Inclusive Platform Design

How to cite

Kumar, Madhav; Singh, Ashudeep; Silva, Pedro. 2025. "Inclusive Platform Design." AEA RCT Registry. AEARCTR-0015853. https://doi.org/10.1257/rct.15853-1.0

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

Madhav Kumar* , Ashudeep Singh , Pedro Silva

* Primary investigator

Registry ID
AEARCTR-0015853
DOI
https://doi.org/10.1257/rct.15853-1.0
First registered
Last updated
Status
Completed
Trial dates
2023-03-28 – 2023-04-24
Countries
United States of America ()
JEL codes
D22, J15, D83, C93, L86, L25
Registry record
http://www.socialscienceregistry.org/trials/15853

Abstract

This study examines whether online platforms can enhance representation without affecting business performance. Using a large-scale field experiment on a major visual discovery platform, we randomly assign users to see either standard content recommendations or a more balanced representation of skin tones in suggested content. We analyze how this change affects various engagement metrics, including overall platform activity and interaction with diverse content. We also explore how different user segments respond to more inclusive content and whether these effects persist over time. Finally, we test if our approach extends to other visual attributes beyond skin tone to assess the broader applicability of inclusive design strategies.

Experimental design

Users on the platform are randomly assigned to receive either status quo recommendations or recommendations that are more balanced in terms of the skin tone of people depicted in the content. Users are allocated to one of the conditions for 28 days, and they remain in the same condition throughout the experiment.

Design details

Randomization method
Platform's internal computerized experimentation software. The software randomizes individual users to different experimental conditions, and detailed logging ensures that individuals remain in the same experimental condition throughout the experiment.
Randomization unit
Individual users of the platform
Clusters
Not clustered. 640,000 individual users.
Observations
640,000 individual users.
Arms
Approx. 320,000 users in treatment (50%) Approx. 320,000 users in control (50%)
Primary outcomes
Engagement on the platform, as measured by content items saved by each user in the experiment

Cited by

No citation data available for this registration yet.

Keywords