Russian Disinformation in Kenya

How to cite

Berlin, Maria; Olofsgård, Anders; Arbuzova, Anastasiia. 2026. "Russian Disinformation in Kenya." AEA RCT Registry. AEARCTR-0019435. https://doi.org/10.1257/rct.19435-1.0

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

Maria Berlin* , Anders Olofsgård , Anastasiia Arbuzova

* Primary investigator

Registry ID
AEARCTR-0019435
DOI
https://doi.org/10.1257/rct.19435-1.0
First registered
Last updated
Status
On going
Trial dates
2026-07-30 – 2026-09-15
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/19435

Abstract

This study examines how young adults in urban Kenya respond to potentially misleading information on social media and whether short warning messages can help people assess such information. Approximately 5,000 people aged 18–40 in Nairobi and other urban areas of Kenya will participate in an online survey.
Participants will be shown two examples of social media content concerning topics relevant to Kenya and international affairs. They will then see different types of messages designed to alert people to potentially misleading or manipulative information. These warnings vary in who provides the warning, how the source of the potentially misleading information is described, and how the warning is presented, for example as social media content or a radio advertisement. Some participants will not receive a warning, allowing us to compare responses across groups.
The study will examine whether these different approaches affect people's willingness to share social media content, how credible they find the information, and their views on international actors. The broader aim is to better understand how people in Kenya respond to foreign influence and misleading information online, and which types of warnings may help people identify and resist such content.

Intervention

The study fields a cross-sectional survey with an embedded experiment among approximately 5,000 individuals aged 18–40 in Nairobi and other urban centres across Kenya. Respondents are asked to read two social media posts, one questioning the motive behind western companies selling gene-modified crops in Kenya, and the second mimicking the language used in posts trying to attract Kenyans to sign up for work in the Russian military. To encourage respondents to read the messages carefully they are asked to rate the posts based on spelling and grammar. Respondents are then randomized into a control group and eight treatment arms receiving variations of disinformation-debunking warnings. Warnings differ across three dimensions: the source of the disinformation warning (European Union vs. a local independent fact-checking organization), whether the spread of disinformation is attributed explicitly to Russia or to a generic foreign power, and the medium of the message (radio ad versus a social media post). The key outcome variable is social media sharing intention, but we also look at perceived message credibility, and opinions on global powers.

Experimental design

The study uses a 2×2×2 factorial design, crossing three binary treatment dimensions, plus a control group, yielding nine experimental arms. The three treatment dimensions are:
• Warning source: European Union (EU) vs. local independent fact-checking organization (hereafter 'Local NGO')
• Actor attribution: the false message is attributed explicitly to Russia vs. attributed to a generic unspecified foreign power
• Warning medium: the warning is delivered as a social media post or a radio add.

Design details

Randomization method
Randomization is implemented at the survey programming level, with 1,000 respondents assigned to the control group and 500 with equal probability to one of the eight treatment groups. Assignment to each group is determined before the respondent answers any questions. Randomization is stratified by age group, education, admin1 region, and gender to ensure balance on these dimensions across arms. Treatment assignment is recorded automatically by the survey platform.
Randomization unit
Individual.
Clusters
The treatment is not clustered, 5,000 respondents.
Observations
5,000 respondents.
Arms
1,000 respondents assigned to the control group and 500 with equal probability to one of the eight treatment groups.
Primary outcomes
Social media sharing intention.
Secondary outcomes
Intention to discuss the messages with friends, family and colleagues; opinions towards global powers; suspicion about the message; credibility of the warning.

Cited by

No citation data available for this registration yet.

Keywords