241 Legal and Criminological Psychology (2009), 14, 241–252 q 2009 The British Psychological Society The British Psychological Society www.bpsjournals.co.uk Forced confabulation more strongly influences event memory if suggestions are other-generated than self-generated Kathy Pezdek*, Shirley T. Lam and Kathryn Sperry Claremont Graduate University, Claremont, California, USA Purpose. Virtually all eyewitnesses to a crime, who eventually testify in court, are interviewed by police officers at least once. How do these interviews affect what the eyewitnesses are subsequently likely to report? The purpose of this study is to compare the relative impact of self- versus other-generated misinformation on confabulated memory about an event. Self-generated misinformation can occur by encouraging eyewitnesses to guess or speculate about possible answers to questions about which they report having no memory. Other-generated misinformation can occur by having an investigator suggest an answer to an eyewitness. Methods. After viewing a 5 minutes crime video, participants answered written questions. One week later these same questions were answered again. We specifically focused on individuals’ answers to unanswerable questions that probed information not actually presented in the video. If a participant answered an unanswerable question, we know that their answer was confabulated because the information was not presented in the video. Results. If an answer to an unanswerable question was forced confabulated at time 1, that answer was more likely to be repeated at time 2 if it had been other-generated (suggested in the question) rather than self-generated (fabricated by the participant). Conclusions. Pressuring eyewitnesses to answer questions about an event, when they indicate that they do not know the answer, can result in false confabulations. Answers suggested by the investigator are more likely to be repeated later than are answers that are simply self-generated or speculated by the eyewitness. These results are consistent with the reality monitoring framework and ‘recollect-to-reject’ metacognitive reasoning strategies. On the afternoon of September 10, 2003, Anna Lindh, the Swedish Foreign Minister, was stabbed to death while shopping in the ladies’ department of Nordiska Kompaniet department store in Stockholm. A large number of people witnessed the attack, many of whom were subsequently seated together in a room before and while they were * Correspondence should be addressed to Dr Kathy Pezdek, Department of Psychology, Claremont Graduate University, Claremont, CA 91711, USA (e-mail: Kathy.Pezdek@cgu.edu). DOI:10.1348/135532508X344773 242 Kathy Pezdek et al. interviewed by the police. There were numerous opportunities for cross-contamination of witness memory. In fact, the investigation was obstructed by the fact that the initial police bulletins contained information obtained from witnesses that was later dismissed when a video of the incident was reviewed. It turned out that many of these errors were the result of witness memory contagion. Further, photographs of the suspected perpetrator were circulated 3 days after the assassination. Granhag, Ask, and Rebelius (2005) interviewed 29 witnesses to the assassination and reported that descriptions of the perpetrator obtained after publication of the pictures were significantly more accurate (i.e. they better matched the video) than those obtained prior to publication of the pictures, evidencing another source of witness memory contagion. Numerous studies have demonstrated that eyewitnesses can be misled by post-event suggestions following an observed event (Loftus, 1975; Loftus, Miller, & Burns, 1978; Pezdek, 1977; Pezdek & Blandon-Gitlin, 2005). In most of this research, the post-event information has been other-generated, for example, the information in an interviewer’s questions about an event or a narrative that followed viewing an event. Conversations among co-witnesses is another potential source of forensically relevant other-generated suggestion (Gabbert, Memon, & Wright, 2006; Wright, Self, & Justice, 2000). A few studies have demonstrated that self-generated post-event information can also suggestively influence event memory (Hastie, Landsman, & Loftus, 1978; Schreiber, Wentura, & Bilsky, 2001; Zaragoza, Payment, Ackil, Drivdahl, & Beck, 2001). The purpose of this study is to compare the relative impact of self- versus othergenerated misinformation on confabulated memory about an event. This is an important issue because the research on other-generated post-event information is often generalized to particular real world situations that are more likely to involve self-generated post-event information. These situations include, for example, the affect of speculation on the accuracy of eyewitnesses’ memory, and the role of imagination as a catalyst for false memories. There may be important constraints on the extent to which the research on other-generated post-event information can be generalized to self-generated post-event information, yet there is little research specifically investigating this issue. Pezdek, Sperry, and Owens (2007) recently examined whether information is more likely to be incorporated into memory if it is (a) spontaneously self-generated or (b) forcibly self-generated, for example by pressing eyewitnesses to answer questions about events that they are unsure of. Participants viewed a crime video and then answered answerable and unanswerable questions. Those in the ‘voluntary guess’ condition had an ‘I don’t know’ response option, so if they answered a question, it was considered ‘volunteered’. Participants in the ‘forced guess’ condition did not have the ‘I don’t know’response option, so they were forced to provide an answer to each question. One week later the same questions were answered with an ‘I don’t know’ option available for everyone. Information self-generated from forced confabulation at time 1 was less likely to be repeated at time 2 than information that had been voluntarily self-generated. Further, when the same answer was given to an unanswerable question both times, the confidence expressed in the answer increased over time in both the forced and the voluntary guess conditions. The results of Pezdek et al. (2007) are forensically relevant because virtually all eyewitnesses to crimes, who eventually testify in court, are interviewed by police officers at least once, and more typically, multiple times (Cutler & Penrod, 1995; Poole & White, 1995). Guidelines for forensic interviewing warn police officers about the potential for suggestively influencing witness memory by the wording of their questions (see Memon & Bull, 1999; Milne & Bull, 1999, for extensive reviews of this literature). This type of suggestion – other-generated confabulation – can be induced by having an investigator Forced confabulation 243 suggest to an eyewitness what the answer to a question might be (‘The other eyewitness said that the car at the end of the alley was white. What colour do you think the car was?’). Less is known, however, about the suggestive effect of encouraging witnesses to speculate answers to questions. If encouraging eyewitnesses to guess answers or speculate about possible answers to questions about which they report having no memory leads to false confabulation, forensic interviewers should also be advised against pressing witnesses to answer questions under these circumstances; these circumstances could produce self-generated false confabulations. In this study we compared the relative impact of self- versus other-generated suggestion on memory confabulations. No data exist documenting how frequently officers in real police interviews press an eyewitness to answer questions after the eyewitness indicates that he or she does not know the answer to the question. However, Gudjonsson (1992), Kassin (1997), and Leo (1996) have reported that this is not an unusual practice. They reported that eyewitnesses are often pressed to describe events that the interviewer believed occurred, even when witnesses report that they cannot remember the event or report that they did not even witness the event. False confessions commonly occur under similar circumstances. In this study, the methodology of Pezdek et al. (2007) was replicated, however, the post-event information was either self-generated by each participant or other-generated in the questions presented. This study assesses event memory and specifically compares the relative impact of self-generated versus other-generated misinformation on subsequent recall. In most suggestibility studies, the focus is on how memory for information that was actually presented is affected by post-event suggestion. In this study, we specifically focus on individuals’ answers to unanswerable questions about information that was not actually presented in the video. If a participant answers an unanswerable question, we know that the information in his or her answer was confabulated, because it was not presented in the video. Methods Participants and design One hundred forty-four participants volunteered in eight psychology classes in the Los Angeles metropolitan area. This was a 2 (forced guess vs. voluntary guess condition) £ 2 (self- vs. other-generated) £ 2 (time 1 vs. time 2) mixed factorial design with the first two variables manipulated between participants. Materials and procedure Individuals participated with their classmates in two 15 minutes sessions, 1 week apart, in their psychology class. In the first session, participants viewed a 5 minutes video of a carjacking (produced for the study by Clark & Tunnicliff, 2001, and also the same video used by Pezdek et al., 2007). Participants were told to pay very close attention to the video because afterward they would be asked some questions. The video was followed immediately by 22 open-ended questions including 16 answerable and 6 unanswerable questions. The questions were provided in writing; participants provided answers in writing as well. The questions used in this study are identical to those used by Pezdek et al. (2007) and are included in the Appendix. Answerable questions probed information that was presented in the video; unanswerable questions probed information not presented in the video. For example, an unanswerable question is, ‘What was the logo on the 244 Kathy Pezdek et al. perpetrator’s shirt?’ and the perpetrator’s shirt did not have a logo. Questions were presented in the same order to all participants and followed the chronological order in which the relevant information appeared in the video. The first three questions were answerable questions selected from the list presented in the Appendix. The next 19 questions were a mix of answerable and unanswerable questions with no consecutive unanswerable questions. Although this study focuses only on the answers to unanswerable questions, answerable questions were included as well so that the participants would take the task seriously and not be suspicious about nature of the questions. Definition of the other-generated and self-generated conditions In the other-generated condition, all six of the unanswerable questions and six (out of the 16) answerable questions included suggested answers in the question itself. For example, an unanswerable question in the other-generated condition asked, ‘The type of soda the co-worker was drinking looked like a Coke; what do you remember as the type of soda?’ An answerable question asked, ‘The colour of the backpack the victim was carrying looked like it was purple; what do you remember the colour to be?’ For unanswerable questions, the suggested answer to each question was the most commonly reported answer to that question produced in the study by Pezdek et al. (2007). We did this to ensure that the suggested answer was not an improbable one. For answerable questions, the other-generated suggested answer was always the correct answer. In the self-generated condition, all questions were stated just as they are indicated in the Appendix. Definition of forced guess and voluntary guess conditions Participants in the forced guess condition (N ¼ 78) were provided written instructions to answer all 22 questions. They were instructed not to leave any questions unanswered, and that they should make their best guess of each answer even if they were unsure. Participants in the forced guess condition were not given an ‘I don’t know’ response option at time 1. On the other hand, participants in the voluntary guess condition (N ¼ 66) were provided written instructions to answer each question, however they were instructed that if they did not know an answer they should circle the ‘I don’t know’ response option. After each of their 22 responses, participants in all conditions also rated their confidence in the accuracy of their response on a 1 (low) to 7 (high) scale. Participants did not give confidence ratings for questions to which they responded, ‘I don’t know.’ One week later at time 2, participants were retested on their memory for the video. At time 2, participants in all four conditions were provided with the same 22 questions, but this time all participants were given the ‘I don’t know’ response option and were instructed to check this option if they did not know the answer to a question. The ‘other-generated’ suggested answers were not included on the questionnaire at time 2. Again, participants provided confidence ratings for each answer provided. Results There was one exclusion criterion. Ten participants in the forced guess condition were excluded from the analyses because they indicated ‘I don’t know’ at time 1 to at least Forced confabulation 245 one of the six unanswerable questions.1 Of these participants, five were in the selfgenerated condition and five were in the other-generated condition. Analyses were performed on the remaining 134 participants (M age ¼ 22:7 years; 51 males; 83 females). The analyses focus on the responses to unanswerable questions. Responses to these questions are especially revealing, as these are the questions that probed information that was not actually presented in the video, and thus any answer provided was confabulated. In the other-generated condition, the other-generated suggested response was provided 8.8% of the time 1 responses in the voluntary guess condition and 19.2% of the time 1 responses in the forced guess condition. Most of the responses to unanswerable questions in the voluntary guess condition were responses of ‘don’t know’ at time 1 (75.8% of all responses), and of these, 81.9% received a ‘don’t know’ response at time 2 as well. Despite the high rate of ‘don’t know’ responses in the voluntary guess and even the forced guess condition, we would argue that the results that follow have significant forensic relevance. This is first, because in police interviews, unlike in the present study in which participants simply responded to written questions, real witnesses are more likely to be pressured to answer questions; they are consequently less likely to respond that they don’t know the answer to a question. Second, even if our findings are only applicable to a minority of eyewitness cases, if these result in erroneous convictions or even the obstruction of police investigations, these are serious consequences. The primary analysis addressed the mean proportion of unanswerable questions per subject that were responded to at time 2 with the self-generated answer in the selfgenerated condition and the other-generated answer in the other-generated condition given that this same answer had been provided by the subject at time 1. The conditional probability for each of the four conditions in this study is presented in Table 1. In the study by Pezdek et al. (2007), the critical results were also analysed in terms of this conditional probability. Four of the 68 participants in the forced guess condition were not included in this analysis because they produced different answers at time 1 and time 2 to all six unanswerable questions; this analysis is thus conducted on 64 participants in the forced guess condition (35 in the self-generated condition and 29 in the othergenerated condition). In the voluntary guess condition only 31 participants (17 in the self-generated condition and 14 in the other-generated condition) gave an answer to at least one unanswerable question (i.e. they did not respond, ‘don’t know’). This is comparable to the proportion of participants in the voluntary guess condition who responded similarly (56%) in the study by Pezdek et al. (2007). A 2 (voluntary vs. forced guess condition) £ 2 (self- vs. other-generated condition) analysis of variance (ANOVA) was conducted on the data in Table 1. Only the interaction 1 In this study, we examine in each condition, the pattern of responses at time 2 given that participants provided an answer to each question at time 1. There are two ways to achieve this conditional probability. In some previous studies (Ackil & Zaragoza, 1998; Zaragoza et al., 2001) the experimenters made sure that everyone in the forced confabulation condition provided an answer to each question at time 1. Zaragoza et al. (2001), for example, reported that, ‘When participants resisted answering these questions, the experimenter prompted them to provide their best guess (repeatedly, if necessary) until they eventually acquiesced’ (p. 474). They could do this because each participant was interviewed separately. However, in our study individuals participated with other volunteers from their class. So although participants were instructed to provide a response to every question, and the experimenters walked around the room and reminded participants of this, nonetheless, some participants turned in their protocol without answering all questions. Thus, to examine the pattern of responses at time 2 given that participants provided an answer to each question at time 1, we had to exclude from the analysis participants in the forced guess condition who indicated, ‘I don’t know’ to any question at time 1. 246 Kathy Pezdek et al. Table 1. Mean proportion of unanswerable questions (and standard deviation) that were responded to at time 2 with the self-generated answer in the self-generated condition or the other-generated answer in the other-generated condition given that this same answer had been provided at time 1, in the voluntary versus forced guess conditions Voluntary guess condition (N ¼ 31) Self-generated condition Other-generated condition Forced guess condition (N ¼ 64) Self-generated answer 0.59 (0.41) Other-generated answer 0.54 (0.46) 0.33 (0.24) 0.57 (0.39) was marginally significant, Fð1; 95Þ ¼ 3:63, p ¼ :06, h2 ¼ :04. If an answer was volunteered at time 1, there was no significant difference between the probability that that answer would be repeated at time 2 as a function of whether the time 1 response had been self-generated (M ¼ 0:59) or other-generated (M ¼ 0:54), tð29Þ ¼ 0:34, p . :05, d ¼ :12. On the other hand, if the time 1 answer had been forced, that response was more likely to be repeated at time 2 if it had been an other-generated response (M ¼ 0:57) rather than a self-generated response (M ¼ 0:33), tð62Þ ¼ 3:12, p , :01, d ¼ :79. The means in the two self-generated conditions reveal the same pattern of results reported by Pezdek et al. (2007) (M ¼ 0:54 vs. 0.40, respectively). This result suggests that in responding to unanswerable questions, self-generated responses that were spontaneously volunteered at time 1 were equally likely to be repeated at time 2 as other-generated responses that were volunteered at time 1. On the other hand, when forced to guess answers to unanswerable questions, self-generated answers were less likely to be repeated at time 2 than other-generated answers. If an unanswerable question was responded to with an answer at time 1, even though the ‘I don’t know’ response option was available at time 2, that question was equally likely to be responded to with the same answer if the participant’s time 1 response was (a) othergenerated forced (M ¼ 0:57), (b) other-generated volunteered (M ¼ 0:54), or (c) selfgenerated volunteered (M ¼ 0:59). In these three conditions, if a question was answered at time 1, the modal response at time 2 was the same response given at time 1. However, for self-generated responses in the forced guess condition, if a question was answered at time 1, the modal response at time 2 was ‘I don’t know’ (M ¼ 0:48) rather than the answer that had been self-generated at time 1 (M ¼ 0:33). Analyses of confidence ratings We next assessed the change in confidence from time 1 to time 2 in the forced guess versus the voluntary guess conditions for self- versus other-generated answers to unanswerable questions. A 2 (forced guess vs. voluntary guess condition) £ 2 (self- vs. other-generated) £ 2 (time 1 vs. time 2) mixed factorial ANOVA was conducted on the confidence data (range in confidence ratings ¼ 1–7). For this analysis we specifically looked at the responses to questions for which the same self-generated answer was provided at time 1 and time 2 in the self-generated condition, and the same othergenerated answer was provided at time 1 and time 2 in the other-generated condition to see if the confidence expressed in these responses differed across conditions. These data are presented in Table 2. Note that the sample sizes across the four conditions vary considerably. Forced confabulation 247 Table 2. Mean confidence ratings (and standard deviations) for responses to unanswerable questions that were responded to at time 1 and time 2 with the self-generated answer in the self-generated condition or the other-generated answer in the other-generated condition, in the voluntary versus forced guess conditions Self-generated Voluntary guess (N ¼ 12) Time 1 Time 2 2.83 (1.99) 2.42 (1.68) Other-generated Forced guess (N ¼ 29) 2.50 (1.44) 3.07 (1.49) Voluntary guess (N ¼ 9) 4.44 (1.67) 4.11 (1.27) Forced guess (N ¼ 22) 3.03 (1.93) 3.77 (1.84) In this analysis, the other-generated responses that were consistently provided at time 1 and time 2 were more confidently held (M ¼ 3:65) than the self-generated responses (M ¼ 2:74), Fð1; 68Þ ¼ 9:27, p , :01, h2 ¼ :12. There was also a significant interaction of time by the forced versus voluntary guess condition, Fð1; 68Þ ¼ 5:09, p , :05, h2 ¼ :07. Whereas in the voluntary guess condition, confidence did not significantly change from time 1 (M ¼ 3:52) to time 2 (M ¼ 3:14), tð20Þ ¼ 0:98, p . :05, d ¼ :21, in the forced guess condition, the confidence ratings actually significantly increased from time 1 (M ¼ 2:72) to time 2 (M ¼ 3:37), tð50Þ ¼ 2:71, p ¼ :01, d ¼ :39. These results suggest that when participants repeat an answer at time 2 that had been confabulated at time 1, their confidence in this answer was higher if it had been other-generated than self-generated. In addition, when participants were forced to confabulate answers to unanswerable questions, although their confidence in the accuracy of their forced answers was initially low, after 1 week, their confidence in the answers that were repeated was significantly higher – within the confidence range expressed for voluntarily generated answers. The increase over time in the confidence expressed in repeated forced guess answers occurred for both self-generated and othergenerated responses. The analyses of responses to answerable questions, questions for which the relevant information was provided in the video, replicated the comparable effects reported by Pezdek et al. (2007). No effects involving the self-generated versus other-generated condition were predicted for responses to answerable questions, nor were any of the results involving this variable significant. Discussion The purpose of this study was to assess the relative impact of self-generated versus other-generated misinformation on event memory, specifically under the conditions in which individuals are forced to confabulate answers to unanswerable questions. The results replicate the findings reported by Pezdek et al. (2007) with self-generated postevent information and extend these findings to other-generated post-event information. The results suggest that under the conditions of the present study, other-generated forced confabulated information is more likely to be incorporated into memory than self-generated forced confabulated information. As can be seen in the right column of Table 1, if an answer to an unanswerable question was forcibly confabulated at time 1, that answer was more likely to be repeated at time 2 if it had been other-generated than 248 Kathy Pezdek et al. self-generated. Further, the other-generated answers that were repeated were more confidently held than those that were self-generated, although this was true for both the voluntary and forced guess conditions. This finding is important in light of the fact that jurors (Brigham & Bothwell, 1983; Noon & Hollins, 1987) and the courts (Manson v. Braithwaite, 1977; Neil v. Biggers, 1972/1979) often assume that confident eyewitnesses are accurate eyewitnesses. According to the primary interpretation of how post-event information affects memory, when individuals answer unanswerable questions about a target event, the answers that they provide become incorporated into their memory for the target event. On a subsequent test then, individuals are likely to make a source monitoring error and confuse the information in their previous answers with information actually observed. Source monitoring errors occur when individuals integrate new information about an event into their memory for the target event, but then do not remember that the new information did not come from the originally experienced target event ( Johnson, Hashtroudi, & Lindsay, 1993; Lindsay, 1990). Distinguishing between externally and internally generated information represents a specific type of source monitoring known as reality monitoring ( Johnson & Raye, 1981). In studies of reality monitoring, it has been reported that individuals are more accurate differentiating between one external information source and one internal information source than between two external information sources (Raye & Johnson, 1980). According to Johnson and Raye (1981), this is because externally generated memories typically include more sensory information than do internally generated memories. Thus, discriminating between an externally generated memory and an internally generated memory should be relatively easy because the former will typically be accompanied by more sensory information than the latter. However, because two externally generated memories typically cannot be differentiated on this general dimension, individuals would have to assess the content of the information and not just the sensory qualities to assess source. Because this is a more difficult task, more reality monitoring error would likely result. This would suggest in the present study that the self-generated forced confabulated information would be easier to discriminate from the information in the originally viewed target event. Thus, at time 2, the self-generated answer forced at time 1 would less likely be repeated and the ‘I don’t know’ response option would more likely be indicated. Consequently, event memory was less likely to be suggestively influenced by the self-generated information. On the other hand, the other-generated forced confabulated information would be more similar to and thus harder to discriminate from the information in the target event. The source of presented information and othergenerated information would more likely be confused and thus memory for the presented event would more likely be suggestively influenced by the other-generated information. This prediction fits the obtained pattern of results in the forced confabulation condition with unanswerable questions in the present study. Another factor that would come into play and explain why self-generated items are more memorable and less likely to be confused with presented items is the ‘recollect-toreject’ process (Brainerd, Reyna, Wright, & Mojardin, 2003; Rotello, Macmillan, & Van Tassel, 2000). According to the recollect-to-reject process, individuals use metaknowledge to reason, ‘I remember that I generated the words “Coke” when I was forced to guess, so I know that it was not something that I saw’. The generated items in the selfgenerated, forced guess condition are those to which individuals are likely to have devoted more cognitive effort. Accordingly, these are the items more likely to be Forced confabulation 249 remembered and thus more likely to lead to accurate recollect-to-reject reasoning. As a consequence, the self-generated items in the forced guess condition were less likely to be repeated at time 2. This study examined the extent to which the research on other-generated postevent information can be generalized to real world situations that involve selfgenerated post-event information. Self-generated forced confabulation might occur (a) when individuals are encouraged to imagine a traumatic event of which they appear to have no memory or (b) when eyewitnesses are pressed to describe events that the interviewer believed occurred, even when witnesses report that they cannot remember the event or report that they did not even witness the event. The results of this study suggest that although self-generated forced confabulation does occur, forced confabulations are more likely to be produced by other-generated than self-generated information. Numerous methodologies have been used to experimentally induce selfgenerated misinformation (cf. Mazzonni & Memon, 2003; Pezdek, Blandon-Gitlin, & Gabbay, 2006). Certainly there may be conditions under which it is possible to produce self-generated forced confabulations that affect memory more than some other conditions that produce other-generated forced confabulations. However, when comparable conditions were included in the present study, self-generated forced confabulations were less likely to suggestively influence event memory than othergenerated confabulations. Are generalizations from this study to forensic interviews likely to be constrained by the fact that most of the questions that we asked participants were not actually about the car-jacking itself, but rather about the victim and details from the video that preceded the crime? We would argue, no. First, from a cognitive point of view, there is no reason to suggest that forced confabulation of information would differentially affect memory for crime-relevant versus crime-neutral information. In fact, in the study by Yuille and Cutshall (1986), A Case Study of Eyewitness Memory of a Crime, real witnesses to a shooting were questioned about three types of memory that differed in forensic relevance: what occurred; person descriptions; and object descriptions. Similar patterns of results were reported across these three types of memory, some of which were crime-relevant and some crime-neutral. Similar findings were reported by Fahsing, Ask, and Granhag (2004) concerning offender descriptions by 250 witnesses to actual bank robberies. Second, it is often difficult to tell in advance what the forensically relevant details will be in any particular case. For example, in the bombing of the Murrah Federal Building in Oklahoma City in 1995, thousands of people were questioned by the FBI about whether they had seen a yellow Ryder truck on the streets of Oklahoma City. It is likely that these people had seen hundreds of cars and trucks in the days preceding the bombing. And, in the case of People of the State of California v. Sara Jane Olson, the eyewitness was linked to the case because he claimed to have remembered a female, who had accompanied a male who had purchased a particular pipe fitting seven weeks prior in his plumbing supply store. This fitting was the type that had been used in the 1975 attempted bombing of a Los Angeles police car by the Symbionese Liberation Army that claimed Patty Hearst as a member. The forensic relevance of memory for a yellow Ryder truck and memory for a specific pipe fitting would have been difficult to predict in advance. Wright et al. (2000) have made this same argument and called for research using a broader range of stimuli in eyewitness memory research. Even so, the findings in this study were based on peripheral details; additional research is necessary to assess if more central details can be changed in a similar fashion. 250 Kathy Pezdek et al. There are caveats to consider. First, there may be limits in generalizing the findings from any research setting to the real world context of forensic interviewing due to procedural and contextual differences. This includes the fact that the recalled event in this study was less emotional and less personally relevant than are the typical events recalled in police interviews. Second, because the suggestions made in this study were conveyed in written text rather than face-to-face interviews, the results of this study probably present a conservative estimate of the degree of forced confabulation that is likely to occur in real police interviews. Nonetheless, the results of this study and those of Pezdek et al. (2007) suggest that when interviewing eyewitnesses, investigators should be especially careful not to suggest information to the eyewitnesses. Pressuring eyewitnesses to answer questions when they indicate that they do not know the answer can result in false confabulation, especially if the investigator suggests a specific answer to the eyewitness. Answers suggested by the investigator are more likely to be repeated later than are answers that were simply self-generated or speculated by the eyewitness, and these other-generated answers are likely to be more confidently held. Acknowledgements We are very grateful to Steve Clark for providing us with the car-jacking video that we used in this study. This project was supported in part by a grant from the Fletcher Jones Foundation to the first author. References Ackil, J. K., & Zaragoza, M. S. (1998). Memorial consequences of forced confabulation: Age differences in susceptibility to false memories. Developmental Psychology, 34, 1358–1372. Brainerd, C. J., Reyna, V. F., Wright, R., & Mojardin, A. H. (2003). Recollection rejection: False memory editing in children and adults. Psychological Review, 110, 762–784. Brigham, J. C., & Bothwell, R. K. (1983). The ability of prospective jurors to estimate the accuracy of eyewitness identifications. Law and Human Behavior, 7, 19–30. Clark, S. E., & Tunnicliff, J. L. (2001). Selecting lineup foils in eyewitness identifications: Experimental control and real-world simulation. Law and Human Behavior, 25, 199–216. Cutler, B. L., & Penrod, S. D. (1995). Mistaken identification: The eyewitness, psychology, and the law. Cambridge: Cambridge University Press. Fahsing, I., Ask, K., & Granhag, P. A. (2004). The man behind the mask: Accuracy and predictors of eyewitness offender descriptions. Journal of Applied Psychology, 89, 722–729. Gabbert, F., Memon, A., & Wright, D. B. (2006). Memory conformity: Disentangling the steps towards influence during a discussion. Psychonomic Bulletin and Review, 13, 480–485. Granhag, P. A., Ask, K., & Rebelius, A. (2005). ‘I saw the man who killed Anna Lindh’. A case study of witnesses’ offender descriptions. Paper presented at the 2nd Meeting of the Nordic Network for Research on Psychology and Law. Kristianstad, Sweden. Gudjonsson, G. (1992). The psychology of interrogations, confessions and testimony. New York: Wiley. Hastie, R., Landsman, R., & Loftus, E. F. (1978). Eyewitness testimony: The dangers of guessing. Jurimetrics Journal, 19, 1–8. Johnson, M. K., Hashtroudi, S., & Lindsay, D. S. (1993). Source monitoring. Psychological Bulletin, 114, 3–28. Johnson, M. K., & Raye, C. L. (1981). Reality monitoring. Psychological Bulletin, 88, 67–85. Forced confabulation 251 Kassin, S. M. (1997). The psychology of confession evidence. American Psychologist, 52, 221–233. Leo, R. A. (1996). Inside the interrogation room. Journal of Criminal Law and Criminology, 86, 266–303. Lindsay, D. S. (1990). Misleading suggestions can impair eyewitnesses’ ability to remember event details. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 1077–1083. Loftus, E. F. (1975). Leading questions and the eyewitness report. Cognitive Psychology, 7, 560–572. Loftus, E. F., Miller, D. G., & Burns, H. J. (1978). Semantic integration of verbal information into visual memory. Journal of Experimental Psychology: Human Learning and Memory, 4, 19–31. Manson v. Braithwaite, 432 U.S. 98, 112, 97 S.Ct. 2243, 53 L.Ed.2d 140 (1977). Mazzonni, G., & Memon, A. (2003). Imagination can create false autobiographical memories. Psychological Science, 14, 186–188. Memon, A., & Bull, R. (Eds.), (1999). Handbook of the psychology of interviewing. Chichester: Wiley, (Published in paperback 2001). Milne, R., & Bull, R. (1999). Investigative interviewing: Psychology and practice. Chichester, UK: Wiley. Neil v. Biggers, 409 U.S. 188, 34 L.Ed.2d 401 (1972), cert. denied, 444 U.S. 909, 100 S.Ct. 221, 62 L.Ed.2d 144 (1979). Noon, E., & Hollins, C. R. (1987). Lay knowledge of eyewitness behaviour: A British survey. Applied Cognitive Psychology, 1, 143–153. Pezdek, K. (1977). Cross-modality semantic integration of sentence and picture memory. Journal of Experimental Psychology: Human Learning and Memory, 3, 515–524. Pezdek, K., & Blandon-Gitlin, I. (2005). When is an intervening lineup most likely to affect eyewitness identification accuracy? Legal and Criminological Psychology, 10, 247–263. Pezdek, K., Blandon-Gitlin, I., & Gabbay, P. (2006). Imagination and memory: Does imagining implausible events lead to false autobiographical memories? Psychonomic Bulletin and Review, 13, 764–769. Pezdek, K., Sperry, K., & Owens, S. (2007). Interviewing witnesses: The effect of forced confabulation on event memory. Law and Human Behavior, 31, 463–478. Poole, D. A., & White, L. A. (1995). Two years later: Effects on the eyewitness testimony of children and adults. Developmental Psychology, 29, 844–853. Raye, D. L., & Johnson, M. K. (1980). Reality monitoring vs. discrimination between external sources. Bulletin of the Psychonomic Society, 15, 405–408. Rotello, C. M., Macmillan, N. A., & Van Tassel, G. (2000). Recall-to-reject in recognition: Evidence from ROC curves. Journal of Memory and Language, 43, 67–88. Schreiber, N., Wentura, D., & Bilsky, W. (2001). ‘What else could he have done?’ Creating false answers in child witnesses by inviting speculation. Journal of Applied Psychology., 86, 525–532. 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. Yuille, J. C., & Cutshall, J. L. (1986). A case study of eyewitness memory of a crime. Journal of Applied Psychology, 71, 291–301. Zaragoza, M. S., Payment, K. E., Ackil, J. K., Drivdahl, S. B., & Beck, M. (2001). Interviewing witnesses: Forced confabulation and confirmatory feedback increase false memories. Psychological Science, 12, 473–477. Received 21 May 2007; revised version received 25 May 2008 252 Kathy Pezdek et al. Appendix List of the answerable and unanswerable questions used Answerable questions: (1) Describe the victim’s shirt. (2) Describe the victim’s tattoo. (3) What building material were the walls in the computer laboratory made of? (4) What colour was the mouse pad in the computer lab? (5) What time was indicated on the clock in the hallway outside the computer lab? (6) In the lab, what colour was the shirt worn by the victim’s co-worker? (7) Describe the handrail in the interior stairway. (8) What colour backpack was the victim carrying? (9) Inside the building, how many flights of stairs did the victim walk down? (10) How many floor mats were in front of the exit to the building? (11) Just outside the building was an archway. What material was this archway made of? (12) What shape were the planters in the courtyard just outside the building? (13) What traffic sign was posted at the entrance to the parking lot? (14) What type of car was stolen? (15) Describe the shorts the perpetrator was wearing. (16) Which direction did the perpetrator drive as he exited the parking lot? Unanswerable questions: (1) (2) (3) (4) (5) (6) What type of soda was the victim’s co-worker drinking in the computer lab? What was depicted in the framed picture on the wall in the computer lab? Describe the victim’s watch. How many panels did the window in the stairway have? What was in the bed of the white pick-up truck parked in the parking lot? What was the logo on the perpetrator’s shirt?