Tocqueville’s "Free School": How Adjudicative Engagement Reshapes User Behaviors in Online Marketplaces?

Among eligible users of an online secondhand marketplace who have never served as community adjudicators, we randomly select a treatment group to receive an invitation to participate in the platform's peer dispute-resolution feature. The invitation is delivered through the platform's normal notification system. Users in the treatment group are free to accept or decline; the remaining eligible users form the control group and receive no such invitation during the study period. The intervention therefore assigns the opportunity and encouragement to serve as an adjudicator.

2026-08-10

2026-08-31

Transaction scale. Number of purchase orders placed; number of sales orders completed; average value per purchase; average value per sale; and the share of orders that are initiated but not finalized.
Trading conduct as a seller. The service terms committed to when listing (return terms such as no reason returns, and shipping terms such as free shipping), and, when facing a buyer's refund request, the probability of granting a return and refund or a refund only claim.
Refund behavior as a buyer. The incidence of return and refund requests and of refund only (no return) requests initiated by the user.
Formal dispute behavior. Whether the user brings a dispute into the platform's peer adjudication process as a party.

Engagement intensity. Active login time and the volume of private messages exchanged with counterparties.
Realized dispute resolution. The completion of return and refund cases and of refund only cases, distinguishing claims that are merely initiated from those actually carried through.
Pro social behavior. Whether the user donates a share of their sales proceeds to charitable causes through the platform.

We conduct an individually randomized controlled trial on the peer dispute resolution feature of a large online secondhand marketplace, where ordinary users can be invited to help decide disputes between buyers and sellers. From a pool of recently active eligible users who have never previously served as adjudicators, we randomly assign users to a treatment group and a control group at a 1:1 ratio. Treatment users receive an invitation, delivered through the platform's standard notification system, encouraging them to serve as an adjudicator. Control users receive no such invitation during the study period. Take up is voluntary, so the invitation functions as an encouragement rather than compelled participation. After the intervention, we track and compare the two groups' subsequent behavior on the platform, focusing on their conduct as buyers and sellers, their behavior in disputes, and their broader engagement with the platform community. The design allows us to estimate the causal effect of the invitation to serve, and, using the random invitation as an instrument for actual participation, the causal effect of adjudication itself on subsequent economic behavior.

Not available

Randomization was performed by computer. Eligible users are stored as records in the platform's database, and treatment assignment is generated in SQL using its built in random number function, which assigns each eligible user a pseudo random value that is then used to sort users into the treatment and control groups.

The individual user. Treatment is assigned at the level of the individual eligible user, with no clustering or higher level (group) randomization.

No

Randomization is at the individual level, so each user is its own cluster. The treatment group consists of approximately 900,000 users. A control group of comparable size will be drawn by random sampling from the remaining eligible users at approximately a 1:1 ratio, for a planned total on the order of 1,800,000 users.

Randomization is at the individual level, so each user is its own cluster. The treatment group consists of approximately 900,000 users. A control group of comparable size will be drawn by random sampling from the remaining eligible users at approximately a 1:1 ratio, for a planned total on the order of 1,800,000 users.

Treatment (invited to serve as adjudicator): approximately 900,000 users. Control (not invited): approximately 900,000 users, drawn by random sampling from the remaining eligible users at a 1:1 ratio.