Introduction

Autism Spectrum Disorder (ASD) is a developmental disorder that is characterised by a triad of impairments in social interaction, communication and flexible imagination (American Psychiatric Association 2000). It affects how an individual makes sense of the world around them and how they communicate with and relate to other people. Memory difficulties are also well-documented in the disorder (discussed in more detail below) allowing novel predictions to be made regarding the capabilities of individuals with ASD as eyewitnesses. Moreover, a number of ‘risk’ factors indicate that people with autism may be more highly represented in the Criminal Justice System than their 1 % representation in the general population, as a victim, witness, or even perpetrator of a crime (e.g., Allen et al. 2008; Howlin 1997; Petersilia 2001; Woodbury-Smith et al. 2005, but see Woodbury-Smith et al. 2006). It is therefore important to understand not only how well individuals with ASD recall events that they have passively observed, but also how well they recall events in which they played an active role.

ASD is associated with an uneven memory profile: rather than having a poor memory per se, some functions, including semantic memory (e.g., Bowler and Gaigg 2008), recognition memory and cued recall (e.g., Bennetto et al. 1996) tend to be preserved, at least in high-functioning individuals with the disorder. Other memory abilities on the other hand, such as recalling the source of memories (e.g., Bowler et al. 2004), the spontaneous employment of organisational strategies to aid memory (e.g., Gaigg et al. 2008) and the episodic recollection of personally experienced events (e.g., Crane and Goddard 2008; Crane et al. 2009; Klein et al. 1999; Millward et al. 2000), are often reported to be diminished in ASD compared to their typical counterparts, at least when unsupported test procedures are used. Several studies have now explored how this patterning of memory affects eyewitness testimony in ASD (see Maras and Bowler in press, for a review), however the pattern of findings is mixed. Some studies report that witnesses with ASD freely recall fewer correct details (Bruck et al. 2007; Maras and Bowler 2011; Maras et al. 2012; McCrory et al. 2007; North et al. 2008, but see Maras and Bowler 2010, 2012), and some report they make more errors or are less accurate (Maras and Bowler 2011; Maras et al. 2012, but see Bruck et al. 2007; Maras and Bowler 2010, 2012; McCrory et al. 2007; North et al. 2008).

Two studies to date have explored eyewitness testimony by children with ASD using a live event, although the event in these was passively observed rather than enacted. McCrory et al. (2007) used a live classroom event and reported that whilst children with ASD freely recalled around a third less information than typically developing children did, they were no less accurate with regards to the proportion of errors or incorrect details that they reported. Bruck et al. (2007) also reported that ASD children reported fewer correct details than comparison children in response to both free recall and specific questions about a previously witnessed magic show. Of the studies with adults, none to date have used a live eyewitness event, or an event in which the witness has actively participated. This is pertinent given that it is now well established that individuals with ASD experience difficulties in reflecting on the self (e.g., Crane et al. 2009, and see Lind 2010), which extend to impairments in episodic memory (e.g., Crane and Goddard 2008; Klein et al. 1999). Indeed, a number of studies have demonstrated that individuals with ASD experience particular difficulties recalling specific and personally experienced autobiographical events (e.g., Bruck et al. 2007; Goddard et al. 2007). This impairment is in the absence of a personal semantic memory deficit (Crane and Goddard 2008), indicating that it is a deficit related to episodic memory and autonoetic awareness, rather than simply a poor memory per se (Bowler et al. 2000, 2007). These findings have led a number of researchers to suggest that deficits in self-awareness in ASD lead to impairments in episodic memory and a failure to use self-involvement to facilitate their memory (e.g., Crane et al. 2009; Klein 2001; Millward et al. 2000; Powell and Jordan 1993). If correct this has important implications for the eyewitness abilities of individuals with ASD.

The role of the self in facilitating memory in typical individuals is purported to be due to more effective encoding by use of a highly organised structure of self-concept (Symons and Johnson 1997). The self-enactment effect refers to better memory for actions that are self-performed than actions that are observed being performed by another person (e.g., Baker-Ward et al. 1990). Whilst self-concept inevitably plays a role in enhancing encoding, it has also been suggested that a self-enactment effect results from the additional motoric component of self-performed actions leading to more salient memory traces (Engelkamp 1985, Engelkamp and Zimmer 1989). Given the motor difficulties (e.g., Ming et al. 2007) coupled with diminished self-awareness in ASD (see Lind and Bowler 2010), it may come as little surprise that a number of researchers have reported a diminished or absent self-enactment effect in ASD (e.g., Dunphy-Lelii and Wellman 2012; Farrant et al. 1998; Hare et al. 2007; Millward et al. 2000; Russell and Jarrold 1999; Wojcik et al. 2011). Russell and Jarrold (1999), for example, asked children with ASD, children with moderate learning difficulties (to act as IQ matches for the ASD group) and typically developing children to remember whether they or the experimenter had placed a picture card on a grid, either on their own behalf or on behalf of a doll partner. The children with ASD were worse than both the typically developing children and children with learning disabilities at recalling whether they or the other person had placed a card onto the grid. Moreover, the children with ASD were actually worse at recalling which cards they had placed themselves compared to cards that were placed by the experimenter.

It has been argued that individuals with ASD may not benefit from memory enhancement for self-performed actions because they fail to fully integrate visual and sensori–motor signals (Wilson et al. 2007). However, Williams and Happé (2009) adapted a previously used paradigm by Russell and Hill (2001) to include a more specific test of the action monitoring deficit hypothesis. In this task, participants held a computer mouse and either moved it intentionally themselves, or kept hold of it whilst the experimenter moved it. Some of the different coloured squares on the screen moved consistently with the movements of the mouse, and some were moved randomly by the computer and thus the mouse movements on these trials were unrelated. If self-performed actions do not result in better encoding then individuals with ASD should show no difference in identifying which of the different coloured squares had moved when they intentionally moved the mouse themselves or whether the experimenter moved the mouse that they held. It was predicted that typical individuals, by contrast, who have a well-developed experienced of the self as an agent, should find the ‘other’ condition significantly more difficult. However, both groups found it easier to identify the target square when they were in control of the movement than when they simply felt the experimenter move it. In a second experiment, Williams and Happé (2009) failed to replicate Russell and Jarrold’s (1999) findings of an absent self-enactment effect and reported that the ASD group, as was the case with their typical comparisons, found it easier to monitor and recalled more of their own actions than those of another person.

Several other researchers have also reported an intact enactment effect for self- performed actions in ASD (e.g., Hare et al. 2007; Lind and Bowler 2009; Summers and Craik 1994; Zalla et al. 2010), which is problematic for an action monitoring deficit account. It is possible that the reported difficulties in utilising the role of the self in facilitating memory may be dependent on the test procedure used. Findings of intact semantic autobiographical memory but diminished personal episodic memory (e.g., Crane and Goddard 2008) and of an increased reliance on “knowing” alongside diminished autonoetic awareness as measured by “remember” responses in ASD (Bowler et al. 2000), have led Bowler and colleagues to propose the task support hypothesis (Bowler et al. 1997, 2004). According to this account, difficulties in retrieval by individuals with ASD, as evidenced by their poorer performance compared to typical individuals, are largely eliminated when more support for retrieval is provided at test, such as that in the form of cued recall or recognition tests (cf. environmental support hypothesis, see Craik and Jacoby 1996). Individuals with ASD tend not to differ from typical individuals in performance on these tests.

Consistent with the task support account, Zalla et al. (2010) reported that high-functioning individuals with ASD did not show the enactment effect for self-performed actions on tests of free recall, however they did show the effect on a recognition test, whereby both ASD and comparison groups had similarly higher correct recognition for enacted than observed items, with no difference between groups. Similarly, Hare et al. (2007) reported that low-functioning individuals with ASD did not show superior free recall for self- over other-experienced events, but that they did show enhanced recall for self-experienced events when recall was cued. Lind and Bowler (2009) also observed an enactment effect in the ASD group when they tested participants using a recognition test. In contrast to the previous suggested action monitoring deficit account, Lind and Bowler argue their findings indicate that differences observed in ASD are likely to be due to a more general episodic impairment rather than specific self-memory deficit per se. Because their task was one of recognition, Lind and Bowler suggested that the ASD group may have been able to utilise their intact semantic memory to compensate for their impaired episodic memory.

Difficulties with monitoring the source of memories are now fairly well established in the ASD memory literature (see Boucher et al. 2012, for a review). With regards to monitoring the source of self-other memories (whether the action was self-performed or performed by the experimenter), however, research to date paints a mixed picture. Some have reported diminished self-other source memory in ASD (Hala et al. 2005; Lind and Bowler 2009; Russell and Jarrold 1999), whilst others have reported that individuals with ASD perform comparably to typical participants (Farrant et al. 1998; Hill and Russell 2002; Williams and Happé 2009; Zalla et al. 2010). These discrepant findings may be attributable to two factors. First, as mentioned above and accountable by the task support hypothesis, performance in ASD tends to be diminished on free recall but unimpaired on recognition tests. Since the self-other source monitoring paradigms involve tests of recognition (“did you pick this card up or did the experimenter pick this card up?”), invariably the ASD group’s performance will often be better than had they been asked to freely recall both the action and who performed it. This type of more supportive recall test also reduces demand on cognitive load and executive functions, with which individuals with ASD are often reported to show impairments (see Hill 2004). Second, as Lind and Bowler (2009) note, because of the marked difference between “self” and “other”, self-other source judgements are more distinct and thus easier to make than internal (whether an action was imagined or actually performed) or external (which of two individuals performed an action) source judgments (Hashtroudi et al. 1989). This means that studies that have reported unimpaired self-other source monitoring may have simply lacked power or sensitivity to detect a significant difference.

The purpose of the present study is to extend this to-date inconsistent work on monitoring the source of actions and the self enactment effect in ASD using a live eyewitness scenario. No research to date has examined how well adults with ASD recall live eyewitness events in which they actively participated, which is particularly important given the literature suggesting that this might be problematic for them. Moreover, if individuals with ASD are at increased risk of victimisation (e.g., Howlin 1997; Petersilia 2001) recall of self actions and actions that others perform is critical. Thus, the aim of the present study is threefold: to examine (a) how well adults with ASD recall actually experienced eyewitness events in which they personally participated; (b) whether adults with ASD show a facilitative effect of self- over other-performed actions of an eyewitness event, as typical individuals do; (c) whether they show impaired source monitoring for who performed which actions.

Adults with ASD and their typical counterparts participated in a live eyewitness scenario whereby they assisted the experimenter in carrying out some first aid on a manikin-victim. Within this scripted scenario there were a number of actions that the experimenter always performed, and a number of actions that the participant always performed. Participants were later asked to freely recall what happened, before being questioned further about what happened. Based on the aforementioned pattern of findings to date, we predicted that the ASD group would show a diminished self-other enactment effect on the free-recall memory phase of the interview, but that there would be no difference between groups on the questioning phase, which is analogous to a cued recall test procedure. Given that free-recall provides no cues and therefore no task support for memory recall, we also expected the ASD group to make more source misattributions for whether they or the experimenter had performed the actions in their free recall, but not in the more supported questioning phase.

Method

Participants

Eighteen participants with ASD (16 males and 2 females) who were formally diagnosed by qualified clinicians were recruited predominantly in London and the South East of the UK from autism support groups and societies, and from word of mouth. All ASD participants were diagnosed by experienced clinicians with local health authorities according to DSM-IV (American Psychiatric Association 2000) criteria for Autistic Disorder or Asperger Disorder and diagnoses were confirmed for all participants by assessment with the Autism Diagnostic Observation Schedule (ADOS; Lord et al. 1999).

Eighteen comparison participants were recruited through local newspaper advertisements and comprised 15 males and 3 females who were pairwise matched within 7 points of verbal IQ as measured by the WAIS-R or WAIS-III UK (Wechsler 1997) to the ASD participants. They had no known psychiatric, developmental or neurological disorders. Groups did not significantly differ on age, VIQ, PIQ or FIQ (all ts < 1.15, ps > .26). Table 1 summarises these data. Participants also completed the Autism Spectrum Quotient (AQ, Baron-Cohen et al. 2001). None of the comparison participants exceeded the minimum cut off score for ASD of 32 (M = 16, range = 4–25), and as expected the ASD group scored significantly higher (M = 33, range = 21–45) than the comparison group on this measure, t (32) = 8.01, p < .001, Cohen’s d = 2.67. Participants provided their informed consent and were warned before taking part in the first aid scenario that they were about to see a manikin posing as an accident victim so that they were not alarmed. Ethical approval for the study was obtained from the Research Ethics Committees at City University London and Royal Holloway, University of London.

Table 1 Age and IQ scores for the ASD and comparison groups (standard deviations in parentheses)
Full size table

Materials

The first aid scenario was centred around a manikin, who was a purported car crash victim. The manikin was approximately 180 cm in height and made of flexible grey foam, with facial features but no hair or make-up. The manikin was dressed as a professional male in smart trousers, a shirt (with a red biro in the shirt pocket) and tie, and was also wearing a brown belt, wrist watch and socks. In his left trouser pocket were a set of keys, and in the right trouser pocket was a wallet with an ID membership card inside. The manikin-victim had five notable injuries: (a) wounded right hand, which was represented with fake blood on the top of the hand; (b) burn on the left forearm, again denoted using fake blood; (c) broken right arm, which was twisted up into an unusual position; (d) broken left leg, which was crooked at an odd angle; (e) a wound to the right foot, which was visible through a hole in the sock and fake blood on both the sock and the foot.

A number of first aid items were present in the same far right-hand corner of the room. A hazard triangle, foil blanket, red cotton blanket and green first aid box were all on the floor next to a table. The first aid kit contained a number of items including some bandages, band aids, a pad, scissors, tweezers, gloves and sterile swabs. On the table was a tea towel, some cling film and a fluorescent high visibility vest. The experimenter always wore the same clothes and jewellery, including a scarf which she subsequently removed during the scenario to use as a sling. An A2-size photograph print was hung on the far wall opposite the door depicting a car crash, in which two cars were severely damaged following an obvious collision.

Design and procedure

Participants were tested individually, and to avoid spontaneous context reinstatement at interview the first aid scenario was carried out in a different building from the subsequent memory interview. Participants were informed that their task was to help the experimenter carry out some first aid on a manikin–victim, and that the experimenter would instruct them on what they needed to do. On entering the room the experimenter pointed out the picture on the wall and explained that the manikin-victim had been in a car crash and had sustained a number of injuries. The first aid scenario then followed which comprised a series of actions performed by both the experimenter and participant, which were balanced to ensure that the experimenter and participant each carried out both salient and trivial actions (see “Appendix”). In order to control for any potential effects of verbal scaffolding from participants receiving instructions for their own actions, the experimenter provided the same level of verbal description of her own actions as that of the participant’s (e.g., participant-performed task: “can you check his pockets”; e.g., experimenter-performed task: “I am just going to check if he is breathing”). During the scenario there were 19 scripted actions performed by the experimenter, and 19 that were performed by the participant (e.g., Other-performed action: the experimenter gets a bandage from the first aid kit; e.g., Self-performed action: the participant rolls the bandage up). The scenario began with an initial assessment and precautionary measures (e.g., putting a hazard triangle out and checking breathing), before each of the injuries were dealt with in turn. Full details of the actions that occurred in the scenario can be found in “Appendix”.

Event scenarios were videoed through a two-way mirror for quality control purposes to ensure that they all followed the script accurately, and to double-check for any inconsistencies from participants’ reports of what happened during their interview. Following the first aid scenario participants were engaged in similar unrelated tasks for around 1 h, before being interviewed for their memory of the first aid event by a different experimenter from the one who carried out the first aid event with them.

Interview

Interviews followed the same standard structure recommended by government to professionals who interview witnesses, as outlined by the Home Office (2011) Achieving Best Evidence guidance. Interviews began with rapport building, and then the aims and structure of the interview was explained to the participant, where they were given the opportunity to ask questions. The interviewer explained that they should try to recall everything in as much detail as they could about what happened and what they could see during the first aid scenario, and that they should particularly try to recall the finer details of exactly who did what during the scenario. Participants were reminded that the interviewer did not know what happened and that their task was to describe the event as accurately as they could.

The free recall (FR) phase then followed, whereby the participant was instructed to take their time and to recall as much as they could from the scenario. Once the participant had finished speaking and was waiting for the next instruction, they were asked “can you remember anything else?” When they had responded to this and/or indicated that was all they could recall, the questioning phase began. The participant was once again instructed not to guess and that it was ok to say if they did not know the answer to any questions. Questions followed a structured sequence to probe for information pertaining to each of the actions and who performed them, and for descriptions of what things looked like. All participants were asked the same questions, but only if they had already mentioned that topic in their FR or previously in the questioning. For example, where a participant mentioned that they had put a bandage on the manikin, they were asked who had fetched the bandage. However, if no mention was made of using the bandage this question was not asked. Similarly, if a participant had mentioned that the manikin was wearing a tie they would be asked for more information about what the tie looked like. Questions were witness-compatible (i.e., using the witness’s own terminology) and were predominantly open-ended. Closed questions were kept to a minimum and leading and misleading questions were avoided. All interviews were video and audio-recorded for subsequent transcription and coding.

Coding and preliminary analyses

Interviews were transcribed and each detail that the participant mentioned was coded against a coding template of actions and descriptions of items and people that were present in the scenario. The script contained a total of 19 Self-performed and 19 Other-performed actions ("Appendix"). Details were only coded the first time they were mentioned. However, where a participant mentioned an action in free recall but failed to specify at the time who had carried out the action (e.g., “we put a bandage on”), this was followed up in the questioning phase, and their answer to this prompting (e.g., “I put the bandage on”) was coded as having been mentioned in the free recall phase.

Each detail reported was coded as “correct” (e.g., “the victim’s shirt was blue”), incorrect (e.g., “his shirt was red”), or confabulated (e.g., “the victim was wearing a hoody”), with the exception of the 19 Self- and 19 Other-performed actions, which were coded as correct or incorrect only (confabulations were scored separately from self and other details). Self errors were coded where a self-performed action was misattributed to having been performed by the experimenter, and Other errors were coded where the participant incorrectly attributed that they performed an action that was actually performed by the experimenter.

In addition to the main scripted items, there were also a number of actions that inevitably always occurred (e.g., removing the band aid from its packaging, the participant and experimenter kneeling down etc.) Actions such as these that always occurred in all participants’ scenarios were scored if they were reported correctly, but were not tagged specifically as Self or Other. Actions that occurred only in an individual participant’s scenario (i.e., those that went off-script) were not scored, in order to ensure that all participants had an equal number of potential details to score on. In addition to these action details, recall was also scored for details pertaining to people (i.e., what the manikin and the experimenter were wearing and descriptions of the people present in the photograph of the car crash), objects (e.g., the first aid items) and surroundings (e.g., what the room looked like and the location of items). For example, the statement “Anna got the red blanket from the corner and I lifted his leg and put the blanket underneath it” would be coded as 1 Other-performed action correct (got the blanket), 1 Self-performed action correct (lifted his leg), 1 Other-performed action incorrect (saying that the participant had put the blanket under the leg, when in fact it was the experimenter who had done this) and 3 correct details about the item and its location (it was red, it was a blanket and it was in the corner).

A second independent rater blindly scored eight randomly selected interview transcripts (four in each group) against the event script and the resulting Pearson’s correlations between the two raters were good for each type of detail, and importantly also for those specifically pertaining to self and other details: r correct = .97, p < .0001, r incorrect = .81, p < .05, r confabulations = .92, p < .001, r selfcorrect = .95, p < .001, r self incorrect = .90, p < .005, r other correct = .99, p < .001, r other incorrect = .92, p < .001. We also examined potential differences between groups in interview length, and there was no significant difference between the ASD and comparison groups for interview duration (ASD M = 22 min 19 s, SD = 6 min; comparison M = 19 min 48 s, SD = 4 min), t (34) = 1.36, p = .18, Cohen’s d = .51.

Results

Group differences in completeness and accuracy of recall

Our first step was to examine differences in recall between groups and between the FR and questioning phases. Details were coded as being correct, incorrect or confabulated, and three separate ANOVAs were performed for each detail type, with group as the between participants factor and interview phase as the within participants factor. As can be seen in Table 2, the ASD and comparison groups did not differ in the completeness of their recall (i.e., the number of correct details they recalled), F < 1, and there was no group x interview phase interaction, F (1, 34) = 1.62, p = .21, ηp² = .05. That is, the ASD group reported just as many correct details as their typical counterparts in both the FR and questioning phases. The ASD group did, however report significantly more incorrect details than the typical group, F (1, 34) = 11.08, p < .005, ηp² = .25, and a lack of group x interview phase interaction, F < 1, indicated that they did so in both FR and questioning phases. Analysis of confabulations indicated that groups did not significantly differ in the number of confabulations made, F (1, 34) = 3.16, p = .08, ηp² = .09, and there was no interview phases x group interaction for confabulations, F < 1. Thus, the ASD group made significantly more errors (e.g., reporting that the first aid box was on the table, when actually it was on the floor) than their typical counterparts in both FR and questioning phases, but there were no significant group differences in the tendency to confabulate about details (e.g., reporting that they performed emergency resuscitation techniques on the manikin, when in fact this did not occur at all).

Table 2 Group differences in correct, incorrect and confabulated detailed reported in each interview phase (standard deviations are in parentheses)
Full size table

There was a main effect of interview phase for the number of correct details recalled, F (1, 34) = 11.30, p < .005, ηp² = .25, whereby more correct details were recalled in the first FR phase (M = 49.32, SD = 19.41) than in the second questioning phase (M = 39.21, SD = 13.72). This finding is unsurprising given that coding was only for new items. Interview phase also had an effect on the number of incorrect details reported, F (1, 34) = 41.65, p < .001, ηp² = .55, with significantly fewer incorrect details reported in FR (M = 3.69, SD = 2.49) than in questioning (M = 8.75, SD = 4.17). A similar pattern emerged for confabulations, where more confabulations were made in questioning (M = .69, SD = .95) than in FR (M = .31, SD = .75), F (1, 34) = 4.86, p < .05, ηp² = .13.

Recall of self versus other performed actions

In order to assess the pattern of recall for Self- versus Other-performed actions, we conducted two mixed ANOVAs for correct details for each free recall and questioning phase, where group (ASD vs. Comparison) was the between-participants factor, and detail type (Self vs. Other) was the within-participants factor. There was a main effect of detail type in FR, F (1, 34) = 105.54, p < .001, ηp² = .76, but no group × detail type interaction, F (1, 34) = 1.44, p = .24, ηp² = .04. In contrast to our prediction that the self-other enactment effect would be diminished in the ASD group relative to the comparison group, both groups similarly reported more correct actions that they had performed themselves than actions that they had watched the experimenter perform in their FR (Table 3). A similar pattern emerged for Self- versus Other-performed actions in the questioning phase, however this difference fell short of traditional statistical significance levels, F (1, 34) = 3.10, p = .087, ηp² = .08, which may be related to a limited number of Self and Other details left available to be recalled following the FR phase. Again, there was not a group × detail type interaction, F < 1. These data can be seen in Table 3.

Table 3 Recall of Self- versus Other-performed actions: correct details and source confusions for each interview phase (standard deviations are in parentheses)
Full size table

Source monitoring

Next we assessed source memory for Self- and Other-performed actions. We predicted that the ASD group would make more source confusion errors on the FR, but not in the questioning phase. As reported above, both groups demonstrated the self-enactment effect by reporting significantly more self- than other-performed actions, and here we were interested in whether both groups correctly attributed themselves or the experimenter as having performed each of these types of actions. Thus, we compared differences between groups in the number of Self and Other errors made within each interview phase. As can be seen in Table 3, the ASD group made significantly more Self errors than the comparison group in the FR phase, F (1, 34) = 15.87, p < .001, ηp² = .32, but not in the questioning phase, F < 1. Groups did not differ in number of Other errors made in either the FR phases, F < 1, or in the questioning phase, F < 1. Thus, the only difference between groups in terms of source confusions (who performed which actions) was for Self-performed actions in FR. That is, compared to their typical counterparts, the ASD group incorrectly attributed more actions that they themselves had performed as having been performed by the experimenter when they were asked to freely recall what had happened. This difference between groups in source monitoring errors for self-performed actions was diminished, however, in the questioning phase, and there were no differences between groups in the number of source errors made for actions that were performed by the experimenter. Because the number of errors for self-performed actions was quite low overall, we inspected the proportions of individuals in each group who made such errors. All but one of the participants with ASD (i.e., 94 %) made at least one source error for self-performed actions. In contrast only six individuals in the comparison group (i.e., 33 %) made this type of error.

Discussion

The present study is the first to examine how well adults with ASD recall a personally experienced live eyewitness event. In recalling this event, we were interested in whether the ASD group demonstrated a self-enactment effect and whether they could successfully monitor the source of their own and another’s actions. To this end, three main findings emerged. First, the ASD group’s recall was just as complete as that of their typical comparisons, but they made more errors. Second, both ASD and comparison groups showed an enactment effect: they recalled more actions that they had performed themselves than they did actions that the experimenter had performed. Third, when asked to freely recall what happened, the ASD group made significantly more source confusions than the comparison group in attributing actions that they had actually performed themselves as having been performed by the experimenter. The use of a live event in this study is novel and demonstrates that, from a forensic perspective, if adults with ASD are personally involved in a crime as a witness, victim or perpetrator, they can recall just as many correct details as their typical counterparts, but that their recall may be less accurate (i.e., contain more errors). Moreover, if they are questioned appropriately or provided with environmental support they can provide information about source as accurately as their counterparts. We now consider each of these findings in turn in more detail.

Completeness and accuracy of recall

With regards to the eyewitness abilities of individuals with ASD in recalling a personally experienced event, findings from the present study are positive in that they show that individuals with ASD recall just as many correct details as their typical comparisons in both the FR and questioning phases of interview. However, witnesses with ASD were more prone to making errors when reporting in both stages, despite explicit interview instructions not to guess if they were unsure of anything. As mentioned in the introduction, despite the fact that all previous studies used homogeneous samples of high-functioning adults with ASD, their findings regarding both the completeness and accuracy of eyewitness testimony to date are mixed. Our findings of similar levels of completeness of the reports of witnesses with ASD to their typical counterparts are consistent with some existing studies (e.g., Maras and Bowler 2010; Maras et al. 2012) but inconsistent with others (e.g., Bruck et al. 2007; Maras and Bowler 2011; McCrory et al. 2007). It could be concluded from the present findings that individuals with ASD recall more details if the event is salient, live and personally participated in, but given the inconsistent findings previously reported, future work is needed to clarify and confirm this interpretation. Similarly, the same might be said for errors, where again our findings add to a mixed picture, with more errors reported by some studies (Maras and Bowler 2011; Maras et al. 2012), but not others (Bruck et al. 2007; Maras and Bowler 2010; McCrory et al. 2007). That the questioning phase elicited more errors than the FR phase for both groups replicates and extends to witnesses with ASD the finding of the majority of existing studies that show this effect with typical individuals.

Outside of eyewitness research, the pattern of findings remains somewhat mixed, and it is therefore difficult to pinpoint an explanation for the discrepant findings across studies. Studies of story recall by high-functioning individuals with ASD, for example, have reported both unimpaired (e.g., Ambery et al. 2006; Boucher et al. 2005) and impaired performance (Minshew and Goldstein 2001; Salmond et al. 2005; Williams et al. 2006). It is possible that high-functioning individuals are inherently impaired in recalling an event, but that they develop compensatory strategies, for example with an increased reliance on the semantic memory system, to compensate for this deficit (see Boucher et al. 2012, for further discussion), leading to a mixed pattern of findings across studies regarding both the completeness and the accuracy of their reports. Nevertheless, findings from the present study indicate that, from a forensic perspective, victims or suspects with ASD may recall just as many correct details as their typical counterparts, but that forensic professionals might seek to exercise caution in verifying the details that are given in an account by a witness with ASD.

Self-enactment effect

Contrary to our initial predictions, the ASD group showed a similar self-enactment effect to the comparison group in both their FR and questioning. That is, they recalled more actions that they had performed themselves than they did actions that the experimenter had performed. This is an important finding practically because it indicates that if an individual with ASD is the victim or perpetrator of a crime they will be able to recall what happened and what they did. Theoretically this finding is also important because it indicates that individuals with ASD do lay down a stronger memory trace for self-performed actions. Based on previous research (Hare et al. 2007; Zalla et al. 2010) and the task support hypothesis (Bowler et al. 2004) we expected that the ASD group would only demonstrate an undiminished enactment effect on the questioning phase, and not in the unsupported FR phase. It should be noted, however, that not all studies have reported a diminished self-enactment effect in FR for ASD: Summers and Craik (1994) also reported that their ASD group recalled more self-performed tasks, although the comparison task in their study was a list of word items, rather than similar action-related tasks being performed by another person. The conditions in their study could therefore not control for potential scaffolding from the nature of a visual action-related task irrespective of whether it was self-performed or not.

There are two interpretations of the present data. The first, and the interpretation that we favour, is that individuals with ASD genuinely do benefit from self-enactment and that a more general episodic deficit explains previous findings of a diminished enactment effect, whereby differences between groups are quantitative, rather than qualitative in nature (see Lind 2010). The second is that the effect is diminished in ASD, but that we failed to detect a difference in the present paradigm. We will briefly consider each of these possibilities in turn. Some have argued that action monitoring is, in fact, intact in ASD and that the difficulties observed in previous studies have reflected a more general episodic memory deficit, as opposed to a specific difficulty with personally experienced events (Lind and Bowler 2009, Williams and Happé 2009; and see Lind 2010, for a review). Our overall findings support this interpretation: if it were a deficit specific to personally experienced events, then one would expect the ASD group in the present study to recall fewer details overall than the comparison group from the personally experienced event. Such a difference should be particularly marked given that some previous eyewitness studies that have used event stimuli that were not personally participated in have reported diminished completeness of recall (e.g., Bruck et al. 2007; Maras and Bowler 2011; McCrory et al. 2007). Yet we found the ASD group recalled just as many correct details overall as their typical counterparts from this personally experienced event. Moreover, several studies have previously reported a self-enactment effect in ASD (Hare et al. 2007; Lind and Bowler 2009; Summers and Craik 1994; Williams and Happé 2009; Zalla et al. 2010), indicating that individuals with ASD do lay down a stronger memory trace for self-performed actions.

A second interpretation is that we failed to detect a diminished enactment effect because of the paradigm and/or high-functioning adult sample that took part. Henderson et al. (2009), for example, reported age-related improvements in self-referenced memory in their sample of 8- to 16-year-olds with ASD. Thus, one might argue that participants in the present study might have shown diminished self-referenced memory effects earlier in development, but had improved to such an extent by adulthood that it was not detectable in the current study, possibly also because of the narrative structure of the event. Whilst we were cautious to keep verbal commentary of the task to a minimum and ensured that it occurred equally for self- and other-performed actions, it was nevertheless inevitable in order to instruct participants which actions to perform, which might have enhanced encoding through verbal scaffolding. Williams and Happé (2009) failed to replicate Russell and Jarrold’s (1999) findings and reported that their ASD group did, in fact, demonstrate an enactment effect. Williams and Happé suggest that the discrepant findings might be related to the experimenter engaging the participant in more verbal commentary in their study than in Russell and Jarrold’s experiment. This, they argue may have led to events being encoded as self-experienced and thus recalled accurately from memory. There is some evidence to suggest that people with ASD tend not to use certain forms of inner speech (e.g., Whitehouse et al. 2006, Williams et al. 2012) and thus, Williams and Happé argue, these overt verbal descriptions of the actions at the time may have been a sort of ‘outer speech’ which served to scaffold performance. This is the major interpretation that we consider of the present data to explain the enactment effect in FR as well as questioning. Finally, it is also worth noting that whilst the experimenter was cautious to perform their actions whilst the participant attending to them, the nature of the event—designed to be more ecologically valid—means that we cannot rule out the possibility that factors relating to attention did not play a role in the enactment effect. In light of our precautions to prevent this happening, coupled with previous work showing that individuals with ASD do show a beneficial effect of self-enactment on memory, we do not believe this to be the case. It is, nevertheless, worth noting that a trade off between experimental control and ecological validity is often inevitable.

Source monitoring of self- and other-performed actions

In line with our predictions, but in contrast to the findings reported above that the ASD group did demonstrate a self-enactment effect in both their FR and in questioning is the finding that they made more source errors for self-performed actions than the comparison group in FR. The ASD group confused more actions that they had performed themselves as having been performed by the experimenter than did the comparison group. This finding is in fact in stark contrast to some previous work (e.g., Lind and Bowler 2009), which has reported the enactment effect in terms of better source monitoring for self- than other-performed actions in ASD. This is somewhat of a paradox within the present findings: Whilst individuals with ASD appear to use their self-involvement to lay down a stronger memory trace for their own actions (and hence recall more self than other actions), in free recall they are more likely than typical individuals to confuse self-performed actions as having been performed by the experimenter. If they are using their self-involvement to strengthen their memory for these actions, then it is difficult to explain why they are more likely to confuse the source of who performed them. However, the ASD group do not show a greater source monitoring confusion than their typical counterparts when questioned, and it appears to be independent of a more general source monitoring deficit, because we did not find the ASD group to be more likely to confuse the source of other-performed actions.

So how do we explain such a finding? We only observed this deficit in source monitoring in FR and not questioning, and one possibility relates to executive functioning, which can be further interpreted within the task support framework. In the FR phase, participants were instructed to recall as much as they possibly could about what happened, without any cues from the interviewer. This requires holding a large amount of information of what they could remember about the event “online” whilst simultaneously switching attention between details in order to select what to verbally recall and in what order. The instruction to report everything including the finer details of who did what increases this cognitive load on executive functions. Individuals with ASD have been reported to have a number of executive function deficits including working memory, mental flexibility (spontaneously switching attention between different thoughts or actions) and prepotent inhibition (see Hill 2004, for a review). It is also widely observed that individuals with ASD can show inaccurate production and reversal of pronouns (e.g., Jordan 1989; Lee et al. 1994; Loveland and Landry 1986; Mizuno et al. 2011). The ASD group made more source monitoring errors for self-performed actions in FR (“tell me what happened and who did what”), but not in questioning (“did Anna get the bandage or did you get the bandage?”) In contrast to FR, questioning essentially provides task support for executive functions by breaking down the information that is required for recall into smaller segments, hence directing attention to one source, reducing cognitive load and placing less demand on working memory, inhibition and set shifting. Thus, it is possible that the source monitoring errors observed in FR simply reflect executive function deficits triggering pronoun confusion, rather than a genuinely diminished enactment effect.

It is also possible that the ASD group have a more lenient response criterion than their comparisons and hence freely report more details that they are uncertain about. Of course, these are only conjectures, but it is an important area for future research to explore because if correct it has implications for forensic interviews in eliciting information from witnesses with ASD. One of the most widely-reported findings in the eyewitness literature, which police take on board in practice, is that FR is the optimal method of recalling an event to obtain the most reliable reports and that questioning, whilst eliciting more details, also results in a concomitant increase in errors (Home Office 2011; Loftus 1996). It may be the case that individuals with ASD, however, need more specific direction in interviews to focus their recall into smaller segments and minimise demands on executive functions and thus potentially reduce the likelihood of source confusion errors.

Related to an executive function account, the self-source monitoring difficulties in FR displayed by the ASD group might also be partly explained in the context of simulation theories (e.g., Gordon 1986). These theories posit that understanding of others is achieved through understanding self-representations. Meltzoff (2007), for example, reports that during development social cognition shifts from ‘others are like me’ to a dual acknowledgement that whilst others can be like us, they can also be different. Recalling who did what requires the simultaneous understanding of the differing viewpoints of others, and might account for the ASD group confusing the source of self-performed actions, whilst still benefitting from an enactment effect in recalling more of the self-performed actions themselves. This interpretation is supported by findings from O’Shea et al. (2005) who reported that, whilst unimpaired in source monitoring for impersonal items of information, the ASD group had specific difficulties in recalling the source of the person-related detail. If the distinction between self and other is less explicitly distinctive, individuals with ASD may have more specific difficulties with spontaneously teasing apart and accurately reporting the source of person-performed details on tests of FR.

Conclusions

Previous research has reported mixed findings across all three of our research questions. That is, recall of a past event, the self-enactment effect, and source monitoring. Our findings add to this mixed picture, but from our primary objective of the forensic implications regarding the abilities of individuals with ASD in recalling a previously participated-in event, our findings indicate that witnesses, victims or suspects with ASD are likely to recall just as many details as their typical counterparts. Moreover findings indicate that self-involvement boosts memory for actions that witnesses with ASD perform themselves. However, findings also suggest that forensic professionals might seek to exercise caution in verifying the accuracy of details that are given in an account by a witness with ASD. This is the first study to look at this within an eyewitness context, and findings have implications for the recall of events by victims and suspects with ASD, whom by definition would have played an active role in the event. Of course, a limitation of this study is that, in contrast to a real criminal event, it would have held relatively little emotional valence. It is therefore important for future work to extend this using a negatively valenced event, and with a larger sample. Our data also indicate that, on tests of free recall, individuals with ASD may be more likely to verbally confuse the source of self-performed actions, which may arise from executive function demands. Again, given the forensic implications, this is something that could be followed up.