We used apparent co-witness intoxication as a way to examine the effect of source credibility on the acceptance of misinformation from a co-witness.
We used apparent co-witness intoxication as a way to examine the effect of source credibility on the acceptance of misinformation from a co-witness.
Alongside an experimental confederate, individual participants (N = 100) watched a clip involving two simulated thefts. Immediately beforehand, half of the participants watched the confederate consume what appeared to be three alcoholic beverages. During a subsequent discussion with the participant, the confederate introduced two pieces of misinformation about the clip. In the absence of the confederate, participants were then interviewed before completing a target-absent line-up task.
As expected, misinformation impaired participants’ verbal reports, and misinformation about appearance impaired line-up performance. Overall susceptibility to misinformation was not significantly related to co-witness condition, or to participants’ ratings of the confederate's intoxication or ability to accurately complete the tasks. On individual items, however, co-witness condition appeared to exert some influence on misinformation acceptance if the participant's pre-misinformation response was discrepant with the misinformation, but not when it was ‘don't know’.
It is possible that effects of source credibility on misinformation acceptance may depend, at least to some extent, on the presence of a clear discrepancy between the misinformation and the witness's recollection.
Eyewitnesses to crime often assist investigators by recounting what they saw and, in many cases, by attempting to identify the perpetrator(s). Although eyewitness evidence is highly persuasive (Wells & Olson, 2003), it is by no means 100% reliable. In fact, it is susceptible to numerous external influences (see Davis & Loftus, 2007, for a review). One of these influences is discussion with another witness to the same crime. Numerous studies have demonstrated that misinformation from one witness can contaminate another's report, but less is known about how co-witness characteristics can influence the presence and strength of the misinformation effect. In this study, we used apparent co-witness intoxication as a way to examine some of the conditions that influence misinformation acceptance.
Discussions between eyewitnesses are common, and often occur with the goal of providing information (Paterson & Kemp, 2006). Although the exchange of correct information between co-witnesses does not necessarily threaten an investigation, problems arise when a witness conforms to a co-witness's incorrect recollection of an event (see Memon & Wright, 1999, for a discussion of this issue in relation to the Oklahoma bombing investigation).
Eyewitness memory conformity research has flourished over the past decade, with a variety of paradigms employed to examine this phenomenon. Some studies have used a turn-taking approach, in which two participants (or one participant and one confederate) take turns to answer the same question. Perhaps not surprisingly, the response given by the first witness can influence the second witness's response (e.g., Allan & Gabbert, 2008; Roediger, Meade, & Bergman, 2001; Schneider & Watkins, 1996, Exp. 2; Shaw, Garven, & Wood, 1997; Wright, Mathews, & Skagerberg, 2005; Wright, Self, & Justice, 2000, Exp. 1). Other studies have more closely mimicked the investigative process, demonstrating that misinformation from a co-witness can contaminate verbal testimony even when witnesses are tested individually (e.g., Gabbert, Memon, & Allan, 2003; Gabbert, Memon, Allan, & Wright, 2004; Gabbert, Memon, & Wright, 2006; Garry, French, Kinzett, & Mori, 2007; Wright, Self, & Justice, 2000, Exp. 2).
Recent research suggests that co-witness misinformation effects are not limited to verbal reports. Zajac and Henderson (2009) showed that co-witness misinformation about the perpetrator's appearance can impair participants’ photographic line-up performance. Specifically, when misinformed by a confederate that a thief had blue eyes, participants were twice as likely as non-misinformed participants to make an identification from a line-up comprised entirely of blue-eyed members, despite the thief's absence (Experiment 1). This effect was not evident when the line-up members’ eye colour was digitally altered from blue to brown (Experiment 2), and could not be explained by discussion alone (Experiment 3).
Given the relative ease with which eyewitness reports can be disrupted by co-witness discussion, an important question arises: which factors govern one witness's willingness to accept misinformation from another? Misinformation effects can be mediated by several variables, including witness factors (e.g., Sutherland & Hayne, 2001), the misinformation itself (e.g., Pezdek, Finger, & Hodge, 1997; Pezdek & Roe, 1997), and the source of the misinformation (e.g., Dodd & Bradshaw, 1980; Vornik, Sharman, & Garry, 2003). In this study, we focus on the latter.
Characteristics of the misinformation messenger (Vornik et al., 2003) can play an important role in determining the presence and strength of a misinformation effect (e.g., Ceci, Ross, & Toglia, 1987; Dodd & Bradshaw, 1980; Echterhoff, Hirst, & Hussy, 2005; French, Garry, & Mori, 2008; Lampinen & Smith, 1995; Smith & Ellsworth, 1987; Vornik et al., 2003). Dodd and Bradshaw (1980), for example, showed that witnesses to a simulated car accident were less likely to succumb to misleading questions supposedly prepared by a lawyer representing the driver (non-credible source), as opposed to an innocent bystander (credible source). Lampinen and Smith (1995) eliminated a misinformation effect when they described an adult misinformation source to preschoolers as ‘silly’. Smith and Ellsworth (1987) observed an increased misinformation effect when participants heard that their interviewer had seen the videotaped bank robbery several times, relative to not at all. Even source characteristics that are not explicitly conveyed to participants can influence misinformation acceptance. Vornik et al. (2003), for example, presented misinformation in either a New Zealand or a North American accent. Although accent per se did not influence misinformation acceptance, participants who rated the speaker high in social attractiveness and/or power were more vulnerable to misinformation.
Although the notion that the misinformation messenger will influence the effect of misinformation holds considerable face validity, our understanding of the way in which source credibility exerts its effect is limited. Lampinen and Smith (1995) proposed that source credibility can influence susceptibility to misinformation in two ways. First, witnesses who encounter misinformation from an unreliable source may be more likely to notice that the misinformation is discrepant from their own recollection (the discrepancy detection account), thereby reducing susceptibility to misinformation (Tousignant, Hall, & Loftus, 1986). Second, source credibility may influence the way in which detected discrepancies are resolved (the discrepancy resolution account). In other words, a witness who notices a discrepancy may turn to source credibility information in an attempt to decide which of two conflicting accounts to accept.
One limiting factor of these accounts is that they can only apply to situations where an explicit discrepancy exists between a witness's recollection and the misinformation. In reality, there are numerous instances in which eyewitnesses may not have encoded or retained the detail about which they are misinformed. In these situations, misinformation may ‘fill in the gaps’ in a witness's memory (McCloskey & Zaragoza, 1985). The discrepancy accounts discussed above would predict that source credibility would only exert an effect on misinformation acceptance when an original recollection existed and was discrepant with the misinformation.
We tested this hypothesis by examining the effect of source credibility on misinformation acceptance, both overall and as a function of original recollection. We used apparent alcohol intoxication as our source credibility manipulation, for three main reasons. First, the relation between alcohol and crime is well documented (US Department of Justice, 1998; UK Home Office, 2008); many crimes occur in situations where alcohol is likely to have been consumed by witnesses/victims as well as perpetrators (e.g., Ireland & Thommeny, 1993; Sim, Morgan, & Batchelor, 2005; Teece & Williams, 2000). Second, alcohol is known to impair eyewitness performance by narrowing attention (Read, Yuille, & Tollestrup, 1992), decreasing free recall (Yuille & Tollestrup, 1990), and increasing mistaken identification (Dysart, Lindsay, MacDonald, & Wicke, 2002; Yuille & Tollestrup, 1990). Finally, there appears to be a high expectation among adults that alcohol will impair eyewitness performance. In fact, even an alcohol placebo can increase susceptibility to misinformation (Assefi & Garry, 2003).
In the present study, individual participants watched a clip of two simulated thefts, alongside an experimental confederate. Immediately beforehand, half of the participants drank lemonade while they watched the confederate consume what appeared to be alcoholic beverages (intoxicated co-witness condition); for remaining participants, both the participant and the confederate drank lemonade (sober co-witness condition). During a subsequent discussion with the participant, the confederate introduced two pieces of misinformation about the clip. On the basis of previous studies using the same items (e.g., Zajac & Henderson, 2009), we knew that many participants would not have encoded these details from the clip. In the absence of the confederate, participants were then interviewed before viewing a target-absent line-up.
We had three major hypotheses. First, we expected an overall misinformation effect on participants’ verbal reports, but that this effect would be reduced or eliminated in the intoxicated co-witness condition, relative to the sober co-witness condition.
Second, on an item-by-item level, we predicted that source credibility would moderate the conformity effect, based on a person's initial belief. Specifically, we expected to see an effect of co-witness condition only when the participants’ pre-misinformation response and the misinformation were discrepant.
Finally, we expected that overall misinformation effects would extend to performance on the target-absent line-up task. That is, like Zajac and Henderson (2009), we expected that participants who were misinformed that the thief had blue eyes would be more likely than the other participants to mistakenly identify someone from our blue-eyed target-absent line-up. Again, however, we expected that this effect would be reduced or eliminated in the intoxicated co-witness condition, relative to the sober witness condition.
Our participants were undergraduate students from the University of Otago, New Zealand (N = 100; 37 male students, 63 female students; M age = 20.6 years, SD = 2.7, range = 18–35 years). Recruitment information for the experiment included a warning that participants may be required to consume enough alcohol to place them over the legal driving limit. All participants gave written consent to participate.
Individual participants took part in the experiment alongside an experimental confederate playing the role of a naïve participant. To facilitate participants’ belief that the experiment was investigating the effects of alcohol, the bench surfaces in the testing room were wiped down with an alcohol solution prior to each session, creating a distinct alcohol smell.
Half of the participants (n = 50) were randomly assigned to the sober co-witness condition. After reiterating that one or both participants might be required to consume alcohol as part of the experiment, the experimenter asked the participant to select one of two envelopes. Each contained a slip of paper saying ‘no alcohol’. The same process occurred for the confederate. The experimenter explained that both participants had been assigned to the control condition. To maintain the illusion of a controlled trial, the participant and confederate were each required to consume three 150 ml tumblers of lemonade within a 5-min period, and then wait for a period of 20 min.
Next, the pair viewed a 2½-min video, filmed in the University of Otago's Central Library (a highly familiar place to the majority of participants). The clip depicted a male and his female accomplice committing two thefts (see Zajac & Henderson, 2009, for more details). The film had no auditory component, and was displayed on a 15″ LCD computer monitor.
After the clip was played, the experimenter explained to the pair that he needed to print some more forms, but that they could discuss the film clip while he was away. The confederate engaged the participant in a discussion of the video, during which she misled the participant on two aspects of the clip, chosen from a list of four: the accomplice's eye colour (blue rather than brown), the gender of the student(s) seen directly behind the thieves when they entered the library (one male student rather than one female student), the colour of the male thief's jeans (black rather than blue), and the first item that was stolen (MP3 player rather than phone). The confederate casually elicited the participant's viewpoint on the item in question before misinformation was given. Participants were randomly assigned to misinformation items, with the provision that each combination of misinformation items occurred an equal number of times. No participant ever coincidentally mentioned a different critical item during the discussion. The experimenter was blind to which pieces of misinformation had been presented.
When the experimenter returned, the confederate was excused from the room on the pretence that participants would complete the subsequent tasks individually. After a short computer-based filler task, the participant answered a series of cued recall questions, four of which related to the critical items.
After another short filler task, the participant was presented with a six-member target-absent photographic line-up. The line-up photographs were 6″ × 4″ standard colour head and shoulders prints, shot against a plain background. All line-up members were similar in appearance to the female accomplice (i.e., female, medium build, 25–30 years, dark brown hair). Unlike the accomplice, however, all members clearly had blue eyes. The participant was told that the accomplice may or may not be in the line-up, and that it was acceptable to indicate that the accomplice was not there.
In this condition (n = 50), the group assignment process was rigged so that the confederate and participant were always assigned to the ‘alcohol’ and ‘no alcohol’ conditions respectively. The experimenter told the confederate that she would consume enough alcohol to place her over the legal driving limit.
To convince the participants that the confederate was drinking alcohol, we adopted procedures adapted from previous alcohol placebo research (Assefi & Garry, 2003; Rohsenow & Marlatt, 1981). The experimenter took the confederate into another room, ostensibly to weigh her. When they returned, the experimenter seemingly used the confederate's weight to calculate the required amount of alcohol. The experimenter then poured 30 ml of water from a vodka bottle into each of three 150 ml tumblers, filled each glass with lemonade, and asked the confederate to consume the drinks over a 5-min period. The participant was asked to drink three glasses of lemonade in the same fashion. The participant and confederate then waited for 20 min, ostensibly to allow the alcohol to take effect.
Next, the experimenter tested the confederate's breath alcohol level using a small hand-held breath alcohol meter. The breathalyser had been altered so that it always gave a ‘fail’ reading, equating to over 400 μg per litre of breath (i.e., over most legal driving limits). The experimenter explained that the alcohol had taken effect. The experimental procedure then continued as for the sober co-witness condition, beginning with the theft clip.
To avoid a confound between the confederate's supposed level of intoxication and her behaviour, the confederate ensured that her behaviour was consistent across conditions. In both the sober and intoxicated co-witness conditions, the confederate was always cheerful and friendly, but became a little more gregarious as the session progressed.
After the experimental tasks, a subset of participants (sober co-witness condition, n = 21; intoxicated witness condition, n = 34) answered a questionnaire comprising four questions to be answered on a 5-point scale. Two questions pertained to the participant's view of his/her own intoxication and that of the confederate (1 = not at all intoxicated, 5 = very intoxicated). The remaining questions pertained to the participant's view of his/her own ability to accurately perform the experimental tasks, and that of the confederate (1 = not at all able, 5 = very able).
Before full debriefing, participants were asked for their thoughts on the experimental aims, and whether they had noticed anything unusual during the procedure. No participants revealed suspicions about the nature of the experiment, the alcohol manipulation, or the identity of the confederate.
Preliminary analyses revealed no significant effects of gender, and gender did not enter into any interactions. Subsequently, the data were collapsed across gender.
With the exception of one participant in the sober co-witness condition (who gave himself a score of 2 due to an ‘extreme hangover’), all participants receiving the manipulation check (n = 55) rated themselves as ‘not at all intoxicated’ (a score of 1). Furthermore, those in the sober co-witness condition all rated the co-witness as ‘not at all intoxicated’. In the intoxicated co-witness condition, ratings of the co-witness's intoxication ranged from 1 (not at all intoxicated) to 4 (M = 2.53, SD = 0.71).
We conducted a 2 (rating: self; co-witness) × 2 (co-witness condition) analysis of variance (ANOVA) (repeated measures across rating) on participants’ judgements of their own and the co-witness's ability to accurately complete the experimental tasks. There was no main effect of rating, F(1,53) = .18, p = .67, f = .06,1 but there was a significant main effect of co-witness condition, F(1,53) = 15.94, p < .01, f = .54, which was qualified by a significant Rating × Co-witness condition interaction, F(1,53) = 16.96, p < .01, f = .57 (see Figure 1). In the sober co-witness condition, participants’ ratings of their own ability were lower than their ratings of the co-witness, t(20) = −3.51, p < .01, d = .77. Over one third (38%) of participants in this condition rated the co-witness as more able than themselves to accurately perform the experimental tasks; none rated her as less able. In the intoxicated co-witness condition, the opposite pattern was observed, t(33) = 3.19, p < .01, d = .55. Over one half (53%) of participants rated the co-witness as less able, and only 15% rated her as more able. Participants’ ratings of their own ability did not differ significantly across co-witness condition, t(53) = 1.03, p = .31, d = .20.
Before providing each item of misinformation, the confederate elicited the participant's view on that item. Participants’ responses can be seen in Table 1. Responses differed as a function of critical item, with eye colour the least likely item to be recalled accurately (4%), and most likely to elicit a ‘don't know’ response (92%). Response patterns for the remaining three critical items were similar (response accuracy: 27% for item stolen first, 29% for bystander, and 22% for trousers; ‘don't know’ responses: 63% for item stolen first, 55% for bystander, and 70% for trousers).
|Critical item||Report to the co-witness|
Next, we examined the overall effect of the misinformation on participants’ reports to the experimenter. Because our interview questions did not force participants to choose between two or more options, we could do this in two ways: (1) by assessing whether or not misinformation was reported; and (2) by assessing the accuracy of participants’ responses. Because our data were count data with a restricted range, it was inappropriate to use ANOVA. Instead, we conducted Poisson log linear regression analyses to address our research questions.
We first compared the number of times that participants reported misinformation in response to questions about the two misled items and the two control items. Regardless of co-witness condition, participants were more likely to report the misinformation on misled items (M = 0.86 of a possible 2, SE = .08), relative to control items (M = 0.15 of a possible 2, SE = .04), Wald χ2 (df = 1) = 45.13, p < .01. Neither the main effect of co-witness condition, Wald χ2 (df = 1) = 1.44, p = .23, nor the interaction term, Wald χ2 (df = 1) = 2.65, p = .10, reached significance. The number of pieces of misinformation reported was not significantly related to participants’ ratings of co-witness's intoxication, r = −.16, p = .25, or to participants’ ratings of the co-witness's ability to complete the tasks, r = .03, p = .80.
We also examined the effect of misinformation on participants’ accuracy. As in the analysis described above, regardless of co-witness condition, participants’ accuracy on the misled items (M = 0.56 of a possible 2, SE = .06) was significantly lower than their accuracy on the control items (M = 0.79 of a possible 2, SE = .07), Wald χ2 (df = 1) = 6.29, p < .05. Again, there was no significant main effect of co-witness condition, Wald χ2 (df = 1) = 0.36, p = .55, and no significant interaction, Wald χ2 (df = 1) = 0.43, p = .51. Participants’ accuracy on the critical items was not significantly related to their ratings of co-witness's intoxication, r = −.07, p = .60, or to participants’ ratings of the co-witness's ability to complete the tasks, r = .20, p = .15.
Next, we examined how pre-misinformation response might mediate the effect of co-witness condition on acceptance of misinformation (see Table 2 for data). When a pre-misinformation response was not provided (i.e., when a ‘don't know’ response was given), co-witness condition did not appear to play a role in whether or not the misinformation was reported to the experimenter (sober co-witness condition: 45%; intoxicated co-witness condition: 43%). In contrast, in situations where a pre-misinformation response was clearly discrepant with the misinformation (either by being accurate, or by being inaccurate but discrepant with the misinformation), more items of misinformation were incorporated into participants’ reports when the misinformation had come from a ‘sober’ co-witness (64%) as opposed to an ‘intoxicated’ one (32%).
|Pre-misinformation response||Misinformation items accepted||Misinformation items rejected|
|Discrepant with misinformation|
|Consistent with misinformation|
Because these data violated the independence assumptions of a chi-square test, we assessed our hypothesis by considering only the first piece of misinformation encountered by each participant. Despite the resultant decrease in statistical power, there was a marginally significant effect of co-witness condition on misinformation acceptance when pre-misinformation response was discrepant with the misinformation, χ2 (1, N = 20) = 3.33, Fisher's exact p = .09, w = .41; participants were more likely to accept misinformation in the sober co-witness condition than in the intoxicated co-witness condition. When participants were unable to provide a pre-misinformation response, co-witness condition did not exert a significant effect on misinformation acceptance, χ2 (1, N = 77) = 0.02, p = .89, w = .02.
The accomplice in the film clip had brown eyes. Consistent with the eye colour misinformation that some participants encountered, however, all members of the target-absent line-up had blue eyes. First, we ascertained that the eye colour reported to the experimenter did not significantly influence participants’ line-up decisions, χ2 (1, N = 100) = 1.58, p = .21, w = .12. That is, those who told the experimenter that the accomplice had blue eyes were not significantly more likely to make a selection from the blue-eyed line-up than those who did not. Next, we used logistic regression (forward conditional method) to examine line-up accuracy as a function of eye colour misinformation, co-witness condition, and the interaction between these two variables. The overall model was significant, χ2 (df = 2) = 26.40, p < .01, Nagelkerke R2 = .31, and two independent predictors emerged. The first was eye colour misinformation, β = 1.62, SEβ = .47, Wald χ2 = 11.82, p < .01, OR = 5.08. Participants who were misinformed that the accomplice's eyes were blue were twice as likely (72%) to make an identification from the line-up than those who were not (36%). The second predictor was co-witness intoxication, β = 1.62, SEβ = .47, Wald χ2 = 11.82, p < .01, OR = .20. Participants in the intoxicated co-witness condition were over twice as likely to (correctly) reject the line-up (64%) as those in the sober co-witness condition (28%).
We examined the role that source credibility played in the acceptance of co-witness misinformation, by asking whether, and in what circumstances, participants would accept misinformation from a co-witness who appeared to have been consuming alcohol. Before turning our attention to the effects of source credibility, we first consider the basic effects of misinformation on participants’ verbal reports and line-up performance.
As expected based on numerous co-witness discussion studies (e.g., Allan & Gabbert, 2008; French et al., 2008; Gabbert et al., 2003, 2006; Garry et al., 2007), misinformation exerted a significant negative impact on participants’ reports of what they saw. Although the restriction of available response options has considerable utility in informing theory (e.g., McCloskey & Zaragoza, 1985), we diverged from this paradigm because it does not mimic a police interview. However, regardless of the fact that our participants’ responses were not restricted in any way, we still saw a significant effect of misinformation – not only on participants’ propensity to report the misinformation but also on their accuracy.
The absence of the misinformation messenger during testing allows us to tentatively discount at least one mechanism for this misinformation effect: that participants reported incorrect details with the knowledge that they were incorrect. Although normative influence can play a role in participants’ responses if the misinformation messenger is present during testing (Bond, 2005; Deutsch & Gerard, 1955), our participants were tested in the absence of the confederate, making normative influence unlikely. Instead, it appears that participants who reported misinformation to the experimenter accepted this misinformation as a genuine account of the event. The mechanism by which this occurs has been the subject of a lively debate (Ayers & Reder, 1998; Loftus, 2005), and is not the focus of this study. However, given the large number of participants who did not offer a pre-misinformation response, it is appropriate to suggest that, in the majority of cases, the misinformation ‘filled in the gaps’ in participants’ recollections (McCloskey & Zaragoza, 1985), as opposed to replacing (Loftus, 1979; Loftus & Loftus, 1980) or co-existing with (Berkerian & Bowers, 1983; Christiaansen & Ochalek, 1983) an accurate memory representation of the critical detail.
As in Zajac and Henderson (2009), participants who were told that the accomplice had blue eyes were considerably more likely to make a selection from the blue-eyed target-absent line-up than those who were not. In fact, eye colour misinformation was a significant independent predictor of line-up accuracy. Although the precise mechanism involved in this effect is not yet known, we can rule out some possibilities. First, because all participants engaged in discussion and received misinformation, our findings cannot be due to these aspects of the paradigm alone. Second, because our data suggest that line-up decisions were unrelated to the eye colour reported to the experimenter, we cannot claim a ‘freezing effect’ (Loftus, 1979) whereby participants perceived pressure to remain consistent with their initial verbal reports. Third, because an increase in identifications does not seem to occur when the line-up members’ eye colour is inconsistent with the misinformation (Zajac & Henderson, 2009, Experiment 2), the line-up misinformation effect could be the result of convergent misinformation from the co-witness (i.e., being told that the perpetrator had blue eyes) and from the line-up (i.e., being presented with a blue-eyed line-up). We are currently attempting to clarify these issues.
To our surprise, participants in the sober and intoxicated co-witness conditions were similarly likely to succumb to misinformation. Although our immediate inference was that we failed to convince participants that the confederate was intoxicated, a manipulation check conducted with a subset of participants did not support this hypothesis. In the intoxicated co-witness condition, most participants who completed the misinformation check did not think that the confederate was sober, and thought that the confederate's ability on the experimental tasks would have been lower than their own. Furthermore, vulnerability to misinformation was not significantly related to ratings of intoxication or ability.
Why then, did participants accept misinformation from the ‘intoxicated’ co-witness? We know that a disconnect can exist between risk assessment and behaviour, even in situations where risk-taking has serious ramifications for one's own safety (e.g., unsafe sexual practices: Maswanya et al., 1999; risky driving: Rundmo & Iversen, 2004; and workers’ asbestos-safety behaviours: Stewart-Taylor & Cherrie, 1998). Having said this, our findings go against those of other studies demonstrating an effect of source credibility on the misinformation effect (e.g., Dodd & Bradshaw, 1980; French et al., 2008; Lampinen & Smith, 1995; Smith & Ellsworth, 1987; Vornik et al., 2003).
We need to point out that our failure to find an interaction between co-witness condition and the effect of misinformation could have been due to inadequate statistical power. On the other hand, previous misinformation studies exploring source characteristics have found effects of a size that we should have been able to detect (e.g., French et al., 2008; Smith & Ellsworth, 1987). We therefore need to consider how our study might differ from those studies.
One possibility is that poor encoding of our critical items averted the processes normally involved in source credibility effects. Lampinen and Smith (1995) suggest that social variables such as source credibility may alter the probability that a participant detects a discrepancy between his/her account and that of the co-witness, or the way in which a discrepancy is resolved once detected. In this framework, however, source credibility should only exert an influence in situations where a discrepancy between witness response and co-witness response is present to be detected.
We found preliminary support for this notion. When there was a clear discrepancy between the pre-misinformation response and the misinformation, misinformation was more likely to be accepted from the ‘sober’ co-witness than from the ‘intoxicated’ one. When the pre-misinformation response was ‘don't know’ (i.e., where there was no discrepancy), this effect was not observed. It seems, then, that source credibility characteristics might only come into play – or might be more likely to come into play – when critical items are encoded. Clearly this possibility warrants further investigation.
Finally, it is of note that co-witness condition emerged as a significant independent predictor of line-up performance. Specifically, participants in the intoxicated co-witness condition were over twice as likely to (correctly) reject the line-up than those in the sober co-witness condition. Although there are a number of possible explanations for this finding (e.g., a change in perceived demand characteristics of the line-up as a function of perception of a co-witness), replication of this effect would be crucial before speculation could occur.
All of the issues examined in this study are highly relevant to real-life eyewitness situations. We know, for example, that numerous crimes occur in situations where witnesses have been consuming alcohol (Sim et al., 2005; Teece & Williams, 2000). We also know that co-witness discussions regularly occur (Paterson & Kemp, 2006), and often concern details about the perpetrator (Skagerberg & Wright, 2008). It is clear that the consumption of alcohol can selectively impair recall for peripheral details of a witnessed event, like those studied here (Read et al., 1992). Finally, we know that the construction of a line-up should be driven by verbal descriptions given by witnesses (Clark & Tunnicliff, 2001; Wells et al., 1998). Taken together, these research findings suggest that the concerns raised by this study are valid and could easily translate to real-life eyewitness situations.
Although eyewitness testimony is a highly influential component of criminal investigations, the present findings provide further evidence of the malleability – and therefore fallibility – of this form of evidence. Although our data are preliminary, they add to a growing body of research showing that co-witness misinformation can impair both verbal and identification evidence, and raise important questions about how eyewitnesses evaluate a co-witness's ability to provide reliable information.
For all statistical comparisons involving ANOVAs, t-tests and chi-square tests, Cohen's (1988) measures of effect size (f, d, and w respectively) have been included.
This study was funded by the Marsden Fund Council, from Government funding administered by the Royal Society of New Zealand. The authors are especially grateful to Hannah Moss, our tireless experimental confederate, and to Scott Miller and Matt Healey for statistical advice.