Embedding lies into truthful stories does not affect their quality

Funding information Erasmus Mundus Joint Doctorate in Legal Psychology (EMJD-LP), Grant/Award Number: Specific Grant Agreement (SGA) 2016-1339 Summary When given the opportunity, liars will embed their lies into otherwise truthful statements. In what way this embedding affects the quality of lies, however, remains largely unknown. This study investigated whether lies that are embedded into truthful stories are richer in detail and contain higher quality details compared to lies that are part of entirely fabricated statements. Participants (N = 111) were asked to provide a statement that was either entirely truthful, entirely fabricated, or had the fabricated element of interest embedded into an otherwise truthful story. Results indicated that lies embedded in a fabricated statement are not qualitatively different from lies embedded in an otherwise truthful statement. Supporting Bayes factors provided moderate to strong evidence for this conclusion. Accordingly, verbal credibility assessment tools based on the verbal content measured in this study may be robust against the embedding of lies.


Summary
When given the opportunity, liars will embed their lies into otherwise truthful statements. In what way this embedding affects the quality of lies, however, remains largely unknown. This study investigated whether lies that are embedded into truthful stories are richer in detail and contain higher quality details compared to lies that are part of entirely fabricated statements. Participants (N = 111) were asked to provide a statement that was either entirely truthful, entirely fabricated, or had the fabricated element of interest embedded into an otherwise truthful story. Results indicated that lies embedded in a fabricated statement are not qualitatively different from lies embedded in an otherwise truthful statement. Supporting Bayes factors provided moderate to strong evidence for this conclusion. Accordingly, verbal credibility assessment tools based on the verbal content measured in this study may be robust against the embedding of lies.

K E Y W O R D S
content analysis, deceivers' strategies, embedded lies, information management, verbal deception detection Whereas people typically report to rely on nonverbal behaviour to detect deception (Vrij, 2008), the accuracy of credibility assessments actually improves when judges rely only on the verbal content of a statement (Bond & DePaulo, 2006). Moreover, good lie detectors report a higher reliance on verbal cues when making credibility judgments, while poor lie detectors tend to rely primarily on nonverbal cues (Mann, Vrij, & Bull, 2004). Specifically, the most consistent finding in the verbal deception literature is that truthful statements contain more details than deceptive ones (e.g., Amado, Arce, Fariña, & Vilarino, 2016;DePaulo et al., 2003;Luke, 2019;Oberlander et al., 2016). A recent meta-analysis estimated this effect at d = 0.55 (Amado et al., 2016), while additional meta-analytical findings support the usefulness of temporal, visual, and auditory details for differentiating truthful from false accounts (Masip, Sporer, Garrido, & Herrero, 2005).
In studies investigating the effect of deception on the content of statements, participants are typically instructed to report a truthful account, or to fabricate one. Real-life deceptive statements will, however, rarely be complete fabrications, as liars prefer to embed their lies into otherwise truthful statements (e.g., Leins, Fisher, & Ross, 2013;Nahari, 2018a;Nahari & Vrij, 2015;Vrij, 2008;Vrij, Granhag, & Porter, 2010). Yet, it remains unknown how embedded lies are influenced by surrounding truthful information. Reason to believe that the embedding of lies affects their quality stems from research on beliefs about cues to deception and liars' strategies. This research suggests that laypeople and legal professionals alike believe that inconsistency is symptomatic of deception (Blair, Reimer, & Levine, 2018;Strömwall & Granhag, 2003;Vredeveldt, van Koppen, & Granhag, 2014). Accordingly, one of the main concerns of liars-and one of their most frequently reported strategies-is to maintain consistency (Deeb et al., 2017;Hartwig, Granhag, Strömwall, & Doering, 2010). Specifically, Deeb et al. (2017) instructed liars to provide a statement containing a mix of a deceptive event and a truthful event. When asked about their strategies for appearing credible, 45% of liars mentioned maintaining consistency, and many liars reported to have maintained consistency by strategically lowering their "baseline consistency" by including fewer repetitions in specific portions of the interview.
If participants would successfully maintain consistency when embedding their lies into an otherwise truthful statement, their lies could become more richly detailed when surrounded by truthful information. Indeed, recent research compared how interviewees strategically regulate the information they provide when their accounts contain one truthful and one deceptive component and found that lies became more richly detailed when preceded or followed by truthful information compared to when preceded or followed by other lies (Verigin, Meijer, Vrij, & Zauzig, 2019). The present study aimed to extend these findings by examining the extent to which the embedding of lies into otherwise truthful stories affects the quality of information. In line with previous research (e.g., Amado et al., 2016;DePaulo et al., 2003;Luke, 2019;Masip et al., 2005), we predicted that completely truthful statements would be richer in detail compared to completely fabricated statements (Hypothesis 1). Based on our consistency assumption, we also predicted that lies embedded into an otherwise truthful statement would be richer in details than lies embedded into an entirely fabricated statement (Hypothesis 2).
Besides the differences in detail richness, we also investigated whether the consistency hypothesis extends to a number of other content cues that have been shown diagnostic of deception (see Table 1). As secondary cues, we first examined the amount of verifiable information. This cue stems from the Verifiability Approach (VA; Nahari et al., 2014aNahari et al., , 2014b, which works on the assumption that liars, on the one hand, are inclined to provide detailed statements to be perceived as cooperative and credible, but, on the other hand, want to minimize the chances that investigators can falsify their statement (Masip & Herrero, 2013;Nahari et al., 2014a). A strategy that meets these aims is to provide information that cannot be verified. Second, we examined liars' tendency to report fewer complicationsoccurrences that make a situation more difficult than necessary, often characterized by disrupted activity or failing efforts (e.g., missing the bus; Steller & Köhnken, 1989;Vrij, Leal, Jupe, & Harvey, 2018)compared to truth tellers. This makes sense given liars' preference for simple stories (Hartwig, Granhag, & Strömwall, 2007). We also investigated common knowledge details (i.e., strongly invoked stereotypical information about events; . Whereas truth tellers have personal, unique experiences of an event (DePaulo, Kashy, Kirkendol, Wyer, & Epstein, 1996), liars typically lack such information and their reports tend to be characterized by more general, impersonal knowledge (Sporer, 2016). Another cue relates to self-handicapping strategies (i.e., implicit or explicit justifications as to why someone is unable to provide certain information; . Although liars prefer to keep their stories simple, they also realize that admitting a lack of knowledge and/or memory may generate suspicion from investigators (Ruby & Brigham, 1998).
Liars, then, are inclined to provide justifications for their inability to provide certain information.
Lastly, we explored the clarity and plausibility of statements. Clarity, or vividness, refers to the extent to which a statement is clear, sharp, and vivid (instead of dim and vague; e.g., Sporer, 2004). How plausible a statement is refers to whether the story is probable, realistic, and makes sense. Both critera derive from the reality monitoring method of credibility assessment (RM; Johnson & Raye, 1981), which reasons that truthful accounts represent experienced memories and are likely to be, for example, more clear and plausible compared to lies, which are formed from imagination. There is previous empirical support regarding these critera (DePaulo et al., 2003;Leal et al., 2015;Sporer & Küpper, 1995;Zhou et al., 2004).
Altogether, our secondary hypotheses were that completely truthful statements would be characterized by more verifiable details, a higher proportion of complications, and higher ratings of statement clarity and plausibility compared to completely fabricated statements (Hypothesis 3), and that lies embedded into an otherwise truthful statement would be characterized by these higher quality details moreso

| Procedure
Upon arriving to the lab and providing informed consent, participants completed a Pre-Interview Questionnaire followed by a demographic form measuring their age, sex, ethnicity, education, and native-language.
Participants then received a letter instructing them to imagine that they had been called into a police interview as a suspect in a home invasion investigation, and that they must provide an alibi for their whereabouts during the day of the crime. Three conditions were created by providing participants with additional instructions. First, truth tellers were told that they were innocent, and their task was to convince the interviewer of their innocence by providing a completely truthful alibi. In addition, two lie conditions were created. Liars were told to imagine they were guilty of the hypothetical crime in question, and that they must lie about their whereabouts during the time of the home invasion, that took place between 1:00 p.m. and 3:00 p.m. Embedded liars were instructed to embed the critical 1:00 p.m. to 3:00 p.m. period in an otherwise truthful account, whereas in the complete lie condition, liars were asked to fabricate the entire account (see Supporting Information for the instructions).
All participants were told it was important to be convincing because it would earn them a chance to win a £50 voucher and it would prevent them from having to stay an additional twenty minutes to provide a written account. After receiving these instructions, participants were given up to ten minutes alone to prepare.
The assignment to either the truthful, embedded lie, or complete lie condition was done in a pseudo-random manner. The first five of every fifteen participants were assigned to the truth teller condition, whereas the remaining participants (e.g., participants 6 to 15, 21 to 30, etcetera) were assigned to either the embedded lie or complete lie conditions. This was done so we could match the content of the critical period in the two lie conditions to that in the truth teller condition. Specifically, the alibi activity that participants were instructed to lie about was generated based on the truth tellers' responses to the Pre-Interview Questionnaire that asked them to briefly describe, in approximately one sentence, their activities between 1:00 p.m. and 3:00 p.m. on the previous three days.
The first author selected the activity that had the most unique, contextual detail, and this selected activity was used for all three conditions, while making sure the assigned alibi activity differed from any of the liars' reported events. This pseudo-randomized design allowed us to experimentally control the type of activity reported and length of time between the experience and reporting (i.e., one, two or three days) across participants, thereby reducing heterogeneity across statements.
Next, a second researcher (blind to participants' conditions) began the interview by stating that her goal was to obtain as much information as possible, and to determine how credible the participant's account was. The interviewer instructed the participant to report as many details as possible, even if she/he did not think they were important. Each interview followed a structured format (see Supporting Information) and was video-recorded. The interview began with the elicitation of a free narrative of the participants' activities from morning to evening on the day in question. The researcher then asked several questions, such as "What else can you tell me about that day?", "Did anything unexpected happen or perhaps something that did not go as planned?" Interviewees were also asked to report their activities during the 1:00 p.m. to 3:00 p.m. period specifically and were given the opportunity to provide any forgotten or missing information at the end of the interview.
Following the interview, participants were informed that the experimental portion of the study had ended, that their answers to the following questionnaire would not influence how their statement would be assessed, and that they should answer the next questions honestly. Participants then completed the Post-Interview questionnaire, where they were asked to rate several items on 5-point Likert scales (1strongly agree to 5strongly disagree): (i) The instructions clearly explained what I needed to do, (ii) I had enough time to prepare for the interview, (iii) I was motivated to convince the interviewer that I was innocent, (iv) I was successful in convincing the interviewer that I was innocent, (v) I prepared my statements strategically, (vi) The interviewer was friendly. Next, participants evaluated the truthfulness of both the critical and general components of their alibi using a 10-point scale (1not at all truthful to 10completely truthful). Finally, participants were debriefed, and the experiment was concluded. None of the participants were asked to stay longer and all participants were included in the raffle. Participation in the study took approximately 1 hour.

| Coding
Statements were assessed for the presence of spatial information (e.g., "Sitting in the row behind my friend"), temporal information (e.g., "It was 6:00 p.m."), and perceptual information (e.g., "I saw him sitting at the bar"; richness of detail), the verifiability of detail (e.g., a receipt of purchase), the presence of complications (e.g., missing the bus), common knowledge details (e.g., "We went to pick up groceries at the store"), and self-handicapping strategies (e.g., "I can't tell you anything else because my friend did all of the planning"), and the clarity and plausibility of the statement (statement quality). The exact description of the verbal content coding can be found in the Supporting Information. Two scores were created for each dependent measure; the first was the sum of all occurrences in the entire statement, the second constituted the sum within the time period from 1:00 p.m. to 3:00 p.m. To establish reliability, the main coder and a second coder evaluated a randomly selected 20% of the statements.
Using the two-way random effects model measuring consistency (see Koo & Li, 2016), interrater reliability was high for spatial information After confirming the reliability between the two coders, the main coder completed the remaining sample of participants' statements. In the analyses, we used only the scores of the main coder.

| Deviations from preregistration
The analyses reported here deviate from the preregistration in several ways. All deviations were decided upon prior to analyzing the data. First, we preregistered two separate analyses, one based on "quantity of details" (e.g., particular information regarding places, times, persons, objects, and events) and one based on "the richness of detail". Instead, we limit our analysis to "richness of detail," a combination of all spatial, temporal, and perceptual information. Second, we preregistered predictions based on a measure combining the frequency of complications, self-handicapping strategies, and common-knowledge details. Instead, we coded the frequency of each cue separately and calculated the proportion of complications score (complications/ [complications + common knowledge details + self-handicapping strategies]). This is in line with previous literature (see, for example, Vrij, Leal, Jupe, & Harvey, 2018) and has theoretical advantages given that it is a within-subjects comparison that is also sensitive to the different verbal strategies used by liars and truth tellers. Finally, we specified a priori hypotheses regarding participants' self-reported strategies. We report these analyses in the Supporting Information to keep the manuscript within reasonable length.

| Truthfulness measures
We asked participants to rate, on a scale of one to ten (one being not at all truthful and ten being completely truthful) how truthful the

| Statement characteristics
Our primary analyses focused on examining the characteristics between entirely truthful statements versus entirely fabricated statements, and between the deceptive 1:00 p.m. to 3:00 p.m. period embedded in lies and embedded in truths. Additionally, we conducted two exploratory analyses. 1 We compared the characteristics of the
Next, we tested Hypothesis 2 by conducting a second univariate between-subjects ANOVA to compare the detail richness of the 1:00 p.m. to 3:00 p.m. period between the embedded lie and complete lie conditions. We did not observe a significant effect for the richness of detail, F(1, 72) = .21, p = .648, η P 2 = .003, BF 01 = 3.80, meaning that lies embedded into otherwise truthful statements were not significantly richer in detail than lies embedded into entirely fabricated statements. As such, we did not find support for Hypothesis 2.  See Table 2 for the exact values. Thus, our analysis of the secondary content cues revealed no differences between lies embedded in truths and lies embedded in lies, with Bayes Factors demonstrating weak to substantial evidence in favor of the null hypothesis. The only exception was self-handicapping strategies.
Exact values can be found in Table 2. These results indicate that lies embedded into otherwise truthful accounts can be differentiated from truths based on detail richness and statement quality, although the Bayes factors indicate the evidence for this is weak at best.
We preregistered our hypothesis that lies embedded in truths would be richer in detail than lies incorporated into fully fabricated accounts. It is, however, also possible that embedded

| DISCUSSION
In line with previous research (e.g., Amado et al., 2016;Luke, 2019) We also did not find the proportion of complications to be a diagnostic cue to veracity in any of the three comparisons. This may have been due to floor effects in our sample (truthful interviewees in our study reported on average, less than one complication, relative to other studies in which truthful reports typically produce an average of ten or more complications; e.g., Vrij, Leal, Jupe, & Harvey, 2018;. Regarding common knowledge details, it has been suggested that truthful interviewees sound scripted in their reports if they underestimate the amount and type of detailed information they are required to report (Vrij, 2018). A potential reason for not observing significant differences with regard to the proportion of complications was due to the reduced time period, as well as the events, that participants were reporting in their alibi statement.
If participants had reported longer, more dynamic statements, perhaps after exposure to a model statement (e.g., Leal et al., 2015), then the proportion of complications may have been a more effective cue.
We also did not observe differences regarding the statement quality between completely truthful statements and completely fabricated statements when comparing the entire accounts, yet we did replicate this effect when comparing only the embedded 1:00 p.m.
to 3:00 p.m. component of the alibis. Reporting truthfully involves retrieving and reconstructing one's memory, whereas constructing a lie involves fabricating a story based on scripted knowledge about comparable situations and events (Schank & Abelson, 1977). Considering that liars in our study admitted having included some truthful information in their statements, it is possible that this allowed their overall statements to come across equally as clear and plausible as honest interviewees.

| Limitations and future research
The goal of our study was to examine embedded lies and we did so by isolating a critical statement of interest, while manipulating the veracity of the surrounding components. However, the period for which the liars came up with spanned 2 hours. In real life, liars may stay as close to the truth as possible, only fabricating or omitting a few key, incriminating details. Future research could extend our paradigm to accommodate for the dispersion of truths and lies throughout a statement and particularly how interviewees' verbal content may be inconsistent when they lie and tell the truth in a single account.
A second limitation is that the self-reported truthfulness ratings revealed that liars instructed to fabricate their entire account reported still including some truthful information, and the embedded liars reported that their general statement was mostly truthful, but still included some lies. This may be methodologically somewhat awkward but it does reflect what liars typically do: Providing statements that contain a mixture of truths and lies (Leins et al., 2013;Leins, Zimmerman, & Polander, 2017). As such, the finding is high in ecological validity. We did, however, check by self-report that liars did not engage in the assigned activity on the day in question nor on any adjacent days. This does not, of course, exclude the possibility that they engaged in the activity on an earlier occasion, meaning they could still have drawn from this truthful experience, simply displacing it in time. Future research that manipulates the type of lie that interviewees provide, such that it cannot be readily drawn from a potential previous experience, may produce a different pattern of results. Another methodological adjustment that may yield different findings would be allowing participants to choose the topic of their report, rather than constraining their reports to an activity scripted by the experimenter.
This would more appropriately reflect the circumstances of realworld liars, who are typically not forced to report any particular event (e.g., Leins et al., 2013).
Another important consideration relates to ground truth. Our study involved interviewees reporting self-generated stories within a naturalistic alibi scenario. We established partial ground truth via our truthfulness measures, which indicated that participants largely conformed to the experimental instructions. We were unable to further corroborate participants' accounts, however. Although the current experiment ensured that participants were emotionally engaged with the experimental process and similar paradigms have been used extensively by deception researchers (e.g., Elntib, Wagstaff, & Wheatcroft, 2015;Masip et al., 2005;Sporer & Sharman, 2006), future research would benefit from attempting to establish ground truth. A possible way to do so without having to resort to artificial mock crime procedures would be to require participants to wear a video-recording device for a certain duration of hours over a period of several days. Then, the researcher could verify the veracity of the interviewees' reports in the subsequent interview (e.g., Meixner & Rosenfeld, 2014).

| Conclusion
In sum, we showed that truthful statements could be distinguished from fabricated ones, and that lies embedded in otherwise truthful statements did not differ from lies embedded in deceptive statements. We also showed that lies embedded in otherwise truthful statements could be distinguished from truths embedded in truthful statements. Accordingly, verbal credibility assessment tools based on verbal content measured in this study may be robust against the embedding of lies.