Narratives and the communication gap

In the experiment, participants first observe their own data and choose their narrative based on this data, and then are paired to communicate and choose their narrative again. The first stage is a Baseline, and the second stage has three different communication setting. The main intervention is that participants are randomly assigned into one of three treatments, each with differing communication settings. In the Communication treatment, participants only observe their own data during communication. In the Communication+Share Data treatment, participants observe not only their own data but also their partner’s data, as well as a merged data combining both. In the Communication+Reasoning treatment, participants only observe their own data during communication; while before communication, they are asked to write a piece of advice for future experiment participants on how to choose the correct narrative.

The secondary intervention is whether a participant is paired to communicate with a partner who initially chose the same narrative (referred to as a homogeneous pair) or a differing narrative (referred to as a heterogeneous pair). The participants are randomly paired without regard to their initial narrative choices in the experiment.

2025-04-08

2025-12-31

There are two main outcomes variables: (1) Consensus rate. (2) Correct rate.

(1) Consensus rate: It is defined as the number of heterogeneous (homogeneous) pairs in which both participants choose the same narrative after communication, divided by the total number of heterogeneous (homogeneous) pairs.
(2) Correct rate: It is defined as the number of participants who choose the correct narrative (i.e., true narrative), divided by the total number of participants. This variable is calculated both before and after communication.

Rate of change (from stage 1 to stage 2)

The rate of change is defined as the fraction of participants who change their narrative choice before and after communication. It reflects participants’ willingness to revise their narratives.

The experiment consists of two main decision stages. At the beginning of each session, the computer randomly selects one of two possible narratives, referred to as “narrative A” and “narrative B”, to serve as the true narrative. This true narrative remains fixed throughout both stages and is the same for all participants within the same session. In the first stage, each participant observes a random subset of data generated based on the true narrative and is asked to choose the narrative they believe to be true. No feedback is provided after the first stage. This stage serves as a Baseline for the second stage.

In the second stage, each participant is randomly paired with another participant from the same session. The pair is allowed to communicate freely via a chatbox on the computer for up to 15 minutes. During communication, participants see both their own and their partner’s narrative choices in the first stage. After communication, participants are asked to choose the narrative they believe to be true again. These two stages (one without communication, one with communication) are within-subject treatments.

There are three between-subject treatments in the main experiment. In the Communication treatment, participants can only review their own data in the first stage during communication. The Communication+Share Data treatment differs from the Communication treatment in that participants can review three sets of data: their own data, their partner’s data, and a merge of both data. In the Communication+Reasoning treatment, participants review only their own data during communication, as in the Communication treatment. However, the Communication+Reasoning treatment includes an additional stage before communication, in which participants are asked to write a piece of advice for future experiment participants about how to identify the true narrative, which requires explicitly stating their reasoning of the narrative.

After we finish the three treatments of the main experiment, we will also run an additional follow-up experiment in a different university. In this experiment, each participant also sees a dataset from Narrative A or B, and is asked to choose which one is the true narrative. Moreover, they are provided with three pieces of advice (elicited from the Communication+Reasoning treatment), and are asked to select the most helpful advice. If an advice is selected to be the most helpful one, the corresponding advisor will receive an additional payment.

In all three treatments, participants are incentivized to correctly choose the true narrative in both stages, earning 40 RMB for a correct choice and zero otherwise. At the end of the experiment, one of the two stages is randomly selected for payment.

In the Communication+Reasoning treatment, participants are also incentivized to write helpful advice. Their advice is evaluated by third-party participants from another university, each of whom read three anonymized pieces of advice and selects the most helpful one. Participants whose advice is selected as the most helpful earns an additional 20 RMB.

In the experiment, Narrative A says “for the observed gender pay gap, less than 10% can be explained by observable difference in men and women, and more than 90% cannot be explained by observable differences.” Narrative B says “for the observed gender pay gap, explainable part and unexplainable part each account for about 50%.” The content of narrative A and narrative B, as well as the corresponding datasets, are constructed and generated based on results from the two following papers:
(1) Zhang C, Tian X, Yang X, et al. The iron-out effect of digital economy: A discussion on gender wage rate discrimination for working hours[J]. Journal of Business Research, 2023, 156: 113399.
(2) Ma X. Internet use and gender wage gap: evidence from China[J]. Journal for Labour Market Research, 2022, 56(1): 15.

Each experimental session randomly implements one of the three treatments. Within a session, subjects are randomly paired.

Individual-level randomization

No

We aim to collect about 20-30 heterogeneous pairs in each of the three between-subject treatment arms. Since we expect that around 40-50% of randomly matched pairs are heterogeneous pairs, we aim to collect about 100-150 subjects for each treatment.

In total about 300-450 individuals, recruited in the subjects’ pool of the Shanghai University of Finance and Economics. The number of third-party participants from a different university depends on the number of observations in the Communication+Reasoning treatment. We guarantee that for each piece of advice elicited in the Communication+Reasoning treatment, it needs to be used at least once in the third-party experiment.

About 100-150 subjects per between-subject treatment arm.