PresentationPDF Available

Detecting deception in insurance claims - How effective are verifiability approach and model statement?

Authors:

Abstract

Deception detection can be a difficult task. Based on mere subjective evaluations, people can hardly distinguish between true and fabricated stories. This is even the case for professionals, whose accuracy rates only minimally outperform chance level (Vrij, 2008). Content-based techniques of deception detection, i.e. CBCA, are able to discriminate between true and fabricated statements with an accuracy rate of approximately 70% (Oberlader et al., 2016). The lately introduced verifiability approach (VA; Nahari, Vrij, & Fisher, 2014) holds promising results: As a verbal veracity tool, this approach has proven to be a highly diagnostic instrument for detecting deception in insurance settings. At least in conjunction with an information protocol (IP), the VA correctly classifies true versus fake insurance claims with a discrimination rate of 80%. If the IP is additionally supplemented with a model statement (MS), the VA even attains a discrimination rate of 90% (Harvey, Vrij, Leal, Lafferty, & Nahari, 2017). Trying to replicate and extend previous lab results, we conduct a pre-registered online study using a 2 (veracity: truth vs. lie) x 3 (additional information given to participants: IP vs. MS vs. IP + MS) between-subjects design. Specifically, we want to test whether the VA combined with IP meets its expected high diagnostic value in distinguishing between true and fabricated reports. Furthermore, we will test the effect on detection rates when adding a mere MS without IP, or when a full combination is presented. We will critically discuss the results of our study and outline potential practical applications.
A preview of the PDF is not available
... To date, there have been 17 published studies using a MS and two studies reported at conferences (Hirn et al., 2012;Körner & Urban, 2018). One study used pairs of participants, and although the MS appeared useful for facilitating lie-detection, this has only been tested once . ...
... This combination was later retested by Körner and Urban (2018) ...
... The studies in which a MS enhanced lie-detection, researchers employed highly subjective measures, such as plausibility (see Leal et al., 2015), as well as measures that are difficult for investigators to determine, such as peripheral information ; for failure to replicate see; Leal et al., 2019a), and common knowledge details (Vrij et al., 2018b). Harvey et al. (2017) established that combining a MS containing verifiable information, in conjunction with the IP (of the verifiability approach) appeared effective, though this needs to be more robustly tested, as this effect has failed to be replicated (Bogaard et al., 2020;Körner & Urban, 2018). One of the key problems with these comparisons are that the researchers have all used different MS scripts, making any accurate replication-and meaningful comparison-difficult. Future research could more robustly examine the differences between the MS scripts and could provide a collection of such scripts for more independent replications to take place. ...
Article
Full-text available
Investigators need to elicit detailed statements from interviewees to find potential leads, whilst simultaneously judging if a statement is genuine or fabricated. Researchers have proposed that the Model Statement (MS) can both (a) increase information elicitation from interviewees and (b) amplify the verbal differences between liars and truth tellers, thereby enhancing lie‐detection accuracy. Based upon a critical analysis of the MS literature, we argue that this tool is not currently ready for practical usage, as its utility has not been fully established. We highlight a diverse range of existing MS scripts, and a greater diversity in the dependent measures examined in conjunction with this tool. More robust replications of these procedures are needed. We also highlight why some measures of verbal content may not be suitable as outcome measures and suggest that new research could use the well‐established reality monitoring criteria to allow for standardisation across studies.
Article
Full-text available
The Verifiability Approach predicts that truth tellers will include details that can be verified by the interviewer, whereas liars will refrain from providing such details. A meta‐analysis revealed that truth tellers indeed provided more verifiable details (k = 28, d = 0.49, 95% CI [0.25; 0.74], BF10 = 93.28), and a higher proportion of verifiable details (k = 26, d = 0.49 95% CI: 0.25, 0.74, p < .001, BF10 = 81.49) than liars. We found no evidence that liars would include more unverifiable details than truth tellers (k = 20, d = −0.31, 95% CI [−0.02; 0.64], BF10 = 1.12) Moderator analysis revealed the verifiable detail effect was substantially larger when the statement is the suspect's alibi, but smaller when an incentive to appear credible was used. Our findings support the main prediction behind the Verifiability Approach, but also stress the need for larger sample sizes and independent replications.
Article
Full-text available
Within the scope of judicial decisions, approaches to distinguish between true and fabricated statements have been of particular importance since ancient times. Although methods focusing on “prototypical” deceptive behavior (e.g., psychophysiological phenomena, nonverbal cues) have largely been rejected with regard to validity, content-based techniques constitute a promising approach and are well established within the applied forensic context. The basic idea of this approach is that experience-based and non-experience-based statements differ in their content-related quality. In order to test the validity of the most prominent content-based techniques, Criteria-Based Content Analysis (CBCA) and Reality Monitoring (RM), we conducted a comprehensive meta-analysis on English- and German-language studies. Based on a variety of decision criteria, 56 studies were included revealing an overall effect size of g = 1.03 (95% CI [0.78, 1.27], Q = 420.06, p < .001, I² = 92.48%, N = 3429). There was no significant difference in the effectiveness of CBCA and RM. Additionally, we investigated a number of moderator variables such as characteristics of participants, statements, and judgment procedures, as well as general study characteristics. Results showed that the application of all CBCA criteria outperformed any incomplete CBCA criteria set. Furthermore, statement classification based on discriminant functions revealed higher discrimination rates than decisions based on sum scores. Finally, unpublished studies showed higher effect sizes than studies published in peer-reviewed journals. All results are discussed in terms of their significance for future research (e.g., developing standardized decision rules) and practical application (e.g., user training, applying complete criteria set).
Article
Full-text available
We analyze the accuracy of deception judgments, synthesizing research results from 206 documents and 24,483 judges. In relevant studies, people attempt to discriminate lies from truths in real time with no special aids or training. In these circumstances, people achieve an average of 54% correct lie-truth judgments, correctly classifying 47% of lies as deceptive and 61% of truths as nondeceptive. Relative to cross-judge differences in accuracy, mean lie-truth discrimination abilities are nontrivial, with a mean accuracy d of roughly .40. This produces an effect that is at roughly the 60th percentile in size, relative to others that have been meta-analyzed by social psychologists. Alternative indexes of lie-truth discrimination accuracy correlate highly with percentage correct, and rates of lie detection vary little from study to study. Our meta-analyses reveal that people are more accurate in judging audible than visible lies, that people appear deceptive when motivated to be believed, and that individuals regard their interaction partners as honest. We propose that people judge others' deceptions more harshly than their own and that this double standard in evaluating deceit can explain much of the accumulated literature.
Article
Purpose: The Verifiability Approach (VA) is verbal lie detection tool that has shown promise when applied to insurance claims settings. This study examined the effectiveness of incorporating a Model Statement comprised of checkable information to the VA protocol for enhancing the verbal differences between liars and truth tellers. Method: The study experimentally manipulated supplementing (or withholding) the VA with a Model Statement. It was hypothesised that such a manipulation would (i) encourage truth tellers to provide more verifiable details than liars and (ii) encourage liars to report more unverifiable details than truth tellers (compared to the no model statement control). As a result, it was hypothesized that (iii) the model statement would improve classificatory accuracy of the VA. Participants reported 40 genuine and 40 fabricated insurance claim statements, in which half the liars and truth tellers where provided with a model statement as part of the VA procedure, and half where provide no model statement. Results: All three hypotheses were supported. In terms of accuracy, the model statement increased classificatory rates by the VA considerably from 65.0% to 90.0%. Conclusion: Providing interviewee's with a model statement prime consisting of checkable detail appears to be a useful refinement to the VA procedure.
Article
Lie detection in insurance claim settings is difficult as liars can easily incorporate deceptive statements within descriptions of otherwise truthful events. We examined whether the Verifiability Approach (VA) could be used effectively in insurance settings. According to the VA, liars avoid disclosing details that they think can be easily checked, whereas truth tellers are forthcoming with verifiable details. The study experimentally manipulated notifying claimants about the interviewer's intention to check their statements for verifiable details (the ‘Information Protocol’). It was hypothesized that such an instruction would (1) encourage truth tellers to provide more verifiable details than liars and to report identifiable witnesses who had witnessed the event within their statements, and (2) would enhance the diagnostic accuracy of the VA. Participants reported 40 genuine and 40 fabricated insurance claim statements, in which half the liars and truth tellers were notified about the interviewer's intention to check their statements for verifiable details. Both hypotheses were supported. In terms of accuracy, notifying claimants about the interviewer's intention to check their statements for verifiable details increased accuracy rates from around chance level to around 80%. The VA, including the information protocol, can be used in insurance settings.
Article
Deception research regarding insurance claims is rare but relevant given the financial loss in terms of fraud. In Study 1, a field study in a large multinational insurance fraud detection company, truth telling mock claimants (N = 19) and lying mock claimants (N = 21) were interviewed by insurance company telephone operators. These operators classified correctly only 50% of these truthful and lying claimants, but their task was particularly challenging: Claimants said little, and truthful and deceptive statements did not differ in quality (measured with Criteria-Based Content Analysis [CBCA]) or plausibility. In Study 2, a laboratory experiment, participants in the experimental condition (N = 43) were exposed to an audiotaped truthful and detailed account of an event that was unrelated to insurance claims (a day at the motor races). The number of words, quality of the statement (measured with CBCA), and plausibility of the participants' accounts were compared with participants who were not given a model statement (N = 40). The participants who had listened to the model statement provided longer statements than control participants, truth tellers obtained higher CBCA scores than liars, and only in the model statement condition did truth tellers sound more plausible than liars. Providing participants with a model statement is thus an innovative and successful tool to elicit cues to deception. Providing such a model has the potential to enhance performance in insurance call interviews, and, as we argue, in many other interview settings.
Article
SUMMARY According to the verifiability approach, liars tend to provide details that cannot be checked by the investigator and awareness of this increases the investigator's ability to detect lies. In the present experiment, we replicated previous findings in a more realistic paradigm and examined the vulnerability of the verifiability approach to countermeasures. For this purpose, we collected written statements from 44 mock criminals (liars) and 43 innocents (truth tellers), whereas half of them were told before writing the statements that the verifiability of their statements will be checked. Results showed that ‘informing’ encouraged truth tellers but not liars to provide more verifiable details and increased the ability to detect lies. These findings suggest that verifiability approach is less vulnerable to countermeasures than other lie detection tools. On the contrary, in the current experiment, notifying interviewees about the mechanism of the approach benefited lie detection. Copyright © 2013 John Wiley & Sons, Ltd.
Article
Background We examined the hypothesis that liars will report their activities strategically and will, if possible, avoid mentioning details that can be verified by the investigator. MethodA total of 38 participants wrote a statement in which they told the truth or lied about their activities during a recent 30-minute period. Two coders counted the frequency of occurrence of details that can be verified and that cannot be verified. ResultsLiars, compared with truth tellers, included fewer details that can be verified and an equal number of details that cannot be verified in their statement, and the ratio between verifiable and unverifiable details was smaller in liars compared with truth tellers. High percentages of truth tellers and liars were classified correctly based on the frequency counting of verifiable details (79%) or the ratio between verifiable and unverifiable details (71%). Those percentages were higher than the percentage that could be classified correctly (63%) based on verifiable and unverifiable detail combined. We compared our verifiability approach with other theoretical approaches as to why differences in detail between truth tellers and liars emerge.
Versicherungsbetrug: aktuelle Entwicklungen, Muster und ihre Abwehr
  • K John
John, K. (Stand 2011). Versicherungsbetrug: aktuelle Entwicklungen, Muster und ihre Abwehr. Pressekonferenz des Gesamtverbandes der Deutschen Versicherungswirtschaft e.V. Retrieved from http://www.gdv.de/wpcontent/uploads/2011/11/PK_Versicherungsbetrug_2011_Praes1_Versicherungsbetrug_in_Deutschland_GfK_KarstenJohn_n4.pdf Gesamtverband der Deutschen Versicherungswirtschaft e.V. 2017
Management von Betrugsrisiken in Versicherungsunternehmen
  • J Knoll
Knoll, J. (2011). Management von Betrugsrisiken in Versicherungsunternehmen (1. Aufl. 2011). Baden-Baden: Nomos. https://doi.org/10.5771/9783845232874
Wiley series in the psychology of crime, policing and law
  • A Vrij
Vrij, A. (2008). Detecting lies and deceit: Pitfalls and opportunities (2nd ed.). Wiley series in the psychology of crime, policing and law. Chichester, England, Hoboken, NJ: John Wiley & Sons.