Labeling (Mis)information

How to cite

Louis-Sidois, Charles. 2026. "Labeling (Mis)information." AEA RCT Registry. AEARCTR-0019254. https://doi.org/10.1257/rct.19254-1.1

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

Charles Louis-Sidois*

* Primary investigator

Registry ID
AEARCTR-0019254
DOI
https://doi.org/10.1257/rct.19254-1.1
First registered
Last updated
Status
In development
Trial dates
2026-07-31 – 2027-07-31
Countries
France ()
JEL codes
D83, D72
Registry record
http://www.socialscienceregistry.org/trials/19254

Abstract

This study investigates how individuals label pieces of information they encounter, with a particular focus on labels associated with misinformation such as conspiracy theory. Using a survey experiment conducted on a representative sample of French adults, I examine how the identity of the alleged actor and the political affiliation of the source of a claim influence labeling decisions. Respondents are exposed to short claims and asked to classify them, while the actor or source attributed to each claim is randomly varied. This design allows me to identify whether labeling decisions are driven purely by respondents' perception of the claim - such as its perceived veracity - or also reflect their political preferences.

Intervention

Respondents are exposed to four short claims, each structured around a common template in which an actor secretly engages in a potentially harmful action. For each claim, one of three versions is randomly and independently assigned to each respondent. Two claims vary the identity of the alleged actor across versions; two vary the political source attributed to the claim. The order in which respondents see the claims is randomized. After each claim, respondents answer questions about their labeling decisions and their perception of the claim.

Experimental design

The study is a survey experiment conducted on a representative sample of French adults recruited via an online platform. Participants first answer background questions on socio-demographic characteristics, political preferences, and support for various actors and public figures.

Participants are then exposed to four short claims, each presenting an actor secretly engaging in a potentially harmful action. For each claim, one of three versions is randomly and independently assigned: two claims vary the identity of the alleged actor, and two vary the political source attributed to the claim. The order of claims is randomized across respondents.

After each claim, respondents answer two sets of questions. The first elicits labeling decisions: respondents indicate whether they would classify the claim as a conspiracy theory, misinformation, a rumor, or credible information. The second elicits assessments of the claim's plausibility, potential harm, and prior familiarity. An attention check is included at the end of each claim block. The order of the labeling questions relative to the assessment questions is randomized across respondents, as is the order of the labeling categories.

At the end of the survey, respondents complete an incentivized belief elicitation task in which they estimate the share of other respondents who labeled each claim as a conspiracy theory.

The study is conducted in two waves: a pilot study followed by a main study, the size of which will be determined based on pilot results and a formal power calculation.

Design details

Randomization method
Randomization is performed automatically by the survey software (Qualtrics) at the time of data collection. For each respondent, one of three versions of each claim is selected independently and with equal probability using Qualtrics' built-in randomization function. The order of claims and the order of labeling categories are also randomized by the survey software.
Randomization unit
Randomization is at the individual level.
Observations
Approximately 1,000 to 3,000 respondents (individuals), covering both a pilot study and a main study. The final sample size of the main study will be determined based on a formal power calculation conducted after the pilot and subject to the availability of French respondents on the recruitment platform.
Arms
Not applicable
Primary outcomes
(main) Binary indicator equal to 1 if the respondent labels the claim as a conspiracy theory. Binary indicator equal to 1 if the respondent labels the claim as misinformation. Binary indicator equal to 1 if the respondent labels the claim as a rumor. Binary indicator equal to 1 if the respondent labels the claim as credible information. Respondent's belief about the share of other participants who labeled the claim as a conspiracy theory.
Secondary outcomes
Perceived veracity, perceived harm, familiarity.

Cited by

No citation data available for this registration yet.

Keywords