis, predicts that confidence will track accuracy only for people who are sober during the rape scenario. The optimality account specifies that confidence and accuracy are related to the extent that the conditions of the crime and the ID are optimal for the witness in being able to remember the perpetrator. In contrast, the information-based theory of confidence put forward by Palmer and colleagues (2013) predicts that confidence will track accuracy for intoxicated participants if they take into account that alcohol can impair memory. The information-based theory specifies that confidence tracks accuracy when people consider the factors can impair their memory for the culprit and adjust their confidence accordingly.
.
So, what did we find? We found support for the information-based theory. As shown below, ID accuracy increased as confidence increased, regardless of alcohol-intoxication during the rape scenario. In fact, participants who were alcohol-intoxicated during the rape were less likely to be overconfident (i.e., to show higher confidence than warranted given one’s actual accuracy), indicating that participants who had consumed alcohol were better calibrated than their sober counterparts.
So, what do we make of a highly confident complainant at trial? If we can extrapolate from the research above to real world cases, provided that the complainants’ ID confidence was obtained immediately after the ID and there were no extra-memory influences on the complainant's confidence, then it is likely that the ID evidence is accurate. Whether we can extrapolate depends on the extent to which features of cases map onto the features of the research on which we are basing our conclusions. Further, theories about confidence and accuracy are actively being researched and debated in psychology (for examples, see this recent paper by John Wixted and this other one, c.f. this one by Wade and colleagues and this one by Berkowitz urging caution).
For now, some tentative conclusions:
-Blanket statements about alcohol's effects on memory accuracy are not proper. Expert testimony to suggest that victim alcohol-intoxication during rape has necessarily impaired the victim’s memory accuracy are not helpful to juries, especially if the complainant was highly confident in the accuracy of her memory at the time of the ID, all other things being equal.
-Police and others should take on board a victims’ confidence at the time of the ID to determine likely accuracy, all other things being equal.
Endnotes: Technical Details on Accuracy and CAC
Some technical details. Semmler et al. (2018) point out that we have to bear in mind that there are two types of ID accuracy. The first is discrimination accuracy, or the ability of the victim to distinguish between guilty and innocent suspects in a lineup. Usually, lineups are composed of the police suspect and filler faces (normally 5 persons in US lineups and 8 persons in UK parades), who are people who appear physically similar to the victim’s description of the culprit. (The role of fillers is to reduce the odds that the suspect is identified based on guessing alone.) If the police have arrested someone who is innocent of the crime, and the victim positively identifies the innocent suspect from the lineup, this would be a false positive ID or a false alarm. On the other hand, if the police have arrested someone who is guilty of the crime and the victim positively identifies the guilty suspect, this would be an accurate positive ID or a hit. To maximize discrimination accuracy, we should adopt policies for questioning and testing rape victims that maximize the hit rate and minimize the false alarm rate to the greatest extent possible.
The other type of accuracy is the positive predictive value (PPV) which we use to create a confidence accuracy characteristic curve (CAC). The PPV is the number of times that victims accurately identify the guilty culprit from the lineup out of the total number of positive identifications that victims make. For instance, if 5 victims positively identify a guilty suspect and 5 victims positively identify innocent suspects, then the PPV is 5 / 5 guilty IDs + 5 innocent IDs = 50%. PPV is the measure of accuracy that is relevant when police, prosecutors and jurors are faced with the task of evaluating a positive identification of a suspect. That is, given that the victim has identified the suspect from a lineup, PPV measures the probability that the ID is accurate. We can calculate PPV at every level of confidence a victim expresses immediately after they make their ID to draw the CAC. What we find is that PPV increases with accuracy. Victims are well-calibrated to the extent that their confidence tracks their ID accuracy.
Comments