Our studies were not pre-registered. In our initial study, we leverage an existing conversational database to test whether there is a positive relationship between self-ratings of trustworthiness and reported feelings of closeness with a conversational partner [H1]. Additionally, we test whether perceived trustworthiness of one’s partner mediates this relationship [H2].
Study 1 Corpus
BetterUp Inc. released a multimodal dataset of naturalistic conversations collectively referred to as the CANDOR corpus (Conversation: A Naturalistic Dataset of Online Recordings)28. The corpus includes over 1TB of data, including raw and processed recordings, transcripts, behavioral measures, and survey responses from a large, diverse sample of participants based in the United States. This study received approval by Ethical & Independent Review Services.
Participants were recruited using Prolific; recruitment targeted individuals based in the United States and 18 years of age or older. Between January and November 2020, six rounds of data collection yielded a total of 1656 dyadic conversations that were recorded over video chat. Participants provided informed consent to have a conversation with a stranger that would last at least 25 minutes, complete survey ratings of their experience, and have their data made publicly available. Dyads included in our analyses conversed for 29 to 113 minutes with an average conversation length of 30.01 minutes (SD = 7.79). Participants were paid $0.85 for completing an initial scheduling survey and an additional $14.15 upon full completion of the recorded conversation and post-conversation survey.
We received the CANDOR corpus in March 2022. Dyads were included in analyses if both participants completed the post-conversation ratings of personal trustworthiness, perceived trustworthiness of their partner, and the measure of interpersonal closeness with their partner. No further data exclusions were made. This resulted in a total of 388 complete dyads in the analyses below. Of these 776 participants, 423 identified as female, 321 as male, and 5 as other or prefer not to answer. Participants were an average age of 33.81 (SD = 10.98; range = 19–63). Twenty-seven participants did not provide age information.
Study 1 Procedures
Dyads were matched according to their shared availability, which participants reported in an initial survey. Once matched, participants were notified via email of the time and date of their conversation.
A brief survey was conducted to measure participants’ current mood prior to their conversation28. This survey instructed participants to make sure their webcam and microphone were enabled and to have a conversation lasting at least 25 minutes. Then, a link was provided to the video chat room, where the conversation was to take place.
As soon as the first participant clicked the link from the pre-conversation survey, the recording began. Participants were not given any specific instructions regarding conversation content. Instead, they were told to “talk about whatever you like, just imagine you have met someone at a social event and you’re getting to know each other.” After completing the conversation, participants ended the recording session and returned to their original web browser to complete the post-conversation survey.
Following the conversation, participants reported on their experience of the conversation and perceptions of their conversation partner’s and their own psychological states and traits28. Importantly for the purposes of this investigation, participants responded to the statements, “How would you rate yourself on each of the following traits?—trustworthy?”, and “To what extent does your conversation partner have each of the following traits?—trustworthy?”, on a 1 (not at all) to 9 (extremely) scale. Additionally, participants were asked to rate their agreement with the statement “I felt close to my partner” on a 1 (strongly disagree) to 7 (strongly agree) scale. Following completion of the post-conversation survey, participants were thanked for their time and provided with information on how to make a request for payment.
A limitation of relying on the CANDOR dataset28 to test our hypotheses is that participants responded to a single item, trait-measure of their personal trustworthiness, but did not provide details on the trustworthiness of their communication in the conversation, specifically. In Study 2, we manipulated sender veracity in the context of a dyadic conversation and subsequently measured feelings of closeness with the conversational partner, allowing us to establish a causal link between telling lies and feeling socially distant [H1]. We also examine ‘deceiver’s distrust’ as a potential mechanism for this relationship [H2].
Study 2 Participants
Our sample size goal (N = 200 dyads) was identified on the basis of similar research, involving stranger dyads in a chat-based, two-group experimental design29. Participants were recruited using Prolific. All online workers in the United States who were 18 years of age and older were eligible to participate. Participants provided informed consent and were welcome to withdraw their participation in the study, for any reason at any point during the study, without penalty or loss of compensation. Data was collected between April 11 and May 25, 2022. Participants received $15 CAD in compensation.
A total of 212 dyads provided complete data. Four dyads were removed prior to analysis for not following instructions (i.e., did not use conversation-starter questions). The final dataset included 208 dyads or N = 416 participants (189 men; 216 women; 8 non-binary; 3 gender not listed; 2 prefer not to say). The mean age of our participants was 37.92 (SD = 13.29; range: 18-84).
Study 2 Materials and measures
Colloquially referred to as ‘fast friends’, this procedure was developed for experimentally manipulating closeness among strangers in a laboratory setting3. In the original experimental condition, stranger dyads alternated in asking each other a list of 36 increasingly personal questions over a 45-minute face-to-face conversation. Reciprocal disclosure of personal details led to increased interpersonal closeness, relative to a small-talk condition3. For the purposes of our study, we selected six increasingly personal questions for our stranger dyads to discuss. We referred to these questions as conversation-starters. Specifically, the following questions were used:
-
What would constitute a “perfect” day for you?
-
If you could change anything about the way you were raised, what would it be?
-
Is there something that you’ve dreamed of doing for a long time? Why haven’t you done it?
-
What is your most treasured memory?
-
If you were going to become a close friend with your partner, please share what would be important for him or her to know.
-
Share with your partner an embarrassing moment in your life.
Following the conversation, participants completed several manipulation checks, and scales relating to their experience, evaluation of their partner, and their personality. Of particular interest was a measure of perceived honesty of their partner, which served as an index of ‘deceiver’s distrust’ (i.e., our mediator), and two single-item scales measuring interconnectedness and closeness with their partner, which served as outcome measures of interpersonal closeness. We also included trait measures of deception frequency and loneliness. Due to experimenter error, these scales were added mid-way through data collection. Accordingly, data on these scales is not available for all participants. Each of these measures are described in detail, below.
The Reysen Honesty Scale30 is an 8-item measure which provides a measure of the extent to which an individual is perceived as honest and includes items such as “I believe what this person says” and “I trust this person will tell me the truth.” Items were rated on a 7-point Likert scale and showed high internal reliability (α = 0.91).
The Inclusion of Other in Self (IOS)31 measure is a single-item measure of interpersonal closeness which uses a series of images of overlapping circles representing the self and the conversational partner. A set of Venn-like diagrams, ranging from completely separate to completely overlapping circles make up this pictorial 7-point scale. This measure of interpersonal closeness was designed to tap into people’s sense of being interconnected with another person. The scale has demonstrated strong convergent validity with lengthier measures of closeness (i.e., Relationship Closeness Inventory)32 and has been used to validate the experimental generation of closeness, using the guided conversation questions that we employed in this study3. As a supplementary measure of closeness, we also included one face-valid item, “On the following scale, please rate how close you feel to your partner”, which participants answered on a 7-point Likert scale from 1 (not at all) to 7 (very).
Additionally, participants were asked to rate the extent to which they engaged in deception during the discussion of each topic-question that guided their conversation. Specifically, they were asked: “Please rate the extent to which your answers to each of the following questions involved deception (i.e., lies meant to mislead your partner)” followed by a list of the topic-questions, and a 1(not at all deceptive) to 5 (completely deceptive) scale with an option to indicate that they did not discuss this topic question (“Ran out of time before we got to this question”). A mean of responses on these items (excluding “Ran out of time …” responses) was calculated to provide a measure of deception by each participant.
The Lying in Everyday Situations (LiES)33 survey consists of 14-items and provides a measure of the extent to which people participate in deception in their daily lives (overall score: α = 0.90). Two 7-item subscales tap the use of vindictive lies (α = 0.93) and relational lies (α = 0.91), specifically. Vindictive lies are generally told to harm others or benefit the self (e.g., “I lie for revenge”) while relational lies were told to maintain social cohesion (e.g., “I tell lies in order to spare another’s feelings”). These scales showed high test-retest reliability and strong convergent validity with related scales, including Machiavellian personality traits and self-reports of deception frequency33.
The UCLA Loneliness-8 (ULS-8)34 scale is a short-form version of the 20-item UCLA Loneliness Scale (ULS-20)35. This 8-item scale provides a brief measure of the subjective sense of loneliness—a deficiency in social contact, relative to what is desired. This short-form version of the scale showed high internal reliability (α = 0.91) in our sample and previous research suggests that it is highly correlated with the original 20-item version (r = 0.91)34, while reducing participant burden.
Participants were also asked to complete the Relational Communication Scale36, rate their partner on basic dimensions of social evaluation (e.g., warmth, competence, morality)37, likeability, and indicate whether they thought their partner was lying to them. It is worth noting that only n = 10 of 209 (i.e., 4.8%) of the receivers responded ‘yes’ to the question, “At any point during the conversation did you think that your partner was lying to you?”, indicating a strong truth bias among receivers24. A full list of measures and the order in which they were asked can be found in the Qualtrics file, posted to OSF (https://osf.io/ezn7p/). These measures were included to test a separate series of hypotheses and are not examined further here.
Study 2 Procedures
Participants completed the consent form and were provided instructions about their upcoming conversation. All participants were told they would be paired with a random stranger for a chat-based conversation using ChatPlat software (www.chatplat.com) embedded in the Qualtrics survey. ChatPlat provided the capacity to match senders with receivers in real time, a forum for text-based conversations to occur, and recording of text data. Participants were told that they would be provided with six conversation-starter questions to guide the discussion. Participants were asked to imagine that they were considering their conversation partner as a potential roommate and to use this conversation to get to know them to determine whether they would be a good fit. They were encouraged to ask follow-up questions and to move on to the next conversation-starter when the previous topic had been exhausted. Participants were also asked to keep the nature of their conversation confidential.
Participants completed several attention-check questions to ensure that they understood these instructions; they could not move forward until they had answered all questions correctly. Participants were then provided with a list of the conversation-starter questions and were given two minutes to review them before proceeding.
Participants were then randomly assigned to be Participant 1 who would always answer the conversation-starter questions first, or Participant 2 who would answer second. For ease of description, we refer to Participant 1 as the sender, and Participant 2 as the receiver, throughout the manuscript. Senders were further randomly assigned to be as complete, open, and honest as possible, or to lie to their partner for the entirety of the conversation. Senders in the lie condition were assured that these were secret instructions and were asked to be as convincing as possible. All receivers were asked to be as complete, open, and honest as possible.
Participants then entered the chat where a sender and receiver were always paired together. Dyads were instructed to chat until they discussed all six questions, or up to 25 minutes (whichever came first). Participants’ conversations ranged from 5.16 minutes to 28.25 minutes, with an average duration of 21.74 minutes (SD = 4.97). An independent samples t-test yielded no statistically significant evidence that senders responded to a different number of questions between the truth (M = 4.38, SD = 1.69) and lie (M = 4.49, SD = 1.40) conditions, t(206) = 0.51, p = 0.610, d = 0.07 [95% CI: −0.34, 0.20]. We used JABApprox.1 function in R to approximate the Bayes Factor, finding a BF of 18.02, indicating strong support for the null hypothesis. After exiting the chat, participants were asked if they were successfully paired with another participant and able to complete the chat session. If not, they were able to exit the study while still receiving compensation for their time. For those who were successfully paired, they proceeded to complete measures about their partner, their experience in the conversation, and themselves. Participants also provided demographic information, were debriefed about the true nature of the study, and were provided compensation. This study was approved by the University of British Columbia, Okanagan Behavioural Research Ethics Board.
In our final study, we further considered whether the dispositional tendency to tell lies is associated with a general distrust of others and a sense of loneliness. In Study 3, we sought to gather a larger sample than in Study 2, include additional control variables, and examine the mediating role of dispositional interpersonal trust. Thus far, loneliness has been characterized by an individual’s subjective lack of social connection, however, in Study 3 we examine the role of lacking social connection objectively. Studies including objective network characteristics such as the number of friends, relatives, and frequency of contact with social network members show an inverse relationship with loneliness38,39,40,41. Specifically, in Study 3 we included objective social network characteristics (i.e., social network size and diversity) as control variables in our analyses. As before, we expected that lies would be associated with a sense of loneliness, even when accounting for the number and type of close social ties [H1]. Additionally, we examined whether interpersonal trust mediated the relationship between dispositional use of deception and the experience of loneliness [H2].
Study 3 Participants
We recruited participants using Prolific. All online workers in the United States and 18 years of age or older were eligible to participate (participant age data is unavailable due to experimenter error). This study received approval from the Ethics board at the University of British Columbia. Participants provided informed consent and could withdraw participation in the study for any reason, at any point during the study, without penalty or loss of compensation. A power analysis indicated that N = 395 participants would be necessary to find a small effect (f2 = 0.02) in a multiple regression with three predictors, setting p = 0.05 and 1-β = 80%. N = 399 participants completed the study, and no data exclusions were made. Of these individuals, 178 participants identified as men, 206 as women, 11 as non-binary, and 4 selected ‘prefer not to say’. One participant selected both ‘woman’ and ‘non-binary’. Data was collected on October 13, 2022. Prolific participants received $3.00 CAD for their time.
Study 3 Measures
Participants completed the following four scales: Lying in Everyday Situations Scale (LiES)3, General Trust Scale (GTS)42, the revised UCLA Loneliness Scale (ULS-20)35, and the Social Network Index Revised (SNI)43.
The UCLA Loneliness Scale (ULS-20)35 includes 20 items (e.g., No one really knows me well), each rated on a 1 (never) to 4 (often) scale. It is widely used to measure loneliness and a mean score of all items has demonstrated strong convergent and discriminant validity with related constructs35. This measure showed high internal consistency in our sample (α = 0.92).
The General Trust Scale (GTS)42 is a 6-item measure and uses general statements to measure beliefs about the honesty and trustworthiness of others. Each item is rated on a 1 (strongly disagree) to 7 (strongly agree) scale. This measure showed high internal consistency in our sample (α = 0.87).
The Social Network Index Revised (SNI)43 measure asks participants to report on participation in 12 types of social relationships (e.g., spouse, parents, parents-in-law, children, other close family members, close neighbors, friends). To assess network diversity, participants are assigned one point for each relationship type (maximum score: 12) for which they indicate they speak to a person (or multiple people) fitting that description at least once every two weeks. We also calculated social network size by summing across the 12 roles. To ensure that this figure was not artificially inflated by individuals reporting large group memberships (e.g., of volunteer organizations), we followed the recommendation of recoding very large groups to have an upper bound of seven43.
Study 3 Procedure
Participants provided informed consent and completed the Lying in Everyday Situations Scale (LiES)3, General Trust Scale (GTS)42, the revised UCLA Loneliness Scale (ULS-20)35, and the Social Network Index Revised (SNI)43 in random order. Participants also provided basic demographic information before being debriefed, thanked for their time, and provided compensation. This study was approved by the University of British Columbia, Okanagan Behavioural Research Ethics Board.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.