CO-WITNESSES AND THE EFFECTS OF

DISCUSSION ON EYEWITNESS MEMORY

Helen M. Paterson

A thesis submitted in fulfilment

of the requirements for the degree of

Doctor of Philosophy

School of Psychology University of New South Wales

December 15, 2004

i

Abstract The research presented in this thesis was designed to investigate the effects of co-witness information on the accuracy and completeness of eyewitness memory. Co-witness information is defined as information that one eyewitness conveys to another about an event that they both observed. Very little research has focused on co-witness discussion, so the first two studies surveyed real eyewitnesses and police officers to determine how often witnesses discuss the event with one another. The results from these surveys suggested that co-witnesses commonly talk about the event with each other and this outcome provided a clear justification for studying the effects of co-witness discussion on memory. Previous research on co-witness discussion has reached inconsistent conclusions, and the possibility exists that these discrepancies are due to methodological differences. Therefore, this research aimed to determine whether co-witness discussion helps or hinders individual recall, and to investigate this within a closely defined methodological set. In a series of five experiments, participants were shown a crime video and then asked to discuss the video in groups (some of which received experimentally induced misinformation from a co- witness). Following the discussion, participants were asked to give their individual accounts of what happened. These experiments showed that exposure to postevent information from a co-witness can cause people to incorporate this information into their individual testimonies, regardless of the accuracy of the information. This phenomenon has become known as ‘memory conformity.’ Relevant theories were tested in order to contribute to knowledge regarding the causes of memory conformity. Furthermore, the experiments also aimed to establish whether it is possible to mediate any negative effects of co-witness discussion by employing our theoretical understanding of the causes of memory conformity. Five approaches were utilized in an attempt to reduce the negative effects of co-witness discussion: warnings about possible misinformation, source monitoring, free recall, confidence ratings, and ‘remember/know judgments’ (Tulving, 1985). Some evidence was found to suggest that when using ‘remember/know judgments’ it may be possible to distinguish ‘real’ memories from information obtained from a co-witness. These results are discussed in terms of theoretical and practical implications.

Discussion and eyewitness memory ii

Table of Contents

ABSTRACT I TABLE OF CONTENTS II ACKNOWLEDGEMENTS IV OVERVIEW OF THESIS V

SECTION 1: LITERATURE REVIEW 1

CHAPTER 1 2 LITERATURE REVIEW OF RELEVANT RESEARCH 2 DIRECT TRANSFER OF CO-WITNESS INFORMATION 5 INDIRECT TRANSFER OF CO-WITNESS INFORMATION 23 SUMMARY OF CHAPTER 1 26 CHAPTER 2 28 THEORETICAL EXPLANATIONS OF MEMORY CONFORMITY 28 WHY DOES MEMORY CONFORMITY OCCUR? 28 EVALUATING THE THEORIES OF MEMORY CONFORMITY 42 FACTORS THAT INFLUENCE CONFORMITY 50 SUMMARY OF CHAPTER 2 55

SECTION 2: SURVEY STUDIES 56

CHAPTER 3 57 STUDY 1: WITNESSES TALK 57 METHOD 59 RESULTS 61 DISCUSSION 68 CHAPTER 4 72 STUDY 2: THE LEGAL PERSPECTIVE 72 METHOD 74 RESULTS 77 DISCUSSION 89

SECTION 3: EXPERIMENTAL STUDIES 95

CHAPTER 5 97 STUDY 3: CO-WITNESS CONTAMINATION 97 METHOD 101 RESULTS 105 DISCUSSION 115 CHAPTER 6 121 Discussion and eyewitness memory iii

STUDY 4: WAYS OF ENCOUNTERING POSTEVENT INFORMATION 121 METHOD 127 RESULTS 132 DISCUSSION 150 CHAPTER 7 159 STUDY 5: UNINTENTIONAL HEARSAY 159 METHOD 166 RESULTS 174 DISCUSSION 190 CHAPTER 8 197 STUDY 6: SOURCE MONITORING 197 METHOD 199 RESULTS 201 DISCUSSION 213 CHAPTER 9 218 STUDY 7: EYEWITNESS INTERVIEW 218 METHOD 222 RESULTS 228 DISCUSSION 234

SECTION 4: GENERAL DISCUSSION 239

CHAPTER 10 240 GENERAL DISCUSSION 240 OVERVIEW AND SUMMARY OF SIGNIFICANT FINDINGS 240 IMPLICATIONS OF THE RESEARCH 252 LIMITATIONS OF THE RESEARCH AND FUTURE DIRECTIONS 257 CONCLUSION 261

REFERENCES 263

APPENDICES 279

Discussion and eyewitness memory iv

Acknowledgements

To my supervisor, Richard Kemp, for being such a wonderful mentor and friend. I couldn’t have asked for a better supervisor. You gave me support and positive feedback when you knew I needed it and constructive criticism when it was deserved. In addition to learning a great deal from you, I have enjoyed our ‘important’ lab meetings in donut factories, caves and aboard sailboats!

I would like to thank Peter Lovibond and the psychology department at UNSW for providing a positive and enriching environment in which to work. I am thankful to Jane Goodman-Delahunty and Kevin McConkey for their reviews of my thesis proposal and mid-candidature report. I appreciate the time and effort they put into my reviews and I am grateful for their suggestions. I would like to give a special thanks to Kevin McConkey for giving me advice on my funding. I am also grateful to my initial supervisor, Joseph Forgas, for helping me devise ideas for my thesis and design Study 3.

I would also like to thank the volunteers who helped with my research. In particular, I would like to thank the police officers who completed my survey, the students who volunteered for my studies, the ‘actors’ who starred in my ‘uber-crap’ crime videos, and Denis Beaubois for producing the videos. Furthermore, I would like to acknowledge the labbies who helped run my experiments, code data, and assist with data entry. I would like to give a special thanks to Joanne Bower and Alice Walsh for proofreading my thesis.

To my parents. Thank you for being so supportive of my decision to travel and study abroad in Australia.

To Rachel, Linda, and Deborah who were (apart from being great friends) people with whom I could talk about my thesis and enjoy post-grad coffees.

And finally, to my good friends…without you, I would have finished my PhD earlier, but I wouldn’t have had so much fun. Thanks for the constant distractions! ☺ Discussion and eyewitness memory v

Overview of Thesis

The present research was designed to investigate the effects of co-witness information on the accuracy and completeness of eyewitness memory. Co-witness information is defined as "information that one eyewitness might pass to another eyewitness regarding an event that they both observed," (Luus & Wells, 1994, p. 714). Whereas the legal system assumes that the testimony given by eyewitnesses should be independent of one another (Levine & Tapp, 1973), this is frequently not the case. “Discussion among victims or witnesses to a crime is difficult, if not impossible to prevent” (Yarmey, 1992, p. 252). Crimes are often ‘remarkable’ in the sense of the event's uniqueness and also in the sense that witnesses repeatedly ‘remark’ on the event to others (Yuille & Daylen, 1998). Because eyewitness information is often conveyed from one witness to another through discussion, it is important to ascertain the effects of co-witness information on the validity of eyewitness testimony. To address this aim, seven studies were conducted as part of a four-phase approach, with each phase designed to inform the next. This thesis was divided into four sections, each describing one phase of the research program.

Section 1: Literature Review Section 1 involved a thorough review of the scientific literature regarding the effects of co-witness information on memory. The literature review was divided into two main sections, which focused on: (1) experimental methodologies employed to investigate the effects of discussion on memory and (2) theoretical explanations of the findings. This literature review informed the design of studies conducted in the subsequent phases of this research program.

Section 2: Surveys The aim of Section 2 was to determine how common co-witness discussion is and to investigate the opinions held by legal authorities regarding discussion between witnesses. To examine these issues, two survey studies were conducted: one of real eyewitnesses and Discussion and eyewitness memory vi

the other of police officers. The outcome of Section 2 provided a clear justification for the subsequent experimental study of co-witness discussion.

Section 3: Experimental Studies The aim of Section 3 was to determine whether co-witness discussion helps or hinders individual recall and to investigate this within a closely defined methodological set. Previous research on eyewitness discussion has reached inconsistent conclusions, and the possibility exists that these discrepancies are due to methodological differences. In Section 3, experiments are described which examined these different methodologies and investigated the impact of co-witness discussion on eyewitness memory. A final aim of Section 3 was to investigate the effects of co-witness discussion within a theoretical framework. Relevant theories were tested in order to contribute to knowledge regarding the causes of memory conformity. Furthermore, these experiments aimed to establish whether it is possible to mediate any negative effects of co-witness discussion by employing our theoretical understanding and empirical outcomes.

Section 4: General Discussion The aim of Section 4 was to summarise significant findings from the research presented in this thesis and describe how the findings contribute to our understanding of the effects of co-witness discussion on memory. Practical and methodological implications of the research are also described in the section. Finally, limitations of the research, as well as possible future directions for the research, are discussed.

1

SECTION 1: LITERATURE REVIEW

Section 1 Aims

• To describe and evaluate different methodologies that have been used to investigate the effects of discussion on memory.

• To describe and evaluate different theories that have been used to explain the effects of co-witness discussion on memory.

Section 1 involves a thorough review of the scientific literature concerning the effects of co-witness information on memory. The literature review is divided into two chapters. Chapter 1 describes different methodologies that have been used to investigate the effects of group discussion on memory. In particular, Chapter 1 summarises the relevant research, highlighting inconsistent findings in the literature and areas requiring further investigation. Chapter 2 reviews existing social and cognitive theories that may be used to explain the effects of co-witness discussion on memory. 2

Chapter 1 Literature Review of Relevant Research

Approximately 77,000 individuals are arrested in the United States each year based primarily on eyewitness testimony (Goldstein, Chance, & Schneller, 1989). Eyewitness testimony is commonly used as evidence in courts of law and is also one of the most persuasive forms of evidence for juries (Wells, Lindsay, & Ferguson, 1979; Wells et al., 1998). Given the pivotal role that eyewitness testimony plays in some trials, it is important to establish whether or not the jury’s faith in this testimony is warranted. Many social scientists argue that eyewitness memory is not as trustworthy as the layperson may believe. One study has shown that eyewitness errors are the most common cause of false convictions (Huff, Rattner, & Sagarin, 1996). Almost all innocent individuals exonerated by DNA evidence had been convicted primarily as a result of erroneous eyewitness evidence (Wells et al., 1998). Consequently, a great deal of research has focused on the unreliability of eyewitness testimony. In a series of influential studies, Loftus (e.g., 1975, 1977, 1979a, 1992) has shown that exposure to incorrect information about an event after it has occurred (i.e., postevent misinformation) often causes people to incorporate this misinformation into their memories. This phenomenon has become known as the misinformation effect and there is extensive laboratory evidence for its existence (e.g., Belli, 1989; Ceci, Ross, & Toglia, 1987; Lindsay, 1990; Loftus & Greene, 1980; Loftus, Miller, & Burns, 1978; Tversky & Tuchin, 1989). Wright and Davies (1999) suggest that for a witness to a real crime, there are three ways in which postevent misinformation may be encountered: (a) being asked questions about the event by police officers, lawyers, friends, or others, (b) media coverage of the event (e.g., reading about the event in the newspaper), and (c) through co-witness information. A great deal of research has focused on the first of these situations: the effect of questions containing misleading information (e.g., Loftus, Miller, & Burns, 1978; Loftus & Palmer, 1974; Loftus & Zanni, 1975). For example, in one study, participants saw pictures of an accident, showing a stop sign at an intersection (Loftus, 1978). Half of the Chapter 1: Literature Review 3

participants were subsequently asked a leading question which suggested that there was a yield (give way) sign at the intersection, while the other half were not presented with such misinformation. When participants were later tested on their memories for the event, those exposed to the false information were more likely to report that the yield sign was, in fact, part of the original accident scene than those who had not received the misleading information. The second way that postevent misinformation may be encountered is through media reports. One study investigated this phenomenon by showing participants pictures of a shoplifting incident (Wright & Stroud, 1998). The participants then read a brief summary of the crime, which included some incorrect details. Just as the participants in the Loftus (1978) study incorporated the misleading details into their memories for the event, participants in this study incorporated the incorrect details from the summary into their memories of the incident. Other research using postevent narratives has found similar findings (e.g., Allen & Lindsay, 1998; Belli, Lindsay, Gales & McCarthy, 1994; Searcy, Bartlett, & Memon, 2000; Shaw, Garcia, & Robles, 1997; Sporer, 1995). The current thesis examines the third way that postevent misinformation may be encountered: through other witnesses. This area has been surprisingly neglected until recently, as the majority of the literature on eyewitness memory has focused on the effect of questions and media reports containing misleading information. Co-witness information is defined as "information that one eyewitness might pass to another eyewitness regarding an event that they both observed," (Luus & Wells, 1994, p. 714). There are various ways in which co-witness information may be passed from one eyewitness to another. Co-witness information may be transferred in ‘direct’ conversation between witnesses or ‘indirectly’ through a third party, such as a police officer, who informs one witness about the evidence given by another (Luus & Wells, 1994). Furthermore, there are two ways in which direct conversation between witnesses may lead to the transfer of co-witness information (e.g., Yarmey, 1992). Witnesses may either discuss the event prior to giving individual statements, or they may collaborate to provide a joint statement (see Figure 1.1 for an illustration of the types of co-witness information).

Chapter 1: Literature Review 4

Co-Witnesses Information

Direct Transfer Indirect (Co-witness Transfer discussion)

Discussion prior to Collaboration Individual in Statement Statement

Figure 1.1. Types of co-witness information.

Investigating the effects of co-witness information is important because once witnesses discuss an event, their testimony should not be treated as independent observations. For example, if four witnesses who have discussed a crime each report the same detail, this should not be regarded as any more informative than if just one witness reports that detail (Wright & Davies, 1999). Bearing in mind the clear importance of studying co-witness information, it is surprising how little research has been conducted in this area. Moreover, the little research that has been done has reached inconsistent conclusions. Conclusions have varied from support for allowing witnesses to discuss the event with one another (e.g., Warnick & Sanders, 1980) to caution against allowing discussion (e.g., Hollin & Clifford 1983; Shaw et al., 1997). Warnick and Sanders (1980) are among the researchers who support the practice of allowing witnesses to discuss the crime with one another. Based on their research, they concluded that, "Group discussion prior to the delivery of eyewitness testimony is probably an advantageous procedure" (p. 257). Contrary to this conclusion, other researchers have reported that the detrimental effects of co-witness information are larger and more robust Chapter 1: Literature Review 5

than those of leading questioning (Shaw et al., 1997). Hollin and Clifford (1983, p. 242) also contended, “it would be preferable to try to obtain uncontaminated individual testimony from each witness.” In a recent survey of 64 eyewitness experts, some experts claimed to have testified in court on the issue of co-witnesses (Kassin, Tubb, Hosch, & Memon, 2001). This is potentially troublesome given how little agreement there is in the literature. Courts require consistency among a body of science before expert testimony can be given (e.g., Daubert v. Merell Dow Pharmaceuticals, Inc., 1993; Frye v. United States, 1923). For example, in the United States expert evidence “must be sufficiently established to have gained general acceptance” to be permissible in court (Frye v. United States, 1923, p. 1014). Such general acceptance does not currently exist in the literature on the topic of co-witness discussion. Because the conclusions drawn from past literature are so inconsistent, it is important that co-witness information be studied in more detail. Recently, it has been argued, “It appears that co-witness information can lead to false reports, but perhaps only under certain conditions. What are these conditions? …Future research needs to explore these possibilities” (Garven, Wood, & Malpass, 2000, p. 46-46). Yarmey and Morris (1998) suggest that, “The capricious results among these investigations are probably due to methodological differences and variability in subject matter” (p. 1638). To appreciate the effects of co-witness information on eyewitness reports, we must examine, in detail, the different methodologies that have been used to investigate this topic. When examining the literature on co-witness information, this literature review focuses first on the direct transfer of co-witness information (i.e., through discussion) followed by the indirect transfer of co- witness information (i.e., through a third party).

Direct Transfer of Co-witness Information As previously mentioned, there are two ways that co-witness information can be transferred directly: (1) through joint collaboration in statement and (2) through discussion prior to individual statement. These two methods are examined separately.

Chapter 1: Literature Review 6

Collaboration in Statement Although collaboration in preparation of a statement is briefly discussed in this section, the main focus of the thesis is discussion prior to individual statement, as this is a more common occurrence. As Hollin states, “In real life several witnesses would not be asked to arrive at a consensus decision” (1980, cited in Hollin & Clifford, 1983, p. 235). Until recently, research on human memory has focused almost exclusively on the individual participant. Some researchers (e.g., Hyman, 1994) argue that much of this literature has limited ecological validity as it has relied heavily on the simple reproduction of memories in an unrealistic, experimental setting. Such an environment disregards the expansive and integrative aspects of memory. For example, it disregards the constructive nature of remembering as well as the incorporation of information that people typically have available to them about at event (Hyman, 1994). Traditionally, researchers in memory have aimed to keep procedures free from contamination, such as other people’s memories (Schneider & Watkins, 1996). However, such a narrow focus may not fully explain how people remember (Meudell, Hitch, & Kirby, 1992). Because such ‘contamination’ is common to memory, understanding its effects enables greater knowledge regarding memory itself (Schneider & Watkins). “Everyday remembering often has a more social interpersonal flavour, particularly when two (or more) people jointly recall and discuss an event they have experienced or witnessed” (Clark, Stephensen, & Kniveton, 1990). Therefore, instead of intentionally avoiding the social aspects of memory, they should be explored in their own right. Some of the first research suggesting that human memory has a fundamentally social nature was conducted over 70 years ago (Bartlett, 1932). In his classic text, Remembering, Bartlett wrote, “Both the manner and the matter of recall are often predominantly determined by social influences” (p. 244). In 1987, Edwards and Middleton re-examined Barlett's studies and argued that in his book, Bartlett often referred to the importance of conversation in memory without studying it directly, or explicitly stating its importance. In reference to Bartlett's work, Edwards and Middleton argued that, "conversational discourse, in which people draw upon and share with each other their past experiences, provides exactly the sort of context in 'daily life' in which everyday remembering can best be studied” (p.77). Chapter 1: Literature Review 7

Bartlett (1932) also argued that accuracy of memory is not a primary goal in conversation. He supplied convincing evidence that when people attempt to recall stories, both integrative and reconstructive changes often occur. As he stated: We discuss with other people what we have seen, in order that we may value or criticize, or compare our impressions with theirs. There is ordinarily no directed and laborious effort to secure accuracy. We mingle interpretation with description, interpolate things not originally present, transform without effort and without knowledge (p. 96). According to Bartlett, the role of memory in a “social context” is to communicate and convey significance, rather than to accurately represent experience (Edwards & Middleton, 1987). With the exception of Bartlett (1932), research did little to address the social aspects of memory until Perlmutter (1953) investigated the effects of collaboration on memory. Following Bartlett’s tradition, participants were told the story, "The War of the Ghosts." Participants were then asked to recall the story 15 minutes after hearing it and again 24 hours later, either individually or in groups of two or three. It was found that collaborative groups recalled more than individuals, although the differences in recall were not significant. A similar study was conducted in which participants were told "The War of The Ghosts" story and asked to recall the story either independently or in pairs, immediately after hearing it or one week later (Stephenson, Brandstatter, & Wagner, 1983). Stephenson et al. found that collaboration increased confidence, completeness, and accuracy. As the authors concluded, “Social recall is an improvement on individual performance” (p. 188). However, while collaborative groups were more confident in their correct responses than individuals, they were also more confident in their incorrect responses. Furthermore, it was also found that that collaboration increased errors that do not contradict the original story, but were nonetheless incorrect. Another methodology used for studying collaborative memory is to test recall for word lists or nonsense syllables. For example, Hoppe (1962) used nonsense syllables as a stimulus to investigate the effects of group collaboration on memory. Participants were shown eight nonsense words and asked to recall them either on their own or with two other Chapter 1: Literature Review 8

participants. Findings indicated that groups who collaborated to recall the word list remembered more than individuals working alone. Groups and individuals did not differ in the mean number of intrusions (i.e., incorrect responses) they produced. While the above studies provide us with valuable information regarding the social aspects of memory, caution needs to be exercised before applying these results to the judicial arena. One should not assume the results obtained from studies using stories and word lists as stimuli can be generalized to forensic contexts. It seems an ill-advised procedure to study the social aspect of memory using word lists and nonsense syllables as they have low ecological validity. To assess whether these results can be generalized, more applied forensic research investigating the social aspects of memory has been conducted. Alper, Buckhout, Chern, Harwood, and Slomovits (1976) were the first researchers to investigate the effects of memory collaboration in a forensic setting. In their experiment, a theft was staged in front of a class, and the students’ memories for the crime were tested. First, the participants were tested individually, and then collaboratively in groups. It was found that while groups had fewer errors of omission than individuals, errors of commission increased. That is, the groups tended to have more complete descriptions of the event, but they also included more fabricated details. However, the within-subjects design meant that individual and group data were confounded. That is, the differences found between individuals and groups could simply be due to the participants giving their reports for a second time, a demand characteristic typical of repeated-measures designs (Warnick & Sanders, 1980), or the time that has elapsed since viewing the incident (Hollin, 1980, cited in Hollin & Clifford, 1983). Warnick and Sanders (1980) attempted to address these problems by following the same general procedure as Alper et al. (1976), with the exception that they used a between- subjects design. Participants viewed a crime video and then were asked, either individually or in groups of three or four, to recall the event. Some of the groups were asked to reach a consensus decision, while others were not. Group discussion, regardless of whether they were asked to reach a consensus decision, led to significantly more accurate accounts than those produced by the average individual working alone. Groups were superior to individuals with respect to the completeness of their reports and the accuracy of their estimations of the duration of the incident. Groups also showed a non-significant trend Chapter 1: Literature Review 9

towards fewer errors of commission than individuals. Although participants' ability to accurately recognize the perpetrator in a lineup did not differ as a function of individuals versus group discussion, there was a non-significant trend in favour of groups. Two studies similar to one another have also investigated the impact of group recall in a forensically relevant context (Clark, Stephenson & Kniveton, 1990; Stephenson, Clark, & Wade, 1986). In these studies, participants listened to an audiotape of a fictional police interrogation. Participants were tested either individually or in groups of two or four on their ability to recall the interrogation. Results indicated that collaborative memory has both advantages and disadvantages and its usefulness is dependent upon the requirements of the situation. Clark et al. summarised their findings by stating, "If maximum accuracy with the minimum of evaluative comment, regardless of the occasional (but consistent) error is required, then collaboration has distinct advantages. However, if the minimization of error is more important than the completeness of accounts, and evaluative comments can be tolerated, then individual recall has distinct advantages" (p. 92). Furthermore, in both studies groups were typically overconfident about wrong answers. In a similar study, instead of overhearing an interrogation, participants listened to a telephone conversation of a kidnapper discussing his crime (Yarmey, 1992). The participants’ recollection for the conversation was tested either 10 minutes later or two days later. Participants either recalled the conversation individually, in collaborative dyads following a discussion, or individually following a discussion in pairs. The results from this study indicated that collaborative recall had no effect on overall accuracy in comparison to the recall of participants who did not discuss the event, but it did produce more consistent and concise accounts of the incident. Yarmey (1998) showed participants a video of a bank robbery and then tested participants’ memories for the event either in collaborative groups, individually following a discussion about the crime, or individually following a non-crime-related discussion, or individually following no discussion. It was found that collaborative groups were more accurate in recall than groups who discussed issues unrelated to the crime, and the no- discussion group. Collaborative groups were also better at rejecting a culprit absent lineup than the other groups. The authors concluded that collaboration can improve eyewitness recall and minimise false identifications. Chapter 1: Literature Review 10

The two studies described above (Yarmey, 1992; Yarmey & Morris, 1998) also included groups that discussed the event prior to giving individual statements. The results of these conditions are discussed in further detail in the section entitled, “Discussion Prior to Individual Statement.” Collaborative Recall and Nominal Groups. A limitation of this research on collaborative memory is that the memory of groups is compared with that of individuals. Basden, Basden, Bryner, and Thomas (1997) made the important point that group performance should not be compared with individual performance but rather with ‘nominal groups’ comprised of pooled, nonredundant data from the same number of people tested individually. For example, instead of comparing the responses generated collaboratively by a group of four with an individual participant, the responses should be compared with the pooled responses of four people tested individually. Using nominal groups, one can assess whether or not groups actually remember more than the sum of the individual members’ memories. This approach is important from a forensic perspective because in a legal context, individual witnesses’ testimonies are often pooled together to help the investigator understand what actually happened. Because many of the studies described above (e.g., Alper et.al 1976; Clark et al., 1990; Stephenson et al., 1986; Hoppe, 1962; Perlmutter, 1953; Stephenson, et al., 1983) have not compared collaborative group performance with nominal groups, they can only conclude that a group of people performs better than one individual. It cannot be said that the group working together is better than same number of people working individually. That is, we cannot be sure that the group can recall more than the sum of its constituent individuals. According to popular wisdom, “two heads are better than one.” One would think that collaborative groups could perform better than nominal groups because members of the group could “cross-cue” one another so as to produce new memories not available to the members on their own. However, Diehl and Stroebe (1987) call this the “illusion of group effectivity” because research consistently shows that, in fact, people produce more ideas in nominal groups than collaborative groups. For example, Basden et al. (1997) compared the memory of interacting collaborative groups and non-interacting nominal groups (i.e., a ‘group’ of people who never had a discussion with one another) each comprising of three members. Participants were shown a list of ninety categorized words and later asked to Chapter 1: Literature Review 11

recall the words either individually or by collaborating in groups. The results suggest that collaborative groups perform much more poorly than the same number of people tested individually as nominal groups. Similarly, Weldon and Bellinger (1997) found that collaborative groups recalled more information than individuals, but less information than nominal groups when they tested participants' recall for pictures, words, and "The War of The Ghosts" story. Research has shown that group members fail to cross-cue each other such that they do not elicit from one another new memories that the members had not initially generated on their own (Meudell et al., 1992). This failure to cross-cue one another occurred even when the opportunity for cross-cuing was manipulated using categorised lists of words (Meudell, Hitch, & Boyle, 1995). That is, no new information was generated by collaborating pairs even when an attempt was made to increase the opportunity for cross- cuing by requiring participants to recall categorised lists of words. A great deal of research supports this notion that collaborative groups perform more poorly than groups of pooled individuals (e.g., Basden et al., 1997; Basden, Basden, & Henry, 2000; Weldon & Belliger, 1997; Weldon, Blair, & Huebsch, 2000). Weldon, Blair, and Huebsch (2000) have used the term collaborative inhibition to refer to this deficit in recall by collaborating groups in comparison to non-interacting nominal groups. Causes of Collaborative Inhibition. It is possible that collaborative inhibition is due to motivational factors such as social loafing in the group situation. Social loafing is the decrease in individual motivation and effort that occurs when individuals perform in a group (Latané, Williams, & Harkins, 1979). Weldon et al. (2000) showed participants slides of unrelated words and then gave them various motivational incentives to perform better (e.g., monetary incentives, personal accountability). Their results indicated that nominal groups still performed better than collaborative groups. That is, increasing motivation failed to eliminate the collaborative inhibition effect. It is therefore unlikely that collaborative inhibition is due to social loafing. Wright and Klumpp (in press) suggested that collaborative inhibition may be due to the process of turn-taking in groups. That is, they suggested that the turn-taking etiquette of most group discussion may result in collaborative inhibition because participants are not able to choose their own timing for retrieving items. However, in their study, when Chapter 1: Literature Review 12

participants took turns recalling items, but were unaware of what the other person recalled, collaborative inhibition disappeared. Therefore, Wright and Klumpp conclude that it is not the process of turn-taking, but rather the product of what the other members recall that results in collaborative inhibition. Another possible explanation is that collaborative inhibition is due to disorganized retrieval within the groups. Basden et al. (1997) found that when participants in groups were forced to use a more organized retrieval strategy of categorizing the words, collaboration did not inhibit the groups’ recall. Furthermore, Finlay, Hitch and Meudell (2000) found that collaborative recall only led to poorer memory after individual learning but not after collaborative learning, in which the members shared similar retrieval strategies. This suggests that collaborative inhibition in recall is due to the mutual disruption of individual retrieval strategies. In sum, several experiments have investigated the effects of collaboration on recall. The results from these experiments typically indicate that collaborative groups remember more than individuals (e.g., Clark et al., 1990; Hoppe, 1962; Perlmutter, 1953), but less than ‘nominal groups’ comprised of the pooled contributions of non-interacting individuals (e.g., Basden et al., 1997; Weldon et al., 2000). This latter finding provides a good justification for the legal practice of seeking to prevent collaborative statements. In a review paper on collaborative testimony, Stephenson (1990) argued that based on social psychological evidence, it is evident that groups do not fully capitalize on individual expertise, they are more conventional than individuals, they suppress evaluative commentary, and they believe in their errors. This would suggest that collaborative statements are less complete than individual statements, but is this also true for individual statements made after group discussion? That is, is it the effects of discussion or collaborative statement that results in the decreased performance? The following section addresses this question.

Discussion Prior to Individual Statement When examining the literature on the effects of discussion prior to individual statement, we examine studies that have used a variety of methodologies. To clarify the differences between the methodologies we will introduce some new terminology. We first Chapter 1: Literature Review 13

examine studies that have investigated co-witness discussion without experimental manipulation or induction of postevent information (i.e., Natural Discussion Groups). Second, we examine studies that have utilised Experimentally Induced Information. Experimentally Induced Information is postevent information provided during the discussion phase by experimental manipulation. Two types of Experimentally Induced Information are described: (1) Confederate Induced Information, and (2) Stimulus Induced Information. Confederate Induced Information is defined as postevent information that is introduced during group discussion by an experimental confederate. In contrast, Stimulus Induced Information occurs when participants are unknowingly shown one of two slightly different versions of a stimulus event and then asked to discuss the event with group members who saw the alternative version of the event. Thus, with Stimulus Induced Information, postevent information is introduced by group members who saw the alternative version of the stimulus event. These methodologies are examined separately. Studies with Natural Discussion Groups. Like many of the studies on collaboration, several of the early studies investigating the effects of Natural Discussion Groups on individual statement have used pictures, words and stories. For example, Yuker (1955) studied the effects of discussion prior to individual statement by telling participants the story, "The War of the Ghosts." Memory for the story was tested three times. First, participants were instructed individually to write down a free recall of the story. Next, participants discussed the story in groups until they reached a consensus account. Finally, participants were instructed to write an individual account of the story once again. The results suggested that group recall was superior to the best initial free recall, and to the average initial free recall by individuals. Group performance was also superior to the average final individual free recall. Furthermore, the final individual free recall was better than the initial individual free recall and there was less variation within the groups on the final individual free recall than the initial individual accounts. This suggests that discussion in groups led to the participants “filling in” parts of their account that they might otherwise have missed. However, in this study individual and group data are confounded with order of recall, with all participants providing individual accounts before and after group accounts. Hence this finding is problematic, as the improved final performance of individuals may have been due to practice effects. Experimental studies have shown that Chapter 1: Literature Review 14

individuals recalling twice will typically recall additional information on the second occasion that was unavailable on the first, a phenomenon known as the reminiscence effect or hypermnesia (Roediger & Payne, 1982). Weldon and Bellinger (1997) overcame the weaknesses of Yuker’s (1955) study. In their study, participants' recall for pictures, words, and stories was tested twice, either individually or collaboratively in groups of three. It was found that individuals performed better after they had collaborated in groups than if they had previously recalled the items alone. The authors admit that this result is “somewhat paradoxical” (p. 1171) in that individuals recalled more after group discussion than those who had previously recalled alone, but collaborating groups recalled less than nominal groups (i.e., there was evidence of collaborative inhibition). The authors suggest that individuals recall more after group discussion due to re-exposure of items to individuals that they did not initially recall, thereby making them more accessible for later retrieval. In a more recent study it was similarly found that members who had been collaborating in groups performed better on a final individual free recall test than those who had been in non-interacting nominal groups (Basden et al., 2000). Therefore, the researchers concluded that group discussion benefits memory when the group members are later tested individually. However, in contrast to these findings indicating that discussion aids subsequent individual recall, another study found that the individual participant’s recall did not benefit from prior collaborative recall (Finlay et al., 2000). The authors admit that these results are ‘counterintuitive’ and recommend further research. Whereas the above research focuses on studies in the socio-cognitive literature, several applied studies have also been conducted which investigate the effects of discussion on individual statement. Stephenson, Abrams, Wagner, and Wade (1986) were the first researchers to investigate the effects of discussion prior to individual statement in a forensic setting. After witnessing a video of an interrogation, participants either first gave an individual recall followed by a dyadic group recall or completed these tasks in reverse order. Stephenson and his colleagues found that the completeness of the report was not affected by the order in which the participants completed these recall tasks. However, those who completed the individual recall followed by group recall had higher confidence in their Chapter 1: Literature Review 15

incorrect answers than those who completed the tasks in the reverse order, but there was no difference in confidence for correct answers. They also found that individual confidence was a better predictor of the groups’ decision than the truth. Results from another study (Yarmey, 1992) also suggest that discussion prior to individual statement is not advantageous. In this study, participants listened to a telephone conversation of a kidnapper discussing his crime. The participants’ recollection for the conversation was tested either individually, in collaborative dyads following a discussion, or individually following a discussion in pairs. The results from this study indicated that discussion prior to individual testimony led to more fabrications and less accurate estimates of duration than the other two groups. However, some studies have found that under certain conditions group recall can have a beneficial effect on eyewitness testimony. In one study, participants viewed a video of a traffic accident (Underwood & Milton, 1993). Some of the participants were told that they were going to witness an accident, whereas others were not. Approximately half of the participants viewed the video in groups of three and were encouraged to talk about the film. The other half were separated from other witnesses and not given the opportunity to discuss the event. All participants were then asked to individually fill out an accident report form. The results indicated that when witnesses were not expecting to see an accident (presumably the more ecologically valid condition), there was no difference in recall performance between those that had discussed the incident with co-witnesses and those who were not allowed to discuss the event. However, when participants were pre-warned that they were going to witness an accident, the witnesses who had discussed the event performed better than those who had not discussed it. Yarmey and Morris (1998) also found that group discussion could have a beneficial effect on eyewitness testimony under certain conditions. In particular, the type of memory task given to the participants influenced whether or not discussion was beneficial to individual memory. In their study, pairs of people who discussed a video they had viewed were more accurate in their recall, regardless of whether they went on to make collaborative or individual statements, than either pairs who discussed issues unrelated to the crime or no-discussion groups. In addition to testing the participants' descriptive memory for the event, their eyewitness identification ability was also tested. Group Chapter 1: Literature Review 16

discussion resulted in more correct rejections of a culprit-absent lineup, but only when the pairs reached collaborative decisions about the lineup, and not when discussion was followed by independent identification decisions. In sum, studies on the impact of co-witness discussion that have used Natural Discussion Groups have yielded mixed findings. An advantage of investigating co-witness discussion without any experimental manipulation or induction of postevent information is that the studies have high ecological validity. However, when investigating co-witness discussion without any experimental manipulation some experimental control is lost during the discussion phase because the experimenters cannot control what is being said by the group members. To address this limitation, several researchers have attempted to manipulate the information encountered during the discussion phase through the use of the Experimentally Induced Information methodology. Studies Utilising Experimentally Induced Information. As previously mentioned, postevent information can be experimentally induced using either Confederate Induced Information or Stimulus Induced Information. These two methodologies are described separately. Confederate Induced Information: A great deal of the research describing the effects of discussion on individual statements draws on early conformity research (Asch, 1951; see Bond & Smith, 1996 for review). In his experiments, Asch discovered that what one person says in response to a perceptual judgment task often affects another person’s judgment, even when the first person is clearly incorrect. Schneider and Watkins (1996, Experiment 2) expanded this literature by investigating response conformity in memory, or memory conformity, as it has become known (Gabbert, Memon, & Allan, 2003; Wright & Mathews, 2004; Wright, Self, & Justice, 2000). Memory conformity refers to “people being influenced by another person’s memory report” (Wright & Mathews, p. 2). In Schneider and Watkins’ experiment, participants were instructed to study a list of words and were later tested in pairs on their memory for the items. The experimenter read out a list of words and instructed the participants to respond “yes” or “no” according to whether they accepted or rejected the word as having been included in the original word list. One member of each pair was a confederate. The participant and confederate then reversed their order of responding as directed by the experimenter. The analysis indicated that when participants Chapter 1: Literature Review 17

responded second in the pair they tended to conform to the confederate’s response, regardless of whether or not it was correct. However, the effect was weaker when the confederate responded “no” than “yes” to items that had been on the original word list. That is, participants were less likely to conform when the confederate incorrectly stated that an item had not been on the word list than to disagree when the confederate incorrectly said that an item had been on the word list (i.e., participants were more likely to conform to a false positive than false negative response). In line with Schneider and Watkins’s (1996; Experiment 2) study, most subsequent studies investigating memory conformity have utilized confederates. Such studies have indicated that there are mitigating factors that influence the strength of the memory conformity effect. For example, in one set of studies (Wright & Mathews, 2004) participants were shown words (Study 1) and pictures of cars (Study 2) and then later tested for their recognition memory of the items in the presence of a confederate. The confederate always responded first, and gave erroneous information for approximately one third of the stimuli presented. Findings suggested that the effect of memory conformity was twice as large for items that had not been seen (false-positive errors) in comparison to items that had been seen (false-negative errors). That is, participants relied more on the information they received from the confederate for items they did not remember themselves than for items they did remember. In another study, participants were presented with a series of objects and their recognition memory of the objects was later tested (Walther et al., 2002). The recognition test was completed on a computer and participants could see responses that were apparently being given by the other participants in the room, but were actually manipulated by the experimenter. The results indicated that memory conformity only occurred when there was an absence of a clear recollection of the stimulus and if the stimulus was not judged to be particularly memorable. The memory conformity decreased with the presence of dissenters who broke the unanimity of the group, and increased with higher uncertainty (i.e., impaired encoding conditions). Hollin and Clifford (1983) were the first researchers to investigate the effects of memory conformity using confederates in a forensic setting. The authors staged an incident in front of a class and then had half of the participants answer questions about the incident Chapter 1: Literature Review 18

in one large group containing two trained confederates who provided incorrect information. The other half of the participants answered the same questions individually. Finally, all participants responded individually to a questionnaire, which included questions about the details for which the confederates had misled the group. Participants agreed in their testimony following group discussion and changed their responses to conform to the confederates' recall. In a second study, the method of questioning was changed from a recognition questionnaire to a narrative free recall format. It was found that changing the method of questioning to free recall eliminated the differences between discussion and non- discussion groups. The authors suggest that only “easy” items were reported in free recall accounts and that “difficult” items are the ones that are susceptible to memory conformity. That is, only the obvious, central details of the event were reported in the free recall accounts; subtle, peripheral details, for which participants were more easily misled, were not reported in the free recall accounts. One limitation of this study is that the confederates were presented as having higher credibility than the other participants because the participants were told that they had “previous experience with this type of experiment.” In addition, the confederates’ credibility would also be inflated because there were two confederates who agreed on the incorrect answer, thus offering support to each other for the validity of the information they provided. In another set of studies (Shaw et al., 1997; Experiments 2 & 3) participants were shown a video of a robbery and then questioned in pairs, with each pair containing one actual participant and one confederate. The participant and the confederate answered the questions alternately. In Experiment 3, participants were tested a second time 48 hours later, without the confederate present. In both experiments, it was found that direct co- witness information had an effect on memory. When participants received incorrect information from a co-witness, they were more likely to give the incorrect response than if they had received no co-witness information. In fact, in this study the effects of co-witness information were larger and more robust than those of suggestive questioning. Furthermore, they found that incorrect information appeared to have a stronger influence on participants’ responses than correct information. The authors note that this aspect of their results is particularly worrying because in a legal context, incorrect co-witness information may potentially have a very harmful effect. Chapter 1: Literature Review 19

Another set of studies (Hoffman, Granhag, See, & Loftus, 2001) also investigated memory conformity using confederates and reached a similar conclusion. Participants were instructed to study pictures of some objects and imagine others. They were later questioned as to whether they had seen a picture of the object, imagined it, or whether it was a new item. Prior to the participant's response, a confederate was instructed to respond to some items out loud in front of the individual. The confederate responded correctly to all of the items that had previously been seen or imagined, but for new items, they were instructed to respond that the item was either perceived, imagined, or new, or give no response. It was found that the confederate influenced the participants' memories, even if the confederate was in error. There was no difference in conformity when there was a short delay (20 minutes) and long delay (1 day) between the study and test phases of the experiment. Participants were less likely to conform (but still did conform) if the credibility of the confederate was undermined, by indicating that the responses were randomly generated. In another study (Gabbert, Memon, Allan, & Wright, 2004), participants were shown a simulated crime on video. Some participants were then exposed to misinformation by discussing the video with a confederate and other participants were exposed to misinformation by reading about the event in a narrative. In the control condition, participants read a narrative about the event that did not contain any misinformation. It was found that participants who had been exposed to misinformation were less accurate than the controls on questionnaire items. Furthermore, it was also found that misinformation encountered through discussion with the confederate was more misleading than misinformation encountered by reading about it in a narrative. In a recent study (Mudd & Govern, 2004), participants viewed a video of a car chase and then individually completed an open-ended questionnaire about the event. After completing the questionnaire, the experimenter collected it and left the room. While the experimenter was gone, a confederate participant began talking to the real participant. In the control condition, the confederate spoke to the participant about classes and the weather, however in the misinformation condition the confederate talked about answers to the questionnaire. In particular, the confederate mentioned that the car chase occurred because it was a stolen vehicle, however in the video the cause of the chase was ambiguous. After the discussion, the participants were then questioned twice using a multiple-choice Chapter 1: Literature Review 20

questionnaire: once immediately after the discussion and once two weeks later. It was found that the misinformation provided by the confederate was reported by the participant and this effect increased in magnitude over time. Furthermore, participants who received misinformation became more confident in their responses after a two-week delay whilst participants in the control condition became less confident. However, this study is limited because a repeated measures design was used and the results regarding delay may have been confounded by the repeated use of the multiple-choice questionnaire. That is, the questions themselves may have been a source of misinformation, which may have caused the increase in magnitude of the misinformation effect after the longer delay. Other research using confederates suggests that people can be misled to report entire events that they have not witnessed, not just small details of an event. One study investigated confederate influence on the likelihood that participants would claim to have witnessed non-existent film footage of the sinking of the passenger liner, the Estonia (Granhag, Strömwall, & Billings, 2003; Study 2). In this study, participants were asked to complete a questionnaire regarding the incident. While they were doing so, they overheard a confederate who was thinking out loud while filling out her own questionnaire. It was found that 76% of participants who heard the confederate say she had seen the footage of the film reported that they, too, had seen this non-existent film, whereas only 36% of participants who heard the confederate say that she hadn’t seen the film reported that they had seen it. In a similar study, participants completed a questionnaire in the presence of a confederate concerning their memory for (non-existent) CCTV footage of an explosion in a Bali nightclub (Ost, Hogbin, & Granhag, 2004). Overall, 40% of the participants claimed falsely to have seen this non-existent footage. The number of false reports increased or decreased in line with confirmatory or disconfirmatory social influence exerted by the confederate. In sum, studies using Confederate Induced Information show clear evidence of memory conformity. Not only can people be misled to report small details of an event, but they can be misled to report entire events that they have not witnessed (Granhag et al., 2003; Ost et al., 2004). Furthermore, two studies have found that Confederate Induced Information has a stronger influence on memory than misinformation encountered through Chapter 1: Literature Review 21

other means (Gabbert et al., 2004; Shaw et al., 1997). Whilst using confederates allows the experimenter to manipulate the information presented during the group discussion, and hence is the most appropriate methodology to measure the misinformation effect, it is not without limitations. An issue of concern when utilizing confederates is that the confederates may not act as true, objective co-witnesses would (although they may act like biased witnesses who are deliberately trying to mislead other witnesses). Compared to real participants, confederates may appear to have increased confidence because they know exactly what they are supposed to say. To address this issue, some researchers have begun to employ the Stimulus Induced Information methodology. Stimulus Induced Information: Schneider and Watkins (1996; Experiment 1) were the first researchers to utilise the Stimulus Induced Information methodology to investigate response conformity in memory. In their experiment, pairs of participants were instructed to study a list of words. The participants were led to believe that they had been given the same list to study, but in fact they had not. This method allows for misinformation to be presented more naturally within the pairs than the studies using delegated confederates, as both participants are real, but the experimenter can manipulate the ‘misinformation’ they present to each other by manipulating the differences in the stimuli that they see. Participants in the study were later tested in pairs on their memory for the items. When participants responded second in the pair they tended to conform to the first participant’s response, regardless of whether or not it was correct. Furthermore, participants were less likely to conform when the first participant stated that an item had not been on the word list than when they said that an item had been on the word list. Wright, Self, & Justice (2000) investigated the effects of Stimulus Induced Information in an applied setting. In their studies, pairs of participants were each shown one of two slightly different eyewitness stimuli, but were informed that they had seen the same event. In their first study, participants were shown a series of photos of cars, some of which were slightly different from those other participants had received. After a 20-minute delay, they were then shown photos of cars and asked if they had seen them in the previous stage of the study. The results indicated that those who responded first, influenced the second participant, regardless of whether they were correct. In their second study, Wright and his colleagues showed slightly different photos of a crime to different participants. Some of the Chapter 1: Literature Review 22

participants witnessed a robbery in which the thief had an accomplice, while others witnessed the same scene without an accomplice. The authors reported that whereas initial individual recollection of the photos was quite accurate, the participants tended to conform when discussing the event with another participant who had viewed the alternative set of photos. Confidence ratings were a good predictor of which person persuaded the other in the group. Although the Stimulus Induced Information methodology has higher ecological validity than studies using trained confederates, the studies by Schneider and Watkins (1996) and Wright, Self, and Justice (2000) are limited by using only word lists and photos as stimuli. Gabbert, Memon, and Allan (2003) have attempted to address this limitation by showing pairs of participants a simulated crime on video. Participants were led to believe they were viewing the same video, however they were actually shown one of two slightly different videos. After viewing the video, participants were asked to recall the event either alone or in pairs (in which members had seen the different versions of the event). Participants were then given an individual recall test to assess their memory of the crime. A significant proportion of participants who had discussed the event reported at least one detail acquired during the discussion with their co-witness that they had not personally witnessed. A new technique for introducing Stimulus Induced Information has recently been invented called ‘Manipulation of Overlapping Rivalrous Images by polarizing filters’ (MORI, Mori, 2003). Using this technology, two different events can be presented on the same screen simultaneously to two groups of viewers without them noticing that different versions exist. This is done by using two video projectors with polarizing filters to project different images onto one screen. Participants viewing the screen wear polarizing sunglasses that allow them to see what is being projected by one of the two video projectors without being aware of the other. Using this method, it has been shown that participants often change their memory of the event after having discussed the event with someone who had seen an alternate version of the video (Kanematsu, Mori, & Mori, 2003; Mori & Mori, 2004). This is especially true if they are asked to come to an agreement about what happened during group discussion (Kanematsu et al., 2003). Participants were less likely to change their answers if there was one other person in their group who had witnessed the Chapter 1: Literature Review 23

same stimulus (Mori & Mori, 2004). Whilst the MORI method is effective at deceiving participants into believing that only one version of the video exists, it can be an expensive and difficult methodology to use. The MORI method appears to be no more effective at deceiving participants into believing that only one version of the video exists than the simpler presentation of videos used by Gabbert, Memon, and Allan (2003). In sum, most of the memory conformity research involving Experimentally Induced Information (both Confederate and Stimulus Induced) suggests that postevent information provided by a co-witness may influence the memories of others. A general feature of most research involving Experimentally Induced Information is that it seeks to identify the influence of misinformation presented by one witness to another. That is, the focus is on errant information presented during discussion, rather than correct information. Therefore, the assumption is made that discussion between witnesses is a negative, detrimental process and no attempt is made to measure any possible positive effects of discussion.

Indirect Transfer of Co-witness Information In comparison to research on the direct transfer of co-witness information, relatively few studies have examined the effects of co-witness information transferred indirectly through a third party. Loftus and Greene (1980) conducted the first study of this topic. Although the primary purpose of their research was to determine the extent to which memory for faces can be altered, Loftus and Greene also inadvertently explored the issue of co-witness information. In three separate experiments (Pilot study, Experiments 1 & 2), participants were shown a face and later asked to read a description of it that had supposedly been written by another witness. The results indicated that the participants were influenced by the written descriptions such that they would incorporate misleading details into their reconstructions of the original face. Participants who did not hear the misleading details rarely incorporated such information. However, the co-witness description was purportedly written by a university professor, which may have given this witness’s statement more credibility than that of a typical witness. Chapter 1: Literature Review 24

In a second study (Betz, Skowronski, & Ostrom, 1996), participants saw bogus answers supposedly given by other participants, whose credibility would be more typical than that of a university professor. In this study participants were instructed to read a story and then take a multiple-choice test about it. They were then shown bogus answers to the questions. When later tested on their memory for the story using a cued recall test, participants changed their responses to match the bogus ones. This was especially true when the agreement on the bogus responses was high and the participants' own original memory was poor. Furthermore, participants continued to show this conformity even when they were explicitly told that the responses they had seen were bogus. Whereas the previous studies investigate the effects of indirect co-witness information on subsequent memory reports, another study was conducted which assessed the immediate effects of co-witness information on witnesses’ memory reports (Shaw et al., 1997; Experiment 1). By ‘immediate,’ the authors refer to responses made by a witness immediately after receiving information about what a co-witness had said. In this study, participants were asked to name people in a highly publicized murder trial. Participants were given bogus information on the questionnaire about responses previously given to the questions. The results indicated that when participants received incorrect information about a co-witness’s response, they were more likely to give incorrect responses than if they received no co-witness information. Corey (2001)1 investigated the immediate and delayed effects of indirect co-witness information. Participants viewed a live, staged robbery and they were interviewed both immediately after the event and again one week later. During the first interview, participants were told what other witnesses to the robbery had supposedly said. It was found that, when interviewed immediately after the robbery and one week later, participants reported the co-witness information. However, if participants were instructed to ignore the co-witness information at the second interview, they were largely effective in doing so. In contrast to the previous studies, Garven, Wood, & Malpass (2000) found that the indirect transfer of co-witness information yielded weak effects that were rarely significant. In their study, a storyteller visited children's classes, and one week later the children were

1 Only the dissertation abstract for this research has been made available and therefore, this study cannot be discussed in much detail and will not be referred to later in this thesis. Chapter 1: Literature Review 25

interviewed on their memory for the visit. In questioning, children were told what other children had supposedly reported. It was found that this co-witness information had little effect on the children’s responses. Although co-witness information had a small effect on accuracy for mundane events (e.g., the storyteller tore a book) in the first interview, no effects were observed for fantastic events (e.g., the storyteller took the child on a helicopter ride), nor were any effects observed when children were re-interviewed a few weeks later. A few studies have used eyewitness confidence, rather than memory accuracy, as the dependent variable to investigate the effects of indirect transfer of co-witness information. It is important to study eyewitness confidence because if a witness’s confidence is artificially inflated by co-witness information then this could mislead the jury. Juries often consider confidence to be an indicator of accuracy (e.g., Penrod & Cutler, 1995), however research suggests that the two are not highly correlated (e.g., Wells, Lindsay, & Ferguson, 1979). In one study investigating eyewitness confidence, a theft was staged in front of participants, who were then asked to attempt to identify the thief from a culprit-absent photo lineup (Luus & Wells, 1994). Participants were then told bogus information about whom the other co-witness had identified from the lineup. It was found that the confidence expressed by the participants regarding their identification changed on the basis of what they were told the other witness said. That is, self-rated confidence increased when the participants were told that the co-witness had identified the same person and decreased when they were told that the other witness had identified someone else or no one at all. The effect of initial co-witness information was not changed by subsequent changes to that information. For example, a witness’s confidence level wouldn’t change even if they were later told that the earlier co-witness information was incorrect. In a second study, Luus and Wells showed videotapes of participants from the first study to neutral observers who rated the participants on their overall credibility. The results from this experiment indicated that the observers' ratings of credibility closely matched the witnesses' self-reported levels of confidence. This shows that the co-witness manipulation in Study 1 was effective in altering the perceived credibility of the witnesses. Two similar studies have also investigated the effects of feedback on how accurate participants are in their confidence judgments (Allwood, Jonsson, & Granhag, 2004; Allwood, Knutsson, & Granhag, in press). These studies investigated confidence for event Chapter 1: Literature Review 26

memory as well as lineup identification. One study investigated the effects of confirmatory and disconfirmatory co-witness feedback on confidence in adults and compared this with a control group that received no feedback (Allwood et al., in press). In this study, it was found that participants were often overconfident in their decisions, especially when they received confirmatory co-witness feedback, compared to the disconfirmatory and no feedback conditions. In another study investigating children’s confidence, it was found that confidence was higher when children received confirmatory feedback than when they received disconfirmatory feedback (Allwood et al., 2004). Feedback that had supposedly come from a teacher did not influence children more than feedback that had supposedly come from another school-peer. In sum, not many studies have investigated the indirect transfer of co-witness information. Most studies that have investigated this topic have found that witness accuracy and confidence can be influenced by the indirect transfer of co-witness information.

Summary of Chapter 1 In an attempt to reconcile some of the inconsistent results that have been obtained with regards to co-witness research, Chapter 1 separately described research investigating the different ways of encountering co-witness information. First, studies investigating the direct transfer of co-witness information through discussion were described. Studies examining collaboration in statement typically showed that collaborative groups remember more than individuals, but less than ‘nominal groups.’ In contrast, studies investigating discussion prior to individual statement have yielded mixed findings depending on the methodology employed. Research on discussion prior to individual statement employing the Experimentally Induced Information methodology (both Confederate and Stimulus Induced) typically finds that postevent information provided by a co-witness may influence the memories of others. Most research involving the Experimentally Induced Information methodology seeks to identify the influence of misinformation presented by one witness to another, and therefore the assumption is made that discussion between witnesses is a detrimental process. It may therefore be advantageous to also investigate the effects of co-witness information using Natural Chapter 1: Literature Review 27

Discussion Groups as this methodology has high ecological validity. However, few studies have used this methodology, and those that have, have yielded mixed findings. Therefore, future investigation using the Natural Discussion Group methodology would be helpful to better understand the effects of discussion on memory. Co-witness information may also be encountered indirectly through a third party who tells a witness what other witnesses have reported. Most studies investigating this have found that witness accuracy and confidence can be influenced by the indirect transfer of co- witness information.

28

Chapter 2 Theoretical Explanations of Memory Conformity

While the misinformation effect is a well-established phenomenon, “what remains in dispute is the nature of a satisfactory theoretical explanation” (Deffenbacher, 1991, p. 382). This was true in 1991 and it is still true now. One critical weakness of many studies investigating the effects of memory conformity is the lack of clarity regarding whether conformity is due to memory distortion or to other factors. Traditionally, the effects of postevent information on memory have been investigated within a cognitive framework. However, when investigating the effects of co-witness discussion on memory, social factors also become relevant. As Wright and Mathews suggest, “Given memory conformity’s intellectual link to both Asch and Loftus, it is not surprising that the mechanisms proposed to account for the phenomenon also span the social-cognitive divide” (2004, p. 3). Therefore, in order to understand why memory conformity occurs, we must draw from both cognitive research on memory and social research on conformity. In this section, relevant cognitive and social theories are discussed in order to (1) explain the occurrence of memory conformity and (2) describe factors that influence memory conformity.

Why does Memory Conformity Occur? Four distinct explanations have been offered for the memory conformity effect: (1) normative social influence, (2) informational influence, (3) biased guessing and (4) modification of the memory. The empirical evidence relevant to each of these explanations is reviewed in this section.

Normative Social Influence Normative social influence occurs when an individual yields to conformity for the sole reason of gaining social approval from others (Deutsch & Gerard, 1955). For example, people may choose to look as if they are in agreement with others in order to appear more likable (Tajfel & Turner, 1986). In this explanation it is argued that people show public compliance with the group without changing their own beliefs. Chapter 2: Theoretical Explanations 29

Normative social influence is a likely explanation for the conformity observed in the studies conducted by Asch (1951, 1955, 1956) in which participants were shown a target line and three comparison lines simultaneously and asked to decide which of the comparison lines was the same length as the target line. This was very simple task and all participants answered correctly when asked individually. However, when participants were tested in groups with confederates who were instructed to give an incorrect answer, participants often conformed to the incorrect answer. When Asch asked his participants why they conformed to an obviously incorrect answer, he found that most of them wanted to avoid appearing “conspicuous” or “queer.” In the case of memory conformity, when normative social influence occurs, it is thought that both the original stimulus (e.g., the crime) and the second-hand information (e.g., information obtained from co-witness) are stored in memory and the person remembers the source of each type of information (Betz, Skowronski, & Ostrom, 1996). Despite remembering the original stimulus information, the individual responds in the socially acceptable manner, repeating the second-hand information. Whilst normative social influence may explain the conformity that occurs in Asch’s earlier studies (1951, 1955), co-witness collaboration in statement, and individual statements given in the presence of a co-witness, it is an unlikely explanation for memory conformity that may occur when people give individual statements following discussion in the absence of their co-witness. When participants give individual accounts of the eyewitness stimuli, other group members have no way of knowing what answers are given by the participants and therefore it is unlikely that the participant would yield to gain their social approval. Asch (1956) suspected that many of his participants did not believe the incorrect answers that they gave, so he replicated his earlier study, but this time, after hearing the majority’s response, the participants submitted their responses in writing. The result of this new procedure was a dramatic reduction in conformity. That is, when the other group members had no way of knowing whether they conformed or not, participants responded with the answer they knew to be correct.

Chapter 2: Theoretical Explanations 30

Informational Influence Informational influence occurs when an individual believes what other people report, but doesn’t have an actual memory for what they reported (Deutsch & Gerard, 1955). This is likely to occur when the correct answer is unclear. For example, in a series of studies Sherif (1935, 1936, 1937) demonstrated that people conform to one another when asked to make judgments about the movement of a light source. Sherif capitalised on a perceptual illusion known as the autokinetic effect, in which a stationary point of light appears to be moving when it is viewed in complete darkness. In this situation, there is no clear answer to the question of how much the light was moving, so people usually give widely different estimates of movement, ranging from a few centimetres to a meter or more, when asked to respond individually. However, when participants were asked to give their judgments in the presence of others, the range of answers given dramatically decreased. When participants were individually asked again one year later, they still gave answers that supported the group norm (Rohrer, Baron, Hoffman, & Swander, 1954). This suggests that in the absence of any clear information of their own, participants changed their beliefs to agree with the position taken by the others. When informational influence occurs in memory conformity, the second-hand information (e.g., information obtained from co-witness) is stored in memory and the individual remembers that the information was obtained second hand (Betz et al., 1996). However, the witness does not remember the original stimulus (i.e., the eyewitness’ own experience of the crime). Therefore, the witness reports and privately accepts the second- hand information because he/she has no reason to doubt it. Informational influence may explain the memory conformity that occurs after co- witness discussion, as people generally assume that information exchanged during the course of a discussion is truthful and accurate (Grice, 1975). In this case, participants may report the co-witness information when asked to remember the original stimulus because they believe it be to correct and are unable to remember their own experience of the event. According to Betz et al. (1996) both normative social influence and informational influence require that people remember the co-witness information and that it was obtained from a second-hand source. For this reason, they refer to these first two mechanisms as “source-tagged mechanisms.” That is, the memories are retrieved with a “tag” that Chapter 2: Theoretical Explanations 31

identifies that the information is second-hand. Thus, in these two explanations, no memory distortion occurs.

Biased Guessing The third explanation of memory conformity is similar to informational influence. In this explanation, memory conformity is considered to be a result of the demand characteristics of the experiment (McCloskey & Zaragoza, 1985). This argument rests on the assumption that many participants in the experiment may fail to notice or retain the critical detail in the original event. In the traditional misinformation experiment, some of these participants are then given false information regarding the critical detail, whereas others are not. When they are later questioned about the critical detail using a forced choice recognition test, they are often asked to choose the correct answer from two alternatives: (1) the detail that was in the original event or (2) the detail given to some participants as misinformation. Therefore, the participants who did not remember the original event and who received no misleading information would select the correct answer 50% of the time. However, participants who did not remember the original event and who received misleading information would often select the incorrect answer (i.e., the detail given to them as misinformation) because they had no reason to doubt the misleading information presented to them. Therefore, McCloskey and Zaragoza argue that the misinformation effect is due to an inherent response bias in the traditional experiment, and not due to memory change. In experiments investigating co-witness discussion, this response bias may be exacerbated because participants typically believe that the co-witnesses they talk to are also regular participants in the study. That is, they are typically unaware of any experimental manipulation involving confederate or stimulus induced information. Because the participants believe their co-witnesses are independent of the experiment, they may reason that the probability of two independent sources (i.e. the co-witness and the experimenter) both generating the exact same misinformation about the event is very low, and therefore, the probability that the information is correct would be high. This response bias may not be as strong in typical experiments investigating the misinformation effect because participants in those experiments generally receive both the leading questions or postevent Chapter 2: Theoretical Explanations 32

narratives and the final questions from the experimenter and therefore the participants may not perceive the two as being from independent sources. Thus, it is possible that the biased guessing explanation may be stronger in co-witness studies than traditional experiments on the misinformation effect. Another possible explanation for the misinformation effect offered by McCloskey and Zaragoza (1985) is that participants may remember both the original and the misleading information, but choose to report only the misinformation from the narrative because they believe this is what is the experimenter wants. In other words, they suggest that the misinformation effect may be solely explained by the demands of the experimental task used to demonstrate the effect, and not due to memory impairment. To test this explanation, McCloskey and Zaragoza (1985) altered the traditional experimental procedure by changing the forced-choice recognition test. Rather than requiring the participants to select the correct detail and reject the misinformation given to some participants, all participants had to choose between a detail in the original event and a novel stimulus. For example, if the item in the original event was a ‘Coke’ can and the misleading information referred to ‘7 Up’, the test would be to discriminate between the ‘Coke’ can from a can of ‘Sunkist.’ They argued that response bias would be eliminated using this modified testing procedure, but choice of the novel stimulus should still be higher among misled participants if the original memory had been significantly degraded by the postevent information. Using this procedure, McCloskey and Zaragoza were unable to find any difference in responses between control and misled participants. To further test this explanation, Zaragoza, McCloskey, and Jamis (1987) questioned participants using a cued recall procedure instead of the traditional forced-choice recognition test. This cued recall procedure involved asking questions, in which the items used as misleading information were not appropriate responses to the test questions. For example, if participants saw a ‘7 Up’ can on a desk and were later misled to believe it was a can of ‘Planter’s peanuts,’ participants were asked, “What brand of soft drink was it?” In this experiment, the misinformation effect was eliminated when participants were tested using cued recall. The authors argued that the misinformation effect was not found with a cued recall test because the biases present in the traditional recognition test were eliminated using this methodology (Zaragoza et al.). However, a criticism of this study is that the Chapter 2: Theoretical Explanations 33

misinformation suggesting a peanut can would not necessarily have degraded the original memory for the soft drink because there could have been many cans on the desk. Furthermore, given that participants were unlikely to answer “Planter’s Peanuts” in response to a question about soft drinks, this does not tell us whether participants would have reported the misinformation if the question had been more general such as, “What items were on the desk?” (Sutherland & Hayne, 2001). The suggestion that memory conformity is a result of biased guessing is similar to the informational influence explanation because in both instances the participants may remember the second-hand information and remember that it was obtained second-hand, but report it because they have no memory of the original event and therefore no reason to doubt the new information. However, the distinguishing feature between the two explanations is that the biased guessing explanation focuses more on response biases and demand characteristics. The biased guessing explanation would predict that the misinformation effect would not occur when there are no response biases or demand characteristics in the experiment (e.g., when tested using cued recall or free recall). On the other hand, the informational influence procedure would predict that memory conformity could occur in the absence of demand characteristics and response bias. Whilst biased guessing may account for the misinformation effect that occurs in some instances (e.g., McCloskey & Zaragoza, 1985; Zaragoza et al., 1987), research suggests that it is not the only reason for the occurrence of the misinformation effect. In fact, several studies have found that misinformation effect occurs despite using the modified testing procedure (e.g., Belli, Windschitl, McCarthy, & Winfrey, 1992; Ceci, Ross, & Toglia, 1987; Chandler, 1989, 1991; Windschitl, 1996). These findings are supported by a meta-analysis of 44 published experiments using modified recognition tests (Payne, Toglia, & Anastasi, 1994). The meta-analysis showed that a misinformation effect can be obtained with a modified test, with the effect size of all 44 cases being small, but significant. This suggests that the misinformation effect may be due at least partially to memory impairment, rather than just biased guessing.

Chapter 2: Theoretical Explanations 34

Modification of the Memory In explanations discussed so far, no memory distortion is believed to have occurred; the participants remember the co-witness information and that it was obtained from a second-hand source. However, it is also possible that people conform and report postevent information because their memory for the original event has changed. For example, an individual may remember the postevent information, but forget that it was obtained second- hand. In this case, individuals may report the co-witness information and believe it is their recollection of the stimulus. Several theories have been used to explain eyewitness susceptibility to postevent information and these theories may also be used to explain the effects of co-witness information on eyewitness memory. Ainsworth (1998) identifies two broad categories under which these explanations fall: (1) Alteration hypotheses and (2) Coexistence hypotheses. Experimental evidence for each of these explanations is described below. Alteration Hypothesis. The alteration hypothesis was the first explanation offered for the misinformation effect. According to this hypothesis, the misinformation effect occurs simply because the original information does not exist in memory. There are three alteration explanations for the misinformation effect: (1) Vacant slot explanation, (2) Overwriting explanation, and (3) Blend explanation. These explanations are described below. Vacant Slot Explanation: The vacant slot explanation (as named by Wright, 1995) asserts that the misinformation effect occurs because some participants may fail to encode the original information. Because these participants have not encoded the original information, the misinformation does not contradict any existing beliefs and is therefore easily accepted. This explanation differs from the informational influence explanation because the vacant slot explanation suggests that participants accept the information and believe they saw it in the original event, however the informational influence explanation assumes that participants know the information was obtained from a different source. Overwriting Explanation: Loftus (1979a) argued that misleading postevent information ‘overwrites’ or permanently alters the original memory for a witnessed event. Loftus claimed that memory for a witnessed event is stored in memory as an integrated whole. If new information about the event is introduced to the witness, it can be Chapter 2: Theoretical Explanations 35

incorporated into the original memory (as in the case of the Vacant Slot Explanation). However, if the new information contradicts the previous information, it can update or replace the original memory. This can have two consequences for behaviour. First, people will have difficulty discriminating between their memory for the original event and their memory for the postevent information. Second, some aspects of the original trace will be overwritten by later information and therefore be permanently lost. Therefore, in the case of co-witness information, this explanation would predict that information presented by co- witnesses would change the witness’s memory for the original event. Furthermore, this explanation would predict that the witness would never be able to recover the original memory and he/she would not be aware of the memory change. Loftus’s (1979a) overwriting theory challenged the model of memory held by most psychologists at the time. Loftus and Loftus (1980) conducted a survey and found that most psychologists believed that once information is stored in long-term memory it remains there permanently, and any forgetting that occurs is a result of retrieval failure. This belief was based on evidence that people who were not able to retrieve memories in one situation were later able to retrieve the memories in a different situation, for example under hypnosis or in different physical surroundings. However, Loftus and Loftus argued that even though some instances of forgetting may be due to retrieval failures, this does not mean that all instances are a result of it. In support of the overwriting theory, Loftus and Ketcham (1991) described a real case in which a witness’s memory was altered. In this case, a woman who had been raped was pressured by police authorities to give the name of her attacker. The woman eventually reported that the husband of one of her work colleagues had raped her, and this man was convicted. However, shortly afterwards the true rapist was caught and he confessed to the crime. Despite this true confession, the victim maintained that the first man had committed the rape. Loftus and Ketcham suggested that over time, the woman’s memory for the rapist’s face had been overwritten and permanently altered. There is also experimental support for the overwriting theory. Loftus (1979a) found that offering participants $25 to respond correctly did not lessen the misinformation effect, nor did it help to allow participants to make a second guess among several alternatives (Loftus & Hoffman, 1989). Loftus and Loftus (1980) reasoned that participants would be Chapter 2: Theoretical Explanations 36

able to recall the original information when financially motivated or when given a second opportunity if it still existed in memory, and that failure to be able to do so offers support for the overwriting explanation. In further support of the overwriting theory, Loftus also found that witnesses who have been misled are often just as confident in their wrong answers as those who have not been misled and thus answer correctly (Loftus, Donders, Hoffman, & Schooler, 1989). However, criticism of the overwriting theory is based on evidence that there is a co-existence of the original memory trace and the misled information (e.g., Bekerian & Bowers, 1983; Christiaansen & Ochalek, 1983; Pirolli & Mitterer, 1984). For example, it has been found that the misinformation effect occurs with explicit memory tests, but not with implicit memory tests (Dodson & Reisberg, 1991). That is, in this study misinformation impaired performance on a conventional recognition test, but not on a lexical decision task. Thus, it appears that the original memory is not altered in such a way that it is irretrievable. Loftus also faced criticisms for her memory overwriting theory from McCloskey and Zaragoza (1985) who suggested that the misinformation effect may in part be due to biased guessing (see above). In light of these findings, Loftus modified her position and offered a less extreme view. Loftus conceded that even if the original memory was not completely overwritten, as she had originally suggested, the findings of McCloskey and Zaragoza could not preclude the possibility of a blend occurring between the original information and postevent information (Loftus, Schooler, & Wagenaar, 1985). Blend Explanation: The blend explanation assumes that participants encode misinformation within the same cognitive structure as the original information (Wright, 1995). This results in a composite representation that is in some way a combination of the old and new information. When a “blend” occurs, the original memory trace is damaged and when given forced-choice recognition tests, participants choose the item that is most like the resulting blend. Loftus et al. (1985) argued that McCloskey and Zaragoza’s (1985) findings from their modified recognition test can be explained using a blend explanation. In particular, it was suggested that the original stimulus would likely be more similar to a blend of the original information and misinformation than an unseen foil. Therefore, participants with a Chapter 2: Theoretical Explanations 37

blend representation would be more likely to select the correct alternative than the foil on the recognition test. The Composite Holographic Associative Recall Model (CHARM, Metcalfe, 1990) is an example of a blend explanation. According to this model, misinformation is superimposed or blended with the original trace. The most compelling evidence for CHARM comes from experiments in which the remembered item was not the original item nor the misinformation, but rather a compromise between the two. For example, in one study participants were shown a green car, but later received misinformation suggesting that the car was blue (Loftus, 1977). When participants were tested on their memory for the car, they were given a series of 15 continuously varying colour swatches from which to select the colour of the car. A significant blue shift was found in the colour judgments given by the participants who received the misinformation in comparison with the controls. Misled participants chose a blue-green colour, which was a blend between the two pieces of information they were given. Similarly, Weinberg, Wadsworth, and Baron (1983) also found that participant’s reports of the colour of a traffic sign can be changed by subsequent misleading information. In another study, participants were shown a video of eight demonstrators interrupting a lecture (Loftus, 1975). Participants were then asked leading questions which either suggested that there had been four demonstrators or twelve demonstrators. One week later, participants were asked to report how many demonstrators they had seen interrupting the lecture. Participants who were misled to believe that there were twelve demonstrators were more likely to overestimate the number of demonstrators, while those who were told there were four demonstrators were more likely to underestimate the number of demonstrators. Loftus suggested that her participants had produced a compromise memory. That is, what they had reported was a compromise between what they had originally seen and what they were later told. When estimating numbers or when remembering colours, it is possible for witnesses to compromise between what they actually saw and what they were told. However, in many cases the witness will not have the opportunity to compromise between the original and postevent information. For example, if a witness doesn’t see a barn and is later misled to Chapter 2: Theoretical Explanations 38

believe there was one, it would be difficult for a witness to report a compromise of the two (Loftus, 1979a) Coexistence Explanation. A final explanation of the misinformation effect is that the original information and misinformation coexist and are stored separately. According to this explanation, each memory trace is capable of being recovered. Supporters of this explanation point out that research shows it is sometimes possible to recover some parts of an original memory, even after misleading information has been presented (e.g., Bekerian & Bowers, 1983; Christiaansen & Ochalek, 1983). They argue that although it is sometimes very difficult to retrieve the original memory, this does not in itself prove that the original memory does not exist. Rather, they suggest that failure to remember should be seen as a (possibly temporary) retrieval difficulty. Therefore, the coexistence explanation has a slightly more optimistic outlook on the possibility of recovering original information than the alteration explanation. According to the coexistence explanation, even though both original information and misinformation are stored in memory, a participant may choose to report the misinformation for several reasons. Some researchers (e.g., Bekerian & Bowers, 1983; Christiaansen & Ochalek, 1983; Pirolli & Mitterer, 1984) believe that the original memory is less accessible than the subsequent misleading information. Participants may report misleading information because it is retrieved more quickly and easily than the original memory. Supporters of this explanation claim that the misleading information is often more accessible for recall because it is more recent. As suggested by the ‘recency effect,’ recent memories are often more easily accessed than are more distant recollections (Murdock, 1962). Alternatively, the original memory may be rendered inaccessible due to retroactive interference (e.g., Bekerian & Bowers, 1983; Christiaansen & Ochalek, 1983). Retroactive interference occurs when information acquired more recently disrupts the retrieval of information acquired earlier (McGeoch, 1932). There are four main coexistence explanations for the misinformation effect: (1) Headed records model, (2) Fuzzy-trace theory, (3) Activation-based memory model, and (4) Source monitoring framework. These explanations are described below. Headed Records Model: The headed records model (Morton, Hammersley, & Bekerian, 1985) is one example of a coexistence explanation for memory conformity. Chapter 2: Theoretical Explanations 39

According to the headed records model, all memories leave records. These records are unchangeable and, in theory, accessible. The headed records model assumes that memory is organised as a system of records, and each record has a heading. Access to a record is achieved not directly, but indirectly through its heading and people do not necessarily have direct access to the headings. Records can only be accessed through headings by matching a description (a combination of internal search strategy and retrieval cues in the environment) to the headings. If the headings of several records match a retrieval cue, the most recently activated record is given precedence for the response. In the case of the misinformation effect, participants may retrieve the wrong record because its heading was created more recently. In other words, retroactive interference causes participants to retrieve the misinformation. In support of the headed records model, Morton et al. (1985) refer to findings that the misinformation effect disappears when questions on the final memory test are in the same order as the original sequence of events (Berkian & Bowers, 1983). They argue that when the test is given in sequential order, information about sequence is used to access the original memory and the sequential information differentiates it from the misinformation. Fuzzy-Trace Theory: Fuzzy-trace theory is another coexistence explanation that has been proposed to describe memory change following the acquisition of postevent information (e.g., Brainerd & Reyna, 1998). According to this theory, there are two types of memory trace: “gist” and “verbatim.” These refer to feelings of familiarity and episodic memories respectively and are analogous to Tulving’s (1985) distinction between knowing and remembering (Wright & Loftus, 1998). That is, “gist” memories preserve only general meanings, senses, or patterns, and “verbatim” traces preserve exact detail. The two types of traces are created in parallel and independently of one another. Gist traces are thought to be quite durable, whereas verbatim traces are thought to decay more rapidly with time and are more susceptible to interference. According to the fuzzy-trace theory, the misinformation effect may arise in two ways. First, it may occur when people rely on gist memories alone. People may have no verbatim representation of misleading postevent information because it was not presented as part of the original stimuli, and therefore they may rely on their gist memories. If gist memories are strong enough, they may be experienced as verbatim memories at the time of retrieval. Second, this theory also suggests that the misinformation Chapter 2: Theoretical Explanations 40

effect can occur when verbatim memories from one source are mistaken as memories from another. If the wrong verbatim memories are retrieved this may result in false alarms and intrusion errors. Activation-Based Memory Model: An activation-based memory model has also been used to understand the misinformation effect in eyewitness testimony (Ayers & Reder, 1998). According to this model, concepts are represented by linked nodes that form semantic networks. The nodes and the links between the nodes can have various activation strengths that get stronger or weaker in memory as a function of how frequently and recently they have been encountered. Because a concept’s strength decays over time since its last exposure, supporters of this theory suggest that participants often report misinformation because it was seen more recently. They also suggest that memory error can occur because the source of a concept’s activation can be misattributed. According to this model, people have access to the concepts themselves (the nodes) but they do not have access to the links between the nodes along which the activation travels. People can only try to infer the source responsible for the activation. In the case of the misinformation effect, people may be unaware that the source of the misinformation was not the original event that they witnessed. Source Monitoring Framework: This notion of source misattribution is also evident in Johnson’s source monitoring framework (e.g., Johnson, Hashtroudi, & Lindsay, 1993; Lindsay & Johnson, 1989). According to this framework, people do not typically retrieve a ‘tag’ with a memory that specifies its source, or “the conditions under which the memory was acquired” (Johnson et al., 1993, p. 3). Instead, people access their memories and attribute them to particular sources through decision processes performed during remembering. Therefore, during remembering it is possible for a person to misattribute a memory’s source. In the case of co-witness information, the source monitoring theory would predict that participants exposed to co-witness information may erroneously attribute the memories derived from the co-witnesses as memories derived from the actual witnessed event. Source misattribution can occur in several ways. First, source misattributions can occur if the source of the original information does not exist in memory, either because it has decayed or was never encoded. In these cases, people try to infer the source of the Chapter 2: Theoretical Explanations 41

information and they generate a plausible, but inaccurate source. Source misattributions can also occur when participants respond based on familiarity with the cue rather than the source information, or when there is a high similarity between the possible sources. Finally, source misattributions may occur when people monitor their sources in a careless, superficial manner. According to Echterhoff, Hirst, and Groll (2001) the last reason mentioned may be susceptible to social and conversational factors. Echterhoff and his associates argued that in social settings, people may draw on conversational rules in order to determine whether they can rely on the accuracy of the information conveyed by a source. According to Grice (1975) one such conversational assumption is that people often assume that what others say is true. Echterhoff et al. maintains that if people fail to realize that this conversational assumption was not maintained, then they may not make an effort to carefully monitor their sources. In support of the source monitoring framework, it has been found that the more similar the circumstances in which the original and misleading information are presented, the larger the misinformation effect observed (Lindsay, 1990). Likewise, a key prediction of the source monitoring theory is that any manipulation that increases the extent to which memories from one source resemble those from another increases the likelihood of source misattribution. In support of this prediction, Allen and Lindsay (1998) found that source misattributions were more common for details that were similar between the original event and postevent information than for details that were quite different between the two. For example, if participants were shown a ‘Pepsi’ can in the original slide show and then a ‘Coke’ can in a subsequent show, participants would often report that the ‘Coke’ can was in the original show because the two soft drinks are quite similar. However, these findings were only obtained when there was an interval of 48 hours between the two events, and not when there was no time delay. Thus, source monitoring confusions are more likely to occur after a delay. Additional support for the source monitoring framework comes from studies showing that the misinformation effect is decreased if participants are instructed to monitor the source of their information. For example, participants are better able to discriminate objects they have seen in a picture as opposed to reading about in a postevent narrative if they are given instructions to assign items by source (picture/description) rather than simply Chapter 2: Theoretical Explanations 42

asked whether they were in the picture (Lindsay & Johnson, 1989). This suggests that to some extent participants are able to differentiate between information they obtained from the original source and that which they obtained second-hand. However, other studies show that even when participants are encouraged to monitor the source of their information, they are still vulnerable to misleading information (see Weingardt, Toland, & Loftus, 1994).

Evaluating the Theories of Memory Conformity Four main explanations have been discussed that may describe why memory conformity occurs: (1) normative social influence, (2) informational influence, (3) biased guessing, and (4) modification of the memory. The explanations offered are not necessarily mutually exclusive. Indeed, there is no reason to expect that the original memory should always suffer the same fate (Cohen, 1989), and different sources of postevent information may have different effects on memory. For example, memory conformity to co-witness information may occur for a different reason than the misinformation effect observed after exposure to misleading questions. The research described in this thesis compares predictions made by the alternative explanations in an attempt to identify the mechanism most likely responsible for memory conformity in individual recall following co-witness discussion. As mentioned earlier, normative social influence (Deutsch & Gerard, 1955) is an unlikely explanation for memory conformity when, following discussion, statements are given individually, in the absence of other group members. According to the normative social influence explanation, we would expect the misinformation effect to occur when participants are in groups, but not when they are giving individual testimony in the absence of others. That is, when participants are asked to give individual accounts of the original stimulus, it is unlikely that they would feel pressure from others to conform. However, normative social influence may play a factor in memory conformity when groups are asked to give their testimony collaboratively, or members are asked to give individual statements in front of other group members. Informational influence (Deutsch & Gerard, 1955), biased guessing (McCloskey & Zaragoza, 1985), and modification of memory may help explain why memory conformity Chapter 2: Theoretical Explanations 43

occurs when participants are tested individually, as they are in the studies presented in this thesis. In the case of informational influence, participants may report the information they received from a co-witness because they believe it is correct and have no reason to doubt it. In the case of biased guessing, the participants may report the misinformation because of demand characteristics or response biases in the experiment. In both of these instances, the participant has awareness at some level that they obtained the information from a second- hand source. Alternatively, it is also possible that participants may conform and report the information because their memory had changed. The research presented in this thesis compares these alternative explanations to determine which best explains memory conformity in individual recall following co-witness discussion. It is very important to distinguish between these explanations because each explanation provides a different reason for the misinformation effect and may produce a different prediction about whether it would be possible to mediate any negative effects of co-witness discussion. One way to determine whether memory conformity occurs solely because of biased guessing is to eliminate the demand characteristics in the experiment. Demand characteristics and the resulting response bias can be reduced by asking participants to respond to open-ended questions (e.g., free recall) that do not require forced-choice responses. One study investigating the effects of co-witness discussion on memory found that the misinformation effect was eliminated when participants gave free written statements as opposed to interrogative questioning (Hollin & Clifford, 1983). Furthermore, a study investigated the effects of leading questions on eyewitness memory and found that participants were less likely to report the postevent information when asked general, free recall questions, in comparison to more specific questions (Sutherland & Hayne, 2001). There still may be an implied demand that participants should write as much as they can though, and this may result in participants reporting misinformation if they believe it is correct. It therefore becomes important to determine whether participants are reporting this information because of informational influence or because their memories have been modified. Researchers have developed a way to distinguish between informational influence and memory change by warning participants about possible misinformation. That is, participants are warned after they have received misinformation that the information they Chapter 2: Theoretical Explanations 44

received may not have been correct, and then they are tested on their memory for the original event. It is argued that if the misinformation effect ceases to occur after the presentation of a warning, it is more likely that the memory conformity is due to informational influence than to memory change. However, if the misinformation effect still persists after the presentation of a warning, then this implies that participants are unable to distinguish whether the information was from the original source or whether it was obtained second-hand from the co-witness. In this scenario, it is likely that modification of the memory trace occurred. Studies have investigated the effects of warning on participants’ susceptibility to misinformation presented through leading questions and postevent narratives. One study found that warning participants after they received postevent information did not decrease the misinformation effect, but warning them beforehand did (Greene, Flynn, & Loftus, 1982). However, another study found that warnings may reduce the misinformation effect even if they are only given after the presentation of misinformation (Christiannsen & Ochalek, 1983). In this study, participants were told that a few of the statements in the narrative they read were incorrect. It was found that when participants were warned after receiving misinformation they were able to edit out errant details. However, the warnings used in both studies were quite general in that they questioned the overall accuracy of the postevent information. These warnings are not specific enough to assume that the misled participants would be aware that the misinformation was indeed false (Wright, 1995). In order to address this criticism, a new warning paradigm was applied to the study of the misinformation effect by Lindsay (1990). Lindsay warned his participants that the information they had read in a postevent narrative was wrong and that it should not be reported on the memory test. Specifically, he told them, “there is no question on this test for which the correct answer was mentioned in the story” (p. 1080). Therefore, if participants were planning to report the information from the postevent narrative for informational reasons or task demands of the experiment, they would no longer do this, provided they could remember the source of the information. In fact, it was found that participants continued to base their responses on the misinformation, despite explicit instructions against doing so. This implies that the misled participants experienced a memory change Chapter 2: Theoretical Explanations 45

and they actually believed the misinformation that was in the postevent narrative was in the original story. To date, just one study has used warnings to investigate the effects of co-witness discussion on memory (Meade & Roediger, 2002). In this experiment, it was found that warnings reduced the effect of co-witness misinformation, but did not eliminate it, suggesting that some distortion of the original memory had occurred. Similarly, a study investigating indirect co-witness information found that when participants were warned that co-witness information they received from the interviewer was ‘bogus,’ the effect of the co- witness misinformation was reduced, but not eliminated (Corey, 2001). Experiments described in this thesis (Studies 5-7) include a warning for some participants about possible misinformation in an attempt to determine whether participants report misinformation because of informational influence or memory change. So far, this section has discussed ways to differentiate the normative social influence, informational influence, biased guessing, and modification of memory trace explanations for memory conformity. Table 2.1 summarises the predictions regarding the misinformation effect made by each of these explanations considered. None of the methods described so far (i.e., individual recognition tests, individual free recall, and warnings) help to differentiate between the various modification of memory trace explanations.

Chapter 2: Theoretical Explanations 46

Table 2.1 Theoretical Predictions of Whether the Misinformation Effect will Occur Under Differing Circumstances

Theory Collaborative Individual Individual Warnings about Testimony Recognition Free recall misinformation tests Normative X X X Influence Biased X X Guessing Informational X Influence Memory Change: Alteration Memory Change: Coexistence Note. = Misinformation effect would be expected to occur; X = misinformation effect would not be expected to occur.

In order to differentiate between the various memory change explanations, the present research compares predictions made by the alternative explanations, but focuses mostly on the source monitoring framework. Although other theoretical interpretations are plausible, it is the source monitoring theory that has received the most attention with respect to co-witness information (e.g., Echterhoff et al., 2001; Meade & Roediger, 2002; Roediger, Meade, & Bergman, 2001). According to Roediger and his associates (2001) this is because, “The source monitoring framework provides a natural and compelling interpretation that is rich in suggesting future research”(p. 369). Most importantly for this thesis, the source monitoring framework also provides predictions that are useful for differentiating it from the other memory change explanations. According to the source monitoring framework, the misinformation effect should decrease if participants are instructed to monitor the source of their information. However, the vacant slot, overwriting, and blend explanations would predict that source monitoring would have no influence on the misinformation effect (see Table 2.2).

Chapter 2: Theoretical Explanations 47

Table 2.2 Theoretical Predictions on the Effect of Source Monitoring Instructions

Theory Source Monitoring Instructions Memory Change: No effect Alteration Memory Change: Coexistence: Source ↓ Misinformation effect Monitoring Framework

In support of the source monitoring framework, one study has shown that the misinformation effect decreases if participants are given instructions to indicate the source of the item they recall rather than to simply state what was in the original scene (Lindsay & Johnson, 1989). Similarly, Zaragoza and Koshmider (1989) also asked participants to monitor the source of their information. Specifically, they asked participants to indicate whether the information they remembered had come from the slides they had seen or a postevent narrative. In accordance with the previous findings, they found that, “subjects are able to distinguish between information they specifically remember seeing at a witnessed event and information relevant to the event that they learned about subsequently” (p. 254). These findings suggest that the misinformation effect may be due to source misattributions, and that there may be a coexistence of memory traces. To date, one set of studies has included source monitoring instructions when investigating co-witness memory conformity (Meade & Roediger, 2002). In these studies, participants were shown a series of slides and then asked to recall items from the slides with a confederate who erroneously reported some misinformation. Participants were then asked to individually complete a recognition questionnaire and indicate whether they remembered the information they reported from: (1) the slide only, (2) the confederate only, (3) both the slide and confederate, or (4) neither the slide nor the confederate. It was found that participants reported the misinformation despite the fact that they were asked to monitor the source of their information. The authors report that the source monitoring did not reduce or eliminate the misinformation effect in their studies because memory conformity is a “powerful enough phenomenon to persist even under these stringent testing conditions” (p. 1007). Chapter 2: Theoretical Explanations 48

The research in this thesis compares predictions made by the alternative memory change explanations by investigating whether the misinformation effect occurs when people are instructed to monitor their sources. The alteration explanations (i.e., memory overwriting, vacant slot, and blend explanations) predict that witnesses would be unable to distinguish what they actually observed from information obtained second-hand from a co- witness. However, the source-monitoring theory would predict that it might be possible for witnesses to distinguish between their ‘real’ memories and information obtained from a co- witness if they are encouraged to monitor the source of their information. These different theoretical predictions may have important policy implications, so it is necessary to find out which prediction is best supported by the evidence. As Ainsworth (1998) describes, if the alteration hypothesis is correct, then it is likely that a witness’s memory for the original event may be permanently altered if he/she is given any subsequent information about the event. It would therefore be important for police officers to establish whether a witness has discussed the event with others or whether he/she has seen reports about the event in the media. Police officers must also be aware that their form of questioning may itself contaminate a witness’s memory. Alternatively, if the coexistence hypothesis is correct, then it is likely that a witness would be able to recover original event memories. It would therefore be important for police officers to instruct witnesses to think carefully about the source of their memories, or to mentally reinstate the context of the event because context reinstatement may trigger memories from the event (Tulving & Thompson, 1973). A further purpose of the research in this thesis is to investigate whether memories obtained second-hand from a co-witness are in some way different from memories for events actually observed. One method of doing this has been to instruct participants to produce remember/know judgments (Tulving, 1985). For example, in a study investigating the effects of discussion on memory, participants were asked to recall items they had observed in a scene and then make remember/know judgments for each item (Roediger et al., 2001). If the participants specifically remembered seeing an aspect of the scene, they would indicate that they “remembered” it. In contrast, “know” responses were used to indicate aspects for which the participants did not have any specific recollection, but which they believed to be true. Although participants in Roediger et al.’s study did falsely report Chapter 2: Theoretical Explanations 49

some suggested items, they were more likely to claim they “knew” the suggested items had been in the scenes than to report that they specifically “remembered” seeing them. These results were replicated in studies by Meade and Roediger (2002). This implies that to some extent participants were able to distinguish information obtained from a co-witness from information they obtained first-hand from the scene, offering some further support for the coexistence hypothesis. Debate exists regarding the meaning of remember/know responses. Some researchers argue that they reflect output from two memory systems (e.g., Gardiner & Java, 1993) and others suggest they reflect output from just one (e.g., Donaldson, 1996; Dunn, 2004). This will be described in more detail in the concluding chapter. In the experiments described in the current thesis, participants were asked to make remember/know judgments to further investigate these findings in the context of co-witness memory conformity. A second way to investigate whether memories obtained second-hand from a co- witness are different in some way from memories for events actually observed is to investigate eyewitness confidence. Loftus et al. (1989) recorded participants’ self reported confidence levels and found that misled participants responded as confidently to the misled items as they did to their genuine memories. Therefore, it seems as though the level of confidence does not differentiate between memories for events actually observed and those for which people have been misled. In the experiments described in this thesis, participants were asked to indicate their confidence to further investigate this finding. In sum, there are various theories that have been offered to explain why memory conformity occurs. The research described in this thesis attempts to compare predictions made by the alternative explanations in order to identify the causes of memory conformity in individual recall following co-witness discussion. In particular, the present research uses individual testing, open-ended questioning, warnings, source monitoring tests, confidence ratings, and remember/know judgments in order to gain an enhanced understanding of the mechanisms responsible for memory conformity.

Chapter 2: Theoretical Explanations 50

Factors that Influence Conformity Whilst it has been shown that in some circumstances many people tend to conform to the opinions of others, we also know that some people are able to resist conforming in some situations. For example, in Asch’s original research (1955) about one quarter of the participants never yielded to the erroneous judgments of others. This section of the literature review examines factors influencing whether or not a person is likely to conform that are (1) in the situation, and (2) within the individual.

Factors in the Situation There are several situational factors that increase the likelihood of an individual conforming. The first, and most obvious factor is the number of people who report the information. Conformity increases as the number in the majority increases, but only up to a point. Research has shown that participants yield to three or four confederates almost as often as they yield to fifteen (Asch, 1956). This may be because the participants believed that the others in the group were colluding with one another, or being influenced by the others. That is, if we believe that other group members are not acting independently of one another, then the effect of a larger group decreases. Therefore, it is not just the number of people providing the information that produces conformity, but rather the number of apparently independent sources of the information (Wilder, 1977). One study has investigated the effect of group size on the indirect transfer of co- witness information (Walther et al., 2002). Participants were required to complete a recognition test about a series of slides they had been shown. They were given answers that had supposedly been given by either five or ten other people. It was found that participants conformed more to the larger group than the smaller group, but only when the participants had clear encoding conditions (i.e., the stimulus slides had been presented to them slowly). When they had impaired encoding conditions (i.e., the stimulus slides had been presented to them quickly), participants conformed to the smaller group, and the larger group size did not strengthen the effect any further. Other researchers (Schneider & Watkins, 1996) have noted that whilst a majority it needed for most conformity studies, this is not necessarily Chapter 2: Theoretical Explanations 51

true for memory conformity. In their study on memory conformity, they found that participants conformed to just one confederate in a group of two. A second situational factor that increases the likelihood of conformity is group cohesion. Cohesive groups are committed to shared goals and feel positively about each other and their group. Studies have shown that the pressure to conform is greater in cohesive groups than in non-cohesive groups (Forsyth, 1983; Schachter, 1951). However, one study investigated the effects of cohesiveness in pairs (stranger, friend, partner) on conformity during co-witness discussion and found that contrary to expectations, there were no differences found as a function of the participants’ self-rated level of acquaintance (Foster & Ost, 2003). Uncertainty is a third factor that increases the likelihood of conformity. As mentioned earlier, when the original stimulus is unclear (e.g., Sherif, 1935, 1936, 1937) participants look to others for information. Therefore, conformity increases with decreasing certainty. In a study investigating indirect transfer of co-witness information, it was found that the influence of responses given by others increased with higher uncertainty (Walther et al, 2002). A fourth factor that increases the likelihood of conformity is the credibility of the information source. Individuals are more likely to conform when the information source has high credibility. In one study, the credibility of the confederate was manipulated by telling the participant that their partner (the confederate) either had a same, better, or worse view of the original stimulus in comparison to the participant (Brimacombe, Allison, & Garrioch, 2003). It was found that participants were more likely to report co-witness information when it came from a credible source (i.e., when the confederate supposedly had a better view of the stimulus than the participant). In another study, participants received indirect co-witness information about an accident that was either from an innocent bystander (credible source) or a participant in the accident (non-credible source). It was fond that participants were more influenced when the information came from the credible source, presumably because they realized that the accident participant might have been biased (Dodd & Bradshaw, 1980). Thus far we have discussed factors that increase the likelihood of conformity, however there are also factors that decrease the likelihood of an individual conforming and Chapter 2: Theoretical Explanations 52

reporting postevent information. Asch (1952) showed that the impact of the group decreased dramatically (by almost 80%) if just one of the confederates agreed with the participant and gave a correct answer. Other studies investigating conformity (Allen & Lavine, 1969, 1971) have shown that a confederate who disagreed with the group reduced the participants’ conformity even when the confederate dissenter did not agree with the participant. Furthermore, confederate dissenters also led to a reduction in participants’ conformity even when the confederate appeared to be incompetent at the task. In a study investigating indirect transfer of co-witness information, it was found that the presence of dissenters also decreased conformity (Walther et al, 2002). Furthermore, a study investigating the effects of co-witness discussion has also found that participants were less likely to change their answers if there was one other person in their group who agreed with the participant (Mori & Mori, 2004). A second factor that decreases the likelihood of conformity is prior public commitment. Individuals are more likely to resist a majority view if they have already committed themselves to a contrary position. As one study has shown, participants who made a prior public commitment about their position before hearing answers given by other group members conformed 25% less than participants who did not make this prior commitment (Deutsch & Gerard, 1955). A final situational factor that may decrease the likelihood of conformity is the collective information sharing bias (Wittenbaum, Hubbell, & Zuckerman, 1999). The collective information sharing bias is the tendency for group discussions to focus on the reiteration of information that group members share in common, with little attention being paid to the exposure of information that is unique to individual members (e.g., Stasser & Titus, 1985; 1987). If unshared information is mentioned, it is less likely to be repeated and reconsidered later in the discussion than shared information (Stasser, Taylor, & Hanna, 1989). Therefore, it is possible that participants will steer the group discussion away from topics on which people do not agree (such as those about which a confederate is supplying misinformation), and focus on details that they agree on, although this has not been demonstrated in the context of co-witness memory conformity.

Chapter 2: Theoretical Explanations 53

Factors within the Individual There are several factors within individuals that influence the likelihood of whether they will conform to others. Personality characteristics of the individual are one such factor. Crutchfield (1955) was one of the first researchers to investigate the effects of personality characteristics on conformity. He found that in comparison to people who do conform, those who don’t demonstrate more “intellectual effectiveness, ego strength, leadership ability, and maturity of social relations” and less “inferiority feelings, rigid and excessive self-control, and authoritarian attitudes” (p. 194). Other researchers have found that people who don’t conform have a reduced need for affiliation and less of a tendency to blame themselves compared to those who do conform (Costanzo, 1970; Ellis, Nel & Van Rooyen, 1991; McGhee & Teevan, 1967). Furthermore, individuals high on private self- consciousness (i.e., awareness of their internal states, feelings, attitudes, and opinions) are less likely to conform (Froming & Carver, 1981; Scheier, 1980) than those who are low on private self-consciousness. Very little research has been conducted that investigates the effects of individual differences on memory conformity, although one study has shown that individuals who score higher on a measure of trait dissociation (i.e., the Dissociative Experiences Scale, Bernstein & Putnam, 1986) are more likely to accept confederate induced misinformation than those who score lower on this scale (Ost, Hogbin, & Granhag, 2004). Another factor that may influence an individual’s tendency to conform is their gender. Early studies reported that women tend to conform more than men (e.g., Crutchfield, 1955), and it was suggested that the demands of different gender roles were responsible. Women were expected to be submissive, while men were expected to be more dominant and independent. A meta-analysis of conformity studies revealed that conformity rates were significantly higher when there were greater proportions of female respondents (Bond & Smith, 1996). However, one study has shown that conformity is influenced by one’s familiarity with the issue at hand, not one’s gender (Sistrunk & McDavid, 1971). Men conformed less on matters relating to stereotypically ‘masculine’ topics (e.g., sports, warfare) and women conformed less on matters relating to stereotypically ‘feminine’ topics (e.g., fashion, domestic issues). On gender–neutral topics, there were no differences found between men and women. Another study found gender differences in conformity, but only Chapter 2: Theoretical Explanations 54

when the participants thought that other people would see their responses (Eagley, Wood, & Fishbaugh, 1981). Women conformed more and men conformed less when they thought they were being observed. This tendency may result from pressure to behave in ways viewed as acceptable within traditional gender-role constraints. Age is a third factor that may influence an individual’s likelihood of conformity. Young adolescents appear to be more vulnerable to peer pressure than younger children and adults (Iscoe, Williams, & Harvey, 1963). A study investigating memory conformity found that there were no differences in memory conformity for inaccurate information between younger (18-30 years) and older (60-80 years) adult participants, however older participants conformed more to correct information than younger participants (Gabbert, Memon, & Allan, 2003). A final factor that may influence an individual’s tendency to conform is their culture. Cultures have been described as either individualist or collectivist (Hui & Triandis, 1986). Individualist cultures emphasize the importance of independence, autonomy, and self-reliance, whereas collectivist cultures emphasize interdependence, cooperation, and social harmony. A meta-analysis revealed that conformity rates are generally higher in collectivist cultures than individualist cultures (Bond & Smith, 1996). Furthermore, the study revealed that the impact of cultural variables was greater than any other factor studied, including variables such as majority size, which is typically identified as being important. Whilst the factors that influence conformity raise some interesting empirical and theoretical questions, they are not explored further in this thesis. These factors are not studied in detail because it is not possible to mediate any negative effects of co-witness discussion by manipulating them. Eyewitnesses commonly discuss the event prior to police arrival, and it is therefore impossible to change what occurs during the discussion phase. A distinction in the literature has been drawn between system variables and estimator variables (Wells, 1978). System variables include variables that are under the control of the judicial system (e.g., conduct of the eyewitness interview) and can therefore be changed to increase the accuracy of eyewitnesses. On the other hand, estimator variables refer to variables that cannot be controlled by the judicial system (e.g., cohesiveness of group, personality of witness) and can only be used in the courtroom to enhance or discount the Chapter 2: Theoretical Explanations 55

credibility of eyewitnesses. This thesis attempts to focus on relevant system variables in the context of co-witness discussion in the hope that a greater knowledge of the impact of these variables can lead to policy change, which in turn can increase the accuracy of eyewitness evidence. Although the experiments described in this thesis do not attempt to manipulate and test the factors that influence conformity, they are used to help understand the results obtained and consider implications of the findings.

Summary of Chapter 2 In Chapter 2, theories are described that may be used to explain why memory conformity occurs. Traditionally, cognitive theories have been used to explain the effects of postevent information on memory. However, the different sources of postevent information may have different explanations. For example, when investigating the effects of co-witness discussion on memory, social factors also become relevant. Therefore, in order to understand why memory conformity occurs in individual recall following co-witness discussion, it is important to draw from both cognitive research on memory as well as social research on conformity. The four main explanations described in this chapter are: (1) normative social influence, (2) informational influence, (3) biased guessing, and (4) modification of the memory. These explanations are not necessarily mutually exclusive, however the research described in this thesis attempts to identify the mechanism most likely responsible for memory conformity following co-witness discussion by comparing predictions made by the alternative explanations. It is hoped that with this increased understanding, it will be possible to mediate any negative effects of co-witness discussion at the stage of recall. Chapter 2 also presented a brief summary of factors that influence memory conformity. These factors include factors that are within the individual and those which are in the situation. Although the experiments described in this thesis do not attempt to manipulate these factors, they are used to help interpret the findings.

56

SECTION 2: SURVEY STUDIES

Section 2 Aims

• To investigate the importance of studying co-witness discussion by measuring its prevalence.

• To investigate the opinions that police officers hold regarding discussion between witnesses.

Before describing experiments that investigate the methodological and theoretical implications of co-witness discussion on eyewitness memory, an important first step is to measure the prevalence of co-witness discussion and to investigate the opinions held by legal authorities regarding discussion between witnesses. Previously, it has been assumed that co-witnesses commonly discuss the event with one another, but to date there is no quantitative evidence to support this. It has also been assumed that legal authorities, such as police officers, would discourage co-witness discussion, but again this has never been assessed. To examine these issues, two surveys were conducted: one of real eyewitnesses and the other of police officers. Study 1 was designed with the aims of determining the proportion of real eyewitnesses who report discussing the event with co-witnesses, and the nature and purpose of their discussions. Study 2 investigated the opinions that police officers hold regarding discussion between co-witnesses to a crime. The aims of Study 2 were to identify whether police officers are aware of the prevalence of co-witness discussion and to assess their perception of the impact of co-witness discussion on eyewitness memory. The findings from these surveys were used to inform subsequent experimental studies of the effects of co-witness discussion on memory and to increase their ecological validity. In addition, given the applied nature of this research, these findings are also used later in the thesis to consider the practical implications of the research.

57

Chapter 3 Study 1: Witnesses Talk2

Before the police arrive on the scene, witnesses talk to one another… I personally witnessed this when I entered a shop in Cambridge, Massachusetts, moments after a robbery had occurred and before the police arrived. In the immediate aftermath, customers and employers shared their recollections, providing fuel for influencing the thoughts of one another. (Loftus, 2003, p. 232-233)

Within the legal system the ideal is that testimony given by eyewitnesses should be independent of one another, however this is frequently not the case. As Yarmey observed, “Discussion among victims or witnesses to a crime is difficult, if not impossible to prevent” (1992, p. 252). Crimes are often ‘remarkable’ in the sense of the event's uniqueness and also in the sense that witnesses repeatedly ‘remark’ on the event to others (Yuille & Daylen, 1998). In one study, all of the crime witnesses interviewed claimed that they had told almost everyone they had met about the event (Yuille & Daylen). It is clear that witnesses often do talk to each other about the event. Several real life instances demonstrate how co-witness information is passed from one witness to another. For example, after the Oklahoma bombing in 1995, one witness, Tom Kessinger, may have influenced several other witnesses because of his confident recollection (Memon & Wright, 1999). Kessinger reported that he had seen the suspect with an accomplice, who became known as ‘John Doe 2.’ The other witnesses did not initially report having seen an accomplice, however they, too, came to believe there was one. This prompted the Federal Bureau of Investigation (FBI) to launch an intense search for this mysterious accomplice who may never have existed. In testimony, Kessinger admitted that the witnesses had discussed their recollections of the incident prior to the arrival of the FBI. Another high profile case in which co-witness information was passed from one witness to another occurred when the Swedish foreign affairs minister, Anna Lindh, was murdered (Granhag, Memon, Gabbert, & Allwood, 2004). Following the murder, witnesses were brought together in a small shop nearby and allowed to talk to one another and hear

2 Study 1 is described in a paper by Paterson & Kemp (in press). Chapter 3: Witnesses Talk 58

each other’s statements to the police. One witness confidently reported that the offender was wearing a camouflage jacket, which led other witnesses to believe this. However, CCTV footage revealed that the offender was not, in fact, wearing such a jacket. In actual fact, he was wearing a grey, ‘Nike’ sweatshirt. Other examples of co-witness discussion have occurred in the United Kingdom. Peter Hain (1976), the current Leader of the House of Commons, was falsely charged with bank robbery when he was young. One of the reasons to which he attributed this was co- witness discussion. Hain also identified two other instances in which discussion between witnesses prior to individually viewing an identification parade may have contributed to wrongful convictions. As he states, “In the cases of both Ince and Meehan…witnesses were allowed to communicate, and the possibility must exist that the subsequent choices were affected by information passed on by other witnesses” (p. 136). In one of these cases, an innocent man had to serve nearly seven years in prison before being released. Therefore, it appears that co-witness discussion can lead to profound miscarriages of justice. Whereas the previously mentioned examples illustrate anecdotal evidence suggesting that witnesses commonly discuss the event with others, the discussion behaviour of real eyewitnesses has never been measured. Therefore, the purpose of this study is to measure the discussion behaviour of real witnesses. In addition to identifying the prevalence of discussion amongst witnesses, this study also aims to investigate how common it is for a witness to view a serious event in the presence of a co-witness. This can help us gain an understanding of how often the opportunity for co-witness discussion arises in the first place. To date, two studies have investigated this issue, however the conclusions drawn from these two studies are quite different. One study found that that only about 17% of police lineups involved a sole witness (Wright & McDaid, 1996). However, the other study found that 71% of the cases investigated involved the victim as the only witness. The present research investigates how many witnesses report having at least one other co-witness, and discusses possible reasons for the apparent discrepancy between the two studies. Chapter 3: Witnesses Talk 59

Method

Participants A Screening Questionnaire was given to 1000 undergraduate psychology students and was completed by 773 students (71% female). The average age of respondents was 20.1 (SD = 4.0) years. Respondents to the Screening Questionnaire were invited to complete the Follow-up Questionnaire if: (1) they had experienced a serious event, (2) there was at least one other witness present at the event, and (3) they indicated that they were willing to participate in research related to the initial questionnaire. Respondents were not aware of the conditions that would make them eligible for the Follow-up Questionnaire. A total of 178 respondents were eligible for the Follow-up Questionnaire and they were contacted by phone or e-mail. The Follow-up Questionnaire was completed by 60 respondents (73% female) with a mean age of 20.9 years (SD = 5.1). A further six respondents completed the Follow-up Questionnaire, but were excluded from the analysis; three because there were no co-witnesses in the incident they described in the Follow-up Questionnaire, two because they arrived after the event and did not actually witness it, and one because the event was not deemed serious enough. Respondents received credit for their psychology course upon completion of each questionnaire.

Materials Screening Questionnaire. A Screening Questionnaire was developed (see Appendix A) to identify individuals who were eligible for the Follow-up Questionnaire. Respondents were asked to indicate whether they had witnessed any of a series of events listed (homicide; attempted murder; manslaughter; physical assault; sexual assault; event/accident causing serious injury/death; abduction; property vandalism; breaking and entering; motor vehicle theft; fraud, blackmail, or extortion; arson; robbery) or a similar type of event. Respondents who indicated that they had witnessed such an event completed several further questions and indicated whether anyone else had witnessed the event. If participants had witnessed more than one event, they were asked to answer the questions with regards to the most serious or personally relevant event that they witnessed. Interested Chapter 3: Witnesses Talk 60

participants who had witnessed a serious event for which there was at least one other witness present were invited to complete a Follow-up Questionnaire. Follow-up Questionnaire. The Follow-up Questionnaire (see Appendix B) was primarily developed to assess the discussion behaviour of witnesses. The questionnaire was divided into four main parts. First, respondents were asked to report some basic demographic information including their age, ethnicity, and sex. The second part of the questionnaire asked respondents to report details of one serious event they had witnessed. Respondents were required to briefly describe the serious event, and indicate how much they feared for their personal safety during the incident (using a six-point scale; 1= Not at all; 6 = Very strongly) and the impact the event had on their life (using a six-point scale; 1= No influence; 6 = Extreme). Respondents were also asked to indicate whether they were both an eyewitness and victim of the event, or whether they were an eyewitness, but not a victim of the event. In addition, respondents were asked whether the police had investigated the event, whether they had been interviewed by the police, and whether the case had gone to trial. The third part of the questionnaire investigated the respondents’ discussions with others. Specifically, the questionnaire asked to whom the witnesses talked about their experience (other witnesses, friends, family, police officers, lawyers, religious authorities, strangers, medical doctors, psychologists, or other), the amount of time spent discussing the event (using an eight-point scale; 1 = Never, 2 = Once or twice, 3 = Infrequently, 4 = Occasionally, 5 = Often, 6 = Frequently, 7 = Very Frequently, 8 = Almost continuously), and the purpose of their conversations (providing information, emotional support, advice- seeking, story-telling, or other). This information was requested for each of three separate time periods following the incident (24 hours, one week, and six months after the event). The fourth part of the questionnaire focused on discussion with other witnesses. In this section, respondents were asked to state how many others had witnessed the event, and whether they had discussed the event with their co-witnesses. They were also asked whether police or lawyers had either encouraged them or discouraged them from talking about the event with other witnesses. Chapter 3: Witnesses Talk 61

Procedure Respondents were given the Screening Questionnaire at the end of a psychology class, as part of a battery of questionnaires. They were under no obligation to complete the questionnaires, however those who did were awarded credit toward their psychology class. The experimenter contacted respondents who were deemed eligible for the Follow-up Questionnaire (based on their screening questionnaire responses and indicated willingness to participate in further research) by phone or e-mail. Respondents were informed that they were invited to take part in a psychology study investigating the experiences of eyewitnesses. Respondents who agreed to complete the study were asked to pick up the questionnaire and complete it at their own convenience. Respondents were instructed to return the completed questionnaire to the experimenter who would award them credit toward their psychology class. Although the respondents were required to provide their name in order to receive credit, the questionnaires themselves were anonymous. After returning their questionnaires, respondents were fully debriefed about the purpose of the study. Contact information for a local victim support agency was also made available. To the best of our knowledge, none of the respondents made use of this service.

Results

Screening Questionnaire Statistical analysis of the Screening Questionnaire focused on three questions. First, what percentage of the respondents had witnessed a serious event? Second, what were the most frequently witnessed events? Third, what proportion of eyewitnesses reported the presence of at least one other co-witness? The majority of the respondents to the Screening Questionnaire (75%) reported that they had witnessed at least one serious event. The average number of years since witnessing the event was 3.43 years (SD = 3.81). As can be seen on Table 3.1, the most frequently witnessed event was physical assault (47%), followed by property vandalism (38%) and events causing serious injury/death (23%). Manslaughter (2%), abduction (1%) and homicide (1%) were the least frequently witnessed events. Chapter 3: Witnesses Talk 62

Table 3.1 Type of Event Witnessed by Screening Questionnaire Respondents

Number of % of respondents who Type of Event respondents who have witnessed event witnessed event Physical assault 360 47 Property vandalism 297 38 Event causing serious injury/death 175 23 Robbery 143 18 Breaking and entering 107 14 Motor vehicle theft 89 12 Sexual assault 81 10 Fraud, blackmail, or extortion 60 8 Arson 49 6 Other 22 3 Attempted murder 18 2 Manslaughter 16 2 Abduction 10 1 Homicide 8 1 Note. N=773; some respondents reported witnessing more than one event. A total of 1435 events were reported with a mean of 1.85 (SD = 1.83) events reported per respondent.

The majority of those who witnessed an event (86%) reported that there was at least one other witness at the event. On average, there were 6.77 (SD = 9.66) other witnesses present (median = 3; mode = 0; range = 0-100). The very large standard deviation relative to the mean suggests a non-normal distribution and this is confirmed by Skewness and Kurtosis values of 3.91 and 24.58 respectively. A frequency distribution of this data is shown in Table 3.2.

Chapter 3: Witnesses Talk 63

Table 3.2 Number of Co-witnesses Reported by Screening Questionnaire Respondents

Number of co-witnesses Frequency of times reported Percent of times reported 0 69 16 1 52 12 2 62 14 3-5 104 23 6-10 94 21 11 or more 64 14 Note. N=445. N is lower than expected because several respondents did not complete the questionnaire in its entirety. When respondents indicated a range (e.g., 1-2 co-witnesses), the lower number was used for this classification.

In order to determine whether the number of co-witnesses varies with the seriousness of the event, comparisons were made between the total sample of respondents and those who indicated they were interviewed by the police or who testified in court. It was found that respondents who indicated they were interviewed by the police or who testified in court reported a similar number of co-witnesses to that reported by the sample of all respondents. Of the 91 respondents who reported being interviewed by the police, 80% stated that there was at least one other witness present at the event. On average, they reported 3.83 other witnesses being present (SD = 5.33). Of the 9 respondents who reported testifying in court, 78% stated that there were co-witnesses present, with an average of 1.22 other witnesses present at the event (SD = 0.44).

Follow-up Questionnaire Analyses of the Follow-up Questionnaire concentrated on the following questions: First, what were the most frequently witnessed events? Second, with whom did witnesses report discussing the event? In particular, what proportion of eyewitnesses report discussing the event with co-witnesses? Third, what were the primary motivations for their discussions? Finally, did legal authorities either encourage or discourage witnesses from discussing the event with one another? Chapter 3: Witnesses Talk 64

As can be seen in Table 3.3, the most frequently witnessed events were physical assault (30%), event causing serious injury or death (27%), and robbery (18%). These percentages broadly reflect those of the sample who completed the Screening Questionnaire. In the follow-up questionnaire, there were no witnesses to abduction, manslaughter, or homicide. The average number of years since witnessing the event was 3.31 years (SD = 4.02).

Table 3.3 Type of Event Witnessed by Follow-up Questionnaire Respondents, and Percent who Discussed Event with Co-Witnesses

Number (and percent) Number (and percent) Type of Event of respondents who who discussed event witnessed the event with co-witnesses Physical assault 18 (30%) 16 (89%) Event causing serious 16 (27%) 13 (81%) injury/death Robbery 11 (18%) 10 (91%) Other 5 (8%) 3 (60%) Sexual assault 2 (3%) 2 (100%) Arson 2 (3%) 2 (100%) Motor vehicle theft 2 (3%) 1 (50%) Breaking and entering 1 (2%) 1 (100%) Property vandalism 1 (2%) 1 (100%) Fraud, blackmail, or 1 (2%) 1 (100%) Extortion Attempted murder 1 (2%) 1 (100%) Abduction 0 (0%) n/a Manslaughter 0 (0%) n/a Homicide 0 (0%) n/a Note. N=60. Each respondent described only one event.

The average level of fear for personal safety reported by the respondents was 3.4 (SD = 1.90) on a scale of 1-6 (1 = Not at all; 6 = Very strongly). The average level of impact the event had on the participant’s life was 3.17 (SD = 1.38) on a scale of 1-6 (1 = No influence; 6 = Extreme). When asked about their involvement in the event, 18 (30%) of Chapter 3: Witnesses Talk 65

the respondents indicated that they were victims of the event and 42 (70%) of the respondents indicated that they witnessed the event, but were not victims of it. Respondents were asked when they had first spoken about the event. The majority of the 60 respondents (63%) reported that they spoke about it “immediately.” The second most common answer selected by 22% of the respondents was “later the same day.” Only 7% of the respondents indicated that they had never spoken about the event. Of the respondents, 28% reported that they actively avoided talking about the event with others at some point. The majority of the 60 respondents (86%) reported discussing the event with a co- witness. In order to ascertain whether this discussion pattern is typical of respondents who were more closely affected by the event (as opposed to bystanders who were the majority of respondents), a series of chi-square analyses were conducted. There was no association between the level of fear for personal safety reported (classified as either high, ≥4, or low, <4) and whether or not the respondent reported discussing the event with a co-witness (χ2 (1, N = 58) = 0.43, p = .51). Similarly, there was no association between the reported impact of the event (impact classified as either high, ≥4, or low, <4) and whether or not the co-witnesses discussed the event (χ2 (1, N = 58) = 0.28, p = .59). No association was found between witness/victim status of the respondents and whether or not they discussed the event (χ2 (1, N = 58) = 0.08, p = .77). This suggests that victims and bystanders were equally likely to discuss the event with their co-witness. Finally, no association was found between whether witnesses had been interviewed by the police and whether they discussed the event (χ2 (1, N = 57) = 0.33, p = .57). This suggests that those respondents who were interviewed by the police and those not interviewed were equally likely to discuss the event with their co-witnesses. Table 3.4 indicates how frequently respondents reported discussing the event with various individuals or groups at three different time periods: 24 hours, one week, and six months after the event. Respondents were asked to report how frequently they discussed the event on an 8-point rating scale (1 = Never, 2 = Once or twice, 3 = Infrequently, 4 = Occasionally, 5 = Often, 6 = Frequently, 7 = Very Frequently, 8 = Almost continuously). Respondents most commonly reported discussing the event with friends, family, and other Chapter 3: Witnesses Talk 66

witnesses. Up to 24 hours after the event, respondents, on average, reported discussing the event with other co-witnesses “Infrequently” (M = 3.0, SD = 2.2). One week after the event, respondents reported talking about the event with a co-witness “Once or twice” on average (M = 1.8, SD = 1.4). Six months after the event, respondents reported talking about the event with co-witnesses almost “Never” (M = 1.3, SD = 0.62).

Table 3.4 Mean Reported Frequency of Discussion with Others

Time of discussion Discussed event with: 24 hours after One week after 6 months after event (n = 53) event (n = 53) event (n = 51) Other Witnesses 3.0 (2.2) 1.8 (1.4) 1.3 (0.62) Friends 4.0 (2.4) 3.0 (1.7) 2.0 (1.1) Family 3.5 (2.2) 2.2 (1.4) 1.4 (0.83) Police Officers 1.3 (0.64) 1.1 (0.43) 1.0 (0.14) Lawyer 1.0 (0) 1.0 (0.0) 1.0 (0.0) Religious Authority 1.0 (0.19) 1.0 (0.19) 1.0 (0.14) Stranger 1.3 (1.0) 1.1 (0.23) 1.1 (0.27) Medical Doctor 1.3 (1.0) 1.1 (0.41) 1.0 (0.28) Psychologist 1 (0) 1.0 (0.14) 1.2 (0.63) Other 1.1 (0.83) 1.1 (0.55) 1.0 (0) Note. Scores range from 1 (never) to 8 (almost continuously). Standard deviations are in brackets.

Table 3.5 indicates the reasons respondents gave for discussing the event with co- witnesses. Respondents were asked to indicate which of the following purposes applied to their conversations (emotional support, advice-seeking, providing information, story- telling) or state an alternative purpose. Up to 24 hours after the event, the most frequent response, given by 44% of respondents, was “providing information.” One week after the event, this number fell to 12% and six months later it fell to 8%.

Chapter 3: Witnesses Talk 67

Table 3.5 Reasons stated for discussing event with co-witnesses

Time of discussion Purpose of Discussion 24 hours after One week after 6 months after event (n = 34) event (n = 17) event (n = 12) Emotional Support 14 (41) 8 (47) 3 (25) Advice-Seeking 6 (18) 3 (18) 3(25) Providing Information 15 (44) 2 (12) 1 (8) Story-Telling 7 (21) 5 (29) 3 (25) Other Purpose 6 (18) 3 (18) 2 (17) Note. The values represent the number (and percentage) of respondents who reported having discussions for the listed reasons. The numbers vary according to how many respondents reported having discussions at the various time intervals. Respondents may have stated more than one reason for the discussions.

Twenty-one respondents reported that there was a police investigation of the event. Of these, 24% reported that the police had encouraged them to discuss the event with other witnesses. Respondents were asked to elaborate on how the police had encouraged discussion, and one participant reported that it was explained to her that, “It may help us to cope if we discussed it.” Another participant stated, “we all talked to attending police in a group” and a third participant stated that she had been encouraged to discuss the event, “to confirm what had happened (if what I saw was what they saw etc).” When asked whether the police had ever discouraged them from discussing the event with other witnesses, 14% reported that they had been discouraged. Respondents were asked to describe why the police had discouraged them, and one reported that talking about the event could “obscure facts from our view” and another reported that it could “contaminate the story in our heads.” Ten respondents reported that there were legal proceedings following the event. None of those respondents reported that they had ever been encouraged to discuss the event with other witnesses during the legal proceedings. Two respondents reported that they had been discouraged from discussing the event with other witnesses during the legal proceedings. Chapter 3: Witnesses Talk 68

Discussion

Screening Questionnaire Results from the Screening Questionnaire suggest that the majority of the undergraduate population sampled had witnessed a serious event at some point in their lives. Furthermore, the majority of those who had witnessed a serious event reported that there was at least one other witness present (Median = 3), with only 14% reporting that they were the sole witness. Even if we examine only the 91 eyewitnesses in our study who were interviewed by the police, only 20% stated that they were sole witnesses. This finding is compatible with a study by Wright and McDaid (1996) who found that only approximately 17% of police lineups involved a sole witness. Indeed, this may overestimate the number sole witnesses because the police do not necessarily invite all witnesses to attend lineups. However, these results differ slightly from the conclusions drawn by Yuille and Tollestrup (1992) who state, “For the crimes directly witnessed (robbery, sexual assault, and nonsexual assaults), 71% of the cases involved the victim as the only witness. Thus, the modal witness to a crime of violence is not a passive observer but rather a victim of that crime” (p. 205). While it is also true for the current study that the modal number of co- witnesses was zero, in the current study the majority of witnesses had one or more co- witnesses. One possible explanation for the apparent discrepancy between the data of the current study and that of Yuille and Tollestrup (1992) is that the witnesses were observing different types of events. However, this does not appear to be the case, as a comparison of the percentage of cases witnessed (as shown in Table 1) with the figures reported by Yuille and Tollestrup suggest that the groups of witnesses were remarkably similar (e.g., for the current study and Yuille and Tollestrup’s study respectively: Non-sexual (Physical) assault: 47%, 59%; Sexual assault: 11%, 15%; Robbery 19%; 24%). Another possible explanation for the discrepancy is that although Yuille and Tollestrup’s (1992) results suggest that the typical crime has only one witness (the victim), this is different from their conclusion that the typical witness is the only witness to a crime. For example, in the Yuille and Tollestrup study, six homicides were investigated for which four had no witnesses, one had a single witness, and one had sixteen witnesses. Whereas the typical homicide (83%) in Yuille and Tollestrup’s study may not have had multiple Chapter 3: Witnesses Talk 69

witnesses, the typical witness (94%) had many co-witnesses. Therefore, the typical experience of witnesses is to be in the presence of co-witnesses. This finding has implications for the way in which research on eyewitnesses should be conducted. Whereas Yuille and Tollestrup argue that the typical witness is a victim and the sole witness, and therefore unlike the typical laboratory participant, this may not be the case. It may be that the ecological validity of laboratory experiments is not as low as Yuille and Tollestrup would suggest and therefore the results of laboratory experiments may be more readily generalised to the forensic setting. The finding that the typical witness has co-witnesses may have important implications for eyewitness memory. Vicario and Tomat (1998) found that if a witness knows that other witnesses are being questioned, the accuracy of their testimony will be improved. In their study, participants were shown a short film either individually or in groups of two. Participants who viewed the scene individually also gave their reports individually. However, those who viewed the event in pairs simultaneously gave their testimonies to two different experimenters in two adjacent rooms separated by a glass wall. Participants were not allowed to discuss the event with the other witness. It was found that the reports given by the participants in pairs were more reliable and contained fewer false elements than those given by individual participants. Vicario and Tomat interpret these findings by suggesting that participants who are aware that their reports could be compared with that of another are more cautious and avoid inserting doubtful elements. This means that even if a real witness does not receive any co-witness information, the fact that he/she knows of the existence of a co-witness may change testimonial performance.

Follow-up Questionnaire The Follow-up Questionnaire surveyed only respondents who reported having witnessed a serious event for which there was at least one other witness present. The results from this questionnaire show that when there is a co-witness present, the vast majority of respondents discuss the event with the co-witness. Furthermore, the most frequently stated reason for discussing the event with a co-witness was “providing information” about the event. This could be problematic if the witnesses are contaminating one another’s memories of the event. Research has shown that errant information presented through Chapter 3: Witnesses Talk 70

discussion with a co-witness may cause witnesses to incorporate this misinformation into their memories (e.g., Gabbert, Memon, Allan, 2003; Gabbert, Memon, Allan, & Wright, 2004; Hoffman, Granhag, See, & Loftus, 2001; Hollin & Clifford, 1983; Shaw et al., 1997; Wright, Self & Justice, 2000). However, other research suggests that group discussion prior to individual testimony can be beneficial to memory (Basden, Basden, & Henry, 2000; Stephenson, Abrams, Wagner, & Wade, 1986; Warnick & Sanders, 1980; Weldon & Bellinger, 1997; Yuker, 1955). Because the conclusions drawn from past literature are inconsistent, it is important that co-witness discussion be studied in more detail. Furthermore, relatively little is known about the mechanisms responsible for memory conformity and whether it is possible to reduce any negative effects of co-witness discussion. For these reasons, it is vital that the effects of co-witness discussion are investigated more thoroughly. In order to decrease the amount of co-witness discussion, it may be beneficial for police officers to instruct witnesses not to discuss details of the event with one another. In the jurisdiction where the survey was conducted, guidelines exist for identification parades, which recommend separating co-witnesses. However, there are no similar guidelines known for interviewing eyewitnesses. Only 14% of the respondents in this questionnaire reported that the police had discouraged them from discussing the event with other witnesses. Furthermore, it is notable that a larger percentage of the respondents (24%) reported that the police had encouraged them to discuss the event with other witnesses. Whilst care should be taken when interpreting these results as the sample size of those who were interviewed by the police is relatively small (n = 21), this finding provides some justification for further investigation regarding the opinions that police officers hold about co-witness discussion. This study also suggests that lawyers are more likely to warn witnesses not to discuss than are police officers. When the witnesses were asked whether they had ever been discouraged from discussing the event with other witnesses during the legal proceedings, 20% reported that they had been discouraged. None of the respondents reported that they had ever been encouraged to discuss the event with other witnesses during the legal proceedings. However, a lawyer may not talk to a witness until months after the event, and even police officers may not arrive for several minutes or even on the same Chapter 3: Witnesses Talk 71

day, depending on the location and severity of the event. Therefore even if a warning is given and even if the eyewitnesses follow the advice contained in the warning, it may be too late to prevent co-witness contamination of memory from occurring. This study is limited by the fact that a relatively homogeneous student sample was used. Students may be more likely to witness certain crimes and more likely to talk about them than the general population. Furthermore, the survey relies on self-report and respondents were self-selected. Only those who were willing to complete the survey did so, and the possibility exists that the respondents may not be representative of all eyewitnesses. For example, it may be the case that the respondents who volunteered to answer questions about their experiences were the ones who were also more likely to discuss the events with co-witnesses. An additional limitation of this study is that the number of respondents who were interviewed by the police and/or a lawyer is relatively small. The fact that not many of the respondents were interviewed by the police may indicate that they were peripheral to the event. The possibility exists that more distant, bystander witnesses may be more numerous and may behave differently from those more closely involved in the event. However, our research suggests that this is probably not the case because witnesses who were interviewed by the police and who testified in court were equally likely to discuss the event with a co-witness to those who had not been interviewed by legal authorities. Furthermore, self-report measures of fear for personal safety, involvement in the event, and impact of event on life were not significantly associated with whether or not witnesses reported engaging in co-witness discussion, further suggesting that peripheral and central witnesses do not differ in this respect. While this survey has some limitations, it does provide valuable insights into the discussion behaviour of real eyewitnesses. This survey has revealed that the majority of respondents had witnessed a serious event at some point during their lives, that most eyewitnesses had at least one other witness present at the event, and that they commonly discussed details of the event with their co-witnesses. These findings provide a justification for the need to undertake further research on the effects of co-witness discussion on eyewitness memory, and suggest that the laboratory research investigating the non-victim witness may be a valid methodology for studying this issue. 72

Chapter 4 Study 2: The Legal Perspective

What a court wants is the independent recollection of the individual witness. It does not want a number of observers to get together and agree among themselves upon a common version…The Court of Appeal must grasp the nettle by taking an appropriate opportunity to condemn uncompromisingly the insidious and dangerous practice of … collaboration. (Heaton-Armstrong, 1987, p. 472).

The results reported in the previous chapter suggest that witnesses commonly discuss an event with one another, however within the legal system the ideal is that testimony given by eyewitnesses should be independent of one another. For example, the legal concept of ‘hearsay’ embodies the notion that “a witness’s assertions of relevant facts should be based upon his or her own experiences” (Forbes, 2003, p. 59) and not those of another. Legal procedures are made with the presumption that dangers arise when witnesses discuss the event with one another. These dangers include collusion and the tainting of one witness’s recollection by something he/she has heard another witness say (Heaton- Armstrong, 1985, 1987). Therefore, it is standard legal procedure to attempt to prevent eyewitnesses from discussing the crime with other observers of the incident. For example, witnesses are often prohibited from hearing the testimony of other witnesses during a trial. Furthermore, judges routinely instruct witnesses not to discuss the event with other witnesses, and lawyers will often question the witnesses regarding whether or not they have discussed the incident with others (Yarmey & Morris, 1998). A recent set of guidelines on interviewing has been written for police in the United States (United States National Institute of Justice, 1999). In the guidelines, there are several instructions suggesting that witnesses should be separated and instructed not to discuss the event with each other. There are similar guidelines in the United Kingdom (Home Office, 2003, p. 131), which recommend that, “Where there is more than one witness, every effort should be made to keep them separate.” In at least one state in Australia, police officers are instructed not to allow any communication between witnesses and to keep all witnesses Chapter 4: Legal Perspectives 73

separate before and after an identification parade (Field Operations, 1997). It should be noted that in both Australia and the United Kingdom these guidelines apply only to identification parades, and there appear to be no formal policies or guidelines about separating co-witnesses when obtaining eyewitness statements. In a recent high profile incident, attempts were made by legal authorities to discourage co-witness discussion in the United States (Morello & Davis, 2002). After a series of sniper attacks in the Washington area, Montgomery County Police Captain, Nancy Demme, was reported to have warned the population that witnesses should not contaminate their memories by comparing notes with other witnesses or reporters. (However, ironically the same article that published these warnings proceeded to give descriptions given by witnesses about the appearance of the sniper, thus providing indirect co-witness information to the public). Despite the fact that courts appear to prefer witnesses not to discuss the event with each other, it is unclear whether police officers commonly discourage such discussion. In Study 1, only a small percentage of the eyewitnesses surveyed reported that they had been discouraged by the police from discussing the event with other eyewitnesses. Furthermore, it is notable that a larger percentage of the participants reported that the police had encouraged co-witness discussion. As part of a different research project, the author of this thesis was able to observe police officers at work. During approximately 300 hours of observation, police officers were often seen to question several witnesses simultaneously and in front of one another when they arrived at the scene of an incident. However, no instances were observed of police officers encouraging or discouraging co-witness discussion. Unfortunately, for reasons of confidentiality, it is not possible to further describe details of these observations. With exception of the information obtained in Study 1 there has been no research investigating the practices of police officers with regards to co-witness discussion. The purpose of Study 2 was to investigate the opinions police officers hold regarding co-witness discussion and to find out about the practices in which officers engage. In particular, the main aim of Study 2 was to identify whether police officers are aware of the prevalence of co-witness discussion and to assess their perception of the impact it has on eyewitness memory. Specifically, the survey was designed to answer to following research questions: Chapter 4: Legal Perspectives 74

1) How frequently do police officers believe co-witnesses discuss the event with each other? 2) What opinions do police offers hold regarding co-witness discussion? 3) What instructions have police officers received regarding co-witness discussion and are these instructions practical? 4) What type of practices do police officers engage in when dealing with multiple witnesses? 5) How does rank and experience affect the opinions and practices of police officers?

Method

Participants A questionnaire was sent electronically to approximately 14,500 police officers. For reasons of confidentiality, the police organisation involved cannot be identified. Only 145 police officers (112 male, 31 female3) responded to the questionnaire, which means there was a relatively low response rate of approximately 1%. Participants received no incentives for completing the survey and their participation was completely voluntary. Anonymity was guaranteed. The respondents worked in rural (n = 48) and metropolitan areas (n = 71) as well as special units (n = 18). The average age of the officers was 34.7 years (SD = 8.19) and the average length of service was 11.5 years (SD = 8.62). The questionnaire was completed by 102 constables (probationary constables, constables, senior constables, leading senior constables) and 36 more senior officers (sergeants, senior sergeants, inspectors, chief inspectors). Despite the low response rate, the respondents formed a highly representative sample of the total population of the police force. Table 4.1 compares the percentages of survey respondents with the total police population surveyed. From Table 4.1, it can be

3 For this variable and others, numbers may not add up to the total number of respondents (N =145) because some respondents did not fill out all of their demographic information. Chapter 4: Legal Perspectives 75

seen that the sample used in this study is reasonably representative, with the exception of respondents being of slightly higher rank than the total population.

Table 4.1 Comparison of Survey Respondents with Total Police Population Surveyed

Demographic Information Respondent Police Sample Population Sex Male 78% 72% Female 22% 28% Region Metropolitan 52% 52% Rural 35% 35% Specialist 13% 13% Rank Constables 74% 80% More Senior 26% 20% Age in years Up to 25 13% 12% 26-30 23% 21% 31-35 24% 24% 36-40 15% 18% 41-45 13% 12% 46-50 9% 7% Over 50 3% 6% Years of 0-5 33% 31% Service 6-10 22% 15% 11-15 19% 26% 16-20 9% 11% 21-25 11% 8% Over 26 7% 9%

Materials A questionnaire was developed for this study to investigate police officers’ opinions regarding discussion between co-witnesses (see Appendix C). The questionnaire comprised of three main sections. First, respondents were asked to report basic demographic information regarding their age, sex, present rank, jurisdiction, and number of years they had worked for the police. Chapter 4: Legal Perspectives 76

Second, respondents answered open-ended questions about events witnessed by more than one person. For example, respondents were asked their opinion about discussion between witnesses, whether they had ever encouraged or discouraged co-witness discussion and if so, their reasons for doing so. Furthermore, respondents were asked if they had received any instructions regarding discussion between witnesses and if so, whether these instructions were practical. Respondents were also asked what topics eyewitnesses should and should not discuss with each other. Third, respondents were asked to answer a series of questions using rating scales. Respondents were asked to state their opinion of co-witness discussion on a 6-point scale (1 = Discussion can be beneficial to witness memory; 6 = Discussion can be detrimental to witness memory). Additionally, respondents were asked to estimate the percentage of eyewitnesses who have discussed details about the event with one another prior to the arrival of the police on a 10-point scale (ranging from 10%-100%). Respondents were also asked to indicate on a 5-point scale (0 = Definitely not; 1 = Maybe not; 2 = Not sure; 3 = Maybe; 4 = Definitely) if discussion between eyewitnesses could help refresh their memories and if discussion between eyewitnesses could distort their memories. Finally, respondents were asked how often they engaged in a series of practices. For example, respondents were asked if they had ever physically separated eyewitnesses at the scene of the incident, conveyed information they had learned from one eyewitness to another, or asked whether witnesses had discussed the event with other witnesses. All these items were answered using a 5-point scale (0 = Almost never; 1 = Not often; 2 = Often; 3 = Very often; 4 = Almost always).

Procedure All officers in a police organisation were e-mailed a copy of the questionnaire by the Head of School of Operational Policing. Officers were informed in writing that they were under no obligation to complete the questionnaire, and that their consent was completely voluntary. Officers were told that there were no right or wrong answers and were encouraged to answer all questions honestly. Furthermore, they were informed that all of the information they provided would remain strictly confidential and that all documents would be anonymous. Chapter 4: Legal Perspectives 77

Officers were given one month to complete the questionnaires. Respondents could return the completed questionnaire to the researchers by e-mail. Although it is possible for the respondents to be identified by their e-mail address, the questionnaires themselves were anonymous. When a questionnaire was received, it was printed and the e-mail was destroyed to ensure anonymity. Alternatively, if the respondents were concerned about their confidentiality, they had the option of printing the completed form and posting it to the researchers.

Results First, the data from all respondents were examined using descriptive statistics. Second, a comparison was made between constables and more senior officers to examine the effect of rank on responses to the questionnaire. Finally, a comparison was made to examine the effect of years of experience in the police force on questionnaire responses.

Questionnaire Responses from All Police Officers Sampled When there are multiple witnesses to a serious event, police officers estimated that on average 68% (SD = 21.9) of witnesses talk about the event with one another prior to the arrival of the police. When asked to indicate their opinion regarding discussion between co-witnesses on a 6-point scale (1 = Discussion can be beneficial to witness memory; 6 = Discussion can be detrimental to witness memory) the average response given was 4.12 (SD =1.57), suggesting that officers believe that discussion is marginally more detrimental than beneficial to witness memory. Responses to this question were widely distributed across all possible answers as shown in Table 4.2.

Chapter 4: Legal Perspectives 78

Table 4.2 Responses to the Question, “What is your opinion about discussion between witnesses?”

Discussion Discussion can be can be beneficial detrimental Question to to eyewitness eyewitness memory memory 1 2 3 4 5 6 What is your opinion about discussion 8% 10% 19% 14% 25% 24% between witnesses?

When asked to explain their opinions regarding discussion between co-witnesses using an open-ended question, the responses were varied, ranging from support for discussion between witnesses to condemnation of it (see Tables 4.3 and 4.4 for examples of responses). All qualitative responses were coded by one person, who reviewed all of the open-ended answers given and developed a series of categories representing items which seemed to go together. Once these categories were developed, the coder reviewed all of the answers once again and counted the number of times each category was mentioned.

Chapter 4: Legal Perspectives 79

Table 4.3 Responses to the Open-Ended Question, “What is your opinion about discussion between witnesses?” Indicating that Discussion Can Be Beneficial

# of times What is your opinion about discussion between co-witnesses? mentioned Discussion - Discussion can prompt eyewitnesses to recall details they 40 previously did not remember can be - Can help reinforce witness memory 20 beneficial - Discussion helps recovery from trauma 13 - Discussion helps recovery after speaking to the police 8 - Discussion helps recovery after all court matters are finalized 3 - You obtain a more accurate overall picture of the event. 9 - Police witnesses discuss evidence before making a statement to ensure they cover all the evidence and get 6 the facts in order - After gaining an individual account I would ask that they discuss it and let me know of anything else they recall 2 later - Situations in which a witness may feel obliged to say they saw something they in fact didn’t isn’t usually a 2 problem - Witness versions, unless uniform, are a great impedance 1 on a criminal case - Witnesses start to doubt what they think they saw if 1 others aren’t around to confirm it - With strong witnesses it can be good 1 - Multiple offenders can discuss their versions therefore it 1 is only fair that witnesses be afforded the same privilege

Chapter 4: Legal Perspectives 80

Table 4.4 Responses to the Open-Ended Question, “What is your opinion about discussion between witnesses?” Indicating that Discussion Can Be Detrimental

# of times What is your opinion about discussion between co-witnesses? mentioned Discussion - Independent recollection may be contaminated by other 84 witnesses’ perceptions can be - According to the courts it is detrimental so it may weaken 48 detrimental the prosecution case. - Stronger witnesses may lead weaker ones thus altering 34 what people thought they saw - Discussion could cause an eyewitnesses to believe they 22 saw something that they actually did not - Witnesses may become unsure of themselves if their 21 memory differs from another’s - Different witnesses see and hear different things. Discussion leads to reports of ‘common things’ seen and 18 heard. - The problem of collusion 13 - Differences between witness statements adds to their 12 credibility - Discussion causes loss of evidence/overall picture 12 - Discussion can cloud testimony 6 - At worst, it can misdirect a whole investigation based on 1 false evidence - Discussion can exhaust memory detail by the time police 1 begin to obtain a version

Respondents were asked if they thought discussion between eyewitnesses could help refresh eyewitness memory. Officers responded using a 5-point scale (0 = Definitely not; 1 = Maybe not; 2 = Not sure; 3 = Maybe; 4 = Definitely). The average response given by police officers was 2.75 (SD = 1.23). Using the same 5-point scale, respondents were also asked if discussion between eyewitnesses could distort their memories. The average response given by police officers for this question was 3.43 (SD = 0.88). Table 4.5 shows a frequency distribution of the responses given for these two questions.

Chapter 4: Legal Perspectives 81

Table 4.5 Frequency Distribution of Responses Given for the Questions, “Can discussion between eyewitnesses help refresh their memories” and “Can discussion between eyewitnesses distort their memories?”

Definitely Maybe Not Maybe Definitely M Question Not Not sure (SD) 0 1 2 3 4 Can discussion between 2.75 eyewitnesses help 10% 6% 13% 42% 29% (1.23) refresh their memories? Can discussion between 3.43 2% 4% 3% 33% 58% eyewitnesses distort (0.88) their memories?

The majority of police officers (74%) reported they had received instructions regarding discussion between witnesses. The types of instructions they reported receiving are summarised in Table 4.6.

Chapter 4: Legal Perspectives 82

Table 4.6 Types of Instructions that Police Officers Receive Regarding Co-Witness Discussion

Number of Describe the instructions you have been given regarding discussion times between eyewitnesses mentioned - Prevent discussion 23 - Prevent discussion until after the statement has been obtained 2 - Prevent discussion by encouraging witnesses not to talk 30 - Prevent discussion by separating witnesses 36 - Prevent discussion for eyewitness identification 2 - As directed by magistrate in court when only some evidence has 5 been given - Take statement independently 4 - Take statement independently in different rooms/ends of rooms 2 - Tell witnesses to write down what they remember 2 - Refrain from encouraging discussion 1 - For traffic incidents, each driver should be allowed to hear the 1 version supplied by the other driver - Don’t let a parent speak for the child 1

Police officers were not explicitly asked to state where they received instructions from regarding co-witness information, however some police officers volunteered this information. Their responses are summarised in Table 4.7.

Chapter 4: Legal Perspectives 83

Table 4.7 Sources of Information Regarding Co-witness Discussion Instructions

Number of Comments regarding source of instructions times mentioned - From senior officers 10 - Training (type of training unspecified) 4 - Detective Training Course 7 - Investigative training 4 - Initial training at the academy 3 - ID evidence training 1 - Bachelor of Policing (Investigations) 1 - Extensive discussions on various course and training activities 1 - At court 6 - Tertiary studies 1 - Own experience 1 - Law notes 1 - Mostly self-education (e.g. university) 1 - Commonsense 1

Respondents were asked to indicate whether the instructions they received regarding co-witnesses were practical, using a 5-point scale (0 = Almost never; 1 = Not often; 2 = Often; 3 = Very often; 4 = Almost always). The average response given to this question was 2.75 (SD = 1.28). Responses to the open-ended question, “Are these instructions practical?” are shown in Tables 4.8 and 4.9.

Chapter 4: Legal Perspectives 84

Table 4.8 Negative Responses to the Open-ended Question, “Are these instructions practical?”

Number of Are these instructions practical? times mentioned No - Witnesses will have time to speak prior to the arrival of the 17 police - It is almost impossible to prevent the discussion from occurring as it is natural for witnesses to want to discuss 13 - As above especially if witnesses know each other 2 - As above especially at a later time when police not around 1 - There is sometimes a lack of facilities/space for separating 6 witnesses - Sometimes there are not enough police and it can be difficult 4 to make one person stand alone while other witnesses are spoken to - Debriefing is necessary after a traumatic incident and it’s 3 good to do that with someone else who knows or was there - Often the number of witnesses is greater than the number of police so one police officer will take the details of multiple 2 witnesses at the same time - Separation may inhibit free recall of events 1 - At the station whilst waiting for a statement to be obtained, 1 the witnesses may have to wait together - It is very difficult to stop witnesses from talking as there are crucial tasks which must be performed by police, like 1 preserving the crime scene, which are more important than stopping witness discussion - Barristers use discrepancies between witness evidence to their 1 advantage - There is no formal procedure to separate and interview 1 witnesses - After a statement has been obtained there is no instruction 1 given as to whether they can talk about it or not - Separating children and parents isn’t always a good idea 1

Chapter 4: Legal Perspectives 85

Table 4.9 Affirmative Responses to the Open-ended Question, “Are these instructions practical?”

Number of times Are these instructions practical? mentione d Yes - Witnesses are happy to comply when the reasons are explained 5 - If police attend the crime soon after it has been committed, 2 witnesses can be kept from talking to one another - It’s not hard to stop people talking about the incident and 2 separate them - There are very few occasions when there are so many witnesses that they cannot be separated prior to speaking to the 1 investigator - Usually numerous police assist 1 - Asking witnesses to restrict conversation to matters other than 1 the incident is not asking too much - Even if the witnesses ignore what they are told and do talk about it, hopefully they will have the sense to deny it if asked 1 about it while giving evidence - Most witnesses to serious crimes almost always attend a police 1 station to have a version of events recorded

Police officers were asked how often they engaged in a series of practices regarding co-witnesses. All of the items were answered using the same 5-point scale (0 = Almost never; 1 = Not often; 2 = Often; 3 = Very often; 4 = Almost always). Table 4.10 gives a frequency distribution of the responses given to the questions.

Chapter 4: Legal Perspectives 86

Table 4.10 Frequency Distributions of How Often Police Officers Engage in a Series of Practices Regarding Co-Witnesses.

Almost Not Often Very Almost M Question Never Often Often Always (SD) 0 1 2 3 4 Do you ever encourage co- 0.21 86% 10% 3% 0% 1% witness discussion? (0.62) Do you ever discourage co- 3.26 5% 7% 6% 21% 61% witness discussion? (1.15) Do you physically separate 3.14 2% 8% 14% 26% 50% witnesses? (1.07) Do you instruct witnesses 3.23 not to discuss before taking 4% 9% 10% 15% 62% (1.16) statements? Do you instruct witnesses 1.65 not to discuss after taking 29% 29% 12% 8% 22% (1.51) statements? Do you take statements of 0.28 one witness in front of 87% 6% 2% 1% 4% (0.86) others? Do you ask witnesses if they 2.43 10% 19% 23% 13% 35% have discussed the event? (1.40) Do you convey information from one eyewitness to 0.29 75% 22% 2% 0% 1% another and tell them the (0.58) source? Do you convey information from one eyewitness to 0.61 54% 34% 1% 8% 4% another without telling them (0.76) the source? Note. If participants gave two or more responses to the same item, then the lower number was accepted.

Respondents were asked to provide examples of topics that witnesses should discuss and those that they should not discuss with one another. The majority of respondents (85%) provided examples of topics that eyewitnesses should not discuss with one another, however less than half of the respondents (45%) provided examples of topics that eyewitnesses should discuss with one another. Topics which eyewitness should and should not discuss are summarised in Table 4.11. Chapter 4: Legal Perspectives 87

Table 4.11 Topics Which Eyewitnesses Should and Should Not Discuss with One Another

Number of times mentioned Topic of Discussion Should Should Not Discuss Discuss - Their feelings/welfare after the traumatic event 20 -- - Their feelings if police officer present to stop discussion from heading into details of the incident 2 -- - Their feelings after court proceedings conclude 1 -- - Their feelings after police statements 7 -- - What they saw 1 24 - Time frames/dates 10 1 - Descriptions of offender/ID evidence 4 21 - Description of vehicle 3 2 - Weather 3 1 - Exact words in a conversation 2 10 - Certain basic descriptions 2 -- - Names of streets/location 1 -- - Where they were standing 1 -- - Sequence of events 1 1 - Actions of offenders -- 6 - Descriptions of weapons -- 1 - Speed -- 1 - The incident 3 32 - The incident when witnesses are independent and have no involvement in the matter 1 -- - The incident after getting individual accounts 2 -- - The incident especially if proceeding to court -- 9 - The incident before statements taken by police -- 3 - Anything other than the incident (e.g. football) 3 -- - The need to stand together and not allow fear or 2 -- intimidation to prevent them from testifying - Support services available 3 -- - Everything 3 13 - Court procedures 1 -- Police should discuss when getting to the facts of the 1 -- matter - Supposition 1 -- - The property that was stolen 1 -- - Any conversations or actions that have not been heard -- 3 by all the participants Chapter 4: Legal Perspectives 88

Number of times mentioned Topic of Discussion (continued) Should Should Not Discuss Discuss - Any matter when the witness and suspect know each -- 2 other - Domestic details -- 2 - How the police conduct their investigations -- 1 - Hearsay evidence -- 1 - Compensation -- 1 Note. Dashes indicate that the topic was not mentioned.

The Effect of Rank on Questionnaire Responses A comparison was made between the responses of 81 Constables (probationary constables, constables, senior constables, leading senior constables) and 62 more Senior Officers (detectives, sergeants, senior sergeants, inspectors, chief inspectors) to examine the impact of rank on responses to the questionnaire. For this analysis, all detectives were included as Senior Officers (even if they indicated that their rank was detective constable) because detectives receive more training than regular constables. A chi-square analysis revealed an association between rank and whether or not officers had received instructions regarding discussion between co-witnesses, suggesting that Senior Officers were more likely to report that they had received instructions than Constables (χ2 (1, N = 143) = 6.60, p = .01). It was found that 85% of Senior Officers in comparison with 67% of Constables reported receiving instructions regarding discussion between co-witnesses. Independent t-tests were conducted to compare the quantitative responses given by Constables and Senior Officers. Only two significant differences were found. Senior Officers were more likely than Constables to report that they instruct witnesses not to discuss the event with one another before taking their statements at the scene of the incident, t (143) = 2.77, p < .01. Senior Officers were also more likely than the Constables to report that they ask the witnesses if they had discussed the event with one another, t (143) = 2.70, p < .01. Table 4.12 summarises the differences found between Constables and Senior Officers. However, once Bonferroni corrections were made (for the nine tests on Chapter 4: Legal Perspectives 89

police officer practice ratings), there were no significant differences found between Constables’ and Senior Officers’ responses.

Table 4.12 Summary of the Differences Found Between Constables and Senior Officers

Constables Senior Officers Question M SD M SD Do you instruct witnesses not to discuss before taking 3.01 1.29 3.52 0.92 statements? Do you ask witnesses if they 2.17 1.41 2.79 1.32 have discussed the event? Note. Respondents answered the questions using a 5-point scale (0 = Almost never; 1 = Not often; 2 = Often; 3 = Very often; 4 = Almost always).

The Effect of Experience on Questionnaire Responses Comparisons were made between officers based on how long they had been working within the police force. Once Bonferroni corrections were made, there were no significant differences found between the 44 officers with 0-5 years of experience compared with the 71 officers with 10 or more years of experience; between the 44 officers with 0-5 years of experience compared with the 101 officers with more than 5 years of experience; or between the 75 officers with 0-10 years of experience compared with the 70 officers with more than 10 years of experience.

Discussion The current research investigated the opinions that police officers hold regarding discussion between witnesses. The first research question asked, “How frequently do police officers believe co-witnesses discuss the event with each other?” Police officers in our survey estimated that 68% of witnesses talk with one another about the event prior to the arrival of the police when there are multiple witnesses present. This estimate is lower, but not incompatible with that provided by the witnesses in Study 1, who indicated that 86% had discussed details of the event with a co-witness. While it is possible that the police officers were underestimating the prevalence of co-witness discussion, it is also possible Chapter 4: Legal Perspectives 90

that the witnesses in Study 1 were not representative of all witnesses, as those who volunteered to answer questions about their experiences may have been more likely to discuss the events with co-witnesses than those who did not volunteer. Alternatively, the police officers may have given a lower estimate because they were asked to estimate how often witnesses talk to one another prior to the arrival of the police, whilst the witnesses where simply asked whether they had discussed the event with a co-witness. However, despite the fact that the estimate provided by the police is lower than that provided by the witnesses in Study 1, both studies strongly suggest that the majority of witnesses discuss details of the event with each other. The second research question asked, “What opinions do police offers hold regarding co-witness discussion?” Responses to this question on rating scales suggest that officers are somewhat critical of discussion. They were more likely to indicate that discussion can be detrimental to eyewitness memory than beneficial to it. When asked to explain their opinions regarding discussion between co-witnesses, the responses were varied, and included both condemnation and support of discussion between witnesses. Officers were critical of discussion between witnesses mainly because they believed that it may contaminate independent recollection, it may weaken the prosecution case, witnesses may become unsure of themselves if their memory differs from another’s, discussion leads to reports of ‘common things’ experienced, and witnesses might collude with one another (i.e., witnesses may agree on a version of the event for a deceitful purpose). On the contrary, some police officers noted that there were advantages of co- witness discussion. For example, they suggested that discussion between witnesses is beneficial because it may prompt eyewitnesses to recall details they previously did not remember, it may help reinforce witness memory; it may help recovery from trauma, and it may help provide a more accurate picture overall. These advantages and disadvantages of co-witness discussion mentioned by police officers are similar to those summarised by Stephenson (1990). Stephenson noted that arguments in favour of collaborative testimony include the beliefs that collaborative testimony facilitates accuracy, is more comprehensive than individual testimony, and is administratively convenient. However, Stephenson also noted arguments against collaborative testimony such as that disagreement between witnesses may be camouflaged, Chapter 4: Legal Perspectives 91

hearsay may be rendered acceptable, evidence may be suppressed, and effective cross- examination prevented. While this survey did not specifically investigate police officers as eyewitnesses, some officers stated that they found it beneficial to discuss their memory for events with other police officers who were at the scene. The officers would discuss the event with one another before preparing reports in order to ensure that they had covered all the evidence and coherently presented the facts. Despite the fact that barristers have been critical of police officers discussing events with one another (Heaton-Armstrong, 1985; 1987), some police officers were of the opinion that discussing matters with one another is not problematic. They appeared to be of the opinion that police officers are more objective than civilian witnesses and are not susceptible to the same influences as other eyewitnesses. As one officer noted: Police as witnesses talk about incidents on a regular basis and in my experience, it increases memory of events that have taken place…As a police officer, if I did not hear a conversation… the fact that a colleague or witness has heard the conversation does not influence me when preparing or giving a statement. However, civilian witnesses are not so like-minded.

Despite this belief among some police officers that they are superior witnesses due to their training and practical experience, research has shown that police officers do not differ from students on memory accuracy for a video depiction of a crime (Ost, Green, & Cherryman, 2004). However, police officers are significantly more confident in their accounts than students (Ost et al.). Furthermore, in another study pairs of police trainees were shown slightly different versions of a crime video (without their awareness that two versions existed) and then asked to discuss the event with one another. The results showed that the majority of police officers who had discussed the event (67%) mistakenly recalled items acquired during discussion (Granhag, Memon, Gabbert, & Allwood, 2004). That is, there was evidence of memory conformity among police officers similar to that which occurs among civilian eyewitnesses (e.g. Gabbert, Memon, & Allan, 2003; Wright, Self, & Justice, 2000). However, it is not known whether experienced officers would respond in a similar manner to the police trainees participating in Granhag et al.’s study. Our third research question asked, “What instructions have police officers received regarding co-witness discussion and are these instructions practical?” The majority of Chapter 4: Legal Perspectives 92

police officers (74%) reported that they had received instructions regarding discussion between witnesses. The most common instructions emphasised the importance of preventing discussion, mostly by discouraging witnesses from talking about the event and separating them. Senior Officers were more likely to report having received such instructions than Constables. The fact that Senior Officers were somewhat overrepresented in the sample relative to the population suggests that the officers completing the survey may be more educated about co-witnesses than average police officers. Furthermore, most of the officers who indicated the source of their instructions regarding co-witnesses stated that they learned from senior officers, court, or advanced training courses (e.g., Investigative training and Detective training). Only three officers indicated that they had received instructions about co-witnesses from their initial academy training. This suggests that there may be a lack of formal training regarding multiple witnesses in the early stages of a police officer’s career. Constables are more likely than senior officers to be the first to arrive at the scene of an incident, and it is therefore important for constables to be trained about discussion between eyewitnesses, and that this training be informed by empirical research investigating the real positive or negative consequences of co-witness discussion. On average, police officers thought the instructions they received were ‘Often’ or ‘Very Often’ practical, however they did suggest some ways in which the instructions were not practical. The most common ways that instructions could be impractical were that witnesses speak to one another before the police have time to arrive at the incident, and that it is almost impossible to prevent discussion from occurring, as it is natural for witnesses to want to talk to each other about the event. If witnesses will discuss the event with one another despite instructions not to, then perhaps in addition to trying to prevent discussion, researchers and the police should focus on improving interview techniques that elicit accurate, first-hand memories for the event that are not potentially contaminated by co- witness discussion. Other issues officers reported that made instructions impractical included a lack of facilities/space for separating witnesses and a lack of police officers at the scene in comparison to witnesses. Some officers mentioned that because of this, they would end up interviewing multiple witnesses together at the same time, as was seen during the observations of police officers by the author. Chapter 4: Legal Perspectives 93

Other police officers mentioned that the instructions are impractical because there is no formal procedure to separate and interview witnesses, nor are there instructions regarding whether witnesses can discuss the event after a statement has been obtained. These comments suggest that there may be a need to formalise training and policy in this regard. As mentioned previously, neither the United Kingdom nor (as far as can be established) Australia has formal policies regarding the separation of co-witnesses when obtaining eyewitness statements. Guidelines only exist in these countries for identification parades (Field Operations, 1997; Home Office, 2003). Given this paucity of formal policy, it is interesting to investigate the practices that officers engage in when dealing with multiple witnesses. On the whole, police officers reported that they engaged in practices that were in accordance with the informal instructions they had received. On average, police officers reported that they “Almost Never4” encouraged co-witness discussion or took statements from one witness in front of others, and they “Almost Never” conveyed information from one eyewitness to another and told them the source. Officers reported that they would “Not Often” convey information from one eyewitness to another without telling them the source. Officers reported that they would “Often” instruct witnesses not to discuss the event after taking their statements and “Often” ask witnesses if they have discussed the event with one another. Police officers reported that they would “Very Often” discourage co-witness discussion, physically separate witnesses, and instruct witnesses not to discuss the event before taking their statements. These responses present an encouraging picture. It appears that even without formal policy, most officers are being cautious about co-witness discussion. However, it is important to investigate co-witness discussion empirically to determine the validity of the assumptions made by legal authorities regarding the dangers of co-witness discussion. Despite this encouraging picture, when looking at the practices in which police officers engage, it is important to also consider the range of responses. It is interesting to note that virtually every possible answer on the scales was endorsed by at least one officer. Furthermore, conflicting responses were given for several of the open-ended questions (e.g., some police officers mentioned that certain topics should be discussed while other

4 Means were rounded to the closest whole number in order to be described in the terms used on the rating scale. Chapter 4: Legal Perspectives 94

police officers thought that the very same topics should not be discussed). This suggests that there is confusion in the minds of at least some police officers regarding the best practices. This confusion may be a result of inconsistent training and lack of policy or it might be because police officers know that courts are not in favour of co-witness discussion, and yet they believe the possibility exists that discussion can be beneficial to memory. On rating scales, police officers responded that discussion between eyewitnesses could “Maybe” refresh their memories. The confusion in the minds of police officers may be rectified by further research into the effects of co-witness discussion, which could lead to formal training and policy. It may be the case that assumptions made by the courts regarding the detrimental consequences of co-witness discussion are unsubstantiated. Additional research on co-witness discussion is required before its effects are fully understood and appropriate policy can be formulated. One limitation of the current study is that it relies on self-report. Police officers may have given the answers they believed to be the correct (i.e., socially desirable responses in accordance with policy or training), rather than the answers that most accurately reflected their practices. It is unlikely that this is the case, however because the questionnaires were anonymous and the police officers were encouraged to answer honestly. This survey provides useful insights into the opinions and practices of police officers. First, survey responses indicated that police officers believe the majority of witnesses discuss the event with one another. Some officers conceded that this discussion exists even before the police arrive at the scene of the incident. This suggests that research should focus on improving interview techniques that elicit accurate, first-hand memories for the event. Second, the survey results suggested that even though police officers were inclined to limit co-witness discussion, they could also think of several advantages of it. Third, the survey indicated that the majority of police officers had received instructions regarding co-witnesses. Police officers most commonly reported receiving instructions to prevent discussion by discouraging witnesses from talking about the event and separating them. However, many of the police officers reported receiving these instructions informally from other officers. Finally, the survey responses suggest that police officers, on the whole, engage in practices that are in accordance with the informal instructions they had received.

95

SECTION 3: EXPERIMENTAL STUDIES

Section 3 Aims

• To determine whether co-witness discussion helps or hinders individual recall and to investigate this within a closely defined methodological set.

• To investigate the relative impact of co-witness discussion on eyewitness memory in comparison to other ways of encountering postevent information.

• To test existing theories on the misinformation effect and hence contribute to knowledge regarding the causes of memory conformity.

• To establish whether it is possible to mediate negative effects of co-witness discussion by employing our theoretical understanding of the causes of the effect.

From Section 2, it was established that witnesses often discuss details of the event with one another. While most police officers discouraged this practice, they conceded that witnesses sometimes discuss the event before they arrive at the scene and that it can be difficult to prevent co-witness discussion because it is natural for witnesses to want to talk amongst themselves. Because it is evident that eyewitness information is frequently conveyed from one witness to another, it is critical to ascertain the effects of co-witness information on the accuracy and completeness of eyewitness testimony. Since previous research has been inconclusive, it is important to determine whether co-witness discussion helps or hinders individual recall and to investigate this within a closely defined methodological set. Furthermore, another aim of Section 3 was to establish whether it is possible to mediate any negative effects of co-witness discussion by employing our theoretical understanding of the causes of the effect. To investigate these issues, five experiments were conducted (Studies 3-7). Study 3 was an exploratory study that sought to investigate the effects of co-witness discussion and delay on eyewitness memory. In this experiment, two methodologies were employed to investigate the social influences on memory: (1) Confederate Induced Information, (2) Natural Discussion Groups. 96

After investigating the effects of co-witness discussion on eyewitness memory in Study 3, Study 4 was designed to examine the relative size of the effects. Psychological research has focused on how postevent information encountered through sources such as leading questions and media reports affects eyewitness memory, however very little research in comparison has examined the impact of discussion between co-witnesses. To address this matter, Study 4 compared the effects of co-witness discussion on memory with the more widely studied methods of encountering postevent information. Study 5 was conducted in order to further examine the causes of memory conformity following discussion. Specifically, two techniques were employed in an attempt to identify the mechanisms responsible for memory conformity: remember/know judgments (Tulving, 1985) and warnings about possible misinformation. The aim of Study 6 was to build on the findings obtained from Study 5. In Study 6, a source-monitoring test was employed instead of remember/know judgments. Participants in Study 6 were asked to indicate whether they remembered the item from the video only, from the discussion only, or from both the video and discussion. Like Study 5, participants in Study 6 were also given warnings about possible misinformation. Finally, Study 7 aimed to determine whether the effects of co-witness misinformation could be reversed during an interview. Specifically, this study examined whether remember/know judgments and warnings in the context of an interview could affect participants’ abilities to distinguish what they actually witnessed from co-witness information.

97

Chapter 5 Study 3: Co-witness Contamination

It is evident from the results described in Section 2 that eyewitness information is frequently conveyed from one witness to another, and therefore it is critical to determine the effects of co-witness information on the accuracy and completeness of eyewitness testimony. In particular, it is important to determine whether co-witness discussion helps or hinders individual recall. As outlined in Section 1, previous studies investigating the effects of co-witness discussion on eyewitness memory have reached inconsistent results. Some studies have found that discussion can be advantageous to eyewitness reports (e.g., Warnick & Sanders, 1980), while others have found it to be detrimental (e.g., Hollin & Clifford 1983; Shaw, Garven, & Wood, 1997). Yarmey and Morris (1998) have suggested that the variation in results is due to methodological differences. To further our understanding of the effects of co-witness discussion on eyewitness reports, Study 3 examines two different methodologies that have been used to investigate this topic: (1) studies utilizing Confederate Induced Information, and (2) studies involving Natural Discussion Groups. Following the tradition of Asch (1951; see Bond & Smith, 1996 for review) many memory conformity studies utilize trained confederates. Asch found that participants would conform to a confederate’s response on a perceptual judgment task, regardless or whether the confederate was correct. Schneider and Watkins (1996, Experiment 2) demonstrated a similar effect with response conformity in memory for words finding that memory responses given second often conformed to those given first, even when the first person was clearly incorrect. Although this study lacks ecological validity due to its use of word lists rather than an event, similar results have also been found in more applied studies. In a typical applied experiment, participants are shown an event and then instructed to discuss the event with a confederate who provides misinformation. Participants are later asked to give individual accounts of what happened during the event and the accounts are analyzed for evidence that the participants incorporated the misleading information provided by the confederate into their testimonies. Typically, a comparison is made either between items for Chapter 5: Co-witness Contamination 98

which participants had been misled by the confederate, and those for which they had not been misled by the confederate, or between responses given by participants who were in groups with confederates and participants who did not discuss the event with anyone. Several researchers have reported that groups with confederates supplying misinformation show less accurate memory for the misled items than non-misled items (e.g., Hoffman et al., 2001; Schneider & Watkins, 1996; Shaw et al., 1997; Wright & Mathews, 2004) and that these groups are less accurate than individuals who did not discuss the event (e.g., Gabbert, Memon, Allan, & Wright, 2004; Hollin & Clifford, 1983). A general feature of research involving Confederate Induced Information is that it relies on information that is artificially provided by a confederate, and therefore the group discussion may not progress in a natural way. To address this, the current research included groups that did not include confederates (i.e. Natural Discussion Groups). No previous studies have compared responses given by individuals who were in groups with confederates to those who were in discussion groups that did not include any confederate. This is surprising considering that the Natural Discussion Group methodology has potentially the highest ecological validity. Furthermore, when participants in groups with confederates are compared with individuals who have not discussed the event, there is a confound because it is unclear whether the effects are due to the act of discussing the event, the social presence of another, the elaboration and rehearsal of event-related memories, the impact of the misinformation, or some other aspect of the confederate’s influence. There are also confounds that exist when all participants are misled by a confederate and a comparison is made between items for which they had been misled, and those for which they had not been misled. As with any within-subjects design, there can be the problem of carryover effects. For example, if all participants are receiving some misinformation from a confederate, this may affect credibility of the confederate, and therefore the extent to which participants accept subsequent information presented to them by the confederate. Research has shown that participants may not be influenced by blatant misleading information and that a ‘spillover effect’ may occur in which they become less susceptible to subsequent misinformation introduced by the same source (Loftus, 1979b). Although no previous studies have compared groups with confederates to those without confederates, a few applied studies have included groups without confederates. Chapter 5: Co-witness Contamination 99

These studies that do not include confederates, but rather rely on naturally occurring conversations between genuine co-witnesses, have found inconsistent results. In one study, participants’ recollection for a conversation they had heard earlier was tested (Yarmey, 1992). It was found that discussion prior to individual testimony led to more fabrications and worse duration estimates than individuals or collaborative dyads. Thus, Yarmey found that discussion prior to individual recall was a disadvantageous procedure. Other studies involving Natural Discussion Groups have found that under certain conditions, group recall can have a beneficial effect on eyewitness testimony. For example, Underwood and Milton (1993) tested participants’ memories for the details of a traffic accident. They found that when participants were informed that they were going to witness an accident, those who had discussed the event prior to giving an individual statement performed better than participants who did not discuss the event. However, when witnesses were not expecting to see an accident, there was no effect of discussion on recall performance. In another study (Yarmey & Morris, 1998), participants who discussed a video they had seen were more accurate in their subsequent individual recall than those who had previously discussed issues unrelated to the video and no-discussion groups. However, this advantage of group discussion was not evident when the participants completed a culprit-absent lineup task. Because inconsistent results have been found in studies involving Natural Discussion Groups, it is unclear what the impact of natural (i.e., non-confederate-induced) discussion on memory for an event is. The aim of the current research was to compare the effects of discussion among members of groups that contain trained confederates with Natural Discussion Groups, and with individuals who had not discussed the event. Based on previous studies (e.g., Gabbert, Memon, Allan, & Wright, 2004; Hoffman et al., 2001; Hollin & Clifford, 1983; Schneider & Watkins, 1996; Shaw et al., 1997; Wright & Mathews, 2004) it was hypothesized that groups with confederates supplying misinformation would have less accurate memory for the misled items than individuals not in a discussion group. It was also hypothesized that individuals who had discussed the event in groups with confederates would have less accurate memory on the misled items than individuals in Natural Discussion Groups, however this has not been investigated before. Given that past research has yielded mixed findings (e.g., Underwood & Milton, Chapter 5: Co-witness Contamination 100

1993; Yarmey, 1992; Yarmey & Morris, 1998) it was not possible to make any predictions regarding the performance of Natural Discussion Groups relative to individuals not in a discussion group. However, it seems likely that discussion among group members might re- expose participants to details of the event that they would otherwise have forgotten, thus resulting in greater recall for individuals in groups without confederates in comparison to individuals not in a discussion group. A second objective of this study was to assess the effects of time delay on susceptibility to misleading information. In traditional studies investigating the misinformation effect, it has been found that when initial memory is good (i.e., after a short delay), participants are less readily persuaded by misinformation than after a long delay (Loftus, Miller, & Burns, 1978). That is, the shorter the time interval between the witnessed event and the occurrence of the misleading information, the less likely it is that the original memory will be vulnerable to misinformation. However, this finding has not been replicated in the context of co-witness memory conformity. For example, one study found that there was no difference in conformity for short (20 minutes) versus long (1 day) delay groups (Hoffman, Granhag, See, & Loftus, 2001). Hoffman et al.’s study showed that delay affected memory accuracy, but not the amount of memory conformity. Even after a short delay, when participants’ memory for the event was quite good, they conformed to misleading co-witness information. As the authors concede, “The fact that the delay manipulation successfully affected memory accuracy, but did not affect the amount of conformity appears to contradict both the social literature and predictions made based on the misinformation paradigm” (p.403). However, they explain their results by suggesting that although the participants were more accurate in the short delay condition than the long delay condition, they were no more confident in their memories. According to Strack and Bless (1994) the participant’s impressions about how accurately he or she can remember items (i.e., their confidence) is better at determining whether people will conform than memory accuracy itself. Because traditional studies investigating the misinformation effect have shown different results from Hoffman et al.’s (2001) study investigating memory conformity, it was unclear what the effect of delay would be on susceptibility to misleading information in the current experiment. Chapter 5: Co-witness Contamination 101

Method

Design There were four phases in the study: exposure, delay, discussion, and recognition. During the exposure phase, participants were shown two crime videos. In the delay phase, participants either experienced a 20-minute or 2-week delay, depending on the condition to which they were randomly assigned. In the discussion phase, participants were assigned to one of three conditions: (1) discuss the videos in a group that included an experimental confederate (confederate-present), (2) discuss the videos in a group without a confederate (confederate-absent), or (3) individually write down details of the videos with no discussion. During the recognition phase, all participants were asked to complete a true/false questionnaire. Some of the questions assessed participants’ memories for details about which the confederates had introduced misinformation during the discussion phase. Other questions concerned details that had not been mentioned by the confederates (i.e. for which the participants did not receive misinformation). Participants were also asked to indicate their confidence in their answers. The study employed a 3x2 experimental design that investigated the effects Discussion Type (confederate-present, confederate-absent, no discussion) and Time Delay after witnessing the crime (twenty minutes, two weeks) on memory for misled and non- misled items. The dependent variables were accuracy and confidence of memory for the videos, measured using a true or false recognition questionnaire.

Participants Participants were 171 undergraduate students (116 females, 55 males) from the University of New South Wales. The average age of the participants was 20.5 years (SD = 3.1). Participants were invited to take part in the experiment as part of their second year psychology course. Chapter 5: Co-witness Contamination 102

Materials Eyewitness Stimuli. The eyewitness stimuli used in this study were two short crime videos, each under a minute in duration. Both videos depicted robberies: one of a store robbery and the other, a street robbery. The store robbery scene was extracted from a relatively unknown feature film entitled, “The Arrival” (Schmoeller, 1990). The selected scene takes place in a store and shows a woman arguing with the storeowner. Two Caucasian males then enter the shop and hold up the storekeeper. The two men steal some money and cigarettes, and then fire a gun at the storekeeper, narrowly missing him. When asked, none of the participants indicated that they had seen the film before, so no one was excluded from the study for this reason. The street robbery scene was taken from a training video developed by the Metropolitan Police in London. In this video, a black man is seen looking into a parked car. An older woman sees him looking in her car and yells at him. He pushes her onto the ground and his accomplice, a Caucasian male, runs up and steals her handbag. The woman yells for help and a third man comes to her rescue and chases after the black man. All participants were shown both videos, with the presentation order counter- balanced. Postevent Information. Participants in the confederate-present condition were exposed to four items of misinformation about each video. The items were introduced by the confederate and are summarised in Table 5.1. Eight different confederates were used in order to decrease the possibility that the results obtained would be dependent on distinct characteristics of the confederate used. The confederates were male and female university students who were around the same age as the participants in the study, but unknown to the participants. The confederates were instructed to take part in the experiment as though they were actual participants. They were instructed to introduce the eight items of misinformation during the discussion in any order they liked. They were also allowed to contribute other correct information to the discussion so that they didn’t appear implausible. Confederates were not instructed about the number of correct responses to give; rather they were just instructed to report as many as they thought were necessary. If Chapter 5: Co-witness Contamination 103

someone questioned the confederates about what they said, confederates were instructed to reply, "Well, that's what I remember seeing on the video." They were instructed to reply confidently, without being overly assertive.

Table 5.1. Misleading Postevent Information Used in Study 4 Video Misleading Postevent Information Provided Correct by Confederate Information From Videos Store Robbery The thief that shot the gun was named His name was “Chuck.” Hank.

The thief that shot the gun was wearing a red The bandana was bandana. navy blue.

The thief who pointed the handgun at the lady He yelled at her, kicked her to get her away from the counter. but didn’t kick her. The woman in the store was wearing a large She was wearing brooch. a large necklace. Street The man who pushed the woman stole her bag. His accomplice Robbery stole the bag. The black man who pushed the woman tried to He looked in her smash her car window. window, but didn’t try to smash it. The man who came to help the lady was He was wearing a wearing a dark green shirt. royal blue shirt. The woman’s car had two doors. The car had four doors.

Recognition Questionnaire. The recognition questionnaire (see Appendix D) consisted of a total of forty statements (twenty relating to each of the two videos) to be rated as true or false. Of the twenty statements about each video, twelve were true, four were false, and four were false statements regarding an issue about which participants in the confederate-present group received misleading information. Example questionnaire Chapter 5: Co-witness Contamination 104

items relating to misleading information supplied by the confederate include statements such as: “The thief that shot the gun was named Chuck,” (when in fact he was identified as ‘Hank’ in the video, but the confederate had said he was named ‘Chuck’ during the discussion) or “The man who pushed the woman stole her bag,” (when in fact it was his accomplice who stole the bag, but the confederate had stated it was the man who pushed the woman). For each statement, participants were asked to indicate whether they thought the statement was true or false, and how confident they were in this decision, using a six point scale where a score of “1” indicated “True” and “6” indicated “False.” Discussion Questions. Participants who were in confederate-present and confederate-absent discussion groups were also asked three questions about their experience in the groups. They were asked, “Did you find the group discussions interesting and useful”; “Did you learn anything new about the episodes as a result of the discussions”; and “Did your group have an open and free exchange of ideas?” Participants were asked to indicate their level of agreement with each of these questions using a six-point scale where a score of “1” indicated “Yes” and “6” indicated “No.”

Procedure In class groups of approximately twenty people, participants were shown the two short crime scene videos. Prior to viewing the video, the experimenter stated, “I will be showing you a video of two short crime scenes. Imagine that these are real life crimes and that you are actual witnesses to the crimes. Just like real witnesses, you will be asked questions about the crimes later, so please pay close attention to the videos.” Half of the groups saw the store robbery video first, and half saw the street robbery video first. After a delay of either twenty minutes (during which they did their regular course work) or two weeks, depending on the condition to which the groups were randomly assigned, participants were given the Recognition Questionnaire. Participants were instructed, “You saw a video of two short crime scenes. Please take a few minutes to complete these questions about the scenes. Please work individually.” After completing the questionnaire, participants were split into groups of four to discuss the crime scenes. Unbeknown to the group members, some groups (i.e., confederate-present groups) included a trained confederate who introduced four pieces of Chapter 5: Co-witness Contamination 105

misleading information about each scene into the discussion. Members of the confederate- absent discussion groups were assigned to groups that did not include a confederate. Participants in the no-discussion control group were required to work individually and write an account of the videos without discussion. All participants were given five minutes to discuss or write about each of the videos. They were instructed, “[Discuss/write about] what you remember seeing in the videos [with one another/individually]. Specifically, [discuss/write about] the sequence of events, what was said, the setting, what the people looked like, and any other information you can remember.” Participants were then given the Recognition Questionnaire (identical to the one completed earlier) and Discussion Questions to complete individually. Finally, participants were thanked for their participation and fully debriefed about the study. Participants were asked not to discuss the experiment with other potential participants.

Results Statistical analysis concentrated on the following questions: First, was the experimental manipulation successful? Second, how did Discussion Type (confederate- present, confederate-absent, no discussion) and Time Delay after witnessing the event (twenty minutes or two weeks) influence the accuracy of the participants’ memories for misled and non-misled items? Finally, how did Discussion Type and Time Delay influence participants’ confidence in their memories for misled and non-misled items?

Experimental Manipulation Check Separate 2x2 (Discussion Type x Time Delay) ANOVAs were conducted, with participants’ responses to each of the three questions relating to their perception of the discussion that occurred in their group acting as the dependent variable. The individuals in the no-discussion control group were not given these questions, so they were not included in this analysis. Analysis of responses to the first question, “Did you find the group discussions interesting and useful?” did not show a main effect of Discussion Type (F (1, 100) = 2.83, p = .10), Time Delay (F (1, 100) = 0.93, p = .34) or an interaction between Discussion Type and Time Delay (F (1, 100) = 1.68, p = .20). For the question, “Did you Chapter 5: Co-witness Contamination 106

learn anything new about the episodes as a result of the discussions?” there was an identical pattern of results. Responses did not show a main effect for Discussion Type (F (1, 100) = 1.10, p = .30), Time Delay (F (1, 100) = 1.67, p = .20) or the interaction of the two (F (1, 100) = 0.004, p = .95). Finally, for the question “Did your group have an open and free exchange of ideas?” there were once again no significant effects of Discussion Type (F (1, 100) = 0.28, p = .60), Time Delay (F (1, 100) = 3.35, p = .07) or the interaction (F (1, 100) =1.16, p = .28). That is, there were no significant main effects or interactions for any of the three questions indicating that the participants in groups with confederates did not perceive the discussions differently to those in groups without confederates. Table 5.2 shows the average responses given by participants for the three discussion questions.

Table 5.2

Response to Discussion Questions

20 Minute Delay 2 Week Delay Question Condition M SE n M SE n Confederate- Did you find the group 2.42 0.30 19 2.33 0.27 24 present discussions interesting and Confederate- useful? 2.52 0.28 21 3.13 0.21 40 absent Confederate- Did you learn anything 1.84 0.30 19 2.21 0.27 24 present new about the episodes as Confederate- a result of the discussions? 2.14 0.29 21 2.48 0.21 40 absent Confederate- Did your group have an 1.42 0.18 19 1.54 0.16 24 present open and free exchange of Confederate- ideas? 1.33 0.17 21 1.80 0.12 40 absent Note. Participants responded using a six-point scale where a score of “1” signified “Yes” and “6” signified “No.”

Separate one-way ANOVAS were conducted to investigate whether the eight confederates differently influenced the participants’ responses to each of the three questions relating to their perception of the group discussion. It was found that there were no differences between the groups who had different confederates on responses to the first question, “Did you find the group discussions interesting and useful?” (F (7, 35) = 1.11, p = Chapter 5: Co-witness Contamination 107

0.38). Similarly, there were no differences between the groups who had different confederates on responses to the question, “Did you learn anything new about the episodes as a result of the discussions?” (F (7, 35) = 0.76, p = .63). Finally, there were no differences between the groups who had different confederates on responses for the question “Did your group have an open and free exchange of ideas?” (F (7, 35) = 0.59, p = .76). Furthermore, separate one-way ANOVAS were conducted to investigate whether the eight confederates influenced the participants differently. It was found that there was no effect of the different confederates on accuracy and confidence for misled and non-misled items. That is, there were no differences between the groups who had different confederates on accuracy for misled items (F (7, 39) = 1.75, p = .13) and accuracy for non-misled items (F (7, 39) = 2.02, p = .08). Similarly, there were no differences between the groups who had different confederates on confidence for misled items (F (7, 39) = 2.04, p = .07) and confidence for non-misled items (F (7, 39) =1.63, p = .16).

Memory Accuracy Participants’ responses to the recognition questions were recoded such that a correct answer was scored “1” and an incorrect answer scored “0.” Scores were then combined to give a total accuracy score for the eight items (across both videos) for which we had attempted to mislead some participants. A separate score for the thirty-two items that were not subject to misinformation was also calculated. All results were then converted into percentage accuracy scores. Preliminary analyses revealed that the three discussion groups differed significantly on the first recognition questionnaire, F (2, 168) = 5.03, p < .01, (see Figure 5.1). Given that the first questionnaire was completed prior to the discussion/writing task (and therefore before the experimental manipulation occurred) this difference must be due to chance differences not eliminated by the random allocation of participants to groups. To overcome this problem, scores on the first questionnaire were used as a covariate in subsequent analyses. Chapter 5: Co-witness Contamination 108

on i

t 67.50 a

pul ni

a 65.00

m

o o

t

or i

62.50

pr

y

c a

ur

c 60.00

c

A

%

57.50 Confederate-present Confederate-absent No discussion Discussion Type

Figure 5.1. Mean accuracy of the three discussion groups on the first recognition questionnaire. Error bars indicate 95% Confidence Intervals.

Memory accuracy for misled and non-misled items on the second questionnaire were analysed with a (2) x 2 x 3 (Video x Time Delay x Discussion Type) MANCOVA with accuracy score on the first recognition questionnaire serving as a covariate. There were no main effects for the factor “Video” (store or street robbery), F (1, 164) = 1.75, p = .19. There were also no interactions between Video and Discussion Type, F (2, 164) = 2.45, p = .09, Video and Time Delay, F (1, 164) = 0.89, p = .35, or Video, Discussion Type, and Test Time, F (2, 164) = 0.91, p = .41. Since there were no main effects or interactions for the factor ‘Video,’ data from the two videos were combined for all subsequent analysis. A 2 x 3 MANCOVA was conducted which omitted the factor ‘Video.’ However, before conducting the MANCOVA, the two accuracy scores (i.e., for misled and non- misled items) were correlated to check for possible multicollinearity. It was found that the dependent variables were not unacceptably highly correlated (r = -.33, p < .05) and Chapter 5: Co-witness Contamination 109

therefore it was appropriate to proceed with the MANCOVA (Tabachnick & Fidell, 2001). The mean scores for memory accuracy are shown in Table 5.3. The MANCOVA revealed a significant effect of Discussion Type on the combined dependent variable of accuracy, F (4, 326) = 10.74, p < .0005; Wilks’ Lambda = .8; partial eta squared = .12. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was a significant main effect of Discussion Type on non-misled items (i.e., items for which we had not attempted to mislead any participants, F (2, 164) = 5.39, p < .01). Planned contrasts using a Bonferroni correction revealed that participants from confederate-present groups had significantly more accurate memory for the non-misled items (M = 68.7, SE = 1.37) than the no-discussion control group (M = 62.1, SE = 1.48, p < .01). There were no significant differences on the non- misled items between the confederate-absent groups (M = 65.5, SE = 1.09) and the no- discussion control group (p = .07) or the confederate-present groups (p = .07). The MANCOVA also revealed a significant main effect of Discussion Type on misled items (i.e., items for which we had attempted to mislead some participants, F (2, 164) = 20.6, p < .0005). Planned contrasts using a Bonferroni correction revealed that participants from confederate-present groups had significantly less accurate memory for the misled items (M = 45.0, SE = 2.76) than those in confederate absent (M = 65.4, SE = 2.20, p < .0005) and the no-discussion control groups (M = 66.7, SE = 2.97, p < .0005). There were no significant differences on the misled items between the confederate-absent discussion groups and the no-discussion control group (p = .72). Time Delay also significantly affected the combined dependent variable of accuracy, F (2, 163) = 25.1, p < .0005; Wilks’ Lambda = .8; partial eta squared = .24. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was no effect of Time Delay on non-misled items, F (1, 164) = 1.26, p = .26. However, there was an effect of Time Delay on misled items, F (1, 164) = 30.0, p < .0005. Participants’ memory was more accurate for the misled items 20 minutes after witnessing the event (M = 68.6, SE = 2.19) than two weeks later (M = 49.5, SE = 2.44). There was no interaction between Discussion Type and Time Delay, F (4, 326) = 1.95, p = .10. Chapter 5: Co-witness Contamination 110

Table 5.3

Mean Memory Accuracy for Recognition Questionnaire

Condition Non-misled Items Misled Items M SE M SE Discussion Confederate-present 68.7 1.37 45.0 2.76 Type Confederate-absent 65.5 1.09 65.4 2.20 No-discussion 62.1 1.48 66.7 2.97 Time Delay 20-minutes 66.4 1.09 68.6 2.19 2 weeks 64.5 1.22 49.5 2.44

The accuracy scores were also recoded so that a comparison could be made between the number of participants who reported misinformation both before and after the discussion manipulation. That is, scores were recoded in such a way that those who reported more, less, or the same amount of misinformation following the manipulation could be compared across the three Discussion Types. In the confederate-present group, a much higher percentage of the participants (83%) reported more misinformation after the discussion manipulation than before it in contrast to participants in the confederate-absent groups (38%) and the no discussion controls (31%). A chi-square analysis revealed that there was an association between Change in Misinformation Reported (more, less, same) and Discussion Type (confederate-present, confederate-absent, no discussion; χ2 (4, N = 171) = 33.40, p < .0005).

Memory Confidence The six-point scale used for the recognition questionnaire combined accuracy and confidence ratings such that a score of “1” indicated “True” and “6” indicated “False.” To look at confidence alone, the data were recoded so that a score of ‘3’ represented high confidence and a score of ‘1’ represented low confidence regardless of accuracy. Scores were then combined to provide a total confidence score for the eight items (across both videos) for which we had attempted to mislead some participants. A separate score for the thirty-two items for which we had not attempted to mislead participants was calculated in the same way. All results were then rescaled in the range 0-100. Chapter 5: Co-witness Contamination 111

After noting that the dependent variables with regards to confidence (for misled and non-misled items) were not unacceptably highly correlated (r = -.10), a 2 x 3 (Time Delay x Discussion Type) MANCOVA was conducted, with confidence score on the first recognition questionnaire serving as a covariate. The mean scores for memory confidence are shown in Table 5.4. The 2 x 3 MANCOVA revealed a significant effect of Discussion Type on the combined dependent variable of confidence, F (4, 326) = 12.56, p < .0005; Wilks’ Lambda = .8; partial eta squared = .13. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was a significant main effect of Discussion Type on confidence for non-misled items, F (2, 164) = 8.35, p < .0005. Planned contrasts using a Bonferroni correction revealed that participants from confederate-present groups had significantly higher confidence for the non-misled items (M = 73.7, SE = 0.67) than those in confederate-absent groups (M = 71.0, SE = 0.53, p < .01) and the no- discussion control group (M = 69.9, SE = 0.72, p < .0005). There were no significant differences on the non-misled items between the confederate-absent discussion groups and the no-discussion control group (p = .20). The MANCOVA also revealed a significant main effect of Discussion Type on confidence for misled items, F (2, 164) = 23.9, p < .0005. Planned contrasts using a Bonferroni correction revealed that participants from confederate-present groups had significantly lower confidence for the misled items (M = 56.3, SE = 1.30) than the confederate-absent groups (M = 67.5, SE = 1.03, p < .0005) and the no-discussion control groups (M = 65.3, SE = 1.39, p < .0005). There were no significant differences on the misled items between the confederate-absent discussion groups and the no-discussion control group (p = .22). There was also a significant effect of Time Delay on the combined dependent variable of confidence, F (2, 163) = 31.2, p < .0005; Wilks’ Lambda = .7; partial eta squared = .28. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was no effect of Time Delay on non-misled items, F (1, 164) = 1.64, p = .20. However, there was an effect of Time Delay on misled items, F (1, 164) = 38.1, p < .0005. Participants’ confidence was higher for the misled items 20 minutes Chapter 5: Co-witness Contamination 112

after witnessing the event (M = 68.4, SE = 1.06) than two weeks later (M = 57.6, SE = 1.20).

Table 5.4

Mean Memory Confidence for Recognition Questionnaire

Condition Non-misled Items Misled Items M SE M SE Discussion Confederate-present 73.7 0.67 56.3 1.30 Type Confederate-absent 71.0 0.53 67.5 1.03 No-discussion 69.9 0.72 65.3 1.39 Time Delay 20-minutes 72.1 0.55 68.4 1.06 2 weeks 71.0 0.62 57.6 1.20

There was a significant interaction found between Discussion Type and Time Delay, F (4, 326) = 2.88, p < .05; Wilks’ Lambda = .9; partial eta squared = .03. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was no interaction between Discussion Type and Time Delay on non- misled items, F (2, 164) = 2.85, p = .06. However, there was a significant interaction between Discussion Type and Time Delay for confidence on misled items, F (2, 164) = 5.39, p < .01. See Figure 5.2. Inspection of the graph suggests that confidence for misled items dropped dramatically after a two-week delay for participants in the confederate- present and no-discussion conditions. Although participants in the confederate-absent groups had a decrease in confidence after a two-week delay, it was less pronounced. Chapter 5: Co-witness Contamination 113

80

70

Confederate-present 60 Confederate-absent No discussion

50

40 20 minutes 2 weeks

Figure 5.2. Interaction between Discussion Type and Time Delay for confidence on misled items.

Confidence in Correct and Incorrect Decisions In a second confidence analysis, confidence ratings for correct and incorrect decisions were computed separately. 2 x 3 MANCOVAs were conducted to investigate the effects of Time Delay and Discussion Type separately for correct and incorrect decisions. Correct Decisions. First, after establishing that the two dependent variables for confidence in correct decisions (for misled and non-misled items) were not unacceptably highly correlated (r = .62, p < .01), a 2 x 3 MANCOVA was conducted to investigate the effects of Discussion Type and Time Delay on confidence scores for correct decisions made for misled and non-misled items. Confidence scores for items that the participants answered correctly on the first recognition questionnaire served as the covariate. The mean confidence scores for correct decisions are shown in Table 5.5. The MANCOVA revealed a significant effect of Discussion Type on the combined dependent variable of confidence in correct decisions, F (4, 326) = 3.54, p < .01; Wilks’ Lambda = .9; partial eta squared = .04. Analysis of each individual dependent variable, Chapter 5: Co-witness Contamination 114

using a Bonferroni adjusted alpha level of .025, showed that there was no main effect of Discussion Type on non-misled items, F (2, 164) = 0.76, p = .47. However, there was a significant main effect of Discussion Type on misled items, F (2, 164) = 6.70, p < .01. Planned contrasts using a Bonferroni correction revealed that participants in the no- discussion control group had significantly lower confidence on misled items that they got correct (M = 1.84, SE = 0.64) than participants in the confederate-present discussion groups (M = 2.16, SE = 0.06) and the confederate-absent groups (M = 2.03, SE = 0.05, p < .05). There were no significant differences in the number of misled items that the participants got correct between the confederate-present and confederate-absent groups (p = .09). There was no effect of Time Delay on the combined dependent variable of confidence in correct decisions, F (2, 163) = 0.70, p = .50. There was a significant interaction found between Discussion Type and Time Delay, F (4, 326) = 2.70, p < .05; Wilks’ Lambda = .9; partial eta squared = .03. However, analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was no interaction between Discussion Type and Time Delay on non-misled items that the participants got correct, F (2, 164) = 3.24, p = .04. There was also no interaction between Discussion Type and Time Delay on misled items that the participants got correct, F (2, 164) = 0.99, p = .38.

Table 5.5

Mean Confidence in Correct Decisions

Condition Non-misled Items Misled Items M SE M SE Discussion Confederate-present 2.24 0.04 2.16 0.06 Type Confederate-absent 2.19 0.03 2.03 0.05 No-discussion 2.20 0.04 1.84 0.64 Time Delay 20-minutes 2.23 0.03 2.04 0.05 2 weeks 2.18 0.03 1.98 0.05

Incorrect Decisions. After establishing that the dependent variables for confidence in incorrect decisions (for misled and non-misled items) were not unacceptably highly correlated (r = .39, p < .01), confidence scores for incorrect decisions were also Chapter 5: Co-witness Contamination 115

analysed with a x 2 x 3 (Time Delay x Discussion Type) MANCOVA. Confidence scores for items that the participants got incorrect on the first recognition questionnaire served as the covariate. The mean scores for confidence in incorrect decisions are shown in Table 5.6. The MANCOVA revealed a significant effect of Discussion Type on the combined dependent variable of confidence in incorrect decisions, F (4, 326) = 2.65, p < .05; Wilks’ Lambda = .9; partial eta squared = .03. However, analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was no main effect of Discussion Type on non-misled items that the participants got incorrect, F (2, 164) = 2.85, p = .06. There was also no main effect of Discussion Type on misled items that the participants got incorrect, F (2, 164) = 2.04, p = .13. There was no effect of Time Delay on the combined dependent variable of confidence in incorrect decisions, F (2, 163) = 1.89, p = .16. There was also no interaction found between Discussion Type and Time Delay on the combined dependent variable of confidence in incorrect decisions, F (4, 326) = 1.42, p = .23.

Table 5.6 Mean Confidence in Incorrect Decisions

Condition Non-misled Items Misled Items M SE M SE Discussion Confederate-present 1.54 0.05 2.01 0.10 Type Confederate-absent 1.68 0.04 1.77 0.08 No-discussion 1.55 0.06 1.80 0.11 Time Delay 20-minutes 1.63 0.04 1.80 0.07 2 weeks 1.55 0.05 1.92 0.08

Discussion The findings reported in the current experiment suggest that when participants talk to a confederate about an event that they both witnessed, they are likely to report misinformation introduced by the confederate when later questioned individually. Chapter 5: Co-witness Contamination 116

Furthermore, the findings suggest that such misinformation has a particularly strong effect on the individual’s testimony when it is presented after a longer delay. Our first research aim was to determine whether or not the experimental manipulation was successful. Participants were asked three questions in order to investigate whether those involved in confederate-present discussions perceived the group discussions differently to those in confederate-absent groups. We found no significant differences on any of the three questions relating to the discussion experience. This implies that participants in confederate-present groups were unaware of the manipulation, as they reported their discussions to be just as ‘interesting, useful, informative, open, and free’ as the participants in groups without confederates. Participants also appeared to be surprised when they were debriefed and informed about the confederates, which again suggests that they were unaware of the experimental manipulation. Our second research aim was to investigate how Discussion Type (confederate- present, confederate-absent, no discussion) and Time Delay after witnessing the event (twenty minutes or two weeks) influenced the accuracy of the participants’ memories for misled and non-misled items. Our results supported the hypothesis that confederate-present groups who received misinformation would show less accurate memory for the misled items than individuals in the no discussion control group. That is, participants incorporated misinformation provided by the confederate into their own individual account. These findings support those from previous studies (e.g., Gabbert, Memon, Allan, & Wright, 2004; Hoffman et al., 2001; Hollin & Clifford, 1983; Schneider & Watkins, 1996; Shaw et al., 1997; Wright & Mathews, 2004). Although it has not been investigated before, it was also predicted that confederate- present groups would be less accurate on the misled items than confederate-absent groups, which turned out to be the case. That is, participants who were in groups with confederates who introduced misinformation were more likely to report the misled items than those who were in groups without confederates. As mentioned previously, the confederate-absent discussion groups have potentially the highest ecological validity, and are a methodologically sound comparison group, and yet they have been neglected by most applied studies. Chapter 5: Co-witness Contamination 117

It was unclear how confederate-absent groups would perform relative to individuals not in a discussion group, as previous research has yielded mixed findings (e.g., Underwood & Milton, 1993; Yarmey, 1992; Yarmey & Morris, 1998). The results from this study found that there were no differences between confederate-absent discussion groups and the no-discussion controls on either the misled or non-misled items. That is, the results from this study suggest that their accuracy was no better or worse than those of the no-discussion controls. This finding is particularly interesting because many studies involving confederates have experimentally induced misinformation, and therefore have focused on the negative effects of co-witness discussion. However, as this study shows, when misinformation is not experimentally induced, there appear to be no effects, either positive or negative, on subsequent memory accuracy in comparison to people who do not discuss the event. Our finding that participants from confederate-present groups had more accurate memory for the non-misled items than participants from the no-discussion control groups was unexpected. There are several possible explanations for this. First, the confederates may not have acted like true co-witnesses and therefore their presence in the group may have altered the behaviour of other members. For example, their presence may have led to better discussion in which more information was elicited from the other group members. Research has shown that when a group member voices a differing opinion (e.g., a devil’s advocate) this can help generate group discussion (e.g., Janis, 1972). Alternatively, participants may have steered the group discussion away from topics on which people did not agree, such as the misleading information, and instead focused on information that they agreed on, thus enhancing recall for non-misled items. Research on the collective information sharing bias suggests that there is a tendency for group discussions to focus on the reiteration of information that group members share in common, with little attention being paid to the exposure of information that is unique to individual members (Wittenbaum, Hubbell, & Zuckerman, 1999). Furthermore, when faced with a differing opinion, group members may have questioned their own recollections and attempted to place themselves back at the time of viewing the incident to refresh their memories. This self-imposed context reinstatement may have improved their memories for non-misled items because according to the encoding specificity principle, recall is enhanced when Chapter 5: Co-witness Contamination 118

conditions match as closely as possible, the conditions present during the learning (Tulving & Thomson, 1973). However, it is also possible that the confederates assisted the group too much because they were not limited concerning how much correct information they were allowed to provide about the video. That is, confederates were allowed to introduce correct information to the group, so they didn’t appear implausible, and this may have helped the group remember a point they would have forgotten otherwise. If the confederates were over-assisting the participants, this may have increased their credibility within the group, thus making their misinformation seem more plausible. Even though confederate-present groups did not report having had different discussion experiences from the participants in confederate absent groups, the possibility exists that the group experiences were different.

Future research in this thesis attempts to control for this confound. Study 3 also assessed the effect of time delay on susceptibility to postevent information. Because traditional studies investigating the misinformation effect (e.g., Loftus, Miller, & Burns, 1978) have shown that longer delays increase susceptibility to misinformation, and yet a study investigating memory conformity has failed to replicate this result (Hoffman et al., 2001), it was unclear what the effect of delay would be on susceptibility to misleading information. In the current study it was found that co-witness information for both misled and non-misled items had a particularly strong effect on the individual’s testimony when it was presented after a two-week delay compared to a delay of only twenty minutes. These findings are in line with previous research investigating the misinformation effect (e.g., Loftus, Miller, & Burns, 1978), however they differ from the results obtained by Hoffman et al. (2001). This difference may occur because a longer delay was used in the current study (two weeks) than Hoffman et al.’ s study (one day). Our final research aim was to investigate how Discussion Type and Time Delay after witnessing the event influenced the participants’ confidence in their memories for misled and non-misled items. The results from this experiment showed that participants in confederate-present groups were less confident for the misled items and more confident for the non-misled items than participants in the confederate-absent groups and individuals who did not discuss the event. There were no significant differences in confidence between the participants in the confederate-absent groups and individuals who did not discuss the event. Chapter 5: Co-witness Contamination 119

There was a significant interaction between Discussion Type and Time Delay for confidence on misled items, irrespective of accuracy. Confidence for misled items dropped dramatically after a two-week delay for participants in the confederate-present and no- discussion conditions. Although participants in the confederate-absent groups also had a decrease in confidence after a two-week delay, it was less pronounced. This interaction may be due to the fact that the participants in the confederate-absent groups were able to discuss the event with others who agreed with them. This discussion confirmed their beliefs about what happened and therefore their confidence did not decrease as dramatically. On the other hand, confidence dropped more dramatically for those in the confederate-present groups because they heard misinformation that was contrary to what they saw. Furthermore, confidence dropped more dramatically for those in the no-discussion control condition because they were unable to discuss the event with others who could verify what occurred in the video. One potential problem with the interpretation of the results from this study is that the recognition questions given to the participants before the discussion/writing exercise may have, themselves, been acting as a source of misleading information. That is, participants may have incorporated the incorrect statements they read in the initial questionnaire into their memories. For example, studies investigating the effect of leading questions on witness memory have shown that misleading questions can cause people to incorporate the misinformation into their memories (e.g., Loftus & Palmer, 1974; Loftus &

Zanni, 1975; Loftus, 1979a). Furthermore, the use of forced-choice recognition tests can be problematic because they can create demand characteristics (McCloskey & Zaragoza, 1985). Participants had no reason to doubt the misleading information presented to them by the confederate, so they may only have reported the misinformation because it was a possible response on the questionnaire. Thus, it is possible that the memory conformity evident in this study may simply have been due to a response bias inherent in the experiment. This study did not investigate why participants change their responses following the acquisition of co-witness information. Of the possible explanations for memory conformity discussed in Section 1, only normative social influence (i.e., conforming to gain social approval, Deutsch & Gerard, 1955) can be ruled out as an explanation of the results for this Chapter 5: Co-witness Contamination 120

study. This is because participants were asked to give individual accounts of the videos, separate from other group members. That is, other group members had no way of knowing what answers were given by the participant and therefore it is unlikely that the participant would yield to gain their social approval. However, beyond excluding normative social influence as a possible explanation for the memory conformity observed, this study did not examine whether participants change their responses because their original memory trace is altered or because of other factors, such as informational influence. This is a topic that is investigated in later studies reported in this thesis. Despite its limitations, this study has yielded some important findings. First, this study has demonstrated that confederate induced co-witness information does influence eyewitness memory. Participants who were misled by a confederate co-witness were less accurate on the misled items than those who didn’t receive the misinformation (i.e., confederate-absent groups and individuals not in a discussion group). Second, this study has also shown that the longer the delay, the stronger the effect of co-witness information. Furthermore, the inclusion of confederate-absent groups improves on previous methodologies because these groups potentially provide a more realistic environment for testing the impact of co-witness discussion. This study revealed that discussion in confederate-absent groups did not show any effect on memory (either positive or negative) when compared with individuals who did not discuss the event. That is, without experimentally induced information, there appear to be no advantages or disadvantages of co-witness discussion. The results from the current study have shown that participants who receive misinformation from a confederate are likely to report this misinformation when later questioned individually about the event. The next step is to examine the relative size of this effect, in comparison to other, more widely investigated methods of encountering postevent information. The following study described in this thesis (Study 4) compares the effects of Confederate Induced Information with information obtained from leading questions, a media report, and through the indirect transfer of co-witness information.

121

Chapter 6 Study 4: Ways of Encountering Postevent Information

As outlined in Section 1, Wright and Davies (1999) suggest that there are three ways in which an eyewitness might encounter postevent information about the crime: (a) being asked questions about the event by police officers, lawyers, friends, or others, (b) media coverage of the event (e.g., reading about the event in the newspaper), and (c) through co-witness information. Furthermore, there are different ways in which co-witness information may be passed from one eyewitness to another (Luus & Wells, 1994). Co- witness information may be transferred in 'direct' conversation between witnesses or 'indirectly' through a third party, such as a police officer that informs one witness about what another witness has said (see Figure 6.1).

Ways of Encountering Postevent Information

↓ Leading Co-Witnesses Media Questions Information

Direct Transfer Indirect (Co-witness Transfer discussion)

Figure 6.1. Ways of Encountering Postevent information

A great deal of psychological research has focused on how postevent information encountered through leading questions (e.g., Loftus & Palmer, 1974; Loftus & Zanni, 1975; Loftus, 1979a ) and postevent narratives such as media reports (e.g., Allen & Lindsay, 1998; Belli, Lindsay, Gales & McCarthy, 1994; Searcy, Bartlett, & Memon, 2000; Shaw, Garcia, & Robles, 1997; Sporer, 1995; Wright & Stroud, 1998) affects eyewitness memory. Chapter 6: Ways of Encountering Postevent Information 122

By comparison, little research has examined the impact of co-witness information and almost nothing is known about the relative impact of these different ways of encountering postevent information. Since a great deal of research has been done on misleading questions and postevent narratives, one might argue that researchers could generalize the results from these studies to the effects of co-witness information. However, this is not advisable for several reasons (Betz, Skowronski & Ostrom, 1996). First, an interviewer or media account might be regarded as a more credible source than a co-witness. Eyewitnesses may be more likely to believe information from someone in a position of authority than a fellow witness. Research has shown that the misinformation effect is weaker when the source has low credibility, than if it has high credibility (Underwood & Pezdek, 1998). Therefore, witnesses may ignore the co-witness information or even consciously recognise the errors and disregard them (Roediger, Meade, & Bergman, 2001). Second, in order to formulate a response, participants in eyewitness research are motivated to pay close attention to the information presented to them in leading questions (Betz et al., 1996). Under circumstances in which such attention is not required (such as casual conversation with a co-witness), misleading information may not have the same effect. Thus, as Betz et al. state, "Although the literature on eyewitness testimony is certainly consistent with the notion that information obtained from second-hand sources can affect recall for the original stimulus information, it would be a mistake to generalize such findings" (1996, p. 115). Therefore, the effects of co-witness information need to be examined in their own right and compared to other ways of encountering misinformation. While the majority of the literature on postevent information has concentrated on the effects of leading questions and media reports, some studies have shown that misleading information presented through the indirect transfer of co-witness information (e.g., Betz et al., 1996; Garven, Wood, & Malpass, 2000; Loftus & Greene, 1980; Shaw, Garven, & Wood, 1997, Experiment 1) and discussion with a co-witness (e.g., Gabbert, Memon, Allan, 2003; Gabbert, Memon, Allan, & Wright, 2004; Hoffman, Granhag, See, & Loftus, 2001; Hollin & Clifford, 1983; Shaw et al., 1997; Wright, Self & Justice, 2000) may also cause witnesses to report this misinformation. For example, in Study 3 it was found that exposure to erroneous information encountered through discussion with a Chapter 6: Ways of Encountering Postevent Information 123

confederate co-witness would cause people to incorporate this misinformation into their statements. Study 4 was conducted in order to compare the impact of postevent information encountered through co-witness discussion with other ways of encountering postevent information. The other ways of encountering postevent information compared in the current study were leading questions, media reports, and indirect transfer of co-witness information. In particular, the purpose of this study was to determine which of these ways of encountering postevent information has the strongest impact on the accuracy of participants' memory for the original event. In order to do this, all experimental factors were held constant with the exception of the source of postevent information. Very few studies have compared the relative impact of different ways of encountering postevent information. Shaw et al. (1997) have conducted three experiments comparing the effects of co-witness information on memory with those of suggestive questions. In Experiment 1, the co-witness information came from multiple anonymous co- witnesses (i.e., indirect co-witness information). Participants in Experiment 1 were asked to fill out an anonymous questionnaire regarding the names of people involved in a highly publicised murder trial. Answers supposedly given by previous participants were displayed on the questionnaire. In Experiments 2 and 3 the information was from a single identified co-witness in a face-to face encounter (i.e., co-witness discussion). Participants were asked to answer questions about a crime video they had seen in the presence of a confederate. The interviewer called on them to answer one at a time and would alternate who answered first. In Experiment 3, participants were asked to return to the lab two days later to be questioned about the incident individually. In all three experiments the detrimental effects of co- witness information were larger and more robust than those of suggestive questioning. In another study, Gabbert et al. (2004) compared the effects of postevent information encountered through co-witness discussion with those encountered through indirect co-witness information. In their study, participants were shown a video of a simulated robbery and they then received postevent information either by discussing the video with a confederate or reading a narrative supposedly given by a previous participant. The results from this experiment showed that co-witness discussion was significantly more misleading than the indirect co-witness information. Chapter 6: Ways of Encountering Postevent Information 124

Meade and Roediger (Experiment 4, 2002) conducted a similar study to that of Gabbert et al. (2004), however they found different results. Meade and Roediger (Experiment 4, 2002) showed participants pictures of complex scenes and later had them recall items from the scenes either with a ‘live’ confederate (i.e., direct co-witness discussion) or a ‘virtual confederate’ (i.e., indirect co-witness information). In the virtual confederate condition, participants read index cards with responses that had supposedly been written by previous participants. Finally, participants were asked to recall the items individually. In this study, the method of encountering co-witness misinformation (i.e., through virtual or live confederates) did not affect the number of false answers given on a cued recall test. There are some differences between the current study and those previously conducted. In the study described in this paper, the indirect co-witness information is operationalised slightly differently from those conducted by Gabbert et al. (2004) and Meade and Roediger (Experiment 4, 2002). Participants in the current study were given sentences that had supposedly been written by other people who had viewed the same video. Participants were then asked to state whether they agreed or disagreed with each sentence. This is analogous to a situation in which a police officer tells a witness what another witness has said and asks for his/her opinion. It is unlikely that in real life an officer would give a complete witness statement to another witness to read, especially without asking them to state their opinion of it. By asking the witnesses to indicate whether or not they agree with the statement, this may shed some doubt about the accuracy of the statement. Therefore it is possible that the indirect co-witness information encountered this way may not be as influential as when witnesses are not asked to state their opinion. On the other hand, when witnesses are asked to state whether or not they agree with the statement, they are forced to attend to the statement more carefully and this may result in a greater influence than if they just read the information. In the current study, the term “co-witness discussion” is also operationalised slightly differently from the Shaw et al. (1997) studies. Participants in the current study freely discussed the crime with a confederate, as they did in the Gabbert et al. (2004) study rather than answer an interviewer’s questions in the presence of a confederate co-witness. This is a more realistic reflection of a real life situation because witnesses often freely Chapter 6: Ways of Encountering Postevent Information 125

discuss the event with one another (as indicated by Study 1). Furthermore, witnesses are often questioned individually (as indicated by Study 2) and not in the presence of others. The possibility exists that co-witness information encountered in a casual, free discussion may not be as influential as when encountered through formal questioning in the presence of a co-witness. This is because when questioning a participant in the presence of a confederate, witnesses may feel greater pressure to conform and when tested later individually they may feel pressure to give answers consistent with their previous ones. A further difference of the current study is that most previous research seeks to identify the influence of misinformation presented by one witness to another. Therefore, the assumption is made that discussion between witnesses is a negative, detrimental process and no attempt is made to measure any possible positive effects of discussion. This approach is inappropriate because it is possible that memory conformity can have positive as well as negative effects on memory. For example, if one person has poor memory for an event, discussing it with a co-witness who remembers more may help to cue the first person’s memory. Some experimental research (Roediger et al., 2001) suggests that co- witness discussion may increase memory accuracy, however appropriate counterbalancing was not used, so caution should be taken when interpreting the findings. Another study has shown that both leading (i.e., correct) and misleading (i.e., incorrect) questions can affect memory accuracy, increasing and decreasing it respectively (Sutherland & Hayne, 2001). In the current study, we look at the effects of correct postevent information as well as misleading postevent information. It may be the case that correct postevent information can cue a witness’s memory resulting in an increase in both information that was cued and non- cued information. Despite the differences between the current study and those previously conducted, the findings from the preceding studies were used to formulate our hypotheses. Based on the findings of Shaw et al. (1997), it was hypothesised that co-witness information (whether encountered indirectly or directly through co-witness discussion) would have a stronger influence on the participants' accuracy than the leading questions. It was unclear how accurate participants in the co-witness discussion condition would be in comparison to those in the indirect transfer condition because previous studies have found mixed findings (Gabbert et al., 2004; Meade & Roediger, 2002). No predictions were made regarding the Chapter 6: Ways of Encountering Postevent Information 126

relative effects of the media article to those of leading questions and co-witness information because this comparison has not been made previously. In accordance with the previous research on leading questions (e.g., Loftus & Palmer, 1974; Loftus & Zanni, 1975; Loftus, 1979a), postevent narratives (e.g., Allen & Lindsay, 1998; Belli et al., 1994; Searcy et al., 2000; Shaw, Garcia, et al., 1997; Sporer, 1995; Wright & Stroud, 1998), indirect co-witness information (e.g., Betz et al., 1996; Garven at al., 2000; Loftus & Greene, 1980; Shaw, Garven et al., 1997, Experiment 1) and co-witness discussion (e.g., Gabbert et al., 2003; Gabbert et al., 2004; Hoffman et al., 2001; Hollin & Clifford, 1983; Shaw, Garven, et al., 1997; Wright et al., 2000), it was hypothesised that the participants receiving postevent information would be more likely to report this information than controls who received no postevent information. In addition to assessing the accuracy of participants’ memories, the current study also investigates how confident the witnesses are about their memories. It is important to assess eyewitness confidence because a witness’s confidence may influence future events in the legal process (Wells et al., 2000). For example, confidence levels may affect a witness’s willingness to contact the police or to testify in court. Furthermore, it may also affect the way other people (such as police investigators or jury members) view the reliability of the witness statement. Research investigating levels of confidence has found that misled participants have as much confidence in their often-mistaken recognition judgments as non-misled participants who are considerably more accurate (Loftus, Donders, Hoffman, & Schooler, 1989). However, previous studies that have compared different ways of encountering postevent information have not assessed eyewitness confidence (Gabbert et al., 2004; Meade & Roediger, 2002; Shaw, Garven, et al., 1997). The current study aims to expand our understanding of the effects of postevent information by providing information about the confidence in their memories that witnesses have, in addition to their likelihood of reporting the postevent information. Chapter 6: Ways of Encountering Postevent Information 127

Method

Design There were three phases in the study: exposure, postevent information, and individual recall. During the exposure phase, participants were shown a video of a crime scene. In the postevent information phase, participants received postevent information about the video in one of four different ways: (1) leading questions, (2) media report, (3) indirect co-witness information, or (4) co-witness discussion. There was also a control condition in which participants received no postevent information. During the recall phase, participants were asked to give a free recall account of the event shown on the video and then complete prompted recall (short answer) and recognition (true or false) questionnaires. Some of the items on the questionnaires assessed participants’ memories for items for which they had received correct postevent information, others for items for which they had received misleading postevent information, and others for items for which they had received no postevent information. This study investigated the effects of five Mode of Postevent Information conditions (leading questions, media report, indirect co-witness information, co-witness discussion, no postevent information). The dependent variables were memory accuracy and confidence on questions for which the participants had received either correct, misleading, or no postevent information.

Participants Participants were 105 undergraduate Psychology students (81 females, 24 males) from the University of New South Wales. The average age of the participants was 19.7 (SD=3.6) years. Participants received credit for their psychology course upon completion of the study. Chapter 6: Ways of Encountering Postevent Information 128

Materials Eyewitness Stimulus. The eyewitness stimulus used in this study was a short video depicting a store robbery. The store robbery was the same scene extracted from the film, “The Arrival” (Schmoeller, 1990) used in Study 3. General Knowledge Test. A general knowledge test was included in the study as a filler task. The test contained 100 questions. Sample items include: What is 54 divided by 9; When was the Sydney Opera House opened; What is the largest human organ; How many days are there in January? Participants were given ten minutes to complete as many questions as they could and to guess if they did not know the answer to any items. Postevent Information. Table 6.1 summarises postevent information given to the participants. Four pieces of information given to the participants were misleading (i.e., contrary to what happened in the video) and four pieces were correct (i.e., true to what happened in the video). The postevent information was given to participants in one of four ways (leading questions, media report, indirect co-witness information, co-witness discussion), depending on the condition. There was also a control condition that did not receive any postevent information. For the leading question condition, postevent information was presented as a series of written questions that included the correct or misleading information (see Appendix E). For example, the video showed the thief wearing a navy blue bandana, and one of the misleading questions was, “Did the thief wearing the red bandana shoot the gun?” The participants were required to indicate whether each statement was true or false. The media report was presented as an article in a mock newspaper (see Appendix F). The newspaper article included only the correct and misleading postevent information. Participants were required to read the article. Indirect co-witness information was given to participants in the form of sentences that had supposedly been written by other people who had viewed the same video (see Appendix G). No information regarding the credibility of the other people was given. Participants were asked to state whether they agreed or disagreed with each statement. Chapter 6: Ways of Encountering Postevent Information 129

For the co-witness discussion condition, participants discussed the event with an experimental confederate who was trained to provide the postevent information. Ten confederates were used to decrease the possibility that the results obtained would be specific only to distinct characteristics of one confederate. The confederates were male and female university students who were around the same age as the participants in the study, but unknown to the participants. The confederates were instructed to take part in the experiment as though they were actual participants. They were instructed to introduce the eight items of postevent information into the discussion in any order they liked. They were not allowed to add any other details to the discussion. If someone questioned a confederate about what they said, confederates were instructed to reply, "Well, that's what I remember seeing on the video." They were instructed to reply confidently, without being overly assertive.

Table 6.1. Correct and Misleading Postevent Information Used in Study 4 Postevent Information Information Type The thief that shot the gun was named “Chuck.” Misleading. His name was Hank. The thief that shot the gun was wearing a red bandana. Misleading. The bandana was navy blue. The thief who pointed the handgun at the lady kicked her to get Misleading. He yells at her away from the counter. her, but doesn’t kick her. The woman in the store was wearing a large brooch. Misleading. She was wearing a large necklace, but no brooch. The name of the store was “Chalet Liquor and Deli”. Correct

The woman in the store was wearing a peach-coloured floral Correct coat. The thief who pointed the handgun at the lady was wearing a Correct white singlet top. The thief who pointed the handgun at the lady had blonde hair. Correct

Chapter 6: Ways of Encountering Postevent Information 130

Free Recall Form. A Free Recall Form was developed for this study. Participants were instructed to write down as much as they could remember about the crime video. Specifically, they were asked to write about the sequence of events, what was said, the setting, what the people looked like, what was stolen, and any other details they could remember. Short Answer Questionnaire. The short answer questionnaire consisted of 14 questions about the video (see Appendix H). Four of the questions were concerning details of the event about which some participants had received correct postevent information. Conversely, four other questions concerned details of the event about which some participants received misleading information. For example, participants were asked, “What was the name of the thief who shot the gun?” Participants were asked to answer the questions and rate their confidence in this answer on a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” Recognition Questionnaire. The recognition questionnaire about the store robbery was the same 20-item questionnaire used in Study 3. Of the twenty statements, twelve were true and eight were false. Four of the true statements concerned details of the event about which some participants had received correct postevent information. Conversely, four of the false statements concerned details of the event about which some participants received misleading information. An example of a false statement for which some participants received misleading information is: “The thief that shot the gun was named Chuck,” (when in fact he was identified as ‘Hank’ in the video, but the postevent information had referred to him as ‘Chuck’). For a full description of the correct and misleading postevent information, please refer to Table 6.1. The other 12 statements related to events about which no postevent information had been provided. For each statement, participants were asked to indicate whether the statement was true or false and indicate their confidence on a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” Experimental Manipulation Check. A questionnaire was developed to assess whether participants (except those in the control group) were aware of the experimental manipulation. In particular they were asked, “Did you find the [e.g., media report] Chapter 6: Ways of Encountering Postevent Information 131

interesting and useful”; “Did you learn anything new about the episodes as a result of the [e.g., media report]”; and Do you think your answers were influenced by the [e.g. media report]? In addition, participants in the co-witness discussion condition were asked, “Did your group have an open and free exchange of ideas?” Participants were asked to indicate their agreement with each question using a six-point scale where a score of “1” indicated “Yes” and “6” indicated “No.” All participants were asked if they had discussed the video with anyone else who had participated in the experiment. They were also asked what they thought the purpose of the study was, and whether they had ever seen the movie that the scene came from.

Procedure This experiment took place over two sessions, one week apart. In the first session, participants were invited to take part in an experiment on eyewitness memory. Participants were told that their participation was completely voluntary and that they were free to withdraw from the study at any point. After informed consent was obtained (see Appendix I), participants were shown the robbery video. Prior to viewing the video, participants were instructed, “I will be showing you a short video of a crime scene. Imagine that this is a real life crime and that you are actual witnesses to the crime. Like real witnesses, you will be asked questions about the crime later, so please pay close attention to the video.” After viewing the video, a meeting was scheduled with the participants for the following week at the same time. Participants were asked not to talk about the video with anyone else. Participants completed this phase either individually or in a group of up to six people. After a delay of one week, participants returned to the psychology laboratory and were randomly assigned to one of the five Postevent Information conditions. That is, they either encountered postevent information about the crime through leading questions, a media report, statements supposedly written by a co-witness, discussion with a co-witness (who was an experimental confederate) or received no postevent information. The discussions in the co-witness discussion condition were tape-recorded and these recordings confirmed that all of the confederates had introduced the information required, and had not provided any other information. Chapter 6: Ways of Encountering Postevent Information 132

After receiving the postevent information, participants were given ten minutes to complete the General Knowledge Test as a filler task. Upon completion of the filler task, participants were given a series of questionnaires to assess their memory for the event. First, they were given as much time as they needed to complete the Free Recall Form. Next, participants were given as much time as they needed to complete the Short Answer Questionnaire followed by the Recognition Questionnaire. Finally, participants were given the questions that made up the Experimental Manipulation Check. When participants had completed all of the questionnaires, they were fully debriefed about the study, thanked for their participation, and asked not to discuss the experiment with other potential participants.

Results Statistical analysis concentrated on the following questions: First, how successful was the experimental manipulation? Second, how did Mode of Postevent Information (leading questions, media report, indirect co-witness information, co-witness discussion, no postevent information) influence the accuracy of the participants’ memories for the event? Finally, how did Mode of Postevent Information influence the participants’ confidence in their memories?

Experimental Manipulation Check Separate one-way ANOVAs were conducted for each of the three questions relating to the participants’ perceptions of the postevent information. Participants were asked to indicate their agreement with each question using a six-point scale where a score of “1” indicated “Yes” and “6” indicated “No.” The individuals in the no-postevent information control group were not given these questions, so were not included in the analysis. The mean scores for these experimental manipulation questions are shown in Table 6.2. Responses to the questions, “Did you find the [e.g., media report] interesting and useful” and “Do you think your answers were influenced by the [e.g. media report]” did not show a main effect for Mode of Postevent Information (F (3, 81) = 0.84, p = .48 and F (3, 81) = 0.45, p = .72 respectively). However, for the question, “Did you learn anything new about the episode as a result of the [e.g., media report]?” there was a significant effect of Mode of Chapter 6: Ways of Encountering Postevent Information 133

Postevent Information, F (3, 81) = 4.45, p < .01. LSD post hoc tests revealed that participants in the co-witness discussion condition (M = 1.55, SD = 1.01) were more likely to report that they had learned something new than participants in the leading question (M = 2.90, SD = 1.67), media report (M = 3.14, SD = 1.90), and indirect transfer (M = 2.86, SD = 1.68) conditions, all p’s < .01

Table 6.2

Response to Experimental Manipulation Check

Question Condition M SD n

Leading Questions 2.90 1.14 21

Did you find the [e.g., Media Report 3.24 1.48 21 media report] interesting Co-witness and useful? 2.59 1.05 22 Discussion Indirect Transfer 3.00 1.67 21

Leading Questions 2.43 1.60 21

Do you think your answers Media Report 2.90 1.95 21 were influenced by the Co-witness [e.g., media report]? 2.45 1.47 22 Discussion Indirect Transfer 2.43 1.33 21

Leading Questions 2.90 1.67 21 Did you learn anything Media Report 3.14 1.90 21 new about the episode as a result of the [e.g., media Co-witness 1.55 1.01 22 report]? Discussion Indirect Transfer 2.86 1.68 21 Note. Responses were given on a six-point scale in which a score of “1” signified “Yes” and “6” signified “No.”

For participants who discussed the event with a co-witness, separate one-way ANOVAs were conducted to investigate whether the ten confederates differently influenced the participants’ responses to each of the three questions relating to their perception of the Chapter 6: Ways of Encountering Postevent Information 134

group discussions. It was found that there were no differences between the groups who had different confederates on responses to the first question, “Did you find the group discussion interesting and useful?” (F (9, 12) = 1.03, p = 0.47). Similarly, there were no differences between the groups who had different confederates on responses to the question, “Do you think your answers were influenced by the group discussion” (F (9, 12) = 1.12, p = 0.42). For the question, “Did you learn anything new about the episodes as a result of the discussions?” there were no differences between the groups who had different confederates (F (9, 12) = 0.81, p = .62). Finally, there were no differences between the groups who had different confederates on responses for the question “Did your group have an open and free exchange of ideas?” (F (9, 12) = 1.46, p = .27). Furthermore, separate one-way ANOVAS were conducted to investigate whether the ten confederates influenced the participants’ confidence and accuracy differently. It was found that there were no effects of the different confederates on accuracy and confidence for questions for which they received correct, misleading, and no postevent information. That is, there were no differences between the groups who had different confederates on accuracy for items for which they received misleading postevent information (F (9, 12) = 0.85, p = .59) correct postevent information (F (9, 12) = 0.69, p = .71), and no postevent information (F (9, 12) = 0.95, p = .52). Similarly, there were no differences between the groups who had different confederates on confidence for items for which they received misleading postevent information (F (9, 12) = 1.35, p = .31) correct postevent information (F (9, 12) = 1.40, p = .29), and no postevent information (F (9, 12) = 0.57, p = .80) All participants were asked if they had seen the movie from which the film clip had been extracted. None of the participants reported being familiar with the scene or the movie, so nobody was excluded from the study for this reason. Participants were asked if they had discussed the video with anyone else who had participated in the experiment (outside the experimental discussions). Eight participants indicated that they had discussed details of the study with another participant (four participants from the indirect transfer condition, two from the control condition, one from the media report condition, and one from the co-witness discussion condition). A chi-square analysis revealed that there was no association between Mode of Postevent Information and whether or not the participants reported discussing the study with another participant (χ2 (4, Chapter 6: Ways of Encountering Postevent Information 135

N = 105) = 6.33, p = .18). When the eight participants who reported discussing the videos were removed from the analyses, the results were not greatly affected, so all participants were included in the analyses reported. Where differences do exist between the two analyses, they are acknowledged in footnotes. Finally, participants were asked what they thought the purpose of the study was, and their answers were analysed to determine if the participants were aware of the nature of the experimental manipulation. Thirteen participants were considered aware of the manipulation, as they reported that the purpose was something to do with investigating whether people would be influenced by incorrect information given to them. A chi-square analysis revealed that there was an association between Mode of Postevent Information and whether or not the participants were aware of the purpose of the study (χ2 (4, N = 105) = 13.70, p < .01). Inspection of the contingency table suggests that the leading question and indirect transfer conditions were more transparent to the participants than the co-witness discussion, media report, and control conditions. When these thirteen participants were removed from the analyses, this had no great effect on the results, so all participants were included in the analyses reported. Where differences do exist between the two analyses, they are acknowledged in footnotes.

Recognition Questionnaire Memory Accuracy. Participants’ responses to the recognition questions were recoded in such a way that a correct answer was scored “1” and an incorrect answer scored “0.” Scores were then added together to get separate memory accuracy scores for items for which they received correct, misleading, or no postevent information. Accuracy scores for the correct and misleading information were both divided by four because there were four questions pertaining to each. Scores for the items in which participants received no postevent information were divided by twelve because there were twelve pertinent questions on the questionnaire. All results were then rescaled in the range 0-100. A one-way MANOVA was conducted in which the effects of Mode of Postevent Information (leading questions, media report, indirect co-witness information, co-witness discussion, no postevent information), on memory accuracy (correct, misleading, no Chapter 6: Ways of Encountering Postevent Information 136

postevent information) were analysed. Following the procedure recommended by Tabachnick and Fidell (2001), before conducting the MANOVA, the three dependent variables were correlated to check for possible multicollinearity. It was found that the dependent variables were not unacceptably highly correlated (r’s ranged from -.20 to .13, all p’s > .05) and it was therefore appropriate to proceed with the MANOVA. The MANOVA revealed a significant effect of Mode of Postevent Information on the combined dependent variable of memory accuracy, F (12, 244) = 3.87, p < .0005; Wilks’ Lambda = .6; partial eta squared = .14). The mean scores for memory accuracy are shown in Table 6.3. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .017, showed that there was no effect of Mode of Postevent Information on items for which the participants received no postevent information, F (4, 94) = 1.05, p = .38. There was a significant effect of Mode of Postevent Information on accuracy for items in which participants received correct postevent information, F (4, 94) = 6.43, p < .0005. Employing the LSD post-hoc test revealed that participants in the control group (M = 56.9, SD = 29.5) were significantly less accurate than participants who received correct postevent information in the leading question (M = 75.0, SD = 22.0, p < .05), media report (M = 72.5, SD = 21.3, p < .05)5, co-witness discussion (M = 89.8, SD = 14.8, p < .0005), and indirect transfer conditions (M = 85.0, SD = 22.1, p < .0005). Furthermore, it was also found that participants in the co-witness discussion condition were more accurate than those who received correct postevent information from leading questions (p < .05)6 and media reports (p < .05). There were no significant differences between any of the other conditions. There was a significant effect of Mode of Postevent Information on accuracy for items in which participants received misleading postevent information, F (4, 94) = 4.81, p < .01. Post Hoc tests with an LSD correction revealed that participants who received misleading postevent information through co-witness discussion were significantly less

5 When participants who had discussed the experiment with another participant were excluded, participants who received correct information from the media report were no more accurate (M = 73.7, SD = 21.2) than the controls who didn’t receive any postevent information (M = 59.4, SD = 27.2, p =.06). 6 When participants who had guessed the purpose of the study were excluded from the analyses, participants who received correct co-witness information from a confederate were no more accurate (M = 89.8, SD = 14.8) than those who received the information from leading questions (M = 76.8, SD = 22.9, p =.10). Chapter 6: Ways of Encountering Postevent Information 137

accurate (M = 21.6, SD = 23.5) than the control group (M = 51.4, SD = 25.0, p < .0005). They were also significantly less accurate than participants who received misleading postevent information from leading questions (M = 43.4, SD = 24.8, p < .01) and the media report (M = 50.0, SD = 28.1, p < .0005). There were no significant differences between any of the other conditions. 7

Table 6.3 Mean Memory Accuracy for Recognition Test

Correct Misleading No Postevent Condition Postevent Postevent Information Information Information M SD M SD M SD Leading 64.9 15.4 75.0 22.0 43.4 24.8 Questions Media 58.8 16.6 72.5 21.3 50.0 28.1 Report Co-witness 64.4 14.6 89.8 14.8 21.6 23.5 Discussion Indirect 65.0 13.7 85.0 22.1 36.3 25.0 Transfer Control 68.1 11.5 56.9 29.5 51.4 25.0

Memory Confidence. Participants stated their confidence on a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” Scores were added together to get separate confidence scores for items for which they received correct, misleading, or no postevent information8. Scores for the correct and misleading information were divided by four because there were four questions pertaining to each on the recognition questionnaire. Scores for the items in which participants received no postevent information were divided by twelve because there were twelve pertinent questions.

7 When participants who had guessed the purpose of the study were excluded from the analyses, participants who received misleading information through indirect transfer were less accurate (M = 32.8, SD = 21.8) than the controls who didn’t receive any postevent information (M = 51.4, SD = 25.0, p < .05). 8 In this analysis, confidence scores were pooled for ease of analysis, however some researchers suggest that each item should be analysed separately (e.g., Wheatcroft, Wagstaff, & Kebbell, 2004). Chapter 6: Ways of Encountering Postevent Information 138

After establishing that the dependent variables for memory confidence were not unacceptably highly correlated (r’s ranged from .36 to .58, all p’s < .01) a MANOVA was conducted in which the effects of Mode of Postevent Information on items for which participants received correct, misleading, and no postevent information were analysed. The mean scores for memory confidence are shown in Table 6.4. There was a significant effect of Mode of Postevent Information on the combined dependent variable of memory confidence, F (12, 246) = 3.26, p < .0005; Wilks’ Lambda = .7; partial eta squared = .12). Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .017, showed that there was no effect of Mode of Postevent Information on items for which the participants received no postevent information, F (4, 95) = 2.12, p = .08. There was a significant effect of Mode of Postevent Information on confidence for items in which participants received correct postevent information, F (4, 95) = 4.08, p < .01. Employing the LSD post-hoc test revealed that participants in the co-witness discussion condition were more confident (M = 4.39, SD = 0.88) than participants who received correct postevent information in the leading question condition (M = 3.63, SD = 0.88 p < .01) and the control group (M = 3.39, SD = 0.79, p < .01). It was also found that participants in the media condition were more confident when they received correct postevent information (M = 4.20, SD = 0.85) than the control group (p < .01). There were no significant differences between any of the other conditions.9 There was a significant effect of Mode of Postevent Information on accuracy for items in which participants received misleading postevent information, F (4, 95) = 6.45, p < .0005. Post Hoc tests with an LSD correction revealed that participants who received misleading postevent information through co-witness discussion were significantly more confident (M = 4.45, SD = .89) than the control group (M = 3.16, SD = 0.96, p < .0005). They were also significantly more confident than participants who received misleading postevent information from leading questions (M = 3.59, SD = 1.15, p < .01) and the indirect transfer of co-witness information (M = 3.48, SD = 0.95, p < .01). Participants who

9 When participants who had guessed the purpose of the study were excluded from the analyses, participants who received correct co-witness information from a confederate were more confident (M = 4.39, SD = 0.89) than those who received the co-witness information through indirect transfer (M = 3.78, SD = 1.05, p <.05).

Chapter 6: Ways of Encountering Postevent Information 139

received misleading postevent information from the media source were significantly more confident (M = 4.24, SD = .88) than the control group (p < .01). They were also significantly more confident than participants who received misleading postevent information from leading questions (p < .05) and the indirect transfer of co-witness information (p < .05). There were no significant differences between any of the other conditions.

Table 6.4

Mean Memory Confidence for Recognition Test

Correct Misleading No Postevent Condition Postevent Postevent Information Information Information M SD M SD M SD Leading 3.93 0.64 3.63 0.88 3.59 1.15 Questions Media 4.41 0.51 4.20 0.85 4.24 0.88 Report Co-witness 3.96 0.77 4.39 0.88 4.45 0.89 Discussion Indirect 4.14 0.57 3.91 1.07 3.48 0.95 Transfer Control 3.95 0.63 3.39 0.79 3.16 0.96

Confidence in Correct and Incorrect Decisions. Confidence was also calculated so that it reflected confidence in correct responses separately from incorrect responses. Therefore, in this analysis there were six dependent variables (confidence for correct decisions when given correct information, confidence for incorrect decisions when given correct information, confidence for correct decisions when given misleading information, confidence for incorrect decisions when given misleading information, confidence for correct decisions when given no postevent information, confidence for incorrect decisions when given no postevent information). For this analysis, the N’s were quite low because it is not possible to calculate the confidence scores for correct or incorrect decisions for participants who had not answered any questions correctly or incorrectly. Chapter 6: Ways of Encountering Postevent Information 140

After establishing that the dependent variables for confidence in correct and incorrect decisions were not unacceptably highly correlated (r’s ranged from 0.13 to 0.56) a one-way MANOVA was conducted to investigate the effect of the five Modes of Postevent Information on the six dependent variables. The mean scores for confidence in correct and incorrect decisions are shown in Tables 6.5 and 6.6. The MANOVA revealed that there was a significant effect of Mode of Postevent Information on the combined dependent variable of confidence in correct and incorrect decisions, F (24, 130) = 1.62, p < .05; Wilks’ Lambda = .4; partial eta squared = .20). Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .008, showed that there was a significant effect of Mode of Postevent Information on confidence in incorrect decisions for misled items, F (4, 42) = 7.67, p < .0005. Post Hoc tests with an LSD correction revealed that participants who received misleading postevent information through co-witness discussion were significantly more confident in their incorrect decisions (M = 4.81, SD = 0.93) than the control group (M = 2.71, SD = 1.29, p < .01). They were also significantly more confident in their incorrect decisions than participants who received misleading postevent information through the indirect transfer of co-witness information (M = 2.67, SD = 0.92, p < .01). Furthermore, participants who received misleading postevent information from the media report were significantly more confident in their incorrect decisions (M = 4.79, SD = 1.02) than the control group (p < .0005) and participants who received incorrect information through the indirect transfer of co-witness information (p < .01). Finally, participants who received misleading postevent information from leading questions were significantly more confident in their incorrect decisions (M = 3.80, SD = 1.39) than the control group (p < .05). There were no significant differences between any of the other conditions. There were no significant effects of Mode of Postevent Information any of the other dependent variables relating to confidence in correct and incorrect decisions. That is, there was no effect of Mode of Postevent Information on confidence for correct decisions when given correct information (F (4, 42) = 2.25, p = .08), confidence for incorrect decisions when given correct information (F (4, 42) = 1.11, p = .36), confidence for correct decisions when given misleading information (F (4, 42) = 1.48, p = .23), confidence for correct Chapter 6: Ways of Encountering Postevent Information 141

decisions when given no postevent information (F (4, 42) = 2.55, p = .05), or confidence for incorrect decisions when given no postevent information (F (4, 42) = 0.87, p = .49).

Table 6.5 Mean Confidence in Correct Decisions

Correct Misleading No Postevent Condition Postevent Postevent Information Information Information n M SD M SD M SD Leading 9 4.40 0.76 3.96 1.35 2.91 1.72 Questions Media 12 5.02 0.59 4.56 1.06 4.32 1.29 Report Co-witness 6 4.13 0.52 4.58 0.81 2.86 1.27 Discussion Indirect 6 4.40 0.59 3.53 1.71 3.42 1.91 Transfer Control 14 4.46 0.72 3.45 0.78 3.39 1.51

Table 6.6

Mean Confidence in Incorrect Decisions

Correct Misleading No Postevent Condition Postevent Postevent Information Information Information n M SD M SD M SD Leading 9 3.03 1.03 2.56 1.49 3.80 1.39 Questions Media 12 3.64 1.15 3.35 1.19 4.79 1.02 Report Co-witness 6 3.47 1.36 2.00 1.26 4.81 0.93 Discussion Indirect 6 2.96 1.24 2.58 1.80 2.67 0.92 Transfer Control 14 2.88 1.20 2.90 1.30 2.71 1.29

Chapter 6: Ways of Encountering Postevent Information 142

Short Answer Questionnaire Short Answer Coding. Participants’ short answers were coded for accuracy using a standard coding system and scored by two independent judges. The short answers were scored based on whether they (a) accurately reported the correct postevent information they were given, (b) inaccurately reported the misleading postevent information they were given, and (c) accurately reported information for questions about which they received no postevent information. There were four questions regarding the correct postevent information, four regarding the misleading postevent information, and six regarding items for which they received no postevent information. Scores were added together to give separate scores for items for which participants received correct, misleading, and no postevent information. Two independent judges repeatedly viewed the crime video and independently coded the short answer questions. The judges were unaware of the conditions to which the participants were assigned. Each participant’s responses were randomly assigned to one of the two coders. Twenty of the short answer questionnaires were scored by both coders in order to assess inter-rater reliability. Their scoring was found to be reliable with one another on every measure. That is, there was a significant correlation between the two coders on the items for which participants were given correct postevent information (r = 0.90, n = 20, p < 0.01, two-tailed), misleading postevent information (r = 0.80, n = 20, p < 0.01, two-tailed), and no postevent information (r = 0.97, n = 20, p < 0.01, two-tailed). According to Elmes, Kantowitz & Roediger (1991), high inter-rater reliability occurs when the correlation coefficient is greater than +0.70. All of the above correlations met this criterion. Short Answer Memory Accuracy. After establishing that the dependent variables for memory accuracy were not unacceptably highly correlated (r’s ranged from 0.10 to .20) a one-way MANOVA was conducted in which the effects of Mode of Postevent Information (leading questions, media report, indirect co-witness information, co-witness discussion, no postevent information) on memory accuracy for items for which participants were given correct, misleading, and no postevent information were analysed. The mean scores for memory accuracy are shown in Table 6.7. Chapter 6: Ways of Encountering Postevent Information 143

There was a significant effect of Mode of Postevent Information on the combined dependent variable of accuracy, F (12, 257) = 4.92, p < .0005; Wilks’ Lambda = .6; partial eta squared = .17). Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .017, showed that there was no effect of Mode of Postevent Information on items for which the participants received no postevent information, F (4, 99) = 0.98, p = .42. There was a significant effect of Mode of Postevent Information on accuracy for items in which participants received correct postevent information, F (4, 99) = 3.85, p = .006. Employing the LSD post-hoc test revealed that participants who were in the control group were significantly less likely to report correct postevent information (M = 0.85, SD = 0.75) than those who received the correct postevent information through the media report (M = 1.50, SD = 1.19, p < .05), co-witness discussion (M = 1.73, SD = 0.83, p < .05), and indirect transfer conditions (M = 1.76, SD = 1.09, p < .05). Furthermore, it was also found that participants in the co-witness discussion condition were more likely to report correct postevent information than those who received correct postevent information from leading questions (p < .05). Similarly, participants who received correct postevent information through indirect transfer were more likely to report correct postevent information than those who received correct postevent information from leading questions (p < .05)10. There were no significant differences between any of the other conditions. There was a significant effect of Mode of Postevent Information on accuracy for items in which participants received misleading postevent information, F (4, 99) = 10.74, p < .0005. Post Hoc tests with an LSD correction revealed that participants who were in the control group were significantly less likely to report misinformation (M = 0.30, SD = 0.47) than those who received misleading postevent information through co-witness discussion (M = 2.23, SD = 1.19, p < .0005), leading questions (M = 1.33, SD = 1.02, p < .05), the media report (M = 1.45, SD = 1.05, p < .0005), and indirect transfer (M = 1.48, SD = 0.87, p < .005). Participants who received misleading postevent information through co-witness discussion were also significantly more likely to report this misinformation than

10 When participants who had discussed the experiment with another participant were excluded, participants who received correct information through indirect transfer were no more likely to report the correct postevent information (M = 1.65, SD = 1.17) than those who received it through leading questions (M = 1.10, SD = 0.70, p =.07). Chapter 6: Ways of Encountering Postevent Information 144

participants who received misleading postevent information from leading questions (p < .01), indirect transfer (p < .05), and the media report (p < .05). There were no significant differences between any of the other conditions.

Table 6.7

Mean Memory Accuracy for Short Answer Questionnaire

Correct Misleading No Postevent Condition Postevent Postevent Information Information Information M SD M SD M SD Leading 1.98 1.08 1.10 0.70 1.33 1.02 Questions Media 2.45 0.83 1.50 1.19 1.45 1.05 Report Co-witness 2.00 1.01 1.73 0.83 2.23 1.19 Discussion Indirect 2.45 1.24 1.76 1.09 1.48 0.87 Transfer Control 2.23 1.14 0.85 0.75 0.30 0.47

Short Answer Memory Confidence. Participants stated their confidence in their answers to the short answer questions on a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” Scores were added together to get separate confidence scores for items for which participants were given correct, misleading, and no postevent information. Scores for the correct and misleading information were divided by four because there were four questions pertaining to each on the recognition questionnaire. Scores for the items in which participants received no postevent information were divided by six because there were six pertinent questions. After establishing that the dependent variables for memory confidence were not unacceptably highly correlated (r’s ranged from .58 to .70), a MANOVA was conducted in which the effects of Mode of Postevent Information on memory confidence were analysed. The mean scores for memory confidence are shown in Table 6.8. There was a significant effect of Mode of Postevent Information on the combined dependent variable of confidence, F (12, 119) = 1.98, p < .05; Wilks’ Lambda = .6; partial Chapter 6: Ways of Encountering Postevent Information 145

eta squared = .15).11 However, analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .017, showed that there was no effect of Mode of Postevent Information on items for which the participants received no postevent information, F (4, 47) = 1.83, p = .14, correct postevent information, F (4, 47) = 1.75, p = .16 and misleading postevent information F (4, 47) = 2.22, p = .08.

Table 6.8

Mean Memory Confidence for Short Answer Questionnaire

Correct Misleading No Postevent Postevent Postevent Information Information Information M SD M SD M SD Leading 3.00 0.91 2.96 0.75 3.16 1.16 Questions Media 3.26 0.92 3.53 0.75 4.19 1.25 Report Co-witness 3.50 0.93 3.13 0.89 3.97 1.20 Discussion Indirect 3.74 0.63 3.50 1.05 3.34 0.80 Transfer Control 2.93 0.90 2.61 0.95 2.86 1.25

Free Recall Free Recall Coding. Participants’ written narratives of the event were coded by two independent judges for accuracy using a standard coding system. This coding system was loosely based on one used by Yarmey and Morris (1998). Each narrative was scored in terms of: (a) number of accurate propositions (regarding sequence of events, character descriptions, dialog, setting, and items stolen); (b) conciseness (number of accurate propositions divided by number of words written); (c) fabrications (spontaneous errors);

11 When participants who had discussed the experiment with another participant were excluded, there was no significant effect of Mode of Postevent Information on the combined dependent variable of confidence, F (12, 106) = 1.74, p = .07; Wilks’ Lambda = .6; partial eta squared = .15). Furthermore, when participants who had guessed the purpose of the study were excluded, there was no significant effect of Mode of Postevent Information on the combined dependent variable of confidence, F (12, 98) = 1.39, p = .19; Wilks’ Lambda = .7; partial eta squared = .13). Chapter 6: Ways of Encountering Postevent Information 146

and (d) Experimentally Induced Information (correct and misleading postevent information). Two independent judges repeatedly viewed the crime video and independently coded the free recall. The judges were unaware of the conditions to which the participants were assigned. Each participant’s responses were randomly assigned to one of the two coders. Ten of the free recall narratives were scored by both coders in order to assess inter- rater reliability. There was a significant correlation between the two coders on accurate propositions (r = 0.97, n = 10, p < 0.01, two-tailed), items for which participants were given misleading postevent information (r = 1.00, n = 10, p < 0.01, two-tailed), and the items for which participants were given correct postevent information (r = 0.84, n = 10, p < 0.01, two-tailed). Inter-rater reliability was not assessed for conciseness because it only differed from the variable, “accurate propositions,” based on the number of words written and it was assumed that the judges would be able to count words accurately. According to Elmes, Kantowitz & Roediger (1991), high inter-rater reliability occurs when the correlation coefficient is greater than +0.70. All of the above correlations met this criterion. However, despite best efforts of the coders, it was not possible to achieve correlations of over +0.70 for Fabrications (r = 0.60, n = 10). The correlations for fabrications may be low because there were not many fabrications reported. When there was disagreement between the judges, one score was randomly chosen to use for the analysis. Free Recall Memory Accuracy. After establishing that the dependent variables for free recall accuracy were not unacceptably highly correlated (r’s ranged from -0.09 to 0.44) a one-way MANOVA was conducted in which the effects of Mode of Postevent Information (leading questions, media report, indirect co-witness information, co-witness discussion, no postevent information) on free recall accuracy (accurate propositions, conciseness, fabrications, misleading postevent information reported and correct postevent information reported) were analysed. The mean scores for memory accuracy are shown in Table 6.9. There was a significant effect of Mode of Postevent Information on the combined dependent variable of free recall accuracy, F (20, 316) = 2.18, p < .01; Wilks’ Lambda = .7; partial eta squared = .10). Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .01, showed that there was no effect of Mode of Chapter 6: Ways of Encountering Postevent Information 147

Postevent Information on items for which participants received no postevent information. That is, there was no effect of Mode of Postevent Information in accurate propositions, F (4, 99) = 1.23, p = .30, conciseness, F (4, 99) = 1.06, p = .38, and fabrications, F (4, 99) = 1.26, p = .29. There was a significant effect of Mode of Postevent Information on accuracy for items in which participants received misleading postevent information, F (4, 99) = 4.35, p < .01. Post Hoc tests with an LSD correction revealed that participants who were in the control group were significantly less likely to report misinformation (M = 0.05, SD = 0.22) than those who received misleading postevent information through co-witness discussion (M = 0.86, SD = 0.85, p < .0005) and the media report (M = 0.62, SD = 0.86, p < .05). Participants who received misleading postevent information through co-witness discussion were significantly more likely to report this misinformation than participants who received misleading postevent information from leading questions (M = 0.24, SD = 0.54, p < .01) and indirect transfer (M = 0.38, SD = 0.74, p < .05)12. There were no significant differences between any of the other conditions. There was a marginally non-significant effect of Mode of Postevent Information on accuracy for items in which participants received correct postevent information, F (4, 99) = 3.20, p = .016. Because Bonferroni corrections can be quite conservative, it was decided to further investigate this effect using post-hoc tests. Employing the LSD post-hoc test revealed that participants who were in the control group were significantly less likely to report correct postevent information (M = 0.85, SD = 0.75) than those who received the correct postevent information through co-witness discussion (M = 1.62, SD = 0.97, p < .05) and indirect transfer (M = 1.48, SD = 1.21, p < .05)13. Furthermore, it was also found that

12 When participants who had guessed the purpose of the study were excluded, participants who received misleading information through co-witness discussion were no more likely to report the misinformation (M = 0.86, SD = 0.85) than those who received it through indirect transfer (M = 0.44, SD = 0.81, p =.08). Furthermore, when participants who had discussed the experiment with another participant were excluded, participants who received misleading information through co-witness discussion were no more likely to report the misinformation (M = 0.75, SD = 0.72) than those who received it through indirect transfer (M = 0.41, SD = 0.80, p =.13). 13 When participants who had guessed the purpose of the study were excluded, participants who received correct information through indirect transfer were no more likely to report the correct postevent information (M = 1.38, SD = 1.20) than those in control group (M = 0.85, SD = 0.75, p =.13). Furthermore, when participants who had discussed the experiment with another participant were excluded, participants who Chapter 6: Ways of Encountering Postevent Information 148

participants in the co-witness discussion condition were more likely to report correct postevent information than those who received correct postevent information from leading questions (M = 0.71, SD = 0.72, p < .01). Similarly, participants who received correct postevent information through indirect transfer were more likely to report the correct postevent information than those who received it from leading questions (p < .05)14. There were no significant differences between any of the other conditions.15

Table 6.9

Mean Memory Accuracy for Free Recall

Misleading Correct Fabrication Accurate Condition Concise Postevent Postevent Proposition Information Information M SD M SD M SD M SD M SD Leading 26.3 9.84 0.12 0.04 0.24 0.54 0.71 0.72 6.48 2.91 Questions Media 31.2 8.12 0.11 0.03 0.62 0.86 1.29 1.23 6.86 2.52 Report Co-witness 29.8 8.31 0.13 0.04 0.86 0.85 1.62 0.97 6.67 3.07 Discussion Indirect 32.2 10.5 0.11 0.03 0.38 0.74 1.48 1.21 5.24 2.59 Transfer Control 29.3 9.52 0.13 0.03 0.05 0.22 0.85 0.75 7.20 3.93

Number of People Who Reported Experimentally Induced Information The accuracy scores were recoded so that the number of participants who reported at least one item of correct postevent information could be compared with those who didn’t report any. Percentages of those who reported at least one item of correct postevent information can be seen in Table 6.10. For the responses to the Recognition Questionnaire,

received correct information through indirect transfer were no more likely to report the correct postevent information (M = 1.41, SD = 1.23) than those in control group (M = 0.89, SD = 0.76, p =.13). 14 When participants who had guessed the purpose of the study were excluded, participants who received correct information through indirect transfer were no more likely to report the correct postevent information (M = 1.38, SD = 1.20) than those who received it through leading questions (M = 0.67, SD = 0.82, p =.06). 15 When participants who had discussed the experiment with another participant were excluded, those who received correct information through the media report were significantly more likely to report the correct information (M = 1.35, SD = 1.23) than those who received it through leading questions (M = 0.71, SD = 0.72, p <.05). Chapter 6: Ways of Encountering Postevent Information 149

a chi-square analysis revealed that there was no association between Mode of Postevent Information and whether or not participants reported at least one item of correct postevent information (χ2 (4, N = 105) = 8.67, p = .07). Similarly, there was no association between Mode of Postevent Information and whether or not participants reported at least one item of correct postevent information on the Short Answer Questionnaire (χ2 (4, N = 105) = 9.27, p = .06) and the Free Recall Form (χ2 (4, N = 105) = 7.07, p = .13). The accuracy scores were also recoded so that the number of participants who reported at least one item of misleading postevent information could be compared with those who didn’t report any such information. Percentages of those who reported at least one item of misleading postevent information can be seen in Table 6.10. For the recognition questionnaire, a chi-square analysis revealed that there was no association between Mode of Postevent Information and whether or not participants reported at least one item of misleading postevent information (χ2 (4, N = 105) = 4.13, p = .39). However, there was an association between Mode of Postevent Information and whether or not participants reported at least one item of misleading postevent information on the Short Answer Questionnaire (χ2 (4, N = 105) = 31.79, p < .0005) and the Free Recall Form (χ2 (4, N = 105) = 17.14, p < .01).

Table 6.10 Percentages of Participants who Reported at Least One Item of Experimentally Induced Information across the Modes of Postevent Information

Type of Mode of Postevent Information Memory Postevent Leading Media Co-witness Indirect Task Control Information Question Report Discussion transfer Recog- Correct 100% 100% 100% 100% 90% nition Misleading 95% 86% 100% 95% 90% Short Correct 81% 81% 100% 86% 65% answer Misleading ** 76% 76% 86% 86% 30% Free Correct 57% 62% 91% 71% 65% Recall Misleading * 19% 44% 59% 29% 5% Note. *χ2 (4, N = 105) = 17.14, p < .01. **χ2 (4, N = 105) = 31.79, p < .0005.

Chapter 6: Ways of Encountering Postevent Information 150

Discussion In general, the results from this experiment suggest that co-witness information, whether encountered directly through co-witness discussion or indirectly through a third party, has a particularly strong influence on subsequent testimony. That is, in general participants were more likely to report co-witness information than postevent information encountered through leading questions or a media report. This was true for both accurate and misleading postevent information. The first aim of our research was to determine whether the experimental manipulation was successful. Participants who received postevent information were asked three questions about the media report, leading questions, co-witness discussion, or indirect transfer of co-witness information, depending on the Mode of Postevent Information condition to which they were randomly assigned. The questions were asked at the end of the experiment in order to determine whether participants in the various conditions had similar experiences. We found no significant differences for Mode of Postevent Information on the questions, “Did you find the [e.g., media report] interesting and useful” and “Do you think your answers were influenced by the [e.g. media report].” This implies that to some extent participants in the different conditions were unaware of the manipulation, as they reported their conditions to be equally interesting, useful, and influential. However, for the question, “Did you learn anything new about the episode as a result of the [e.g., media report]?” there was a significant effect of Mode of Postevent Information. Participants in the co-witness discussion condition were more likely to report that they had learned something new than participants in the leading question, media report, and indirect transfer conditions. Interestingly, although compared to the other conditions participants in the co-witness discussion condition were more likely to claim to have learned something new, they were not more likely to claim that their answers had been influenced by the new information. Participants in some conditions (i.e., leading question, indirect transfer) were more likely to guess the purpose of the study. However, this had minimal impact on the results, so all participants were included in the analyses reported. Similarly, some participants admitted that they had discussed the experiment with another Chapter 6: Ways of Encountering Postevent Information 151

participant in the study (outside the experimental discussions), but again, this had had minimal impact on the results. The second aim of our research was to investigate how Mode of Postevent Information (leading questions, media report, indirect co-witness information, co-witness discussion, no postevent information) influences the accuracy of participants’ memory for the event. Results from the recognition, short answer, and free recall questionnaires revealed that there were no significant differences in accuracy between the leading question, media report, co-witness discussion, indirect transfer, and control (i.e., no postevent information) conditions when participants answered questions for which they had received no postevent information. This suggests that the groups were equal with regards to their memory accuracy, and that any differences found between the groups (on items for which they had received correct or misleading information) were a result of the experimental manipulation. When responding to questions for which they had received correct postevent information, participants in the leading question, media report, co-witness discussion, and indirect transfer conditions were significantly more accurate than the controls on the recognition questionnaire. Similar results were found on the short answer questionnaire (although for this dependent variable participants in the leading question condition were not significantly more accurate than the controls) and the free recall form (although for this dependent variable participants in the leading question and media report conditions were not significantly more accurate than the controls). On the other hand, when responding to questions for which they had received misleading postevent information, participants in the leading question, media report, co- witness discussion, and indirect transfer conditions were significantly less accurate than controls on the short answer questionnaire. Similar results were found on the free recall form (although for this dependent variable participants in the leading question and indirect transfer conditions were not significantly less accurate than the controls) and the recognition questionnaire (although in this case participants in the leading question, indirect transfer, and media report conditions were not significantly less accurate than the controls). It was predicted that the participants who received misleading postevent information would be less accurate than the controls because previous research suggests that Chapter 6: Ways of Encountering Postevent Information 152

participants receiving misleading postevent information from leading questions (e.g., Loftus & Palmer, 1974; Loftus & Zanni, 1975; Loftus, 1979a), media reports (e.g., Allen & Lindsay, 1998; Belli et al., 1994; Searcy et al., 2000; Shaw, Garcia, et al, 1997; Sporer, 1995; Wright & Stroud, 1998), co-witness discussion (e.g., Gabbert et al., 2003; Gabbert et al., 2004; Hoffman et al., 2001; Hollin & Clifford, 1983; Shaw, Garven, et al., 1997; Wright et al., 2000), and indirect co-witness information (e.g., Betz et al., 1996; Garven et al., 2000; Loftus & Greene, 1980; Shaw, Garven, et al., 1997, Experiment 1) are more likely to report this information than those who receive no postevent information. For the most part, the results from the present study were consistent with the findings from these previous studies, with the exception that the misinformation effect was not consistently found for all conditions across all of the dependent variables employed in the present study. However, the means in the current study were all in the predicted direction. That is, the mean accuracy scores for participants who received misleading postevent information were always lower than the controls when responding to questions for which they had received misleading postevent information. Even though the misinformation effect did not consistently occur for all conditions on all of the questionnaires, it is notable that participants in the co-witness discussion condition consistently reported more misinformation than the controls. This provides evidence regarding the strength of the influence that co-witness discussion has on subsequent testimony. Further evidence regarding the powerful influence that co-witness discussion has on memory becomes apparent when comparisons are made between different methods of encountering postevent information. The results from this study suggest that participants who discussed the event with a confederate co-witness were more likely to report correct postevent information in their recognition questionnaire responses, in their short answers, and in their free recall accounts of the event than those who encountered the information through leading questions. Furthermore, participants who discussed the event with a confederate co-witness were also more likely to report correct postevent information on the recognition questionnaire than those who encountered the information through the media report. On the other hand, participants who discussed the event with a confederate co- witness were also more likely to report misleading postevent information than those who encountered the misinformation through leading questions, the media report (although this Chapter 6: Ways of Encountering Postevent Information 153

difference was not significant on the free recall form), and indirect transfer (although this difference was not significant on the recognition form). The results from this study also showed that the indirect transfer of co-witness information has a powerful influence on memory when compared to other ways of encountering postevent information. It was found that participants who received indirect co-witness information were more likely to report correct postevent information on the short answer questionnaire and free recall form than those who received the information through leading questions. The finding that co-witness information, whether received directly through discussion or indirectly through a third party, has a stronger influence on memory than suggestive questioning is compatible with the results reported by Shaw, Garven, et al. (1997). Both studies have shown that despite all the attention that has been paid to leading questions, the effects of co-witness information are larger. It seems that witnesses are influenced by the information they receive either directly or indirectly from co-witnesses and this influence occurs irrespective of the accuracy of the co-witness information. That is, co-witness discussion can have both positive and negative impacts on memory accuracy depending on whether the information provided was correct or incorrect respectively. However, this finding needs further investigation because it is unclear whether co-witness discussion, without the influence of experimentally induced information, results in the same advantages and disadvantages. That is, future research should include Natural Discussion Groups in order to determine whether co-witness discussion has both positive and negative impacts on memory accuracy without the influence of experimentally induced information. There are several reasons why postevent information encountered through discussion with a co-witness would be stronger than information from a non-social, written source, such as the leading questions and media report. Gabbert et al. (2004) suggest that written postevent information does not convey additional information that may influence memory conformity such as non-verbal cues (e.g., eye-contact, facial expressions etc) and subtle social cues (e.g., perceived credibility, trustworthiness, confidence, etc). In addition, Gabbert and her colleagues suggest that co-witness information encountered through discussion may have a larger effect than when encountered through a non-social source because people generally assume that information exchanged during discussion is truthful Chapter 6: Ways of Encountering Postevent Information 154

and accurate (Grice, 1975; Swann, Giuliano & Wegner, 1982). Furthermore, they suggest that people may choose to agree with others in order to appear more likeable (Tajfel & Turner, 1986). However, these accounts do not explain why for the most part in the current study there were no differences found between those who discussed the event with a co- witness and those who encountered the postevent information indirectly by reading statements that had purportedly been written by a co-witness. These findings are consistent with those of Meade and Roediger (Experiment 4, 2002), however Gabbert et al. (2004) found that co-witness discussion was significantly more misleading than the indirect co- witness information. Whilst the results from the current experiment are different from those of Gabbert et al.’s study, they are not necessarily contradictory. In the current study information encountered through discussion had a slightly stronger effect than indirect co- witness information, but the results were not significant. The possibility exists that the power of the study was not high enough to detect a difference between the co-witness discussion group and the indirect transfer group. Alternatively, it may be the case that there are no differences between information obtained through co-witness discussion and indirect transfer. The possibility exists that even a hypothetical co-witness can have as strong of a social influence as co-witness discussion even though there are no non-verbal or subtle social cues. A final aim of Study 4 was to investigate how the Mode of Postevent Information influences participants’ confidence in their memories. It is important to consider eyewitness confidence because the confidence that witnesses have about their testimonies may influence future investigative and legal proceedings (Wells et al., 2000). Confidence levels were reported on the short answer and recognition questionnaires only. On both of these measures, it was found that there were no significant differences in confidence between each of the leading question, media report, co-witness discussion, indirect transfer, and control conditions when participants answered questions for which they had received no postevent information. This suggests that the groups did not differ with regards to their confidence levels, and that any disparities found between the groups (on items for which they had received correct or misleading information) were a result of the experimental manipulation. Chapter 6: Ways of Encountering Postevent Information 155

When participants responded to recognition questions for which they had received correct postevent information, those in the co-witness discussion condition were more confident in their answers than those in the leading question and control conditions. Furthermore, participants in the media report condition were also more confident than the controls. The increased confidence levels suggest that participants in the co-witness discussion and media report conditions are more likely to believe the correct postevent information they have been provided with. However, on the short answer questionnaire, there were no differences in confidence levels between the conditions. It may be the case that confidence differences weren’t detected on the short answer questionnaire because the questions were broad enough that participants could answer the questions without referring to the postevent information. For example, in response to the question, “What was the thief who pointed the handgun at the lady wearing?” participants may have reported the correct postevent information they were given (a white singlet top) or they may have reported an alternative correct response (e.g., jeans). On the other hand, when responding to questions for which they had received misleading postevent information, participants in the co-witness discussion condition were more confident in their answers than those in the leading question, indirect transfer, and control conditions. Furthermore, participants in the media condition were also more confident than those in the leading question, indirect transfer, and control conditions. The increased confidence levels suggest that participants in the co-witness discussion and media report conditions are more likely to believe the misleading postevent information to be true when in fact it is not. That is, participants were particularly more likely to be taken in by misleading postevent information when encountered during discussion with a co-witness or when written in a media report. This finding is consistent with past research, which suggests that misled participants often have as much confidence in their recognition judgments as non-misled participants who are considerably more accurate (Loftus, Donders, Hoffman, & Schooler, 1989). The current study expands on the literature because previous studies comparing different ways of encountering postevent information have not assessed eyewitness confidence (Gabbert et al., 2004; Meade & Roediger, 2002; Shaw, Garven, et al., 1997). However, on the short answer questionnaire, there were no differences in confidence levels between the conditions. Once again, it may be the case that Chapter 6: Ways of Encountering Postevent Information 156

that confidence differences weren’t detected on the short answer questionnaire because the questions were too broad. Confidence levels for correct and incorrect answers were also analysed. It was found that participants in the co-witness discussion and media report conditions were more confident in their incorrect answers than those in the indirect transfer and control conditions when responding to questions for which they had received misleading postevent information. Participants in the leading question condition were also more confident in their incorrect answers than the controls when responding to questions for which they had received misleading postevent information. High levels of confidence in incorrect decisions are particularly worrisome, because they suggest that people are not only reporting the misinformation, but that they also have a high level of confidence in the accuracy of this information and may therefore be more likely to influence the direction of an investigation. The current study addresses several limitations of Study 3. In Study 4, participants were not given a recognition questionnaire prior to encountering the postevent information, as they were in Study 3. While the absence of a pre-manipulation questionnaire prevented us from determining whether there were any differences between the groups prior to manipulation, it had the advantageous effect of decreasing the extent to which the questions themselves could operate as a source of misleading information in conditions for which we were not trying to manipulate this factor. A second limitation of Study 3 regarded the type of questioning the participants were given. In Study 3, participants were only given a true/false recognition questionnaire to complete. The use of forced-choice recognition tests can render an inherent response bias in the experiment (McCloskey & Zaragoza, 1985). In Study 4, participants were given free recall and short answer questionnaires prior to the recognition questionnaire in order to reduce the demand characteristics of the study. To date, only a few empirical studies have used more open-ended questioning procedures to examine the misinformation effect (e.g., Belli, Lindsay, Gales, & McCarthy, 1994; Hammersley & Read, 1986; Hollin & Clifford, 1983; Lindsay, 1990; Sutherland & Hayne, 2001; Zaragoza, McCloskey, & Jamis, 1987). Only three of these studies have directly compared recognition and recall procedures, and all studies have shown that the misinformation effect decreases when participants are asked more open-ended questions (Hollin & Clifford, 1983;Sutherland & Hayne, 2001; Zaragoza Chapter 6: Ways of Encountering Postevent Information 157

et al., 1987). In the present study, co-witness memory conformity decreased, but was not eliminated when participants answered free recall and short answer questions. This finding suggests that memory conformity is not entirely due to response biases. A final limitation of Study 3 that was addressed in the current study regarded what information the confederates were instructed to tell the participants. In Study 3, participants were not limited concerning how much correct information they were allowed to provide about the video. Therefore, in Study 3 the confederates may have been assisting the participants too much, increasing their credibility within the group, and thus making their misinformation seem more plausible as well. In Study 4, confederates were told to only provide information as instructed, and they were tape recorded to ensure that this was done. While Study 4 improved on many of the weaknesses of Study 3, there are a few limitations that are apparent in both studies. First, both of these studies are limited by their use of confederates. While confederates can be useful for adding experimental control to studies, they may not act like true co-witnesses and their presence in the group may alter the behaviour of other members. For example, confederates may appear to have increased confidence because they know exactly what they are supposed to say. A second limitation of both studies is that they make little attempt to address the theoretical explanations regarding why participants change their responses following the acquisition of co-witness information. The future experiments in this thesis attempt to address these limitations. Despite minor limitations, this study has yielded some important findings. This study has shown that co-witness information has a particularly strong influence on subsequent testimony, whether encountered directly through co-witness discussion or indirectly through a third party. This effect is evident whether measured using true/false recognition questions, short answer questions, or free recall. In general, participants were more likely to report co-witness information than postevent information encountered through leading questions or a media report. This was true for both accurate and misleading postevent information. Furthermore, after having been misled, participants who discussed the event with a co-witness were often overconfident in their incorrect answers. The finding that co-witness information has a particularly strong influence on witness accuracy and confidence suggests that we should be cautious about generalising from studies investigating other ways of encountering postevent information to co-witness information Chapter 6: Ways of Encountering Postevent Information 158

as social factors may have a strong influence on witnesses. For this reason, co-witness information should be studied in its own right.

159

Chapter 7 Study 5: Unintentional Hearsay

Studies 3 and 4 have shown that exposure to co-witness information about an event after it has occurred can cause people to incorporate this information into their testimony. Several other studies have demonstrated similar memory conformity effects. These studies suggest that participants who have discussed an event with a co-witness who supplies misinformation show less accurate memory for the misled items than non-misled items (e.g., Hoffman, Granhag, See, & Loftus, 2001; Schneider & Watkins, 1996 Shaw, Garven, & Wood, 1997; Wright & Mathews, 2004) and than individuals who did not discuss the event (e.g., Gabbert, Memon, Allan, & Wright, 2004; Hollin & Clifford, 1983). However, many of these studies are limited, as they do not examine why participants frequently report misinformation presented to them. That is, it is unclear whether the conformity is due to memory distortion or to other factors. The investigation as to why participants often report misinformation has important legal implications. If witnesses are unable to distinguish what they actually experienced from information obtained from a co-witness, then they may be reporting hearsay without knowing it. The legal concept of ‘hearsay’ embodies the notion that “a witness’s assertions of relevant facts should be based upon his or her own experiences” (Forbes, 2003, p. 59) and not those of another. However, implicit in this concept is the assumption that witnesses are able to distinguish their own experiences from information they obtain second-hand (e.g., from a co-witness). That is, the legal system makes the assumption that witnesses would only report hearsay evidence because of informational influence and not because of memory change, but this may not be the case. If witnesses are unable to distinguish their ‘real’ memories from second-hand information then their testimonies are considered ‘contaminated’ and this may lead to the unintentional presentation of hearsay evidence. As previously described in Section 1, there are four ways in which memory conformity can occur. First, an individual may yield for the sole reason of gaining social approval, a mechanism known as ‘normative social influence’ (Deutsch & Gerard, 1955). This mechanism is unlikely to be a factor in Studies 3 and 4 because participants gave their Chapter 7: Unintentional Hearsay 160

final accounts in the absence of other group members. Second, an individual may assume that the co-witness information they received is correct, but they accurately remember that it was obtained second-hand. This mechanism is known as informational influence (Deutsch & Gerard, 1955). In this case, a witness may report the co-witness information when asked to remember the original stimulus because they believe it be to correct. Similar to the informational influence, participants may report misinformation because of biased guessing (McCloskey & Zaragoza, 1985). That is, a witness may report the co-witness information because of demand characteristics of the experiment. According to Betz, Skowronski, and Ostrom (1996) these mechanisms require that witnesses remember the co- witness information and that it was obtained from a second-hand source. For this reason, they refer to these mechanisms as “source-tagged mechanisms.” In these situations, no memory distortion occurs. A final explanation for memory conformity suggests that memory change does occur. For example, according to the source monitoring theory (e.g., Johnson, Hashtroudi, & Lindsay, 1993; Lindsay & Johnson, 1989), it is possible for an individual to remember the co-witness information, but forget that it was obtained second- hand. Source-monitoring theory suggests that people may sometimes have trouble identifying the source of information that is stored in memory. When people are unable to identify the source of their information, they may report the co-witness information and believe it is their recollection of the stimulus. Whilst these explanations of memory conformity are not necessarily mutually exclusive, Study 5 aims to investigate which explanation best accounts for the effect. Researchers have employed two techniques in an attempt to identify the mechanism responsible for memory conformity (Meade & Roediger, 2002; Roediger, Meade, & Bergman, 2001). The first method has been to instruct participants to produce a remember/know judgment (Tulving, 1985) for each item they recall following the discussion. In one study (Roediger et al., 2001), participants viewed common household scenes containing many items. In pairs, participants recalled as many items as they could, unaware that they were working with a confederate, who was supplying misinformation. After a “brief delay” (p. 365; exact time not specified), participants were instructed to recall as many items as they could individually and produce a remember/know judgment for each item they recalled. If they consciously remembered seeing the object in the scene they Chapter 7: Unintentional Hearsay 161

would indicate that they “remembered” it. In contrast, “know” responses were used to indicate items for which the participants did not have any specific recollection, but which they believed to be true. This method was used in an attempt to assess whether the participants remembered that the information was actually obtained second-hand. It was found that individuals recalled the misinformation supplied by the confederates more than participants in the control condition, who received no misinformation. Although participants did falsely report some suggested items, they were more likely to claim that they “knew” the suggested items had been in the scenes than to report that they specifically “remembered” seeing them there. These results were replicated in studies by Meade and Roediger (2002). This implies that there is a difference between information obtained from a co-witness and that obtained first-hand from the scene. The second experimental technique used to determine the mechanism most likely responsible for memory conformity is to warn participants about the presence of misinformation. For example, participants may be warned prior to starting their individual accounts that the confederate with whom they discussed earlier may have been in error, and that they should try to report only what they remember seeing (Meade & Roediger, 2002). If the participants are not able to distinguish between what they saw and what the confederate told them (that is, if they report the misinformation despite the warning) then this would support the notion that memory change is responsible for the conformity observed. Meade and Roediger found that the warnings significantly reduced the effect of the co-witness misinformation, but did not eliminate it, suggesting that some distortion of the original memory occurred. While the studies described above (Meade & Roediger, 2002; Roediger et al., 2001) provide valuable insights on the mechanisms underlying memory conformity, they do have some limitations. First, in these studies there was only a short time delay between the presentation of misinformation and the memory testing. Real witnesses are commonly interviewed after longer delays. In a study of actual witnesses, it was found that 52% were interviewed more than three days after the event and 37% were not interviewed until five to six days after the crime (Kapardis, in preparation, cited in Kapardis, 1997). Source misattribution is more likely to occur after a longer delay. For example, Underwood and Pezdek (1998) found that source information became less strongly associated in memory Chapter 7: Unintentional Hearsay 162

after a delay of one month than after a ten-minute delay, so it is important to test the effect of warnings and remember/know judgments after longer delays. Another limitation of these studies is their use of static photos as stimuli, which may not have high ecological validity. Furthermore, in these studies, the discussion in which participants engaged is much more confined than a natural conversation because the participants were required to take turns recalling items from the scenes. This may further decrease the ecological validity of the experiment. The present study aimed to improve on these limitations and increase ecological validity by increasing the delay period between co-witness discussion and individual testing, using videos to present a more realistic crime scenario, and by allowing more natural conversations. The first objective of Study 5 was to determine whether memory conformity is due to memory distortion or to other ‘source-tagged mechanisms’ (Betz et al., 1996). To achieve this, participants in Study 5 received warnings that they may have been exposed to misinformation. Explicit warnings about misinformation are meant to reduce the experimental demand that may encourage conformity. It was hypothesised that memory conformity would be reduced, but not eliminated by the warning, as suggested by other research on co-witness discussion (Meade & Roediger, 2002) and other forms of misinformation (e.g., Betz et al., 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, 2001; McDermott & Roediger, 1998; Wright, 1993). Furthermore, participants were asked to state remember/know judgments (Tulving, 1985). In accordance with previous research (Meade & Roediger, 2002; Roediger et al., 2001) it was hypothesized that participants would be more likely to claim that they “know” the items mentioned by the co-witness than to report that they specifically “remembered” seeing them. Another aim of Study 5 is to improve on the experimental manipulation utilised in Studies 3 and 4. These studies were limited due to the use of confederates to induce the experimental manipulation of postevent information. Confederates may not act as true co- witnesses would and may alter the behaviour of other witnesses. For example, confederates may appear to have increased confidence because they know exactly what they are supposed to say. To address this issue, a new methodology has been developed, which we refer to as ‘Stimulus Induced Information.’ Chapter 7: Unintentional Hearsay 163

When using the Stimulus Induced Information technique, two similar versions of a stimulus event are produced and participants are shown one version of the stimulus, without their knowledge that a second, slightly different version exists. Each version contains items that are unique to it. Participants are then asked to discuss the stimulus event with one another in groups, which contain members who have seen different versions of the stimuli. Because the participants are unaware that they have seen different stimuli, their interactions with the group are hopefully more natural than those of the delegated confederates, and consequently their confidence and credibility will not be any higher than those of an ordinary participant. The methodology is similar to a real situation in which eyewitnesses may have seen slightly different views of the crime. This methodology has been developed and used recently in research investigating the effects of co-witness discussion on memory (Gabbert, Memon, & Allan, 2003; Kanematsu, Mori, & Mori, 2003; Mori & Mori, 2004; Wright, Self, & Justice, 2000). In accordance with the findings from studies using Confederate Induced Information, it has been found that participants who received Stimulus Induced Information during discussion erroneously recall items from the alternative version of the stimulus when later giving their individual accounts (Gabbert, Memon, & Allan, 2003; Kanematsu, Mori, & Mori, 2003; Mori & Mori, 2004; Wright, Self, & Justice, 2000). To date, studies that have used the Stimulus Induced Information methodology have not used appropriate control groups. There are experimental confounds when participants in groups with members who have seen different versions of the stimulus are compared with individuals who have not discussed the event. Using this paradigm it is impossible to separate the effects from the act of discussing the event, the social presence of another, the elaboration and rehearsal of event-related memories, and the confederate’s influence. There are also confounds when all participants are in groups with members who have seen different versions and a comparison is made between items that differed between the two stimuli, and those that did not (e.g., carryover effects associated with within-subjects design). In order to address this limitation, the current study included a control condition where participants discussed the event, but received no experimentally induced misinformation. That is, participants discussed the event in groups in which all of the members had seen the same version of the stimulus event. The inclusion of these “Natural Chapter 7: Unintentional Hearsay 164

Discussion Groups” allows positive or negative effects of discussion on memory in general to be investigated. These Natural Discussion Groups that do not involve the experimental introduction of postevent information (either by a confederate or any other method), have potentially the highest ecological validity, and yet few applied studies have employed such controls. Furthermore, the little research that has been conducted has yielded inconsistent results. Results from some experiments without experimentally induced information suggest that under certain conditions, group recall can have a beneficial effect on eyewitness testimony (e.g., Underwood & Milton, 1993; Yarmey & Morris, 1998), while others suggest that discussion prior to individual recall is not an advantageous procedure (e.g., Yarmey, 1992). Moreover, Study 3 indicated that Natural Discussion Groups were equally accurate and confident as the control group who didn’t talk about the event. That is, in Study 3 there was no advantage or disadvantage of group discussion prior to individual testimony if no postevent misinformation was deliberately introduced by the experimenter, thus it is important to investigate this issue further. The second objective of the current study was to compare the memory of individuals in groups who receive Stimulus Induced Information to individuals who have not discussed the event, and individuals in Natural Discussion Groups. Based on previous studies (e.g., Gabbert et al., 2001; Wright et al., 2000) it was hypothesized that groups who receive Stimulus Induced Information would show less accurate memory for the misled items than individuals not in a discussion group. It was also hypothesized that groups with Stimulus Induced Information would perform worse on the misled items than Natural Discussion Groups (i.e., groups in which all participants saw the same stimuli), however this has not been investigated before. It was unclear how Natural Discussion Groups would perform in comparison to individuals not in a discussion group, as previous research has yielded mixed findings (e.g., Underwood & Milton, 1993; Yarmey, 1992; Yarmey & Morris, 1998). In addition to the free recall and recognition tests used in previous studies in this thesis, the current study also assesses the effects of co-witness discussion on lineup identification accuracy. To date, two studies have investigated the effects of group discussion on lineup identification accuracy. One study found group discussion resulted in Chapter 7: Unintentional Hearsay 165

more correct rejections of a culprit-absent lineup, but only when the pairs reached collaborative decisions about the lineup. There were no advantages when discussion was followed by independent identification decisions (Yarmey & Morris, 1998). The other study found that although participants' ability to accurately recognize the perpetrator in a lineup did not differ as a function of individual versus group discussion, there was a non- significant trend in favour of groups (Warnick & Sanders, 1980). Based on these previous findings, it was hypothesised that co-witness discussion would either have no effect, or slightly increase the accuracy of subsequent individual lineup identifications. We were also interested in investigating effects of group discussion on lineup identification confidence. One study which investigated the effects of indirect co-witness information on identification confidence found that identification confidence changed on the basis of what participants were told another witness had said (Luus & Wells, 1994). That is, the participants’ confidence in their identification decisions increased when they were told that the co-witness had identified the same person, and decreased when they were told that the other witness had identified someone else or no one at all. On the basis of these results, we would hypothesise that participants who were in Natural Discussion Groups would give higher confidence ratings after the discussion than the no discussion control group because they would presumably hear confirmatory evidence from their co-witnesses. It is also likely that participants in groups who receive Stimulus Induced Information would give lower confidence ratings than the no discussion controls because they would presumably hear disconfirmatory evidence from some of their co-witnesses. However, the possibility also exists that the participants will not want to change their post-discussion confidence ratings because they do not want to seem easily influenced by others. For example, research on eyewitness memory has shown that people are less likely to accept postevent misinformation if they are required to make a commitment to a response prior to receiving the misinformation (Loftus, 1977).

Chapter 7: Unintentional Hearsay 166

Method

Design There were three phases in this study: the exposure phase, the discussion phase, and the recall phase. During the exposure phase, participants saw one of two slightly different versions of a crime video (either Version A or B). In the discussion phase, participants were assigned to one of three conditions. Participants in the ‘same-video’ group discussed the crime in a group made up of members who had all seen the same video (all A or all B). Participants in the ‘mixed-video’ group discussed the crime in a group where half the group had seen Version A and half had seen Version B. Participants in the ‘no discussion’ control group did not discuss the crime, but were told to think about what they had seen and to write an individual account of the details they could recall. At the start of the recall phase, participants were given one of three warnings: (1) specific warning, (2) general warning, or (3) no warning. The specific warning explicitly stated that some members in their discussion group may have been show a different version of the crime, whereas the general warning stated that participants should disregard what other group members told them and try to answer the questions based on what they remember from the video. Participants in the ‘no warning’ condition were not given any warning. During the recall phase, participants were asked to give a free recall account of what happened during the video and then to complete a recognition questionnaire. Some of the items on the questionnaire assessed participants’ memories for details that differed between the two versions of the video, and some assessed memory for details that were consistent across the two videos. Participants were asked to indicate their confidence in their answers, and to make a remember/know judgment (Tulving, 1985) regarding the nature of their memory. Finally, participants were asked to identify the thief from a target-present lineup. The study employed a 3 x 2 + 1 experimental design in which Warning (specific warning, general warning, no warning) and Discussion Type (same-video, mixed-video) were manipulated. There was also a ‘no discussion’ control condition that did not discuss the crime and received no warnings. The dependent variables were accuracy and confidence of memory for the video. Chapter 7: Unintentional Hearsay 167

Participants Participants were 119 undergraduate psychology students (97 female, 22 male) from the University of New South Wales. The average age of the participants was 20.9 (SD = 1.40) years. Participants were invited to volunteer to take part in the experiment during a class tutorial. They received no incentives for their participation in the study.

Materials Eyewitness Stimuli. The eyewitness stimulus used in this study was a short video (under two minutes in duration) depicting a robbery. In the video, a Caucasian thief follows someone into a secured apartment block and knocks on the door of one of the apartments. A woman answers the door and the thief attempts to persuade her to allow him to enter the apartment by saying that he needs to fix some problems in the building. When the woman resists, he pushes her out of the way, runs into the apartment and begins stealing various items. The woman runs out of the apartment to get help and returns with a man. The man confronts the thief, but the thief pushes his way past him. The thief runs out of the building and the man chases him, but he is unable to keep up with the thief. There were two versions of the video that varied slightly. For example, in Version A the thief poured jewellery out of a jewellery box, while in Version B he looked through the jewellery box without emptying it, and two different actors played the role of the thief in Version s A and B (see Appendix J). Table 7.1 summarises the differences between the videos. The videos were displayed to the participants on individual computer monitors. Each participant viewed only one of the two videos.

Chapter 7: Unintentional Hearsay 168

Table 7.1. Summary of Differences between Video A and Video B

Version A of the Video Version B of the Video Conflicting Thief poured jewellery out of Thief looked through jewellery information jewellery box. box. between Actor 1 as thief Actor 2 as thief videos Item in Thief was drinking beer. Thief was not seen drinking beer. Video A Thief stole camera. Thief didn’t steal camera. only Thief was wearing a watch. Thief was not wearing a watch. Thief sent to fix water problems. Thief didn’t say why he was sent. Woman is wearing a headband. Woman was not wearing a headband. Item in Woman was not wearing a brooch. Woman was wearing a brooch. Video B The rescuer was not wearing a The rescuer was wearing a leather only leather jacket. jacket.

Discussion Questions. Participants who were in discussion groups (either mixed or same-video conditions) were asked four questions about their experience in the groups. They were asked, “Do you think your answers were influenced by the group discussion; Did the discussion help you remember things from the video that you would otherwise have forgotten; Did you learn anything new about the video as a result of the discussions,” and finally, “Did your group have an open and free exchange of ideas?” Participants were asked to indicate their response to each question using a six-point scale in which a score of “1” indicated “Yes” and “6” indicated “No.” Free Recall Form. A Free Recall Form was developed for this study. Participants were instructed to write down as much as they could remember about the crime video. Specifically, they were asked to write about the sequence of events, what was said, the setting, what the people looked like, what was stolen from the flat, and any other details they could remember. Recognition Questionnaire. A recognition questionnaire containing 32 questions about the video was developed for the purpose of this study (see Appendix K). Each question required a true/false response. Of the 32 questions, 12 were true and 6 were false-- Chapter 7: Unintentional Hearsay 169

these questions were non-misled (i.e., no participants received misleading information regarding these items during the group discussion). Six items were “spontaneously false” (i.e., they contained incorrect information about the video that was spontaneously generated in some groups while discussing the video).16 Finally, eight items were false statements for which some people received stimulus induced misinformation-- these eight items regarded aspects which differed between Version A and Version B of the video (see Table 7.1). When the videos were compressed and digitized for viewing on computer monitors, some of the differences between the two videos became very unclear. To obtain an idea of which differences were noticeable, coders were asked to view the two versions of the video back to back and note the differences between them. None of the coders noted that: (1) The woman was wearing a brooch in Version B of the video; (2) The rescuer was wearing a leather jacket in Version B of the video; and (3) The thief was wearing a watch in Version A of the video. When these three items were removed from the analyses, the results were not greatly affected, so the results were analysed without the three questions pertaining to these items. For each question, participants were asked to indicate whether the statement was true or false, rate their confidence on a 6-point scale (1= low confidence; 6= high confidence), and categorize the answers they had indicated as “true” answers as either a “remember” or a “know” judgment (Tulving, 1985). Verbally and in writing, participants were given definitions of “remember” and “know” (see Table 7.2) based on instructions given by Rajaram (1993).

Table 7.2

16 These items were not included in the analyses because it was difficult to assess the spontaneous misinformation mentioned during group discussions because the audiotapes were very poor quality. Chapter 7: Unintentional Hearsay 170

Instructions for Remember/Know Judgments.

Instructions Remember If your recognition of the item is accompanied by a conscious judgments recollection of its occurrence in the video, then state that you “Remember” it. “Remember” is the ability to become consciously aware again of some aspect or aspects of what happened or what was experienced at the time the video was presented (e.g., aspects of the physical appearance of the item in the video, or of what you were thinking at the time). In other words, the “remembered” statement should bring back to mind a particular association, image, or something more personal from the time of viewing the video, or something about its appearance or position. Know “Know” responses should be made when you believe that the item was judgments in the video, but you cannot consciously recollect anything about its actual occurrence or what happened or what was experienced at the time of its occurrence. In other words, indicate, “know” when you believe the statement to be true, but it fails to evoke any specific recollection from the video.

To further clarify the difference between these two judgments (i.e., “remember” vs. “know”), participants were given the following examples: If someone asks you the last movie you saw, you would typically respond in the “remember” sense, that is, becoming consciously aware of the experience. On the other hand, there are times when we meet someone on the street who appears familiar. Although we may “know” we have met this person before, we may not remember the circumstances in which we met him/her. Photo Lineup. A photo lineup task was also developed to assess the ability of the participants to identify the thief (see Appendix L). Two different versions of the lineup were developed which showed the same photographs, but in different orders. Each lineup contained nine photographs. Of these photographs, one showed the thief from Version A of the video, and one showed the thief from Version B. The remaining seven photographs were of “foils” who had not appeared in the videos and were unfamiliar to the participants. Participants were instructed that the thief may or may not be present in the line up. A Lineup Questionnaire was developed which first asked whether the thief was present in the lineup. If they indicated, “Yes” to this question, participants were asked to identify the perpetrator from the lineup. They were also asked to rate their confidence in this decision Chapter 7: Unintentional Hearsay 171

on a 6-point scale, with a score of “1” indicating “Low confidence” and “6” indicating “High confidence.”

Procedure This study took place during two class tutorials, one week apart. In the first session, the students congregated in a computer laboratory and were each seated at a computer to view a video of a crime scene. After informed consent was obtained, participants were instructed that they would each be viewing the same crime scene on the individual computers terminals. However, unbeknown to the participants, there were two slightly different versions of the crime scene (versions A and B), and each individual witnessed only one version. Participants were told, “Today I will be showing you a video of a crime scene. In the past, I have done this by showing the videos using TV’s, however I have noticed that the viewing conditions were not very good for many people, especially those seated at the back. Instead, we will show you the video on computers, so that each of you can see the video under equal and optimal viewing conditions.” Participants were randomly seated at the computer terminals and were instructed to put on headphones. The computer monitors were slanted slightly away from each other to ensure that participants could only attend to their own video. The experimenter inconspicuously took note of which terminal each student sat at so they would know which version of the video the participant had observed. Participants were instructed, “In just a moment you will be witnessing a video of a crime scene. Imagine that this is a real life crime and that you are an actual witness to the crime. Just like real witnesses, you will be asked questions about the crime later, so please pay close attention to the video.” The experimenter then checked to make sure that everyone was wearing their headphones and knew how to begin the video. Participants were instructed to begin their videos at exactly the same time so that they would not finish ahead of others and have the opportunity to view any aspect of the video again by watching someone else’s screen. After seeing the video, there was a delay of twenty minutes (during which the participants did their regular course work). Participants were then split into groups of four to discuss the crime scene. Unbeknown to the participants, groups were arranged so that Chapter 7: Unintentional Hearsay 172

some groups contained individuals who had all seen the same version of the video (same- video groups) These same-video groups had participants who had either all seen Version A of the video or all seen Version B. Alternatively, some groups contained equal numbers who had seen slightly different versions of the video (mixed-video groups). These mixed- video groups had two participants who had seen Version A of the video and two who had seen Version B. The group discussions were tape-recorded. Participants were told, “We would like you to tape-record your conversations because we are interested in the information you remember. Having someone write everything down would be inconvenient, so we believe this method will be easier for you, and allow you to have more natural, free-flowing conversations. Please be aware that what you say will be kept strictly confidential, and there will be no way of identifying what each of you says.” Participants in the control group were asked not to discuss the video with anyone else, but rather to write about the video individually. All participants were given ten minutes to discuss or write about the video using a guided recall procedure. Specifically, the participants were instructed to discuss/write about the sequence of events, the setting, what the people looked like, and any other information they could remember. At the end of ten minutes, participants were asked not to discuss the event any further. After a delay of one week, participants in the discussion groups (both same- and mixed-video groups) were asked to complete the Discussion Questions. All participants were then asked to state what they believed to be the purpose of the experiment. Participants in the discussion conditions were then randomly assigned to one of three warning conditions (see Table 7.3). One third of the participants were given a ‘specific warning’ that some students in their group may have seen a slightly different video and that they should answer the questions solely on the basis of what they themselves remember from the video. Another third of the participants were given a more ‘general warning’ that they should make a specific effort to disregard what others in their discussion group told them and answer the questions solely on the basis of what they themselves remember from the video. The remaining participants received no warning.

Table 7.3 Chapter 7: Unintentional Hearsay 173

Warning Instructions. Warning Type Warning Instructions When answering these questions make a specific effort to disregard General what others in your discussion group told you and answer them warning solely on the basis of what you remember from the video. In your discussion groups last week, some members of your group may have been shown a slightly different version of the video. Specific Therefore, when answering these questions you should make a warning specific effort to disregard what they told you and answer them solely on the basis of what you remember from the video. No warning No warning given.

After the warning manipulation, participants were given a series of questionnaires to complete individually to assess their memory for the event. First, they were given as much time as they needed to complete the Free Recall Form, followed by the Recognition Questionnaire. When participants had completed both of these questionnaires, they were given the Photo Lineup Questionnaire to complete. They were instructed, “I would like you to individually decide which person from the photo lineup you think is the thief from the video. The person you saw may, or may not be depicted in one of these photographs. Please fill out these question sheets individually.” After handing the completed questionnaires to the experimenter, participants in the discussion conditions (both same- and mixed-video groups) were put back into their original groups. They were instructed, “I would like you to take a moment to discuss in your groups which person from the photo lineup you think is the thief from the video. The person you saw may, or may not, be depicted in one of these photographs. Each group will receive a different lineup, so it won’t help you to listen to what the other groups are saying.” These discussions were tape-recorded. After a couple of minutes of discussion, the experimenter passed out a second Lineup Questionnaire that was identical to the first. The students were instructed, “Please fill out this second sheet individually and hand it in when you are done.” Chapter 7: Unintentional Hearsay 174

Individuals who were in the no discussion control condition were not asked to discuss the lineup with anyone. Instead, they were asked to take a few moments to visualize what the thief looked like before completing the second, identical Lineup Questionnaire. Once the Lineup Questionnaires had been collected, participants were thanked for their participation and fully debriefed about the study. Participants were also asked not to discuss the experiment with other potential participants.

Results Statistical analysis focussed on the following questions: First, how successful was the experimental manipulation? Second, how did Discussion Type (mixed-video group, same-video group, no discussion) and Warning (general, specific, or no warning) influence participants’ memory accuracy and confidence for misled and non-misled items? Third, did remember/know judgments differ for those participants whom we had attempted to mislead and those whom we had not? Finally, what were the effects of Discussion Type on participants’ ability to identify the perpetrator from a lineup?

Experimental Session The data for this experiment were collected during two separate experimental sessions. The first session comprised of more mixed-video groups and controls than same- video groups; the second session comprised of more same-video groups and controls than mixed-video groups. In order to determine whether there were any differences between the participants from the two experimental sessions, the results of the controls were compared. Using a one-way ANOVA, it was found that there were no differences between the controls in the first session and those in the second. That is, there were no significant differences for the participants’ accuracy on the non-misled items, F (1, 27) = .01; p = 0.92, or the misled items, F (1, 27) = 2.78; p = 0.11. Similarly, there were no significant differences for the participants’ confidence on the non-misled items, F (1,27) = 1.57; p = 0.22, or the misled items, F (1, 27) = 1.23; p = 0.28. It was therefore decided that it was safe to disregard the session in which participants were tested and analyse the data from both sessions together. Chapter 7: Unintentional Hearsay 175

Discussion Questions Four independent t-tests were used to compare responses given by participants in the same- and mixed-video groups to four questions relating to the participants’ experiences in their discussion groups (the individuals in the no-discussion control group were not given these questions, so they were excluded from this analysis). There were no significant differences between the two discussion groups for any of the four discussion questions (see Table 7.4). Responses to the first question, “Do you think your answers were influenced by the group discussion?” did not differ significantly for Discussion Type, t (76) = 1.94, p = .06. Likewise, for the second question, “Did the discussion help you remember things from the video that you would otherwise have forgotten?” responses did not differ significantly for Discussion Type, t (87) = 1.35, p = .18. For the question, “Did you learn anything new about the video as a result of the discussions?” responses did not differ significantly for Discussion Type, t (87) = 1.87, p = .07. Finally, for the question “Did your group have an open and free exchange of ideas?” responses did not differ significantly for Discussion Type, t (87) = 0.16, p = .87.

When asked to state the purpose of the study, only one participant gave an answer which suggested that he thought he might have been shown a slightly different video from the other members of his group. However, this particular participant was in a same-video group, so the experimental manipulation was considered successful and no participants were excluded from the analyses.

Chapter 7: Unintentional Hearsay 176

Table 7.4 Response to Discussion Questions.

Question Condition M SD n

Mixed-Video 3.05 1.11 20 Do you think your answers Group were influenced by the Same-Video group discussion? 2.50 1.05 58 Group Mixed-Video Did the discussion help 2.51 1.34 59 you remember things from Group the video that you would Same-Video 2.90 1.18 30 otherwise have forgotten? Group Mixed-Video 2.31 1.34 59 Did you learn anything Group new about the video as a Same-Video result of the discussion? 2.87 1.33 30 Group Mixed-Video 1.51 0.73 59 Did your group have an Group open and free exchange of Same-Video ideas? 1.53 0.68 30 Group Note. Responses were given on a six-point scale in which a score of “1” signified “Yes” and “6” signified “No.”

Recognition Questionnaire Memory Accuracy. Participants’ responses to the recognition questions were recoded such that a correct answer was scored “1” and an incorrect answer scored “0.” Scores were then combined to give a total accuracy score for the eighteen items that were not subject to misinformation (i.e., non-misled items). A separate score was also calculated for the items for which we had attempted to mislead some participants (i.e., misled items). All results were then rescaled in the range 0-100. As mentioned previously, three of the eight items were very difficult to discern in the final version of the two videos. For this reason, the data for the misled items were analyzed excluding these items. When these three items were removed from the analyses, the results were not greatly affected, however where differences do exist, they are acknowledged in footnotes. Chapter 7: Unintentional Hearsay 177

Preliminary analyses were conducted in order to assess whether there were any differences between individuals in groups who had all seen Version A of the video and those in groups who had all seen Version B. Independent t-tests revealed that there were no significant differences between the two groups on accuracy for non-misled items, t (32) = 2.02, p = .05, or accuracy for misled items, t (32) = 0.39, p = .70. Therefore the data from the two conditions were combined as “same-video group” for all further analysis. A 2 x 2 MANOVA was conducted to investigate the effects of Discussion Type and Warning on participants’ accuracy for non-misled items (i.e., items for which we had not attempted to mislead any participants) and misled items (i.e., items for which we had attempted to mislead some participants; that is, items which differed between the two versions of the video). This analysis excluded the no-discussion control group because it did not make sense to give this group a warning when they had not discussed the video with anyone. Consequently only the mixed-video groups and the same-video groups were included in this analysis. Before conducting the MANOVA, the dependent variables for memory accuracy were correlated to check for possible multicollinearity. It was found that the dependent variables were not unacceptably highly correlated (r = -.05) and it was therefore appropriate to proceed with the MANOVA (Tabachnick & Fidell, 2001). The mean scores for memory accuracy are shown in Table 7.5. The MANOVA revealed that there was a significant effect of Discussion Type on the combined dependent variable of memory accuracy, F (2, 80) = 8.02, p < .01; Wilks’ Lambda = .8; partial eta squared = .17. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was a significant main effect of Discussion Type on non-misled items, F (1, 81) = 7.79, p < .01. Participants in the mixed-video group were more accurate on non-misled items (M = 81.68, SD = 9.27) than participants in same-video groups (M = 75.56, SD = 9.85). There was also a main effect of Discussion Type on misled items, F (1, 81) = 6.14, p < .025.17 Participants in the mixed- video groups were less accurate on misled items (M = 58.95, SD = 21.85) than participants in same-video groups (M = 70.67, SD = 17.99).

17 When all eight misled items were included in the analysis (i.e., including the three items that were difficult to discern in the final videos), there was no main effect of Discussion Type on misled items, F (1, 81) = 3.14, p = .08.

Chapter 7: Unintentional Hearsay 178

The MANOVA did not reveal a significant main effect of Warning on the combined dependent variable of memory accuracy, F (4, 160) = 1.05, p = .39. Despite the fact that warning had no effect on the mixed-video groups’ accuracy for misled items, it is interesting to note that the means were in the predicted direction. That is, when a specific warning was given, participants in the mixed-video group were more accurate (M = 61.11, SD = 18.75) than those who received a general warning (M = 58.00, SD = 19.36) and more accurate than those who received no warning at all (M = 57.89, SD = 27.40). There was no interaction between Discussion Type and Warning on the combined dependent variable of accuracy, F (4, 160) = 0.57, p = .69. A separate one-way MANOVA was conducted which included the no-discussion control group, and excluded the factor, Warning. The MANOVA revealed a significant effect of Discussion Type on the combined dependent variable of accuracy, F (4, 226) = 5.22, p < .0005; Wilks’ Lambda = .8; partial eta squared = .09. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was a significant main effect of Discussion Type on non-misled items, F (2, 114) = 5.60, p < .01. LSD post hoc tests revealed that participants in the mixed-video group were significantly more accurate on non-misled items (M = 81.68, SD = 9.27) than those in the same-video group (M = 75.56, SD = 9.85, p < .01) and the no discussion controls (M = 75.37, SD = 11.55, p < .01). The difference between the same-video group and no discussion controls on accuracy for non-misled items was not significant (p = .94). There was also a main effect of Discussion Type on misled items, F (2, 114) = 5.04, p < .0118. Participants in the mixed-video group were less accurate on misled items (M = 58.95, SD = 21.85) than participants in same-video groups (M = 70.67, SD = 17.99, p < .05) and the no discussion controls (M = 72.00, SD = 22.65, p < .01). The difference between the same- video group and no discussion controls on accuracy for misled items was not significant (p = .81).

18 When all eight misled items were included in the analysis (i.e., including the three items that were difficult to discern in the final videos), there was no main effect of Discussion Type on misled items, F (2, 114) = 2.38, p = .10.

Chapter 7: Unintentional Hearsay 179

Table 7.5 Mean Memory Accuracy for Recognition Questionnaire

Condition Non-misled Items Misled Items M SD M SD No Warning 82.46 9.85 57.89 27.40 Mixed-Video General Warning 80.83 10.11 58.00 19.36 Group Specific Warning 81.79 8.04 61.11 17.75 Total 81.68 9.27 58.95 21.85 No Warning 79.44 7.88 64.00 18.38 Same- Video General Warning 74.31 8.37 75.00 14.14 Group Specific Warning 73.15 11.80 73.33 19.69 Total 75.56 9.85 70.67 17.99 Control Group 75.37 11.55 72.00 22.65

Percentage of Participants Who Reported Misinformation on Recognition Test. The accuracy scores were recoded to indicate which participants had reported at least one item of misinformation on the recognition questionnaire. It was found that the vast majority of participants inaccurately reported at least one misled item across all three conditions. That is, 93% of the participants in the same-video group, 93% of those in the mixed-video group, and 83% of those in the control group inaccurately responded to at least one question for which we had attempted to mislead some participants. A chi-square analysis revealed that there was no association between Discussion Type and whether the participants reported at least one item of misinformation on the recognition questionnaire (χ2 (2, N = 119) = 2.64, p = .27). Memory Confidence. Participants recorded their confidence using a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” Scores were added together to obtain separate confidence scores for misled and non-misled items. After noting that the dependent variables with regards to memory confidence were not unacceptably highly correlated (r = .59), a 2 x 2 MANOVA was conducted which analyzed the effects of Discussion Type and Warning on participants’ confidence for misled and non-misled items. Only the mixed-video groups and the same-video groups Chapter 7: Unintentional Hearsay 180

were used in this analysis because the no-discussion control group did not receive any warnings. The mean scores for memory confidence are shown in Table 7.6. The analysis revealed that there was no main effect of Discussion Type on the combined dependent variable of confidence, F (2, 73) = 0.32, p = .73. There was also no main effect of warning on the combined dependent variable of confidence, F (4, 146) = 0.42, p = .79. Furthermore, there was no significant interaction on the combined dependent variable of confidence between Discussion Type and Warning, F (4, 146) = 1.59, p = .18. A separate one-way MANOVA was conducted which included the no-discussion control group, but omitted the factor, Warning. This analysis revealed that there was no main effect of Discussion Type on the combined dependent variable of confidence, F (4, 208) = 1.11, p = .35.

Table 7.6

Mean Memory Confidence for Recognition Questionnaire

Condition Non-misled Items Misled Items M SD M SD No Warning 4.74 0.39 4.44 0.73 Mixed-Video General Warning 4.45 0.68 3.77 1.19 Group Specific Warning 4.51 0.65 4.45 0.98 Total 4.55 0.60 4.19 1.04 No Warning 4.47 0.66 3.96 1.16 Same-Video General Warning 4.31 0.29 4.37 0.83 Group Specific Warning 4.56 0.80 4.03 1.21 Total 4.47 0.65 4.09 1.09 Control Group 4.44 0.69 3.87 1.04

Confidence in Correct and Incorrect Decisions. Confidence was also calculated so that it reflected confidence in correct responses separately from incorrect responses. Therefore, in this analysis there were four dependent variables (confidence for correct items when given misleading information, confidence for incorrect items when Chapter 7: Unintentional Hearsay 181

given misleading information, confidence for correct items when given no postevent information, confidence for incorrect items when given no postevent information). After noting that the dependent variables were not unacceptably highly correlated (r’s ranged from 0.28 to 0.58), a 2 x 2 MANOVA was conducted to investigate the effects of Discussion Type and Warning on the four dependent variables relating to confidence in correct and incorrect judgments. Only the mixed-video groups and the same-video groups were used in this analysis because the no-discussion control group did not receive any warning. The mean scores for confidence in correct and incorrect decisions are shown in Tables 7.7 and 7.8. The analysis revealed that there was no main effect of Discussion Type on the combined dependent variable of confidence in correct and incorrect judgments, F (4, 63) = 0.93, p = .45. There was also no main effect of warning on the combined dependent variable of confidence in correct and incorrect judgments, F (8, 126) = 0.53, p = .84. Furthermore, there was no significant interaction between Discussion Type and Warning on the combined dependent variable of confidence in correct and incorrect judgments, F (8, 126) = 1.08, p = .38. A separate one-way MANOVA was conducted which included the no-discussion control group, but omitted the factor, Warning. The MANOVA revealed that there was no effect of Discussion Type on the combined dependent variable of confidence in correct and incorrect judgments, F (8, 172) = 1.60, p = .13.

Chapter 7: Unintentional Hearsay 182

Table 7.7

Mean Confidence in Correct Decisions

Condition Non-misled Items Misled Items n M SD M SD No Warning 12 5.04 0.39 4.51 1.01 Mixed- General Warning 17 4.68 0.73 3.76 1.58 Video Group Specific Warning 16 4.87 0.57 4.70 0.97 Total 45 4.84 0.60 4.29 1.29 No Warning 10 4.91 0.51 3.93 1.58 Same- General Warning 7 4.90 0.27 4.79 0.70 Video Group Specific Warning 10 5.01 0.56 4.23 1.26 Total 27 4.94 0.47 4.27 1.28 Control Group 25 4.61 0.85 3.64 1.20

Table 7.8

Mean Confidence in Incorrect Decisions

Condition Non-misled Items Misled Items n M SD M SD No Warning 12 3.56 1.09 2.24 2.02 Mixed- General Warning 17 2.72 1.26 1.87 0.96 Video Group Specific Warning 16 3.09 1.35 2.23 1.46 Total 45 3.08 1.27 2.10 1.45 No Warning 10 2.91 1.17 2.78 1.47 Same- General Warning 7 2.71 0.87 2.98 1.66 Video Group Specific Warning 10 3.52 1.16 2.74 2.00 Total 27 3.08 1.11 2.82 1.67 Control Group 25 3.07 1.16 3.01 1.04

Free Recall Coding. A coding system for the free recall was developed based on one used by Yarmey and Morris (1998). Each narrative was scored in terms of: (a) number of accurate Chapter 7: Unintentional Hearsay 183

propositions (regarding sequence of events, character descriptions, dialog, setting, and items stolen); (b) conciseness (accuracy divided by number of words written); (c) fabrications (spontaneous errors); and (d) amount of misleading postevent information reported. Four independent judges repeatedly viewed the crime video and independently coded the free recall. The judges were unaware of the conditions to which each participant was assigned. Each participant’s responses were randomly assigned to one of the four coders. Twenty of the free recall narratives were scored by all coders in order to assess inter-rater reliability. There was a significant correlation between the four coders on accurate propositions and misleading postevent information (see Table 7.9 for correlation coefficients). According to Elmes, Kantowitz & Roediger (1991), high inter-rater reliability occurs when the correlation coefficient is greater than +0.70. The correlations between the scorers for accurate propositions and misleading postevent information met this criterion. However, despite best efforts of the coders, the correlations for fabrications were very low. The correlations for fabrications may be low because there were not many fabrications reported. When there was disagreement between the judges, one score was randomly chosen to use for the analysis.

Table 7.9 Inter-rater Reliability for Free Recall Coding

Correlations between scorers: 1&2 1&3 1&4 2&3 2&4 3&4 Accurate .89** .93** .92** .87** .80** .89** propositions Fabrications .10 .36 .24 .08 .44 .33 Misleading postevent .85** .90** .83** .95** .90** .94** information Note. n = 20, ** p < 0.01, two-tailed

Memory Accuracy. After establishing that the dependent variables for Free Recall Accuracy were not unacceptably highly correlated (r’s ranged from -0.11 to 0.14) a 2 x 2 Chapter 7: Unintentional Hearsay 184

MANOVA was conducted in which the effects of Discussion Type and Warning on the combined dependent variable of free recall accuracy (accurate propositions, conciseness, fabrications, misleading postevent information reported) were analysed. Only the mixed- video groups and the same-video groups were included in this analysis because the no- discussion control group did not receive any warning. The mean scores for memory accuracy are shown in Table 7.10. The analysis revealed that there was a significant main effect of Discussion Type on the combined dependent variable of free recall accuracy, F (4, 80) = 3.82, p < .01; Wilks’ Lambda = .8; partial eta squared = .16). Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .013, showed that there was a significant main effect of Discussion Type on accurate propositions, F (1, 83) = 6.55, p < .013. Participants in the mixed-video group were more likely to report accurate propositions (M = 49.81, SD = 13.19) than participants in same-video groups (M = 43.63, SD = 7.65). There was also a main effect of Discussion Type on misleading postevent information, F (1, 83) = 6.65, p < .013. Participants in the mixed-video group were more likely to report misleading postevent information (M = 0.32, SD = 0.68) than participants in same-video groups (M = 0.00, SD = 0.00). There was no main effect of Discussion Type on conciseness, F (1, 83) = 1.51, p = .22, or fabrications, F (1, 83) = 0.36, p = .55. There was no main effect of warning on the combined dependent variable of accuracy, F (8, 160) = 0.84, p = .57, and no significant interaction between Discussion Type and Warning on the combined dependent variable of accuracy, F (8, 160) = 1.24, p = .28. A separate one-way MANOVA was conducted which included the no-discussion control group, but omitted the factor, Warning. The analysis revealed that there was a main effect of Discussion Type on the combined dependent variable of free recall accuracy, F (8, 224) = 3.19, p < .01; Wilks’ Lambda = .8; partial eta squared = .10). Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .013, showed that there was a significant main effect of Discussion Type on accurate propositions, F (2, 115) = 7.02, p < .01. LSD post hoc tests revealed that participants in the mixed-video group were significantly more likely to report accurate propositions (M = 49.81, SD = 13.19) than participants in same-video groups (M = 43.63, SD = 7.65, p < .05) and the no discussion Chapter 7: Unintentional Hearsay 185

control group (M = 40.72, SD = 10.76, p < .01). There were no significant differences between the same-video groups and the control group (p = .33). There was also a main effect of Discussion Type on misleading postevent information, F (2, 115) = 5.04, p < .01. LSD post hoc tests revealed that participants in the mixed-video group were significantly more likely to report misleading postevent information (M = 0.32, SD = 0.68) than participants in same-video groups (M = 0.00, SD = 0.00, p < .01) and the no discussion control group (M = 0.07, SD = 0.26, p < .05). There were no significant differences between the same-video groups and the control group (p = .60). There was no main effect of Discussion Type on conciseness, F (2, 115) = 0.95, p = .69, or fabrications, F (2, 115) = 0.33, p = .72.

Table 7.10

Mean Memory Accuracy for Free Recall

Misleading Accurate Condition Conciseness Postevent Fabrication Propositions Information M SD M SD M SD M SD No 54.60 8.97 0.15 0.03 0.35 0.67 2.40 2.19 Warning General Mixed- 44.52 15.43 0.15 0.03 0.19 0.51 2.86 2.57 Warning Video Specific Group 50.67 12.70 0.16 0.04 0.44 0.86 1.94 1.76 Warning Total 49.81 13.19 0.15 0.03 0.32 0.68 2.42 2.21 No 40.90 8.70 0.16 0.03 0.00 0.00 2.10 1.73 Warning General Same- 42.88 4.16 0.18 0.03 0.00 0.00 2.25 2.12 Warning Video Specific Group 46.42 8.11 0.15 0.02 0.00 0.00 2.00 1.60 Warning Total 43.63 7.65 0.16 0.03 0.00 0.00 2.10 1.73

Control Group 40.72 10.76 0.16 0.03 0.07 0.26 2.17 1.61

Chapter 7: Unintentional Hearsay 186

Percentage of Participants Who Reported Misinformation on Free Recall. The accuracy scores were recoded to indicate those participants who, during free recall, mentioned at least one item of misleading postevent information. It was found that 22% of participants in the mixed-video group reported at least one misled item, while only 10% in the control group and 0% in the same-video group inaccurately reported at least one misled item. A chi-square analysis revealed a significant association between Discussion Type and whether or not participants mentioned at least one item of misleading postevent information (χ2 (2, N = 119) = 8.71, p < .05).

Remember/Know Judgments Only data from those in the mixed-video group were used in the analysis to determine whether Remember/Know Judgments differ for those participants whom we had attempted to mislead and those whom we had not. The analysis included items for which participants who had seen Version B had potentially been misled by those in the group who had seen Version A. The number of times each participant reported they “remembered” an item to be true as opposed to “knew” it to be true were calculated for the items. Next, a one-way ANOVA was conducted to investigate the effects of Video Version on remember judgments. The factor, Video Version, was found to be significant, F (1, 57) = 31.73, p < .0005. Participants who viewed Version A (i.e., those who had seen the items in the video and were therefore not misled) were more likely to give a remember judgment (M = 2.88, SD = 1.13) than those who viewed Version B (i.e., those who had not seen the items in the video and were misled; M = 1.26, SD = 1.06). A second ANOVA was conducted to investigate the effects of Video Version on know judgments. Once again the factor, Video Version, was found to be significant F (1, 57) = 4.49, p < .05. Participants who viewed Version B (i.e., those who had not seen the items in the video and were misled) were more likely to give a know judgment (M = 1.78, SD = 1.19) than those who viewed Version A (i.e., those who had seen the items in the video and were therefore not misled; M = 1.22, SD = 0.83). In a separate analysis that included all participants, the proportion of times that participants were accurate when they indicated they ‘remembered’ an item to be true was Chapter 7: Unintentional Hearsay 187

calculated (i.e., remember accuracy = the number of times participants indicated ‘remember’ for an accurate decision divided by the total number of times they indicated ‘remember’). Similarly the proportion of times that participants were accurate when they indicated they ‘knew’ an item to be true was calculated (i.e., know accuracy = the number of times participants indicated ‘know’ for an accurate decision divided by the total number of times they indicated ‘know’). A paired samples t-test was then conducted which compared remember accuracy and know accuracy and a significant difference was found, t (116) = 8.42, p < .0005. Participants were more likely to be accurate if they indicated ‘remember’ (M = 90.82, SD = 11.77) than if they indicated ‘know’ (M = 72.53, SD = 23.13). Inspection of the confidence intervals indicated that participants were 14-23% more likely to be accurate if they indicated ‘remember’ than if they indicated ‘know.’

Eyewitness Identification Participants were asked to identify the perpetrator from a lineup at two separate times: before and after the manipulation of Discussion Type. Identification Accuracy. Line-up decisions provided by participants were categorised as: (1) correct selection (i.e., participant correctly selected the perpetrator from the version of the video they viewed); (2) other target (i.e., participant incorrectly selected the perpetrator from the alternative version of the video to the one they viewed); (3) incorrect selection (i.e., participant incorrectly selected one of the foils); or (4) no selection (i.e., participant incorrectly indicated that the perpetrator was not in the lineup). The pre- and post-discussion manipulation identification decisions can be seen in Table 7.11. A chi-square analysis was conducted to determine whether there was an association between Discussion Type and identification accuracy. When participants were asked to individually identify the perpetrator from a lineup prior to the manipulation of Discussion Type, there was no association between Discussion Type and identification accuracy, χ2 (6, N = 119) = 1.80, p = .94. Similarly, there was no association between Discussion Type and identification accuracy when participants were asked to individually identify the perpetrator from a lineup after the Discussion Type manipulation, χ2 (6, N = 119) = 4.98, p = .55.

Chapter 7: Unintentional Hearsay 188

Table 7.11

Identification Decisions Made Before and After the Discussion Manipulation

Correct Other Incorrect Condition No choice Selection target selection

Mixed-Video 19 (32%) 3 (5%) 13 (22%) 24 (41%) Pre- Group Discussion Same-Video 9 (30%) 2 (7%) 9 (30%) 10 (33%) Manipulation Group Controls 10 (33%) 2 (7%) 5 (17%) 13 (43%) Mixed-Video 18 (31%) 5 (8%) 10 (17%) 26 (44%) Post- Group Discussion Same-Video 10 (33%) 1 (3%) 10 (33%) 9 (30%) Manipulation Group Controls 10 (33%) 1 (3%) 6 (20%) 13 (43%) Note. Values represent the number (and percentage) of participants who made each type of identification decision.

Memory Confidence. Participants recorded their confidence in the lineup decision using a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” A one-way ANOVA was conducted to investigate the effects of Discussion Type (mixed-video group, same-video group, no discussion) on lineup decision confidence. It was found that Discussion Type did not have any effect on lineup decision confidence either before the manipulation of Discussion Type, F (2, 112) = 0.66, p = .52, or after the manipulation, F (2, 115) = 1.01, p = .37. Identification confidence can be seen in Table 7.12.

Chapter 7: Unintentional Hearsay 189

Table 7.12

Identification Confidence Before and After the Discussion Manipulation

Condition Confidence Mixed-Video Group 2.93 Pre-Discussion Same-Video Group 3.24 Manipulation Controls 3.17 Mixed-Video Group 2.78 Post-Discussion Same-Video Group 3.13 Manipulation Controls 3.20 Note. A score of “1” indicates “Low confidence” and “6” indicates “High confidence.”

Decision Change. A chi-square analysis was conducted to determine whether Discussion Type (mixed-video group, same-video group, no discussion) was associated with whether or not participants changed their identification decision between the first and second lineups. The analysis revealed that Discussion Type was not associated with decision change, χ2 (2, N = 119) = 3.21, p = .20. When the same-video groups and no- discussion controls were combined and compared against the mixed-video group, decision change approached significance, χ2 (1, N = 119) = 3.21, p = .07. Participants in the mixed- video groups were slightly more likely to change their identification than those not in a mixed-video group, however this was not significant. Decision changes were categorised as: (1) erroneous change (i.e., participant initially made a correct identification decision and changed to an incorrect decision after the discussion manipulation); (2) correct change (i.e., participant initially made an incorrect identification decision and later changed to a correct decision); or (3) no change (i.e., participant selected the same identification decision both before and after the manipulation). The number (and percentage) of participants who made a decision change after the discussion manipulation can be seen in Table 7.13. A chi-square was used to determine the association between Discussion Type and the type of decision change that was made (erroneous change, correct change, no change). Again, Discussion Type was not associated with the type of decision change made, χ2 (4, N = 119) = 4.23, p = .38. When the same-video groups and control groups were combined and Chapter 7: Unintentional Hearsay 190

compared against the mixed-video group, the type of decision change made was still not significant, χ2 (2, N = 119) = 3.55, p = .17.

Table 7.13

Identification Decision Changes made by Participants After the Discussion Manipulation

Erroneous Correct Condition No Change Change Change Mixed-Video Group 2 (3%) 3 (5%) 54 (92%) Same-Video Group 1 (3%) 0 (0%) 29 (97%) Controls 0 (0%) 0 (0%) 30 (100%) Note. Values represent the number (and percentage) of participants who made a decision change after the discussion manipulation.

Discussion The findings reported in the current experiment suggest that participants often report misinformation that had previously been suggested to them by a co-witness. Furthermore, even when participants are warned that the co-witness information they received may be incorrect they still report it. However, the findings also suggest that there may be a difference between ‘real’ memories for an event and those obtained from a different source as indicated by remember/know judgments (Tulving, 1985). That is, participants were more likely to indicate that the misinformation they reported was “known” rather than “remembered.” Our first aim was to determine whether the experimental manipulation was successful. Participants were asked four questions relating to their discussion experiences in order to determine whether participants in the mixed-video group had similar experiences to those in the same-video groups. We found no significant differences on any of the four questions. This implies that participants in mixed-video groups were unaware of the manipulation, as they reported their discussions to be just as “influential, informative, open, and free” as the participants in same-video groups. Participants were also asked to state what they believed to be the purpose of the study. Only one participant guessed that he may have been shown a slightly different video from other members in his group, however, this Chapter 7: Unintentional Hearsay 191

participant was actually in a same-video group, which again suggests that the experimental manipulation was successful. Our second research question asked: how does Discussion Type (mixed-video group, same-video group, no discussion) and Warning (general, specific, or no warning) influence participants’ memory accuracy and confidence for misled and non-misled items? As hypothesised, participants in the mixed-video groups were more likely to report misled items than those in the same-video groups and the no-discussion controls. That is, participants who discussed the video with people who had seen an alternate version, erroneously incorporated information from the alternate version into their own individual account. This was true for both the recognition and free recall questionnaires. The finding that participants report Stimulus Induced Information when later giving their individual accounts supports findings from previous studies (e.g., Gabbert, Memon, & Allan, 2003; Kanematsu, Mori, & Mori, 2003; Mori & Mori, 2004; Wright, Self, & Justice, 2000). Whilst participants in mixed-video groups were less accurate for misled items than those in the same-video groups and the no-discussion controls, they were more accurate for non-misled items. That is, participants in the mixed-video groups gave more accurate accounts for non-misled items than those in the same-video groups and the no-discussion controls. This was true for both the recognition and free recall questionnaires. This suggests that disagreement about some items appears to have been beneficial for the recall of other non-misled items. This finding supports the finding from Study 3 that participants who received misinformation were more accurate on the non-misled items than those who did not receive misinformation. In Study 3, the possibility existed that this finding was a result of the confederates over-assisting the group. However, because confederates where not used in the current study, this explanation cannot be used here. A possible explanation for the finding in the current study is that disagreement in groups can generate better discussion in which more information is elicited from the other group members. Research on “devil’s advocates” has shown that group discussion can be enhanced when a member voices a differing opinion (e.g., Janis, 1972). Alternatively, when faced with differing information, participants may have steered the group discussion away from topics on which people did not agree (i.e., misled items). Research on the collective information sharing bias suggests that there is a tendency for group discussions to focus on the reiteration of information that Chapter 7: Unintentional Hearsay 192

group members share in common, with little attention being paid to the exposure of information that is unique to individual members (Wittenbaum, Hubbell, & Zuckerman, 1999). Finally, another explanation for the findings is that group members may have questioned their own recollections when faced with a differing opinion about some items, and attempted to place themselves back at the time of viewing the incident to refresh their memories. This self-initiated context reinstatement (Tulving & Thomson, 1973) may have improved their memories for non-misled items. There were no differences in memory accuracy found between participants in same- video groups and the no-discussion controls on either the misled or non-misled items. That is, if postevent information is not deliberately introduced by the experimenter, there is no advantage or disadvantage of group discussion prior to individual testimony. This finding that Natural Discussion Groups are just as accurate as no-discussion controls is in accordance with the findings from Study 3. Natural Discussion Groups have potentially the highest ecological validity, and yet have been completely neglected as a comparison group by studies investigating the effects of Stimulus Induced Information. Whilst memory accuracy was influenced by the type of discussion in which participants took part, the same was not true for memory confidence. Participants’ confidence in their memories was unaffected by the type of discussion they experienced. This was true for straightforward confidence judgments as well as confidence in correct and incorrect answers. This means that even though participants in the mixed-video discussion groups were more likely to report misinformation, they were equally confident in their false memories as those who were not given misinformation. The finding that misled participants often have as much confidence in their recognition judgments as non-misled participants is consistent with past research (Loftus, Donders, Hoffman, & Schooler, 1989). This suggests that eyewitness confidence cannot be used to distinguish real memories from memories obtained second-hand from a co-witness. This study also assessed the effect of Warning on memory accuracy and confidence. As hypothesized, the results from this experiment indicated that even when people are warned that the co-witness information they received may be incorrect they still believe they saw it. That is, warning the participants about possible co-witness misinformation did not eliminate the reporting of misinformation. This finding is in line with research on co- Chapter 7: Unintentional Hearsay 193

witness discussion (Meade & Roediger, 2002) as well as other ways of encountering postevent misinformation (e.g., Betz et al., 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, in press; McDermott & Roediger, 1998; Wright, 1993). The fact that participants report co-witness information despite warnings against doing so suggests that they may be unable to distinguish what they actually witnessed from the information they discussed with the co-witness. This would support the notion that memory change is responsible for the conformity observed. However, despite the fact that warnings did not eliminate the effects of inaccurate co-witness information, it did help reduce its effects (although not significantly). That is, participants who received specific warnings tended to be slightly less likely to report misinformation than those who did not receive warnings. Other research on co-witness discussion has found that warnings could significantly reduce, but not eliminate the deleterious effects of co-witness discussion (Meade & Roediger, 2002). The discrepancy between the results of the current study and those of Meade and Roediger may be due to the fact that participants in the current study were given a longer delay between the presentation of postevent information and the warning. Source misattribution is more likely to occur after a longer delay (Underwood & Pezdek, 1998) and therefore the warnings may not be as effective after a longer delay because participants are unable to decipher whether the memory was from the original scene or co-witness discussion. A third question this study was designed to investigate was whether Remember/Know Judgements (Tulving, 1985) differ for those participants whom we had attempted to mislead and those whom we had not. It was found that participants whom we had attempted to mislead were more likely to give “know” judgments than those whom we did not misled. “Know” responses are typically believed to reflect retrieval from generic (semantic) memory (Tulving, 1985) or retrieval based on global familiarity of the event (e.g., Jacoby, Yonelinas, & Jennings, 1997). Conversely, participants who were not misled were more likely to give “remember” judgments than those whom we had attempted to mislead. In a separate analysis, it was found that participants were 14-23% more likely to be accurate if they indicated they consciously ‘remembered’ an item was in the video than if they indicated they ‘knew’ the statement was true, but it failed to evoke any specific recollection from the video. Chapter 7: Unintentional Hearsay 194

The remember/know findings from this experiment are consistent with those of Roediger et al. (2001) who found that when participants reported misinformation, they were more likely to report they “knew” the misled item was in the original scene, rather than reporting they consciously “remembered” it was there. “Describing false recognition judgements as known rather than remembered is the typical outcome of most experiments” (Roediger et al., p. 368). Roediger et al. use the source monitoring theory (e.g., Johnson et al., 1993; Lindsay & Johnson, 1989) to explain this finding. They suggest that most reported misinformation tends to be “known” rather than “remembered” because participants have difficulty locating the source of their memories. They argue that if a source could be confidently attributed, the item should be judged as “remembered.” Whilst people were more likely to report that misinformation was “known” rather then “remembered” some participants in the current study did report that they consciously “remembered” a misled item. Other studies have shown that misinformation can be accompanied by remember judgments (e.g., Higham & Vokey, 2004; Roediger & McDermott, 1995). When false memories are accompanied by “remember” judgments, this implies that memory change has occurred, (Meade & Roediger, 2002). However, false “know” responses are more ambiguous in their meaning, as they may imply either public or private conformity (Meade & Roediger). However, as Meade and Roediger suggest, if participants give “know” responses, it is less likely that their responses would reflect public conformity if they are following the remember/know instructions typically given (e.g., Rajaram, 1993; Tulving, 1985) as they were in the current study. A limitation of the remember/know methodology utilised in the current study is that participants were not allowed the option of “guess.” That is, participants were asked to indicate either “know” or “remember” to items they indicated were true, when in fact, the participants may have simply guessed the answer. Therefore, “know” responses may have been confounded when participants were merely guessing (Gardiner, Richardson-Klavehn, & Ramponi, 1997). Future studies in the current thesis take this experimental confound into account. Finally, this study was designed to investigate the effects of Discussion Type on participants’ ability to identify the perpetrator from a lineup. It was found that Discussion Type was not associated with participants’ identification accuracy. These results are Chapter 7: Unintentional Hearsay 195

consistent with previous findings (e.g., Yarmey & Morris, 1998) although we did not find a slight trend in favour of groups, as Warnick and Sanders (1980) observed. Discussion Type was also not associated with participants’ confidence in their identification decisions. This finding was contrary to those of Luus and Wells (1994), who found that witness confidence changed on the basis of co-witness information. It is possible that witnesses in the present study did not change their confidence ratings or identification decisions because the members of their group may have been so unsure about their own decisions that they did not readily influence one another. Furthermore, the participants in the current study were questioned about their identification before the discussion manipulation, as well as after. This means that they may have been reluctant to change their confidence and identification decisions following the discussion, not wanting to appear easily influenced. Research in social psychology has shown that individuals are more likely to resist conformity if they have already committed themselves to a position (Deutsch & Gerard, 1955). Furthermore, research on eyewitness memory has shown that the misinformation effect is reduced if people are required to make a commitment to a response prior to receiving misleading information (Loftus, 1977). However, despite this reason for reluctance, participants in the mixed-video groups were slightly (but not significantly) more likely to change their identification following co-witness discussion than those not in a mixed-video group. Despite the statistically non-significant results, it is interesting to note that five of the seven participants who selected the target from the alternative video sequence after the discussion manipulation were in the mixed-video group. Study 5 addressed some of the limitations of the previous experiments described in this thesis. For example, Study 5 improved on the methodology used in Studies 3 and 4 in that instead of using confederates, Study 5 used a Stimulus Induced Information methodology. Despite the use of a different methodology, similar results were obtained. That is, participants who received postevent information from a co-witness were more accurate for non-misled items and less accurate for misled-items than participants who didn’t receive postevent information. These results confirm the findings from Studies 3 and 4 and strengthen the argument that co-witness discussion can be both beneficial and detrimental because the problems associated with the use of confederates were eliminated from this study. Chapter 7: Unintentional Hearsay 196

Another strength of Study 5 is that it investigated why participants change their responses following the acquisition of co-witness information. In order to investigate the mechanisms responsible for memory conformity, the present study used remember/know judgments (Tulving, 1985) and warnings. The results from the remember/know judgments suggest that it may be possible to distinguish between ‘real’ memory and information obtained from a co-witness. In sum, the results from this study suggest that there are both advantages and disadvantages of co-witness discussion, when misinformation is experimentally induced. Participants who were in mixed-video groups were less accurate on the misled items than those who didn’t receive the misinformation. However, the participants who were in mixed- video groups were also more accurate on the non-misled items than those who weren’t in mixed-video groups. When misinformation is not experimentally induced, there was no difference between individuals who discussed the event (i.e., same-video groups) and those who did not. The results from this experiment indicate that even when people are warned that the co-witness information they received may be incorrect, they still cannot edit it from their personal account of the event. This suggests that to some extent people are unable to distinguish what they actually witnessed from the information they discussed with the co- witness. However, participants were more likely to indicate that the misled items were “known” rather than “remembered” which suggests that there may be a difference between ‘real’ memories for an event from those based on information obtained from a different source.

197

Chapter 8 Study 6: Source Monitoring

Results of Study 5 suggest that it may be possible to use remember/know judgments (Tulving, 1985) to distinguish between real memories and those obtained second-hand. While this finding has important theoretical implications, it has limited practical implications. The distinction between “remember” and “know” can be difficult to explain and police officers may not feel comfortable with the procedure because it does not have high face validity. In Study 6, a different type of test was used to see if there is a more explicit way to distinguish between real memories and those obtained from a co-witness. Participants were given a source monitoring test that was similar to that used by Lindsay and Johnson (1989). Source monitoring tests give participants the option of saying that they remember an item only from the postevent suggestion, and not from the original stimulus. This option can help determine whether participants are able to distinguish between information that they witnessed in the original scene and information that they learned second-hand. For example, participants in the current study were asked to indicate whether they remembered the item from the video only, from the discussion only, or from both the video and discussion. Studies have shown that source monitoring tests such as these tend to eliminate the standard misinformation effect (Lindsay & Johnson, 1989; Zaragoza & Koshmider, 1989). However, some evidence suggests that co-witness discussion is strong enough to persist even when participants are asked to monitor the source of the information they recall (Meade & Roediger, 2002). In their studies, Meade and Roediger showed participants a series of slides. They then asked them to recall items from the slides with a confederate who erroneously reported some misinformation. After recalling the items with a confederate, participants were asked to individually complete a recognition questionnaire and indicate whether they remembered the information they reported from: (1) the slide only, (2) the confederate only, (3) both the slide and confederate, or (4) neither the slide nor the confederate. It was found that participants reported the misinformation despite the fact that they were asked to monitor the source of their information, and they incorrectly Chapter 8: Source Monitoring 198

reported that they had seen the misinformation in the original slide. The aim of the current thesis was to investigate this finding within a more ecologically valid context. As previously mentioned, the studies reported by Meade and Roediger use photos as stimuli and do not allow for natural, free flowing discussion, as participants are required to answer questions in the presence of a confederate. In the current study, videos will be used as stimuli and participants will be able to freely discuss the event with a co-witness. In accordance with the findings of Meade and Roediger, it was hypothesised that source monitoring would not eliminate the misinformation effect, but may help reduce it. This study also attempted to investigate, under slightly different conditions, the result found in Study 5 that warning participants about possible co-witness misinformation does not have an effect. In Study 5, there were three levels of warning (no warning, general warning, specific warning), however for Study 6 the warnings were dichotomized (i.e., specific warning, no warning). Some of the participants were warned prior to completing the individual memory tasks that members of their discussion group might have seen a slightly different video. The other participants did not receive any warning at all. Previous research on co-witness discussion (Meade & Roediger, 2002) as well as other modes of encountering misinformation (e.g., Betz et al., 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, in press; McDermott & Roediger, 1998; Wright, 1993) suggests that the misinformation effect may be reduced, but not eliminated when warnings are used. It was therefore hypothesised that memory conformity in the present study would be reduced, but not be eliminated by the warning. Because memory conformity may be reduced when participants are given a warning (e.g., Betz, Skowronski, & Ostrom, 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, in press; McDermott & Roediger, 1998; Wright, 1993), and may be eliminated when they are given a source monitoring test (Lindsay & Johnson, 1989) it is possible that the memory conformity effect which occurred in the prior studies in this thesis may be eliminated altogether under these rigorous testing conditions. One further distinction between Study 5 and Study 6 is that the current study excludes the ‘No Discussion’ control group. The ‘No Discussion’ control group was omitted from the study because both Study 3 and Study 5 found no differences between the no discussion control group and the groups who received no induced information (i.e., the Chapter 8: Source Monitoring 199

Natural Discussion Groups). As mentioned previously, the Natural Discussion Group methodology may be a more appropriate control group for these studies. Furthermore, excluding the ‘No Discussion’ control group simplifies the design because it does not make sense to give them any warnings when they have not discussed the video with anyone. In accordance with the findings from Study 5, it was hypothesized that groups who received Stimulus Induced Information (i.e., ‘mixed-video groups’ in which members of the groups saw different versions of the eyewitness video) would have less accurate recall on the misled items than Natural Discussion Groups (i.e., ‘same-video groups’ in which all participants saw the same video). In Study 5 it was also found that groups who received Stimulus Induced Information unexpectedly performed better on non-misled items than Natural Discussion Groups. It was therefore hypothesized that groups with Stimulus Induced Information would have more accurate recall on the non-misled items than groups with no induced misinformation.

Method

Design This study employed a design similar to that used in Study 5, except that the ‘General Warning’ condition and the ‘No Discussion’ control group were excluded. Therefore, the study employed a 2x2 experimental design, which investigated the effects of Discussion Type (same-video group, mixed-video group) and Warning (no warning, specific warning) on memory for misled and non-misled items. The dependent variables were memory accuracy measured using free recall and recognition questionnaires, memory confidence, and participants’ judgments about the source of their memory.

Participants Participants were 65 undergraduate psychology students (49 female, 16 male) from the University of New South Wales. The average age of the participants was 20.9 (SD = 2.32) years. Participants were invited to volunteer and take part in the experiment during a class tutorial. They received no incentives for their participation in the study. Chapter 8: Source Monitoring 200

Materials Eyewitness Stimuli. The current study used the same eyewitness stimuli used in Study 5. As in Study 5, there were two slightly different versions of the video (Version A and Version B), but each participant only saw one version. Recognition Questionnaire. The current study used the same 32 questions used in Study 5, however the instructions were changed slightly (see Appendix M). As in Study 5, participants were asked to indicate whether the statement was true or false and rate their confidence on a 6-point scale (1= low confidence; 6= high confidence). However, instead of making remember/know judgments, participants were asked to make a source monitoring judgment. That is, for the items they said were “true”, they were asked to indicate whether they were remembered from: (1) the video only, (2) the discussion only, or (3) both the video and the discussion. Free Recall Form. The current study used the same Free Recall Form that was used in Study 5. Discussion Questions Manipulation Check. Participants were asked the same Discussion Questions used in Study 5. Post-Debriefing Check for Awareness of Manipulation. After being partially debriefed about the study, participants were given a sheet that asked, “Do you think you were in a discussion group in which other members saw a different version of the video than the one you saw?” If they agreed with this statement, participants were asked to write why they believed other members saw a different version of the video.

Procedure Study 6 took place during a class tutorial and followed the same procedure as Study 5. Briefly summarised, students were led to believe that they were all viewing the same crime video, however in fact they were each shown one of two slightly different versions of the video. After a delay of twenty minutes (during which the participants did their regular course work), they were split into groups of four to discuss the crime scene. Unbeknown to the participants, groups were arranged so that some groups contained individuals who had Chapter 8: Source Monitoring 201

all seen the same version of the video (same-video groups) and some groups contained two participants who had seen Version A and two who had seen Version B of the video (mixed- video groups). Participants were given ten minutes to discuss the video using a guided recall procedure at the end of which participants were asked not to discuss the event any further. The discussions were tape-recorded. After a delay of one week, participants were asked to complete the Discussion Questions and state what they believed to be the purpose of the experiment. Participants were then assigned to one of two warning conditions. Some of the participants were given a ‘specific warning’ that some students in their group may have seen a slightly different video and that they should answer the questions solely on the basis of what they themselves remember from the video. The other students did not receive any warning at all. Participants were then asked to complete the Free Recall Form, followed by the Recognition Questionnaire, which included source monitoring judgments. Once participants had completed the questionnaires, they were partially debriefed about the study. They were told that they might have been shown a slightly different version of the video than the other members in their discussion group. After the partial debriefing, participants were given the Post-Debriefing Awareness of Manipulation questionnaire to complete. Finally, participants were fully debriefed, thanked for their participation, and asked not to discuss the experiment with other potential participants.

Results Statistical analysis focussed on the following questions: First, were the experimental manipulations successful? Second, how did Discussion Type (mixed-video group, same- video group) and Warning (specific or no warning) influence participants’ accuracy and confidence for misled and non-misled items? Finally, did source monitoring judgments differ between participants whom we had attempted to mislead and those whom we had not? Chapter 8: Source Monitoring 202

Experimental Manipulation Checks Discussion Questions Manipulation Check. Independent t-tests were used to explore responses to four questions relating to the participants’ experiences in their discussion groups. There were no significant differences between the two discussion groups for any of the four discussion questions (see Table 8.1). Responses to the first question, “Do you think your answers were influenced by the group discussion?” did not differ significantly for Discussion Type, t (63) = 0.44, p = .66. Likewise, for the second question, “Did the discussion help you remember things from the video that you would otherwise have forgotten?” responses did not differ significantly for Discussion Type, t (62) = 0.59, p = .55. For the question, “Did you learn anything new about the video as a result of the discussions?” responses did not differ significantly for Discussion Type, t (62) = 1.92, p = .06. Finally, for the question “Did your group have an open and free exchange of ideas?” responses did not differ significantly for Discussion Type, t (62) = 0.55, p = .58.

Table 8.1

Response to Discussion Questions.

Question Condition M SD n

Mixed-Video 3.37 1.21 31 Do you think your answers Group were influenced by the Same-Video group discussion? 3.24 1.28 34 Group Mixed-Video Did the discussion help 2.65 1.23 31 you remember things from Group the video that you would Same-Video 2.82 1.10 33 otherwise have forgotten? Group Mixed-Video 2.29 1.04 31 Did you learn anything Group new about the video as a Same-Video result of the discussion? 2.81 1.16 33 Group Did your group have an Mixed-Video 1.58 0.76 31 open and free exchange of Group Chapter 8: Source Monitoring 203

ideas? Same-Video 1.48 0.62 33 Group Note. Responses were given on a six-point scale in which a score of “1” signified “Yes” and “6” signified “No.”

Participants were asked if they had discussed the video with anyone else in the tutorial class in which they participated in the experiment, other than when they were instructed to do so. Only one participant indicated that she had discussed the video with someone in the tutorial. When the participant who reported discussing the video was removed from the analyses, the results were not affected; so all participants were included in the analyses reported. Participants were also asked if they had discussed the video with anyone in their psychology course, but from other tutorials. Four participants indicated that they had discussed the video with someone in another tutorial, two from mixed-video groups, and two from same-video groups. When the participants who reported discussing the video with someone from another tutorial were removed from the analyses, the results were not affected, so all participants were included in the analyses reported here. Participants were asked what they thought the purpose of the study was. Only one participant guessed that she might have been shown a slightly different video from the other members of her group. As she stated, “Maybe people saw slightly different versions of the video and you wanted to see if we would dispute the new information brought in by the other members of the group or if we would just work the details into our own account.” Because this participant was from the mixed-video group and she was aware of the manipulation, her data were excluded from the analyses. Post-Debriefing Check for Awareness of Manipulation. After being partially debriefed about the study, participants were asked, “Do you think you were in a discussion group in which other members saw a different version of the video than the one you saw?” A chi-square analysis revealed that there was no association between whether participants thought they had been in a mixed–video group and whether they were in fact in a mixed- video group (χ2 (1, N = 63) = 0.01, p = .95). When debriefed, the majority of witnesses (65%) thought that others in their group had seen a different version of the video. Of the Chapter 8: Source Monitoring 204

participants who were in the same-video groups, 65% incorrectly thought that they were in a mixed-video group. Of the participants who were in the mixed-video groups, 66% correctly thought that they were in a mixed-video group.

Recognition Questionnaire Memory Accuracy. Participants’ responses to the recognition questions were recoded such that a correct answer was scored “1” and an incorrect answer scored “0.” Scores were then combined to give a total accuracy score for the eighteen items that were not subject to misinformation (i.e., non-misled items). A separate score was also calculated for the items for which we had attempted to mislead some participants (i.e., misled items). All results were then rescaled in the range 0-100. As in Study 5, three of the items were very difficult to discern in the final version of the two videos. For this reason, the data for the misled items were analyzed excluding these items. When these three items were removed from the analyses, the results were not greatly affected, however where differences do exist, they are acknowledged in footnotes. Preliminary analyses were conducted in order to assess whether there were any differences between individuals in groups who had all seen Version A of the video and those in groups who had all seen Version B. Independent t-tests revealed that there were no significant differences between the two groups on accuracy for non-misled items, t (32) = 2.02, p = .05, and accuracy for misled items, t (32) = 0.89, p = .38. Therefore the data from the two conditions were combined as “same-video group” for further analysis. A 2 x 2 MANOVA was conducted to investigate the effects of Discussion Type and Warning on participants’ accuracy on misled and non-misled items. Following the procedure recommended by Tabachnick and Fidell (2001), before conducting the MANOVA, the dependent variables were correlated to check for possible multicollinearity. It was found that the dependent variables were not unacceptably highly correlated (r = .30) and it was therefore appropriate to proceed with the MANOVA. The mean scores for memory accuracy are shown in Table 8.2. The MANOVA revealed that there was a significant effect of Discussion Type on the combined dependent variable of memory accuracy, F (2, 57) = 4.81, p < .05; Wilks’ Chapter 8: Source Monitoring 205

Lambda = .9; partial eta squared = .14.19 Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .025, showed that there was a significant main effect of Discussion Type on misled items (i.e., items for which we had attempted to mislead some participants), F (1, 58) = 5.38, p < .025. Participants in the mixed-video groups were less accurate on misled items (M = 55.00, SD = 17.74) than participants in same-video groups (M = 67.06, SD = 22.50). There was no main effect of Discussion Type on non-misled items (i.e., items for which we had not attempted to mislead any participants) F (1, 58) = 1.22, p = .27. The main effect of Warning was not significant on the combined dependent variable of accuracy, F (2, 57) = 0.39, p = .68. There was no interaction between Discussion Type and Warning on the combined dependent variable of accuracy, F (2, 57) = 0.23, p = .80.

Table 8.2 Mean Memory Accuracy for Recognition Questionnaire

Condition Non-misled Items Misled Items M SD M SD No Warning 79.17 10.54 52.73 20.54 Mixed-Video Specific Warning 79.41 5.79 56.47 16.18 Group Total 79.32 7.81 55.00 17.74 No Warning 78.24 8.03 66.67 16.80 Same- Video Specific Warning 75.52 9.61 67.50 28.17 Group Total 76.96 8.78 67.06 22.50

Percentage of Participants Who Reported Misinformation on the Recognition Test. The percentage of participants who had reported at least one item of misinformation on the recognition questionnaire was compared with those who did not report any. It was found that there was no association between Discussion Type and whether or not participants reported misinformation on the recognition test, (χ2 (1, N = 65)

19 When all eight items of the items that differed between Video A and Video B were included in the analysis (i.e., including the items that were difficult to discern in the final videos), there was no effect of Discussion Type on the combined dependent variable of accuracy, F (2, 57) = 2.72, p = .07. Chapter 8: Source Monitoring 206

= 2.87, p = .09). The vast majority of participants inaccurately reported at least one misled item across both conditions. In total, 91% of the participants in the same-video group, and 100% of those in the mixed-video group inaccurately responded to one question for which we had attempted to mislead some participants. Comparisons were also made across the two Warning conditions. It was found that there was no association between Warning and whether or not participants reported misinformation on the recognition test (χ2 (1, N = 65) = 2.70, p = .10). That is, 91% of participants who received a warning and 100% of those who did not receive a warning reported at least one misled item. Memory Confidence. Participants stated their confidence on a six-point scale where a score of “1” indicated “Low confidence” and “6” indicated “High confidence.” Scores were added together to obtain separate confidence scores for misled and non-misled items, and these were converted into mean confidence scores. After establishing that the dependent variables for memory confidence were not unacceptably highly correlated (r = .24) a 2 x 2 MANOVA was conducted to investigate the effects of Discussion Type and Warning on participants’ confidence for misled and non-misled items (see Table 8.3). The analysis revealed that there was no main effect of Discussion Type on the combined dependent variable of confidence, F (2, 54) = 0.69, p = .51. There was also no main effect of warning on the combined dependent variable of confidence, F (2, 54) = 0.34, p = .71. Furthermore, there was no significant interaction between Discussion Type and Warning on the combined dependent variable of confidence, F (2, 54) = 0.10, p = .90.

Table 8.3

Mean Memory Confidence for Recognition Questionnaire

Condition Non-misled Items Misled Items M SD M SD No Warning 3.44 0.42 4.07 1.14 Mixed-Video Specific Warning 3.43 0.33 4.16 0.90 Group Total 3.43 0.36 4.13 0.98 Same- Video No Warning 3.34 0.38 4.03 0.76 Group Specific Warning 3.31 0.43 4.36 1.50 Chapter 8: Source Monitoring 207

Total 3.33 0.39 4.18 1.14

Confidence in Correct and Incorrect Decisions. Confidence was also calculated so that it reflected confidence in correct responses separately from incorrect responses. Therefore, in this analysis there were four dependent variables (confidence for correct items when given misleading information, confidence for incorrect items when given misleading information, confidence for correct items when given no postevent information, confidence for incorrect items when given no postevent information). After noting that the dependent variables were not unacceptably highly correlated (r’s ranged from 0.21 to 0.57), a 2 x 2 MANOVA was conducted to investigate the effects of Discussion Type and Warning on the four confidence in correct and incorrect judgments. The mean scores for confidence in correct and incorrect decisions are shown in Tables 8.4 and 8.5. The analysis revealed that there was a main effect of Discussion Type on the combined dependent variable of confidence in correct and incorrect judgments, F (4, 49) = 3.89, p < .01; Wilks’ Lambda = .8; partial eta squared = .24.20 Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .0125, showed that there was a significant main effect of Discussion Type on confidence for incorrect items when given misleading information, F (1, 52) = 10.19, p < .0125. Participants in the mixed-video groups had lower confidence on incorrect items for which we had tried to mislead some participants (M = 1.90, SD = 1.11) than participants in same-video groups(M = 3.04, SD = 1.57). There were no main effects of Discussion Type on the other dependent variables. There was also a main effect of Warning on the combined dependent variable of confidence in correct and incorrect judgments, F (4, 49) = 3.09, p < .05; Wilks’ Lambda = .8; partial eta squared = .20.21 Analysis of each individual dependent variable, using a

20 When all eight items were included in the analysis (i.e., including the items that were difficult to discern in the final videos), there was no effect of Discussion Type on the combined dependent variable of confidence in correct and incorrect judgments, F (4, 51) = 0.58, p = .68. 21 When all eight items were included in the analysis (i.e., including the items that were difficult to discern in the final videos), there was no effect of Warning on the combined dependent variable of confidence in correct and incorrect judgments, F (4, 51) = 2.19, p = .08.

Chapter 8: Source Monitoring 208

Bonferroni adjusted alpha level of .0125, showed that there was a significant main effect of Warning on confidence for correct items when given no postevent information, F (1, 52) = 12.17, p < .0125.Participants who received a warning had higher confidence on correct items for which they were given no postevent information (M = 5.02, SD = 0.50) than participants who received no warning (M = 4.61, SD = 0.48). There were no main effects of Warning on the other dependent variables. There was no significant interaction between Discussion Type and Warning on the combined dependent variable of confidence in correct and incorrect judgments, F (4, 49) = 0.42, p = .80.

Table 8.4

Mean Confidence in Correct Decisions

Condition Non-misled Items Misled Items n M SD M SD Mixed- No Warning 10 4.48 0.53 4.21 1.48 Video Specific Warning 18 4.93 0.46 4.41 0.98 Group Total 28 4.77 0.52 4.34 1.16 Same- No Warning 18 4.67 0.45 3.92 1.04 Video Specific Warning 10 5.18 0.56 4.86 1.47 Group Total 28 4.85 0.54 4.25 1.27

Table 8.5

Mean Confidence in Incorrect Decisions

Condition Non-misled Items Misled Items n M SD M SD Mixed- No Warning 10 2.62 1.16 1.65 0.91 Video Specific Warning 18 2.77 1.00 2.05 1.20 Group Total 28 2.72 1.04 1.90 1.11 Same- No Warning 18 3.07 1.00 2.96 1.40 Video Specific Warning 10 3.38 1.07 3.19 1.92 Group Total 28 3.18 1.02 3.04 1.57

Chapter 8: Source Monitoring 209

Free Recall Coding. Participants’ written narratives of the event were coded for accuracy by three independent judges using a standard coding system. This coding system was loosely based on one used by Yarmey and Morris (1998). Each narrative was scored in terms of: (a) number of accurate propositions (regarding sequence of events, character descriptions, dialog, setting, and items stolen); (b) conciseness (number of accurate propositions divided by number of words written); (c) fabrications (spontaneous errors); and (d) amount of misleading postevent information reported. Three independent judges repeatedly viewed the crime video and independently coded the free recall. The judges were unaware of the conditions to which the participants were assigned. Each participant’s responses were randomly assigned to one of the three coders. Eleven of the free recall narratives were scored by all coders in order to assess inter-rater reliability. There was a significant correlation between the three coders on all items (see Table 8.6 for correlation coefficients). According to Elmes, Kantowitz & Roediger (1991), high inter-rater reliability occurs when the correlation coefficient is greater than +0.70. All of the correlations between the scorers met this criterion.

Table 8.6 Inter-rater Reliability for Free Recall Coding

Correlations Correlations Correlations between between between Scorers 1 & 2 Scorers 2 & 3 Scorers 1 & 3 Accurate 0.95** 0.96** 0.96** propositions Fabrications 0.86** 0.70* 0.76** Misleading 0.77** 0.89** 1.00** postevent information Note. *n = 11, p < 0.05, two-tailed **n = 11, p < 0.01, two-tailed

Chapter 8: Source Monitoring 210

Memory Accuracy. After establishing that the dependent variables for Free Recall Accuracy were not unacceptably highly correlated (r’s ranged from 0.09 to 0.38) a 2 x 2 MANOVA was conducted in which the effects of Discussion Type and Warning on free recall accuracy (accurate propositions, conciseness, fabrications, misleading postevent information reported) were analysed. The mean scores for memory accuracy are shown in Table 8.7. The analysis revealed that there was no main effect of Discussion Type on the combined dependent variable of free recall accuracy, F (4, 56) = 2.16, p = .09. There was also no main effect of Warning on the combined dependent variable of accuracy, F (4, 56) = 1.00, p = .42. Furthermore, there was no significant interaction between Discussion Type and Warning on the combined dependent variable of accuracy, F (4, 56) = 0.11, p = .98. Because the effect of Discussion Type was marginally non-significant (p = .09) and because it was hypothesised that there would be a significant effect of Discussion Type on misleading postevent information reported, this effect was analysed further. A univariate ANOVA investigating the effects of Discussion Type and Warning on misleading postevent information reported revealed a significant effect of Discussion Type, F (1, 59) = 6.55, p < .05. Participants in the mixed-video group were more likely to report the misleading information (M = 0.40, SD = 0.67) than those in the same video group (M = 0.06, SD = 0.35). There was no effect of warning on misleading postevent information reported, F (1, 59) = 0.55, p = .46, and no interaction between Discussion Type and Warning, F (1, 59) = 0.01, p = .91.

Table 8.7

Mean Memory Accuracy for Free Recall

Misleading Accurate Condition Conciseness Postevent Fabrication Propositions Information M SD M SD M SD M SD Mixed- No 37.45 7.35 0.13 0.03 0.45 0.82 4.64 3.04 Video Warning Group Specific 40.58 12.69 0.15 0.04 0.37 0.60 4.53 2.32 Warning Chapter 8: Source Monitoring 211

Total 39.43 11.00 0.14 0.03 0.40 0.67 4.57 2.56 No 36.18 12.51 0.14 0.03 0.12 0.49 3.65 2.74 Warning Same- Specific Video 36.06 7.71 0.15 0.03 0.00 0.00 3.38 2.99 Warning Group Total 36.12 10.30 0.15 0.03 0.06 0.35 3.52 2.82 Percentage of Participants Who Reported Misinformation during Free Recall. The accuracy scores were recoded to indicate which participants had, during free recall, recalled at least one item of misinformation. It was found that 29% of participants in the mixed-video group reported at least one misled item, while only 6% in the same-video group inaccurately reported at least one misled item. A chi-square analysis revealed that there was a significant association between Discussion Type and whether or not participants reported at least one misled item, (χ2 (1, N = 65) = 6.18, p < .05). A similar comparison was also made across the two Warning conditions. It was found that 17% of participants who received a warning and 17% of those who did not receive a warning reported at least one misled item. A chi-square analysis revealed that there was no association between whether or not the participants reported misinformation and whether or not they received a warning (χ2 (1, N = 65) = 0.003, p = .96).

Source Monitoring Judgments Only data from participants in the mixed-video group were involved in the analysis to assess whether source monitoring judgments differ for those participants whom we had attempted to mislead and those whom we hadn’t. The analysis included items for which participants who had seen Version B of the video had potentially been misled by those in the group who had seen Version A. The number of times each participant reported they remembered the item from the “video only,” “discussion only,” or “both the video and the discussion’ were calculated for both misled items and non-misled items. After establishing that the dependent variables for source monitoring were not unacceptably highly correlated (r’s ranged from -0.75 to 0.36) a 2 x 2 MANOVA was conducted in which the effects of Video Version and Warning on source-monitoring Chapter 8: Source Monitoring 212

(misled item video only, misled item discussion only, misled item both, non-misled video only, non-misled discussion only, non-misled both) were analysed. The analysis revealed that there was no main effect of Video Version on the combined dependent variable of source monitoring, F (6, 22) = 3.38, p = .06. There was also no main effect of Warning on the combined dependent variable of source monitoring, F (6, 22) = 1.55, p = .21. Furthermore, there was no significant interaction between Video Version and Warning on the combined dependent variable of source monitoring, F (6, 22) = 1.98, p = .11. Given that the above analysis was only marginally non-significant, a second analysis was conducted that omitted the factor, Warning. A one-way MANOVA revealed that there was a significant main effect of Video Version on the combined dependent variable of source monitoring, F (6, 24) = 2.98, p < .05; Wilks’ Lambda = .6; partial eta squared = .43. Analysis of each individual dependent variable, using a Bonferroni adjusted alpha level of .008, showed that there was a significant main effect of Video Version on misled both (i.e., participants responded they remembered items for which we had attempted to mislead some participants from both the video and the discussion), F (1, 29) = 9.75, p < .008. Participants who saw Version A (i.e., those who actually saw the items in the video and were therefore not misled) were more likely to say they remembered misled items from both the discussion and the video (M = 1.80, SD = 1.15) than people who saw Version B (i.e., those who did not see the items in the video and were misled; M = 0.75, SD = 0.68). There were no differences for the other dependent variables (see Table 8.8).

Table 8.8

Mean Scores for Source Monitoring Judgments

Sequence A Sequence B Total Participant response M SD M SD M SD Video only 1.07 1.10 1.44 0.81 1.26 0.96 Misled Discussion only 0.33 0.62 0.56 0.63 0.45 0.62 Items Both Video and 1.80 1.15 0.75 0.68 1.26 1.06 Discussion Chapter 8: Source Monitoring 213

Video only 5.93 2.58 7.56 3.03 6.77 2.89 Non- Misled Discussion only 0.47 0.74 1.06 1.00 0.77 0.92 Items Both Video and 5.53 1.81 4.31 2.39 4.90 2.18 Discussion

Discussion In agreement with previous studies reported in this thesis, the results from Study 6 suggest that witnesses often report misleading information that is presented to them by a co-witness. In this study we have shown that they do this despite explicit warnings against doing so, and despite instructions to monitor the source of their information. The first aim of our research was to determine whether the experimental manipulation was successful. In order to determine whether participants in the mixed-video group had similar experiences to those in the same-video groups, participants were asked four questions relating to their discussion experiences. There were no differences found between the two groups on any of the questions, which suggests that participants in mixed- video groups were unaware of the manipulation. Participants were also asked to state what they believed to be the purpose of the study. Only one participant guessed that she may have been shown a slightly different video to other members in her group, so her data were excluded from the analyses. Because the vast majority of participants did not guess the purpose of the study, the experimental manipulation was deemed successful. After being partially debriefed about the study, participants were asked whether they believed they were in a discussion group in which the other members saw a different version of the video than the one they saw. No association was found between whether participants thought they had been in a mixed–video group and whether they were in fact in a mixed-video group. The majority of witnesses thought that others in their group had seen a different version of the video, regardless of whether they were in a same-video or mixed- video group. The fact that participants were unable to tell whether they were in a same- video or mixed-video group, even after being partially debriefed, also suggests that the experimental manipulation was successful. Chapter 8: Source Monitoring 214

Our second research question investigated how Discussion Type (mixed-video group, same-video group) and Warning (specific or no warning) influence participants’ accuracy and confidence for misled and non-misled items. In accordance with our hypotheses, participants in mixed-video groups were more likely to report misled items than those in same-video groups. That is, participants who discussed the video with people who had seen an alternate version of the video, erroneously incorporated information from the alternate version into their own individual account. This was true both for memory assessed using a true/false recognition questionnaire and for free recall accounts. The finding that participants report errant co-witness information when later tested individually supports findings from the present thesis (Study 5) as well as previous studies employing the Stimulus Induced Information methodology (e.g., Gabbert, Memon, & Allan, 2003; Kanematsu, Mori, & Mori, 2003; Mori & Mori, 2004; Wright, Self, & Justice, 2000). Whilst participants in mixed-video groups were less accurate for misled items than those in the same-video groups, there were no differences between the two groups for non- misled items. This finding contrasts previous findings in this thesis (e.g., Studies 3 & 5), which have shown that participants who receive Experimentally Induced Information are more accurate on non-misled items than those who do not received it. However, although the results were not significant, the means were in the same direction as those in the previous studies. That is, participants in the mixed-video groups were slightly more accurate on non-misled items on the recognition questionnaire than those in the same-video groups. Similarly, on the free recall questionnaire participants in the mixed-video groups reported slightly (but not significantly) more accurate propositions than those in the same- video groups. In addition to memory accuracy, this study also investigated how the type of discussion in which participants took part affected memory confidence. It was found that participants’ confidence in their memories was generally unaffected by the type of discussion they experienced. However, this was only true for confidence judgments irrespective of accuracy. There was a significant effect of Discussion Type on confidence for incorrect items when participants were given misleading information. Participants in the mixed-video groups had lower confidence on incorrect items regarding aspects that differed between Versions A and B of the video than participants in same-video groups. This means Chapter 8: Source Monitoring 215

that, whilst participants who had been misled reported the misleading information, they weren’t as confident in their responses as those who had not been misled. This study also assessed the effect that warning participants about possible misinformation has on memory accuracy and confidence. It was found that warning participants about possible misinformation had no effect on memory accuracy measured using either free recall or true/false recognition questionnaires. That is, even when people are warned that the co-witness information they received may be incorrect, they still report that they saw it. This finding supports those reported in Study 5. For both studies it was found that there wasn’t a significant decrease in misinformation reported, however there was a slight trend in that direction. That is, participants in the mixed-video groups who received a warning were slightly less likely to report misinformation on the recognition and free recall questionnaires than those who did not receive warnings. Previous research on co- witness discussion (Meade & Roediger, 2002) as well as other forms of misinformation (e.g., Betz et al., 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, in press; McDermott & Roediger, 1998; Wright, 1993) has found that the misinformation effect can be decreased, but not eliminated using warnings. Whilst in the present study it was also found that the misinformation effect could not be eliminated using warnings, it is possible that warnings in the present study and in Study 5 did not significantly decrease the misinformation effect because there was a relatively long delay of one week before the warning was given. This long delay may have made it more difficult for participants to disregard the co-witness information. As suggested in Study 5, the fact that participants report co-witness information despite warnings against doing so may support the notion that memory change is responsible for the conformity observed. However, like Study 5, even though the warnings did not eliminate the effects of co-witness information, they did help reduce its effects (although not significantly). This suggests that to a small extent, informational influence may also account for some memory conformity. This study also investigated the effects of warnings about misinformation on confidence judgments. In general, it was found that there were no effects of warning on confidence judgments. That is, participants who were warned about possible Chapter 8: Source Monitoring 216

misinformation were no more or less confident in their decisions than those who were not warned. A final research question addressed by the current study was whether the source monitoring judgments of participants whom we had attempted to mislead differed from those whom we had not. It was found that, for the most part, these groups did not differ in their source monitoring judgments. There was some indication that participants who actually saw the items in the video (and were therefore not misled) were more likely to say they remembered the items from both the discussion and the video than people who did not see the items in the video (and were therefore potentially misled). However, there was a non-significant finding, which suggested that misled participants who did not see the items in the video were slightly more likely to indicate that they remembered the item from the “video only” than those who actually saw the items in the video. This very odd finding suggests that participants may have had difficulty monitoring the source of their information, despite explicit instructions to do so. Meade and Roediger (2002) also found that memory conformity persists despite the use of recognition/source monitoring tests in which participants were given choices among possible sources of information. The fact that memory conformity occurs even when participants are asked to monitor the source of the memory is particularly notable because prior research has shown that source monitoring can eliminate the misinformation effect when postevent information is encountered in other ways (Lindsay & Johnson, 1989). Thus it appears that when postevent information is encountered through postevent narratives and leading questions, the misinformation effect is eliminated if participants are asked to complete source monitoring tests, but not when it is encountered through co-witness discussion. Furthermore, Meade & Roediger (Experiment 4, 2002) found that participants were more likely to misattribute the source of misinformation to the original stimulus when it was encountered through co-witness discussion than when it was encountered through the indirect transfer of co-witness information. This issue will be explored in more detail in the General Discussion. In this study the final recognition/source monitoring test was preceded by a free recall test for which participants were not required to monitor their sources. This prior recall may have induced the source confusion (see Roediger, Jacoby, & McDermott, 1996). Chapter 8: Source Monitoring 217

However, Meade & Roediger (Experiment 2, 2002) also found that participants had trouble identifying information sources following co-witness discussion when the recognition test was not confounded by prior recall. In sum, the findings from this experiment suggest that memory conformity may result from memory change. This claim is supported in that: a) participants were tested individually and therefore it is unlikely that normative social influence caused the conformity; b) participants were encouraged to report what they remembered from the original stimulus and were warned about possible misinformation; and c) memory conformity occurred on a source monitoring test that explicitly called participants’ attention to the various possible sources of information. These findings are consistent with those obtained by other researchers (Meade & Roediger, 2002) and suggest that the effects of memory conformity must be particularly strong in order to persist even under these strict testing conditions (Meade & Roediger).

218

Chapter 9 Study 7: Eyewitness Interview

The previous studies in this thesis have shown that misinformation presented by co- witnesses can have a detrimental effect on eyewitness memory for the event. The objective of Study 7 was to determine whether these detrimental effects of co-witness misinformation can be reversed during an interview. Specifically, this study examines whether participants still report co-witness misinformation during the ecologically valid context of an interview. The eyewitness interview in this study was structured to reflect a good police interview. The interviewer had observed several real police interviews and tried to create an interview environment similar to that of the police to increase ecological validity. In Australia, officers are trained to use some aspects of the cognitive interview (Fisher & Geiselman, 1992). These aspects of the cognitive interview were incorporated into the study. In particular, the interviewer attempted to build rapport with the witnesses, listen effectively, transfer control to the witnesses, allow for pauses, ask only open-ended questions, and encourage the witnesses to report as much information as they could. The cognitive interview was developed to enhance the accuracy of witnesses as it ensures that poor interview practice does not limit their recall. Furthermore, the free recall format of the interview reduces the demand characteristics and subsequent response bias, which can be brought about by the use of forced-choice recognition tests of the sort used in previous studies of the misinformation effect (McCloskey & Zaragoza, 1985). There is reason to believe that participants may be less likely to report co-witness information within the context of a cognitive interview than a standard interview. Newlands and Skaltsis (2004) compared participants’ memories for a videotaped crime scene using either a cognitive interview or a structured interview. It was found that participants in the cognitive interview group recalled more correct information and were significantly less likely to report misinformation presented to them than participants interviewed with a structured interview. As the current study only uses some components of the cognitive interview, it is unclear whether it will have the same advantages. Chapter 9: Eyewitness Interview 219

It is also interesting to investigate the effects of co-witness discussion within an interview context because research suggests that there may be differences between written and spoken recall of an event. In one study, participants were shown a series of slides depicting an accident and they were later asked to freely recall everything they could remember about the event (Berkerian & Dennett, 1990). Half of the participants were asked to speak their free recall accounts and the other half were asked to write them down. It was found that spoken recall was more accurate than written recall. Therefore, if spoken reports are more accurate, it may be the case that participants are less likely to report erroneous co- witness information in the context of a verbal interview than a written narrative, as used in the previous studies described in this thesis, and most other studies investigating the effects of co-witness discussion on memory. As in Studies 5 and 6, participants in the current experiment were warned about possible misinformation in order to investigate the mechanisms responsible for memory conformity. In Studies 5 and 6, warning participants that they may have received incorrect information from their co-witness tended to reduce the negative effects of co-witness misinformation, although not to a significant effect. Study 7 aims to investigate this finding within an interview context. Because warnings about hearsay evidence are typically used (and apparently assumed to work) in the legal setting, the finding that warnings don’t significantly protect against co-witness contamination warrants further investigation. Before commenting on the success of the current legal process, it is important to investigate the effects of warnings within a forensically relevant context. It is, for example, possible that warnings may have a greater effect within the interview context due to increased emphasis of verbal instructions given one-on-one. Participants in Study 7 were given the same warnings as those in Study 6. That is, participants were either warned prior to the interview that members in their group may have seen a slightly different video or they did not receive any warning at all. It was hypothesised that memory conformity would be reduced, but would not be eliminated by the warning, as suggested by the previous studies in this thesis as well as other research on co-witness discussion (Meade & Roediger, 2002) and other modes of encountering misinformation (e.g., Betz et al., 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, in press; McDermott & Roediger, 1998; Wright, 1993). Chapter 9: Eyewitness Interview 220

The current study was also designed to investigate whether it is possible to distinguish ‘real’ memories from information obtained second-hand. In Study 6, it was found that source monitoring tests (Lindsay & Johnson, 1989) were not effective at distinguishing between real memories and co-witness information. Therefore, this methodology was not used in Study 7. However, in Study 5 there was some evidence to suggest that using remember/know judgements (Tulving, 1985) may be a good way to distinguish between ‘real’ memories and co-witness information. In Study 5, participants were more likely to indicate that misled items were ‘known’ rather than ‘remembered.’ The current study aims to explore this finding within the interview context. In order to do so, the procedure had to been changed slightly. The procedure was adapted in such a way that the remember/know judgments could be applied to complete statements rather than true/false questions or word lists as has been done previously (e.g., Roediger, Meade, & Bergman, 2001). Also, it was important that the procedure be quick and easy for police officers to administer within an interview. In the current study, participants were asked to read through their statements and indicate the details that they consciously ‘remembered’ to be true using a highlighter in one colour and indicate those that they ‘knew’ to be true using a different colour highlighter. Because participants were asked to give remember/know judgments, context reinstatement was not used in the interview. Although context reinstatement is one of the most valuable techniques of the cognitive interview, research has shown that when participants engage in context reinstatement, this can negate the general finding that “know” responses predominate for misinformation (Berkerian & Dennett, 1994, cited in Memon & Stevenage, 1996), and therefore, using remember/know judgments would not be a helpful way of distinguishing between true memory for an event and co-witness information. Berkerian and Dennett found that when participants were interviewed using a cognitive interview, they gave more ‘remember” judgments for misinformation than would be expected by chance. Because the cognitive interview encourages context reinstatement, Berkerian and Dennett suggest that misinformation recalled in a cognitive interview is erroneously judged as being “remembered” as it is accompanied by contextual detail. A secondary reason for not using context reinstatement was that real police officers were not observed to use the mnemonic. Chapter 9: Eyewitness Interview 221

The current study also aims to improve on the remember/know methodology used in Study 5. A limitation of the remember/know methodology used in Study 5 was that participants were asked to indicate either “remember” or “know” to items they indicated were true, when in fact, the participants may have simply guessed the answer on the true/false questionnaire. Therefore, the “know” responses may have been confounded when participants were merely guessing (Gardiner, Richardson-Klavehn, & Ramponi, 1997). The current study is an improvement on Study 5 because participants give a free recall account of what happened, so the likelihood that they might guess an answer is decreased relative to a situation in which they respond to a forced choice questionnaire. In light of previous research in this thesis as well as other research on co-witness discussion (Roediger, Meade, & Bergman, 2001), it was hypothesized that participants would be more likely to claim that they ‘know’ the misled items than to report that they specifically ‘remembered’ seeing them. As in Study 6, in the current study we compared groups who received Stimulus Induced Information (mixed-video groups) with Natural Discussion Groups (same-video groups). However, there are a few differences between the two studies. Unlike Study 6, the participants in the current study discussed the stimulus with only one other person, instead of three. This change was made to address the difficulties experienced in organising for four participants to show up for the experiment at the same time. The effects of co-witness memory conformity may be diminished when there are only two participants in the group because there are no other participants to support the witness supplying misinformation. Conformity increases with the number of independent sources of the information (Wilder, 1977). Furthermore, if one participant doesn’t talk much, there may be a very limited exchange of information. A further difference between the previous research and Study 7 is the creation of new eyewitness stimuli. This was done for three reasons. First, it was important to determine that the findings could be generalised, and were not just the product of one set of stimuli. Second, when the stimuli used in Studies 5 and 6 were compressed and digitized for viewing on computer monitors, some of the differences between the two videos became very unclear. Some of the differences were so unclear that people viewing the two versions of the video consecutively were unable to spot the differences between them. Third, new Chapter 9: Eyewitness Interview 222

stimuli were created because the stimuli used in the previous studies did not allow for a bidirectional flow of misinformation. That is, in the old stimuli Version A of the video contained items that were not in Version B, but Version B did not contain items that were not in Version A (once the items that were difficult to discern in the final videos were eliminated). This meant that participants who had seen Version A were equipped with more information than the those who saw Version B and therefore they may have appeared more knowledgeable and credible. This may have had some impact on the size of the memory conformity effect. In the current study, both videos contained items that were exclusive to that version and therefore the participants would be equally likely to influence one another. Furthermore, the new stimuli also contained items that were conflicting between the two versions. For example, in Version A, the thief refers to himself as “Joe” while in Version B he refers to himself as “James.” The effects of memory conformity are weakened when participants are told that something they saw didn’t occur (Wright & Mathews, 2004) and therefore the effects of memory conformity may be decreased when the information conflicts with something they saw. Despite differences between the current study and previous research, it was hypothesized that the findings would be consistent with those from Studies 5 and 6. That is, it was hypothesized that groups with Stimulus Induced Information would be less accurate on the misled items than Natural Discussion Groups. No hypotheses were made regarding accuracy on non-misled items because results from past experiments in this thesis have been inconsistent. In Study 5 it was found that groups with Stimulus Induced Information were more accurate on the non-misled items, however in Study 6 this difference was not significant.

Method

Design There were three phases in this study: the exposure phase, the discussion phase, and the recall phase. During the exposure phase, participants saw one of two slightly different versions of a crime video (Version A or B). In the discussion phase, participants were Chapter 9: Eyewitness Interview 223

assigned to one of two conditions: (1) discuss the crime in a group of two in which both members had seen the same video (both A or both B) or (2) discuss the crime in a group where the two members had seen different versions of the video (one member had seen Version A and one, Version B). After the discussion phase, participants were given one of two warnings: (1) specific warning, or (2) no warning. During the recall phase, participants were individually interviewed about the video. In addition, participants were also asked to make remember/know judgments (Tulving, 1985) for each item in their statement. Therefore, this study employed a 2x2 experimental design in which Discussion Type (same-video group, mixed-video group) and Warning (no warning, specific warning) were manipulated. The dependent variables were amount and accuracy of information recalled during the interview regarding misled and non-misled items, and remember/know judgments.

Participants Participants were 64 undergraduate psychology students (40 female, 24 male) from the University of New South Wales. The average age of the participants was 19.20 (SD = 1.88) years. Participants received credit for their psychology course upon completion of the study.

Materials Eyewitness Stimuli. The eyewitness stimulus used in this study was a short video (just under three minutes in length) depicting a robbery. The video was created for the purpose of this study. In the video, a male of Middle Eastern appearance (the thief) talks to a building resident over an intercom and convinces the resident that he needs to fix some problems in the building. When the resident lets him into the building, the thief begins listening at the doors of the units. The scene then cuts to a man and a woman preparing dinner. They have a discussion about what night the garbage has to go out. The woman decides to take out the garbage and as she is exiting the unit, she bumps into the thief in the stairwell. He says he is looking for his friend and she says she thinks the friend lives one level up. The woman then leaves to take out the garbage and the thief walks into her unit. Once inside the unit, he proceeds to steal various items from the living room and bedroom. Chapter 9: Eyewitness Interview 224

As he is on the way out of the unit, the man in the kitchen sees the thief and chases him. The thief runs out of the building, jumps on a motorcycle and drives away. There were two different versions of the video that varied slightly. For example, in Video A the thief refers to himself as “Joe” while in Video B he refers to himself as “James.” Table 9.1 summarises all of the differences between the videos. The videos were shown on computer monitors and were viewed by participants individually. Each participant viewed only one of the two videos.

Table 9.1. Summary of Differences between Video A and Video B

Video A Video B Conflicting Thief calls himself Joe. Thief calls himself James. information Thief wearing grey shirt. Thief wearing striped shirt. between Garbage collected on Thursdays. Garbage collected on Tuesdays. videos Door left unlocked. Thief keeps door open. Item in Thief steals mobile phone. Thief doesn’t steal phone. Video A Man shouts, “Bastard!” at thief. Man doesn’t say anything. only View of motorcycle license plate. No view of license plate. Item in No view of intercom number View of intercom number Video B pressed. pressed only Thief doesn’t say why he was sent Thief says he was sent to fix to the building. water problems. Woman is not wearing a jacket Woman is wearing a jacket.

General Knowledge Test. A general knowledge test was included in the study as a filler task. The test contained 100 questions. Sample items include: What is 54 divided by 9; When was the Sydney Opera House opened; What is the largest human organ; How many days are there in January? Participants were given ten minutes to complete as many questions as they could. They were encouraged to guess if they did not know the answer to a question. Discussion Question Manipulation Check. Participants were asked three questions about their experiences in the groups. They were asked, (1)“Did the discussion help you remember things from the video that you would otherwise have forgotten?” (2) Chapter 9: Eyewitness Interview 225

“Did you learn anything new about the video as a result of the discussion?” (3) “Did your group have an open and free exchange of ideas?” Participants were asked to indicate their response to each question using a six-point scale (1=Yes; 6=No). Participants were also asked to state what they believed to be the purpose of the study. Post-Debriefing Awareness of Manipulation. After being partially debriefed about the study, participants were given a sheet that asked, “Do you think you were in a discussion group in which the other member saw a different version of the video than the one you saw?” If they agreed with this statement, participants were asked to write why they believed the other member saw a different version of the video.

Procedure This experiment took place over two sessions, one week apart. Like Studies 5 and 6, participants were shown a video of a robbery on individual computer terminals with headphones. Students were instructed that they would be viewing the same crime scene on the computers to ensure that they each have “equal and optimal” viewing conditions. However, unbeknown to the participants, there were two slightly different versions of the crime scene, and each individual witnessed only one version. Participants were given the same instructions and cover story as used in Study 5. After viewing the video, participants completed the General Knowledge Test as a filler task for ten minutes. While completing the task, the experimenter assigned the participants into groups of two to discuss the crime scene. Half of the groups contained students who had both seen the same version of the video (either both had viewed Version A or both had viewed Version B), and the other half of the groups contained members who had unknowingly viewed different videos. When participants met in their groups, they were given ten minutes to discuss the video using a guided recall procedure (see Study 5 for details). They were asked to discuss the sequence of events, what was said, what the characters looked like, the setting, and any other information they could remember. These conversations were tape-recorded. Following the discussion, participants were asked not to talk about the event any further. An individual appointment was made with each participant for the following week. Chapter 9: Eyewitness Interview 226

After a one-week delay, participants returned to a quiet office to be individually questioned about the crime video. The interviewer was not the same experimenter whom the participants had met in the first session, and the interviewer was blind to the condition to which the participants were allocated in the first session. First, participants were asked to complete the Discussion Questions and state what they believed to be the purpose of the experiment. After the manipulation check, the interviewer began questioning the participant. The interviewer had observed several real police interviews and tried to create an interview environment similar to that observed. That is, the witness was seated in a comfortable chair, and the interviewer sat at a computer. The interviewer typed the responses as the participant gave them, using a real police statement form as a template (see Appendix N). The interview format was designed to reflect a good police interview. Participants were put at ease during the rapport building stage. They were seated in a comfortable chair and offered tea or coffee. When the participant appeared comfortable, the experimenter began gathering some background information on the participant (i.e., name, age, sex). When the background information had been gathered, the formal interview began. Participants were instructed, “I am going to ask you some open-ended questions about the crime video you witnessed last week. I will ask you to tell me about the sequence of events, what was said, the setting, what the people looked like, and any other information you can remember. Please keep in mind that I have never seen the video, so try to be as detailed as possible in your description so I can get a good idea of what happened. Please take as much time as you need to answer the questions and try to report everything you can remember, no matter how trivial it may seem.” The participants were then randomly assigned to one of two warning conditions. Half of the participants were given a ‘specific warning’ that the other student in their group may have seen a slightly different video. These participants were told, “In your discussion groups last week, the other member of your group may have been shown a slightly different version of the video. Therefore, when answering these questions you should make a specific effort to disregard what they told you and answer the questions solely on the basis of what you remember from the video.” The other half of the participants did not receive any warning at all. Chapter 9: Eyewitness Interview 227

Next, within the interview context, participants were then asked the following questions: 1. What was the sequence of events from beginning to end? Describe this in detail. 2. What did the characters look like and what were they wearing? 3. What dialog was exchanged between the characters? 4. What did the setting look like? (e.g., layout, furniture, props, etc.) 5. What was stolen from the flat? 6. Can you remember any other details? If required, additional questions were asked for clarification. Once all questions had been answered, the interviewer instructed the participants to go over the crime in their minds and think about anything else they might like to add to the statement. While they were doing this, the interviewer made corrections to the statement and fixed up any typing errors. When the interviewer had finished making corrections and adding anything that the participant had later recalled, the statement was printed. The participant was instructed to read through the statement and make any changes they would like. Once the participant was happy with the statement, the participant was asked to sign it to verify that it was accurate and complete. Next, participants were asked to read through their statements again and indicate whether they remembered each item from the video or just knew on some other basis that the item was in the video. Participants were given the same definitions of “remember” and “know” as used in Study 5. If the participants “remembered” an item in their statement, they were asked to highlight it using a pink highlighter. Conversely, if they “knew” an item they were asked to highlight it using a yellow highlighter. Once participants had finished highlighting their statements, they were partially debriefed about the study. They were told that they might have been shown a slightly different version of the video than the other members in their discussion group. After the partial debriefing, participants were given the Post-Debriefing Awareness of Manipulation questionnaire to complete. Finally, participants were fully debriefed, thanked for their participation, and asked not to discuss the experiment with other potential participants. The entire interview usually lasted between 45 minutes and an hour. Chapter 9: Eyewitness Interview 228

Results Statistical analysis focussed on the following questions: First, was the experimental manipulation successful? Second, how did Discussion Type (mixed-video group, same- video group) and Warning (specific or no warning) influence participants’ memories for misled and non-misled items from the event? Finally, did participants’ remember/know judgments differentiate between misled and non-misled items?

Experimental Manipulation Checks Discussion Question Manipulation Check. Independent t-tests were used to explore responses to three questions relating to the participants’ experiences in their discussion groups. For each question, a score of “1” indicated “Yes” and a score of “6” indicated “No.” Responses to the first question, “Did the discussion help you remember things from the video that you would otherwise have forgotten?” were not significantly different for the two Discussion Types, t (62) = 0.62, p = .54. For the question, “Did you learn anything new about the video as a result of the discussions?” responses differed significantly for Discussion Type, t (62) = 2.23, p < .05. Participants in the mixed- discussion groups were more likely to say that they had learned new information during the group discussion (M = 2.28, SD = 1.65) than those in the same-video groups (M = 3.19, SD = 1.60). Finally, for the question “Did your group have an open and free exchange of ideas?” there were no significant differences for Discussion Type, t (62) = 0.82, p = .42. Table 9.2 shows the average responses given by participants for the three discussion questions. Participants were asked what they thought the purpose of the study was. Only one participant thought that he might have been shown a slightly different video to that seen by the other person in his group. He stated that the purpose of the study might have been to investigate “the possibility of tainting their memory or making them think they saw something similar that they did not actually see.” Because this participant was from the mixed-video group and he appeared to be aware of the manipulation, his data were excluded from the analyses.

Chapter 9: Eyewitness Interview 229

Table 9.2

Response to Discussion Questions

Question Condition M SD n

Mixed-Video Did the discussion help 2.09 1.28 32 you remember things from Group the video that you would Same-Video 2.28 1.14 32 otherwise have forgotten? Group Mixed-Video 2.28 1.65 32 Did you learn anything Group new about the video as a Same-Video result of the discussion? 3.19 1.60 32 Group Mixed-Video 1.78 1.26 32 Did your group have an Group open and free exchange of Same-Video ideas? 1.56 0.84 32 Group Note. Responses were given on a six-point scale in which a score of “1” signified “Yes” and “6” signified “No.”

Post-Debriefing Awareness of Manipulation. After being partially debriefed about the study, participants were asked, “Do you think you were in a discussion group in which the other member saw a different version of the video than the one you saw?” A chi- square analysis revealed that there was no association between whether participants thought they had been in a mixed–video group and whether they were in fact in a mixed-video group (χ2 (1, N = 64) = 3.00, p = .08). The majority of witnesses (75%) did not think that their partner had seen a different version of the video. Of the participants who were in the same-video groups, 16% incorrectly thought that they were in a mixed-video group. Of the participants who were in the mixed-video groups, 34% correctly thought that they were in a mixed-video group.

Free Recall Coding. Participants’ interviews of the event were coded for accuracy using a standard coding system employed by three independent judges. This coding system was Chapter 9: Eyewitness Interview 230

loosely based on one used by Yarmey and Morris (1998). Each narrative was scored in terms of: (a) number of accurate propositions (regarding sequence of events, character descriptions, dialog, setting, and items stolen); (b) conciseness (number of accurate propositions divided by number of words written); (c) fabrications (spontaneous errors); and (d) amount of misleading postevent information reported. Three independent judges repeatedly viewed the crime video and independently coded the free recall. The judges were unaware of the conditions to which the participants were assigned. Each participant’s responses were randomly assigned to one of the three coders. Ten of the free recall narratives were scored by all coders in order to assess inter- rater reliability. There was a significant correlation between the three coders on all items (see Table 9.3 for correlation coefficients). According to Elmes, Kantowitz & Roediger (1991), high inter-rater reliability occurs when the correlation coefficient is greater than +0.70. All of the correlations between the scorers met this criterion.

Table 9.3 Inter-rater Reliability for Free Recall Coding

Correlations Correlations Correlations between between between Scorers 1 & 2 Scorers 2 & 3 Scorers 1 & 3 Accurate 0.92** 0.97** 0.98** propositions Fabrications 0.78** 0.94** 0.73* Misleading 0.77** 0.88** .92** postevent information Note. *n = 10, p < 0.05, two-tailed **n = 10, p < 0.01, two-tailed

Memory Accuracy. Preliminary analyses were conducted in order to assess whether there were any differences between individuals in groups in which both members had seen had seen Version A of the video and those in groups where both members had seen Version B. Independent t-tests revealed that there were no significant differences between the two groups on free recall reporting of accurate propositions, t (30) = 0.50, p = Chapter 9: Eyewitness Interview 231

.62, conciseness, t (30) = 0.09, p = .93, fabrications, t (30) = 0.46, p = .65, and misleading postevent information, t (30) = 1.85, p = .08. Therefore the data from the two conditions were combined into one “same-video” group for all subsequent analysis. A 2 x 2 MANOVA was conducted to investigate the effects of Discussion Type and Warning on participants’ free recall accuracy (accurate propositions, conciseness, fabrications, misleading postevent information reported). Before conducting the MANOVA, the dependent variables were correlated to check for possible multicollinearity. It was found that the dependent variables were not unacceptably highly correlated (r’s ranged from -0.01 to 0.43) and it was therefore appropriate to proceed with the MANOVA (Tabachnick & Fidell, 2001). The mean scores for free recall memory accuracy are shown in Table 9.4. The analysis revealed that there was no main effect of Discussion Type on the combined dependent variable of free recall accuracy, F (4, 56) = 1.67, p = .17. There was also no main effect of Warning on the combined dependent variable of accuracy, F (4, 56) = 0.86, p = .50. Furthermore, there was no interaction between Discussion Type and Warning on the combined dependent variable of accuracy, F (4, 56) = 0.36, p = .84.

Table 9.4

Mean Memory Accuracy for Free Recall

Misleading Fabrication Accurate Condition Conciseness Postevent Propositions Information M SD M SD M SD M SD Warning 71.13 18.10 0.11 0.02 0.50 0.73 5.94 3.09 Mixed- No Video 68.40 20.85 0.10 0.02 0.47 0.52 5.33 3.20 Warning Group Total 69.81 19.20 0.11 0.02 0.48 0.63 5.65 3.10

Warning 74.00 15.56 0.11 0.02 0.19 0.40 6.25 4.86 Same- No Video 67.50 19.16 0.10 0.01 0.19 0.40 7.44 7.88 Warning Group Total 70.75 17.48 0.11 0.02 0.19 0.40 6.84 6.47 Chapter 9: Eyewitness Interview 232

Because it was hypothesised that there would be a significant effect of Discussion Type on misleading postevent information reported, this effect was analysed further. A 2 x 2 ANOVA investigating the effects of Discussion Type and Warning on misleading postevent information revealed a significant effect of Discussion Type, F (1, 59) = 4.89, p < .05; partial eta squared = .08. Participants in the mixed-video group were more likely to report the misleading information (M = 0.48, SD = 0.63) than those in the same video group (M = 0.19, SD = 0.40). There was no effect of warning on misleading postevent information reported, F (1, 59) = 0.02, p = .90, and no interaction between Discussion Type and Warning, F (1, 59) = 0.02, p = .90. Percentage of Participants Who Reported Misinformation on Free Recall. The accuracy scores were also recoded to indicate which participants had reported at least one item of misinformation during free recall. It was found that 44% of participants in the mixed-video group reported at least one piece of misinformation, while only 19% in the same-video group inaccurately reported at least one piece of misinformation. A chi-square analysis revealed that this association between Discussion Type and whether or not participants reported any misinformation was significant (χ2 (1, N = 64) = 4.66, p < .05). A similar comparison was also made across the two Warning conditions. It was found that 28% of participants who received a warning and 34% of those who did not receive a warning reported at least one piece of misinformation. A chi-square analysis revealed that this association was not significant (χ2 (1, N = 64) = 0.29, p = .59). Remember/Know Judgments Only data from those in the mixed-video group were used to assess whether Remember/Know Judgments differ for misleading postevent information and accurate propositions reported. The sample size was reasonably small (n = 14) because only participants who reported some misleading postevent information could be included in the analysis. The proportion of times participants had indicated that they “remembered” misleading postevent information to be true was divided by the total number of misleading postevent information they wrote to calculate the remember misleading information score. The proportion of times participants had indicated that they “remembered” an accurate Chapter 9: Eyewitness Interview 233

proposition to be true was divided by the total number of accurate propositions they wrote to calculate the remember accurate proposition score. A repeated measures ANOVA was then conducted which compared the remember misleading information score and the remember accurate proposition score. The ANOVA revealed a significant differences, F (1, 13) = 5.06, p < .05; Wilks’ Lambda = .7; partial eta squared = .28). Those in the mixed-video group indicated a higher proportion of “remember” judgments for accurate propositions (M = 85.10, SD = 15.78) than misleading postevent information (M = 57.14, SD = 51.36). Conversely, a similar analysis was conducted for “know responses.” The proportion of times participants had indicated that they “knew” misleading postevent information to be true was divided by the total number of misleading postevent information they wrote to calculate the know misleading information score. The proportion of times participants had indicated that they “knew” an accurate proposition to be true was divided by the total number of accurate propositions they wrote to calculate the know accurate proposition score. A repeated measures ANOVA was then conducted which compared the know misleading information score and the know accurate proposition score. The ANOVA revealed a significant differences, F (1, 13) = 5.06, p < .05; Wilks’ Lambda = .7; partial eta squared = .28). Those in the mixed-video group indicated a higher proportion of “know” judgments for misleading postevent information (M = 42.86, SD = 51.36) than for accurate propositions (M = 14.82, SD = 15.80). Data from all participants were used to assess whether participants’ remember/know judgments were different for fabrications and accurate propositions. The proportion of times participants had indicated that they “remembered” fabrications to be true was divided by the total number of fabrications they wrote to calculate the remember fabrication score. The proportion of times participants had indicated that they “remembered” an accurate proposition to be true was divided by the total number of accurate propositions they wrote to calculate the remember accurate proposition score. A repeated measures ANOVA was then conducted which compared the remember fabrication score and the remember accurate proposition score. The ANOVA revealed a significant differences, F (1, 63) = 26.06, p < .0005; Wilks’ Lambda = .7; partial eta Chapter 9: Eyewitness Interview 234

squared = .29. Participants indicated a higher proportion of “remember” judgments for accurate propositions (M = 79.83, SD = 13.91) than fabrications (M = 59.56, SD = 35.44). Conversely, a similar analysis was conducted for “know responses.” The proportion of times participants had indicated that they “knew” fabrications to be true was divided by the total number of fabrications they wrote to calculate the know fabrication score. The proportion of times participants had indicated that they “knew” an accurate proposition to be true was divided by the total number of accurate propositions they wrote to calculate the know accurate proposition score. A repeated measures ANOVA was then conducted which compared the know fabrication score and the know accurate proposition score. The ANOVA revealed a significant differences, F (1, 63) = 29.46, p < .0005; Wilks’ Lambda = .7; partial eta squared = .32). Participants indicated a higher proportion of “know” judgments for fabrications (M = 41.74, SD = 35.84) than accurate propositions (M = 19.99, SD = 13.84). In a separate analysis, the proportion of times that participants were accurate when they indicated they ‘remembered’ an item to be true was calculated (i.e., remember accuracy = the number of times participants indicated ‘remember’ for an accurate decision divided by the total number of times they indicated ‘remember’). Similarly the proportion of times that participants were accurate when they indicated they ‘knew’ an item to be true was calculated (i.e., know accuracy = the number of times participants indicated ‘know’ for an accurate decision divided by the total number of times they indicated ‘know’). A paired samples t-test which compared remember accuracy and know accuracy revealed a significant difference, t (62) = 2.29, p < .0005. Participants were more likely to be accurate if they indicated ‘remember’ (M = 93.62, SD = 5.30) than if they indicated ‘know’ (M = 82.26, SD = 17.80). Inspection of the confidence intervals indicated that participants were 7-16% more likely to be accurate if they indicated ‘remember’ than if then indicated ‘know.’

Discussion Broadly speaking, the results from Study 7 confirm previous findings in this thesis that participants often report errant co-witness information presented to them when later Chapter 9: Eyewitness Interview 235

questioned individually. Furthermore, it expands the findings by showing that the memory conformity persists despite the more stringent testing requirements of an ecologically valid interview. Our first research question asked: was the experimental manipulation successful? Participants were asked three questions relating to their discussion experiences in order to determine whether participants in the mixed-video group had similar experiences to those in the same-video groups. We found no significant differences when participants were asked whether the discussion helped them remember things from the video that they would otherwise have forgotten, and when they were asked if their discussion group had an open and free exchange of ideas. This implies that participants in the two groups had similar discussion experiences. However, when asked if they had learned anything new about the video as a result of the discussion, participants in the mixed-video groups were more likely to say that they had learned new things than the same-video group. This finding is different from previous studies in this thesis. This may be due to the fact that the new videos allowed for bidirectional flow of information, which would mean that everyone in the mixed-video groups could potentially report that they learned something new, instead of just half of the group. Although this finding implies that participants in mixed-video groups and same- video groups had different experiences in their group discussions, for the most part participants were unaware of the manipulation. When participants were asked to state what they believed to be the purpose of the study, only one participant guessed that he might have been shown a slightly different video from other member in his group, so his data were excluded from the analyses. After being partially debriefed about the study, participants were asked if they thought they were in a discussion group in which the other member saw a different version of the video than the one they saw. There was no association between whether participants thought they had been in a mixed–video group and whether they were in fact in a mixed- video group. This suggests that participants were unaware of the manipulation. The majority of witnesses did not think that their partner had seen a different version of the video. In Study 6, the opposite was true: most participants believed that they were in a group in which the other members saw a different version of the video. This apparent discrepancy may be explained because the two studies used different group sizes. In the Chapter 9: Eyewitness Interview 236

current study, participants were in groups of two, so one would expect less variation in responses than studies with more group members (such as Study 6). An alternative explanation for this dicrepancy is that it is due to the use of new videos as stimuli. The new videos allowed for the bidirectional flow of information and the differences between the two videos may have been more subtle, and thus gone undetected. However, this explanation can only be used to explain why the mixed-video groups were more likely to say that they thought they had seen the same video; it does not explain why the same video groups were more likely to say that they thought they had seen the same video Our second research question asked: how did Discussion Type (mixed-video group, same-video group) and Warning (specific or no warning) influence participants’ memories for misled and non-misled items from the event? There were no differences found between participants in mixed-video groups and same-video groups for accurate propositions, conciseness, or fabrications. However, there was a significant difference between mixed- video groups and same-video groups for misleading postevent information, when analyzed separately from the other variables. Participants in the mixed-video group were more likely to report the misleading information than those in the same-video group. These results are consistent with the previous studies in this thesis and with previous research investigating Stimulus Induced Information (e.g., Gabbert, Memon, & Allan, 2003; Kanematsu, Mori, & Mori, 2003; Mori & Mori, 2004; Wright, Self, & Justice, 2000). Furthermore, it was found that almost half of the participants in the mixed-video group (44%) reported at least one misled item, while significantly fewer in the same-video group (19%) inaccurately reported at least one misled item. Even though the finding that misled participants are more likely to report misinformation than non-misled participants is in accordance with previous research, it is still somewhat surprising that the results persist within the interview context. For example, participants were tested using some aspects of the cognitive interview, which was designed to increase eyewitness accuracy (Fisher & Geiselman, 1992).22 There is some evidence that participants are less likely to report co-witness information within the context of a cognitive

22 Despite the fact that the cognitive interview was designed to increase eyewitness accuracy, more recent research suggests that it may increase detail, but not accuracy (e.g., Newlands, George, Towell, Kemp, & Clifford, 1999). Chapter 9: Eyewitness Interview 237

interview than a standard interview (Newlands & Skaltsis, 2004). Furthermore, the free recall format of the interview decreases demand characteristics and responses biases that may be at least partially responsible for the misinformation effect in some cases (McCloskey & Zaragoza, 1985). Previous research on co-witness discussion has shown that the misinformation effect was eliminated when participants gave free written statements as opposed to interrogative questioning (Hollin & Clifford, 1983). Finally, within the interview participants were asked to give a verbal account of what they witnessed and research suggests that spoken recall is more accurate than written recall (Berkerian, & Dennett, 1990). Thus we should expect that the interview format should reduce or eliminate the misinformation effect, but the fact that the effect persists despite such stringent testing requirements is a testament to the strength of memory conformity. This study also assessed the effect of warnings on memory accuracy. Although it has not been investigated before, it was thought that warnings may have an increased effect when emphasised within an interview context. However, this was not the case. There was no effect of warning on accurate propositions, conciseness, fabrications, or misleading postevent information reported. These results suggest that even when people are warned that the co-witness information they received may be incorrect, the misinformation effect is not completely eliminated. These findings are consistent with previous studies on the misinformation effect both within this thesis (Studies 5 & 6) and elsewhere (e.g., Betz et al., 1996; Gallo, Roberts & Seamon, 1997; Gallo, Roediger & McDermott, in press; McDermott & Roediger, 1998; Meade & Roediger, 2002; Wright, 1993). As in previous studies, the fact that participants reported misinformation presented to them by co-witnesses despite explicit warnings against doing so suggests that they may have been unable to distinguish what they actually witnessed from the information they discussed with the co-witness. This would support the notion that a modification in memory is responsible for the conformity observed. Despite the fact that warnings do not decrease co-witness memory conformity, there appears to be some sort of difference between information obtained from a co-witness and memories obtained from first-hand experience. Our final research question addressed this topic by investigating whether remember/know judgments differentiate between misled and non-misled items. It was found that participants in mixed-video groups indicated a higher Chapter 9: Eyewitness Interview 238

proportion of “remember” judgments for accurate propositions than misleading postevent information. Conversely, they indicated a higher proportion of “know” judgments for misleading postevent information than accurate propositions. When the data from all participants were examined, it was found that participants indicated a higher proportion of “remember” judgments for accurate propositions than spontaneous fabrications. In contrast, they indicated a higher proportion of “know” judgments for fabrications than accurate propositions. Participants were 7-16% more likely to be accurate if they indicated ‘remember’ than if then indicated ‘know.’ These findings suggest that ‘remember’ judgments may be more trustworthy that ‘know’ judgments. These results are compatible with previous findings (e.g., Meade & Roediger, 2002; Roediger et al., 2001) which have found that when participants report misinformation, they are more likely to report they “know” the misled item was in the original scene, rather than reporting they consciously “remember” it was there. One limitation of this experiment is that many factors were changed from the previous experiment. For example, in the current experiment an interview format was used instead of written questionnaires, smaller groups were used, a new video was made, and the new video allowed for the bidirectional flow of information. Because so many factors were changed we would not be able to tell which factor(s) was responsible for any variation in the findings. Fortunately, most of the findings were consistent with the previous research presented in this thesis. In sum, this study has shown that the effects of errant co-witness information can be so strong that they persist despite the most stringent testing requirements. That is, witnesses report misinformation even when they provide a verbal, free recall account within the context of an interview and even when they are explicitly warned that they may have encountered misinformation. However, despite these discouraging findings, it is encouraging to know that there may be a difference between memories obtained from a co- witness and memories obtained from first-hand experience. True memories are more likely to be ‘remembered’ and false memories are more likely to be ‘known.’ Therefore, using remember/know judgments could potentially reduce the amount of misinformation reported (by approximately 10%) if only “remembered” information were relied upon.

239

SECTION 4: GENERAL DISCUSSION

Section 4 Aims

• Summarise significant findings from the research presented in this thesis.

• Discuss implications of the research.

• Discuss limitations of the research and possible future directions.

Section 4 brings together findings from the seven studies presented in this thesis in an attempt to gain a broad understanding of the results obtained, and explain any inconsistencies. The first aim of Section 4 was to summarise significant findings from the research presented in this thesis. Findings from the seven studies are discussed in terms of how they help contribute to our understanding of the effects of co-witness discussion on eyewitness memory. The second aim of Section 4 was to discuss the implications of the research presented in this thesis. In particular, Section 4 discusses the methodological and practical implications of the findings. Finally, it is important to acknowledge any shortcomings of the research when interpreting the results. Therefore, the third aim of Section 4 was to discuss the limitations of the studies, together with ways in which the research may be improved upon.

240

Chapter 10 General Discussion

The general discussion is divided into three main sections: (1) overview and summary of significant findings; (2) implications of the research; and (3) limitations of the research and possible future directions.

Overview and Summary of Significant Findings There are a number of important findings presented in this thesis that make a significant contribution to our understanding of the effects of co-witness discussion on eyewitness memory. In particular, the studies were designed to expand our knowledge of the area by investigating six main aims. Each aim and the findings relevant to the aims are discussed separately.

AIM 1: To investigate the importance of studying co-witness discussion by measuring its prevalence.

Previous to this research, there was no empirical evidence relating to the prevalence of eyewitness discussion; only anecdotal evidence of co-witness discussion was available. Therefore, it was unclear how often witnesses talk to one another about the event, and whether this is a relevant topic of research. The research presented in this thesis is unique in that it is the only attempt that has been made to measure the discussion behaviour of real eyewitnesses. Study 1 surveyed real eyewitnesses about how often they discussed the event with one another and what the purposes of their discussions were. It was found that the vast majority of eyewitnesses reported that there was at least one other witness present at the event, and of these, 86% reported that they discussed details of the event with their co- witness. Furthermore, the most frequently stated reason for discussing the event with a co- witness was “providing information” about the event. In Study 2, surveyed police officers confirmed that the majority of witnesses discuss the event with one another prior to the arrival of the police. The finding that most eyewitnesses discuss the event with one another Chapter 10: General Discussion 241

provides a clear justification for the importance of studying the effects of co-witness discussion on eyewitness memory.

AIM 2: To investigate the opinions that police officers hold regarding discussion between witnesses.

To the best of our knowledge, no prior research has ever investigated the opinions and practices of police officers regarding co-witness discussion. In Study 1, survey responses given by eyewitnesses suggest that police officers are slightly more likely to encourage co-witness discussion than discourage it. This suggests that police officers may believe that discussion has beneficial effects, however the sample of respondents who were interviewed by the police was small, and may not be representative. Therefore, a survey of police officers was conducted in Study 2. The survey revealed that the police officers were somewhat critical of discussion. They were more likely to indicate that discussion can be detrimental to memory than beneficial to it. However, they could also think of some advantages of discussion, such as that it may refresh memory, reinforce memory, aid in trauma recovery, or help create a more accurate, overall picture of what happened. Most police officers reported that they had received instructions regarding discussion between witnesses. In particular, they reported that they were instructed to prevent discussion, mostly by discouraging witnesses from talking about the event and separating them. However, some officers conceded witnesses often discuss the event even before the police arrive at the scene of the incident.

AIM 3: To determine whether co-witness discussion helps or hinders individual recall and investigate this within a closely defined methodological set.

There is no single answer to the question of whether co-witness discussion helps or hinders individual recall. The results of the experimental studies presented in this thesis revealed that it can help, hinder, or have no effect on recall, depending on the methodology employed. It is important to understand how different methodologies can yield different results, as this can help us to understand why seemingly inconsistent findings have been reported in the literature. Chapter 10: General Discussion 242

Studies in this thesis demonstrated both advantages and disadvantages of experimentally induced co-witness information (both confederate and stimulus induced) depending on whether the postevent information provided to participants was correct or erroneous respectively. For example, in Study 4, when participants were told correct postevent information by a confederate co-witness, they were more likely to report this correct information than those who did not receive the information. Conversely, it was found in Studies 3, 4, 5, 6 and 7 that when participants were told misleading postevent information by a co-witness, they were more likely to report this erroneous information than those who did not receive the information. Furthermore, in Study 3 it was found that erroneous co-witness information had a particularly strong effect on the individual’s testimony when it was presented after a delay. An unexpected advantage of co-witness discussion was found in Studies 3 and 5. In these studies, it was found that although participants in groups who received experimentally induced co-witness information (both confederate and stimulus induced) were less accurate for misled items than those who did not receive experimentally induced information, they were more accurate for non-misled items. That is, in an experimental situation in which witnesses discussed the event with others who had seen slightly different events, or discussed with a confederate who was deliberately providing misinformation, witnesses had more accurate memories for the non-misled information. This was an unexpected and interesting finding. In Study 3, the possibility existed that this finding was a result of the confederates over-assisting the group. However, because confederates where not used in Study 5, this explanation cannot be used to account for the finding. Another possible explanation is that when group members were faced with a differing opinion they may have tried to reinstate the context of the original event to refresh their memories, which may have improved their memories for non-misled items (Tulving & Thomson, 1973). Furthermore, it may be the case that disagreement in groups can generate better, more elaborate discussion. Research has shown that group discussion can be enhanced when a member voices a differing opinion (e.g., Janis, 1972). Alternatively, participants may have steered the group discussion away from topics on which people did not agree (i.e., the misled items), and instead focused on information about which they agreed (i.e., the non- misled items). Research suggests that there is a tendency for group discussions to focus on Chapter 10: General Discussion 243

the reiteration of information that group members share in common, with little attention being paid to the exposure of information that is unique to individual members (Wittenbaum, Hubbell, & Zuckerman, 1999). However, it is important not to place too much emphasis on the finding that groups with experimentally induced information are more accurate on non-misled items because it was not replicated in Studies 6 and 7 (although the means were typically in the same direction). These advantages and disadvantages of Experimentally Induced Information are evident when the participants are compared with individuals who have not discussed the event as well as Natural Discussion Groups. To date, Natural Discussion Groups have been neglected as a comparison group by studies utilising the Experimentally Induced Information methodology. This is surprising considering that they are the most appropriate control group for studies utilising the Experimentally Induced Information methodology because they are not confounded to the extent that the other comparison groups are (i.e., studies using individuals who did not discuss the event and with-in subject designs). The current research makes an important contribution because it shows that the effects of Experimentally Induced Information persist when using Natural Discussion Groups as a comparison group. Whilst the effects of Experimentally Induced Information on recall are important to investigate, it is also important to investigate whether co-witness discussion per se, without the influence of Experimentally Induced Information, also affects accuracy of recall. This can be done by comparing Natural Discussion Groups with individuals who have not discussed the event. Natural Discussion Groups have high ecological validity and can be used to directly investigate the impact of discussion on memory. In Study 3 it was found that there were no differences in accuracy between the groups without confederates and the no-discussion control groups. Furthermore, in Study 5 it was found that there were no differences in accuracy between those in same-video groups and the no-discussion controls on either the misled or non-misled items. That is, when information was not experimentally induced there were no advantages or disadvantages of co-witness discussion prior to individual testimony. Previous studies have found mixed findings when comparing Natural Discussion Groups to individuals not in a discussion group, (e.g., Underwood & Milton, Chapter 10: General Discussion 244

1993; Yarmey, 1992; Yarmey & Morris, 1998), so the current research can add to the debate.

AIM 4: To investigate the relative impact of co-witness discussion on eyewitness memory in comparison with other ways of encountering postevent information.

Study 4 revealed that co-witness information has an exceptionally strong influence on subsequent individual testimony. That is, participants were generally more likely to report co-witness information than postevent information encountered through leading questions or a media report. This was true regardless of the accuracy of the postevent information (correct or misleading) and regardless of the type of questioning (recognition, short answer, or free recall). Furthermore, after having been misled, participants who discussed the event with a co-witness were often overconfident in their incorrect answers. That is, participants not only reported the misinformation introduced to them by a co- witness, but they were also confident in the accuracy of the erroneous information. To date, no studies have systematically investigated the variety of ways of encountering postevent information examined in Study 4. A few studies have compared the relative impact of co-witness information and other ways of encountering postevent information, and those that have (e.g., Gabbert, Memon, Allan, & Wright, 2004; Meade & Roediger, 2002; Shaw, Garven & Wood, 1997), have not investigated eyewitness confidence. The results from Study 4 that co-witness information has a particularly strong influence on witness accuracy and confidence suggests that findings from studies investigating other ways of encountering postevent information should not be generalised to co-witness information. Because social factors may have a strong influence on witnesses, co-witness information should be studied in its own right.

AIM 5: To establish whether it is possible to mediate negative effects of co- witness discussion by employing our theoretical understanding of the causes of the effect.

In order to establish whether it is possible to mediate negative effects of co-witness discussion, five methodologies were employed: (1) participants were given warnings about possible misinformation, (2) participants were asked to explicitly remember the source of Chapter 10: General Discussion 245

the information they reported, (3) participants were asked to make “remember/know judgments” (Tulving, 1985), (4) participants were asked to give confidence ratings, and (5) participants were asked to give free recall statements. Warnings: In Studies 5, 6, and 7 it was found that when people were warned that the co-witness information they received may be incorrect, they still reported it. That is, warnings about possible misinformation had no significant effect on whether or not participants reported postevent information in comparison to participants who received no such warning. This suggests that participants were unable to distinguish what they actually witnessed from the information they discussed with a co-witness. Whereas Meade and Roediger (2002) found that warnings significantly reduced the effect of the co-witness misinformation, but did not eliminate it, the present research did not find evidence that warnings significantly reduced the effects of co-witness discussion. This discrepancy in results may be due to the fact that participants in the studies presented in the thesis experienced a longer delay between the presentation of postevent information and warnings than those in Meade and Roediger’s study. Source misattribution is more likely to occur after a longer delay (Underwood & Pezdek, 1998) and therefore the warnings may not have been as effective after a longer delay because participants were unable to decipher whether the memory was from the original scene or the co-witness discussion. Source monitoring: In Study 6 it was found that, in general, source monitoring judgments did not differ between participants whom we had attempted to mislead and those whom we had not. This finding suggests that participants may have had trouble monitoring the source of their information, despite explicit instructions to do so. Meade and Roediger (2002) obtained similar findings. They found that memory conformity persisted on a recognition/source monitoring test in which participants were asked to choose among possible sources of information. The fact that memory conformity occurred even when participants were asked to monitor the source of their memory is an important finding because prior research (e.g., Lindsay & Johnson, 1989) has shown that source monitoring can eliminate the misinformation effect when information is obtained through other ways of encountering misinformation than co-witness discussion. It is interesting to note that Meade and Roediger (Experiment 4, 2002) found that participants were more likely to misattribute the source of misinformation to the original stimulus when it was encountered through co- Chapter 10: General Discussion 246

witness discussion than when it was encountered through the indirect transfer of co-witness information. There are a few possible reasons why participants may find it harder to monitor their sources after co-witness discussion than other ways of encountering postevent information. First, co-witness discussion may be more easily confused with the actual event because the two sources are more similar than, for example, a written narrative would be to the event. That is, the co-witness discussion and event may be quite similar because the co- witness was at the original event, which might make the two events easily confused. Furthermore, co-witness discussion may be easily confused with the event because the modalities under which the information is obtained are quite similar (i.e., both interactive verbally and visually) relative to, for example, a video and a written narrative. A further explanation may lie in the fact that participants may view the co-witnesses as more credible than information obtained from the experimenter because they believe they saw the same thing as the co-witness and they are generally unaware of any manipulation when discussing the event with a co-witness. People generally assume that information exchanged during the course of a discussion is truthful and accurate (Grice, 1975). However, if people fail to realize that this conversational assumption was not met, then they may not make an effort to monitor their sources (Echterhoff, Hirst, & Groll, 2001). When people do not make an effort to monitor the source of their information at the time of encoding because they believe the source to be credible, then presumably they would find it very difficult to separate the sources later on. Remember/Know Judgments: Whilst warnings and source monitoring can be used to investigate the fate of the original memory, remember/know judgments (Tulving, 1985) can be used to investigate the phenomenological experience associated with memories. That is, remember/know judgments can be used to study the conscious experience that accompanies the retrieval of true memories versus misinformation. Research from the current thesis suggests that memories for misinformation may be in some way different from true memories for events experienced. That is, through the use of remember/know judgments it may be possible to distinguish memories for what people actually witnessed from information obtained from a co-witness. In Study 5 it was found that participants who received stimulus induced misinformation, were more likely to give “know” judgments than those who did not receive stimulus induced misinformation. In Chapter 10: General Discussion 247

contrast, participants who did not receive stimulus induced misinformation were more likely to give “remember” judgments than those who received stimulus induced misinformation. In a separate analysis, it was found that participants were 14-23% more likely to be accurate if they indicated they consciously ‘remembered’ an item was in the video than if they indicated they ‘knew’ the statement was true, but it failed to evoke any specific recollection from the video. Similarly, in the free recall data from Study 7 it was found that participants who received stimulus induced misinformation indicated a higher proportion of “remember” judgments for accurate propositions than misleading postevent information. In contrast, they indicated a higher proportion of “know” judgments for misleading postevent information than accurate propositions. When the data from all participants in Study 7 were examined, it was found that participants indicated a higher proportion of “remember” judgments for accurate propositions than spontaneous fabrications. Conversely, participants indicated a higher proportion of “know” judgments for fabrications than accurate propositions. Participants were 7-16% more likely to be accurate if they indicated ‘remember’ than if they indicated ‘know.’ This finding suggests that information associated with ‘remember’ judgments may be somewhat more reliable than that associated with ‘know’ judgments. These findings suggest that ‘memories’ for misinformation may be somewhat different from memories for an actual event, possibly because they contain fewer episodic and perceptual details. Alternatively, participants may have given “know” judgments for misinformation because although they did not recall the misinformation from the event, they did recall learning about it after the event from a co-witness. That is, as Meade and Roediger (2002) have also observed, “know” responses are somewhat ambiguous in their meaning, as they may imply either private or public conformity. However, Meade and Roediger suggest that if participants are following the remember/know instructions typically given (e.g., Rajaram, 1993; Tulving, 1985), as the present studies instructed them to do, it is not as likely that their “know” responses would reflect information they were aware came from a co-witness (i.e., they are aware that they did not see the information in the event). Chapter 10: General Discussion 248

The findings from Studies 5 and 7 support previous research, which also suggest that when participants report misinformation, they are more likely to report they “know” the misled item was in the original scene, than to report that they consciously “remember” it was there (e.g., Frost, 2000; Roediger, Jacoby, & McDermott, 1996). These results have also been found for studies investigating co-witness information (e.g., Meade & Roediger, 2002; Roediger et al., 2001). However, as previously mentioned, the studies reported by Meade and Roediger use photos as stimuli and do not allow for natural, free flowing discussion, as participants are required to answer questions in the presence of a confederate. Studies 5 and 7 investigated remember/know judgments within a more ecologically valid context. In the studies reported in this thesis, crime videos were used as stimuli and participants were able to freely discuss the event with a co-witness. Although the current research investigated the judgments within a more forensically relevant context, similar results were obtained. Roediger et al. use the source monitoring theory (e.g., Johnson et al., 1993; Lindsay & Johnson, 1989) to explain the finding that most reported misinformation tends to be “known” rather than “remembered.” They suggest that this finding is due to the fact that participants have difficulty locating the source of their memories. They argue that if a source could be confidently attributed, the item should be judged as “remembered.” When the source can’t be confidently attributed, the item is judged as “know.” It is important to note that whilst participants were more likely to report that misinformation was “known” rather then “remembered” many participants in the current thesis did report that they consciously “remembered” a misled item. Some other studies have shown that misinformation can be accompanied by remember judgments (e.g., Roediger & McDermott, 1995). According to Meade and Roediger (2002), when false memories are accompanied by “remember” judgments, this implies that memory change has occurred. Frost (2000) suggested that misinformation may be more likely to be “remembered” after longer delays (of one or two days) than short delays, because the memory for the original event loses its perceived details, and may therefore become more similar to the misinformation. That is, after a longer delay, participants may be more likely to state that the misinformation is “remembered” because it would be more comparable to the original event in terms of perceptual details than after a shorter delay. In Studies 5 and Chapter 10: General Discussion 249

7, long delays were used comparable to other studies, which may have contributed to the tendency for participants to judge the misinformation as “remembered” rather than “known.” Debate exists regarding the meaning of remember/know judgments. Some researchers (e.g., Gardiner & Java, 1993) argue that remember and know responses reflect output from two different memory systems (i.e., the two-factor theory). That is, they argue that the distinction between remembering and knowing forms the basis of the distinction between episodic and semantic memory. “Remember’ responses are typically believed to reflect retrieval from episodic memory. In contrast, “know” responses are typically believed to reflect retrieval from generic (semantic) memory (Tulving, 1985) or retrieval based on global familiarity of the event (e.g., Jacoby, Yonelinas, & Jennings, 1997). Evidence supporting this view suggests that experimental manipulations influence remembering and knowing in different ways (see Gardiner & Java, 1993 for review). That is, it has been found that some experimental manipulations affect remember judgments but not know judgments, and vice versa. On the other hand, some researchers argue that a one-factor theory, rather than a two-factor theory should be used to explain the pattern of remember/know judgments (e.g., Donaldson, 1996; Dunn, 2004). For example, Donaldson (1996) argues that remember/know judgments do not necessarily reflect separate memory systems. Rather, he believes that the findings can be explained by simply assuming that individuals require more evidence to produce a “remember” response than a “know” response. That is, according to this view, the instructions to respond, “remember” or “know” are interpreted by participants as reflecting different levels of confidence concerning the memory retrieval, rather than different processes. Whilst the results from the current thesis suggest that there is some sort of difference between true memories for an event and co-witness misinformation, it is unclear whether this difference is qualitative or quantitative. That is, it is unclear whether the data support a two-factor or a one-factor theory. This is an important issue, but unfortunately it is one that goes beyond the scope of this thesis. Confidence Ratings. The findings from confidence ratings given by participants were somewhat ambiguous. In Study 3 it was found that groups with confederates were less Chapter 10: General Discussion 250

confident on misled items than participants in Natural Discussion Groups and individuals who did not discuss the event. However, in Study 4 it was found that participants who discussed the event with a confederate were more confident in their answers to misled items than individuals who did not discuss the event. In studies using Stimulus Induced Information (Studies 5 & 6), it was found that there were no differences in confidence ratings for misled items between individuals who were in mixed-video groups and those who were in same-video groups, or individuals who did not discuss the event. The finding that misled participants often have as much confidence in their recognition judgments as non-misled participants is consistent with past research (Loftus, Donders, Hoffman, & Schooler, 1989). It is unclear why such different results were obtained for the four studies investigating eyewitness confidence, although it is interesting to note that the two studies that used the Stimulus Induced Information methodology found no effect of co-witness discussion on confidence, while the two studies that used the Confederate Induced Information methodology did. It is possible that the confederates were not acting as true witnesses would, and that their influence either inflated or diminished the participants’ confidence depending on how the confederate behaved in the group. A benefit of the Stimulus Induced Information methodology is that participants behave in a more natural way than confederates and therefore, it may be that real co-witnesses do not greatly affect each other’s confidence. Free Recall. In Studies 4-7, participants were asked to give free recall accounts of the event because the use of forced-choice recognition tests can render an inherent response bias in the experiment (McCloskey & Zaragoza, 1985). Even though the use of free recall questionnaires reduces the demand characteristics of the studies, it was still found that participants who received Experimentally Induced Information were more likely to report this information than those who did not receive the information. Although co-witness misinformation was still reported in free recall, the extent to which it was reported was greatly diminished. This is consistent with previous research, which has found that the misinformation effect decreases when participants are asked more open-ended questions (Hollin & Clifford, 1983; Sutherland & Hayne, 2001; Zaragoza et al., 1987).

Chapter 10: General Discussion 251

AIM 6: To test existing theories on the misinformation effect and hence contribute to knowledge regarding the causes of memory conformity.

Theories that may be used to explain the causes of memory conformity were reviewed in Section 1. Of these theories, normative social influence (Deutsch & Gerard, 1955) can be ruled out as a possible explanation for the memory conformity observed in the studies presented in the current thesis. This explanation cannot account for the findings because it is unlikely that participants would feel pressure from others to conform when they are asked to give individual accounts of the original stimulus, away from their group members. The results from the current thesis also suggest that informational influence (Deutsch & Gerard, 1955) and biased guessing (McCloskey & Zaragoza, 1985) cannot completely account for the memory conformity observed in misled participants. These explanations cannot account fully for the findings because the free recall format of questioning, combined with warnings about possible misinformation, would be expected to significantly reduce the amount of misinformation reported if participants were able to distinguish their true memories from the co-witness information. However, in the present thesis it was found that participants included co-witness information in their free recall accounts, despite warnings against doing so. The finding that participants continued to report co-witness misinformation despite stringent testing requirements, (i.e., warnings about possible misinformation and free recall statements) suggests that their memory for the original event may have been modified. On the other hand, because the amount of co-witness information reported decreased in the free recall format and after warnings (although not significantly), it is also possible that informational influence and biased guessing may provide a partial explanation for some of the memory conformity observed under less stringent testing conditions. In order to differentiate between the various memory change explanations, the research presented in this thesis compared predictions made by the alternative explanations, but focused mostly on the source monitoring framework. According to the source monitoring framework, the misinformation effect should decrease if participants are instructed to monitor the source of their information. In contrast, the memory alteration explanations would predict that source monitoring would have no influence on the Chapter 10: General Discussion 252

misinformation effect. In Study 5, it was found that memory conformity persisted on a recognition test in which participants were asked to monitor the source of the information they reported. In general, source monitoring judgments did not differ between misled participants and those who had not been misled. Therefore, the current research does not support the source monitoring framework. Rather, the alteration hypothesis appears to provide the best explanation about why some participants in Study 5 conformed and reported the memories of their co-witnesses, despite source monitoring instructions. However, one note of warning regarding this conclusion is that the source monitoring instructions used in Study 5 were given after a fairly long delay (one week) and this long delay may account for the participants’ inability to monitor their sources despite instructions to do so. It is unclear whether participants would continue to find it difficult to monitor their sources if there was a shorter delay. It may be the case that after a short delay, the source monitoring framework can be used to explain co-witness memory conformity, however after a longer delay, participants may no longer be able to monitor the source of the information, despite instruction to do so, and thus the alteration hypothesis can be used to explain the memory conformity.

Implications of the Research

The findings from the current thesis have many implications. In this section, the methodological and practical implications of the research are discussed separately.

Methodological Implications In Study 1, the finding that witnesses typically have co-witnesses has important methodological implications. Whilst Yuille and Tollestrup (1992) argue that real eyewitnesses are unlike the typical laboratory participant because eyewitnesses are often the soul witness and also the victim, this may not be the case. Research from this thesis suggests that the ecological validity of laboratory studies may not be as low as Yuille and Tollestrup would suggest because often real witnesses are bystanders and have multiple co- witnesses. This suggests that laboratory research investigating the non-victim witness may be valid and that it is important to study the effects of co-witness discussion on eyewitness memory. Chapter 10: General Discussion 253

A second methodological implication of this research concerns the use of Natural Discussion Groups. To date, most research investigating the effects of co-witness discussion on memory has not utilised Natural Discussion Groups. The argument is made in this thesis that Natural Discussion Groups should be used, for two main reasons. First, Natural Discussion Groups have potentially the highest ecological validity, as they do not rely on the experimental induction of misinformation. Most research focuses solely on the effects of misinformation on witness memory, and the assumption is made that discussion between witnesses is a negative, detrimental process and no attempt is made to measure any possible positive effects of discussion. Natural Discussion Groups should be used as they make it possible to assess the effects of discussion, both positive and negative, in a natural setting without Experimentally Induced Information. Second, Natural Discussion Groups are an appropriate control group for studies investigating experimentally induced information. Studies which employ individuals who did not discuss the event as a control are inferior in this respect because it is unclear whether the effects are due to the act of discussing the event, the social presence of another, the elaboration and rehearsal of event- related memories, the impact of the misinformation, or some other aspect of the confederate’s influence. There are also problems associated with the use of a methodology that contrasts items for which participants had been misled by a confederate, and those for which they had not been misled. As with any within-subjects design, there can be the problem of carryover effects. For example, if all participants are receiving some misinformation from a confederate, this may affect the credibility of the confederate, and therefore the extent to which participants accept subsequent information presented to them by the confederate. A third methodological implication of this research concerns the use of Experimentally Induced Information. In this thesis, both Confederate Induced Information and Stimulus Induced Information were used, and each has its advantages and disadvantages. The Confederate Induced Information methodology allows for more experimental control, and there is certainty that the misled items will, in fact, be presented to the participant. However the Stimulus Induced Information methodology has higher ecological validity because it does not rely on a confederate, who may not act like a real witness would. The experiments reported here show that participants conform and report Chapter 10: General Discussion 254

what their co-witnesses had said, regardless of the methodology employed. It is recommended that the Stimulus Induced Information methodology be employed in preference to other methodologies commonly used in this field because it has higher ecological validity, and yet still gives rise to the misinformation effect. A fourth methodological implication of this research concerns the type of test used to assess witness memory. The studies reported in this thesis used recognition tests, cued recall tests, and free recall. In general, the effects of co-witness information were found for all types of memory test. Given the response biases and demand characteristics associated with recognition tests or even some cued recall tests (McCloskey & Zaragoza, 1985) it is recommended that free recall be used in future experiments. A final methodological implication of this research concerns the use of experimental manipulation checks. In the current thesis, care was taken to check that the experimental manipulations were successful and that the participants were not aware of the purpose of the studies. For example, participants were given questions relating to their discussion experiences in order to ensure that participants in groups who received Experimentally Induced Information did not perceive the discussion any differently than those in the Natural Discussion Groups. Furthermore, participants were questioned about what they believed to be the purpose of the experiment, and the data from the few who guessed the purpose were omitted from the analyses. Many studies are not so thorough with their manipulation checks, and it is advised that they be used.

Practical Implications Although there is some evidence in this thesis which suggests that discussion can be beneficial to eyewitness memory, the research has consistently shown that there are also disadvantages. In this thesis it has been demonstrated that co-witness discussion is the most powerful method of inducing misinformation and it is very difficult to differentiate accurate information from co-witness induced information. Therefore, in order to eliminate the negative influences of co-witness discussion on memory, it is probably necessary to sacrifice the advantages that it may have. For this reason it is recommended that co-witness discussion is discouraged. Chapter 10: General Discussion 255

The different theoretical explanations of memory conformity have different practical implications. Because the alteration hypothesis was supported to some extent in the current thesis, then it is likely that a witness’s memory for the original event may be permanently altered if he/she is given any subsequent information about the event. It is therefore important for police officers to establish whether a witness had discussed the event with others prior to the arrival of the police. Furthermore, it would be beneficial for police officers to separate witnesses and warn them not to discuss the event with one another as soon as they arrive at the scene. It would also be advisable for police to refrain from questioning the witnesses in front of one another, or conveying any information from one witness to another. However, in Study 2 police officers reported that witnesses often speak to one another before the police have time to arrive at the incident, and that it is almost impossible to prevent discussion from occurring. The implications of this are that if witnesses will discuss the event with one another despite instructions not to, then perhaps in addition to trying to prevent discussion, researchers and the police should focus on improving interview techniques that attempt to elicit accurate, first-hand memories of the event and exclude possible misinformation. One such interview technique investigated in the current thesis was to ask witnesses to give free recall accounts of the event, as opposed to asking direct, closed questions. Although co-witness misinformation was still reported in free recall, the extent to which it was reported was greatly diminished. It is therefore advisable that legal authorities try to ask broad, open-ended questions when interviewing a witness rather than direct, closed questions. Another interview technique investigated in the current thesis was to warn witnesses about possible misinformation. However, it was found that warnings do not significantly decrease the negative effects of misleading co-witness information, which suggests that the legal system should not rely too heavily on such warnings. For example, when a police officer warns a witness not to provide hearsay evidence, the witness may report information he/she heard from a co-witness without the intention of doing so. Despite the fact that witnesses may unintentionally report co-witness information when warned against doing so, it is probably still advisable to give such warnings because the warnings tended to Chapter 10: General Discussion 256

slightly, but not significantly decrease co-witness misinformation reported and they had no discernable negative effects. Thus, it appears that whilst we cannot rely on warnings to significantly decrease the reporting of co-witness information, there may be a slight, but non-significant benefit in doing so, and no disadvantage. A third interview technique that was investigated in the current thesis was to ask participants to think carefully and identify the source of their memories. However, it was found that after a one-week delay participants still reported co-witness information and the technique was not effective in distinguishing between true memories of and event and those obtained through co-witness discussion. It is unknown whether witnesses would have the same trouble monitoring the source of their information after a shorter delay. In general, it is recommended that eyewitness interviews should take place as soon as possible after the event. This is confirmed by Study 3, which found that co-witness information has a stronger effect on witness memory after a longer delay. It is possible that after a shorter delay source monitoring may be more successful and therefore worth utilising. However, until this is investigated further, this technique is not recommended. A fourth technique that was investigated was to assess the participants’ confidence in their memories. It was found that confidence was not a reliable indicator of witness accuracy, and thus it is not advisable to attempt to use expressions of confidence to differentiate between real and co-witness induced memories. That is, the findings that misled participants may have higher, lower, or equal confidence to those who were not misled suggests that eyewitness confidence cannot be used to distinguish real memories from memories obtained second-hand from a co-witness. A final interview technique that was investigated was the use of remember/know judgments (Tulving, 1985). Studies 5 and 7 found that when using remember/know judgments it may be possible to distinguish between true memories and those obtained from a co-witness. Furthermore, remember/know judgments could potentially reduce the amount of misinformation reported (by approximately 10%) if only “remembered” information were relied upon. Therefore, it is advisable that this technique be employed to help differentiate between real and co-witness induced memories. Whilst this technique may be reasonably difficult to apply and may have low face validity, Study 7 has shown that it can be effectively employed within an interview context. Chapter 10: General Discussion 257

In sum, our practical recommendations are as follows: i. Police officers should be trained about the dangers of co-witness contamination. ii. Police officers should interview witnesses individually, as soon as possible after the event. iii. Police officers should separate witnesses and warn them not to talk about the event with one another. iv. Police officers should warn witnesses about possible misinformation. v. Police officers should ask for a free recall account of what happened and not ask any closed, direct questions. vi. Police officers should ask witnesses to make remember/know judgments.

Limitations of the Research and Future Directions

A possible limitation of the research in this thesis is that a relatively homogenous student sample was used. This limitation is particularly notable in Study 1, in which student eyewitnesses were surveyed. Students may be more likely to witness certain crimes and may be more likely to talk about them than the general population. Furthermore, in the experimental studies student participants may not have behaved as typical witnesses would. For example, research on conformity has found that age (Iscoe, Williams, & Harvey, 1963) and possibly gender (e.g., Bond & Smith, 1996) may affect the likelihood that a witness conforms. Given the fact that the participants in the experiments were all young and predominantly female, this may affect the generalisability of the results. Whilst one experiment (Gabbert, Memon, & Allan, 2003) has looked at the effects of age on memory conformity (comparing young and older adults) and found no effect, it may be interesting to investigate age further, as well as the effects of gender on memory conformity in the context of co-witness discussion. A further limitation of the research is that the experiments were all conducted within a laboratory setting and this may limit the ecological validity of the findings. However, research in this thesis suggests that the ecological validity of laboratory studies may not be as low as some researchers (e.g., Yuille & Tollestrup, 1992) would suggest. Chapter 10: General Discussion 258

Furthermore, an attempt has been made in this thesis to increase ecological validity by using videos instead of photos or wordlists as stimuli. Using staged events would have increased the ecological validity further, however it was not possible to use staged events when utilizing the Stimulus Induced Information methodology because it would not be possible to ensure that witnesses would see slightly different stimuli whilst viewing the event simultaneously. In this thesis the Stimulus Induced Information methodology was utilised in order to increase ecological validity. A concern about utilizing confederates is that they may not act like real witnesses and their behaviors may alter those of other group members. However, the Stimulus Induced Information methodology, whilst increasing ecological validity, also resulted in a decrease of experimental control. In this methodology, there is no control of whether participants were, in fact, discussing the items that differed between the different versions of the stimulus. Group conversations were recorded in order to encourage them to discuss the event and to make them believe that the tapes would be used to assess their recall, however the recordings were not of high enough quality to actually do this. It would be valuable for future research to investigate what actually happens during the group discussions (i.e., how experimentally induced misinformation is received by the group and what effect this has on memory conformity as well as whether spontaneous misinformation is presented by group members and if this results in memory conformity). A further limitation of this thesis relating to group conversations involves the Natural Discussion Groups. Whilst it was found in this thesis that there were no differences between Natural Discussion Groups and the no-discussion controls, it may be the case that the measurements weren’t sensitive enough to detect differences between the groups. For example, when investigating the possible deleterious effects of misinformation, it would be beneficial to examine any specific misinformation that was spontaneously introduced to the group. This would involve listening to audio recordings of the group discussions, recording any misinformation that was spontaneously introduced by the group members (i.e., incorrect information that is not experimentally induced), and checking individual statements to see if the group members report their own spontaneous misinformation. This analysis was not done in the current thesis because some of the audio recordings were of Chapter 10: General Discussion 259

very poor quality and difficult to understand. Future research should investigate this issue further. One way to increase experimental control within the group discussions would be to restrict the conversations by having participants answer specific questions in front of one another. However, this methodology would have resulted in a decrease in ecological validity because the conversations would not have progressed naturally. A further threat to ecological validity is that all of the participants in the studies were aware that they were going to witness a crime scene. This was done in order to entice participants to sign up for the study. However, real witnesses are often unaware that a crime is going to take place, and thus may not be paying very close attention. Therefore, if anything, the memory conformity effects in the current studies may be underestimated because participants may have paid closer attention to the original stimulus and have been less easily persuaded by others. Furthermore, the Natural Discussion Groups may have performed differently if they had not been told that they were going to witness a crime scene. Research by Yarmey and Morris (1998) has shown that when participants in Natural Discussion Groups were not expecting to see an accident, there was no effect of discussion on recall performance. However, when warned that they were going to witness an accident, those who had discussed the event in Natural Discussion Groups performed better than whose who had not discussed the event. Further research should investigate the effects of warning people that they will see a crime on memory conformity. It would also be valuable to undertake further research on the effects of warnings about possible misinformation. In the studies presented in this thesis no attempt was made to assess the impact of delay on whether participants could respond to warnings or source monitoring instructions after a relatively brief period of time. In the current thesis, participants did not complete the source monitoring test and were not warned about possible misinformation until one week after encountering it. In a forensic context, a delay of one week between the event and interview is not an abnormally long period of time (Kapardis, in preparation, cited in Kapardis, 1997). However, the possibility exists that participants would find it easier to disregard postevent information and monitor the source of their information if there was a shorter delay between the misinformation and the warning or source monitoring instructions. Research investigating ways of encountering Chapter 10: General Discussion 260

postevent information other than through co-witness discussion has shown that warnings may reduce the misinformation effect if they are given immediately after the postevent information (Wright, 1993) but not after a longer period of time (e.g., Lindsay, 1990). Wright suggests that multiple memory traces may exist for a limited period, but not indefinitely. That is, after a period of time, the individual memory traces may decay or blend into some other representation. For example, studies have shown that memory sources decay more quickly than their contents (Pratkanis, Greenwald, Lieppe, & Baumgardner, 1988). Furthermore, research on the ‘sleeper effect’ suggests that participants sometimes remember the contents of a message, but forget the source of the message over time (Hovland & Weiss, 1951). Future research investigating the effects of immediate warnings and source monitoring tests on memory conformity would be valuable to help resolve this issue. Another possible limitation of the warning studies is that the warnings used were quite general in that they questioned the overall accuracy of the co-witness information. These warnings may not have been specific enough to inform the misled participants that the misinformation was indeed false (Wright, 1993). This criticism has been addressed by Lindsay (1990) who warned his participants that the information they had read in a postevent narrative was wrong and that it should not be reported on the memory test. Specifically, he told them, “there is no question on this test for which the correct answer was mentioned in the story” (p. 1080). Therefore, if participants were planning to report the information from the postevent narrative for informational reasons or task demands of the experiment, they would no longer do this, provided they could remember the source of the information. Unfortunately, this methodology could not be employed here in experiments utilising the Stimulus Induced Information methodology because it is the nature of this method that some of the information provided by the co-witness will be correct. This, it is difficult to control for whether any of the correct answers were mentioned in the discussions because participants were allowed to freely discuss the event. However, it would be possible for future research to investigate the effects of explicitly telling the participants which items were misled during the discussion and then later assessing their memory for the items. Without further research, it is difficult to make any strong assertions about the causes of memory conformity. Chapter 10: General Discussion 261

Without further research, it is also difficult for us to understand the meaning of remember/know judgments. Whilst the results from the current thesis suggest that there is some difference between true memories for an event and co-witness misinformation, it is unclear whether this difference is qualitative or quantitative. That is, it is unclear whether the data support a two-factor or a one-factor theory of memory. A further limitation of the remember/know methodology utilised in the current study is that participants were not allowed the option of “guess.” That is, participants were asked to indicate either “know” or “remember” to items they indicated were true, when in fact, the participants may have simply guessed the answer. Therefore, “know” responses may have been confounded when participants were merely guessing (Gardiner, Richardson-Klavehn, & Ramponi, 1997). Although this was addressed to some extent in Study 7 by assessing remember/know judgments for free recall accounts (thus eliminating the possibility that participants might guess an answer) it is still possible that participants were confusing know judgments with guesses, despite the explicit instructions they were given. Future research should investigate remember/know judgments further in order to ascertain whether there is a qualitative or quantitative difference between “remember” and “know” judgments.

Conclusion

The research presented in this thesis makes an original and significant contribution to our understanding of the effects co-witness discussion on eyewitness memory. Study 1 is the first known study to investigate the discussion behaviour of real eyewitnesses. By surveying eyewitnesses it was found that the vast majority of witnesses discuss the event with one another. Study 2 is the first known study to investigate the opinions and practices of police officers regarding co-witness discussion. From this study, it was found that although police officers report that they are generally critical of discussion between witnesses, they do concede that witnesses commonly discuss the event before their arrival at the scene. The findings from these surveys provide a clear justification for the importance of studying co-witness discussion. Furthermore, findings from the survey studies were used to inform the design of the experimental studies. Chapter 10: General Discussion 262

The experimental studies in this thesis investigated whether co-witness discussion is advantageous or disadvantageous to individual recall. In order to do this, two sound methodologies were employed, with thorough manipulation checks done. It was found that participants often reported Experimentally Induced Information (either Confederate or Stimulus Induced) that they had heard their co-witness report earlier, regardless of the accuracy of the information. Furthermore, it was found that co-witness information had a stronger influence on eyewitness reports than leading questions or written postevent narratives. Whilst the above results describe the effects of Experimentally Induced Information, this thesis also investigates whether co-witness discussion, without the influence of Experimentally Induced Information, results in the same findings. Very few studies on eyewitness recall have investigated Natural Discussion Groups and they have been completely neglected as a comparison group for studies investigating Experimentally Induced Information. It was found that Natural Discussion Groups showed no advantages or disadvantages in recall in comparison to individuals who did not discuss the event. In order to assess whether negative effects of co-witness discussion could be reduced, stringent testing requirements were employed. Participants were warned about possible misinformation, instructed to monitor the source of their information, and were asked to give free recall accounts. However, despite these stringent testing requirements, participants continued to report misinformation that they had heard from their co-witness. By employing a range of methodological refinements, the research presented in this thesis advances our understanding of the impact of co-witness discussion on individual testimony within a forensically relevant context.

263

REFERENCES

Ainsworth, P. B. (1998). Psychology, law, and eyewitness testimony. Chichester: John Wiley & Sons.

Alper, A., Buckhout, R., Chern, S., Harwood, R., & Slomovits, M. (1976). Eyewitness identification: Accuracy of individual vs. composite recollections of a crime. Bulletin of the Psychonomic Society,8, 147-149.

Allen, V. L., & Levine, J. M. (1969). Consensus and conformity. Journal of Experimental Social Psychology, 5, 389-399.

Allen, V. L., & Levine, J. M. (1971). Social support and conformity: The role of independent assessment of reality. Journal of Experimental Social Psychology, 7, 48-58.

Allen, B. P., & Lindsay. D. S. (1998). Amalgamations of memories: Intrusion of information from one event into reports of another. Applied Cognitive Psychology, 12, 277-285.

Allwood, C. M., Jonsson, A. C., & Granhag, P. A. (2004). The effects of source and type of feedback on child witnesses’ metamemory accuracy. Article submitted for publication.

Allwood, C. M., Knutsson, J., & Granhag, P. A. (in press). Eyewitnesses under influence: How feedback affects the realism in confidence judgements. Psychology, Crime, & Law.

Asch, S. E. (1951). Effects of group pressure upon the modification and distortion of judgment. In H. Guetzkow (Ed.), Groups, leadership, and men. Pittsburgh, PA: Carnegie.

Asch, S. E. (1952). Social psychology: NY: Prentice-Hall.

Asch, S. E. (1955). Opinions and social pressure. Scientific American, 193, 31-35.

Asch, S. E. (1956). Studies of independence and conformity: I. A minority of one against a unanimous majority. Psychological Monographs, 70 (Whole No. 416).

Ayers, M. S., & Reder, L. M. (1998). A theoretical review of the misinformation effect: Predictions from an activation-based memory model. Psychonomic Bulletin & Review, 5, 1-21.

264

Bartlett, Sir F. C., (1932). Remembering: A study in experimental and social psychology. London: Cambridge University Press.

Basden, B. H., Basden, D. R., Bryner, S., & Thomas, R. L., III. (1997). A comparison of group and individual remembering: Does collaboration disrupt retrieval strategies? Journal of Experimental Psychology: Learning, Memory, & Cognition, 23, 1176- 1189.

Basden, B. H., Basden, D. R., & Henry, S. (2000). Costs and benefits of collaborative remembering. Applied Cognitive Psychology, 14, 497-507.

Bekerian, D. A., & Bowers, J. M. (1983). Eyewitness testimony: Were we misled? Journal of Experimental Psychology: Learning, Memory, & Cognition, 9, 139-145.

Bekerian, D. A., & Dennett, J. L. (1990). Spoken and written recall of visual narratives. Applied Cognitive Psychology, 4, 175-187.

Belli, R. F. (1989). Influences of misleading postevent information: Misinformation interference and acceptance. Journal of Experimental Psychology: General, 118, 72-85.

Belli, R. F., Lindsay, D. S., Gales, M. S., & McCarthy, T. T. (1994). Memory impairment and source misattribution in postevent misinformation experiments with short retention intervals. Memory & Cognition, 22, 40-54.

Belli, R. F., Windschitl, P. D., McCarthy, T. T., & Winfrey, S. E. (1992). Detecting memory impairment with a modified test procedure: Manipulating retention interval with centrally presented items. Journal of Experimental Psychology: Learning, Memory, & Cognition, 18, 356-367.

Bernstein, E. M, & Putnam, F. W. (1986). Development, reliability, and validity of a dissociation scale. .Journal of Nervous & Mental Disease, 174, 727-735.

Betz, A. L., Skowronski, J. J., & Ostrom, T. M. (1996). Shared realities: Social influence and stimulus memory. Social Cognition, 14, 113-140.

Bond, R., & Smith, P. B. (1996). Culture and conformity: A meta-analysis of studies using Asch's (1952b, 1956) line judgment task. Psychological Bulletin, 119, 111-137.

Brainerd, C. J., & Reyna, V. F. (1998). Fuzzy-trace theory and children’s false memories. Journal of Experimental Child Psychology, 71, 81-129.

265

Brimacombe, E., Allison, M., & Garrioch, L. (2003, July) “Who do you choose?”: The effect of social influence on eyewitnesses’ and lineup administrators’ confidence in identification decisions. In L. Shapiro (Chair), Criminal identification. Symposium conducted at the Society for Applied Research in Memory and Cognition, Aberdeen, Scotland.

Ceci, S. J., Ross, D. F., & Toglia, M. P. (1987). Suggestibility in children’s memory: Psycholegal implications. Journal of Experimental Psychology: General, 116, 38- 49.

Chandler, C. C. (1989) Specific retroactive interference in modified recognition tests: Evidence for an unknown cause of interference. Journal of Experimental Psychology: Learning, Memory, & Cognition, 15, 256-265.

Chandler, C. C. (1991). How memory for an event is influenced by related events: Interference in modified recognition tests. Journal of Experimental Psychology: Learning, Memory, & Cognition, 19, 115-125.

Christiaansen, R. E., & Ochalek, K. (1983). Editing misleading information from memory: Evidence for the coexistence of original and postevent information. Memory & Cognition, 11(5), 467-475.

Clark, N. K., Stephenson, G. M., & Kniveton, B. H. (1990). Social remembering: Quantitative aspects of individual and collaborative remembering by police officers and students. British Journal of Psychology, 81, 73-94.

Cohen, G. (1989). Memory in the real world. Hillsdale, NJ: Erlbaum.

Corey, D. (2001). The effect of co-witness influence on witnesses’ memory reports in investigative interviews. Unpublished doctoral dissertation, University of Texas at El Paso.

Costanzo, P. R. (1970). Conformity development as a function of self-blame. Journal of Personality and Social Psychology, 14, 366-374.

Crutchfield, R. A. (1955). Conformity and character. American Psychologist, 10, 191-198.

Daubert v. Merell Dow Pharmaceuticals, Inc. 113 S.Ct. 2786 (1993).

Deffenbacher, K. A. (1991). A maturing of research on the behaviour of eyewitnesses. Applied Cognitive Psychology, 5, 377-402.

Deutsch, M., & Gerard, H. B. (1955). A study of normative and informational social influence upon individual judgment. Journal of Abnormal and Social Psychology, 51, 629-636.

266

Diehl, M., & Stroebe, W. (1987). Productivity loss in brainstorming groups: Toward the solution of a riddle. Journal of Personality and Social Psychology, 53, 497-509.

Dodd, D. H., & Bradshaw, J. M. (1980). Leading questions and memory: Pragmatic constraints. Journal of Verbal Learning and Verbal Behaviour, 21, 207-219.

Dodson, C., & Reisberg, D. (1991). Indirect testing of eyewitness memory: The (non)effect of misinformation. Bulletin of the Psychonomic Society, 29, 333-336.

Donaldson, W. (1996). The role of decision processes in remembering and knowing. Memory & Cognition, 24, 523-533.

Dunn, J. C. (2004). Remember-know: A matter of confidence. Psychological Review, 111, 524-542.

Eagley, A. H., Wood, W., & Fishbaugh, L. (1981). Sex differences in conformity: Surveillance by the group as a determinant of male nonconformity. Journal of Personality and Social Psychology, 40, 384-394.

Echterhoff, G., Hirst, W., & Groll, S. (2001). The communication of misinformation: Conversational assumptions affect source monitoring strategies. Unpublished manuscript, University of Cologne.

Edwards, D., & Middleton, D. (1987). Conversation and remembering: Bartlett revisited. Applied Cognitive Psychology, 1, 77-92.

Ellis, E. G., Nel, E. M., & Van Rooyen, J. (1991). Conformity behaviour of Afrikaans- and English-speaking adolescents in South Africa. Journal of Social Psychology, 131, 875-879.

Elmes, D. G., Kantowitz, B. H., & Roediger, H. L. (1991). Research methods in psychology (5th ed.). St. Paul, MN: West Publishing.

Field Operations (1997). Procedures for the Evidence Act. Retrieved November 30, 2004 from http://www.policensw.com/pdf/evidproc.pdf

Finlay, F., Hitch, G. J., & Meudell, P. R. (2000). Mutual inhibition in collaborative recall: Evidence for a retrieval-based account. Journal of Experimental Psychology: Learning, Memory and Cognition, 26, 1556-1567.

Fisher, R. P., & Geiselman, R. E. (1992). Memory enhancing techniques for interviewing witnesses: The cognitive interview. Springfield, IL: Thomas.

Forbes, J. R. (2003). The Laws of Australia, Testimony 16.4 [66].

Forsyth, D. R. (1983). An introduction to group dynamics. Monterey, CA: Brooks/Cole. 267

Foster, S., & Ost, J. (2003, July). The Effect of Discussion and Acquaintance on Individual Recall of Contradictory Details of a Simulated Crime. In F. Gabbert (Chair), Memory conformity: How do people influence each other’s memories? Symposium conducted at the Society for Applied Research in Memory and Cognition, Aberdeen, Scotland.

Froming, W. J., & Carver, C. S. (1981). Divergent influences of private and public self- consciousness in a compliance paradigm. Journal of Research in Personality, 15, 159-171.

Frost, P. (2000). The quality if false memory over time: Is memory for misinformation “remembered” or “known”? Psychonomic Bulletin & Review, 7, 531-536.

Frye v. United States, 293 F. 1013, 34 A.L.R. 145 (D.C. Cir. 1923).

Gabbert, F., Memon, A., & Allan, K. (2003). Memory conformity: Can eyewitnesses influence each other’s memories for an event? Applied Cognitive Psychology, 17, 533-543.

Gabbert, F., Memon, A., Allan, K. & Wright, D.B. (2004). Say it to my face: Examining the effects of socially encountered misinformation. Legal and Criminological Psychology, 9, 215-227.

Gallo, D. A, Roberts, M. J., & Seamon, J. G. (1997). Remembering words not presented in lists: Can we avoid creating false memories? Psychonomic Bulletin & Review, 4, 271-276.

Gallo, D. A., Roediger, H. L. III, & McDermott, K. B. (2001). Associative false recognition occurs without strategic criterion shifts. Psychonomic Bulletin & Review, 8, 579- 586.

Gardiner, J. M., & Java, R. I. (1993). Recognising and remembering. In A. F. Collins, S. E. Gathercole, M. A. Conway, & P. E. Morris (Eds.), Theories of memory. Hove, UK: Psychology Press.

Gardiner, J. M., Richardson-Klavehn, A., & Ramponi, C. (1997). On reporting recollective experiences and “direct access to memory systems.” Psychological Science, 8, 391- 394.

Garven, S., Wood, J. M., & Malpass, R. S. (2000). Allegations of wrongdoing: The effects of reinforcement on children's mundane and fantastic claims. Journal of Applied Psychology, 85, 38-49.

268

Goldstein, A. G., Chance, J. E., & Schneller, G. R. (1989). Frequency of eyewitness identification in criminal cases: A survey of prosecutors. Bulletin of the Psychonomic Society, 27, 71-74.

Granhag, A., Memon, A., Gabbert, F., & Allwood, C. M. (2004, July). Contagious contacts: How police trainees influence each other’s memories for a criminal event. Paper presented at the 14th European Conference on Psychology and Law, Krakow, Poland.

Granhag, P.A., Strömwall L.A., & Billings, J.F. (2003). "I'll never forget the sinking ferry!" How social influence makes false memories surface. In M. Vanderhallen, G. Vervaeke, P.J. van Koppen, & J. Goethals. Much ado about crime. Chapters on Psychology and Law. (pp. 129-140). Brussel: Politeia.

Greene, E., Flynn, M. S., & Loftus, E. F. (1982). Inducing resistance to misleading information. Journal of Verbal Learning and Verbal Behavior, 21, 207-219.

Grice, H. P. (1975). Logic and conversation. In P. Cole & J.L. Morgan (Eds.), Syntax and semantics 3: Speech acts. San Diego, CA: Academic Press.

Hain, P. (1976). Mistaken identity: The wrong face of the law. London: Quartet Books.

Hammersley, R., & Read, J. D. (1986). What is integration? Remembering a story and remembering false implications about a story. British Journal of Psychology, 77, 329-341.

Heaton-Armstrong, A. (1985). Police officers’ notebooks. The Criminal Law Review, 781- 784.

Heaton-Armstrong, A. (1987). Police officers’ notebooks: Recent developments. The Criminal Law Review, 470-472.

Higham, P.A., & Vokey, John R. (2004). Illusory recollection and dual-process models of recognition memory. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 57, 714-744.

Hoffman, H. G., Granhag, P. A., See, S. T. K., & Loftus, E. F. (2001). Social influences on reality-monitoring decisions. Memory & Cognition, 29, 394-404.

Hollin, C. R., & Clifford, B. R. (1983). Eyewitness testimony: The effects of discussion on recall accuracy and agreement. Journal of Applied Social Psychology, 13, 234-244.

Home Office. (2003). Police and Criminal Evidence Act 1984: Code D. (Version 1.4.2). Retrieved November 30, 2004 from http://www.homeoffice.gov.uk/docs/pacecodedrevised703.pdf

269

Hoppe, R. A. (1962). Memorizing by individuals and groups: A test of the pooling-of- ability model. Journal of Abnormal & Social Psychology, 65, 64-67.

Hovland, C. I., & Weiss, W. (1951). The influence of source credibility on communication effectiveness. Public Opinion Quarterly, 15, 635-650.

Huff, C. R., Rattner., A., & Sagarin, E. (1996). Convicted but innocent: Wrongful conviction and public policy. London: Sage.

Hui, C. H., & Triandis, H.C. (1986). Individualism-collectivism: A study of cross-cultural researchers. Journal of Cross-Cultural Psychology, 17, 225-248.

Hyman, I. E. (1994). Conversational remembering: Story recall with a peer versus for an experimenter. Applied Cognitive Psychology, 8, 49-66.

Iscoe, I., Williams, M., & Harvey, J. (1963). Modification of children’s judgments by a simulated group technique: A normative developmental study. Child Development, 34, 963-978.

Jacoby, L. L, Yonelinas, A. P., & Jennings, J. M. (1997). The relation between conscious and unconscious (automatic) influences: A declaration of independence. In J. D. Cohen & J. W. Schooler (Eds.), Scientific approaches to consciousness (pp. 13-47). Hillsdale, NJ: Erlbaum.

Janis, I. L. (1972). Victims of groupthink. Boston: Houghton Mifflin.

Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin, 114, 3-28.

Kanematsu, H., Mori, K., & Mori, H. (2003). Memory distortion in eyewitness pairs who observed nonconforming events and discussed them. Journal of The Faculty of Education, Shinshu University, 109, 75-84.

Kapardis, A. (1997). Psychology and law: A critical introduction. New York: Cambridge University Press.

Kassin, S. M., Tubb, V. A., Hosch, H. M., & Memon, A. (2001). On the "general acceptance" of eyewitness testimony research. American Psychologist, 56, 405-416.

Latané, B., Williams, K., & Harkins, S. (1979). Many hands make light the work: The causes and consequences of social loafing. Journal of Personality & Social Psychology, 37, 822-832.

Levine, F. J., & Tapp, J. (1973). The psychology of criminal investigation. University of Pennsylvania Law Review, 121, 1079-1131.

270

Lindsay, D. S. (1990). Misleading suggestions can impair eyewitnesses' ability to remember event details. Journal of Experimental Psychology: Learning, Memory, & Cognition, 16 1077-1083.

Lindsay, D. S., & Johnson, M. K. (1989). The eyewitness suggestibility effect and memory for source. Memory & Cognition, 17(3), 349-358.

Loftus, E. F. (1975). Leading questions and the eyewitness report. Cognitive Psychology, 7, 560-572.

Loftus, E. F. (1977). Shifting human color memory. Memory & Cognition, 5, 696-699.

Loftus, E. F. (1979a). Eyewitness testimony. Cambridge, MA: Harvard University Press.

Loftus, E. F. (1979b). The malleability of memory. American Scientist, 67, 312-320.

Loftus, E. F. (1992).When a lie becomes a memory’s truth: Memory distortion after exposure to misinformation. Current directions in Psychological Science, 1, 121- 123.

Loftus, E. F. (2003). Our changeable memories: Legal and practical implications. Nature Reviews: Neuroscience, 4, 231-234.

Loftus, E. F., Donders, K., Hoffman, H. G., & Schooler, J. W. (1989). Creating new memories that are quickly accessed and confidently held. Memory & Cognition, 17, 607-616.

Loftus, E. F., & Greene, E. (1980). Warning: Even memory for faces may be contagious. Law and Human Behaviour, 4, 323-334.

Loftus, E. F., & Hoffman, H. (1989). Misinformation and memory: The creation of new memories. Journal of Experimental Psychology (General), 118, 100-104.

Loftus, E. F., & Ketcham, K. (1991). Witness for the Defence. New York: St. Martin’s Press.

Loftus, E. F., & Loftus, G. R. (1980). On the performance of stored information in the human brain. American Psychologist, 35, 409-420.

Loftus, E. F., Miller, D. G., & Burns, H. J. (1978). Semantic integration of verbal information into a visual memory. Journal of Experimental Psychology: Human Learning & Memory, 4, 19-31.

Loftus, E. F., & Palmer, J. C. (1974). Reconstruction of automobile destruction: An example of the interaction between language and memory. Journal of Verbal Learning & Verbal Behavior, 13, 585-589. 271

Loftus, E. F., Schooler, J. W., & Wagenaar, W. A. (1985). The fate of memory: Comment on McCloskey and Zaragoza. Journal of Experimental Psychology: General, 114, 375-380.

Loftus, E. F., & Zanni, G. (1975). Eyewitness testimony: The influence of the wording of a question. Bulletin of the Psychonomic Society, 5, 86-88.

Luus, C. A. E., & Wells, G. L. (1994). The malleability of eyewitness confidence: Co- witness and perseverance effects. Journal of Applied Psychology, 79, 714-723.

McCloskey, M., & Zaragoza, M. (1985). Misleading postevent information and memory for events: Arguments and evidence against memory impairment hypotheses. Journal of Experimental Psychology: General, 114,1-16.

McDermott, K. B., & Roediger, H. L. III. (1998). Attempting to avoid illusory memories: Robust false recognition of associates persists under conditions of explicit warnings and immediate testing. Journal of Memory & Language, 39, 508-520.

McGeoch, J. A. (1932). Forgetting and the law of disuse. Psychological Review, 39, 352- 370.

McGhee, P. E., & Teevan, R. C. (1967). Conformity behaviour and need for affiliation. Journal of Social Psychology, 72, 117-121.

Meade, M. L., Roediger, H. L., III. (2002). Explorations in the social contagion of memory. Memory & Cognition, 30, 995-1009.

Memon, A., & Stevenage, S. V. (1996). Evidence on the remember/know status of errors: A reply to Higham & Roberts on witness memory, Psycoloquy: 7, #19 Witness Memory, (6).

Memon, A., & Wright, D. B. (1999). Eyewitness testimony and the Oklahoma bombing. Psychologist, 12, 292-295.

Metcalfe, J. (1990). Composite Holographic Associative Recall Model (CHARM) and blended memories in eyewitness testimony. Journal of Experimental Psychology: General, 119, 145-160.

Meudell, P. R., Hitch, G. J., & Boyle, M. M. (1995). Collaboration in recall: Do pairs of people cross-cue each other to produce new memories? The Quarterly Journal of Experimental Psychology, 48A, 141-152.

Meudell, P. R., Hitch, G. J., & Kirby, P. (1992). Are two heads better than one? Experimental investigations of the social facilitation of memory. Applied Cognitive Psychology, 6, 525-543. 272

Morello, C. & Davis, P. (2002, October 17). Shooter’s Description Still Vague, The Washington Post, p A01.

Mori, K. (in press). Surreptitiously projecting different movies to two subsets of viewers. Behavior Research Methods, Instruments, and Computers.

Mori, K., & Mori, H. (2004). Memory distortion of isolated eyewitnesses in an experimental collaborative eyewitness testimony. Manuscript submitted for publication.

Morton, J., Hammersley, R. H., & Berkerian, D. A. (1985). Headed records: A model for memory and its failures. Cognition, 20, 1-23.

Mudd, K., & Govern, J. M. (2004). Conformity to misinformation and time delay negatively affect eyewitness confidence and accuracy. North American Journal of Psychology, 6, 227-238.

Multhaup, K. S. (1995). Aging, source, and decision criteria: When false fame errors do and do not occur. Psychology & Aging, 10, 492-497.

Murdock, B. B., Jr. (1962). The serial position effect for free recall. Journal of Experimental Psychology, 64, 482-488.

National Institute of Justice (1999). Eyewitness evidence: A guide for law enforcement. Retrieved November 30 from http://www.ojp.usdoj.gov/nij/pubs-sum/178240.htm

Newlands, P. J., George, R. C., Towell, N. A., Kemp, R. I., & Clifford, B.R. (1999). An investigation of description quality from real-life interviews. Psychology, Crime & Law, 5, 145-166.

Newlands, P., & Skaltsis, M (2004). Eyewitness interviewing: Can interview technique moderate the effects of misleading post-event information? Paper accepted for presentation at the Society for Applied Research in Memory and Cognition Conference, Wellington, New Zealand.

Ost, J., Green, E., & Cherryman, J. (2004, July). Are three heads better than one? Police officers’ and students’ reports of a video-presented criminal event. Poster session presented at the 14th European Conference on Psychology and Law, Krakow, Poland.

Ost, J., Hogbin, I., & Granhag, P.A. (2004). False reports of a highly-charged public event: Dissociation and confederate influence. Paper accepted for presentation at the Society for Applied Research in Memory and Cognition Conference, Wellington, New Zealand.

273

Paterson, H. M. & Kemp, R. I. (in press). Co-witnesses talk: A survey of eyewitness discussion. Psychology, Crime & Law.

Payne, D. G., Toglia, M. P., & Anastasi, J. S. (1994). Recognition performance level and the magnitude of the misinformation effect in eyewitness memory. Psychonomic Bulletin & Review, 1, 376-382.

Penrod, S., & Cutler, B.L. (1995). Witness confidence and witness accuracy: Assessing their forensic relation. Psychology, Public Policy, & Law, 1, 817-845.

Perlmutter, H. V. (1953). Group memory of meaningful material. Journal of Psychology, 35, 361-370.

Pirolli, P. L., & Mitterer, J. O. (1984). The effect of leading questions on prior memory: Evidence for the coexistence of inconsistent memory traces. Canadian Journal of Psychology, 38(1), 135-141.

Rajaram, S. (1993). Remembering and knowing: Two means of access to the personal past. Memory & Cognition, 21, 89-102.

Rohrer, J. H., Baron, S. H., Hoffman, E. L., & Swander, D. V. (1954). The stability of autokinetic judgments. Journal of Abnormal and Social Psychology, 49, 595-597.

Roediger, H. L., III, Jacoby, J. D., & McDermott, K. B. (1996). Misinformation effects in recall: Creating false memories through repeated retrieval. Journal of Memory & Language, 35, 300-318.

Roediger, H. L. III, & McDermott, K. B. (1995). Creating false memories: Remembering words not presented in lists. Journal of Experimental Psychology: Learning, Memory, & Cognition, 21, 803-814.

Roediger, H. L., III, Meade, M. L., & Bergman, E. T. (2001). Social contagion of memory. Psychonomic Bulletin & Review, 8(2), 365-371.

Roediger, H. L., III, & Payne, D. G. (1982). Hypermnesia: The role of repeated testing. Journal of Experimental Psychology: Learning, Memory, & Cognition, 8, 66-72.

Schachter, S. (1951). Deviation, rejection and communication. Journal of Abnormal and Social Psychology, 46, 190-207.

Scheier, M. F. (1980). Effects of public and private self-consciousness on the public expression of personal beliefs. Journal of Personality and Social Psychology, 39, 514-521.

Schmoeller, D. (Director). (1990). The arrival [Film]. Produced by Del Mar Entertainment. Released by Prism Entertainment. 274

Schneider, D. M., & Watkins, M. J. (1996). Response conformity in recognition testing. Psychonomic Bulletin & Review, 3, 481-485.

Searcy. J., Bartlett, J. C., & Memon, A. (2000). Influence of postevent narratives, line-up conditions and individual differences on false identification of young and older eyewitnesses. Legal & Criminological Psychology, 5, 219-235.

Shaw, J. S., Garcia, L. A., & Robles, B. E. (1997). Cross-language postevent misinformation effects in Spanish-English bilingual witnesses. Journal of Applied Psychology, 82, 889-899.

Shaw, J. S., III, Garven, S., & Wood, J. M. (1997). Co-witness information can have immediate effects on eyewitness memory reports. Law & Human Behavior, 21, 503-523.

Sherif, M. (1935). A study of some social factors in perception. Archives of Psychology, 27, 1-60.

Sherif, M. (1936). The psychology of social norms. New York: Harper.

Sherif, M. (1937). An experimental approach to the study of attitudes. Sociometry, 1, 90-98.

Sistrunk, F., & McDavid, J. W. (1971). Sex variables in conforming behaviour. Journal of Personality and Social Psychology, 17, 200-207.

Sporer, S. L. (1995). Experimentally induced person mix-ups through media exposure and ways to avoid them. In G. Davies, S. Lloyd-Bostock, M. McMurran, 7 C. Wilson (Eds.), Psychology, law, and criminal justice: International developments in research and practice (pp 64-73). Berlin: Walter de Gruyter.

Stasser, G., Taylor, L. A., & Hanna, C. (1989). Information sampling in structured and unstructured discussions of three and six person groups. Journal of Personality and Social Psychology, 57, 67-78.

Stasser, G., & Titus, W. (1985). Pooling of unshared information in group decision making: Biased information sampling during discussion. Journal of Personality and Social Psychology, 58, 1467-1478.

Stasser, G., & Titus, W. (1987). Effects of information load and percentage of shared information on the dissemination of unshared information during group discussion. Journal of Personality and Social Psychology, 53, 81-93.

Stephenson, G. M. (1990). Should collaborative testimony be permitted in courts of law? The Criminal Law Review, 302-314.

275

Stephenson, G. M., Abrams, D., Wagner, W., & Wade, G. (1986). Partners in recall: Collaborative order in the recall of a police interrogation. British Journal of Social Psychology, 25, 341-343.

Stephenson, G. M., Brandstatter, H., Wagner, W. (1983). An experimental study of social performance and delay on the testimonial validity of story recall. European Journal of Social Psychology, 13, 175-191.

Stephenson, G. M., Clark, N. K., & Wade, G. S. (1986). Meetings make evidence? An experimental study of collaborative recall of a simulated police interrogation. Journal of Personality and Social Psychology, 50, 1113-1122.

Strack, F., & Bless, H. (1994). Memory for non-occurrences: Meta-cognitive and presuppositional strategies. Journal of Memory & Language, 33, 203-217.

Sutherland, R., & Hayne, H. (2001). The effect of postevent information on adults’ eyewitness reports. Journal of Applied Cognitive Psychology, 15, 249-263.

Swann, W. B., Giuliano, T., & Wegner, D. M. (1982). Where leading questions can lead: The power of conjecture in social interaction. Journal of Personality & Social Psychology, 42, 1025-1035.

Tabachnick, B. G., & Fidell, L. S. (2001). Using Multivariate Statistics (2nd ed.). New York: Harper Collins.

Tajfel, H., & Turner, J. C. (1986). The Social Identity Theory of inter-group behaviour. In S. Worchel & W.G. Austin (Eds.), Psychology of inter-group relations, 2nd edition (p. 7-24). Chicago; Nelson-Hall.

Tulving, E. (1985). Memory and consciousness. Canadian Psychology, 26(1), 1-12.

Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80, 352-373.

Tversky, B., & Tuchin, M. (1989). A reconciliation of the evidence on eyewitness testimony: Comments on McCloskey and Zaragoza. Journal of Experimental Psychology: General, 118, 86-91.

Underwood, G., & Milton, H. (1993). Collusion after a collision: Witnesses' reports of a road accident with and without discussion. Applied Cognitive Psychology, 7, 11-22.

Underwood, J., & Pezdek, K. (1998). Memory suggestibility as an example of the sleeper effect. Psychonomic Bulletin & Review, 5, 449-453.

Vicario, G. B., & Tomat, L. (1998). Testimony in a social setting. Perceptual and Motor Skills, 86, 1075-1082. 276

Walther, E., Bless, H., Strack, F., Rackstraw, P., Wagner, D., & Werth, L. (2002). Conformity effects in memory as a function of group size, dissenters, and uncertainty. Applied Cognitive Psychology, 16, 793-810.

Warnick, D. H., & Sanders, G. S. (1980). The effects of group discussion on eyewitness accuracy. Journal of Applied Social Psychology, 10, 249-259.

Weinberg, H. I., Wadsworth, J., & Baron, R. S. (1983). Demand and the impact of leading questions on eyewitness testimony. Memory & Cognition, 11, 101-104.

Weingardt, K. R., Toland, H. K. & Loftus, E. F. (1994). Reports of suggested memories: Do people truly believe them? In D. F. Ross, J. D. Read, & M. P. Toglia Eds), Adult eyewitness testimony: Current trends and developments. New York: Cambridge University Press.

Weldon, M. S., Blair, C., Huebsch, P. D. (2000). Group remembering: Does social loafing underlie collaborative inhibition? Journal of Experimental Psychology: Learning, Memory, and Cognition, 26, 1568-1577.

Weldon, M. S., & Bellinger, K. D. (1997). Collective memory: Collaborative and individual processes in remembering. Journal of Experimental Psychology: Learning, Memory, & Cognition, 23, 1160-1175.

Wells, G. L. (1978). Applied eyewitness testimony research: System variables and estimator variables. Journal of Personality and Social Psychology, 36, 1546-1557.

Wells, G. L., Lindsay, R. C., & Ferguson, T. J. (1979). Accuracy, confidence, and juror perceptions in eyewitness identification. Journal of Applied Psychology, 64, 440- 448.

Wells, G. L., Malpass, R. S., Lindsay, R. C. L., Fisher, R. P., Turtle, J. W., & Fulero, S. M. (2000). From the lab to the police station: A successful application of eyewitness research. American Psychologist, 55, 581-598.

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. Law & Human Behavior, 22 603-647.

Wheatcroft, J. M., Wagstaff, G. F., & Kebbell, M. R. The influence of courtroom questioning style on actual and perceived eyewitness confidence and accuracy. Legal & Criminological Psychology, 9, 83-101.

Wilder, D. A. (1977). Perception of groups, size of opposition and social influence. Journal of Experimental Social Psychology, 13, 253-268.

277

Windschitl, P. D. (1996). Memory for faces: Evidence of retrieval-based impairment. Journal of Experimental Psychology: Learning, Memory, & Cognition, 22, 1101- 1122.

Wittenbaum, G. M., Hubbel, A., & Zuckerman, C. (1999). Mutual enhancement: Toward an understanding of the collective preference for shared information. Journal of Personality and Social Psychology, 77, 967-978.

Wright, D. B. (1993). Misinformation and warnings in eyewitness testimony: A new testing procedure to differentiate explanations. Memory, 1, 153-166.

Wright, D. B. (1995). Misinformation methodologies: Explaining the effects of errant information. In G. Davies, S. Lloyd-Bostock, M. McMurran, 7 C. Wilson (Eds.), Psychology, law, and criminal justice: International developments in research and practice (pp 64-73). Berlin: Walter de Gruyter.

Wright, D. B., & Davies, G. M. (1999). Eyewitness testimony, In F.T. Durso (Ed.), Handbook of applied cognition (pp. 789-818). Chichester, England: John Wiley & Sons Ltd.

Wright, D. B., & Klumpp, A. (in press). Collaborative inhibition is due to the product, not the process, of recalling in groups. Psychonomic Bulletin & Review.

Wright, D. B., & Loftus, E.F. (1998). How misinformation alters memories. Journal of Experimental Child Psychology, 71, 155-164.

Wright, D. B., & Mathews, S. A. (2004). Asymmetric Effects of Memory Conformity. Manuscript submitted for publication.

Wright, D. B., & McDaid, A. T. (1996). Comparing system and estimator variables using data from real line-ups. Applied Cognitive Psychology, 10, 75-84.

Wright, D. B., Self, G., & Justice, C. (2000). Memory conformity: Exploring misinformation effects when presented by another person. British Journal of Psychology, 91, 189-202.

Wright, D. B., & Stroud, J. N. (1998). Memory quality and misinformation for peripheral and central objects. Legal & Criminological Psychology, 3, 273-286.

Yarmey, A. D. (1992). The effects of dyadic discussion on earwitness recall. Basic & Applied Social Psychology, 13, 251-263.

Yarmey, A. D., & Morris, S. (1998). The effects of discussion on eyewitness memory. Journal of Applied Social Psychology, 28(17), 1637-1648.

278

Yuille, J. C., & Daylen, J. L. (1998). The impact of traumatic events on eyewitness memory. In Thompson, C., Herrmann, D., Read, D., Bruce, D., Payne, D., & Toglia, M. (Eds.). Eyewitness memory: Theoretical and applied perspectives (pp. 155-178). New Jersey: Lawrence Earlbaum Associates.

Yuille . J. C., & Tollestrup, P. A. (1992). A model of the diverse effects of emotion on eyewitness memory. In S. Christianson (Ed.), The handbook of emotion and memory: Research and theory (pp. 201-215). Hillsdale, NJ: Lawrence Erlbaum Associates.

Yuker, H. E. (1955). Group atmosphere and memory. Journal of Abnormal & Social Psychology, 51, 17-23.

Zaragoza, M. S., & Koshmider, J. W. (1989). Misled subjects may know more than their performance implies. Journal of Experimental Psychology: Learning, Memory, & Cognition, 15, 246-255.

Zaragoza, M. S., McCloskey, M., & Jamis, M. (1987). Misleading postevent information and recall of the original event: Further evidence against the memory impairment hypothesis. Journal of Experimental Psychology: Learning, Memory, & Cognition, 13, 36-44. 279

APPENDICES

Appendix A Study 1 Screening Questionnaire for Eyewitnesses Appendix B Study 1 Follow-up Questionnaire for Eyewitnesses Appendix C Study 2 Police Officer Questionnaire Appendix D Study 3 Recognition Questionnaire Appendix E Study 4 Leading Questions Appendix F Study 4 Media Report Appendix G Study 4 Indirect Co-witness Information Appendix H Study 4 Short Answer Questionnaire Appendix I Study 4 Consent Form Appendix J Study 5 Pictures from Video A and Video B Appendix K Study 5 Recognition Questionnaire Appendix L Study 5 Lineup Appendix M Study 6 Recognition Questionnaire Appendix N Study 7 Police Interview Template Appendix A: Study 1 Screening Questionnaire for Eyewitnesses 280

Eyewitness Survey

Please complete the following details:

Name: Student No: Sex: M F

1. Have you ever witnessed any of the following events happening to you or someone else? Tick all that apply:

Homicide Property vandalism Attempted murder Breaking and entering Manslaughter Motor vehicle theft Physical assault (non-sexual) Fraud, blackmail, extortion Sexual assault Arson (fire-setting) Event/accident causing death Robbery (by physical force or serious injury or threat of force) Abduction Other similar event (please specify):

If you have ticked more than one event, please fill out the following questions with regards to the most serious or personally significant event that you witnessed. If you have not witnessed any of the above events, you do not need to complete any more of this questionnaire.

2. Did anyone else witness the event? Yes No

If yes, how many others?

3. Did police investigate the event? Yes No Don’t Know

4. Did the police interview you? Yes No

5. Did the case go to court? Yes No Don’t Know

6. Did you testify in court? Yes No

7. Approximately how long ago did this event occur?

If you are willing to be contacted about participation in research related to this questionnaire, please provide your contact information below:

Phone: E-mail:

If you have any questions about this questionnaire, please contact Helen Paterson by phone (9385 3049) or e-mail ([email protected]). 281

Eyewitness Survey

Dear Friend,

The purpose of this study is to determine how eyewitnesses cope with their experience. We, Helen Paterson and Richard Kemp, Ph.D., would be grateful if you could take a few moments to fill out this questionnaire. Your involvement in this study is voluntary and should only take about half an hour of your time. You will receive 0.5 credits for your participation. You may refuse to participate or withdraw from this study at any time without penalty. If the questionnaire has been completed and returned to us, then it will be assumed that your consent has been given.

All of the information you provide us will remain strictly confidential and will only be used for the purposes of this study. All documents will be anonymous, and we ask that you refrain from writing your name on the paper to ensure your confidentiality. We plan to publish the results in the form of group means in a peer-reviewed journal. In any publication, information will be provided in such a way that you cannot be identified.

Thinking about a traumatic event that you witnessed can be stressful. If you feel any emotional discomfort, please feel free to stop completing the questionnaire.

If you have any comments or questions about the study, please contact Helen Paterson at (02) 9385 3049. If you wish to know our findings, please write to us at the School of Psychology, University of New South Wales.

If you have any concerns about your rights or treatment as a research participant, you may contact the Ethics Secretariat, University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).

Please return the completed questionnaire in the provided envelope to Helen Paterson (Mathews 437) who will assign your participation credit. We ask that you return your completed questionnaire to Helen in person so she knows to whom she should assign credit, as the questionnaires themselves are anonymous. Thank you for your time and cooperation.

With thanks,

Helen Paterson Richard Kemp, Ph.D.

You may keep this page. 282

Eyewitness Survey

Part A: Demographic Information

1. Date of Birth (day/month/year)

2. Country of Birth:

3. Sex (circle one): M F

4. Nationality:

5. Which of the following best describes your ethnic origin? (circle): African Asian Australian Aboriginal / Torres Strait Islander European South American Other:

6. What is your current living arrangement? Circle all that apply. Live alone Live with parents Live with friend/flatmate Live with partner/spouse Live with children Other

283

Part B: Details of Witnessed Incident

*From this point forward, you will be asked questions about a serious event that you witnessed. If you have witnessed more than one event, please fill out the questions with regards to the most serious or personally significant event that you witnessed.

1. Thinking about this incident, please indicate the nature of the event you witnessed. Tick all that apply:

Homicide Property vandalism Attempted murder Breaking and entering Manslaughter Motor vehicle theft Physical assault (non-sexual) Fraud, blackmail, extortion Sexual assault Arson (fire-setting) Event/accident causing death Robbery (by physical force or serious injury or threat of force) Abduction Other similar event (please specify):

2. Briefly describe the event you witnessed.

3. How long ago did you witness this event? State the date if you remember it.

4. Which of the following would best describe your involvement in the event? Tick one:

I was an eyewitness (i.e. viewed the incident), but I was not a victim of the incident. I was an eyewitness (i.e. viewed the incident) and I was also a victim of the incident. I was a victim of the incident, but I did not witness the event.

284

5. In which state/country did this event take place?

6. At any point during the incident did you fear for your own personal safety? Circle one. 1 2 3 4 5 6 Not at all Very strongly

7. Please rate the impact this incident has had on your life.

1 2 3 4 5 6 No influence Extreme

8. With respect to the previous question, please describe the way in which the incident had an impact on your life:

9. Did the police investigate the event? Yes No Don’t Know If yes, how long after the event did they begin the investigation?

10. Did the police interview you? Yes No If yes, how long after the event did they interview you?

11. Did the case go to court? Yes No Don’t Know If yes, how long after the event did it go to court?

12. Did you testify in court? Yes No

285

Part C: Discussion with Others

1. When did you first talk to someone else about the event you witnessed? Circle one: Never ==> If never, you may skip ahead to Part D Immediately Later the same day Days later Weeks later A year later More than one year later

2. Read these instructions carefully: In the following section, we are interested in finding out about how often you discussed the incident as well as the purpose of your discussions for 3 separate time periods:

1) The first 24 hours after the event, or prior to initial police contact if there was any, 2) The week after the event 3) Six months after the event.

Each separate time period has a corresponding table for you to complete. Please make sure that you fill out the correct table for each time period as indicated by the heading of the table.

In the first blank column of each table, you will be asked to fill out “How often you discussed the event.” For this column, we would like you to write down the description from those listed (ranging from never to almost continuously) which best describes your discussion behaviour.

In the column titled, “Approximate time spent discussing the event,” we would like you to approximate the average amount of time you spent discussing the event per day. You can specify your answer in either hours/day OR minutes/day, whichever is easiest for you.

For the column titled, “Purposes of Discussions” please list any of the following purposes which apply: Emotional support: (e.g., you discussed how you felt about the event) Advice seeking: (e.g., you asked someone what they would do in your situation) Providing information: (e.g., you told someone facts about the incident for informational purposes) Story-telling: (e.g., you discussed what happened during the event to entertain others) Other: (Please specify and give a brief explanation of the purpose) 286

TABLE 1: The first 24 hours after the event, or prior to initial police contact if there was any

Person you How often you Approximate time Purposes of discussed the discussed the spent discussing the discussion event with: event. Choose 1: event (in hours/day List all that 1. Never OR minutes per day): apply: 2. Once or twice Emotional support 3. Infrequently Advice seeking 4. Occasionally Providing 5. Often information 6. Frequently Story-telling 7. Very frequently Other: Please Specify 8. Almost continuously Friends hours/day min/day Family hours/day min/day Other witnesses hours/day min/day Police Officers hours/day min/day Lawyer hours/day min/day Religious hours/day authority min/day Stranger hours/day min/day Medical doctor hours/day min/day Psychologist hours/day min/day Other (specify): hours/day min/day

287

TABLE 2: The week after the event

Person you How often you Approximate time Purposes of discussed the discussed the spent discussing the discussion event with: event. Choose 1: event (in hours/day List all that 1. Never OR minutes per day): apply: 2. Once or twice Emotional support 3. Infrequently Advice seeking 4. Occasionally Providing 5. Often information 6. Frequently Story-telling 7. Very frequently Other: Please Specify 8. Almost continuously Friends hours/day min/day Family hours/day min/day Other witnesses hours/day min/day Police Officers hours/day min/day Lawyer hours/day min/day Religious hours/day authority min/day Stranger hours/day min/day Medical doctor hours/day min/day Psychologist hours/day min/day Other (specify): hours/day min/day

288

TABLE 3: Six months after the event

Person you How often you Approximate time Purposes of discussed the discussed the spent discussing the discussion event with: event. Choose 1: event (in hours/day List all that 1. Never OR minutes per day): apply: 2. Once or twice Emotional support 3. Infrequently Advice seeking 4. Occasionally Providing 5. Often information 6. Frequently Story-telling 7. Very frequently Other: Please Specify 8. Almost continuously Friends hours/day min/day Family hours/day min/day Other witnesses hours/day min/day Police Officers hours/day min/day Lawyer hours/day min/day Religious hours/day authority min/day Stranger hours/day min/day Medical doctor hours/day min/day Psychologist hours/day min/day Other (specify): hours/day min/day 289

Part D: Discussion With Other Witnesses

1. Did anyone else witness the event? Circle one:

Yes No

If yes, how many others? (Give If no, skip the rest of Part D approximate number if unknown)

2. Did you know any of the other witnesses prior to the event? Yes No If yes, how well did you know them and what was your relationship with them?

3. At any time did you ever discuss the event with any of the other witnesses? Yes No If yes, what did you discuss?

4. Did anyone (e.g., police officer or other legal authority) ever ask you whether or not you had discussed the event with the other witnesses? Yes No If yes, please describe:

5. a. At any point during the police investigation, if any, were you encouraged to discuss the event with other witnesses? Yes No N/A If yes, please describe:

b. At any point during the police investigation, if any, were you encouraged not to discuss the event with other witnesses? Yes No N/A If yes, please describe:

290

6. a. At any point during the legal proceedings, if any, were you encouraged to discuss the event with other witnesses? Yes No N/A If yes, please describe:

b. At any point during the legal proceedings, if any, were you encouraged not to discuss the event with other witnesses? Yes No N/A If yes, please describe:

Part E: Avoidance of Discussion with Others

1. Did you ever actively avoid talking about the event with others? Yes No Is yes, why:

291

Police Questionnaire

We, Helen Paterson and Richard Kemp, Ph.D., are interested in finding out about the experiences of eyewitnesses within the legal system. We have already surveyed eyewitnesses regarding this issue, and we would like to investigate the perceptions of police officers as well. We would be grateful if you could take a few moments to fill out this questionnaire. Your involvement in this study is voluntary and should only take about ten minutes of your time. You may refuse to participate or withdraw from this study at any time. If the questionnaire has been completed and returned to us, then it will be assumed that your consent has been given.

All of the information you provide us with will remain strictly confidential and will only be used for the purposes of this study. All documents will be anonymous, and to ensure your confidentiality we ask that you refrain from writing your name on the survey. We plan to publish the results in the form of group means in a peer-reviewed journal. In any publication, information will be provided in such a way that you or your police organisation cannot be identified.

If you have any comments or questions about the study, please contact Helen Paterson at (02) 9385 3049. If you wish to know our findings, please write to us at the School of Psychology, University of New South Wales, Sydney, 2052.

If you have any concerns about your rights or treatment as a research participant, you may contact the Ethics Secretariat, University of New South Wales, Sydney 2052 Australia (phone 9385 4234, fax 9385 6648, email [email protected]).

If you would like to participate in the study, please save this survey under a different file name, fill it in, then save it again and attach it to a new email addressed to Helen Paterson ([email protected]). Alternatively, you may print the completed form and mail it to Helen Paterson, School of Psychology University of New South Wales, Sydney, 2052. Thank you for your time and cooperation.

With thanks,

Helen Paterson Richard Kemp, Ph.D.

292

Police Survey

1. Age:

2. Sex (tick one): M F

3. What is your present rank?

4. For how long have you worked for the police? Yrs mths

5. In which Local Area Command do you work?

From this point forward, you will be asked questions about events witnessed by more than one person (including the victim). Please fill out the questions in as much detail as you can.

When answering the following questions, please be completely honest. There are no right or wrong answers.

6. Think about jobs you attend in which there is more than one witness. In your estimation, what percentage of eyewitnesses have discussed details about the event with one another prior to the arrival of the police? Tick one.

10% 20% 30% 40% 50% 60% 70% 80% 90% 100%

7. What is your opinion about discussion between co-witnesses? Tick one.

1 2 3 4 5 6 Discussion can Discussion can be beneficial to be detrimental to witness memory witness memory

Please explain you opinion.

293

8a. Do you ever encourage eyewitnesses to discuss the event with one another? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

Please explain your answer.

8b. In your opinion, what is the effect of encouraging witnesses to discuss the event with one another?

9a. Do you ever discourage eyewitnesses from discussing the event with one another? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

Please explain your answer.

9b. In your opinion, what is the effect of discouraging witnesses from discussing the event with one another?

294

10a. Have you ever been given any instructions regarding discussion between eyewitnesses (e.g., either to allow or prevent the discussion)? Tick one. Yes No

If yes, please describe.

10b. Are these instructions practical? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

Please justify your opinion.

11. Can you think of any topics which eyewitnesses should discuss with one another? Tick one. Yes No

If yes, please describe.

12. Can you think of any topics which eyewitnesses should not discuss with one another? Yes No

If yes, please describe.

295

13. Do you physically separate eyewitnesses at the scene of the incident prior to taking their statements? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

14. Do you instruct eyewitnesses not to discuss the event with one another before you take their statements at the scene of the incident? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

15. Do you instruct eyewitnesses not to discuss the event with one another after you take their statements at the scene of the incident? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

16. When you know there were multiple eyewitnesses, do you ever take the statement of one eyewitness in front of other eyewitnesses at the scene of the incident? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

17. Do you ever ask witnesses whether they discussed the event with other witnesses? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

18. Do you ever convey information you learned from one eyewitness to another eyewitness and tell them the source of the information? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

19. Do you ever convey information you learned from one eyewitness to another eyewitness without telling them the source of the information? Tick one.

0 1 2 3 4 Almost Never Not often Often Very often Almost always

296

20. Can discussion between eyewitnesses help refresh their memories? Tick one.

0 1 2 3 4 Definitely not Maybe not Not sure Maybe Definitely

21. Can discussion between eyewitnesses distort their memories? Tick one.

0 1 2 3 4 Definitely not Maybe not Not sure Maybe Definitely

22. Do you have any additional comments?

Please save your completed questionnaire and e-mail it as an attachment to [email protected]. Alternatively, you may print the completed form and mail it to Helen Paterson, School of Psychology University of New South Wales, Sydney, 2052.

Thank you for taking the time to complete our survey!

297

Name Tutorial Time Sex Age

Recognition Questionnaire I-B

Instructions: In your tutorial, you saw a video of two short crime scenes. Please take a few moments to complete these questions about the scenes. Please rate the following statements according to how confidently you feel each statement is true or false.

True False 1 2 3 4 5 6 A) Store Robbery:

1) The name of the store was "Chalet Liquor and Deli." 1 2 3 4 5 6 2) One of the thieves shot and killed the storekeeper. 1 2 3 4 5 6 3) The storekeeper had a beard. 1 2 3 4 5 6 4) There were three men robbing the store. 1 2 3 4 5 6 5) The storekeeper was wearing a red and black check flannelette shirt. 1 2 3 4 5 6 6) The woman in the store was wearing a peach-coloured floral coat. 1 2 3 4 5 6 7) The thief who pointed the handgun at the lady had a tattoo on his arm. 1 2 3 4 5 6 8) The thief who pointed the handgun at the lady had blond hair. 1 2 3 4 5 6 9) The thief that shot the gun was wearing a red bandana on his head. 1 2 3 4 5 6 10) In the first scene, a black car pulls up in front of the shop. 1 2 3 4 5 6 11) The thief that shot the gun was named "Chuck." 1 2 3 4 5 6 12) The thief that shot the gun told the storekeeper, "Go straight to Hell." 1 2 3 4 5 6 13) The thief who pointed the handgun at the lady kicked her back from the counter. 1 2 3 4 5 6 14) There was an ice machine beside the door. 1 2 3 4 5 6 15) The storekeeper told the thieves to take whatever they want. 1 2 3 4 5 6 16) The woman in the store was wearing a large brooch. 1 2 3 4 5 6 17) The thief that shot the gun hit the storekeeper with the gun two times. 1 2 3 4 5 6 18) The thief who pointed the handgun at the lady was wearing a white singlet top. 1 2 3 4 5 6 19) The thief that shot the gun stole some cigarettes. 1 2 3 4 5 6 20) The woman in the store had short hair. 1 2 3 4 5 6

Have you seen the movie that this scene came from? Yes_____ No_____ ` 298

B) Street Robbery: True False

1) The woman asked the man, “What are you doing to my car?”. 1 2 3 4 5 6 2) The man who pushed the woman held a knife to her throat. 1 2 3 4 5 6 3) The woman's car was yellow. 1 2 3 4 5 6 4) The man who pushed the lady was looking in her car window. 1 2 3 4 5 6 5) The man who pushed the woman stole her bag. 1 2 3 4 5 6 6) The woman was carrying a white shopping bag. 1 2 3 4 5 6 7) The man who was first seen crossing the street was wearing a denim jacket. 1 2 3 4 5 6 8) The woman was wearing a light brown shirt. 1 2 3 4 5 6 9) The woman's purse was beige. 1 2 3 4 5 6 10) The man who pushed the woman apologized to her. 1 2 3 4 5 6 11) The man who helped the woman asked if she was alright. 1 2 3 4 5 6 12) The woman was wearing a black coat. 1 2 3 4 5 6 13) The man who was first seen crossing the street was wearing a royal blue shirt. 1 2 3 4 5 6 14) The woman's car had two doors. 1 2 3 4 5 6 15) The man who was first seen crossing the street was wearing a baseball cap. 1 2 3 4 5 6 16) The man who was first seen crossing the street was wearing white trainers. 1 2 3 4 5 6 17) The woman had a brown and beige scarf tied around her neck. 1 2 3 4 5 6 18) The man who came to help the lady was wearing a dark green shirt. 1 2 3 4 5 6 19) The man who was first seen crossing the street walked with the man that pushed the lady. 1 2 3 4 5 6 20) The black man who pushed the woman tried to smash her car window. 1 2 3 4 5 6 299

Name Sex Age

Recognition Questionnaire

Instructions: Last week, you saw a video of a store robbery. Please take a few moments to complete these questions about the crime scene. We would like you to state whether each statement is true (T) or false (F).

1) Was the thief who pointed the handgun at the woman wearing a white singlet top?

2) Was the woman in the store wearing a peach-coloured floral coat?

3) Did the thief wearing the red bandana shoot the gun?

4) Did the thief referred to as “Chuck” steal anything from the store?

5) Was the store that was robbed called the Chalet Liquor and Deli?

6) Did the thief who kicked the woman away from the counter point a gun at her?

7) Did the woman wearing the large brooch want to return her bottle?

8) Did the thief who pointed the handgun at the woman have blonde hair? 300

Media Report 301

Name Sex Age

Eyewitness Statements

Instructions: Last week, you saw a video of a store robbery. The following are statements made by other people who viewed the same video. Please indicate whether you agree (A) or disagree (D) with each statement.

1) The thief who pointed the handgun at the lady was wearing a white singlet top.

2) The woman in the store was wearing a peach-coloured floral coat.

3) The thief who shot the gun was wearing a red bandana.

4) The thief who stole from the store was referred to as “Chuck.”

5) The name of the store that was robbed was “Chalet Liquor and Deli”.

6) The thief who pointed the handgun at the lady kicked her to get her away from the counter.

7) The woman who wanted to return her bottle was wearing a large brooch.

8) The thief who pointed the handgun at the lady had blonde hair. 302

Short Answer Questionnaire

Instructions: Last week, you saw a video of a store robbery. Please take a few moments to complete these questions about the crime scene. First, we would like you to state whether each statement is true (T) or false (F). Second, we would like you to rate your confidence about this decision on the following scale:

Low 1 2 3 4 5 6 High Confidence Confidence

Question Answer Confidence 1) What was the name of the store?

2) What did the woman’s hair look like?

3) What was the storekeeper wearing?

4) What type of jewellery was the woman wearing? 5) What was the hair colour of the thief who pointed the handgun at the lady? 6) What was the colour of the bandana worn by the thief that shot the gun. 7) What colour was the car in the first scene? 8) What was the name of the thief who shot the gun? 9) What was the thief that shot the gun wearing? 10) What type of jewellery was the thief that shot the gun wearing? 11) What did the thief do to the woman to get her away from the counter? 12) What was the woman wearing?

13) What was the thief who pointed the handgun at the lady wearing? 14) What did the thief steal from the store?

303

THE UNIVERSITY OF NEW SOUTH WALES SUBJECT INFORMATION STATEMENT AND CONSENT FORM Eyewitness Memory for a Crime

You are invited to participate in a study on eyewitness memory. We, Helen Paterson and Richard Kemp, PhD., hope to learn about the accuracy and amount of information that individuals are able to recall after witnessing a crime.

If you decide to participate in this study, we will show you a video of a crime scene and then ask you some questions about it. The whole procedure will take no more than one hour, and will be spread across one 15-minute and one 45-minute session occurring approximately one week apart.

There are no known discomforts, risks, or inconveniences associated with this study. One possible benefit is that as a participant in this study, you may learn more about experiments conducted in forensic psychology. However, we cannot and do not guarantee or promise that you will receive any benefits from this study.

Any information that is obtained in connection with this study that can be identified with you will remain confidential and will be disclosed only with your permission or except as required by law. If you give us your permission by signing this document, we plan to publish the results in the form of group means in a peer-reviewed journal. In any publication, information will be provided in such a way that you cannot be identified.

Complaints may be directed to the Ethics Secretariat, University of New South Wales, SYDNEY 2052 AUSTRALIA (phone 9385 4234, fax 9385 6648, email [email protected]).

Your decision whether or not to participate will not prejudice your future relations with the University of New South Wales. If you decide to participate, you are free to withdraw your consent and to discontinue participation at any time without prejudice.

If you have any questions, please feel free to ask us. If you have any additional questions later, Helen Paterson (9385 3049) will be happy to answer them.

You will be given a copy of this form to keep. Page 1 of 2 304

THE UNIVERSITY OF NEW SOUTH WALES SUBJECT INFORMATION STATEMENT AND CONSENT FORM (continued)

Eyewitness Memory for a Crime

You are making a decision whether or not to participate. Your signature indicates that, having read the information provided above, you have decided to participate.

Signature of subject

Please PRINT name

Date

Signature of investigator

Please PRINT Name

REVOCATION OF CONSENT

I hereby wish to WITHDRAW my consent to participate in the research proposal described above and understand that such withdrawal WILL NOT jeopardise my relationship with the University of New South Wales .

Signature Date

Please PRINT Name

The section for Revocation of Consent should be forwarded to Helen Paterson at: School of Psychology, University of New South Wales, Sydney, 2052, Australia.

Page 2 of 2 305

Pictures from Version A and Version B of the Video

306

Name Tutorial Time (circle one) Th 9-11 Th 11-1 F 9-11 F 11-1 Sex Age

Recognition Questionnaire

***Please read these instructions carefully***

Instructions: Please take a few moments to complete these questions about the burglary you witnessed last week. First, we would like you to decide whether each statement is true (T) or false (F). Second, we would like to you rate your confidence about this decision on the following scale:

Low 1 2 3 4 5 6 High Confidence Confidence

Finally, if you have indicated that the statement is true, we would like you to indicate whether or not you remember the item from the video or you just know on some other basis that the item was in the video. Please read the following instructions to find out how to make the “Remember” (or “R”) and “Know” (or “K”) judgments.

Remember judgments: If your recognition of the item is accompanied by a conscious recollection of its occurrence in the video, then write “R.” “Remember” is the ability to become consciously aware again of some aspect or aspects of what happened or what was experienced at the time the video was presented (e.g., aspects of the physical appearance of the item in the video, or of what you were thinking at the time). In other words, the “remembered” statement should bring back to mind a particular association, image, or something more personal from the time of viewing the video, or something about its appearance or position.

Know judgments: “Know” responses should be made when you believe that the item was in the video, but you cannot consciously recollect anything about its actual occurrence or what happened or what was experienced at the time of its occurrence. In other words, write “K” (for “know”) when you believe the statement to be true, but it fails to evoke any specific recollection from the video.

To further clarify the difference between these two judgments (i.e., “R” vs. “K”), here are a few examples: When asked the last movie you saw, you would typically respond in the “remember” sense, that is, becoming consciously aware of the experience. On the other hand, there are times when we meet someone on the street whom appears familiar. Although we may “know” we have met this person before, we may not remember the circumstances in which we met him/her. If you have any questions regarding these judgments, please ask the experimenter.

In your discussion groups last week, some members of your group may have been shown a slightly different version of the video. Therefore, when answering these questions, you should make a specific effort to disregard what they told you and answer them solely on the basis of what you remember from the video.

307

Statement True or Confidence If true, False (1-6) Remember or Know The thief pushed the victim aside. The thief held a knife to the victim’s throat. The man who let the thief into the building was wearing a baseball cap. The crime took place during daylight hours. The victim was wearing blue jeans. While waiting outside the building, the thief was drinking from a bottle. The rescuer told the victim to call the police. The thief had a foreign accent. The thief dropped the laptop before he ran out of the apartment. The thief had a tattoo on his lower arm. The victim was wearing a silver brooch. The woman was wearing a black cardigan. The thief rang the doorbell to the flat. The thief looked through a drawer in the flat. The woman was typing on a laptop when the thief knocked on her door. The thief was wearing a watch. The rescuer had a beard. The rescuer asked the thief, “What are you doing in here?” The thief was wearing a white shirt with vertical stripes. The thief told the victim that he had come to fix some water problems. The door to the victim’s flat was red. The thief was wearing a necklace. The rescuer had long hair. The thief told the man who let him into the building that he had forgotten his keys. The thief tipped some jewellery out of its box. The woman took off her glasses before answering the door. 308

Statement True or Confidence If true, False (1-6) Remember or Know The victim was wearing a headband. The woman shouted, “Help, I’m being robbed!” The thief put something from the flat in his pocket before stealing the jewellery. The thief told the victim that he had come to check up on some security problems. The man who let the thief into the building was wearing a blue jumper. The rescuer was wearing a leather jacket.

1) Have you thought about the video at all since last week? Yes No If yes, how many times?______

2) Have you discussed the video with anyone in your tutorial class other than during the class when you were instructed to do so? Yes No If yes, how many times?______

3) Have you discussed the video with anyone from other Psychology and Law tutorials? Yes No If yes, how many times?______309

Study 5 Lineup 310

Name Tutorial (circle one) Mon 11-1 Mon 2-4 Tues 9-11 Wed 9-11 (Chris) Wed 9-11 (Helen) Sex Age

Recognition Questionnaire

Instructions: Please take a few moments to complete these questions about the burglary you witnessed last week. First, we would like you to decide whether each statement is true (T) or false (F). Second, we would like to you rate your confidence about this decision on the following scale:

Low 1 2 3 4 5 6 High Confidence Confidence

Finally, if you have indicated that the statement is true, we would like you to indicate whether you remember the item from the video only, from the discussion you had last week only, or from both the video and discussion.

• If you remember the item from the video only, indicate this by writing “V.” • If you remember the item from the discussion only, indicate this by writing “D.” • If you remember the item from both the video and the discussion, indicate this by writing “B.”

In your discussion groups last week, some members of your group may have been shown a slightly different version of the video. Therefore, when answering these questions, you should make a specific effort to disregard what they told you and answer them solely on the basis of what you remember from the video.

311 Statement True or Confidence If true, False (1-6) Video only (V), Discussion only (D), or Both (B) 1. The thief pushed the woman aside. 2. The thief held a knife to the woman’s throat. 3. The man who let the thief into the building was wearing a baseball cap. 4. The crime took place during daylight hours. 5. The woman was wearing blue jeans. 6. While waiting outside the building, the thief was drinking from a bottle. 7. The rescuer told the woman to call the police. 8. The thief had a foreign accent. 9. The thief dropped the laptop before he ran out of the apartment. 10. The thief had a tattoo on his lower arm. 11. The woman was wearing a silver brooch. 12. The woman was wearing a black cardigan. 13. The thief rang the doorbell to the flat. 14. The thief looked through a drawer in the flat. 15. The woman was typing on a laptop when the thief knocked on her door. 16. The thief was wearing a watch. 17. The rescuer had a beard. 18. The rescuer asked the thief, “What are you doing in here?” 19. The thief was wearing a white shirt with vertical stripes. 20. The thief told the woman that he had come to fix some water problems. 21. The door to the woman’s flat was red. 22. The thief was wearing a necklace. 23. The rescuer had long hair. 24. The thief told the man who let him into the building that he had forgotten his keys. 25. The thief tipped some jewellery out of its box. 26. The woman took off her glasses before answering the door. 312

Statement True or Confidence If true, False (1-6) Video only (V), Discussion only (D), or Both (B) 27. The woman was wearing a headband. 28. The woman shouted, “Help, I’m being robbed!” 29. The thief put something from the flat in his pocket before stealing the jewellery. 30. The thief told the woman that he had come to check up on some security problems. 31. The man who let the thief into the building was wearing a blue jumper. 32. The rescuer was wearing a leather jacket.

1) Have you thought about the video at all since last week? Yes No If yes, how many times?______

2) Have you discussed the video with anyone in your tutorial class other than during the class when you were instructed to do so? Yes No If yes, how many times?______

3) Have you discussed the video with anyone from other Psychology and Law tutorials? Yes No If yes, how many times?______

4) Do you think your answers were influenced by your group discussion? Not 1 2 3 4 5 6 Very at all Strongly

POLICE SERVICE

STATEMENT OF POLICE

In the matter of: Place: Date:

Name:

STATES:

1. This statement made by me accurately sets out the evidence which I would be prepared, if necessary, to give in court as a witness. The statement is true to the best of my knowledge and belief and I make it knowing that, if it is tendered in evidence, I shall be liable to prosecution if I have willfully stated in it anything which I know to be false, or do not believe to be true.

2. My age is years.

3. (Statement typed in here)

Witness: Signature: