Introduction

Imagine the following scenario: Rachel and Martin meet through a dating app, and go on a date to a local pub. At the end of the night, they go back to Martin’s house, where they have sex. Two days later, Rachel makes a complaint to police, alleging that she did not consent to the sex and that Martin sexually assaulted her. Cases such as these (often referred to as “he said/she said” cases) revolve around issues of consent in intimate sexual contexts, to which there are frequently no witnesses. In such a case, there is typically no dispute regarding the fact that sexual intercourse took place; instead, argument often centres on whether both parties were willing participants. In such a situation, the conflicting memories of the parties may be at issue, and if the case goes to trial, it is possible that a memory expert may be called to discuss the ease with which memory can be distorted1.

It is now well accepted that memory is reconstructive in nature, and that exposure to post-event misinformation can distort memory for witnessed or experienced events (the “misinformation effect”)2,3. Research into this topic is inextricably bound up with the criminal justice system, with many hundreds of studies investigating the consequences of memory distortion for eyewitnesses. This research has variously demonstrated that people can be easily led to misremember details of a crime4,5, to misidentify an innocent person as the perpetrator6 or even to remember committing a crime7. As a direct consequence of the focus of this research on averting miscarriages of justice, much eyewitness memory work to date has addressed shortcomings in the memories of those offering evidence against a defendant, including complainants and witnesses to the alleged crime. This can sometimes lead to the unfounded assumption that the memories of victims and witnesses are particularly vulnerable, and more susceptible to distortion than those on the other side of a criminal trial. This is nowhere more evident than in cases relating to sexual assault, where it often appears that it is the putative victim—and their memory—that is on trial.

To take a high-profile example, consider the accusations of sexual assault made by Dr Christine Blasey Ford against U.S. Supreme Court Justice nominee Brett Kavanaugh in 2018. The sexual encounter was alleged to have taken place more than thirty years earlier, and the parties’ recollections of the evening in question were starkly different. The majority of discussion of the case—in the Senate and in the media—revolved around questions of whether Dr Blasey Ford’s memory had been distorted by trauma, inebriation or conversations with a therapist8,9,10. Despite the fact that both parties admitted to having consumed alcohol at the party, there was very little discussion of the nature of Judge Kavanaugh’s memories, with the focus squarely on the complainant.

There are of course many reasons why the memory of a complainant might be inaccurate, including the passage of time11, consumption of alcohol12,13, and exposure to post-event misinformation14. Research has even demonstrated that witnesses may be particularly vulnerable to misinformation when the situation is self-relevant—in one study, participants were more likely to make memory errors when they were cast as the victim rather than a bystander to the incident15. There is not however any obvious reason to believe that these issues apply only the memory of the complainant, or that the memory of the accused is somehow immune.

Nevertheless, if a memory expert is called to give expert witness testimony, it is far more likely that they will have been called on behalf of the defendant, not the prosecution16,17, with the explicit or implicit goal of casting doubt on the memory of the complainant18. As the majority of the evidence relied on by such expert witnesses will have focussed on the memories of witnesses and complainants, jurors may be given the impression that only the complainant’s memory is vulnerable, with potential consequences for their decision making.

To date, there has been little to no research attempting to directly address the question of whether complainants and defendants in “he said, she said” cases differ in their susceptibility to memory distortion. In a first attempt to address this question, Greene and colleagues reported a series of experiments in which participants were cast in the role of victim or wrongdoer while playing an online version of the Prisoner’s Dilemma19. In this game, participants were asked to play against another anonymous player (“Player 2”) who they were led to believe was a real person, but who was in fact a fabrication of the researchers. Participants were told to imagine that they and Player 2 had both been accused of a crime, and that the police had offered each of them a deal if they testified against their partner. If both players chose to stay silent and cooperate with each other, they would each serve a brief prison sentence on a lesser charge, while if they both testified against each other, they would receive a longer prison sentence on the more serious charge. However, if only one player chose to testify against their partner, they would be set free while their partner would serve the full prison sentence. Participants in the game made their choice, either to testify or remain silent, and on this basis were allocated to an experimental condition. Participants who chose to remain silent were told that Player 2 had betrayed them, and that as a result the participant would suffer the consequences for the crime. Participants who chose to testify against their partner were informed that Player 2 had remained silent, and that the participant would therefore be set free while Player 2 was punished.

Prior to playing the game, participants read a brief message in which Player 2 introduced themselves and offered some thoughts about the game. Crucially, Player 2’s message did not indicate how they intended to act in the game, but post-event misinformation was subsequently offered to participants, suggesting that Player 2 had stated an intention either to testify against their partner or to remain silent. Across five experiments, participants who were thus assigned to the role of “victim” or “wrongdoer” were equally likely to accept misinformation about Player 2’s earlier statement, and did not differ in their tendency to accept misinformation that was compatible or incompatible with their account of events.

These experiments provide some initial evidence that people on either side of a single event are equally likely to suffer memory distortion for the details of that event. The Prisoner’s Dilemma game is however a very artificial scenario that offers only a very distant analogy to a real-world sexual assault case. In the present paper, we take this question one step further and investigate memories for events leading up to a sexual assault. Across three experiments, participants were exposed to a video of a date, shot from a first-person point of view, and were encouraged to take the perspective of the person on the date. Participants were then told either that they had been accused of sexually assaulting their date, or that they had accused their date of assaulting them. Thus, participants were assigned the role of accused or complainant in this fictitious “he said, she said” case.

There is copious evidence that memory reconstruction does not occur indiscriminately, but is informed by pre-existing schemas20,21. It is common for past events to be misremembered in ways that align with our biases and preconceptions22,23,24, or cast us in a good light25. We therefore expected that participants might be especially susceptible to post-event misinformation that supported their account of events; in other words, the participants assigned the role of the accused might be motivated to (mis)remember details of the event that seemed to exonerate them, while participants in the role of complainant might be motivated to remember details that seemed to incriminate the other party. Of critical interest was whether this effect might differ between the complainant and accused groups.

Study 1: misinformation distortions for complainants vs. accused

In Study 1, participants viewed a video of a simulated date and were assigned the role of either complainant or accused. They were then exposed to items of post-event misinformation that might credibly support either account before being asked to recall various details of the date. We expected to observe a misinformation effect, but crucially, we hypothesised that there would be no interaction between information groups (misinformation versus control) and role groups (accused versus complainant) in terms of the amount of misinformation reported in the eyewitness memory questionnaire.

Methods

Preregistration and ethical approval

Study 1 was preregistered at https://aspredicted.org/B9B_FY9 and received ethical approval from the Human Research Ethics Committee at University College Dublin. The experiment was conducted in line with the Code of Professional Ethics of the Psychological Society of Ireland.

Participants

Participants were recruited on the research participation website Prolific (https://www.prolific.com/) and were paid at the rate of £6 per hour. We preregistered a sample size of 500 valid responses. In total, 524 individuals took part in the study, however in line with our preregistered criteria, responses were excluded if they were incomplete, if nonsense responses were written to any of the memory questionnaire critical items or if participants failed an audio check before watching the eyewitness video. The final sample (n = 500) comprised 248 men, 248 women and 6 people who indicated their gender as non-binary/other. The age of the participants ranged from 18 to 89 (M = 39.74 years, SD = 13.17). The sample size was determined based on an a priori power analysis for a two-way ANOVA, and provided 95% power to detect small effects of f = 0.08 in the critical interaction term.

Design

The study employed a 2 × 2 mixed between/within-subjects experimental design. The within-subjects factor was exposure to misinformation, where each participant was exposed to misinformation about two details and neutral, control information about two details. The assignment of details to the misinformation and control conditions was counterbalanced. The between-subjects factor was the role participants were assigned to after watching the video of the date. Participants were randomly assigned to the complainant condition (in which they were informed that they had filed a charge against their date for sexually assaulting them) or the accused condition (in which they were informed that they had been accused of sexually assaulting their date). The dependent variable was misinformation score, defined as the number of questionnaire items for which participants recalled the misleading detail.

Materials

All study materials may be viewed online at https://osf.io/8fh4s/.

Video stimuli

Two versions of the video date were filmed, one featuring a male actor and the other featuring a female actor. The two videos were identical in all other respects. Each video was filmed from the perspective of a person going on a date with a man (“Samuel”) or a woman (“Jessica”). Only Samuel/Jessica is visible on the video, and the person behind the camera is not seen or heard at any point. The videos were approximately 7 min in duration, and screenshots are provided in Fig. 1.

Fig. 1
figure 1

Screenshots from the video of a date with Samuel (top row) or Jessica (bottom row). Left: Samuel/Jessica greets their date outside the pub. Middle: Samuel/Jessica takes a phone call from their flatmate. Right: Samuel/Jessica gives directions to the taxi driver.

Full size image

The video begins with a text screen explaining, “You are about to go on a date with [a woman called Jessica Williams/a man called Samuel Miller]. You met [Samuel/Jessica] on a dating app. This is your first in-person meeting. You have arranged to meet at a local pub.” In the first scene, we see Samuel/Jessica greeting their date with a brief hug and complimenting their outfit, before suggesting they go inside. A text screen then reads, “You both head inside the pub, where you have some drinks together and get to know one another. An hour passes…” The next scene take pace inside the pub, and shows Samuel/Jessica speaking about their workplace before taking a phone call from their flatmate. Samuel/Jessica explains that the flatmate has asked for a photo of their date, and asks permission to take a selfie with the date. This is followed by a text screen reading, “Another 2 hours pass. You and [Samuel/Jessica] are still in the pub”. Next, we see Samuel/Jessica returning from the bar with a drink in hand. They place it on the table, saying, “sorry, I just wanted to get one more”, before commenting on the lateness of the hour. Samuel/Jessica then speaks briefly about their relationship history and reason for being on dating apps. A text screen then reads, “After a little while, you and [Samuel/Jessica] leave the pub and walk to a taxi rank. It is very busy, so you decide that it will be easier to share a taxi”. The next scene shows Samuel/Jessica sitting in the back seat of a taxi giving the addresses for two stops to the taxi driver (not visible). A text screen notes that, “After a few minutes, the taxi pulls up outside [Samuel/Jessica]’s house”. In the final scene, Samuel/Jessica speaks directly to the person behind the camera, saying “you’re not too far of a walk from here. Do you want to get out here and walk the rest of the way?” Samuel/Jessica tells the taxi driver that there is no need to go to the second stop and pays him before exiting the taxi. A final text screen states that, “After you leave the taxi, you both end up going into [Samuel/Jessica’s house]. You spend the night and have sex”.

Witness statements

Participants were informed that three witnesses came forward to provide information about what they saw of the date. Three witness statements were provided, one from the security guard who saw the couple meet outside the pub, one from the bartender who witnessed the date inside the pub, and another from the taxi driver who drove them home. The witness statements alluded to four critical details from the video, and participants were randomly assigned to view misinformation about two of these details and control/neutral information about the other two (see Table 1).

Table 1 Misinformation and control details from the witness statements in Study 1, along with the associated critical question.
Full size table
Questionnaire

The memory questionnaire consisted of eight items, including the four critical items (see Table 1) and four filler items. The items were presented in random order, and responses were provided via an open text box. The full questionnaire may be found in online materials at https://osf.io/8fh4s/.

Procedure

The study was administered via a Qualtrics survey (http://www.qualtrics.com). Participants first read a consent form, which informed them that the purpose of the study was to investigate people’s perceptions and opinions of incidents preceding an alleged crime; participants were informed that they would be asked to take the perspective of someone on a date and answer some questions about their experience; there was no mention of misinformation. A content warning advised participants that the study dealt with the topic of sexual assault, and that, while the videos would not contain any graphic images or descriptions of sexual assault, they should not participate if they believed that the study was likely to cause them distress. After providing consent, participants answered some demographic questions and then selected their date. Participants were shown profile pictures of a man and woman, labelled “Samuel, 28” and “Jessica, 28”, and asked to indicate who they would prefer to go on this date with. Participants were asked to “select the person who is closest to who you would date in real life (e.g., based on your sexual orientation)”.

Participants then answered an audio check question to ensure that the sound on their device was working, and watched the video of the date. The button to progress to the next page did not appear until after the 7-minute duration of the video had elapsed. The video was followed by a set of simple arithmetic problems, to impose a delay between the original stimulus and the misinformation. Participants then received a notification regarding the accusation of sexual assault. Those in the complainant condition were told, “Two days pass, and you go to the police to file a sexual assault charge against [Samuel/Jessica]. You did not give consent to have sex with [Samuel/Jessica] and felt manipulated and forced into it by [him/her]. You also believe that you had too much to drink that night and were too drunk to consent. You felt that over the course of the date, [Samuel/Jessica] crossed boundaries and did several things which made you uncomfortable, culminating in sexual assault”. Participants in the accused condition were told, “Two days pass, and you receive a phone call to say that [Samuel/Jessica] has gone to the police to file a sexual assault charge against you. [Samuel/Jessica] is claiming that [he/she] did not give consent to have sex and that [he/she] felt manipulated and forced into it by you. [Samuel/Jessica] also claimed that [he/she] had too much to drink that night and was too drunk to consent. You know that this accusation is false as [he/she] did consent to have sex with you. You felt that [Samuel/Jessica] was extremely flirtatious with you during the date, and did several things which clearly indicated that [he/she] was comfortable taking things further”. Participants were asked to check a box to confirm that they had read this statement and that they understood that they had accused their date or been accused by their date.

Next, participants were told that some witnesses to the date had come forward, and were shown the three witness statements described above. Participants were asked to check a box to confirm that they had read each statement. After another set of filler arithmetic questions, participants completed the memory questionnaire described in Materials. Participants were explicitly instructed to answer the questions based only on their memory of the video representation of the date.

Finally, participants were fully debriefed as to the true purpose of the study, and asked to reconsent to the use of their data. No participants withdrew consent at this stage.

Results

Data coding

Responses to the critical items were coded for the presence of misinformation by two independent raters (0 = misinformation not reported; 1 = misinformation reported). Interrater reliability was excellent (k = 0.99) and conflicts were resolved by a third rater.

A total misinformation score was then calculated separately for the items where participants were shown the misinformation and for the items where participants were shown neutral information. Thus, the misinformation score variable ranged from 0, for participants who did not report misinformation for either of the critical items within the condition, to 2 for those who reported misinformation for both critical items. As the distribution of the data deviated from normality, the ANOVA results reported below were supplemented by non-parametric analyses, reported in the supplemental materials (Table S-1).

Primary analysis

A 2 × 2 mixed within-between subjects ANOVA was conducted with the factors of participant role (between complainant/accused) and misinformation exposure (misinformation/control). A significant misinformation effect was observed (F(1,498) = 276.66, p < .001, ηp2 = 0.36), such that participants reported more misinformation for items where they were misled (M = 0.70, SE = 0.03) than for control items (M = 0.12, SE = 0.02). There was no main effect of role assignment (F(1, 498) = 0.88, p = .35, ηp2 = 0.002) and no interaction effect (F(1, 498) = 0.10, p = .75, ηp2 < 0.001). Post hoc tests indicate that a large misinformation effect was observed for both the complainant (t(247) = 10.86, p < .001, d = 0.69) and accused groups (t(251) = 12.80, p < .001, d = 0.81; see Fig. 2). Equivalence testing confirmed that the difference in misinformation effect between the accused and complainant conditions was statistically equivalent to zero, using upper and lower bounds of d = -0.2 to 0.2. See Table S-2 for full details. Finally, these results were supported by non-parametric equivalents demonstrating a significant main effect of misinformation condition and no significant effect of role, in either the misled or control condition (see Table S-1).

Fig. 2
figure 2

Misinformation effect among participants assigned to the complainant and accused roles in study 1. Error bars represent standard error of the mean.

Full size image

Influence of date gender

Our primary analysis collapsed across participants who chose to go on a date with Samuel (n = 223, 44.6%) or Jessica (n = 277, 55.4%), however we acknowledge that cultural stereotypes regarding sexual assault are much more likely to include a male perpetrator and a female victim26. We therefore repeated the above analysis with the addition of date gender as a factor. The full analysis may be found in supplementary information, Table S-3. There was a significant main effect of date gender condition (F (1,496) = 26.08, p < .001, ηp2 = 0.05), such that participants tended to report more misinformation pertaining to the date with the male character (Samuel; M = 0.52, SE = 0.03) compared with the female character (Jessica; M = 0.32, SE = 0.03). There was no interaction with role group, so the tendency to report more misinformation about the male date did not differ between those in the accused and complainant groups.

Study 1 discussion

As predicted, Study 1 suggested that participants in the accused and complainant conditions were equally susceptible to the misinformation effect. However, the misinformation used in Study 1 was designed to be relatively neutral and equally applicable to both groups of participants. Previous research suggests that we are particularly prone to forming false memories for misinformation that is compatible with our biases or pre-existing beliefs24,27,28. In Study 2, we reasoned that participants who had taken on the perspective of a person on a date and then been assigned a role (as either accused or complainant) might be especially likely to accept misinformation compatible with their account of events. We therefore created separate sets of misinformation items that were compatible either with the account of the accused or the account of the complainant.

In order to examine the effect of compatible misinformation, we manipulated whether or not participants were aware of their role as accused or complainant in the study design. We expected that participants who had been informed that they had been accused of sexual assault (and had denied the charge) or that they had accused someone else would be liable to accept misinformation that was compatible with their account—i.e., information supporting the narrative that they are the innocent or victimised party. However, any differences between accused and complainant groups in their acceptance of this misinformation could simply be due to baseline differences in the plausibility of items compatible with the accused or complainant accounts. To control for this possibility, we included the unaware group, who were exposed to the same misinformation (compatible with either the complainant or accused account) but were not informed about the sexual assault accusation, and who therefore had no particular stake in the outcome. Thus, the role awareness variable manipulated the extent to which the “compatible” misinformation was in fact compatible with the participants’ account of events, since the unaware participants had no reason to think of themselves as the complainant or accused. This approach allowed us to evaluate the effect of misinformation compatibility separately for each role, and to compare this effect between the accused and complainant groups. We hypothesised that participants who were assigned the role of the complainant or the accused in a fictional sexual assault scenario would be equally likely to accept misinformation that was compatible with their account of events, when compared against participants who were not aware of that role.

Study 2: compatible misinformation and role awareness

Methods

Preregistration and ethical approval

The hypotheses, design and analysis for Study 2 are preregistered at https://aspredicted.org/R6T_888, and received ethical approval from the Human Research Ethics Committee at University College Dublin. The experiment was conducted in line with the Code of Professional Ethics of the Psychological Society of Ireland.

Participants

In total, 437 participants took part in Study 2; following exclusion of participants who did not complete the study, wrote nonsense responses, or failed an audio check or two or more attention checks, we were left with 421 valid responses, in line with our preregistered target of 420. The sample comprised 265 women, 153 men and 3 participants who selected non-binary/third gender, with a mean age of 38.05 years (SD = 12.04). This sample provided 80% power to detect effects of less than f = 0.15 in the main effect and interaction terms of the ANOVA.

Design

This study employed a 2 (role) x 2 (role awareness) x 2 (misinformation exposure) between-groups experimental design. Participants were randomly assigned to the role of complainant or accused, and then randomly assigned to be either aware or unaware of their role assignment. Finally, participants were randomly assigned to receive either compatible misinformation or neutral information about three critical details in the video. Misinformation was manipulated between subjects in Study 2 because, in contrast with Study 1, the misinformation was designed to support a particular narrative—i.e. that either “Samuel” or “Jessica” was the aggressor in the sexual encounter. Manipulating misinformation between subjects ensured that participants in the control condition were not exposed to any misinformation that might influence their memory of the encounter as a whole. A schematic of the study design may be seen in supplemental materials, Figure S-1.

Materials

All materials for this study may be found online at https://osf.io/8fh4s/.

The video stimuli used in this experiment were identical to those in Study 1, however the information presented on the text screens differed between conditions. As in Study 1, participants in the aware condition were asked to take the perspective of a person on a date and later told either that they had been accused of sexual assault, or that their date had accused them. Participants in the unaware condition were simply told that they would watch a video of a date, and were not explicitly asked to take the perspective of the person on the date. They did not receive any notification regarding an accusation of sexual assault, and all references to a sexual assault case were removed from the consent form and text screens.

As in Study 1, misinformation was delivered in the form of three witness statements (though these were referred to as ‘observer’ statements in the unaware condition, to avoid any suggestion of judicial proceedings). Misinformation was manipulated between subjects and was compatible with the participant’s assigned role. Thus, if the participant was assigned to the complainant condition, the misinformation was designed to make the sexual assault seem more likely (thereby confirming the complainant’s account). If the participant was assigned to the accused condition, the misinformation was designed to make the sexual assault seem less likely (thereby exonerating the accused). Each specific critical item of misinformation was embedded in commentary from the witness about their opinion of the date that added support to the misinformation. For example, in addition to noting that Samuel/Jessica described their date’s outfit as “sexy”, the security guard in the complainant misinformation condition also observed that the comment was inappropriate and that it appeared to make the date uncomfortable. The full witness statements may be found in online materials; Table 2 shows the critical items for the complainant and accused conditions, with the associated question from the memory test. The final memory questionnaire included all six critical items plus three filler items, presented in randomised order.

Table 2 Critical details from the witness statements and associated items from the memory questionnaire in Study 2.
Full size table

Procedure

The procedure was identical to that of Study 1, with a couple of minor exceptions. First, instead of answering arithmetic questions as in Study 1, participants answered brain teaser questions from the Cognitive Reflection Task29,30 in the delay between the video and misinformation, and misinformation and memory test. Second, to ensure that participants read the witness statements carefully, a multiple-choice question was placed at the end of each statement (e.g., “according to the taxi driver’s statement, how much was the taxi fare?”). Participants who failed two or more checks were excluded from analysis.

Results

Data coding

Responses to the critical items were coded for the presence of misinformation by two independent raters (0 = misinformation not reported; 1 = misinformation reported). Interrater reliability was excellent (k = 0.90) and conflicts were resolved by discussion. Responses that were vague or did not clearly report the misinformation were coded conservatively. For example, in response to the question “What did [Samuel/Jessica] say about your appearance and outfit when you met [him/her] outside of the pub?”, some participants reported the correct information (“cool”) and some reported the misinformation (“sexy”), but others gave answers such as “he liked how I looked” that could be interpreted either way. We coded all such answers as not having reported misinformation. A total score for misinformation reported (0–3) was then calculated separately for the items shown to participants in the complainant and accused conditions.

Primary analysis

In line with our preregistration, we conducted a 3-way ANOVA with the IVs of assignment to role (complainant, accused), awareness of role (aware, unaware) and misinformation exposure (misled, control). The dependent variable was the number of pieces of congenial misinformation reported (0–3). As shown in Table 3, there was a main effect of misinformation condition, such that participants who were exposed to misinformation reported more items of misinformation in their responses (M = 0.42, SE = 0.04) than participants in the control condition (M = 0.06, SE = 0.04). This was confirmed by non-parametric tests, shown in Table S-1.

Table 3 Results of ANOVA examining effects of role, role awareness and misinformation condition on misinformation score in Study 2.
Full size table

While participants in the accused condition reported slightly more misinformation (M = 0.28, SE = 0.04) than those in the complainant condition (M = 0.20, SE = 0.04), there was no significant effect of role or role awareness and no significant interaction effects in the ANOVA. Equivalence tests were conducted to confirm the non-significant effects of role and role awareness (see Tables S-4 and S-5 for details). While the effect of role was statistically equivalent to zero at the upper bound, the lower bound was not equivalent to zero, suggesting that the effect of role was not reliably absent. Similarly, nonparametric analyses demonstrated a borderline significant effect of role, p = .05 (see Table S-1). There was no effect of role awareness in the non-parametric tests, and equivalence tests confirmed that the difference between the aware and unaware groups was statistically equivalent to zero.

Our preregistration called for the ANOVA to be followed by equivalence tests using the two one-sided tests (TOST) procedure, in order to determine whether the misinformation effect (computed as number of pieces of misinformation recalled in misinformation condition—number of pieces of misinformation recalled in the control condition) was statistically equivalent for the complainant and accused groups. However, our preregistration neglected to account for the fact that misinformation was manipulated between subjects in the present experiment, and that we therefore could not compute the misinformation effect for each individual as planned. Instead, we conducted an exploratory analysis in which we performed separate two-way ANOVAs for the complainant and accused groups, evaluating the effect of role awareness and misinformation condition on susceptibility to misinformation.

A main effect of misinformation condition was observed in both groups, such that participants who were exposed to misinformation reported more misinformation in their responses than participants exposed to neutral information (p < .001; see Table S-6). There was no significant effect of role awareness, and no interaction effect in either group (see Fig. 3). These effects were replicated in the nonparametric analyses, Table S-1.

Fig. 3
figure 3

Number of items of congenial misinformation reported by participants assigned to the ‘complainant’ and ‘accused’ roles in Study 2.

Full size image

To further explore the apparent difference in misinformation acceptance between participants assigned to the role of accused and complainant, we examined misinformation acceptance in each individual item—see Table S-7 in the supplementary information for frequencies collapsed across role awareness conditions. The rate of misinformation acceptance varied widely within the misinformation condition, from a low of 1.4% for the detail about the taxi ride in the complainant version, to a high of 14.93% for the detail about the kiss in the accused version. These differences in baseline plausibility or acceptability of the accused and complainant-compatible details may account for the main effect of role.

Influence of date gender

In Study 2, 183 participants (43.47%) chose to go on/view a date with Jessica and 238 participants (56.53%) chose to go on/view a date with Samuel. To investigate the role of date gender, this factor was added to the primary analysis ANOVA described above. The full analysis may be found in supplementary information, Table S-8. A significant interaction between assigned role and date gender was observed (F (1,405) = 5.51, p = .02, ηp2 = 0.013), such that participants in the accused condition tended to report more misinformation about Jessica than about Samuel, while participants in the complainant condition tended to report more misinformation about Samuel than about Jessica. There was no interaction with role awareness or misinformation condition.

Study 2 discussion

Study 2 provided further evidence that participants assigned to both the “complainant” and “accused” roles are susceptible to post-event misinformation. To our surprise, there was no effect of role awareness. There was some indication that participants assigned to the role of the accused reported slightly more misinformation than those in the complainant condition. Baseline difference in the plausibility or acceptability of individual items in the accused or complainant condition witness narratives were anticipated, and were the reason for including the role-unaware control condition. Thus, in the absence of a significant interaction with role awareness or misinformation condition, any effect of role is likely an artefact of the use of different misinformation items across conditions, with some items resulting in very low rates of misinformation reporting. Further replication was however deemed necessary to rule out any true influence of assignment to the accused or complainant conditions.

In the process of coding the responses, we observed that the phrasing of some questions in combination with the open text response format sometimes resulted in responses that were vague or difficult to interpret. We coded all such answers conservatively, which may have resulted in an underestimation of misinformation acceptance in the study. In Study 3, we therefore replicated Study 2 but used multiple choice questions in the final memory test.

The primary hypothesis in this study was, once again, that complainants and accused would be equally likely to accept misinformation that was compatible with their account of events. Based on the results of Studies 1 and 2, we made some specific predictions regarding the results of Study 3:

  1. 1.

    We predicted a main effect of misinformation condition.

  2. 2.

    We made no prediction about the main effect of role, as any effect there may simply be an artefact of the specific items tested.

  3. 3.

    We did not make any specific predictions about the effect of role awareness, however we did expect that if such an effect was observed, it would be of a similar magnitude for both complainant and accused groups.

  4. 4.

    We predicted that there would be no significant three-way interaction (p > .05), indicating that people in the role of complainant and accused are equally susceptible to compatible misinformation.

Study 3: multiple-choice questions

Methods

Preregistration and ethical approval

The hypotheses, design and analysis for Study 3 were preregistered at https://aspredicted.org/4RJ_SY2, and received ethical approval from the Human Research Ethics Committee at University College Dublin. The experiment was conducted in line with the Code of Professional Ethics of the Psychological Society of Ireland.

Participants

In total, 457 participants took part in Study 3. Following exclusions (using the same criteria as in Study 2), the final sample comprised 420 valid responses, in line with our preregistered target. The sample included 209 women (49.76%), 208 men (49.52%) and 3 participants who selected non-binary/third gender, with a mean age of 37.42 years (SD = 12.06, range 18–77).

Design

This study employed a 2 (role) x 2 (role awareness) x 2 (misinformation exposure) between-groups experimental design, identical to that of Study 2, with one minor change. Participants in Study 2 were all presented with the same memory test questionnaire, which included six items designed to test memory for the critical items in the complainant condition and the accused condition. Only three of these items were actually relevant to each participant group, and it is possible that being exposed to the full set of questions may have inadvertently caused confusion. In Study 3 we remedied this issue by exposing participants to just the critical questions relevant to their condition, in addition to some filler questions.

Materials and procedure

The materials and procedure for Study 3 followed those of Study 2 with the exception that the free-response questions were replaced with multiple-choice questions. There were some minor differences in the phrasing of critical details in the witness statements, which were edited to better fit with the multiple-choice questions. The full statements may be found online at https://osf.io/8fh4s/, and the critical details and associated multiple choice questions are listed in Table S-9 in supplementary information. Each multiple-choice question had five potential answers, including the correct answer, the misinformation detail and three other incorrect answers. For example, for the question, “Which of the following terms did [Samuel/Jessica] use to describe your appearance when you first met?”, the correct answer is “cool”, the misinformation answer is “sexy” and the three other incorrect answers are “fun”, “comfortable” and “stylish”.

Results

Main analysis

The primary analysis was a 3-way ANOVA with the IVs of assignment to role (complainant, accused), awareness of role (aware, unaware) and misinformation exposure (misled, control). As shown in Table 4, there was a significant main effect of misinformation condition, such that participants who were exposed to misinformation selected the misinformation option for more items (M = 0.65, SE = 0.05) than did participants in the control condition (M = 0.03, SE = 0.05). There was no significant effect of role or role assignment, and no interaction effects. Of particular relevance to our preregistered hypotheses, the 3-way interaction between role, role awareness and misinformation condition was non-significant (p = .25). These results were replicated in the non-parametric analyses, Table S-1, and equivalence tests confirmed that the effects of role and role awareness were equivalent to zero; see Tables S-10 and S-11.

Table 4 Results of ANOVA evaluating effects of role, role awareness and misinformation condition on misinformation score in study 3.
Full size table

As outlined in our preregistration, we followed this analysis with separate 2-way ANOVAs for the complainant and accused groups, evaluating the effect of role awareness and misinformation condition on misinformation acceptance. The full analysis may be seen in Table S-12 in supplementary information, and the group means may be seen in Fig. 4. A strong misinformation effect was observed in both role groups, but there was no significant effect of role awareness and no significant interaction effect in either group. The magnitude of the non-significant interaction was very similar in both groups (η2p = 0.004 for complainants and 0.003 for accused). Once again, the same pattern was observed in the non-parametric analyses, Table S-1. Thus, we conclude that there is no difference in misinformation susceptibility for participants assigned to the complainant and accused roles.

Fig. 4
figure 4

Number of misinformation answers selected by participants in the ‘complainant’ and ‘accused’ roles in Study 3.

Full size image

Influence of date gender

In Study 3, 226 participants (53.81%) chose to go on a date/view a date with Jessica and 194 (46.19%) chose to go on a date/view a date with Samuel. To evaluate the effect of gender roles on susceptibility to misinformation, date gender was added to the primary ANOVA described above. The full analysis may be seen in supplementary information, Table S-13. There was no main effect of date gender, and no interaction with role, role awareness or misinformation condition.

General discussion

The three experiments described in this paper provide strong evidence that participants assigned to the role of complainant and accused in a fictitious sexual assault scenario are equally likely to misremember details of the events leading up the sexual encounter. In Study 1, participants in both roles received the same misinformation details, whereas Studies 2 and 3 provided misinformation that was tailored to the individual roles—that is, information designed to make the sexual assault appear more likely to have occurred (compatible with the complainant’s account) or less likely to have occurred (compatible with the accused’s account). Strong misinformation effects were observed in both groups, and these were not moderated by participant role or their awareness of that role. Interestingly, there were no consistent effects of date gender on memory, despite deeply embedded social stereotypes regarding sexual assault, which hold that men are almost invariably the perpetrators while women are the victims26. These results are in line with previous findings reported by Greene et al.,19 who used an adapted Prisoner’s Dilemma task to cast participants in the role of ‘wrongdoer’ or victim’.

The present paper represents a significant step forward in the exploration of memory distortion in a situation closely related to a real-world sexual assault. Previous evidence regarding the susceptibility of suspected perpetrators to misinformation is almost entirely lacking, with the overwhelming majority of past research reporting on memory frailty among witnesses and putative victims. The sheer weight of past evidence in this regard naturally leads to the conclusion that memory is not always reliable, and the recognition of this in courtroom settings has undoubtedly averted many miscarriages of justice. Nevertheless, it is crucial that this imbalance in the chosen subjects of eyewitness memory not be allowed to give judges, jurors, or indeed memory experts themselves the impression that complainants are uniquely vulnerable to memory distortion. The results of the present paper help to redress this balance by demonstrating that accused perpetrators are equally likely to suffer from misinformation effects.

Interestingly, there was no effect of role awareness on misinformation acceptance in Studies 2 and 3. We expected that awareness of the role would increase the misinformation effect as participants would be motivated to recall the events in a way that was compatible with their account and that reflected well on their character, in line with previous work23,24,25. This hypothesis was not supported, and awareness of one’s role did not influence susceptibility to compatible misinformation. We are however wary of concluding that parties to a sexual assault trial (fictional or otherwise) are not especially susceptible to congenial misinformation.

While participants were encouraged to take the perspective of the person on the date, and therefore to feel some sense of ownership over the role of accused or complainant, the stakes are of course much lower in this fictitious scenario than they would be in the real world, and the potential consequences of a mistaken memory much less severe. It is possible that motivated reasoning might play a greater role in memory construction in a situation where participants are personally invested in the outcome (though note that Greene et al.19 found no effects of role even when real financial rewards were at stake). Future research should therefore attempt to explore memory distortion among “complainants” and “accused” in real world settings. This presents some significant challenges from both a practical and an ethical perspective (since we cannot expose research participants to situations in which they might experience physical or emotional harm), but this is an important next step in translating this research into practice. Alternatively, future experiments could retain the fictional assault design, but attempt to manipulate or measure participants’ sense of immersion in the scenario or investment in the outcome in order to produce greater motivation to accept congenial misinformation.

A key strength of these studies was the use of first-person perspective of the videos. These were designed to elicit a sense of identification with the person going on a date with “Samuel” or “Jessica”, so that participants would enter into the experience and accept that they are now in the role of accused or complainant. This appears to have been largely successful, judging by open text responses to memory test questions, in which participants describe the events of the video in the first person (e.g., “He said I looked like I did in my photos”). However, the first-person perspective of the videos meant that we were limited in the research design that we could implement and could not directly compare response to misinformation that was compatible or incompatible with each role. As a result of the first-person perspective (in which only one character—Samuel or Jessica—is visible onscreen, the misinformation provided to participants could only refer to the actions of Samuel/Jessica as seen in the video, and not to those of the unseen date (the person in the role taken by the participant).

In initially devising these experiments, we had hoped to manipulate exposure to compatible misinformation (which supported the participant’s account) or incompatible information (which did not support their account). In the event, however, it proved virtually impossible to develop items of misinformation that were incompatible with the participant’s account. To illustrate, if a participant goes on a date with Jessica and is assigned the role of the accused, compatible information (suggesting that Jessica was not assaulted but was in fact a willing participant in the sexual activity) is relatively easy to produce; Jessica’s onscreen actions can be misleadingly described to suggest that she was flirting or sexually provocative. On the other hand, incompatible misinformation (suggesting that the participant had sexually assaulted Jessica) would really have to focus on the actions of the participant (in the role of the accused) in order to make the rape appear more likely. Because of the first-person nature of the videos, it is impossible to mislead participants regarding the actions of the accused in this scenario since we have no record of his or her actions from behind the camera. Thus, we could readily produce items of misinformation that are compatible with the participant’s role, but incompatible items are not easily generated from this stimulus set.

One solution to this issue would be to create a new set of stimuli in which both the accused and complainant are visible onscreen. After careful consideration we determined that use of the first-person perspective videos in the present set of experiments was critical to ensuring that participants took on the perspective of the person on the date, and embodied the role of accused or complainant. Nevertheless, future research should further investigate this question by designing stimuli for which the presentation of compatible and incompatible misinformation can be fully crossed. While perspective taking may be weakened in such an experimental design, converging results from both the present paper and this future research should provide convincing evidence regarding the effect of participant role in a he said/she said case.

In summary, we encourage expert witnesses to be cognisant of the fact that memory distortions are equally likely to affect both parties where memories of a given event are in conflict. In a recent paper outlining guidance for expert witness work concerning memory, victims’ memories are described as subject to weaknesses, distortions and outright fabrications31. Suspects, however, are not discussed in the same vein; rather it is noted that they may be at risk for making a false confession. We encourage further research to correct this long-standing bias in our field.