Pretending to be someone else
Does acting experience diminish the cognitive cost of lying?
Number of EC: 21
Date: March 25, 2013 until March 12, 2014
Name: Chantal Keijlard
Student ID: 5960428
Supervisor: Bruno J. Verschuere PhD. Co-‐assessor: prof. dr. Denny Borsboom UvA Representative: prof. dr. Denny Borsboom
University of Amsterdam, Faculty of Social and Behavioral Sciences, Clinical Psychology
MSc in Brain and Cognitive Sciences, University of Amsterdam Cognitive Neuroscience track
Abstract
Lying typically comes with a cognitive cost as opposed to telling the truth. For instance, lying usually takes longer than truth telling. The boundary conditions of this effect however are unknown. Whereas some have argued that this cognitive cost is hardwired, recent research has suggested that the cognitive cost of lying is malleable through practice. Here, we examined whether individual differences in experience with deception modulate the cognitive cost of lying. In this study we tested individuals who are trained in one specific form of deception: Pretending to be someone else. Hence we examined the differences in cognitive cost of lying regarding denying autobiographical information between actors and non-‐actors with three different reaction time based deception tests. There was no effect of acting experience for any of the deception tests, hence the results pointed in the direction of a fixed cognitive cost of lying. However, two of the three deception tests failed to indicate a cognitive cost of lying altogether. Due to the small sample size and the lack of significant results, conclusions about the malleability of the cognitive cost of lying cannot be made with certainty. Even though the current study failed to provide evidence for the malleability of the cognitive cost of lying, certain methodological adjustments could be made that might yield a different outcome. Implications of a variety of outcomes will be discussed in this paper.
Introduction
Deceiving others is said to be more cognitively constraining than truth telling. Apart from constructing the lie, one has to keep the truth in mind, while at the same time inhibiting it (Walczyk, Roper, Seemann, & Humphrey, 2003; Spence et al., 2004). Previous research indicated that lying takes longer, and is more error prone than telling the truth (Furedy, Davis, & Gurevich, 1988; Seymour, Seifert, Shafto, & Mosmann, 2000; Spence et al., 2001; Farrow et al., 2003; Walczyk et al., 2003; Sartori, Agosta, Zogmaister, Ferrara, & Castiello, 2008; Seymour & Kerlin, 2008; Osman, Channon, & Fitzpatrick, 2009; Walczyk, Mahoney, Doverspike, & Griffith-‐Ross, 2009; Hu, Wu, & Fu, 2011; Verschuere, Spruyt, Meijer, & Otgaar, 2011; G. Visu-‐Petra, Miclea, & L. Visu-‐Petra, 2012; Williams, Bott, Patrick, & Lewis, 2013). According to R. Johnson, Barnhardt, & Zhu (2005) this “cognitive cost of lying” is resistant to practice, as it is caused by inherent limitations of human information processing. However, recent research has provided initial evidence against this statement. Verschuere et al. (2011) found that changing the proportion of the truth and lie responses in a lie detection test affected the ease of lying. When lie trials outnumbered truth trials lying became easier, whereas when truth trials outnumbered lie trials lying became more difficult. These findings provide evidence for the claim that the cognitive cost of lying is not fixed. In fact it may be malleable through training. Additional research replicated this training effect, while also showing that it was specific to practiced questions only and did not generalize to untrained questions (Van Bockstaele et al., 2012). The training however took place in a laboratory, making it difficult to generalize the results outside of a laboratory setting. Additionally, the training was quite short (single training of 540 trials). It has been argued that one needs around 10,000 hours of practice in order to reach an expert level of performance with domain-‐specific learning (Ericsson, Krampe, & Tesch-‐Römer, 1993; Ericsson & Lehmann, 1996). Since it is unrealistic to train people in a laboratory for that amount of time, we suggested a different research approach for examining the malleability of the cognitive cost of lying. Instead of training individuals extensively at lying, it is more efficient to test certain groups of individuals who are trained at deception already (e.g. actors). Such cross-‐sectional designs allow us to examine individuals with different levels of experience with lying.
Before we delve into the current experiment we will first elaborate on the cognitive cost of lying by describing several theories on the topic. Secondly, multiple deception paradigms will be discussed after which research attempting to alter the cognitive cost of lying is described as well as individual differences regarding this cost. Thirdly, an extensive description of the current experiment will be provided after which the implications of the findings will be discussed. Finally, a proposal for future research will be provided.
The cognitive cost of lying
There are several theories about the cognition of deception. In this section some of these theories will be described in order to provide an overview of the opinions on how the enhanced difficulty for lying is expressed. The underlying cognitive tasks these theories propose are involved will also be discussed.
According to the Activation-‐Decision-‐Construction Model of Lying (ADCM) there are three cognitive events (divided into sub-‐steps a to g) that occur during the act of deception (activation, decision, and construction component, Walczyk et al., 2003). When asked a question, relevant information from semantic and episodic memory is activated (step a; Kintsch, 1998, cited in Walczyk et al., 2003). The question itself will
occupy the articulatory loop in working memory (WM; step b). Consequently, information in long-‐term memory (LTM), which is linked to the activated information, will become active, allowing the deceiver to use this additional information to construct a lie (step c; Ericsson & Kintsch, 1995; Walczyk et al., 2003). Based on this information a decision is being made about the favorable response (i.e. whether lying would be more beneficial than telling the truth or vice versa, step d). If a decision to lie has been made, an inhibitory link is established, which prevents the individual from telling the truth (step e; Walczyk et al., 2003; Williams et al., 2013). Additionally, implausible lies or lies that may otherwise jeopardize the believability of the liar are inhibited. After this selection process the lie that is most active will now become fully active in WM (step f) after which it is uttered (step g). The ADCM proposes that the activation component is automatic, whereas the decision to lie or tell the truth is intentional as is the construction of a lie itself (Walczyk et al., 2003). In order to find evidence for the ADCM, Walczyk et al. (2003) examined each of the cognitive components of this model. They found that a question indeed made the truth become active, even when the individual had the intention to lie to the question. When open-‐ended questions were asked, it took participants on average 166 ms to make the decision to lie. Walczyk et al. (2005) replicated this effect. Furthermore, the construction of a lie made participants even slower to respond to the questions as was also found earlier by Spence et al. (2001).
The Parallel Task Set Model (Seymour, 2001, cited in Walczyk, Igou, Dixon, & Tcholakian, 2013) posits that during deceitful answering to multiple choice questions a liar has to remember the correct answer, select the (previously chosen) incorrect (and thus deceitful) alternative from the available answers, prepare to give the response, and finally execute the response. Recognizing the correct response occurs automatically and fast. Recollection of the deceitful response however, occurs slower and is under conscious control. When hiding Guilty knowledge (e.g. the murder weapon from a crime scene), the liar has to inhibit the response for the correct alternative, and execute the previously chosen deceitful response, causing a response conflict, which then in turn results in longer reaction times.
The Interpersonal Deception Theory (Buller & Burgoon, 1996; Burgoon & Buller, 2008, cited in Walczyk et al., 2013) focuses more on the interaction between the deceiver and his or her conversation partner. According to this theory the cognitive load that comes with deception comprises several factors. Deceivers have to make up a believable story, while at the same time monitoring the responses of their conversation partner, as well as keep track of the lies they created during the conversation. Because this is already vexing enough as it is, additional tasks such as controlling for behaviors that cue deception will lead to a cognitive overload. This cognitive overload will result in the deceiver expressing behavioral signs of deception (e.g. vagueness in the story being told, frequent pausing), risking exposure (Walczyk et al., 2003). Skilled liars however, may have better working memory, allowing them to process all the relevant information without causing a cognitive overload. These liars should then be more difficult to expose with the conventional lie detection techniques.
Generally, the theories described above recognize the claim that lying is more difficult than telling the truth. However, there is no clear consensus on how this enhanced difficulty is expressed. Several deception tests have been developed that aim to pinpoint deception. In the next section some deception paradigms and their applications will be discussed.
Deception Paradigms
The most frequently used procedure for the polygraph test is the Control Question Test (CQT; e.g. Ben-‐Shakhar, Bar-‐Hillel, & Kremnitzer, 2002; Walczyk et al., 2013), which consists of critical questions (e.g. relevant to a committed crime), and control questions (usually related to the respondent’s past). If the physiological responses (heart rate, respiration, blood pressure, skin conductance response) to critical items are significantly higher than the responses to control items this indicates a positive test outcome (i.e. indicating that the respondent is lying), if the physiological responses are lower for critical items this indicates a negative test outcome (i.e. indicating that the respondent is telling the truth), and if the physiological responses between item types do not differ the outcome of the polygraph is undecided (e.g. Fiedler, Schmid, & Stahl, 2002). According to several researchers the critical questions of the CQT are emotionally significant to anyone, eliciting a relatively strong response in guilty and innocent interviewees alike. This in turn elevates the risk of a false positive outcome (an innocent individual obtaining a guilty test score, e.g. Fiedler et al., 2002; Ben-‐Shakhar & Elaad, 2003). While the validity of the CQT is heavily contested, there are at least three paradigms that show promise: the Concealed Information Test (CIT, Lykken, 1959; Lykken, 1974)1, the Sheffield Lie Test (SLT), and the autobiographical Implicit Association Test (aIAT).
The CIT aims to detect guilty knowledge, rather than guilt itself (Ben-‐Shakhar et al., 2002; Verschuere et al., 2011). The test consists of questions that focus on a detail of an event, which can only be known to individuals with detailed knowledge of said event (e.g. Your getaway car was a …). After each question, the examinee is presented with a succession of several possible alternatives (e.g. Nissan, Honda, Toyota, Ford, Chrysler). The examinee is typically instructed to say “no” to all alternatives. To an innocent examinee the correct alternative (the probe, e.g. Ford) is indistinguishable from the irrelevant alternatives (e.g. Nissan, Honda, Toyota, Chrysler), whereas to an examinee with guilty knowledge this alternative stands out from the others. In the polygraph version of the CIT the physiological response to the correct alternative is then compared with the physiological responses to the irrelevant alternatives. Typically, individuals with guilty knowledge show a larger skin conductance response (SCR) to the critical information as compared to the irrelevant information (e.g. Gronau, Ben-‐Shakhar, & Cohen, 2005; Verschuere & Crombez, 2008; Verschuere, Crombez, Smolders, & Clercq, 2009). Other versions of the CIT have also proven their use in detecting knowledge of critical information. For example, the Event Related Potential (ERP) based CIT (Farwell & Donchin 1991) detects guilty knowledge by comparing the ERPs to critical information with the ERPs to irrelevant information. Evidence for reaction time differences between critical and irrelevant information was found in ERP studies (e.g.
Rosenfeld, Biroschak, & Furedy, 2006). Additionally, reaction times were found to be highly valid as a separate measure of guilty knowledge (Seymour, Seifert, Shafto, & Mosmann, 2000; Seymour & Kerlin, 2008; Verschuere, Crombez, Degrootte, & Rosseel, 2010). Using reaction times as a measure of guilty knowledge is appealing in part because its application is less complex and less costly than any type of physiological measure (Seymour et al., 2000; Verschuere, Crombez, Degrootte, & Rosseel, 2010). However, reliable measures of both the ERP-‐based CIT and the RT-‐based CIT require an increased amount of trials as compared to the polygraph-‐based CIT (Noordraven & Verschuere, 2013). Furthermore, the stimuli are rapidly presented in succession, which
may lead to items not being processed by the participant. For this reason, inclusion of a third item type (target items) is desirable (Seymour et al., 2000; Noordraven & Verschuere, 2013). Before testing, the examinee is presented with the target items (e.g. Mazda) and is instructed to learn these by heart. During the test, the examinee indicates as quickly as possible whether the presented stimulus is a target, or not. Thus, participants will still answer “no” to both probe items and irrelevant items and “yes” to target items. Although the CIT comes in many forms (e.g. mock crime CIT, autobiographical CIT) a consistent finding is that probe items elicit longer reaction times and higher error rates than irrelevant items for participants who possess guilty knowledge (Seymour et al., 2000; Verschuere et al., 2010; Visu-‐Petra et al., 2012; Noordraven & Verschuere, 2013; G. Visu-‐Petra, Miclea, Bus, & L. Visu-‐Petra, 2014).
Another frequently used reaction time based deception test is the SLT (Spence et al., 2001).2 The visual SLT consists of 36 questions starting with the stem “In the course of today have you done any of the following?” after which certain common daily actions appear on the screen in succession (e.g. Made your bed). Participants are instructed to answer with either “yes” or “no” by pressing the corresponding key on a keyboard. The words yes and no appear at the bottom of the screen below each question as a reminder of which buttons to use for which response. They are presented either in red or green indicating whether the participant is to lie or tell the truth about the question. After an initial template round (in which participants respond truthfully to all items) each question is presented twice; once requiring a lie response and once requiring a truthful response. Reaction times and accuracy scores are recorded (Spence et al., 2001). The comparison of interest is between the Lie and Truth trials (Lie-‐Truth difference, the lie effect). Typically, participants show an increased SCR (Furedy et al., 1988) and longer reaction times (Spence et al., 2001) to Lie trials as compared to Truth trials.
Most recently, the aIAT (Sartori et al., 2008) was introduced. The standard IAT measures the strength of associations through reaction times on classification tasks (Greenwald, Nosek, & Banaji, 2003). The aIAT can be used to assess the truth about autobiographical information (Sartori et al., 2008; Verschuere, Prati, & De Houwer, 2009). The test consists of four categories: True sentences (with known ground truth), False sentences (sentences of clear falsities), Guilty sentences (sentences about real autobiographical events), and Innocent sentences (sentences about fictitious autobiographical events). The categories are combined in such a way that there is a clear distinction between the response patterns for true and false autobiographical sentences. In the congruent block guilty sentences are combined with true sentences and innocent sentences are combined with false sentences. In the incongruent block guilty sentences are combined with false sentences and innocent sentences are combined with true sentences. Generally, participants are slower to respond in the incongruent block (as compared to the congruent block), because this block is incompatible with their implicit (congruent) association (Sartori et al., 2008; Frost et al., 2010). The accuracy of the aIAT was assessed in a meta-‐analysis conducted by Agosta & Sartori (2013). They found a high overall accuracy. Repetition of the aIAT did not seem to decrease the overall accuracy of the test. However, the overall effect size of the aIAT was moderate.
The deception tests mentioned above consistently and repeatedly showed that lying took longer than telling the truth. These findings point out the effectiveness of reaction time based deception tests (e.g. Verschuere et al., 2010). Although it has been
2 The predecessor of the SLT (the differentiation of deception paradigm, DDP) was proposed by Furedy et al. (1988).
argued that reaction time measures are easily deliberately altered by the examinee (e.g. Farwell & Donchin, 1991) research has suggested that fast responses (of 800 ms or less) are very difficult if not impossible to intentionally manipulate (e.g. Seymour et al., 2000).
Altering the cognitive cost of lying
Earlier research has provided evidence indicating that the cognitive cost of lying can be deliberately altered so as to either enhance or impair lie detection. By enhancing the cognitive cost of lying, the behavioral differences between lying and telling the truth within an individual will be magnified, making it easier to distinguish between the two (Vrij et al., 2008) So far, many attempts have been made at altering the cognitive cost of lying. In this section we will describe some of the methods that were used.
In order to increase the difficulty of lying Vrij et al. (2008) instructed participants to recall a story in reversed order. The participants were divided into two groups: innocent or guilty. The “innocent” participants participated in a staged event during which they encountered a person who came looking for their wallet. After finding the wallet in the room the person in question claimed that a £10 note was missing from the wallet. The participant was informed that he/she would be questioned about the missing money. The “guilty” participants committed a mock crime (stealing £10 from the aforementioned wallet). During questioning these participants had to deny the mock crime by using the innocent scenario as their alibi. To increase the cognitive load of this task all participants were instructed to recall the story in reversed order. Since the guilty participants did not really carry out the innocent scenario, telling the story in reversed order should be particularly cognitively constraining for this group. Deceitful parts of the stories were much easier to detect when the story was told backwards instead of in chronological order. This effect was expressed by several deception cues apparent in the guilty participants (i.e. more speech hesitations, slower speech rate, more fidgeting with hands and feet, more speech errors, increased eye blinking) as compared to the innocent participants. These deception cues were less apparent when the stories were told in chronological order. The researchers concluded that recalling an alibi in reversed order would be particularly difficult for liars since they already experience an increased cognitive load due to the construction of the lie (alibi) itself. These results show that increasing cognitive load will indeed impair the ability to lie. Vrij et al. (2012) replicated these results using similar methods. Similar results were found using different methods. For instance, it has been shown that asking unanticipated questions can also amplify the differences between liars (who rehearsed their story) and truth tellers (who give an honest account of the events when asked about them) (Vrij, Mann, Leal, & Fisher, 2009). Furthermore, the instruction of examinees to keep eye contact with their interrogator can add significantly to the cognitive cost of lying (Vrij, Leal, Mann, & Fisher, 2010).
The research described above indicates that the cognitive cost of lying can be enhanced. This is an important notion when fabricating deception tests since deception tests aim to pinpoint differences between liars and truth tellers. However, it is possible that the cognitive cost of lying can also be diminished. It has been noted that lying is not always more difficult than telling the truth (McCornack, 1997, cited in Vrij et al., 2008). In the event that someone fakes a deception test (i.e. appearing innocent when being guilty), the difference between truth tellers and liars will diminish making it more difficult to catch liars. Deception paradigms should be able to indicate, if not prevent such occurrences. As mentioned briefly in the introduction, replicating Verschuere et al. (2011), Van Bockstaele et al. (2012) manipulated the truth and lie proportions in a Sheffield lie test in order to examine the general malleability of the cognitive cost of
lying. When participants were in the “frequent lie condition” (75% lie trials, 25% truth trials), lying became easier (i.e. the difference in reaction times and accuracy scores between truth and lie trials was smaller), whereas when they were in the “frequent truth condition” (75% truth trials, 25% lie trials) lying became more difficult. These results indicated that the cognitive cost of lying is malleable through training of deception skills, albeit only for practiced items (Van Bockstaele et al., 2012). Although this training effect did not transfer to unpracticed items, learning effects have been shown to transfer between working-‐memory tasks, provided that these tasks are mediated by the same neural system (Dahlin, Neely, Larsson, Bäckman, & Nyberg, 2008). According to Osman, Channon, and Fitzpatrick (2009) deception tests require similar cognitive processes as the standard Stroop task, since both tests require the participant to overcome interference of incongruent items. To test this hypothesis, they compared the Stroop scores between participants in a lie condition (100% lie trials) to the Stroop scores of participants in a truth condition (100% truth trials). Participants in the lie condition showed enhanced performance on the Stroop task. Hence, the researchers concluded that deception experience could indeed transfer to performance on cognitive tasks with a similar neural basis. Whether this transfer effect was temporary or permanent remains to be seen.
The research described in this section shows clear evidence for the malleability of the cognitive cost of lying. However, it is not clear whether this cost can be altered permanently through training. Furthermore, individual differences in deception skills were not taken into account as of yet.
Individual differences
Not much is known about the individual differences in the ability to successfully deceive others. It is possible that not everyone is affected by the cognitive cost of lying to the same extent. For instance, certain individuals may already be trained at deception through their work (e.g. poker players, actors) or because they experience a constant urge to lie in daily life (e.g. pathological liars, individuals cheating on their spouses). Moreover, it appears as though lying itself is only more cognitively constraining than telling the truth under specific circumstances. For instance, according to Vrij et al. (2008) the truth needs to be easily accessed from memory, in order to elicit an automatic response that is difficult to inhibit. Furthermore the researchers noted that if the individual is motivated to be believed, credibility will be taken less for granted, resulting in the deceiver trying harder to monitor one’s own behavior and that of the conversational partner. This process will add to the cognitive cost of lying. Individuals with deception experience may respond to these cognitively constraining processes in different ways than people without such experience. Moreover, some individuals may just be “good liars”. According to Vrij, Granhag, and Porter (2010) the best liars possess several important features that may cause them to be able to lie with more ease than others. For instance, they argued that some individuals might have a more trustworthy appearance than others, making it easier for them to get away with deception. This in turn might lead them to experience less emotions during the act of deception.
The best liars are good actors who can behave in a seemingly trustworthy way (Vrij et al., 2010). Actors may have an enhanced sense of self-‐other awareness (Thomson & Jaque, 2012), which may allow them to monitor their own behavior as well as the behavior of others with more ease. When professional actors take on a role, they extensively elaborate on the script through imaginative embellishment (Noice & Noice, 2006), allowing them to “relive” an event that has not really happened to them.
Additionally, they adopt the perspective of the character they are portraying, in part by relating the character’s thoughts and feelings to their own mental state (Noice & Noice, 2006). Since, the actors fully adopt the role they have to play, it may become a second “truth”, making it easy for them to act natural in regard to the role. This in turn may result in actors successfully faking an impulsive response when confronted with information from their previously prepared role, so as to appear convincing. It seems that in order to appear honest a liar must have certain nonchalance toward lying. By integrating the role with their own identity, actors may take their credibility for granted just as any innocent person would, making them appear truthful when they are in fact guilty of the act of deception.
To assess the relationship between private self-‐awareness and deception ability A.K. Johnson et al. (2005) instructed non-‐professional actors to portray varying levels of deception on film. A questionnaire about private self-‐awareness was administered beforehand. There were three conditions for the videos: non-‐deception, faking good (exaggerating positive biographical information), and faking bad (exaggerating negative biographical information). Undergraduate students were presented with the video footage after which they rated the believability of the actors. They were instructed to both indicate whether the actors were lying or telling the truth and how confident the actors seemed to be. Actors with higher overall private self-‐awareness were more successful in deceiving the undergraduates. It has been proposed that enhanced self-‐ awareness may improve one’s Theory of Mind (e.g. Gallup, 1982, cited in A.K. Johnson et al., 2005) possibly resulting in greater deception skills (A.K. Johnson et al., 2005). It is possible that actors in general have enhanced self-‐awareness and thus are better at deceiving others. After all, they are specifically trained at pretending to be someone else. According to Thomson and Jaque (2012) the ability of actors to create characters would be beneficial for their psychological self-‐other awareness. They found that although actors seemed more psychologically vulnerable, they were indeed more psychologically self-‐other aware than non-‐actors.
The current experiment
The research question we aimed to answer with the current experiment was whether the cognitive cost of lying is shaped by an individual’s experience with deception. For this purpose we chose to compare the deception skills of a group with deception experience (actors) to a group without such experience (non-‐actors). We specifically expected actors to be better at concealing their own identity compared to non-‐actors in part because of their experience with pretending to be someone else. Deception skills were measured using three reaction time based autobiographical deception tests: the autobiographical CIT (aCIT), the aIAT, and the autobiographical SLT (aSLT).
Methods
Participants
Eleven actors (3 males), and 22 non-‐actors (4 males) participated in this study (23 of them received €7,50 for their participation, the rest of the participants cooperated voluntarily). Participants were classified as “actors” if they had at least 2 years of acting experience. The participants were recruited through flyers, posters, and through the Internet.
Apparatus
All stimuli were presented on a PC using Inquisit 4 software (2013).
Procedure
Profile
Up to one week prior to testing, participants received a profile containing fictitious autobiographical information (e.g. Your name is Sam; You were bitten in your leg by a Dalmatian when you were little, see Appendix A). They were instructed to memorize this profile so that they could adopt the fictitious identity on the day of testing. Participants were not informed about the questioning format (used on the day of testing) beforehand.
Memory check
On the day of testing, participants’ knowledge of the fictitious identity was first tested by means of a questionnaire (see Appendix B). If participants made mistakes on the questionnaire, they were instructed to read through the profile thoroughly within the next five minutes. If participants still made mistakes after this, the experimenter went over the answers with the participant, after which testing began.
Deception tests
Autobiographical Concealed Information Test (aCIT)
The aCIT (e.g. Rosenfeld, Biroschak, & Furedy, 2006; Verschuere et al., 2010) consisted of 1 practice block (12 trials) and 2 test blocks (108 items per block). Each test block contained 9 target items (about the fictitious identity), 9 probe items (about the real identity of participants), and 36 irrelevant items (see Appendix C), presented in random order. Hence, the total amount of test trials was 216. Inter stimulus intervals were 500, 800, or 1000 milliseconds in randomized order, to prevent participants from preparing their responses and showing rhythmic response patterns (Seymour et al., 2000). On top of the screen the sentence “Do you recognize this stimulus?” appeared. Participants were instructed to hide their true identity during this test by answering with “yes” to all target items and with “no” to all other items (probe items and irrelevant items). Answers were given by pressing the corresponding key (“z” for yes and “m” for no, or vice versa) on a keyboard. The responses were to be made as quickly yet as accurately as possible.
Autobiographical Implicit Association Test (aIAT)
The aIAT consisted of sentences about the true identity of the participants, the fictitious identity from the profile, or sentences with known ground truth that were either true or false (see Appendix D). There were four corresponding categories (Self, Other, True, and False, respectively). Sentences were presented in the middle of a computer screen. Participants were instructed to place each sentence in the corresponding category presented in the top corners of the computer screen. Answers were given by pressing the corresponding key on a keyboard (“e” for categories presented on the left of the screen, “i” for categories presented on the right of the screen). The test consisted of three blocks: attribute practice, congruent, and incongruent. In the attribute practice block (20 trials) participants were familiarized with the buttons. In the congruent block (20 practice trials, 40 test trials) the category True was combined with the category Self (both requiring button press “e”) and the category False was combined with the
category Other (both requiring button press “i”). In the incongruent block (20 practice trials, 40 test trials) the category True was combined with the category Other (both requiring button press “e”) and the category False was combined with the category Self (both requiring button press “i”). Hence, the congruent block required a truth response (i.e. real autobiographical sentences are true; fictitious autobiographical sentences are false) whereas the incongruent block required a lie response (i.e. fictitious autobiographical sentences are true; real autobiographical sentences are false). The inter stimulus interval was 250 milliseconds. There was no response deadline. All trials required a correct response. If participants made a mistake, a red cross would appear at the bottom of the screen, indicating that the participant had to give the correct response in order for the test to continue.
Autobiographical Sheffield Lie Test (aSLT)
This test consisted of two item types: Other Identity (fictitious autobiographical items referring to the profile; see Appendix A) and Own Identity (autobiographical items referring to the participant). There were two practice blocks containing 36 unrelated control items with known ground truth in randomized order borrowed from Verschuere, Prati, & De Houwer (2009) (see Appendix E). The purpose of the practice blocks was to familiarize participants with the buttons. Participants were presented with either an “L” or a “T” indicating whether they had to lie or tell the truth about the upcoming sentence. The sentence was then presented in the middle of the computer screen. The letter (“L” or “T”) remained on the computer screen until the next trial to ensure that participants would not forget whether they had to lie or tell the truth about the presented sentence. Answers were given by pressing the corresponding key on a keyboard (“4” for yes and “6” for no or vice versa). During the first practice block (12 trials) the order of truth and lie trials was fixed. In the second practice block (16 trials) this order was scrambled in order to ensure that participants could no longer anticipate whether they should tell the truth or lie to the stimulus. The test block consisted of 72 truth trials and 72 lie trials in randomized order. During truth trials participants had to answer truthfully (i.e. say yes to all Own Identity items; say no to all Other Identity items). During lie trials participants had to deny their own identity (i.e. say yes to all Other Identity items; say no to all Own Identity items). The time between the presentation of the cue and the presentation of the stimulus ranged between 500 and 1500 milliseconds in order to prevent rhythmic responding (Seymour et al., 2000). There was no response deadline; hence a response was required in order for the next stimulus to appear.
Results
Participants
All reported results for the aCIT are with n = 32 (7 males), for the aIAT with n = 33, and for the aSLT with n = 31 (7 males). The ages of the participants ranged between 19 and 62 years old. All except two participants (1 actor, 1 non-‐actor) spoke Dutch.3 Acting experience for the non-‐actors ranged between 0 hours and 250 hours, whereas acting experience among the actors ranged between 500 hours and 10000 hours (see Table 1).
Table 1
Means and Standard Deviations Per Variable for Actors and Non-‐actors with the p-‐values for the Differences Between Actors and Non-‐actors.
Actors non-‐actors p-‐value total n = 11 n = 22 (t-‐tests) n = 33 Age 38.36 (16.08) 28.91(11.01) .06 32.06(13.44) Gender* (% male) 27.3% 18.2% .56 21.2% Acting experience 3559.09(4193.26) 28.77(55.71) <.01** 1205.55(2890.15) (hours) Fake Identity *** 48.86(89.32) 4.53(6.50) .04 16.46(48.45) (frequency) Preparation time 3.09(2.77) 2.41(5.58) .71 2.64(4.79) (days)
* For gender Spearman’s Chi squared (X2) was used.
** Cohen’s f = .52.4
*** This measure indicates the participants’ estimates of the total amount of times they pretended to be someone else. Due to incomplete questionnaire data, 7 participants were missing for the Fake Identity measure.
aCIT
All practice trials were excluded from the analyses. The response latencies, and error rates were analyzed separately using a 2 by 2 mixed analysis of variance (ANOVA) with Acting (Actor versus Non-‐actor) as a between-‐subjects and Item type (Probe versus Irrelevant) as a within-‐subjects measure.
Preliminary analyses
A correlational analysis was conducted in order to examine whether Age was a confounding factor. If Age did not correlate with both Acting (Actor vs. Non-‐actor) and the CIT effect (RT Probe – RT Irrelevant, RT-‐PI), it would not be taken into account in the main analyses. There was a significant correlation between Age and the CIT effect (r = .35, p < .05. The correlation between Age and Acting was marginally significant, r = .34, p = .06. The correlation between Age and the amount of Acting experience in hours however, was significant, r = .62, p < .01. Despite the fact that the correlation between Age and Acting was only marginally significant, age was included as a covariate in the main analyses due to the significant correlation with the amount of acting experience in hours.
The reliability of the aCIT was assessed by means of the odd-‐even method used in Noordraven and Verschuere (2013). The split-‐half correlation of RT-‐PI was r = .75.
4 According to Cohen (1992) f-‐values from .10 represent small effects, values from .25 represent medium effects, and values from .40 represent large effects.
Main analyses
Reaction Times
Due to a preset response deadline of 800 ms there was no outlier analysis conducted. All time-‐outs were excluded from the analyses. Errors were only discarded for the response latency analyses.
There was a main effect of Item Type, F(2, 58) = 27.68, p < .01, f = .96. Participants responded slower to probe items (M = 600, SD = 45), than to irrelevant items (M = 535, SD = 43). However, the interaction between Item Type and Acting was not significant, F(2, 58) = 1.09, p = .34, f = .04. The difference in reaction times between actors and non-‐actors was also non-‐significant, F(1, 30) = .55, p = .46, f = .02 (see Figure 1).
Figure 1. Mean Reaction Times and Standard Errors Per Item Type of the aCIT for Actors and Non-‐actors.
Error rates
There was a main effect of Item Type, F(2, 58) = 9.41, p < .01, f = .33. Participants had a higher error rate for probe items (M = 3.13, SD = 3.87) as compared to irrelevant items (M = .81 SD = 1.06). The interaction between Item Type and Acting was non-‐significant, F(2, 58) = .70, p = .50, f = .02. The differences in error rates between actors and non-‐ actors were also non-‐significant, F(1, 29) = .21, p = .65, f = .01 (see Figure 2). 5
5 In line with the suggestion made by Simmons, Nelson, and Simonsohn (2011) we checked whether excluding Age as a covariate would yield similar results. When Age was not included as a covariate for both the reaction time analyses and the error rate analyses the main effects of item were still significant. The interactions between Item and Actor were still non-‐significant. The differences in reaction times and error rates between actors and non-‐actors were also still non-‐significant.
400 450 500 550 600 650 Probe Irrelevant R ea ct io n T im es (m s) Item Type Actors Non-‐actors
Figure 2. Mean Error Rates and Standard Errors Per Item Type of the aCIT for Actors and Non-‐actors.
Supplementary analyses
Because of the small sample size, we also conducted dimensional analyses, neglecting the factor actor versus non-‐actor, and relating acting experience with the RT-‐PI across groups. A partial Pearson correlation analysis was conducted controlling for age. The correlation between the RT-‐PI and acting experience was non-‐significant, r = .03, p = .90. These results indicate that actors did not hide their own identity better than the non-‐ actors did.
In order to examine whether Preparation time (during which participants could rehearse their fictitious identity) had a beneficial effect on deception skills a dimensional analysis was conducted relating preparation time (in days) with RT-‐PI. A partial Pearson correlation was conducted controlling for Age. The correlation between the RT-‐PI and preparation time was non-‐significant, r = -‐.18, p = .34, indicating that participants’ deception skills did not benefit from rehearsing their fictitious identity.
Conclusion
In line with what is typically found, the above results indicate that lying is more difficult than telling the truth. Against our expectations however, actors did not show superior deceptive skills as compared to non-‐actors for this test. Furthermore, the amount of preparation time did not have any effect on participants’ performance on the aCIT.
aIAT
The attribute practice block was excluded from the analysis. Since there was a built-‐in penalty for error trials (a correct response was required after an error) adding a penalty to these trials was no longer necessary. This built-‐in penalty allowed us to include error latencies in the analyses (Greenwald et al., 2003). The aIAT was scored using the improved scoring algorithm (D1) described in Greenwald et al. (2003). This algorithm consisted of the following adjustments. The practice blocks were included in the analyses. Latencies over 10,000 ms were discarded.6 Mean reaction times (RTs) were calculated per block. The RT differences between incongruent and congruent practice (block 6 – block 3) and test blocks (block 7 – block 4) were calculated. Pooled standard deviations (SDs) were calculated for the practice (SDs of block 6 and block 3 combined) and test blocks (SDs of block 7 and 4 combined). The difference scores for practice and
6 There were no latencies below 400 ms, hence removal of fast trials was not necessary.
0 1 2 3 4 5 6 Probe Irrelevant Er ro r ra te s (% ) Item Type Actors Non-‐actors