Political Communication on WhatsApp: Mechanisms

How to cite

Carney, Kevin. 2025. "Political Communication on WhatsApp: Mechanisms." AEA RCT Registry. AEARCTR-0015208. https://doi.org/10.1257/rct.15208-1.0

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

Kevin Carney*

* Primary investigator

Registry ID
AEARCTR-0015208
DOI
https://doi.org/10.1257/rct.15208-1.0
First registered
Last updated
Status
In development
Trial dates
2025-01-16 – 2025-02-15
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/15208

Abstract

This online survey experiment builds upon a previous field experiment (Carney, 2024), which studied the causal impact of political WhatsApp groups. The survey uses the content of messages posted to political WhatsApp groups during that study and varies the context in which the content is shared, to understand how people react to particular features of political WhatsApp messages. Participants will see a series of images of WhatsApp messages and be asked a series of questions about how they would view and respond to the message. Each screenshot will feature two messages. The messages will randomly vary in (1) whether they are attributed to a political party or an individual, (2) whether a different individual replied in agreement, disagreement, or with a neutral message, and (3) whether the chat is open to replies from other users.

Intervention

Participants will see a series of 9 screenshots of a hypothetical conversation in a WhatsApp group, each featuring two messages in the group. They will be asked to read the two messages and respond to a series of questions after each of the 9 screenshots.

Experimental design

There are 9 possible combinations of messages:

(1) Party-party (with reply box)
(2) Party-party (no reply box)
(3) Party-individual (individual agrees)
(4) Party-individual (individual neutral)
(5) Party-individual (individual disagrees)
(6) Individual-individual (individual 2 agrees)
(7) Individual-individual (individual 2 neutral)
(8) Individual-individual (individual 2 disagrees)
(9) Individual-individual (both from the same individual)

Comparing 1 vs. 2 estimates the role of "voice" (being able to reply)
Comparing {3, 4, 5} vs. {6, 7, 8} and 1 vs. 9 estimates the role of sender identity (party vs. individual)
Comparing {3, 7} vs. {4, 8} vs. {5, 9} estimates the role of agreement and disagreement
Comparing 7 vs. 9 estimates the role of hearing from multiple people

These are the primary comparisons of interest.

Individuals will be recruited for an online survey using Facebook ads.

Design details

Randomization method
Randomization occurs within subject: each subject will see all 9 of the possible message combinations described above. The subjects' 9 screenshots will be created by taking a bank of 18 possible messages, and randomly assigning them to a position within the 9 possible combinations described above without replacement.
Randomization unit
Individual-message
Clusters
We will recruit approximately 3000 individuals. Each individual will see 9 randomized messages, so we will treat these as unique observations, including individual fixed effects and clustering standard errors at the individual level.
Observations
27000 person-image pairs (3000 individuals will see 9 screenshots each). This may be lower depending on recruitment ad click rates and survey completion rates (respondents who view some screenshots but do not complete all 9 will be retained in the sample).
Arms
3000 individuals for each of the 9 types of message combinations.
Primary outcomes
For each image, respondents will be asked 7 questions: (1) How interesting is the message? (2) How likely would you be to forward one of these messages? (3) How likely would you be to reply to one of these messages? (4) How confident are you that the information shared is true? (5) Does this chat change your opinion about the DMK? (6) Does this chat change your opinion about the AIADMK? (7) How entertaining is the chat? These questions will be answered using a Likert scale. We will analyze these outcomes in isolation and jointly using an inverse covariance index following Anderson (2008).

Cited by

No citation data available for this registration yet.

Keywords