An experiment on algorithmic mediation

How to cite

Dinc, Sukran; Ozyurt, Selcuk; Kesten, Onur. 2026. "An experiment on algorithmic mediation." AEA RCT Registry. AEARCTR-0017830. https://doi.org/10.1257/rct.17830-1.0

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

Sukran Dinc* , Selcuk Ozyurt , Onur Kesten

* Primary investigator

Registry ID
AEARCTR-0017830
DOI
https://doi.org/10.1257/rct.17830-1.0
First registered
Last updated
Status
In development
Trial dates
2026-02-15 – 2026-03-31
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/17830

Abstract

Mediation is widely used because it is faster, cheaper, and more flexible than courts. Yet a growing body of evidence suggests that the informality of mediation can disproportionately disadvantage less powerful parties, particularly those marginalized by gender, race, or ethnicity. In response, demand for more structured and rigorous mediation practices, combined with recent technological advances, has spurred the growth of Online Dispute Resolution (ODR) systems. In a standard ODR system, mediator is an algorithm defined by predetermined rules. Although these protocols are designed to deliver impartial outcomes, we know little about how gender visibility shapes behavior and outcomes in algorithmic mediation. We study this question in a controlled experiment using the algorithmic mediation mechanism proposed by \citet{kesten2025strategy}. We measure negotiation behavior, including participants’ demands, acceptance decisions, and beliefs about others’ behavior, and we collect detailed perceptions of fairness and the salience of gender in decision-making. We manipulate gender visibility across two experimental conditions to examine its effects on evaluation outcomes.

Intervention

The study investigates (i) individuals' negotiation behaviors and beliefs under the algorithmic mediation protocol, and (ii) whether these behaviors, beliefs, and preferences differ across gender visibility conditions. In the gender treatment, negotiators observe gender-revealing icons; in the baseline treatment, only neutral icons are used. Full details are provided in the pre-analysis plan.

Experimental design

This is a three-part incentivized online study on decision-making and negotiation with computer assistance. Participants complete a series of short bargaining tasks for points that affect payment, followed by additional tasks measuring social preferences and beliefs about others’ decisions. In some sessions, participants see simple profile cues about their counterpart, while in others these cues are not shown. Full procedures and analysis plans are documented in the pre analysis plan.

Design details

Randomization method
Randomization is conducted by computer using Qualtrics' built-in randomization logic, based on pre-assigned quotas and random number generators.
Randomization unit
Group-level randomization for gender visibility treatment; individual-level randomization for backup options and matching order.
Clusters
240 participants
Observations
240 participants. Each completes 10 rounds (2,400 participant-round records). Primary outcomes are analyzed at the participant level (averaged across rounds).
Arms
120 participants in gender-visible treatment (balanced: 60 women, 60 men). 120 participants in gender-blind control (balanced: 60 women, 60 men). (Total: 240 participants.)
Primary outcomes
Participants' demand above their backup option (sincerity). Successful resolution of negotiation (compatible demands and mutual acceptance). Fairness of outcomes across matched pairs with swapped backup options. Total welfare (sum of realized payoffs). See the pre-analysis plan for operationalizations and measurement procedures.

Cited by

No citation data available for this registration yet.

Keywords