Contribution
A large preregistered replication of a well-known small study. It finds the original effect is not there, and that intergroup bias in moral judgment takes different forms in minimal and partisan groups. For the record, it is the only preregistered evidence held on whether minimal groups produce favouritism at all. The answer is: only among those who identify with the minimal group.
Key insight
A minimal group is not a group for everyone assigned to it. Labelled “Overestimators” after a dot task and a chat, participants identified with their in-group more than the out-group (d = .64), but not uniformly. Bias followed identification. Politics, a group people care about (d = 1.30), produced derogation of the other side rather than favouritism toward one’s own.
Assumptions
- The original paradigm (Valdesolo & DeSteno 2007), moved online. Participants self-assign an easy (“green”) or hard (“red”) task, leaving the other to a future participant, or use a randomizer. Others judge the fairness of an assigner who takes the easy task.
- Pay was equalized across tasks, so the easy task stayed preferable (p. 12).
- Intent-to-treat was preregistered. The authors made a post hoc switch to excluding fair and altruistic choosers, as the original did, because their inclusion confounds the Self condition (p. 21).
Key results
- Samples. Experiment 1: minimal groups, 610 recruited, 590 analysed, nationally representative. Experiment 2: political groups, 606 recruited, 577 analysed, half Democrats and half Republicans. Power simulations assumed d = .4 (pp. 9–10, 21, 25).
- H1, the original contrast (self and in-group versus other and out-group): Experiment 1 F(1, 536) = 0.15, p = .70, with the 95% CI below the equivalence bound η² = .01, so it is “statistically equivalent to zero”. Experiment 2 F(1, 528) = 1.86, p = .17.
- H2, self versus others: not significant once fair choosers are excluded (Experiment 1 p = .26; Experiment 2 p = .23). It is significant only when they are included, which the authors call uninterpretable.
- H3, in-group versus out-group:
- Minimal groups: F(1, 536) = 3.22, p = .07.
- Political groups: F(1, 528) = 9.04, p = .003, η² = .02, driven by out-group derogation.
- Pooled: F(1, 581) = 11.69, p < .001.
- H6, moderation by identification:
- Minimal groups: the interaction is b = −.39, p < .001. In-group fairness rose with identification (b = .30) and out-group fairness did not change.
- Political groups: no moderation, perhaps because identification was uniformly high (M 2.95 against 1.04).
- H5, stronger in natural than minimal groups: not supported (p = .44).
- Choosers. In the Self condition, 47 (Experiment 1) and 40 (Experiment 2) used the randomizer, and 4 and 6 took the hard task themselves. In the original, 17 of 19 took the easy task.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Valdesolo and DeSteno’s intergroup moral hypocrisy does not replicate in minimal groups | strong: preregistered, high power, equivalence test | Experiment 1, H1 |
| C2 | In minimal groups, in-group favouritism in moral judgment appears among those who identify with the minimal group, and not across the board | moderate: a significant preregistered moderation; the main contrast is p = .07 | Experiment 1, H3 and H6 |
| C3 | Partisans judge an out-party member’s identical selfish act as less fair than an in-party member’s | moderate: one preregistered experiment, a small effect | Experiment 2, H3 |
| C4 | Intergroup bias is favouritism in low-stakes groups and derogation in groups in zero-sum conflict | weak: an interpretation offered from two experiments, citing Brewer 1999 | Discussion, Conclusion |
Concepts
- moral hypocrisy: judging one’s own transgression, or an in-group member’s, as fairer than the same act by others.
- collective identification: here, the difference between in-group and out-group scores on three items (value, pride, importance).
- intent-to-treat: analysing participants as randomized, whatever they chose.
Connections
- Tajfel et al. 1971 (LIT-1697), unread. This is not a replication of that experiment; it uses the same overestimator induction. The finding that favouritism tracks identification is consistent with Hornsey’s report (NOTE-1253) that SCT effects appear under high salience or identification.
- Rubin & Hewstone (LIT-1780), unread, and the self-esteem debate: not addressed here.
- The owner’s thesis (CLAIM-491). About a third of choosers, with pay equalized and no one watching, used a randomizer or took the hard task. That is fair or altruistic choice in a minimal setting where no group was at stake in the choice. It is not evidence about motive, but it is a base rate a selfishness thesis must explain.
Bearing on the record
- The minimal-group effect. It is the record’s only preregistered evidence on it, and it qualifies the 1971 claim that mere categorization suffices: here it sufficed only for those who identified.
- THEORY-1114. It fits the account’s emphasis on identification and salience.
- No THEORY filed. It is a replication of a moral-hypocrisy effect, outside the record’s current questions.
- No ML instruction.
Limitations
- The post hoc switch from intent-to-treat to exclusion was made openly and matched the original.
- Online setting and chat-room induction; Prolific workers only.
- Supplementary materials were not read, including the high-identifier subgroup analysis cited in the discussion.
- A typo in the H5 comparison names both groups “Experiment 2” (p. 28).
Open questions
- Does a minimal-group induction that reliably produces identification produce the original moral hypocrisy? The paper suggests the induction failed for many participants. A manipulation check that selects identifiers in advance would test it.