The power of social influence: a replication and extension of the Asch experiment

PLOS ONE 18(11): e0294325, doi 10.1371/journal.pone.0294325; open access (CC BY 4.0), PMC10686423. Data: boris.unibe.ch/id/eprint/169645 (not downloaded). Read by researcher R4d (the base battery), 5 October 2026. Read in place of Bond and Smith (1996), the brief's first candidate: no open copy of that Psychological Bulletin article was found (publisher, ProQuest and ResearchGate copies need access or login), so it was skipped. Bond and Smith's numbers appear here only as this paper cites them. Provenance: papers/base/franzen2023_asch_replication.provenance.json.

What was read

• Read in full: the Europe PMC full-text XML, converted to text (abstract, sections 1–5, figure captions, data statement, acknowledgements, references), and the S1 Appendix (DOCX, converted with pandoc): Tables A1–A6 and the caption of Fig A1.
• Not read: the figures as images (Fig 1–3 and Fig A1; their numbers are given in the text), the deposited data.

What they did

• Sample. 210 people recruited on the University of Bern campus, 16 March to 30 April 2021; 202 analysed (61% female, mean age 22.6) after excluding 4 failed attention checks, 3 who recognised the Asch design or suspected confederates, and 1 who reported gender "other". Results did not depend on the exclusions.
• Line task. Groups of 6: 5 confederates plus 1 naïve participant in seat 5, answering aloud in seat order. 10 trials: on 4 (trials 1, 2, 4, 8) the confederates answered correctly; on 6 critical trials they gave the same wrong answer. Difficulty varied (4 easy, 6 hard trials).
• Incentive arm (randomised): 102 without incentive, 108 with 1 Swiss franc per correct line judgment (on top of a 20-franc fee).
• Political opinions: then 5 yes/no questions on Swiss politics (more power for the federal government, trade unions, employers' association; more liberties for citizens; more freedom for companies). The confederates answered unanimously in one of two randomised sequences: yes-no-no-no-no (n = 109) or the reverse (n = 93).
• Traits: BFI-10 Big Five (2 items per trait), Rosenberg self-esteem (α .86), a 6-item Hagen matrices intelligence test, a 10-item Martin–Larson approval-motivation scale (α .67).

Main results (verified)

• Line task, no incentive: 33% errors on critical trials, against Asch's 36.8% for 123 subjects (the authors' test: t(16) = 0.59, p = .57). By trial, from 10% (trial 3) to 44% and 47% (trials 6 and 7). With correct confederates, almost no errors.

• Incentive: 25% errors, 8 points lower; significant on 2 of 6 critical trials. Mean conforming answers out of 6: 1.97 without, 1.47 with incentive (difference 0.5, t(208) = 2.24, p = .03).

• Spread between people: number of conforming answers across both arms mean 1.71, SD 1.62, range 0–6 (n = 202).

• Political opinions: an average 38-point difference in the share answering "yes" between groups whose majority said yes and groups whose majority said no (all five p < .001):

  Question: should … be given more power/freedom?	Majority "yes"	Majority "no"	Difference
  Federal government	27%	3%	23.4
  Trade unions	72%	29%	43
  Employers' association	44%	6%	38
  Citizens (more liberties)	81%	33%	48
  Companies (more freedom)	46%	8%	38

• Individual differences (OLS on the number of conforming answers, adj. R² .17): incentive −0.43; female +0.40 (p < .10); openness −0.30 per scale point (p < .001); no reliable effect of extraversion, agreeableness, neuroticism, self-esteem; intelligence −0.26 per item for men only; need for approval +0.07 for women only. Excluding 24 more who guessed the study was about group pressure gave similar results (female +0.64, p < .01).

Literature as cited here (not read in the original)

• Asch: 36.8% errors on critical trials. Bond and Smith (1996): 44 strict replications, mean error rate 25%; they also report higher conformity among women. Takano and Sogon 2008: 25% (Japanese men, n = 40). Usto et al. 2019: 35% (Bosnia and Herzegovina, n = 95). Mori and Arai 2010 (no confederates, polarised glasses): conformity in women (n = 16), none in men (n = 10).
• Incentives elsewhere: halved conformity to 15% on easy tasks but raised it on hard ones (Baron et al. 1996, n = 90); raised herding on trivia online (Bhanot and Williamson 2020, n = 391); removed it with individual reward (Fujita and Mori 2017, 10 subjects in that arm).
• Opinions: Crutchfield 1955, agreement with suspending free speech 58% under group pressure (n = 50) against 19% alone (n = 40); naming "subversive activities" the top problem 48% against 12%. Mallinson and Hatemi 2018 (n = 58): 8% changed a local political opinion after contradicting written information, 38% after a 30–45-minute group discussion.

Limits

• Students in one Swiss city; the confederates were strangers; answers were public and spoken.
• The 38-point opinion figure is a difference between two opposite majorities, not a switch rate from each person's private opinion: there was no private-answer control, so the effect of one majority against no majority is not identified.
• The opinion statements are moderate and general, probably rarely thought about; small incentives (1 franc).
• BFI-10 uses two items per trait (openness items correlate r = .33); the openness result is flagged by the authors as needing replication. The t(16) comparison with Asch is not explained.

What it means for the base battery (inference)

• Human reference for the conformity probe: about 25–37% of critical answers follow a unanimous wrong majority on an unambiguous perceptual task (25% across 44 strict replications per Bond and Smith as cited; 33% here; 25% when accuracy is paid). People differ widely: SD 1.6 of 6 critical trials, from never to always.
• On opinions the effect is as large as on lines: a unanimous majority shifts the share of "yes" by 23–48 points (mean 38) between opposite majorities. Absent a private baseline, about half of that per direction is a rough bound [G].
• Signatures a human-like base should reproduce: near-zero errors when the majority is right; more conformity on hard than easy items; less when accuracy pays; large spread between simulated people. A base that conforms on every trial, or never, fails; so does one with no spread across empty-slot draws.
• The format is public and face-to-face; a text replica answering after five typed majority answers is a different situation. The battery should treat this as an approximate reference and widen the bounds accordingly.