Lineup identification accuracy: The effects of alcohol, target presence, confidence ratings, and response time

One hundred and twenty undergraduate psychology students were offered either course credit or £10 for participation in the study. The size of this sample was determined by an a priori power analysis (β = .8) based on the identification data of Yuille and Tollestrup (1990) , who observed a small effect of alcohol intoxication on identification accuracy for TA lineups ( r m = .245) for a contingency table with 1 df . The sample consisted of 23 males and 97 females, all aged between 18-40 years ( M = 20.40 years, SD = 4.12).

A 3 × 2 between-subjects design was used to examine the effect of alcohol intoxication on subsequent identification accuracy, confidence level, and decision time. The two variables manipulated were alcohol ingestion (no alcohol control, placebo control, and alcohol) and lineup type (TP vs. TA). In both TP and TA lineups, the position of the perpetrator's image (or its replacement in the case of the TA condition) was rotated such that it appeared in each of the six possible positions across participants with the foil faces being presented in a random order for each lineup. Participants were randomly allocated to each of the experimental conditions.

The measure of identification performance was simply whether or not participants correctly identified the presence or absence of the perpetrator from the line-up one week after witnessing him in the video footage. The confidence each participant had in the accuracy of their identification decision was recorded via a 7-point Likert scale (where 1 = not at all confident, and 7 = very confident ). Decision time was defined as the time taken in milliseconds between the lineup being presented on the screen to the time the participant made a key press to either identify the number of the photograph they believed to be the target or to reject the lineup. The BrAC in participants’ deep lung air was recorded using a Lion Alcolmeter 500, a device that is Type Approved by the UK Home Office for police screening and evidential testing. The unit for this measure was milligrams of alcohol per litre of breath (mg/l). Participants were also asked to report their perceived level of intoxication on a scale of 0 ( completely sober) – 100 (extremely drunk).

The stimulus event was a color video of a young man (i.e., the target) furtively inspecting vehicles in a car park, apparently searching for one to break into. The video was a little grainy so as to represent CCTV footage and has been used successfully in previous published research ( Kneller, Memon, & Stevenage, 2001 ). The video was approximately 60 seconds long with the target, the only person visible, shot at close range and from a distance.

For the lineup presentation participants were presented with either a TP or TA lineup consisting of six 10 cm (height) by 7 cm (width) color portrait photographs showing the head and shoulders of six young males displayed in a 2 (rows) × 3 (columns) array. All photographs were taken using identical background, lighting, and distance. The TP condition included the target along with five foil faces that had been previously rated as similar in appearance to the target. In the TA conditions the target was replaced by a similar looking sixth foil. All lineups were presented using Superlab 4.0 Stimulus Presentation Software.

Prior to arriving at the lab all participants were advised that they could consume alcohol in the study and that they should not therefore drive to the experiment venue, or do anything that might be considered inappropriate (e.g., attending a lecture) or risky (e.g., play sport, operate machinery) immediately after the experimental session, as they may still be intoxicated. Additionally, they completed a screening process, conducted by the principal researcher, confirming their eligibility to take part in the study and that they had consumed at least the same amount of alcohol used in the study (at least 1.42 liters of beer with 5% alcohol content, or its equivalent) within a single drinking session in the past month. To assist in the accurate reporting of drinking behaviours, participants were issued with information relating to how much alcohol was equivalent to one standard unit, as provided by the World Health Organisation Alcohol Use Disorders Identification Test (AUDIT; Babor, Higgins-Biddle, Saunders, & Monteiro, 2001 ). Our alcohol screen excluded all participants who reported that they were on medication or had been given medical advice not to drink alcohol. Participants were also asked to report any other reasons why they should not drink alcohol.

On their arrival at the lab, participants were read an experiment information sheet, confirmed that the information they provided on the screening form was correct, that they were not driving that day or engaging in any activity that would be risky whilst intoxicated, and signed a consent form. They were then breathalysed (to confirm their BrAC was zero) and randomly assigned to one of the three alcohol treatment conditions.

Those in the alcohol group were weighed, to determine their alcohol dosage, then given a 450 ml drink containing 0.6 ml of ethanol alcohol per kg of body weight mixed with pure fruit juice. So, for example, a person weighing 50 kg would receive 30 ml of ethanol and 420 ml of orange juice. Participants in the placebo condition were also weighed to maintain the ruse that they would be given alcohol but were just given an equivalent volume of pure fruit juice with a few drops of ethanol floating on the surface. The rim of the glass was spritzed with a 50:50 mixture of ethanol and water to provide an odour of alcohol. This was deemed sufficient to create an illusion of drinking alcohol whilst not actually leading to intoxication (see Fillmore & Vogel-Sprott, 1995 ). Participants in both alcohol and placebo conditions were told their drink contained alcohol. Those in the control group were served 450 ml of pure fruit juice only and were reassured that the drink contained no alcohol. All participants consumed their drink within a 15-minute period after which they relaxed for a further 15-minute alcohol absorption period, during which they were free to read magazines supplied by the researchers, browse on their smartphones, or chat to the experimenter.

A second BrAC measure was then recorded to determine the intoxication level, followed by an additional measure of subjective intoxication. Participants were then asked to simply watch the video without being informed of its purpose. After this initial testing phase, sober participants left the lab but intoxicated and placebo participants were strongly encouraged to stay behind and relax in a comfortable room with magazines and soft drink facilities until their BrAC returned well below 0.35 mg/l (which approximates 0.08% BAC, the legal UK driving limit). Those who elected not to stay behind were required to sign a disclaimer form confirming their awareness that they had recently consumed alcohol and that they might therefore be in excess of the legal limit for driving. All participants were informed that they were to return to the lab a week later to undergo a few further measures and not to discuss the video with anybody.

On their return participants were randomly assigned to one of the two identification conditions and asked to identify the man seen in the video footage from the associated photographic lineup. All six photographs were presented simultaneously on the screen and were numbered from 1 to 6. Participants were asked to either identify, via a key-press, the number of the photograph they believed to be the target, or reject the lineup by pressing the ‘r’ key. For all lineups, participants received unbiased instructions to the effect that the target might or might not be present in the lineup.

Once participants had made their identification decision they were presented with the question on their level of confidence in that decision. After testing, all participants were fully debriefed as to the nature of the study and their performance.

The data met parametric assumptions and were hence tested for significance using analysis of variance (ANOVA) and hierarchical log linear analysis (HILOG) with an alpha value of .05 taken as the criterion of significance. Data were first tested for the effect of the target and his replacement's lineup position on accuracy. No significant effect was identified, χ 2(5, N = 120) = 5.99, p = .31, Cramer's V = .22. Due to an unequal proportion of male and female participants, the effect of gender on accuracy was also examined. No significant effect of gender was found, χ 2(1, N = 120) = 0.09, p = .82, Cramer's V = .03.

The amount of alcohol given to the 40 participants in the alcohol condition ranged from 41.19 to 83.17 ml (M = 59.21 ml, SD = 10.80). Subsequent measures of participant's BrAC taken immediately prior to viewing the video ranged from 0.11 to 0.40 mg/l ( M = 0.23 mg/l, SD = 0.07).

Those who received alcohol reported a mean subjective level of intoxication of 45.18 ( SD = 15.81), compared to the placebo group who reported a mean of 13.13 ( SD = 11.29), and the control group who reported a mean level of 2.77 ( SD = 5.26). A one-way independent ANOVA indicated a highly significant effect of alcohol treatment on the subjective level of intoxication, F(2, 117) = 144.83, MSE = 134.97, p < .001, ¿p 2 = .7. Post hoc pairwise comparisons further revealed that the subjective intoxication levels of all three groups were significantly different from each other (all p 's < .001), indicating that the placebo condition was successful in producing an expectation of intoxication, albeit one significantly lower than that of the alcohol group.

Responses for all lineups were initially analyzed by examining the frequency of correct responses (target hits from the TP lineups and correct rejections of the TA lineups) and incorrect responses (foil identifications from, and incorrect rejections of, TP lineups or incorrect identifications from TA lineups). Table 1 shows the effect of alcohol intoxication on identification performance for both TP and TA lineups. Overall, 37.5% of participants provided a correct response.

An analysis was first conducted to test the hypothesis that accuracy would be significantly impaired by alcohol intoxication. Collapsed across target presence, 40% of the intoxicated group, 40% of the placebo group, and 32.5% of the control group gave correct responses, although subsequent analysis revealed no significant difference in accuracy between these three groups, χ 2(2, N = 120) = 0.21, p = .89, Cramer's V = .04. A hierarchical loglinear analysis (HILOG) with alcohol, lineup type, and accuracy (accurate/inaccurate) as factors revealed no significant three way interaction, χ 2(2, N = 120) = 0.66, ns , and no significant two-way interactions (all p 's > .05). Only the main effect of accuracy was significant, with participants more likely to make an inaccurate decision than an accurate one, χ 2(1, N = 120) = 7.58, p = .006.

Analyses were subsequently conducted separately for TP and TA lineups to see, specifically, if alcohol intoxication caused an increase in false identifications in the TA lineup condition, as per the findings of Dysart et al. (2002).

The data revealed that intoxicated participants were no less likely than sober or placebo participants to make an accurate identification from a TP lineup, χ 2(2, N = 60) = 2.22, p = .69, Cramer's V = .14.

Overall, whilst participants tended to make more false identifications than correct rejections, no significant association between alcohol condition and identification decision was found, χ 2(2, N = 60) = 0.40, p = .82, Cramer's V = .08.

Table 2 displays the mean confidence ratings by alcohol condition, target presence, and accuracy. While alcohol intoxication was expected to have no significant effect on confidence for witnesses shown a TP lineup, we expected a lower level of confidence from those intoxicated witnesses shown a TA lineup. However, a 3 (Alcohol Condition) × 2 (Target Presence) × 2 (Accuracy) independent-measures ANOVA revealed only a significant main effect of accuracy on decision confidence, with accurate witnesses ( M = 4.89, SD = 0.21) being more confident than inaccurate witnesses ( M = 3.98, SD = 0.16), F(1, 108) = 11.89, p = .001, ¿p 2 = .09. The main effect of alcohol, F(2, 108) = 1.56, p = .94, ¿p 2 = .02, the main effect of target presence on confidence, F(1, 108) = 0.07, p = .78, ¿p 2 = .001, nor any of the interactions (all p 's > .05) reached significance.

An additional analysis was conducted to examine confidence between those who made an accurate choice from a TP lineup and those whose decision was incorrect. A 3 (Alcohol Condition) × 2 (Choice) × 2 (Identification Accuracy) ANOVA revealed, again, only a significant main effect of accuracy, with correct decisions ( M = 4.86, SD = 0.21) being made significantly more confidently than incorrect decisions ( M = 3.94, SD = 0.18), F(1, 51) = 3.93, p = .05, ¿p 2 = .07. Both the main effects of alcohol, F(2, 51) = 1.55, p = .22, ¿p 2 = .06, and choice, F(1, 51) = 0.005, p = .94, ¿p 2 = .00, were found to be non-significant, as were all interactions (all p's > .05).

As a result of significant positive skewness the decision time data were log-transformed (i.e., log base 10). However, means are reported for back-transformed values and are displayed in Table 2 . It was predicted that intoxicated witnesses would take the longest time to reach their decisions, particularly when shown a TA lineup. The 3-way ANOVA examining alcohol condition, target presence, and accuracy revealed only a significant effect of accuracy, with accurate witnesses reaching their decisions faster ( M = 14.65, SD = 2.08) than inaccurate witnesses ( M = 20.01, SD = 2.00), F(1, 108) = 5.53, p = .02, ¿p 2 = .05. Neither the main effect of alcohol, F(2, 108) = 1.56, p = .21, ¿p 2 = .03, nor the main effect of target presence on confidence, F(1, 108) = 3.45, p = .07, ¿p 2 = .03, were significant. None of the interactions reached significance (all p's > .05).

Analyses were then conducted to explore the effect of choice alongside alcohol condition and accuracy on decision times. The main effect of accuracy was close to significance, F(1, 108) = 3.70, p = .057, ¿p 2 = .03, with accurate decisions ( M = 14.7, 95% CI [10.6, 18.7]) being made quicker than inaccurate decisions ( M = 20.1, 95% CI [16.6, 23.6]). Neither the main effect of alcohol, F(2, 108) = 0.97, p = .38, ¿p 2 = .02, nor the main effect of choice on decision time, F(1, 108) = 0.01, p = .92, ¿p 2 = .00, were significant. The interaction between accuracy and choice was also found to be significant. Whilst the decision times for correct and incorrect non-choosers were similar (correct non-chooser, M = 16.4 s, 95% CI [11.3, 21.6]; incorrect non-choosers, M = 16.5, 95% CI [10.6, 22.5]), accurate choosers produced faster decisions ( M = 12.9 seconds, 95% CI [6.6, 19.2]) than inaccurate choosers ( M = 23.8, 95% CI [20.1, 27.4]), F(1, 108) = 5.19, p =.02, ¿p 2 = .05. No other interactions were significant (all p's > 0.05).

According to AMT, alcohol intoxication should not impair identification performance when the target face is central to the scene witnessed. Our results are consistent with this prediction, and also with the findings of similar studies published previously ( Hagsand et al., 2013; Harvey et al., 2013; Yuille & Tollestrup, 1990 ). However, these null effects of alcohol on face identification are inconsistent with the results of Dysart et al. (2002) , who found higher levels of intoxication (in a real-world drinking scenario) to be associated with an increased likelihood of making false identifications in a TA showup task. It therefore remains possible that the extent of attentional narrowing under alcohol is dose dependent. At the low to moderate levels of intoxication obtained in the present experiment the scope of the inebriated witness's visual attention may be narrowed, but perhaps remains wide enough to permit the encoding of an entire target face. However, the visual attention of highly intoxicated witnesses, such as those included in Dysart et al's alcohol group, may be narrowed to the extent that only specific salient facial features can be processed, leading to poor identification performance, particularly under more challenging TA showup or lineup conditions. This hypothesis could be tested in future studies, by including a dose manipulation to alcohol-challenge lineup tasks such as the one presented here (as discussed below), and by introducing an additional recognition test for specific facial features. For the latter task, according to AMT, the face identification performance of highly intoxicated viewers should be impaired under TA test conditions, but their ability to identify specific, especially salient, facial features should be unimpaired. Sober controls, on the other hand, are not expected to show this dissociation. Until such studies are conducted we may only conclude that eyewitnesses intoxicated with small to moderate doses of alcohol are no less capable than sober counterparts at ascertaining a perpetrator's presence or absence from a lineup.

The present study is also limited in its use of only a single target to measure identification performance. Future research should employ more than one exemplar to reduce the possibility of target-specific results (see Wells & Windschitl, 1999 , for a discussion of this issue) and consider also the use of multiple lineups in improving identification decisions, as per the work of Sauerland, Stockmar, Sporer, and Broers (2013) .

We also note that the overall rate of correct identifications in the present study was somewhat low, raising the possibility that floor effects reduced the study's sensitivity to detect small alcohol effects. We point out, however, that our data are comparable to those reported by Hagsand et al. (2013) and other researchers examining eyewitness identification (e.g., Brewer et al., 2008 ), who used entirely different stimulus materials. Furthermore, the stimulus video and accompanying lineups employed in the present study have produced higher decision accuracy rates elsewhere ( Kneller et al., 2001).

We sought to examine how the confidence-accuracy relationship was influenced by intoxication, predicting that the strength of this relationship would be weaker for intoxicated witnesses, yet we observed only negligible differences in the level of decision confidence between sober and intoxicated groups. Thus, it appears that mild levels of intoxication cause no reduction in decision confidence scores compared to those reported by sober witnesses. This suggests that the testimony of a mildly intoxicated witness should be regarded as no less reliable than that of a sober witness displaying the same level of confidence in their identification decision.

Our results do reveal, however, that accurate witnesses had significantly more confidence in their identification decisions overall than inaccurate witnesses, regardless of their choosing status. This outcome supports the findings of Bothwell et al. (1987), and Sporer et al. (1995) , whose meta-analyses revealed a small but significant confidence-accuracy relationship. Additionally, and as predicted, we observed a significant relationship between accuracy and confidence for choosers also observed previously by Sporer et al. but, in the present study, this relationship was not mediated by alcohol intoxication.

It was predicted that the alcohol group would have the slowest reaction times of all three groups, particularly in the TA condition, as intoxicated participants may not have encoded enough details of the target face to quickly determine its absence from a lineup. However, intoxicated witnesses were found to be no slower at making an identification decision than control or placebo groups. Although, as mentioned previously, it is possible that the mild levels of intoxication achieved here may have been insufficient to impair the face memory representations of our alcohol group, via the narrowing of visual attention. Nevertheless, findings from previous research demonstrating a negative relationship between decision time and accuracy were replicated in the present study, with accurate decisions being made significantly more quickly than inaccurate decisions. Yet, despite this observation, the decision times of our accurate witnesses were considerably slower than the 10-12 second rule suggested by Dunning and Peretta (2002) . Nonetheless, 95% of accurate witnesses made their decisions within 18.75 seconds, which supports Sauerland and Sporer (2007) accuracy cut-off point of 18 seconds (using the same delay of one week between witnessing the crime and making an identification). Thus, we propose that moderately intoxicated witnesses making an accurate decision do so just as quickly as their sober counterparts.