Can implicit measures detect source information in crime-related amnesia?

ABSTRACT

Participants who are asked to simulate amnesia for a mock crime have a weaker memory for this event when they have to give up their role as a feigner, than those who are not asked to feign memory loss. According to the source monitoring framework (SMF), this memory-undermining effect of simulating amnesia for a crime would be due to misattribution of the right source of information. However, we know that the content of self-generated information (e.g., feigned version of the crime) might be preserved and recognised over time as a result of elaborative cognitive processing. In the present study, we aimed to contrast these two explanations. We showed participants a mock crime video and we instructed them to either feign amnesia (simulators) or confess the mock crime (confessors). Next, a free recall memory test was administered. After one week, participants were asked to perform a personalised source monitoring task using the autobiographical Implicit Association Test (aIAT). As predicted, we found that simulators were able to discriminate the content of their self-generated feigned story of the crime from the original version. Moreover, simulators were quicker than confessors at the aIAT task. Practical and theoretical implications of our results are discussed.

KEYWORDS:

• simulating amnesia
• crime-related amnesia
• source monitoring
• autobiographical implicit association test
• implicit memory

A nontrivial minority of criminal offenders claim amnesia for their deeds (Cima, Merckelbach, Nijman, Knauer, & Hollnack, 2002; Pyszora, Barker, & Kopelman, 2003). Even though genuine crime-related amnesia is not impossible, it seems that most offenders feign their memory loss (Christianson & Merckelbach, 2004). Because of crime-related amnesia, a defendant might be found not completely responsible for a crime or may be rendered incompetent to stand trial (Smith & Resnick, 2007). Therefore, determining the authenticity of memory loss is legally relevant (Giger, Merten, Merckelbach, & Oswald, 2010; Merckelbach & Christianson, 2007).

Although feigning memory loss for a crime is sometimes difficult to determine, a large body of studies supports the idea that pretending to have amnesia impairs an offender’s memory for the crime (the so-called memory-undermining effect of simulating amnesia; Christianson & Bylin, 1999). That is, individuals who were instructed to simulate amnesia for a mock crime on an initial memory test (further referred to as simulators) differed from participants who were instructed to respond honestly (further referred to as confessors) when some time later they were asked to give up simulating and honestly report details about the crime (Bylin & Christianson, 2002; Christianson & Bylin, 1999; Van Oorsouw & Merckelbach, 2004, 2006). Thus, pretending to suffer from memory loss for an offence appears to undermine recollection of that crime.

Source monitoring deficit as an explanation for the simulating amnesia effect

A possible explanation for the memory-undermining effect of feigning amnesia for a mock crime has to do with the source monitoring framework (SMF; Johnson, Hashtroudi, & Lindsay, 1993; Mitchell & Johnson, 2009). As suggested by Christianson and Bylin (1999), simulating amnesia may cause distorted information to be incorporated into the offender’s memory of the crime, thereby leading to memory distortion. In other words, when simulators are asked to feign amnesia for a mock crime they come up with a different (self-generated) story, which might be confused with the original story causing source monitoring errors. Such imagining may confuse memory for what was originally experienced, by increasing qualitative details that are typical of a real event (Johnson et al., 1993; Johnson, Raye, Foley, & Foley, 1981). Previous research has also demonstrated that some offenders might be prompted by vivid memories of their violent acts by repeatedly thinking of the crime that they perpetrated or how it could have been prevented (Evans, 2006; Evans, Ehlers, Mezey, & Clark, 2007). This type of prompt includes a counterfactual thinking (CFT) – reflecting on how past events might have occurred differently (Byrne, 2005, 2016; Roese & Epstude, 2017; Roese & Olson, 1997). For that reason, simulators might increase source confusion for their deeds by thinking how they could have avoided the crime if they would have acted differently. Source confusion might thus increase when imagined events resemble real events (Johnson et al., 1993).

Alternative explanation to the source monitoring paradigm

One could however argue that simulating amnesia might implicate a sort of active elaboration of the memory for the crime (McWilliams, Goodman, Lyons, Newton, & Avila-Mora, 2014), so that individuals who feign memory loss are driven in their simulation by their ideas about how memory loss works (Bylin, 2002). As a matter of fact, previous studies have shown that participants feign amnesia by omitting all information (Iverson, 1995), exaggerating memory impairment (Baker, Hanley, Jackson, Kimmance, & Slade, 1993), or withholding some information while reporting other or distorted information (simulation by omission vs. simulation by commission; Bylin & Christianson, 2002; Van Oorsouw & Merckelbach, 2004). Germane to these findings, it has also been demonstrated that the content of self-generated information (e.g., the simulated version of the crime) might remain relatively clear as a result of elaborative cognitive processing (Chrobak & Zaragoza, 2008; Johnson, Foley, Suengas, & Raye, 1988). Indeed, Chrobak and Zaragoza (2008) showed that individuals had preserved memory of a self-generated event and, one week later, they were unlikely to confuse the same event with another one. Consequently, although feigning amnesia may cause distorted memories for a crime, simulators might still keep in mind which information is self-generated and belongs to their own feigned story and which one belongs to the actual crime. As a result of elaborative cognitive processing, feigners might preserve fictitious crime-related information in order to stick with their personal version of the facts instead of being confused by what they formerly claimed. This effect would take place, for instance, in specific situations where the simulation is repeated, such as during police interrogations.

Hence, in the present study, we aimed to contrast the source monitoring explanation for the feigning amnesia effect (Christianson & Bylin, 1999), to the possibility of distinguishing the content of self-generated information by simulators. Therefore, based on the line of reasoning depicted above, the main research question of the present study was: Are simulators, despite previously feigning amnesia, still able to identify the source of information during a source monitoring task?

Investigating source monitoring ability via explicit vs. implicit task

So far, the SMF explanation in relation to simulated amnesia has been investigated by comparing simulators’ memory errors (i.e., other or distorted information) reported in their free and cued recall at the first and second memory test to errors of participants who were asked to respond honestly (e.g., Christianson & Bylin, 1999; Van Oorsouw & Merckelbach, 2004). Although simulators were encouraged to come up with a self-generated version of the crime at the first memory assessment, this request did not result into a heightened percentage of errors at the second assessment – suggesting a preserved source-monitoring capacity for the crime (Christianson & Bylin, 1999; Van Oorsouw & Merckelbach, 2004). In line with this, Van Oorsouw and Merckelbach (2006) showed that only 5% of the errors that simulators made while feigning amnesia was repeated in the second memory test. The poorer performance of simulators at the second session was indeed not due to self-generated errors, so that the SMF appears to be an unlikely account for the simulating amnesia effect.

However, investigating and measuring simulators’ source monitoring recognition by using explicit self-reported task (e.g., free and cued recall) may be problematic for at least two reasons. First, even if simulators reported information belonging to their own feigned version in a recall task, they could do so by anchoring their statements to some beliefs concerning how simulation works. Indeed, rather than using actual information regarding a specific event, people use their knowledge or beliefs when they provide responses on specific self-reported tasks (Thompson, Skowronski, Larsen, & Betz, 1996). Second, in a recall task, simulators might arbitrarily report a number of errors without discriminating among self-generated information, parts of the crime stimulus, or re-elaborations of both sources. Arguably, even though different pieces of information coexist in memory and are accessible under some circumstance (Wilson, Lindsey, & Schooler, 2000), participants may confuse memory for details with memory for sources when interviewed through explicit measures (Bayen, Murnane, & Erdfelder, 1996).

Memory theorists have made a distinction between explicit and implicit memory tasks. While explicit measures reflect conscious recollections of previous experiences, implicit measures entail automatic, nonconscious changes in task performance that can be attributed to previous experiences (Greenwald & Banaji, 1995; Schacter, 1992). Implicit tasks reduce the role of self-reflective processes and decrease the mental control necessary to provide an explicit response (Nosek, Greenwald, & Banaji, 2007). Furthermore, such measures have been found to provide an index of specific individual properties (e.g., attitude, cognition), even though participants do not have control over the measurement outcome (De Houwer, 2006; Fazio & Olson, 2003). Through implicit measures, information can be identified when memory tasks require item-source associations that are stored together in a single trace (Diana, Yonelinas, & Ranganath, 2007; Mollison & Curran, 2012).

The present study

In the present study, we used an implicit measure in a source monitoring task in order to assess automatic mental associations that are difficult to tap with explicit measures (Hofmann, Gawronski, Gschwendner, Le, & Schmitt, 2005). Among implicit measures, the autobiographical Implicit Association Test (aIAT; Sartori, Agosta, Zogmaister, Ferrara, & Castiello, 2008) has been designed to detect the strength of automatic associations between mental representations of concepts in people’s memory (Curci, Lanciano, Maddalena, Mastandrea, & Sartori, 2015). Preserving the original idea behind the Implicit Association Test (IAT; Greenwald, McGhee, & Schwartz, 1998; Greenwald, Nosek, & Banaji, 2003; Nosek et al., 2007), the underlying assumption of the aIAT is that associated concepts are paired together in memory leading to an easier and faster responding when they are processed at the same time. The aIAT combines stimuli belonging to four categories: two of those categories are formed by logical true (“I am writing a paper”) or false (“I am swimming in the sea”) statements; the other two categories are represented by real (“I was in Italy during Christmas”) or fabricated (“I was in Serbia during Christmas”) autobiographical events that individuals report. In aIAT terms, faster responses should be facilitated when a true logical statement is associated with a real autobiographical event – congruent block – rather than when a true logical statement is associated with a fabricated autobiographical event – incongruent block. The aIAT has been used in different domains such as mock crimes (Sartori et al., 2008), detection of past intentions (Zangrossi, Agosta, Cervesato, Tessarotto, & Sartori, 2015) flashbulb memories (Curci et al., 2015; Lanciano, Curci, Mastandrea, & Sartori, 2013), eyewitness identification (Helm, Ceci, & Burd, 2016), and performed vs. not-performed and imagined vs. not-imagined actions discriminations (Takarangi, Strange, & Houghton, 2015; Takarangi, Strange, Shortland, & James, 2013).

To our knowledge, the only studies using the aIAT as a memory discrimination task were conducted by Takarangi et al. (2013, 2015). These authors pointed out that imagination decreases the ability to discriminate between true and false events on the aIAT, as a result of source confusion between performed vs. not-performed actions. Participants were presented objects (e.g., toothpick) and associated action statements (e.g., break the toothpick), differing whether those statements were seen, imagined or performed. After two weeks, the authors rated participants’ belief and memory for performing a list of actions and requested participants to complete an aIAT task to distinguish performed (real) vs. not-performed (not-real) actions. Findings showed that the more participants remembered or believed having performed a not-performed action, the less the aIAT discriminated between performed and not-performed actions. Thus, the aIAT seems to be susceptible to the effects of imagining a not-real event (Takarangi et al., 2015, 2013). However, unlike Takarangi and colleagues, we aimed to investigate whether individuals would be able to recognise the correct source of information by adopting the aIAT as source monitoring task, rather than using the aIAT to detect real or not-real events. Hence, this brings us to extend our research question: Can the aIAT function as a possible source monitoring detector when simulators are requested to identify their own self-generated information – associated with true statements – from the original event?

Overview and hypotheses

We aimed to determine whether individuals who previously simulated amnesia were capable to accurately recognise the source of information by adopting the aIAT as a source monitoring task. We showed participants a mock crime video and requested them to either feign amnesia (simulator group) or confess the crime (confessor group) during the memory test phase through a free recall test. Next, we asked participants to figure out how they could have avoided the crime if they would have acted in a different way (i.e., CTF). As part of the procedure during the memory test phase, we introduced the CFT task to force participants to make a personal source of confusion. CFT is constrained by reality and it is essentially focused on plausible alternatives to one’s own actions (Roese & Epstude, 2017), thereby increasing the degree of confusion. We used this type of thinking to resemble, as much as possible, the naturalist situation of perpetrators thinking back to their offence. One week later, each participant was invited to come to the laboratory and requested to recognise the source of information represented by statements belonging either to their free recall or not (present vs. absent discrimination). It is important to note that our task requires discrimination between information present vs. absent in the recollection taking place at the first test session, and not between real vs. not-real (performed vs. not-performed) as in the study by Takarangi et al. (2013). We thus administered participants a personalised source monitoring task through the aIAT. Participants who confidently preserved the memory of the correct source would faster categorise true logical statements associated with details that were present in their free recall at the first test session, as compared with true logical statements associated with details that were absent in their free recall. We predicted, thus, that both simulators and confessors exhibit high positive average values of the aIAT effect (Hypothesis 1). Moreover, we expected a significant difference between groups in the aIAT effect. That is, because of having developed a self-generated story as a result of elaborative cognitive processing of the mock crime, simulators would find the aIAT’s discrimination task less difficult than confessors. Hence, we anticipated simulators would be quicker than confessors in the aIAT categorisation when information present in their free recall was associated with true logical statements (congruent block), rather than information present in their free recall was associated with false logical statements (incongruent block). In other words, we predicted simulators would be faster than confessors in identifying source information (Hypothesis 2). Finally, we tested whether the aIAT index is a measure able to discriminate between simulators and confessors (Hypothesis 3).

Method

Participants and design

A group of 119 students participated in the present study for course credit hours. We excluded 11 participants with unusual data patterns, leaving 108 participants (67% women; M_age= 24.03; SD = 2.75). Specifically, eleven participants did not comply with the instructions received during session 1 by reporting an insufficient number of statements either in the free recall or in the CFT task. The lack of details in these reports did not allow us to construct a personalised aIAT for those participants. The study used a 2 × 2 mixed model design with condition (simulators vs. confessors) as a between subjects factor and the aIAT block (congruent vs. incongruent) as a within-subjects factor. Participants were randomly assigned to one of the conditions. The dependent variable was the implicit measure of the aIAT effect calculated as D index.

Measures and procedure

Session 1

Participants involved in the study completed two sessions in the laboratory. Each participant was randomly assigned to one of two conditions. During the screening phase, participants were preliminary assessed with the Positive and Negative Affect Schedule-Trait and State (PANAS–T and –S; Terraciano, McCrae, & Costa Jr, 2003; Watson, Clark, & Tellegen, 1988) and the Structured Inventory of Malingered Symptomatology (SIMS; La Marca, Rigoni, Sartori, & Lo Priore, 2011; Windows & Smith, 2009). The PANAS-T and S were used to assess the participants’ emotional trait and state and to exclude individual differences in terms of the affective state. The PANAS-S was administered a second time to evaluate the emotional impact of the stimulus material. The SIMS was used to assess individual malingering tendency.

The PANAS–T and –S; Terraciano et al., 2003; Watson et al., 1988. The scales require participants to rate on twenty 5-point items how they experience different emotional states along two dimensions, matching to Positive Affect (PA) and Negative Affect (NA). For both PANAS–T and –S item score were summed up. The PA–T scale (α = .75) indicates the individual positive level of emotions generally felt by people, while in contrast the NA–T scale (α = .90) indicates the individual general dimension of aversive affect and distress. The PA–S (α = .87) and the NA–S (α = .92) scales reflect how individuals experience in that precise moment.

The SIMS; La Marca, et al., 2011; Windows & Smith, 2009. The SIMS¹ is a 75 self-report screening questionnaire for malingering of mental disorders. The items are divided into five subscales (affective disorders; amnestic disorders; low intelligence; neurological impairment; psychosis). Answers indicative of malingering are summed to obtain a total SIMS score (α = .87).

Mock crime video

After participants had completed the screening phase, a mock crime video was shown to them. The mock crime video (3 min) was recorded in point of view (pov) perspective in order to avoid the potential confounding effect caused by the offenders’ gender² and was accompanied by background music. Participants were instructed to pay attention to the mock crime and were asked to identify themselves with the character that performed actions in the scene (the offender). The video describes the offender’s day in which s/he woke up and went to the office and after having dinner s/he went to different pubs for drinking. Once in the restroom of the last pub, the offender had an intense fight with a young person. The scene ended with the strangling of the victim.

The mock crime video was divided into one hundred critical information units. A critical information unit was defined as a significant portion of the crime (maximum = 100). Participants’ free recalls were scored by the first author and two research assistants. Participants earned 1 point for every correct unit reported (e.g., “I strongly pushed the victim by the wall”) while half point was assigned for a partial correct answer (e.g., “I pushed the victim”). The Interclass Correlation Coefficient (ICC) average measure for the number of correct free recall information was .966 with a 95% confidence interval from .95 to .97 [F (107,11) = 29.61, p < .001]. Furthermore, the number of errors was identified (i.e., introduction of other or distorted information; for example, “I hit my head against the wall” or “I strangled the victim with my belt”). The ICC average measure for the errors was .80 with a 95% confidence interval from .71 to .86 [F (107,11) = 5.09, p < .001]. After the video presentation, a 10 min filler interval followed during which all participants played a computer game.

Memory test phase

After the filler task, participants were asked to figure themselves in a simulated police interrogation in which they had been arrested on suspicion of homicide. In line with previous research (e.g., Christianson & Bylin, 1999; Van Oorsouw & Merckelbach, 2004), we requested participants to report their statements through a free recall memory test. The simulator group was instructed to feign amnesia to evade punishment by claiming not being able to remember the strangulation. Thus, simulators were free to omit, distort and report other information, pretending they had great difficulties to remember committing the violent crime. In contrast with simulators, confessors were instructed to collaborate with the police by reporting as many details as possible about the crime. Immediately after the free recall, a manipulation check was run by administering the PANAS–S a second time in order to check the affective impact of the mock crime event. Participants then completed another 10-min filler task.

Finally, we invited each participant of both conditions to write down how they could have prevented the murder if they would have behaved differently (i.e., CFT; Roese & Epstude, 2017; Roese & Olson, 1997). By requesting participants to generate an individual source of confusion, we included these statements (i.e., CFT) into the subsequent aIAT statements. The ICC average measure for the number of CFT statements was .90 with a 95% confidence interval from .86 to .94 [F (107,11) = 10.66, p < .001]. Once participants completed this last phase, they were scheduled for a second session one week later.

Session 2

In session 2, participants’ task on the aIAT was to discriminate information among four categories as accurately and as quickly as possible. A tailored aIAT was created for each participant (see Table 1) following the a/IAT methods (Greenwald et al., 2003; Sartori et al., 2008). Simulator and confessor groups followed the same instruction, meaning that simulators were asked to give up their role of a feigner while engaged in the aIAT source monitoring task. Namely, each participant categorised statements as being “true” (five statements; e.g., “I am in front of a computer”) or “false” (five statements; e.g., “I am in the car”), and statements as being “present” or “absent” in their free recall that they had generated during the session 1. The “true” and “false” statements were selected and readjusted from Sartori et al. (2008). True and false sentences were the same for all participants. “Present” statements (five) were obtained from the free recall test by participants during the session 1: For “present” statements we used only statements from the participants’ free recall that were not displayed in the mock crime video. Thus, “present” simulators’ statements concerned their feigned story (e.g., “I hit my head against the wall in the toilet”), whereas “present” confessors’ statements concerned their distorted errors regarding the mock crime video (e.g., “I strangled the victim with my belt”). On the other hand, the “absent” statements (five) were obtained from both mock crime video and CFT³ procedure during the session 1 (see Table 1). Moreover, mock crime video details were those pieces of information that either participants deliberately did not report (i.e., simulators) or did forget to report (i.e., confessors). Each statement that appeared on the screen belonged either to one or to the other category (e.g., “present” vs. “absent”).

Table 1. Example of statements’ list used during the aIAT task.

	Simulators	Confessors
Categories	Statements
True^a	I am in front of a computer I am in Bari It is 2017 I am doing a psychological test I am sitting on a chair
False^a	I am in a car I am in Milan It is 1851 I am doing a card playing game I am in mountain
Present^b	I was sober I just saw the victim for an instant I was pushed by the victim I hit my head against the wall I passed out	I was in a cab I met a friend of mine I drank Cosmopolitan I was punched by the victim I strangled the victim with my belt
Absent^c	I strangled the victim ^(□) I could have controlled my anger ^(…) I vomited in the toilet sink ^(□) I should have stayed at home ^(…) I smoked some cigarettes ^(□)	I could have stopped myself before ^(…) I had an intense fight with the victim ^(□) I should have calmed myself down ^(…) I importuned a young girl ^(□) I just should have said sorry ^(…)

^aTrue and False statements were the same for both groups.

^bPresent statements were obtained from each participant’s free recall.

^cAbsent statements were obtained from both mock crime video ^(□) and counterfactual thinking ^(…) and counterbalanced by order and number.

The aIAT task consisted of five separate blocks of categorisation trials. In each trial, the statement was presented in the centre of a computer screen. In block 1 (“true” vs “false” discrimination; 20 trials), participants categorised statements as belonging to “true” (key E) or “false” (key I). In block 2 (“present” vs. “absent” discrimination; 20 trials), participants had to classify statements as “present” (key E) or “absent” (key I) in their free recall of Session 1. In block 3 (initial double categorisation; 60 trials), participants had to indicate whether statements were either belonging to “true” or “present” (key E) or whether they were either belonging to “false” or “absent” (key I). In block 4 (reversed logical discrimination; 40 trials), participants had to categorise statements as “false” (key E) or “true” (key I). In block 5 (reversed double categorisation; 60 trials), participants had to classify whether statements were either “false” or “present” (key E) or whether statements were either “true” or “absent” (key I). The presentation order and repetitions number of each trial were randomised for each participant within a number of blocks. An error feedback (red “X” letter) appeared when participants made an incorrect response; they were required to correct the responses pressing the other key. The D index was calculated according to Greenwald and colleagues’ procedure (2003). This index expresses the IAT effect in terms of the standard deviation of the latency measures and it includes a penalty for incorrect responses. Specifically, the D index corresponds to the weighted difference of the mean response latencies of critical blocks (i.e., congruent vs. incongruent associations) divided by the standard deviation of all critical trials (Greenwald et al., 2003; Nosek et al., 2007). Finally, participants were thanked for their participation and debriefed (Figure 1).

Figure 1. Procedure adopted in the current study is displayed. Screening measures, mock crime video, filler task and memory test phase, compose Session 1. During Session 2, participants performed a tailored aIAT source monitoring task.

Results

Screening analysis

In order to prevent individual differences in the participants’ affective state, we analysed positive and negative PANAS–T scores through an independent t-test. Participants did not differ neither in the PA-T [M_{PA simulators}= 27.61; SD = 4.60; M_{PA confessors}= 27.44; SD = 4.64; t(106) = .19, p = .85, d. = .03] nor in the NA-T level before the experimental session [M_{NA simulators}= 11.07; SD = 6.97; M_{NA confessors}= 10.76; SD = 8.44; t(106) = .21, p = .83, d = .04]. Further, in order to exclude the individual malingering tendency an independent sample t-test was run on SIMS total score. No significant difference was found between groups [M_simulators = 8.44; SD = 4.38; M_confessors = 8.60; SD = 4.30; t(106) = −.18, p = .86, d. = .04]. To sum up, these findings suggest that participants did not differ in their affective state nor in their malingering tendencies before being exposed to the mock crime event.

Affective impact of the mock crime event

A 2 × 2 repeated measures ANOVA with condition (simulators vs. confessors) as between subjects factor and pre–post mock crime viewing (pre-mock crime vs. post-mock crime) as a within subjects factor was conducted in order to check the affective impact of the mock crime on participants. Only the main effect of the pre–post mock crime viewing was found to be significant on both PA-S and NA-S scores [F(1,105) = 10.17, p = .002, η_p²  = .09; F(1,105) = 23.58, p < .001, η_p²  = .18, respectively]. The initial participants’ positive state decreased after the video was shown to both simulators (M_{PA pre-mock crime}= 26.43; SD = 6.80 vs. M_{PA post-mock crime}= 25.57; SD = 6.27) and confessors groups (M_{PA pre-mock crime}= 26.96; SD = 6.29 vs. M_{PA post-mock crime}= 24.68; SD = 8.01). At the same time, the initial participants’ negative state increased for both simulators (M_{NA pre-mock crime}= 5.80; SD = 6.96 vs. M_{NA post-mock crime}= 7.96; SD = 8.68) and confessors (M_{NA pre-mock crime}= 5.20; SD = 6.64 vs. M_{NA post-mock crime}= 10.19; SD = 10.64). No other main or interaction effects were found [Fs(1,105) < 2.21, p > .15, η_p²  < .02]. These findings suggest that our mock crime video had an affective impact on participants.

Present category: analysis of the free recall

We analysed whether participants complied with their instruction, meaning that simulators would report less correct information and more errors than confessors. With respect to this, an independent sample t-test was run on both correct and error rate scores. Simulators reported fewer correct details and more errors than confessors [t(106) = −12.05, p < .001, d. = 2.32, and t(106) = 5.19, p < .001, d. = .99, respectively; see Table 2]. Findings show both simulators and confessors reported information in their free recall in accordance with the instruction received at the memory test phase.

Table 2. Means of correct information, errors, and CFT statements reported by simulators vs. confessors during the free recall are displayed. Standard deviations are shown between parentheses.

	Simulators	Confessors
Correct	6.39* (4.20)	22.58* (8.93)
Error	11.26* (5.98)	6.67* (2.54)
CFT	3.57 (1.51)	3.33 (1.34)

*p < .001 between groups at the memory test phase.

Absent category: analysis of the CFT

To test whether participants differed on the number of CFT generated- later used as statements in the absent category⁴ – an independent sample t-test was run on the CFT rate score. No significant difference was found between simulators and confessors, meaning that both groups reported a similar amount of CFT statements [t(106) = .87, p = .384, d. = .17; see Table 2].

aIAT measure

Before any additional analysis, we calculated the D index for the IAT design with the built-in error procedure. We follow these steps: (1) use data from congruent and incongruent blocks (blocks 3 and 5); (2) eliminate trials with latencies >10,000 ms and participants who obtained in more than 10% of trials a latency <300 ms; (3) compute the standard deviation for all practice trials in the both crucial blocks (3 and 5 blocks) and the standard deviation for all test trials in both crucial blocks (3 and 5); (4) compute separated means for practice congruent trials, practice incongruent trials, test congruent trials and test incongruent trials; (5) compute two difference scores (one difference between practice congruent trials and practice incongruent trials, and the other between test congruent trials and test incongruent trials); (6) divide each difference score by its associated standard deviation from Step (3); and (7) average the two quotients from Step (6).

In the current study, the D index (Greenwald et al., 2003; Nosek et al., 2007) was employed as main measure of the aIAT effect.⁵ In accordance with our hypothesis (1), positive average D values were observed for both simulators (M_simulators = .90; SD = .25) and confessors (M_confessors = .76; SD = .42) groups, confirming a high aIAT effect for the task. Thus, both groups were fast in identifying the right source of information when statements belonging to their free recall were presented in association with true statements. An independent sample t-test was run in order to compare groups’ average D values of the IAT effect. As predicted, the D value was found to be higher for simulators than for confessors (t(106) = 2.14, p = .03, d = .40). Consistent with our hypothesis (2), simulators were quicker than confessors in discriminating source information (Figure 2).

Figure 2. Participants’ mean D index by simulators and confessors group. Errors bars show standard errors of means.

Receiver-operating-characteristic (ROC) analysis

A ROC analysis was run to test whether and how well the aIAT would discriminate between simulators and confessors at the implicit level (Hypothesis 3). The D index was used as a test variable and condition (simulators vs. confessors) as a state variable. The value of the state variable indicates which category should be considered as positive (in our case simulators). The ROC analysis presents an Area Under the Curve (AUC) which provides a measure of discrimination from 1 (perfect discrimination) to 0 (null discrimination). In our study, the ROC analysis exhibited an AUC of .60 when “simulator” was considered as the positive value of the state variable. At the implicit level, the IAT differentiated simulators from confessors, thereby confirming our hypothesis 3.

Discussion

When individuals feign amnesia for a mock crime and are later requested to give up their role as a feigner, mental confusion caused by an inability to recognise source information may result in impaired and distorted memory of the crime. Some authors suggested that source monitoring errors might cause the memory-undermining effect of simulating amnesia (e.g., Christianson & Bylin, 1999). However, several studies have also shown that the content of self-generated information remains relatively clear in people’s memory (e.g., Chrobak & Zaragoza, 2008; Johnson et al., 1981, 1988). Even though simulating amnesia for a crime can impair memory for the event, it is possible that individuals might distinctly keep in mind which information is self-generated or not. Consequently, in designing the present study, we wondered whether individuals would identify the correct source of information despite they previously simulated amnesia. To this purpose, we applied the aIAT (Sartori et al., 2008) as a possible source monitoring task since we also aimed to investigate whether or not an implicit measure would automatically detect individuals’ source recognition capacity.

The results were in line with our expectations. To begin with, participants of both conditions (simulators vs. confessors) were faster in categorising their self-generated information – belonging to their free recall – associated with true statements and non-self-generate information – belonging either to mock crime video or to the CFT task – associated with false statements (congruent block), than when categorising self-generated information associated with false statements and non-self-generate information associated with true statements (incongruent block). It seems that participants were capable to automatically recognise which displayed information belonged to themselves or came from different sources. For both groups, the positive and high D value indicated that a strong association was obtained when true logical dimension was combined with participants’ statements.

Second, in accordance with our prediction, we found simulators faster than confessors. Precisely, when true logical statements were associated with statements concerning the fake crime-related version (congruent block), simulators were able to recognise the right source of information. Thus, even though simulators were previously asked to feign amnesia, they clearly were able to identify their own self-generated variant of the crime from the original version. On the other hand, confessors also showed a positive D value of the aIAT effect even though their D index was lower than that obtained by the simulator group. This means that confessors seemed to be slower than simulators in the aIAT source monitoring task. Given that confessors’ assignment was to report as many details as possible about the crime, confessors’ free recall statements (“present” statements in the aIAT task) might easily be confused with statements reporting a mere recollection of the original mock crime details (“absent” statements in the aIAT task). Moreover, although assembled into two categories (i.e., “present” vs “absent”), it appears that simulators had to discriminate the displayed statements among three sources – their feigned version of the crime vs. memory for the crime vs. the CFT task – whereas confessors have been pulled by only two – memory for the crime vs. CFT task – rendering their confession and the memory for the mock crime video essentially one source (i.e., memory for the crime). Indeed, drawing on the SMF, source monitoring accuracy decreases when qualitative details are shared between sources (Johnson et al., 1993; Lindsay, 2008). Thus, confessors’ performance was exposed to a higher possibility of source monitoring errors as compared with simulators: Identifying the correct source of information based on similar statements – belonging either to the free recall or to the original mock crime – was more difficult for confessors than for simulators. It follows that the specificity of the statements in participants’ free recall may represent the keystone for understanding the difference between groups in discriminating source information. That is, each simulator generated a personal feigned version of the crime, which was easily differentiated from the original version. As some authors suggested (e.g., Chrobak & Zaragoza, 2008), self-generated information is likely to be remembered because composed by sequences of acts precisely identifiable from the beginning to the end. Moreover, as predicted by the SMF, the more memories require elaborate cognitive processing, the simpler individuals should find to remember the source (Chrobak & Zaragoza, 2008; Johnson et al., 1981). It is a common experience that individuals who simulate memory loss after a crime tend to be consistent with their supposed amnesia state during police investigations, being thus able to identify the correct sequence of events to be coherently reported. In the constancy of their retelling there might be the reason why individuals who feigned a loss of memory were better at accurately distinguishing between sources of information and at the aIAT source monitoring task.

By applying the aIAT as a source monitoring task we extended our research question: Can the aIAT be a possible source monitoring detector? Previous studies (e.g., Van Oorsouw & Merckelbach, 2004) have shown that simulators did not exhibit a heightened percentage of errors in their self-reported free and cued recall at the second memory test, leading the source-monitoring explanation to fail as an adequate interpretation for the memory-undermining effect of simulating amnesia. An important consequence of the present findings is that simulators appear to be able to automatically recognise the correct source of information by using an implicit measure as source monitoring task. Arguably, the poorer performance of simulators in recalling the crime after time, as reported in the literature (e.g., Christianson & Bylin, 1999; Van Oorsouw & Merckelbach, 2004), might be independent on their source monitoring errors. Moreover, asking simulators to recognise source information through aIAT task may have caused an association between item and source, perhaps based on a feeling of familiarity. As a matter of fact, some studies pointed to the importance of the feeling of familiarity in source monitoring recognition (e.g., Diana, Yonelinas, & Ranganath, 2008; Hicks, Marsh, & Ritschel, 2002; Mollison & Curran, 2012; Yonelinas, Kroll, Dobbins, & Soltani, 1999). Familiarity-based recognition implies the contribution of some sub-processes such as those supporting implicit memory (Jacoby, 1991; Yonelinas, 2002). Furthermore, sometimes and to some extent, the IAT effects measure the attributes that they are expected to measure (e.g., De Houwer, Teige-Mocigemba, Spruyt, & Moors, 2009). Hence, in our study, the aIAT was successful in detection of correct information required by the used type of the task. The ROC analysis, indeed, showed that the aIAT was successful at discriminating simulators from confessors (hypothesis 3), implying that the test can differentiate individuals who previously feigned amnesia for a crime from those who honestly reported information about the same event by requesting them to distinguish the right source of information. In accordance with this, we argue that the aIAT might be useful as a task to retrieve crime-related information. By recognising the right source, which might be stored together with the related information in a single trace, offenders might be facilitated in recollecting details of the crime when they are motivated to collaborate with the justice department (e.g., plea bargaining case). Indeed, specific crime-related information is directly related to feelings and mood evoked at the time of the event and, thus, more accessible through implicit association than through traditional methods of assessment (Schacter, 1987; Tobias, Kihlstrom, & Schacter, 1992).

However, the generalizability of our results is related to the aIAT’s ability to detect source information since, for instance, external factors have been shown to affect implicit measures (Greenwald & Nosek, 2001). Moreover, some evidence has shown how it is strategically possible to influence outcomes of implicit measures (De Houwer, 2006). Therefore, in order to better delineate the aIAT’s efficacy as a possible source monitoring detector, the test should be administered in association with an explicit source monitoring task (e.g., multiple discrimination task). That is, future research might focus on comparing the aIAT to an explicit source monitoring measure. Perhaps, individuals who pretend such memory loss might be more successful on an explicit task than confessors at distinguishing among different source of information. Previous studies have shown that comparing explicit vs implicit measures would indeed enable researchers to draw more specific conclusions on the purpose of the aIAT (e.g., Curci et al., 2015). In a related vein, and opposite to our study in which we asked participants to recognise the correct source of information, it would perhaps be wise to investigate in further research how, and to which extent, simulators might perform on the tests (aIAT vs. explicit measure) by requesting them to keep feigning during a source monitoring task.

Finally, some limitations of the present study need to be mentioned. To begin with, we instructed students to either feign amnesia or to confess a violent crime. Given that our sample differs in many ways from individuals who commit a severe crime (Schacter, 1986), our findings may have a limited ecological validity. Second, in our study we were more interested in investigating simulators’ source monitoring ability than in simulators’ memory performance over time. In other words, the present study was not focused on observing the well-known memory-undermining effect of simulating amnesia (e.g., Christianson & Bylin, 1999) since we aimed to investigate whether individuals who feigned a memory loss would accurately discriminate the source of information by using an implicit measure as source monitoring task. For this reason, we did not assess a second delayed free recall memory test as usually it has been done in previous studies in this field (Bylin & Christianson, 2002; Christianson & Bylin, 1999; Van Oorsouw & Merckelbach, 2004, 2006). Futures studies should extend our findings by including a memory retest phase assessing the association between simulators’ source monitoring ability and their memory for the crime. Third, one could argue that the CFT statements might have played a significant role on the aIAT participants’ performance since, in the “absent” category, we brought together CFT statements with crime-related details which participants did not mention in their free recall. Without doubt these two types of statements are very different. Whereas the CFT illustrates acts that participants knew did not happen, information belonging to the mock crime video may be details that, especially within the confessors group, participants knew happened but they did not mention. Although the CFT task correctly applies to the “absent” category, because what perpetrators think of the crime in terms of plausible alternatives to their offence, it is not what they claim when they are interviewed, further research is necessary. Future studies should consider including CFT (Roese & Epstude, 2017; Roese & Olson, 1997) as a condition of the experimental design – thus subjected to corresponding analyses – rather than as part of the procedure, to assess to which extent this way of thinking may affect participants’ source monitoring capacity.

To conclude, our study sheds new light on the debate regarding the memory-undermining effect of simulating amnesia. We aimed to unravel the mechanism behind the simulating amnesia phenomenon by applying the aIAT as possible source monitoring task. Our main findings suggest that simulators may preserve the memory of a crime despite feigning amnesia and seem to be capable to automatically recognise their own self-generated version of the crime from the original one. Although the aIAT still prompts a controversy within the legal context (e.g., Takarangi et al., 2013), this study suggests that an implicit measure might be helpful in retrieving the source of some crime-related information. We believe that the present study would be a step forward for the understanding of both memory-undermining effect of simulating amnesia and aIAT’s forensic relevance.

Acknowledgements

We would like to thank Marco Serpenti for his help in recording the mock crime video for this study and Nausica Cervone for collecting data.

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Ivan Mangiulli http://orcid.org/0000-0002-5409-7325

Antonietta Curci http://orcid.org/0000-0002-0932-7152

Notes

1. With the purpose of checking differences in participants’ malingering tendency, we analysed only the SIMS total scores (Windows & Smith, 2009). Therefore, we did not include the others subscale in our analysis.

2. As the majority of the studies in this field (e.g., Bylin & Christianson, 2002; Van Oorsouw & Merckelbach, 2006) ask female students to identify themselves with a male offender, we created a mock crime video in pov to avoid this problematic aspect of the procedure.

3. We counterbalanced the order and the number of both mock crime video and CFT details in the “absent” category within subject for each personalised aIAT (see Table 1). Both mock crime video and CFT statements were randomly chosen.

4. In the absent category, we also analysed information that was not reported by each group (simulators vs confessors). Specularly to the correct rate investigated in the free recall, simulators omitted more information than confessors [t(106) = 11.74, p < .001, d = 2.30].

5. Additionally, we also calculated error rates for congruent and incongruent block (Blocks 3 and 5) by each group. Simulators and confessors did not differ in error rates neither on congruent (M_simulators = 3.94; SD = 3.01, and M_confessors = 4.78; SD = 4.10; t(106) = −1.20, p = .231, d = .23) nor on incongruent blocks (M_simulators = 10.50; SD = 6.10, and M_confessors = 10.67; SD = 6.57; t(106) = −1.37, p = .892, d = .03). Findings suggest that errors made by participants during the aIAT did not affect their ability at identifying the correct source of information regardless of the group they belonged to, in accordance with our hypothesis (Hp 1).