Algorithmic News Feeds, Personalization, and Democratic Outcomes: Evidence from an App Patient-Preferred Field Experiment in Italy

How to cite

Vecchiato, Alessandro. 2020. "Algorithmic News Feeds, Personalization, and Democratic Outcomes: Evidence from an App Patient-Preferred Field Experiment in Italy." AEA RCT Registry. AEARCTR-0006014. https://doi.org/10.1257/rct.6014-1.2000000000000002

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

Alessandro Vecchiato*

* Primary investigator

Registry ID
AEARCTR-0006014
DOI
https://doi.org/10.1257/rct.6014-1.2000000000000002
First registered
Last updated
Status
In development
Trial dates
2020-08-10 – 2020-09-30
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/6014

Abstract

Algorithmic news feeds provide a combination of news stories with a very high degree of personalization. Personalized political information may affect democratic outcomes by how it bolsters false consensus beliefs, and that, in contexts of high levels of polarization, may create an artificial wedge between political parties. This paper investigates the implications of personalized news feeds, by implementing a pre-registered, globally replicable, lab-in-the-field experiment, in which subjects are treated with a custom-developed news app, and then evaluated with surveys in the changes to their political and democratic beliefs. The experiment is a patient-preferred trial to explicitly account for the impact of self-selection. Results from the pilot show that algorithmic news feed are quite powerful in determining media diets, but ineffective at changing preferences.

Intervention

This intervention requires participants to download a news-feed app that collects real news headlines from all national newspapers in Italy. Subjects are asked to use the app for a period of 15 days. The app provides two manipulations, including a chronological and a personalized news-feed. Outcomes will be evaluated on a number of politically relevant variables, including voting intentions and political knowledge.

Experimental design

The experiment includes two treatment conditions (popular and personalized news feeds) and one control condition (no app). The experimental protocol includes two treatment arms, one with self-selection into treatment and one with randomized treatment. This design will allow the experimenter to develop estimates of the likelihood of users taking up treatment, and therefore estimate the outcome averages were the treatment be implemented outside of the experimental setting.

Design details

Randomization method
Randomization done in office by a computer.
Randomization unit
Treatment is assigned at the participant level.
Clusters
The subject pool is of 700 users.
Observations
700 observations.
Arms
250 control, 550 treatment (50% per treatment arm).
Primary outcomes
The study includes behavioral and surveys outcomes. The app will collect information on users' behavior online, including reading time, clicks, and number, and length of sessions. Survey outcomes include political preferences, political knowledge, and second-order beliefs on others.

Cited by

No citation data available for this registration yet.

Keywords