The validity of the concealed information test : using exemplar items to reduce the effects of leakage

The Validity of the Concealed Information Test: Using Exemplar Items to Reduce the Effects of Leakage

Elisa Cordesius 10366083

Graduate School of Psychology Research Master’s Psychology

University of Amsterdam

Supervisor: Bruno Verschuere Daily supervisor: Linda Geven

Abstract

The Concealed information test (CIT) is a memory detection test that examines whether the examinee has specific knowledge of a crime that only a guilty individual should have. In this study, the RT-CIT will be used, which uses reaction time (RT) to detect concealed

information. A major limitation of the CIT is the phenomenon of leakage of information, in which case innocent individuals could obtain crime-related information. To prevent false positives and counteract the effects of leakage, information that has been leaked to the suspect can be previewed and if necessary can be excluded from the CIT. However, it has not yet been investigated if the nature of CIT items can be altered to better distinguish guilty from informed innocents, thereby improving the validity of the test. Therefore, a modified version of the CIT (i.e. leakage CIT) will be investigated, in which exemplar items are asked. This would offer a solution in case that the leaked information is relatively superficial, and it is plausible that the perpetrator has knowledge of more detailed information than a

knowledgeable innocent individual. The results showed that concealed information, when encoded on an exemplar level, could be detected by the leakage CIT, both immediately after the learning phase and after a one-week delay. These findings suggest that the leakage CIT might be a valuable tool in reducing the effects of leakage.

Introduction

In 1994 three teenagers, known as the West Memphis Three, were convicted for a Satanic murder of three boys. The teenagers seemed to have knowledge about the crime that the prosecution regarded as too close to the facts. However, the three were found innocent and were released in 2011. The information about the crime was later suspected to have originated from police leaks. This case illustrates the issue of leakage of information. In this report, a technique will be discussed that enables us to reveal certain knowledge that individuals hold and more importantly, how we can refine this technique in order to counteract leakage.

People seem to be particularly bad at judging deception, as even professionals (e.g. in mental health or law enforcement) perform largely at chance level (Bond & DePaulo, 2008; Vrij, 2008). In order to help us detect deception, lie detection tests have been developed since the early 20th century. In 1959 a new type of test was introduced to detect deception, called the Guilty Knowledge Test (Lykken, 1959). Today, this test is better known as the Concealed Information Test (CIT). As opposed to other tests that focus on detecting lies, assuming an increased stress response resulting from lying, the CIT focuses on finding out if the examinee has specific knowledge of a crime that only a guilty individual would have. Several measures could be used during the CIT, one of which is response latency-based measures, which uses reaction time (RT) to detect concealed information. In the RT-CIT the examinee is presented with a sequence of items. It involves crime-related familiar items, so-called probe items, and unfamiliar irrelevant items. Subsequently, it is assessed whether the suspect has knowledge about the crime by comparing the RT of the probe and irrelevant items, in which a larger RT to the probe relative to the irrelevant item signifies recognition of the probe item.

The RT-CIT is based on the prediction that deceptive responses produce a larger RT than truthful responses, because a greater cognitive effort is needed to fabricate a lie

DePaulo, & Rosenthal, 1981). According to Seymour and Seifert (1998), this is a result of dual process of recognition. They argue that there are two types of judgements; one that is fast and automatic, based on familiarity, and one that is slow and deliberate, in which memories have to be retrieved. The latter applies to guilty individuals for probe and target items,

because both are familiar, but then it has to be determined if the item has to be denied to avoid detection. The initial response would be to acknowledge recognition, but this response has to be inhibited to appear innocent, thus producing a larger RT.

The CIT relies on the notion that only guilty individuals are aware of the key items of a crime. However, the CIT merely detects the memory of examinees, and not the source of the memories. This forms a potential issue, because innocent individuals could obtain crime relevant details for example through mass media, interrogation or witnessing the crime (they are termed informed innocents). This is known as leakage of information, and undermines the potential of the CIT to discern guilty from innocent individuals. Consequently, this could increase false positive rates (Bradley, Barefoot, & Arsenault, 2011), thereby decreasing validity of the CIT. In forensic applications, this could mean that informed innocents could be falsely accused of the crime. We could imagine a situation in which an individual could obtain information about a crime but still be innocent, for example, if a bystander would witness a murder. The innocent witness could have seen that the murder weapon was a gun. If this innocent bystander would be a suspect in the case, and if ‘gun’ were to be used as an item on the CIT, this person could unjustly be indicated as being guilty of the crime.

To circumvent the effects of leakage, precautions can be taken. When the examiner sets up the CIT, he first assesses if certain information has been leaked to the suspect, by talking to the interrogation team (if the suspect has been interrogated) and to the suspect himself, and inquires what has been revealed about the crime in mass media (Osugi, 2011). The CIT items could also be previewed to the examinee before the test. This way, innocent

suspects could declare if they have knowledge about one or more of these items, and if this is the case, it must be determined how this awareness was established. This strategy may decrease detection efficiency due to habituation, but Verschuere & Crombez (2008) found no evidence of a reduced validity of the CIT. The items that have been leaked can then be excluded from the CIT. However, considering the importance of including central crime details on the CIT (Peth, Vossel, & Gamer, 2012), this solution is not ideal.

Previous research has also attempted to reduce effects of leakage by modification of the CIT. Bradley and Warfield (1984) introduced the Guilty Actions Test, in which the wording of the questions inquired about actions rather than knowledge. It was assumed that informed innocents would not consider it as being deceptive when denying crime-relevant knowledge, because they did not engage in the activity. For example, instead of “Do you know that the murder weapon was a gun?” (CIT) they are asked, “Did you murder the victim with a gun?” (GAT). Although the Guilty Actions Test showed some promising results in reducing false positive rates (Bradley, MacLaren, & Carle, 1996), it fails to consistently differentiate between informed innocent and guilty individuals (Gamer, 2010). Another method includes using target items to avert the attention of informed innocents from key items (Ben-Shakhar, Gronau, & Elaad, 1999). This would aid in the discrimination between guilty and informed innocents, because the crime-relevant knowledge would be less deeply encoded in innocent individuals, thus attracting their attention more in comparison with guilty individuals.

In all the attempts that tried to reduce the effects of leakage, it has not yet been investigated if this can be accomplished by altering the nature of the items of the CIT. It has been demonstrated that when a probe item is replaced by a superordinate item (i.e. an item of a higher category), the responsiveness on the CIT is significantly reduced for guilty

replaces the item ‘apple’. Thus, an item has to be specific enough for a guilty individual to get an adequate response on the CIT. However, it is not recommended to use items that are too specific, as details will probably not be memorized (Osugi, 2011). Therefore, it is essential to strike an optimal balance regarding the specificity of CIT items. Let us consider the earlier mentioned example with the leaked information about the gun; we know that this revolves around relatively superficial information (i.e. that the murder weapon was a gun). However, the perpetrator could have more detailed information, for instance that the gun was a revolver. This information could be used to create a modified version of the CIT, provided that the information that was leaked is relatively superficial (e.g. ‘gun), termed categorical information, and it is plausible that the perpetrator has knowledge of more detailed

information (e.g. ‘revolver’), termed exemplar information. For similar cases, we propose a modified version, as we call the leakage CIT, in which exemplar items will be used. This will be compared to the classical CIT, in which categorical items, that are the superordinate of the exemplar items, will be used. Nevertheless, the leakage CIT has a potential risk of using items that are too specific, and could therefore raise the false negatives rate. Accordingly, the

purpose of this research is to examine the trade-off of false positive and false negative rates, and searching for the optimal balance between the two.

Concealed knowledge can be detected if a CIT item is recognized, and thus, there has to be a match between how an item is encoded and how it is asked on the CIT. Therefore, the relationship between encoding an item (i.e. whether an item is initially encoded at an

exemplar or categorical level) and how it is queried on the CIT will be investigated. Four conditions were created: categorical item encoded and categorical item on the CIT, exemplar item encoded and exemplar item on the CIT, categorical item encoded and exemplar item on the CIT and exemplar item encoded and categorical item on the CIT. These conditions will be given a two-letter abbreviation, in which the first letter represents the level of encoding (either

categorical or exemplar), and the second letter stands for how it was asked on the CIT (either categorical or exemplar). For instance, ‘CE’ means categorical item encoded and exemplar item asked on the CIT. We can assess the validity of the leakage CIT as compared to the classical CIT, as all possible scenarios of a realistic setting are simulated. That is, whether the suspect has knowledge on an exemplar or categorical level and if this is asked on an exemplar (leakage CIT) or categorical level (classical CIT). Furthermore, in realistic settings it rarely happens that the criminal is questioned immediately after the crime. Nahari and Ben-Shakhar (2011) demonstrated that with time delay the response to items on the CIT is reduced. They argued that the way the participants became aware of the information has an effect on memory, and thus recognition of the items. Because time delay could influence memory and thus affect how the item was initially encoded, the effect of time delay will be studied. Previous research has also shown that there is an optimal categorical level in which information is best retained in memory (Pansky & Koriat, 2004). Due to the fact that

exemplar items might be less well remembered than categorical items, it will be investigated whether the decline in memory facilitates the recognition of categorical items, when exemplar items were encoded (i.e. in the EC condition).

It is hypothesised that there will be an interaction effect between encoding and testing. More specifically, it is expected that there will be a significant difference between probe and irrelevant items (with a larger RT to the probe item) for the CC condition and the EE

condition, because in both cases there will be recognition of the probe item. On the other hand, no significant differential response is expected for the CE condition, due to the fact that the encoded item is of a higher category, and hence, no recognition of the exemplar CIT items. For the EC condition, a higher differential response as compared to the CE is expected, because when the participant is presented with the matching categorical item of the encoded exemplar item on the CIT this might spark recognition. Therefore, a significant differential

response is expected. Secondly, it is hypothesised that a delay will reduce RT differences between probe and irrelevant items, as it was previously demonstrated that time delay leads to a reduced response for probe items (Nahari & Ben-Shakhar, 2011). Although a reduction in differential response is expected in the delay condition, this is not expected to revoke the significance in the CC, EE and EC conditions. Figure 1 shows a graphical representation of the expectations in the direct condition and Figure 2 of the delay condition.

Figure 1. Expected results: mean RT (ms) to probe and irrelevant items for the conditions CC, EE, CE and EC in the direct condition.

Figure 2. Expected results: mean RT (ms) to probe and irrelevant items for the conditions CC, EE, CE and EC in the delay condition.

450 460 470 480 490 500 510 520 530 CC EE CE EC Mea n R T ( m s) Condition Probe Irrelevant

*

*

*

*

*

*

Method Participants

A total of 93 participants were recruited for this study. They were recruited through advertisement of the experiment in various ways: advertisement on multiple websites,

including the LAB website of the University of Amsterdam; by distributing flyers and display of posters; and by inviting friends and acquaintances to participate. Participants would be rewarded with either course credits or 10 euros (and a possible bonus of 5 euros or an

equivalent amount of course credits). Data of 10 participants were excluded, leading to a final sample of 83 participants (61.4% female; Mage = 26.48 years; SDage = 10.60), of which 46 were tested immediately (67.4% female; Mage = 25.83 years; SDage = 9.79) and 37 were tested after a one-week delay (54.1% female; Mage = 27.30 years; SDage = 11.62). The same

exclusion criteria were used as in Kleinberg and Verschuere (2015, 2016). Participants with a response error rate of 50% or more on any of the three item types (i.e. probe, target and irrelevant items) were excluded from the analysis. Furthermore, participants that had less than 50% of the trials left after exclusion criteria were applied were excluded as well.

Procedure

The procedure consisted of a learning phase, followed by administration of the CIT, and took about one hour.Participants were invited to come in pairs, to make sure that the procedure was similar between the two participants in the direct and delay condition. The two participants were told that they had to plan a robbery together. They were given the plan of the robbery, in which was described how they were going to rob a bank together. The plan contained eight key items, which were indicated by capitalized letters. There were four versions of the plan; every version included four categorical and four exemplar items. For example, version one and two stated “You met at a SPORTS club” (categorical item), where version three and four said “You met at a VOLLEYBAL club” (exemplar item). The four

conditions CC, EE, CE and EC were randomized over the versions. To illustrate, in version one the participant would encode “SPORTS” and get “VOLLEYBAL” on the CIT (i.e. condition CE), in version two encode “SPORTS” and get “SPORTS” on the CIT (i.e.

condition CC), in version three encode “VOLLEYBAL” and get “SPORTS” on the CIT (i.e. condition EC), and finally in version four encode “VOLLEYBAL” and get “VOLLEYBAL” on the CIT (i.e. condition EC). For another pair of items, the conditions would be distributed differently, creating a counterbalanced design. In Appendix A one of the versions is included as an example.

The experimenter read out the plan of the robbery to the participants and then instructed them to first read it for themselves, then read it out to each other, write down the key items and question each other about the items. In order to stimulate deeper encoding of the items, the participants were asked to discuss how long they knew each other and who was driving the car that would be used to flee the scene of the crime. After ten minutes, the participants were questioned to make sure that the items were properly encoded. In addition, they were given the plan of the robbery with blank spaces where all the key items had to be correctly filled in. Subsequently, they had to write down all the eight key items on a blank paper. If they did not get all the items right, they got additional time to learn the items. After the learning phase, participants were randomly assigned to either do the CIT immediately (direct condition) or to do the CIT after one week (delay condition).

In the second part of the experiment the participant was informed that he would be partaking in a lie detection test. The objective was given to try to beat the test by learning a set of target items. The participant was instructed to press the green ‘YES’ button for every target word, as a way of confirmation of knowledge, and the red ‘NO’ button for every item of the robbery (probe item) and all other irrelevant items, thereby denying knowledge of the items. First, three practicing sessions were conducted to assure that the participant was

accustomed to the speed and requirements of the task, and was aware of the target items. In the first practice session, the participant could only proceed to the next session if their target accuracy was at least 50%, otherwise the practice session had to be repeated. In the second practice session an additional requirement was that their mean RT had to be faster than 800ms, otherwise the next word would automatically be offered. The same was true for the third practice session, but the participant would now also be notified with the message “too slow” if they did not respond within 800ms. After the practice sessions, the CIT was

administered, which had the same format as the third practice session. Participants were told they could earn an additional bonus of 0.5 course credit or 5 euros if they would ‘beat’ the test. They were rewarded with a bonus if they did not have to be excluded and if their cohen’s d effect size was lower than 0.2. The cohen’s d was used to get a within-subject CIT score, and was calculated with the following formula: MRT(probes) – MRT(irrelevants) / SDRT(irrelevants). After

the CIT, participants got a free recall test, in which they had to write down all the items from the robbery they still remembered, and after that a recognition test, in which they had to indicate from a list of items which ones they recognized as items from the robbery plan. The RT-CIT

In the RT-CIT the examinee was presented a sequence of items, among which probe and irrelevant items. In addition, target items were used in the current experiment. For every one of the eight probe items, there was an equivalent target item that could replace the probe item in the robbery plan (e.g. when the probe item was ‘Fiat’, the target item could be

‘Volkswagen’. Target items ensured that attention must be paid to the task so that items were properly processed, as it prevents mindlessly pressing the same button.

The required equipment for testing included a single computer, with Inquisit software installed on it. With this software the stimuli were presented and the RTs were recorded in milliseconds. One trial consisted of a brief presentation of a stimulus (1500ms or until button

press), followed by an inter-trial interval of a randomly varied period (250, 500 or 750ms). A full CIT session comprised a total of 448 irrelevant, 112 probe and 112 target items. Table 1 gives an overview of the types of stimuli and summarizes the design of the test.

Table 1

Types of stimuli and study design Stimulus type Relative

frequency

Description Instructions Stimulus evaluation

Probe 1/6 Relevant to the

robbery

Press red button

Rare and relevant Irrelevant 4/6 Irrelevant to task

and robbery

Press green button

Frequent and irrelevant

Target 1/6 Relevant to task Press green

button

Rare and relevant Data-analysis

For the analysis of the data, the RT of target items was not considered, as they only served as a way to ascertain that participants paid attention to the task. Thus, we only

evaluated the RTs to probe and irrelevant items. The following analyses were also performed for error rates, in order to back up the RT results. Similar results were expected for the error rates; as for the participants it is a trade-off between responding either quickly or accurately. The data were analysed with a 2 x 2 x 2 three-way mixed ANOVA with encoding (exemplar versus categorical) and how it was tested on the CIT (exemplar versus categorical) as within-subjects factors and time at which the CIT is conducted (immediately or after a one-week delay) as a between-subjects factor. Two follow-up one-way ANOVAs were done for both the direct and delay condition in which the differences between the conditions CC, EE, CE and EC were assessed by performing planned contrasts. Lastly, eight t-tests were conducted to identify the difference in RT between probe and irrelevant items, for the conditions CC, EE,

CE and EC in both the direct and delay condition. Again, the same exclusion criteria were used as in Kleinberg and Verschuere (2015, 2016). All trials with errors (i.e. pressing NO for a target item and YES for a irrelevant or probe item) and trials with an RT smaller than 150ms or greater than 800ms will be excluded from the analysis.

Results Reaction times

The 2 x 2 x 2 three-way mixed ANOVA yielded no significant effects, except for the expected interaction effect between encoding and testing, F(1, 81) = 20.00, p < .001, ηp2 = .26. This interaction indicates that how the information was encoded (exemplar versus

categorical) in combination with how the information was tested on the CIT (exemplar versus categorical) had an effect on the differential response.

For the follow-up one-way ANOVA for the direct condition, the assumption of homogeneity of variance was violated, consequently a Welch’s ANOVA was used. This yielded a significant effect of condition (CC, EE, CE or EC) on differential response, F(3, 94.79) = 5.35, p = .002. Planned contrasts revealed that in the CE condition, in which no recognition was possible, there was significantly lower differential response compared to the CC and EE conditions, in which there was recognition, and the EC condition, in which some recognition might be induced as the tested CIT item is the superordinate of the encoded item, t(147.04) = -2.02, p = .045. However, the two conditions in which the encoded items and the tested items were a perfect match (CC and EE) the differential response was significantly higher than in the EC condition, t(114.46) = -3.75, p < .001. Lastly, there was no significant difference in the differential response between the CC and EE conditions, t(82.83) = -0.14, p = .893. The second follow-up one-way ANOVA for the delay condition showed significant effect of condition (CC, EE, CE or EC) on differential response as well, F(3, 144) = 7.60, p < .001. The same three planned contrasts were performed for the delay condition, and

produced similar results. The CE condition had a significantly lower differential response than CC, EE and EC, t(144) = -3.07, p = .003. Furthermore, there was a significantly higher differential response in the CC and EE conditions as compared to the EC condition, t(144) = -3.23, p = .002. There was no significant difference in differential response between the CC and EE conditions, t(144) = -1.72, p = .088.

To compare the RT to probe and irrelevant items in the direct condition for the conditions CC, EE, CE and EC, four paired samples t-tests were conducted. The results are shown Figure 1 and Appendix B. As can be seen in Figure 1, there is a significant difference between the mean RT to probe and irrelevant items for the conditions CC and EE. This finding is consistent with the hypothesis that recognition of the probe item will produce a larger RT.

For the delay condition the same four paired samples t-tests were performed. These results are displayed in Figure 2 and Appendix B. As predicted, the difference between probe and irrelevant items in the EE condition was still significant after the delay. A striking

finding, however, is that participants had a significantly larger RT to irrelevant compared to probe items in the CE condition, in which case no difference was expected. Furthermore, inconsistent with the hypothesis, the differential response for the CC condition after the delay was not significant anymore.

Figure 3. Mean RT (ms) to probe and irrelevant items for the conditions CC, EE, CE and EC in the direct condition. The error bars represent standard errors.

Figure 4. Mean RT (ms) to probe and irrelevant items for the conditions CC, EE, CE and EC in the delay condition. The error bars represent standard errors.

An independent samples t-test was conducted to compare the number of words that were recalled in the free recall and recognition in the direct and delay conditions. In the free recall, there was a significant difference between the direct (M = 7.74, SD = 0.49) and the delay (M = 6.97, SD = 0.18) condition, t(81) = 4.18, p < .001, cohen’s d = 2.00. However, there was no significant difference in recognition between the direct (M = 7.65, SD = 0.80) and the delay (M = 7.68, SD = 0.75) condition, t(81) = -0.14, p = .771, cohen’s d = -0.04.

450 460 470 480 490 500 510 520 530 CC EE CE EC M ea n R T ( m s) Condition Probe Irrelevant

*

*

*

*

Considering that the CIT relies primarily on recognition rather than on free recall, it seems unlikely that this would have affected the results.

Error rates

Overall, similar results were obtained for the error rates. A significant interaction effect between encoding and testing was revealed, F(1,81) = 49.63, p < .001, ηp2 = .38. Besides this interaction, no other significant effects were yielded. This finding supports the results from the RT analysis.

The follow-up one-way ANOVAs produced similar results for the error rates as well. Due to the fact that the assumption of homogeneity of variance was violated for both the direct as the delay condition, two Welch’s ANOVAs were performed. There was a significant effect of condition on differential response in the direct condition, F(3, 79.13) = 8.47, p < .001. The planned contrasts also backed up the RT analysis results: in the CE condition there was a significant lower differential response than in the CC and EE conditions, t(144.58) = -6.00, p < .001, in the CC and EE condition the differential response was significantly higher compared to the EC condition, t(107.98) = -3.23, p = .002, and there was no significant difference in differential response between the CC and EE conditions, t(86.00) = -0.64, p = .522. In the delay condition, a significant effect was yielded of condition on differential response as well, F(3, 65.46) = 5.72, p = .001. Again, similar results were obtained regarding the planned contrasts: there was a significant lower differential response in the CE than in the CC and EE conditions, t(101.81) = -4.69, p < .001, the CC and EE condition had a

significantly higher differential response compared to the EC condition, t(107.42) = -3.24, p = .002, and finally, there was no significant difference in differential response between the CC and EE conditions, t(71.97) = -0.40, p = .689.

Four paired samples t-tests for the direct condition revealed a significant difference between probe and irrelevant items in the CC and EE conditions. These results are shown in

Figure 5 and Appendix C. These results correspond to the findings of the RT analysis. The results of the paired samples t-tests for the delay condition revealed, same as in the RT analysis, a significant difference between probe and irrelevant items in the EE and CE conditions. However, there was also a significant difference in the CC condition, consistent with the hypothesis that recognition produces a significantly larger RT. These results are summarized in Figure 6 and Appendix C.

Figure 5. Mean errors (%) to probe and irrelevant items for the conditions CC, EE, CE and EC in the direct condition. The error bars represent standard errors.

Figure 6. Mean errors (%) to probe and irrelevant items for the conditions CC, EE, CE and EC in the delay condition. The error bars represent standard errors.

0 1 2 3 4 5 6 7 CC EE CE EC E rr o rs ( % ) Condition Probe Irrelevant p < .001 d = 0.88 p < .001 d = 0.81 p = .666 d = -0.08 p = .212 d = 0.21

*

*

*

*

*

Discussion

The aim of this study was to examine whether the leakage CIT might be a good alternative to the classical CIT in the case that the leaked information is relatively superficial and more specific information could be asked on the CIT. The results have shown that in both the direct and the delay condition participants had a higher differential response in the CC and EE condition compared to the EC condition, indicating that when an item is encoded and tested on the same level, this increases the validity of the test. It has also been shown that in the direct as well as in the delay condition there was no significant difference in differential response between the CC and EE condition, suggesting that the differential response in the CC and EE condition was similar. Thus, in a real-life situation, when a suspect would have knowledge on an exemplar level, testing on an exemplar level (i.e. leakage CIT) would

produce similar results as when a suspect had knowledge on a categorical level and was tested on a categorical level (i.e. classical CIT). This result supports the use of the leakage CIT. Furthermore; the hypothesis that there would be a significant difference between probe and irrelevant items for the CC and the EE in the direct condition was confirmed.

In contrast to what was expected, it appears that the differential response for the CC condition was not significant anymore after the delay. However, the reduction in differential response is in line with earlier findings of Nahari and Ben‐ Shakhar (2011), and the effect sizes of the differential responses of the CC and EE condition in the direct condition were relatively small to begin with. Moreover, the error analysis did demonstrate a significant difference between probe and irrelevant items, suggesting that there must have been some recognition of the probes.

Another unexpected result concerns the significant differential response after a delay in the CE condition. In particular, the RT to probe items in this case was lower than the RT to irrelevant items. An explorative analysis was done to examine possible confounding factors

that could explain this result. First, the different versions were considered, but the mean RTs to probes in the versions 1 to 4 (respectively 487.19, 457.31, 453.54 and 502.08) provide no evidence of one version being responsible for the low overall mean RT. Second, there is also no indication that one specific item might have driven the result, as all mean RTs of the eight probe items were equally divided around the mean RT of all probes. Furthermore, the error rates analysis showed the same significant result, but the error rate was very low for both probe and irrelevant items in this condition and this might decrease the validity of the error rates. Altogether, it seems that this unexpected finding could not be explained by extraneous factors. Another possibility is that this finding could have occurred by chance, due to a lack of power.

Lastly, inconsistent with the hypotheses, no significant difference in differential response was found for the EC condition in both the direct and delay condition. This may be due to the artificial nature of the research design, as the participants were instructed to learn the robbery plan and the focus may have been on learning the eight words, instead of

picturing the actual things that those words represented. Perhaps if the items were encoded in another way, such as being exposed to images rather than words, this would have resulted in the expected effect. That is, if items were to be encoded as images, participants would possibly be less focused on particular words and it would be more likely that they show recognition on the CIT for a corresponding superordinate item. The way of encoding could also be related to the fact that after a delay a significant differential response was found for the EE condition but not for the CC condition. Because participants had to learn the words, the exemplar items might be better remembered; for example, “VOLLEYBAL” might have made more of an impression than “SPORTS”, because it may let them picture actually engaging in the act of volleyball, whereas this is more difficult with the more broader term ‘sports’.

This research has been the first to address the issue of leakage of information by using exemplar items, and despite of coming across some unexpected findings, it has provided promising results. Most notably, it has been shown that the leakage CIT can detect concealed information, both immediately after the information was encoded and after a one-week delay. One limitation is the artificial design of this study. In order to create a more realistic scenario during the learning phase, future research could employ virtual reality. This way, a mock crime could be performed instead of merely learning the plan of a crime, which could lead to better encoding, and more importantly, this would be more similar to when a real crime would be committed. Accordingly, future research should continue to investigate the balance

between false positive and false negative rates, so we can eventually prevent cases like that of the West Memphis Three.

References

Ben-Shakhar, G., Frost, R., Gati, I., & Kresh, Y. (1996). Is an apple a fruit? Semantic relatedness as reflected by psychophysiological responsivity. Psychophysiology, 33(1996), 671-679. http://dx.doi.org/10.1111/j.1469-8986.1996.tb02363.x Ben-Shakhar, G., Gronau, N., & Elaad, E. (1999). Leakage of relevant information to

innocent examinees in the GKT: An attempt to reduce false-positive outcomes by introducing target stimuli. Journal of Applied Psychology, 84(5), 651.

http://dx.doi.org/10.1037/0021-9010.84.5.651

Bradley, M. T., Barefoot, C. A., & Arsenault, A. M. (2011). Leakage of information to

innocent suspects. In B. Verschuere, G. Ben- Shakhar, & E. H. Meijer (Eds.), Memory detection: Theory and application of the concealed information test (pp. 187-199). Cambridge, UK: Cambridge University Press.

https://doi.org/10.1017/CBO9780511975196.011

Bradley, M. T., MacLaren, V. V., & Carle, S. B. (1996). Deception and Nondeception in Guilty Knowledge and Guilty Actions Polygraph Tests. Journal of Applied Psychology, 81(2), 153. http://dx.doi.org/10.1037//0021-9010.81.2.153

Bradley, M. T., Warfield, J. F. (1984). Innocence, Information, and the Guilty Knowledge Test in the Detection of Deception. Psychophysiology, 21(6), 683-689.

http://dx.doi.org/10.1111/j.1469-8986.1984.tb00257.x

Bond Jr, C. F., & DePaulo, B. M. (2008). Individual differences in judging deception: accuracy and bias. Psychological Bulletin, 134(4), 477-492.

http://dx.doi.org/10.1037/0033-2909.134.4.477

Gamer, M. (2010). Does the Guilty Actions Test allow for differentiating guilty participants from informed innocents? A re-examination. International Journal of

Kleinberg, B., & Verschuere, B. (2015). Memory detection 2.0: The first web-based memory detection test. PloS one, 10(4), e0118715.

http://dx.doi.org/10.1371/journal.pone.0118715

Lykken, D.T (1959). The GSR in the detection of guilt. Journal of Applied Psychology, 43(6), 385–388. http://dx.doi.org/10.1037/h0046060

Nahari, G., & Ben‐ Shakhar, G. (2011). Psychophysiological and behavioral measures for detecting concealed information: The role of memory for crime

details. Psychophysiology, 48(6), 733-744. http://dx.doi.org/10.1111/j.1469-8986.2010.01148.x

Osugi, A. (2011). Daily application of the CIT in Japan. In B. Verschuere, G. Ben- Shakhar, & E. H. Meijer (Eds.), Memory detection: Theory and application of the concealed information test (pp. 253-275). Cambridge, UK: Cambridge University Press. https://doi.org/10.1017/CBO9780511975196.015

Pansky, A., & Koriat, A. (2004). The basic-level convergence effect in memory

distortions. Psychological Science, 15(1), 52-59. http://dx.doi.org/10.1111/j.0963-7214.2004.01501009.x

Peth, J., Vossel, G., & Gamer, M. (2012). Emotional arousal modulates the encoding of crime‐ related details and corresponding physiological responses in the Concealed Information Test. Psychophysiology, 49(3), 381-390. http://dx.doi.org/10.1111/j.1469-8986.2011.01313.x

Suchotzki, K., Verschuere, B., Van Bockstaele, B., Ben-Shakhar, G., & Crombez, G (2017). Lying Takes Time: A Meta-Analysis on Reaction Time Measures of Deception. Psychological Bulletin, 143(4), 428-453. http://dx.doi.org/10.1037/bul0000087

Verschuere, B., & Crombez, G. (2008). Deja vu! The effect of previewing test items on the validity of the Concealed Information polygraph Test. Psychology, Crime &

Law, 14(4), 287-297. https://doi.org/10.1080/10683160701786407

Verschuere, B., & Kleinberg, B. (2016). ID‐ Check: Online Concealed Information Test Reveals True Identity. Journal of forensic sciences, 61(S1).

http://dx.doi.org/10.1111/1556-4029.12960

Vrij, A. (2008). Detecting Lies and Deceit: Pitfalls and Opportunities. Chichester, UK: John Wiley and Sons. Retrieved from http://eu.wiley.com/

Zuckerman, M., DePaulo, B. M., & Rosenthal, R. (1981). Verbal and nonverbal

communication of deception. Advances in experimental social psychology, 14, 1-59. https://doi.org/10.1016/S0065-2601(08)60369-X

Appendix A Plan een overval en kom er mee weg!

Het komende half uur gaan jullie je verplaatsen in topcriminelen. Jullie zullen samen een overval gaan plannen.

Zorg ervoor dat de informatie goed overeenkomt, dan kan er niets fout gaan. Probeer deze items zo goed mogelijk te onthouden en beeld je in dat je de overval daadwerkelijk gaat plegen.

Jullie hebben elkaar ontmoet op de SPORT club.

Jullie zijn van plan de SNS BANK te overvallen in je woonplaats DELFT. Dit staat gepland voor de maand MEI.

Omdat de buit jullie niet zonder slag of stoot gegeven zal worden, nemen jullie als dreigmiddel een STEEKWAPEN mee.

Doordat een van jullie al eerder stage heeft gelopen in het bedrijf, weten jullie ongeveer wat in welke kluis ligt.

Jullie zijn van plan een dure RING mee te nemen.

Nadat jullie hierin zijn geslaagd is het belangrijk om snel weg te komen.

Daarom zorgen jullie ervoor dat er een AUTO klaarstaat waarmee jullie weg kunnen scheuren.

Om ontdekking te voorkomen besluiten jullie het gestolen goed te bewaren op de ZOLDER.

Zorg ervoor dat jullie alle stappen goed onthouden, er mag tenslotte niets mis gaan! En: Houd de informatie geheim!

Appendix B Table 1

T-test results for the RT to CC, EE, CE and EC conditions in the direct condition Probe Irrelevant Condition M SD M SD t p df Cohen’s d CC 508.95 48.64 493.36 41.47 -3.43 .001* 45 0.34 EE 512.54 61.44 495.91 44.04 -2.70 .010* 45 0.31 CE 484.71 47.09 483.11 42.72 -0.65 .517 45 0.03 EC 485.27 47.530 489.77 39.05 1.14 .260 45 -0.10 Note. *p < .05 Table 2

T-test results for the RT to the CC, EE, CE and EC conditions in the delay condition Probe Irrelevant Condition M SD M SD t p df Cohen’s d CC 497.86 52.32 490.70 49.55 -1.79 .082 36 0.14 EE 504.92 58.74 487.74 48.73 -3.54 .001* 36 0.32 CE 473.11 47.13 480.99 53.44 2.25 .031* 36 -0.16 EC 484.11 60.55 488.28 52.02 1.04 .306 36 -0.07 Note. *p < .05

Appendix C Table 1

T-test results for the error rates of the CC, EE, CE and EC conditions in the direct condition Probe Irrelevant Condition M SD M SD t p df Cohen’s d CC 4.71 6.78 0.41 0.85 -4.53 <.001* 45 0.88 EE 6.23 8.98 0.95 1.62 -4.46 <.001* 45 0.81 CE 0.41 1.18 0.48 0.70 0.44 .666 45 -0.08 EC 2.07 7.21 0.98 1.94 -1.27 .212 45 0.21 Note. *p < .05 Table 2

T-test results for the error rates of the CC, EE, CE and EC conditions in the delay condition Probe Irrelevant Condition M SD M SD t p df Cohen’s d CC 2.98 6.15 0.27 0.52 -2.70 .01* 36 0.61 EE 3.84 6.03 0.58 1.17 -3.33 .002* 36 0.74 CE 0.19 0.82 0.52 0.87 2.14 .039* 36 -0.39 EC 1.07 2.22 0.80 1.44 -0.61 .548 36 0.15 Note. *p < .05