NOTE-1319: A comparison of minimal group induction procedures

Contribution

A methods paper. Forty years of minimal-group research had used induction procedures that “have never been formally evaluated” (p. 81). It compares four of them on the same implicit and explicit measures. It also imports a procedure from implicit-partisanship research in which participants are never put in a group at all. The authors recommend that procedure, and recommend against the classic one’s deception.

Key insight

Explicit favouritism is easy to produce and hard to move. Telling people they prefer one painter, telling them to imagine being assigned, assigning them at random, and having them memorize five names all gave about the same explicit in-group preference. The implicit measures separated the procedures. Studying a group’s names, with no claim that one belongs to it, tied that group to “I”, “me” and “mine” more strongly than being told one belonged to it. The authors’ explanation is that “the focused study of individuating information about one group automatically creates and strengthens links between participants’ self-concepts and the studied group” (p. 95).

Assumptions

• The IAT measures association. The identification IAT pairs the groups with self words (I, me, mine, my, self) and other words (other, their, theirs, them, they); the attitude IAT with pleasant and unpleasant words. Scored with the D measure (p. 86).
• Bonus money to two others is a behavioural measure. Participants rated seven splits of left-over bonus money between two other people in the session, one from each group. They could not pay themselves, but were told they “could receive money from both ingroup and outgroup members” (p. 90).
• “Ingroup” means the memorized, imagined or assigned group (note 1). In the memorization condition it is a group the participant was never told they belonged to.

Key results

• Experiment 1 (pp. 84–88). 44 students at a small public college in the eastern US; 42 analysed (13 memorization, 15 imagination, 14 classic). All conditions heard a cover story about students divided by art preference.
  • Attitude IAT: overall above zero, F(1, 36) = 11.47, p = .002; varied by condition, F(2, 36) = 7.81, p = .002, ηp² = .30. Memorization beat imagination (d = .96) and classic (d = 1.60). Only memorization differed from zero on its own (note 2).
  • Identification IAT: overall F(1, 36) = 13.43, p = .001; by condition F(2, 36) = 13.69, ηp² = .43. Memorization beat imagination (d = 1.55) and classic (d = 2.16). Again only memorization differed from zero (note 3).
  • Explicit attitude: in-group favoured, Mz = .55, ηp² = .30; no condition effect, F(2, 36) = .08, p = .93.
  • Explicit identification: Mz = .65, ηp² = .43; no condition effect, F(2, 36) = .19, p = .83.
  • Correlations: implicit–explicit r = .16 (attitude) and .34 (identification). Attitude–identification r = .66 on the IATs and .71 on explicit measures (p. 88).
• Experiment 2 (pp. 89–93). 74 students, 73 analysed (24 memorization, 24 random, 25 classic). The art cover story was dropped from memorization and random. A manipulation check was answered correctly by 86%.
  • Attitude IAT: by condition F(2, 67) = 13.13, ηp² = .28. Memorization beat random (d = .89) and classic (d = 1.49). Memorization and random differed from zero; classic did not (note 5).
  • Identification IAT: by condition F(2, 67) = 11.50, ηp² = .26. Memorization beat random (d = 1.06) and classic (d = 1.30). Only memorization differed from zero (note 7).
  • Explicit attitude Mz = .66 and identification Mz = .72, neither varying reliably by condition (p = .10, p = .07).
  • Bonus money (Table 1). On a 1–7 scale, splits favouring the in-group were endorsed above the midpoint and those favouring the out-group below it. Maximum relative in-group advantage: classic 4.24, random 5.13, memorization 5.29. Equal split (“min dif”) did not differ from the midpoint, p = .11.
  • Correlations (Table 2). The in-group-preference factor of the allocations correlated with explicit identification r = .37 (p < .01), explicit attitude .27, implicit attitude .24 and implicit identification .23 (each p < .05). Nothing correlated reliably with the out-group-preference factor.
  • Group colour (note 8). Participants assigned to Red gave weaker in-group-favouring ratings on the “own” options. Taken separately, Red-assigned participants did not favour the in-group significantly on maximum relative own, maximum own or maximum joint own. They did reject the out-group-favouring options.
• Recommendations (pp. 94–95).
  • The classic procedure is fine for explicit measures, but “a prudent recommendation” is to use another if implicit favouritism is the interest.
  • Random assignment “does not strongly instill strong identity bonds” on implicit measures.
  • Memorization was “the most broadly successful”.
  • Implicit and explicit favouritism show “little correspondence”.
  • The classic procedure’s deception “should” be discouraged.

Claims

id	claim	strength	support
C1	Classic, imagined, random and name-memorization inductions produce comparable explicit in-group liking and identification	moderate: two experiments, consistent; small cells (13–25), one college	Exps 1–2
C2	All three inductions tested in Experiment 2 produce comparable in-group favouritism in allocating money between two other people	weak to moderate: one experiment; the effect is carried mostly by one colour group	Exp 2, Table 1, note 8
C3	Memorizing one group’s names, with no categorization, produces much larger implicit preference for and self-association with that group than the classic procedure	moderate: replicated across both experiments, large effects, small samples	Exps 1–2
C4	The classic procedure produces no reliable implicit in-group favouritism	weak: null within-condition tests on 13–25 people; the authors note Ashburn-Nardo et al. 2001 found the opposite	notes 2, 3, 5, 7
C5	Implicit and explicit minimal-group favouritism are only weakly related	moderate: both experiments, but with small N	pp. 88, 92–93
C6	Allocation favouritism goes with explicit identification with the minimal group	weak to moderate: correlational, r = .37 pooled, varying by procedure (.10 to .50)	Table 2

Concepts

• implicit partisanship: preference for, and self-association with, a group created by studying its members’ names (Greenwald et al. 2002).
• identification IAT: the speed of pairing a group with self words rather than other words. An implicit measure of the self–group link.
• multiple alternative matrices (MAMs): seven allocation options that separate maximum relative in-group advantage from maximum in-group gain, maximum joint gain and equality (Bornstein et al. 1983).
• minimal group induction: any procedure that creates novel group membership with no history, interaction or stake.

Connections

• Tajfel et al. 1971 (LIT-1697), Unavailable. The paper describes the classic procedure only generically (dots or paintings, false feedback, allocation to others) and states no result of the original.
• Robertson et al. (NOTE-1245). Robertson’s preregistered replication (n = 590) found minimal-group favouritism in moral judgment only as it rose with identification; the main contrast was p = .07. Pinter and Greenwald point the same way on a different outcome. Allocation favouritism correlated with explicit identification (r = .37), and the overall effect rested largely on one colour group. Two differences matter. Here the mean participant identified with the in-group in every condition. And this was a lab study of a few dozen people per cell, with identification measured before the allocations. Neither study shows that favouritism follows from identification; both show it varies with it.
• Aron et al. (LIT-1706) and the fusion item (NOTE-1256) measure the self–group link by drawn overlap. The identification IAT measures it by association with self words. The record holds no study comparing them.
• Otten (2016) summarizes this paper accurately (see the first draft). She gives the authors’ reading of memorization as a self-concept link, and calls it the same factor the self-anchoring work names.
• Self-categorization theory (THEORY-1114). The memorization effect does not come from categorization: nobody was put in a group. If it is a self–group link, it was made by familiarity, not by metacontrast or fit.

Bearing on the record

• The owner’s thesis (CLAIM-491). These are inferences; the paper does not discuss selflessness.

  • The self–group link can be made without being told one belongs. Forty-five seconds of studying five names tied a group to “me” and “mine” more strongly than being told one shared its taste in painting. If that is an extension of the self, the extension is cheap and need not be chosen. It may also be shallow: it did not predict allocation better than the explicit measures did.
  • Favouritism to anonymous others with no gain for oneself appeared under every induction. That is the behaviour the thesis explains as serving an extended self. Here the participant was also told they could receive money from both groups, so a reciprocity motive was not excluded.

• The triviality objection (CLAIM-492). The paper measures the self’s extent and liking separately, which is what the objection asks for. The identification IAT (group with “self”) and the attitude IAT (group with “pleasant”) correlated .66 in Experiment 1 and .48 in Experiment 2. They are related but not the same. The explicit measures, “I identify with” and “I like”, correlated .71 in both experiments. So at the level of report, the self’s extent and caring nearly coincide. At the implicit level they come apart somewhat. Neither predicted allocation much better than the other.

• Batson’s objection (CLAIM-494). Not addressed.

• A dated note should be added to CLAIM-492. Proposed text:

  Note of 2026-10-29: an implicit measure of the self’s extent. Pinter and Greenwald (2011) measured self–group association (an IAT pairing the group with “I”, “me”, “mine”) apart from liking (an IAT pairing it with pleasant words). The two correlated .48 to .66, and the explicit versions .71, so self-extension and caring overlap heavily but are separable implicitly. Memorizing a group’s names, with no membership, made the self-association largest. Allocation favouritism correlated with each measure at about .23 to .37, none clearly ahead.

• No THEORY filed. A methods comparison with small samples.

• No ML instruction.

Limitations

• Small cells: 13 to 25 per condition, at one small public college, for extra credit.
• Comparable is not equal. “No condition effect” on explicit measures rests on F tests with two dozen people per cell. Experiment 2’s explicit identification was close (p = .07).
• The colour effect. In Experiment 2 the allocation result depends on which colour participants were assigned (note 8). The authors give two reasons for confidence, that all 14 effects were in the predicted direction and nine were reliable, but do not explain it.
• The demand argument covers one procedure. The authors say strong implicit effects show minimal-group favouritism is not an artifact of demand (p. 95). That holds for memorization. The classic procedure showed explicit favouritism with no implicit effect, which leaves demand open for it.
• A claim the tables do not show. The conclusion says memorization produced “stronger implicit–explicit correlations” (p. 94). In Experiment 1 its identification correlation (.50) was below the classic’s (.83). In Table 2 its implicit–explicit correlations are −.06 to .13.
• Rounded p-values. Many are printed only as powers of ten (“p = 10-5”), so exact values cannot be checked.
• Typos and slips.
  • “Rabbie & Horwitz, 1969” in the text (p. 81) and “Horowitz” in the references (p. 98).
  • “Wittenbrink & Schwartz, 2009” in the text (p. 82); “Schwarz” in the references.
  • “This change should affect a more rigorous comparison” for “effect” (p. 89).
  • Note 6 gives the main effect of group, a two-level factor, as “F(2, 67)” (p. 95).
  • The Crisp & Turner title reads “Can imagined interactions procedure positive perceptions?” for “produce” (p. 97).
  • The Hofmann et al. 2005 author list is garbled (p. 97).
  • “Pennylvania” in the biographical note (p. 98).

Open questions

• Does the memorization link predict anything costly? The allocation here was other people’s bonus money. A test where favouring the studied group costs the participant would show whether a cheap self–group link moves behaviour.
• Does Robertson’s identification moderation hold for allocation in a preregistered sample, with identification measured before the outcome?