Lying in Groups

In an online experiment, groups jointly identify the veracity of the text-based lies, where the group guess is determined based on the majority after members independently submit their binary guesses. We systematically vary the incentives of the group members with homogeneous or heterogeneous material payoffs (incentives) to mimic real-world social media groups. We measure whether the groups with heterogeneous incentives perform differently at detecting lies than homogeneous groups.

2025-04-23

2025-10-30

We pursue two lines of inquiry. First, do groups outperform individuals in detecting text-based lies when the group decision is decided by a majority-voting rule? Second, we analyze for possible differences in a group's performance when members are incentivized heterogeneously compared to when they receive homogeneous benefits.

We transcribe episodes of the American TV show, To Tell the Truth. Each episode of the show features three contestants, each asserting to be the real John/Jane Doe, while only one of them is the real John/Jane Doe, and four judges who cross-question the contestants based on some publicly known facts about the real John/Jane Doe (a.k.a. ~ objective truth). Our experimental setup mimics a situation similar to social media where a third party seeks to ascertain the truth concerning a particular topic (e.g., the accuracy of economic data or historical events) by observing online discussions among individuals who have conflicting motives related to that topic in the presence of an underlying objective truth. The main task in our experiment is to identify the real John/Jane Doe only from textual cues found in a transcript. A transcript includes the objective truths about the real John/Jane Doe and the conversations between the judges and the contestants. We have three treatments: individual, homogeneous groups, and heterogeneous groups. We recruit human participants from the online platform Prolific.

Participants will have the opportunity to review three sets of transcripts. After reading the objective truth and the conversations, participants will have to submit their guess about who the real John/Jane Doe is for that set. After submitting the guess for the set, participants will be asked how confident they feel that their guess is correct on a scale of 0 (indicating not confident at all) to 100 (indicating absolutely confident) using a slider bar. We ask participants to submit their absolute (Low/Medium/High Difficulty) and relative difficulty levels (Low/Medium/High Difficulty) for each set. Finally, we ask participants to submit their relative confidence levels. For our individual treatment, participants complete all these tasks. Participants earn a flat fee of $0.75 to submit their guess for each set, up to $1 for absolute confidence payments, and $0.50 each for relative difficulty and relative confidence levels. Additionally, after participants complete all the tasks, the computer randomly selects a set for bonus payments. If the participant's guess in that set is correct, they receive a bonus of $1.

For group treatments, we club each participant with two other participants to form a group of three members. A group of three is minimally sufficient to construe a group guess by majority voting rule, since we have two options of contestants to identify the real John/Jane Doe. Each group is randomly assigned a treatment condition. Each participant in the group is assigned a player number (1, 2, 3). For each treatment, participants are incentivised in the same way as in individual treatment to complete the same tasks as an individual. Each member participant of a group independently reads the transcript and submits a guess. For each transcript, participants indicate their absolute confidence in guesses. Additionally, we ask participants their beliefs regarding the probability of success of the other four members in their group - ‘According to you, how likely is Player i to get their guess correct?’. Participants indicate this on a scale of 0 (not likely at all) to 100 (very likely). The group guess is determined by majority voting for each of the three transcripts. For homogeneous group treatment, the group bonus is $3 and the bonus will be divided equally among the participants. Each participant gets $1 if the group guess for the randomly selected set is correct. The group bonus for the heterogeneous treatment condition is $5. For heterogeneous treatment condition, we introduce minimal heterogeneity by restricting a higher bonus to only one member of the group and equally dividing the remaining bonus amount among the other two participants. Of the three members, the member with the highest bonus gets $3, and members with lower bonuses get $1 each (maintaining parity with the homogeneous and individual treatment bonus).

Not available

Randomization is conducted using a random number generator in Python

Transcript level

Yes

For the Individual treatment, we recruit 60 participants.
For the Homogeneous and Heterogeneous treatments, we recruit 180 participants for each treatment. Of the 180 participants, we randomly form groups of 3 to have 60 groups in each treatment.

180 observations for individual treatment. 540 observations at the group-member level for each of the group treatments. 180 group-level observations for each of the group treatments.

Individual treatment: We recruit 60 participants, which is 180 observations (each participant takes three decisions for the three transcripts).
Homogeneous group treatment: We recruit 180 participants to form groups of 3. Each participant takes three decisions for the three transcripts, which is 540 observations at the group member level. We also have 180 group-level observations for the 60 groups of three from the 180 participants.
Heterogeneous group treatment: We recruit 180 participants to form groups of 3. Each participant takes three decisions for the three transcripts, which is 540 observations at the group member level. We also have 180 group-level observations for the 60 groups of three from the 180 participants.