Seen One, Seen Them All

How to cite

Bechdolf, Mathilde Lea Editha; Sadrieh, Abdolkarim; Grossmann, Max. 2026. "Seen One, Seen Them All." AEA RCT Registry. AEARCTR-0018621. https://doi.org/10.1257/rct.18621-1.2

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

Mathilde Lea Editha Bechdolf* , Abdolkarim Sadrieh , Max Grossmann

* Primary investigator

Registry ID
AEARCTR-0018621
DOI
https://doi.org/10.1257/rct.18621-1.2
First registered
Last updated
Status
In development
Trial dates
2026-06-15 – 2027-06-15
Countries
Private
JEL codes
C72, C92, D83, D91
Registry record
http://www.socialscienceregistry.org/trials/18621

Abstract

When interacting with others, individuals are uncertain about whether they will encounter pro-social or anti-social behavior. In such interactions, even a single observation can trigger biased beliefs about entire groups. We conduct a laboratory experiment in which participants simultaneously receive statistically irrelevant information about their in-group and out-group, with treatments varying whether both signals are pro-social or anti-social. Since the informational structure is identical across groups by design, any subsequent difference in cooperative behavior toward in-group versus out-group members must reflect asymmetric biased beliefs. We hypothesize that individuals overweigh signals that confirm existing group stereotypes, generalizing negative information about out-group members to the out-group as a whole, while dismissing equivalent negative information about in-group members as an isolated case. Group identities are induced via the minimal group paradigm. Participants play three cooperation games, a Prisoner's Dilemma, a Stag Hunt, and a Mini Trust Game, separately with in-group and out-group members before and after receiving the anecdotal information. Understanding how statistically irrelevant information generates systematic discrimination is relevant for policy makers seeking to counteract stereotype formation, intergroup hostility, and political polarization.

Intervention

Participants take part in four Game Rounds in which they take several decisions in three 2-player games: a Prisoner's Dilemma, a Stag Hunt, and a Mini Trust Game. After Game Round 1, group identities are induced using the minimal group paradigm: participants are assigned to either the Klee group or the Kandinsky group based on their painting preferences. From Game Round 3 onwards, participants know whether their decisions are matched with a member of their own group (in-group) or the other group (out-group).
Participants are randomly assigned to one of three treatments. In the Good and Bad treatments, participants receive anecdotal information about one decision made by one Klee group member and one Kandinsky group member in Game Round 2. The information indicates either cooperative behavior (Good) or uncooperative behavior (Bad) in the Prisoner's dilemma. The Baseline group receives no information.

Experimental design

Participants take part in four Game Rounds in which they take several decisions in three 2-player games: a Prisoner's Dilemma, a Stag Hunt, and a Mini Trust Game. In each game, participants choose between Action A and Action B; Action B is the non-prosocial option in all three games. The three games are played in random order within each Game Round. In each Game Round, participants make 20 decisions per game, resulting in 60 decisions per Game Round. Before each Game Round, participants' expectations about the share of prosocial decisions are elicited.
In Game Round 1, participants take 60 decisions matched with decisions of other participants in the session of Game Round 1.
After Game Round 1, social identity is induced using the minimal group paradigm.
In Game Round 2, participants know their own social identity but opponent identity is not shown. Participants take 60 decisions matched with decisions of other participants in the session of Game Round 2.
In Game Round 3, opponent identity is revealed. For each game, participants make 10 decisions matched with in-group member decisions and 10 decisions matched with out-group member decisions of Game Round 2.
Between Game Round 3 and Game Round 4, participants are randomly assigned to one of three treatments, subject to a no-deception constraint: the Good and Bad treatments are only available when the corresponding statements are true based on Game Round 2 data.
Baseline: no information shown
Good: participants are informed that one Klee group member and one Kandinsky group member chose the prosocial option in the Prisoner's Dilemma in Game Round 2
Bad: participants are informed that one Klee group member and one Kandinsky group member chose the non-prosocial option in the Prisoner's Dilemma in Game Round 2
In Game Round 4, participants again make 10 decisions matched with in-group member decisions and 10 decisions matched with out-group member decisions of Game Round 2.

Design details

Randomization method
Participants are randomly assigned to treatments by the experimental software. In Sessions 1–9, treatment assignment was performed by independent random assignment (each participant independently and uniformly assigned to one of the available treatments). Starting from Session 10, we use rerandomization to improve covariate balance across treatment arms. Specifically, we apply a simple multi-arm extension of the Euclidean rerandomization scheme proposed by Schindl and Branson (2025) and recommended by List (2026, p. 191), using participants’ prosocial choice rates (share of Action A) from Game Rounds 1–3 as balancing covariates. Treatment assignments are drawn until a Euclidean balance criterion on standardized covariates falls below a threshold calibrated to an acceptance probability of 0.1%. The new algorithm also attempts to place participants into treatments in approximately equal numbers. Should the Euclidean balancing algorithm fail for any reason, the code automatically falls back to a Completely Randomized Experiment (List, 2026, section 6.3.2).
Randomization unit
Individual participant.
Clusters
750 participants
Observations
750 participants
Arms
250 participants Baseline, 250 participants Good, 250 participants Bad
Primary outcomes
The in-group cooperation advantage and mean cooperation rate, both measured in Game Rounds 3 and 4.
Secondary outcomes
Cooperation rate in Game Round 1 and Game Round 2. General preferences (risk willingness, reciprocity, trust), attitudes on social and economic topics, fairness and inequality preferences, personality traits (Mini-IPIP Big Five), and demographic characteristics (age, gender, field of study, country of higher education entrance qualification).

Cited by

No citation data available for this registration yet.

Keywords