Underestimating our influence over others' unethical behavior and decisions

Personality and Social Psychology Bulletin 40(3): 348–362, doi 10.1177/0146167213511825. Read as the authors' accepted manuscript deposited at Cornell eCommons (hdl.handle.net/1813/75150; "Required Publisher Statement: © SAGE"); the published version was not read. Read by researcher R5a, 5 October 2026. Provenance: papers/change/bohns2014_unethical_compliance.provenance.json.

What was read

• Read in full: all 1,922 lines of pdftotext -layout output of the 45-page manuscript. That covers the abstract, introduction, Studies 1–4, general discussion, footnotes 1–4, references, Table 1 (strangers' replies), the captions of Figures 1–4, and Appendices A and B (every scenario).
• Not read: the figures as images. Figures 1, 3 and 4 are charts whose values are in the text; Figure 2 is a photograph.

What they did

• Study 1 (a white lie). 52 student "instigators" (7 quit after reading the task) asked strangers on campus, face to face, to sign a form falsely stating that the instigator had told them about a new course. The script: "…but I really don't want to do it." 198 strangers were approached. Before starting, instigators predicted how many people they would need to ask to get 3 signatures.
• Study 2 (vandalism). 25 instigators (2 quit) asked 108 strangers to write "pickle" in pen in what looked like a library book ("I'm trying to play a prank on someone, but they know my handwriting").
• Study 3 (MTurk, N = 171). Vignettes: buying beer for minors, giving a friend Valium, an illegal download, reading a colleague's private messages.
  • Design: 2 perspectives (actor, instigator) × 2 (an unethical suggestion, or none).
  • Measure: how comfortable the actor would feel doing the right thing (1–7).
• Study 4 (MTurk, N = 155). Work dilemmas that affect only the actor: calling in sick to see a game, taking office paper home, freelancing for a competitor, expensing personal dinners.
  • Design: 3 perspectives (actor, instigator, neutral observer) × 2 (unethical or ethical advice).
  • Measure: how comfortable the actor would feel doing the wrong thing.

Main results (verified)

• Study 1:
  • predicted 8.47 approaches for 3 signatures (SD 5.14), actual 4.39 (SD 2.37); F(1, 43) = 32.87;
  • 40 of 44 (90.9%) underestimated compliance;
  • those who quit predicted 18.33;
  • predictions correlated with expected comfort in refusing, r = .48.
• Study 2:
  • predicted 10.73 (SD 6.58), actual 4.70 (SD 2.53); F(1, 22) = 16.72;
  • 20 of 23 (87.0%) underestimated;
  • r = .31 (p = .11).
  • Strangers' replies show they saw it as wrong: "So this is like vandalism? Okay, there, you got my autograph"; "I don't write on library books."
• Implied compliance rates (my computation): Study 1 about 3/4.39 = 68% actual against 35% predicted; Study 2 3/4.70 = 64% against 28% (23 × 3 = 69 of 108 strangers complied).
• Study 3. Actors' comfort doing the right thing fell from 5.65 with no suggestion to 4.75 after an unethical suggestion (F(1, 99) = 17.66). Instigators' ratings did not move (4.79 against 4.56). Interaction F = 3.98, p = .05.
• Study 4.
  • Actors' comfort doing the wrong thing: 3.13 after unethical advice against 2.38 after ethical advice (p = .056).
  • Instigators (4.08 against 4.32) and observers (3.55 against 3.72) saw no difference.
  • Overall, actors rated themselves less comfortable with wrongdoing (2.84) than observers (3.65) or instigators (4.20) rated them.
• Cited, not read in the original:
  • Flynn and Lake (Bohns) 2008: help-seekers overestimated the number of people they would have to ask "by as much as 50 percent".
  • Bohns et al. 2011: the underestimation was present in the US and China, but attenuated in China.

Limits

• Minor acts, low stakes. Signing a false form and writing one word in a book, asked by student strangers; refusal cost only awkwardness.
• Studies 3–4 are hypothetical comfort ratings, not behaviour.
• The predictions are about others' compliance. No one predicted their own compliance and then acted.
• Possible biases (the authors). Instigators may have picked compliant-looking targets; acting skill may have mattered.
• Inconsistencies found:
  • Study 3: one ANOVA reports F(1, 167) and F(1, 169) for its two main effects.
  • Study 4: Perspective has three levels but is reported as F(1, 149). The interaction "F(1, 149) = 3.63, p = .03" matches its p only with 2 numerator df (my computation: p ≈ .059 on (1, 149), ≈ .029 on (2, 149)).
  • Study 4's discussion says "we replicated our findings from Study 4"; it means Study 3.
  • Study 2: r(24) with 23 completers. The title page misspells the second author ("Roghanziad").

What it means for Kurisutina (inference)

• About two thirds of people agreed to a stranger's direct request to do a minor dishonest or destructive act; requesters expected about a third. What moved people was the social cost of saying no, not reasons.
• A request to act is not a change of belief. A replica needs an action channel (comply, refuse, negotiate) and a human-sized compliance probability for minor acts. That probability presumably falls with severity; this paper does not test severity.
• Self-reports understate compliance. Actors' self-views (Study 4) and requesters' forecasts both point the same way. Interviews asking "would you do X if asked?" cannot set this setting (P6), and human judges' intuitions about what is realistic will err low.