Optimization Incentives, Learning, and Strategy Selection in Indefinitely Repeated Prisoner's Dilemmas

How to cite

Andres, Maximilian; Bao, Yongping; Breitmoser, Yves; Fehrler, Sebastian; Nithammer, Juri Moritz. 2026. "Optimization Incentives, Learning, and Strategy Selection in Indefinitely Repeated Prisoner's Dilemmas." AEA RCT Registry. AEARCTR-0018810. https://doi.org/10.1257/rct.18810-1.0

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

Maximilian Andres* , Yongping Bao , Yves Breitmoser , Sebastian Fehrler , Juri Moritz Nithammer

* Primary investigator

Registry ID
AEARCTR-0018810
DOI
https://doi.org/10.1257/rct.18810-1.0
First registered
Last updated
Status
In development
Trial dates
2026-06-15 – 2026-07-31
Countries
Private
JEL codes
C72, C73, D81, D83
Registry record
http://www.socialscienceregistry.org/trials/18810

Abstract

This pre-analysis plan describes an experimental design on cooperation in indefinitely repeated Prisoner's Dilemmas. The design varies optimization incentives while holding fixed the normalized strategic environment and the payoff kernel within treatment cores, and the normalized strategic environment. Its objective is to separate two central margins that are often confounded in repeated-game data: the normalized strategic environment and the cardinal payoff cost of choosing a suboptimal strategy.

Intervention

This is a laboratory experiment on cooperation in indefinitely repeated Prisoner's Dilemmas. The design employs a 2×2 between-subject factorial structure that independently varies (i) the cardinal optimization-incentive level and (ii) the normalized strategic environment. The optimization-incentive level is either High (Treatment-Condition) or Low (Control-Condition), and the normalized strategic environment is either risk-dominant or not risk-dominant.

Experimental design

Each participant plays multiple infinitely repeated supergames.

Within each supergame, participants choose between two actions in each round. After each round, the supergame continues with a probability implemented as a random draw at the end of each round. If the draw terminates the supergame, a new supergame begins.

After each round, subjects observe the outcome of that round: their own action, their co-player's action, and both payoffs.

Design details

Randomization method
Random subsets of the subject pool are invited to sessions via e-mail and, within each session, participants are randomly assigned to computer terminals---and thus to matching groups---through public shuffling of a deck of cards with numbers. Treatments for each session are initially assigned through two coin flips (one for each treatment dimension) but for later sessions, we will need to assign the treatments where we still need to collect observations (see below).
Randomization unit
Individuals in matching groups.
Clusters
32 matching groups (8 per treatment).
Observations
>= 256 participants (>= 8 participants per matching group).
Arms
32 matching groups (8 per treatment).
Primary outcomes
Round-1 Cooperation.
Secondary outcomes
Strategy Choices, Continuation-play cooperation, and convergence speed.

Cited by

No citation data available for this registration yet.

Keywords