Experimental Insights into Rumor Spread: Manipulation and Filtering Mechanisms

How to cite

charroin, liza; bravard, christophe. 2025. "Experimental Insights into Rumor Spread: Manipulation and Filtering Mechanisms." AEA RCT Registry. AEARCTR-0016910. https://doi.org/10.1257/rct.16910-1.2

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

liza charroin* , christophe bravard

* Primary investigator

Registry ID
AEARCTR-0016910
DOI
https://doi.org/10.1257/rct.16910-1.2
First registered
Last updated
Status
In development
Trial dates
2025-10-13 – 2025-12-19
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/16910

Abstract

This paper investigates how misinformation propagates through social networks using experimental methods that allow us to isolate key mechanisms of message transmission. We focus on the role of individuals in either transferring, blocking, or transforming a message, distinguishing between two treatments: one in which messages can only be transferred or blocked, and another in which individuals can also modify the content before passing it on. These treatments capture real-world dynamics, such as reposting versus quoting on social media, or faithfully relaying versus rephrasing in interpersonal communication.
Our framework departs from classical information cascade models by incorporating biased agents who may manipulate messages, alongside unbiased agents who are incentivized to evaluate credibility and diffuse true information. The central objective is to understand how networks filter information and whether the possibility of manipulation amplifies or mitigates the spread of false rumors. By comparing behavior across treatments, we examine the conditions under which misinformation becomes more or less credible.

Intervention

We design a between-subject experiment with 2 treatments: the No Transformation (NT) treatment and the Transformation (T) treatment.
In each treatment, participants will play the role of an agent in a line-network (cascade) of 6 agents. They may be in position 2, 3 or 4. Agent 1 is the only one who can observe the state of the world (here, a Heads-Tails draw, where Heads only has a 20% chance to be true). Agents in the network are either unbiased (their objective is to find the true state of the world and to diffuse rumors that are credible enough, or biased (their objective is to convince other agents that Heads is true, whatever the true state of the world).
Only the participant is a real agent, the 5 other agents are computerized: this is an individual task. The participant always has the role of an unbiased agent and agents only know the type of their direct friends. We vary the network structure (position of the participant, distribution of biased/unbiased agents, etc.) such that participants face 60 situations. In 42 situations, they receive a message, in 18 they receive nothing because the message was blocked upstream (the situations have been selected before the trial starts to match the probabilities given to participants and to expose them to all the situations possible).
We also designed a theoretical model that predicts what participants should do in each situation (depending on the network and the message they receive).
In both treatments, they have 2 decisions to make:
- Decision 1: based on the network and the message received, they have to guess the state of the world
- Decision 2: they have to decide if they want to transfer, block (or transform in T) the message received.
We link these 2 decisions to avoid incoherent decisions: for instance, if they guess that Heads is the true state of the world, they cannot transfer a Tails message.
Both decisions are incentivized.
In the (18) cases where no message is received, they just have to make Decision 1. In order to shorten the experiment, they will face these 18 situation on one screen at the end of the experiment.

Experimental design

We design a between-subject experiment with 2 treatments: the No Transformation (NT) treatment and the Transformation (T) treatment.
In each treatment, participants will play the role of an agent in a line-network (cascade) of 6 agents. They may be in position 2, 3 or 4. Agent 1 is the only one who can observe the state of the world (here, a Heads-Tails draw, where Heads only has a 20% chance to be true). Agents in the network are either unbiased (their objective is to find the true state of the world and to diffuse rumors that are credible enough, or biased (their objective is to convince other agents that Heads is true, whatever the true state of the world).
Only the participant is a real agent, the 5 other agents are computerized. The participant always has the role of an unbiased agent and agents only know the type of their direct friends. We vary the network structure (position of the participant, distribution of biased/unbiased agents, etc.) such that participants face 60 situations. In 42 situations, they receive a message, in 18 they receive nothing because the message was blocked upstream (the situations have been selected before the trial starts to match the probabilities given to participants and to expose them to all the situations possible).
We also designed a theoretical model that predicts what participants should do in each situation (depending on the network and the message they receive).
In both treatments, they have 2 decisions to make:
- Decision 1: based on the network and the message received, they have to guess the state of the world
- Decision 2: they have to decide if they want to transfer, block (or transform in T) the message received.
We link these 2 decisions to avoid incoherent decisions: for instance, if they guess that Heads is the true state of the world, they cannot transfer a Tails message.
Both decisions are incentivized.
In the (18) cases where no message is received, they just have to make Decision 1. In order to shorten the experiment, they will face these 18 situation on one screen at the end of the experiment.

Design details

Randomization method
The order of the 60 situations has been randomly drawn once, before the trial starts, such that each participant faces the same sequence in both treatments.
Randomization unit
Each experimental session will randomly be a NT or a T treatment.
Clusters
120 participants
Observations
120 participants
Arms
60 participants in each treatment
Primary outcomes
Two key outcomes: - Decision 1: in each situation, which state do they think is the true one. - Decision 2: in each situation, do they transfer or not a message and do they transform it? Our goal is to see whether the filtering of the network makes messages more credible in one treatment or the other.

Cited by

No citation data available for this registration yet.

Keywords