Detecting deceit within a predominantly true statement using two parallel assessment methods: A pilot study

Currently, three major theories are reported to explain the observable signs of deceit: emotional arousal, cognitive load and behavioural control. Variables that can be measured are physiological, verbal and non-verbal (Zuckerman et al. 1981, pp. 7–10; Granhag and Vrij 2005, p. 52, 65; Hart et al. 2009, p. 135).

According to the emotional arousal theory, liars are more excited than truth tellers. This excitement and resulting emotions such as fear, guilt and delight induce noticeable changes in one’s verbal and non-verbal behaviours. The cognitive theory is based on observations that lying is cognitively more demanding than telling the truth. The truth is the norm, and it is automatically activated. For a deceiver, it requires extra effort to suppress the truth and express something that is in contradiction to that. The lie must be plausible and consistent so that the liar does not get caught. In addition to monitoring own behaviour, a liar must also monitor the receiver to get feedback from his or her success. People have quite stereotypical views about how a sincere person behaves. Good liars try to act by these preconceptions and thus must control their behaviour accordingly. This leads to the attempted control theory that explains why liars fail to act naturally and leave a rigid impression while trying to suppress what they think resembles a guilty person’s behaviour (Kirchhübel and Howard 2013, p. 695).

In addition to the emotional arousal theory, the preliminary process theory (PPT) and orienting response explain the changes in physiological reactions during the instrumental assessment process. Stimuli, in this case the questions, initiate cognitive processes, which produce variances in physiological responses. The greater the relative significance of the stimuli is, the larger the variance in the monitored parameters become. The veracity assessment is based on the monitoring and recording of the reactions by measuring the subjects’ electrodermal activity (EDA, Palmatier and Rovner 2015, p. 1).

In lie detection research, such speech-related variables as the pause length, voice pitch, hesitation and the response length have been studied as potential indicators of deception (Hart et al. 2009, p. 135). Some of them have shown to be more potential than others. For example, studies regarding high-stake police interviews with real suspects showed that lying induced increased pauses (Vrij and Granhag 2012, p. 112). Additionally, an increased latency period, ‘ah’ speech disturbances and speech rate have been found to correlate with deception both in laboratory studies and in real-life situations (Vrij et al. 2000, p. 251, 254). Longer latency periods, increased pauses, hesitations, speech errors and slower speech rate have been found to indicate heightened cognitive load, which are interpreted as an attempt to deceive (Vrij et al. 2008, p. 255).

There are also studies that show less correlation between lying and speech hesitations, speech rate and speech errors. In a study concerning the effectiveness of increasing cognitive load to facilitate lie detection, it was found that in the control group, the aforementioned criteria showed no significant discriminating power (Vrij et al. 2008, p. 256, 259). It is perhaps noteworthy that the participants in the control group had received a thorough coaching and very detailed information about the actual event that they lied about and thus could confidently reproduce their version of the event.

As the understanding of the human behaviour that is related to lying and lie detection accumulates, new theories explaining this vast phenomenon are created. Often orienting response, arousal, cognitive load and behavioural control theories can be traced back to establish the foundation that explains observable and measurable human behaviour supporting practical lie detection efforts (Walczyk et al. 2013, pp. 2-3; Vrij et al. 2019, pp. 298-300).

The simultaneous use of two or more lie detection methods has been recognised as a potential avenue of approach to lie detection (Granhag and Vrij 2005, p. 79). Theoretically, the parallel use of two or more methods should increase the overall accuracy by reducing the number of false positives, if two or more simultaneous indications based on different assessment methods are required as a confirmed result. In addition, two indications taking place at the same time should increase the perceived confidence in the finding, when no other supporting information is available.

Thus, the paradigm change towards being able to detect deceit within a single person’s statement and the simultaneous use of multiple assessment methods lead to the following research problem: Will the use of two parallel methods improve the reliability of the lie detection within a predominantly truthful verbal statement?

For this study, the two selected lie detection methods used were orienting response and PPT-based measurement of EDA and cognitive load theory-based changes in speech production including ‘ah’ speech, speech errors and slowed speech rate.

There were two primary reasons for selecting these methods. Firstly, they are widely studied, and there is a fair amount of published research to support their reliability and applicability when they are tested at the group level.

Secondly, the essential requirement for a successful within-statement lie detection is the temporal fidelity of the measured signs of deception in relation to the actual deception. The measured variables must take place, if not immediately, but at least relatively timely with the actual deception. The temporal fidelity between the observation and the occurrence of the deception is highly desired because the aim of this approach is to be able to identify suspected deceitful parts from the likely truthful parts of the statement. If the method, like the general perceived credible demeanour or the immediacy of the sender, would not meet the requirement of the temporal fidelity, the assessment would again be made at the group level, and the within-statement approach would not be tested.

As the task was to detect deceit within freely produced statement containing multiple episodic events, the question–answer latency was excluded as a cognitive load indicator. Illustrating hand gestures was also excluded because the EDA signal was measured from the interviewees’ left hand, and they were instructed to hold them as motionless as possible to insure a proper contact. This was also assessed to hinder the interviewees’ bodily movement on the whole.

At the selection of the used methods, it was assessed that the EDA measurement and the decoding of the speech could be carried out with the least amount of interference from the recording and the person performing the ratings.

Audio-video recording produced a wealth of data, which were not used in this study. Vocal cues, facial expressions, blinking rates, content for the statement validity assessment or reality monitoring are available to be used in future studies. The exclusion was based on the availability of time and other resources reserved for this study.

As the test setup was slightly different from the more common ones in lie detection studies, the applicability of the methods in this study had to be tested first. This was carried out by testing the following hypotheses:

H1: Truthtellers exhibit less EDA-derived indices of orienting response or nervousness than deceivers over the entire interview. The aim of this hypothesis is to validate the use of the method in the current test setup.

Some arguments against the claimed effectiveness of the polygraph testing address the make-shift nature of the test setups in laboratory studies and the analysis of the results. The benefit of a laboratory study, often utilising mock crimes, is the knowledge of the ground truth. However, laboratory studies lack the heightened personal stakes of real-life situations that concern both guilty and innocent suspects, and thus, laboratory tests can be claimed not to represent the real potential of the polygraph test (Iacono 2008, p. 25). However, even if the real fear of getting caught and thus being penalised for lying was missing, the orienting response, or PPT, and cognitive load theory-based approach would overcome the possible issue of the participants not being mentally committed to lying. It can be argued that a potential reward does not work as effectively as a potential punishment, but a reward-based approach is assessed to produce measurable Orienting Response and Cognitive Load derived indices.

During an interview, reports from self-lived activities are assessed to be more plausible than the ones which were pre-scripted by a third party and afterwards reported as own experiences (Verigin et al. 2020, p. 380). To increase the personal involvement, possibly heighten the excitement factor and to increase the test persons’ uncertainty concerning what the interviewer might know, the test persons were put to perform a secret mission consisting of a series of tasks. In principle, the tasks were simple, but they had to be performed successfully without getting exposed to anyone prior to the completion of the mission. The motivation to perform, and if needed to lie, was assessed to be high among all participants. In this test setup, a successful deception would be considered as a positive act, which could lower the perceived negative feelings towards lying (Halevy et al. 2014, p. 56).

Another purpose for the created secret mission type of scenario was to simulate a simple series of events that have at least some resemblances to actual human intelligence collection or counterintelligence operations. Even though the setup was following a typical mock crime scenario, in effect, the test persons playing secret agents were instructed to get into contact with an agent handler or an assisting agent, receive further instructions from them, visit a dead letter box, receive additional instructions and not get caught in the middle of their covert activities. This kind of setup was assessed to be applicable with a secret human asset’s after-action debriefing or a suspected foreign agent’s investigative efforts.

As the research problem dictated, the content of the statement obtained during the intelligence interview had to be mostly, or at least partially, true. The overall duration of the secret mission was estimated to be from 30 to 60 min. This would result in enough content for the intelligence interview to be completed successfully. The desired duration of the intelligence interview was 10–15 min.

Eight participants, six males and two females between 21 years and 41 years (M = 32,4 standard deviation [SD] = 6,0), were randomly divided into test and control groups by the test assistant. They were treated equally in terms of providing instructions and the completion of the intelligence interview. In the morning of the test day, all participants in both groups were given instructions about their role in the study. They were advised that at any circumstances were they not to disclose any information regarding the given mission, even if directly asked. Both the test and control group participants were informed that they would be later interviewed about the test day’s activities. If they would complete the given tasks and keep the secrets successfully, a reward of 20 euros would be granted.

The difference between the groups was that the test group was given a secret mission to be performed during the lunch break. The control group’s mission was to have lunch without any additional directions or tasks. The test group’s secret mission was identical for every participant. The mission was to perform a series of tasks and avoid getting caught or otherwise exposed. The initial task was to establish contact with a person in a coffee corner on the way to the lunch restaurant. The second task was received from the contact person. Instructions were written on a small piece of paper, and they ordered the test group participants to navigate to a given nearby location, an old telephone booth serving as a janitor’s storage, retrieve a wallet and act according to the instructions found in the wallet. A note found in the wallet instructed to perform the third task, which was to take the money and return the wallet to the reception in the building lobby. Depending on the random selection, the wallet was either empty or there was money, which the test person should have taken if the instructions were followed. The participants should also keep and hide the instructions until the secret mission and the test was officially ended.

A research assistant was used to act as the contact person with whom the test group participants were to meet at a coffee corner. Also, the lobby receptionist who monitored the access in and out of the building and to the decommissioned phone booth was assisting the test and received the wallets which were returned as lost and found. After the lunch break, the participants would independently leave for their respective activities until they would be called upon for the interview.

Common to all veracity assessment and lie detection methods is that they are based on observations. It has been suggested that questioning techniques that produce longer answers should be preferred. These techniques potentially produce more information and thus more opportunities for verbal cues of deception to occur (Vrij and Granhag 2012, p. 115).

An identified risk concerning data collection was the potential shortness of the statement. It was anticipated that for the control group, a regular lunch break would not become an especially eventful period. As the control group participants would not have much to talk about, the test group participants were anticipated to keep the story short and simple.

Cognitive interview is a questioning method that aims to increase the statement’s accuracy and the level of detail. Different techniques are applied to make the memory retrieval easier and more accurate. These techniques are the contextual reinstatement, repeating the statement without preliminary screening, recalling the events from another participant’s point of view and recalling the events in mixed or reversed order (Memon and Higham 1999, pp. 178-184).

An enhanced version of the cognitive interview (ECI) was developed with the emphasis on the building rapport and effective communication. During the interview, the free flow of speech and information should not be interrupted, and the interviewee’s effort should be put into active listening. Initially, the communication is facilitated with open-ended questions about neutral topics. The actual interview is started with contextual reinstatement and interviewee’s free narrative of events. The interviewer then reminds that it is important to give a full account of events in as much detail as possible (Memon et al. 2010, p. 5).

The Strategic Use of Evidence (SUE) is a questioning technique that is based on a carefully planned presentation of critical and possibly incriminating evidence. The presentation of the facts, which the interviewer knows and has verified to be true, is supposed to produce different verbal responses from truth tellers than liars. It is thought that the guilty suspect will use more avoidance strategies during the discussion than the innocent suspect who is thought to be more forthcoming. Avoidance strategies include, for example, leaving parts or details of the event untold. Truth tellers are generally more likely to tell the truth as it happened. On the tactical level, the SUE consists of three elements: evidence tactics, question tactics and disclosure tactics. Evidence tactics is a part of the preparation before the questioning. Question tactics is used to systematically exhaust the suspect’s alternative explanations for the presented incriminating evidence. Disclosure tactics is used to increase the diagnostics value of the evidence by assessing the strength of the source and the degree of the precision to be able to present the evidence in a certain way. Research shows that the step-by-step presentation of stronger and more detailed evidence produces more and stronger cues to deception than presenting the most incriminating pieces of evidence straight away (Vrij and Granhag 2012, p. 114).

An applied version of the ECI with an SUE protocol type of incriminating evidence presentation at the end of the interview was created. Compared to the full version of the ECI, it was decided to exclude the contextual reinstatement and recalling the events from another participant’s point of view because the interview took place at the same day, and it was assessed that all the related events were readily available for the accurate memory retrieval. Although the increase in the cognitive load has been shown to increase the ability to detect deception (Vrij et al. 2008, p. 262), recalling the events in a mixed or reversed order was excluded as it could offer an opportunity for a deceiving interviewee to spend time on irrelevant and possibly non-compromising topics. To promote an unhampered flow of information and an undisturbed occurrence of cues to deceit, no measures to add pressure during the interview was taken (Verschuere et al. 2016, p. 916). Although no extra effort to build rapport between the interviewer and the interviewee was taken, a neutral and cooperative atmosphere was sought after to promote an effortless flow of information (Nunan et al. 2020, p. 513).

All the interviews were conducted in Finnish, which was the native language for all the participants.

The free telling information-gathering interview was identical to all the participants. The interviewer did not know whether the interviewee was from the control group or from the test group. The interview consisted of a warmup period covering non-pertinent neutral topics. The pertinent part of the interview consisted of three main phases: free telling initiated with open-ended questions concerning the lunch break, the presentation of potentially compromising light evidence and the presentation of potentially compromising heavy evidence (Figure 1).

Fig. 1:

The warmup period was intended to last from 10 to 15 min. Light topics concerning hobbies, pets and summer vacation activities were covered. The aim of the warmup period was to get the interviewee talking, ease up the atmosphere and form a calm starting point for the assessment during the actual information-gathering interview.

During the free telling phase, the following open-ended questions were used to obtain the statement concerning the lunch break: ‘Please tell me freely, with your own words, what happened during your lunch break today?’ Once the statement was finished, the interviewee was asked to elaborate more by asking: ‘Please cover your lunch break in more detail, chronologically as it happened?’ A third iteration of the statement was initiated by asking ‘Please tell a bit more, and include details and events which you might have excluded earlier as insignificant?’ The last question, which ended the free telling phase, was ‘Once more, did you miss to mention anything that you would wish to add at this point?’ This type of repetition was necessary for two reasons: Firstly, it was anticipated that the statements were not going to be very long. An hour-long lunch break would not likely offer much to talk about. Secondly, deceivers were assessed to keep their statements as short as possible as part of their strategy of not getting caught. All four questions were stated in this way independent of the length or the level of detail of the previous statements. A set of contingency questions was also available in case the interviewee would start to argue against the questions or decided to go numb. Another anticipated scenario was that the interviewee would try to drain the time reserved for the interview by going side tracked or by giving a super detailed statement. This would have been encountered by asking the interviewee first to cover the main events in a more general level and after that go into details. In this study, all participants were cooperative and did not begin to resist the interview.

The presentation of the potentially compromising light evidence was completed with three questions. These questions were presented to all interviewees as they were regardless of their answers or reactions. The first question was ‘During the lunch break you were seen meeting with a person at the coffee corner by the canteen, who did you meet with?’ It was followed by ‘Why did you choose to meet this person?’ The third question ended this phase: ‘What did you talk about?’ If the interviewee denied having met anyone, the following optional question was asked: ‘You were witnessed to meet with someone at the coffee corner, we have certain information about that, who did you meet with?’ If the interviewee still bluntly denied the meeting taking place, the interview transitioned to the next phase.

Potentially compromising heavy evidence was presented with one question: ‘You were reported to have returned a lost wallet at the reception, did you take any money from it?’ If the interviewee denied or gave an ambiguous answer, the follow-up question was asked: ‘Did you take money from the wallet, please answer yes or no?’ (Figure 2).

Fig. 2:

The aim of the secret mission design and the use of an applied version of the ECI with an SUE-protocol type of incriminating evidence presentation was to create suitable conditions to deduce three types of lies. During the free telling phase, the concealment or omission of a detail, fact, event or a person, i.e. not telling the whole truth and that way manipulate the context of the actions, was expected. During the presentation of light evidence, a quick fabrication of a plausible explanation, i.e. a replacement of truth with a small lie, was expected. If the deceitful interviewee was prepared, a credible cover story could have been made up and presented within the initial statement. Another option would be a complete denial of the compromising evidence. During the presentation of potentially compromising heavy evidence, a deliberate production of an untrue statement, i.e. a direct lie, was expected. Ideally, as the interview advances, the lies would develop from subtle lies to exaggerations and outright lies (Vrij et al. 2010, p. 9). All the anticipated lies would be embedded in their nature because they were expected to be told within the otherwise truthful statement (Vrij et al. 2022, p. 8).

All the participants were assessed to be motivated to perform well in the given tasks. They were all soldiers in reserve, non-commissioned officers and officers taking part in the refreshment training course. They did not have a unified military background or basic training. They had not received any prior training related to lie detection, lie detection countermeasures or questioning resistance measures. The participation was voluntary, and it had to be conducted parallel to the refreshment training.

The participants in the test group were expected to be mentally prepared for the deception. They received the instructions in the morning, and they would complete the given tasks during the lunch break. The interview was told to be conducted on the same day. This arrangement would give enough time for the participants in the test group to come up with a deception strategy such as a cover story or a plausible explanation for their potentially unusual activities during the lunch.

Referring to the checklist by Walczyk et al. (2013, p. 7), truthtellers in the control group were assessed to be motivated, they wanted to be perceived as believable and all the related details were easily available to be retrieved to working memory. In addition, the deceivers were expected to keep their statement internally and externally consistent, realistic in terms of time and events to get caught. It was also assessed that the planned deception was at least mentally rehearsed and that the deceivers were monitoring the interviewer for possible feedback concerning their credibility.

The intelligence interview took place in a medium-size office with a window, a room enough to house three people and all the required equipment. The room had minimal furniture: three chairs for the interviewer, the interviewee and the technical assistant; a small table; non-intrusive auxiliary light; a video camera; and a NEXUS-10 biosignal recorder manufactured by Mind Media B.V., Netherlands. The interview was recorded, both video and audio. The equipment was assessed not to be intrusive, although their presence and purpose were neither concealed in any way nor was their function specifically explained. The general atmosphere was business-like neutral, and no deliberate efforts to increase mental pressure were taken.

Data collection produced 32 samples per second EDA time series along with a High Definition (HD) audio-video recording. The EDA measurement formed a master track to which the audio recording was synchronised using manual triggers. Only the audio track of the recording was used in this study.

The interview was the final part of the participants’ secret mission. Before the test persons were allowed to leave, their activities during the test were double-checked from the assistants who were acting as the contact person and the receptionist. Also, the sequence of events was double-checked by asking the participants to state them in the chronological order. After that, their participation to the test was confirmed to be successful, and their rewards were paid.

The question–answer turns for each interview were coded. Every participant’s statement was divided into 15-s slots. Each slot was analysed in terms of cognitive load indicators and nervousness or orienting response indicators. The slot was flagged as a positive indication, if cognitive load speech indicators, including ‘ah’ speech, speech errors and slowed speech rate, occurred. The assessment was based on the verbatim interview transcripts. The speech rate, words per minute, baseline was established within the pertinent part of the interview (Vrij 2016, p. 1114). Individual speech rates were calculated from the transcripts over 5-s intervals, and if the measured speech rate dropped under the set threshold, the corresponding 15-s slot was flagged as a positive indication of deceit. The threshold for slow speech was the participant-specific words per minute average subtracted with the SD calculated over the pertinent part of the interview.

All participants’ measured EDA time series was pre-processed and decomposed into tonic (slow) and phasic (fast) components using Ledalab software (Benedek and Kaernbach 2010, p. 82). The last 3 min of the non-pertinent part of the interview was taken as a reference point, or the threshold, of a non-deceiving statement, for the pertinent part of the interview. Phasic EDA, which had occurred during the last 3 min of the warmup period, formed an envelope of normal variation. Phasic EDA above that individual threshold was considered as an indication of deceit during the interview. The deception was assessed to have ended once the measured EDA had come down to the half of the peak value (Braithwaite et al. 2015, p. 6–7). All the 15-s slots containing deception indicating EDA markers were flagged as positive indications.

Blind to the actual status of the participants, both the EDA and speech-related indicator data were timely synchronised. If an EDA and a speech indicator occurred at the same time, the corresponding 15-s slot for the combined EDA and speech was flagged as a positive indication of deceit. This process produced three columns of indicator data for each participant: independent EDA, independent speech and combined EDA and speech. All these three data sets were compared with the actual deceit during the intelligence interview.

As described earlier, every participant’s statement was broken down to 15-s slots and analysed accordingly. This resulted in a total of 190 assessment points, which were labelled as truths or lies, which was indicated by the studied veracity assessment methods. The correlation calculations were conducted over those 190 assessment points. Instead of having just eight test participants, who would traditionally be assessed as liars or truthtellers, the used method multiplied this study’s effective sample by over 20.

Some limitations must be addressed. The sample size is small (n = 8), of which a test group of three participants and a control group of five participants was established.

However, as the measurement and the analysis were conducted on an individual, within-statement level, the number of the assessment points was much higher than the number of the participants (Levine et al. 2022, p. 191). In a dichotomist approach, one participant constitutes one sample as the choice is between a truthteller or a deceiver. In this study, each statement was sampled in 15-s slots, where measurement and data analysis were applied. Thus, in this setup, eight participants constituted a total of 190 assessment points, which were analysed during the free telling phase. This resulted in an average of 23 assessment points per participant.

The correlation calculations did not interfere with each other as Spearman’s correlation coefficient was calculated individually for every test person. Similarly, accuracy rates were calculated on an individual level over within-statement assessment points. This type of measurement and analysis method is assessed to increase the reliability of the results and to decrease the effect of the small number of the test participants to the overall validity of the results.

Setting the threshold of the phasic EDA on a level that occurred during the last 3 min of non-pertinent part of the interview was based on an assumption that the test persons were not deceiving at that time. In this test setup, the assumption was assessed to be relatively accurate and reliable, but it can be argued to become a significant factor of unreliability outside of the laboratory environment.

As mentioned before, the prevalence of frequent ‘ah’ speech disturbances might be related to the interviewees’ native tongue or some other personal features, which were not known. This phenomenon might have saturated the cognitive load indicators or at least made them less indicative in this study. In addition, in the Finnish language, the sources, meaning and purpose of different interferences are likely to be unique, or their relevance regarding lie detection is unknown. How lying affects the production of Finnish language is not widely studied. To improve the diagnostic value of Finnish language disfluencies to facilitate lie detection, more extensive studies are required. However, disfluencies observed in this study’s interviews follow the general guidelines found in the Finnish population (Penttilä et al. 2018, p. 159, 161). The next challenge is to find out how these data could be operationalised to aid verbal lie detection.

The application of this type of intelligence interview and veracity assessment setup requires the interviewee to be cooperative. Firstly, the interviewee must be willing and able to reply to the open-ended questions and produce relatively long narrative answers. The interviewee must also be willing to repeat the statement as requested and not to suddenly get frustrated or anxious about the repetition. In addition, the interviewee must be willing to get wired to a polygraph or an EDA recorder and thus be able to control his or her bodily movement throughout the interview.