To the group page

Results: Cognitive Science X2P

X2P20262 · Cognitive Science X2P

22 started, 18 completed the whole exercise. Small groups give uncertain estimates: confidence intervals are wide and p-values unstable. That is part of the lesson.

Shepard's tables (visual illusion)

Median = 31 percent, M = 35.6, SD = 27.4, N = 20 · Correct answer: 0 percent

020406080Correct answer: 0Everyone (n = 20)
▍ median
What does this show?

The two table tops are exactly the same size and shape — one is merely rotated. The correct answer is 0%. The visual system reads the picture as three-dimensional and "corrects" for depth, which makes the receding table look longer. The illusion persists even when you know the truth.

Source: Shepard, R. N. (1990). Mind Sights. W. H. Freeman.

Kilimanjaro (incidental anchor)

GroupNMedianMSD
2,000 climbers95,6005,0001,077
9,000 climbers95,8006,0221,337
  • Correct answer: 5,895 metres
  • Difference in means: 1,022 metres, 95% CI [-195, 2,240] · Welch's t-test: t(15.3) = 1.79, p = 0.094, d = 0.84
  • Mann–Whitney (robust to extreme guesses): U = 25.5, p = 0.196
02,0004,0006,0008,000Correct answer: 5,8952,000 climbers (n = 9)9,000 climbers (n = 9)
▍ median
What does this show?

The number of climbers per year says nothing about the mountain's height, yet a larger number in the question still pulls estimates upward. Kilimanjaro is 5,895 metres tall. Compare the groups that saw 2,000 and 9,000.

Source: Tversky, A., & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185, 1124–1131.

Randomisation and drop-out

Assigned → answered, per version. Versions were allocated by permuted-block randomisation so the groups end up equal in size; everyone is analysed in the version they were allocated to.

  • 2,000 climbers: 10 → 9
  • 9,000 climbers: 10 → 9

Gandhi's age (anchoring)

GroupNMedianMSD
Anchor 43 years96769.410.4
Anchor 96 years99084.611.8
  • Correct answer: 78 years
  • Difference in means: 15.1 years, 95% CI [4, 26.2] · Welch's t-test: t(15.7) = 2.88, p = 0.011, d = 1.36
  • Mann–Whitney (robust to extreme guesses): U = 14.5, p = 0.024
  • Anchoring index: 0.43
020406080Correct answer: 78Anchor 43 years (n = 9)Anchor 96 years (n = 9)
▍ median

who answered "older":

Anchor 43 years100 % (8/8)
Anchor 96 years0 % (0/9)

Difference in share: -100 percentage points, 95% CI [-100, -56] · Fisher's exact test: p < 0.001

What does this show?

Everyone first judged an age — 43 or 96 — and then gave their own estimate. The comparison question acts as an anchor from which the estimate is insufficiently adjusted. Gandhi died at 78. The anchoring index is the share of the gap between the anchors that shows up in the estimates: 0 = no effect, 1 = estimates follow the anchor completely.

Source: Strack, F., & Mussweiler, T. (1997). Explaining the enigmatic anchoring effect. JPSP, 73, 437–446. Index: Jacowitz, K. E., & Kahneman, D. (1995). PSPB, 21, 1161–1166.

Randomisation and drop-out

Assigned → answered, per version. Versions were allocated by permuted-block randomisation so the groups end up equal in size; everyone is analysed in the version they were allocated to.

  • Anchor 43 years: 9 → 9
  • Anchor 96 years: 9 → 9

Better than average

Median = 34.5, M = 26.6, SD = 28.5, N = 14 · Reference: 0

Share above 0: 79 % · t-test against 0: t(13) = 3.49, p = 0.004, 95% CI [10.1, 43]

-100-50050100Reference: 0Everyone (n = 14)
▍ median
What does this show?

If everyone judged their company accurately the mean would land on 0, since not everyone can be better than average. A mean clearly above 0 is the better-than-average effect.

Source: Alicke, M. D., & Govorun, O. (2005). The better-than-average effect. In The Self in Social Judgment (pp. 85–106).

Nuclear power: risks and benefits (affect heuristic)

  • Risks × Benefits, N = 18
  • Spearman's rank correlation: ρ = -0.06, p = 0.813
  • Pearson's r (for comparison): r = -0.06, 95% CI [-0.51, 0.42], p = 0.815
1143123111012345012345Risks →Benefits →
▬ least-squares regression line, slope -0.05 per scale step · the numbers in the cells are counts of people
What does this show?

In the world, risk and benefit tend to go together — what brings large benefits often carries large risks. In people's judgments the association is instead negative: those who dislike nuclear power see high risks and low benefits, and vice versa. Feeling drives both judgments. The order of the two questions was randomised.

Source: Finucane, M. L., Alhakami, A., Slovic, P., & Johnson, S. M. (2000). The affect heuristic in judgments of risks and benefits. JBDM, 13, 1–17.

Randomisation and drop-out

Assigned → answered, per version. Versions were allocated by permuted-block randomisation so the groups end up equal in size; everyone is analysed in the version they were allocated to.

  • Risks first: 9 → 9
  • Benefits first: 9 → 9

The Eiffel Tower (wisdom of crowds)

Median = 300 metres, M = 257, SD = 83.5, N = 18 · Correct answer: 330 metres

0100200300400Correct answer: 330Everyone (n = 18)
▍ median
What does this show?

No anchor was given here. Individual guesses scatter widely, but the group's median often lands surprisingly close to the truth: the Eiffel Tower is 330 metres tall.

Source: Galton, F. (1907). Vox populi. Nature, 75, 450–451.

The disease problem (framing)

who chose the sure program (A):

Gain frame ("saved")70 % (7/10)
Loss frame ("die")0 % (0/8)

Difference in share: -70 percentage points, 95% CI [-89, -26] · Fisher's exact test: p = 0.004

What does this show?

Programs A and B have exactly the same expected outcome in both versions: 200 of 600 are saved. Described as gains ("saved") most people choose the sure option; described as losses ("die") most choose the gamble. Same choice, different frames.

Source: Tversky, A., & Kahneman, D. (1981). The framing of decisions and the psychology of choice. Science, 211, 453–458.

Randomisation and drop-out

Assigned → answered, per version. Versions were allocated by permuted-block randomisation so the groups end up equal in size; everyone is analysed in the version they were allocated to.

  • Gain frame ("saved"): 10 → 10
  • Loss frame ("die"): 8 → 8