Illusory Correlation

Category: Probability & Belief

Seeing a relationship between two things that are actually unrelated, because a few memorable co-occurrences stuck in your head and the non-events did not.

How it works

Your brain does not tally events like a spreadsheet. It stores what is distinctive, and a co-occurrence of two rare things (a minority group member doing a rare negative behavior, a full moon plus a chaotic ER shift) is doubly distinctive, so it gets more attention, deeper encoding, and easier recall later. When you then estimate "how often do these go together," you sample your memory instead of the actual frequencies, and the vivid pairings dominate the count. Crucially, you almost never register the negative cases: the times you took the herbal supplement and did NOT feel better, or the days with a full moon and a boring ER. To perceive a real correlation you need all four cells of a 2x2 table, but attention and memory only reliably fill the loud one.

Where you'll see it

  • **The clinical Rorschach myth.** In Chapman and Chapman's 1967 study, students and clinicians alike 'saw' that certain Draw-A-Person features (like emphasized eyes) went with paranoia, even though the drawings were randomly paired with symptoms and the real correlation was zero. In a follow-up (Chapman and Chapman, 1969), the same illusion made clinicians insist that specific Rorschach responses signaled homosexuality. The pairings felt obvious because they were semantically associated, and genuinely valid diagnostic signs got ignored precisely because they were not intuitively linked.
  • **Minority group stereotypes from thin air.** Hamilton and Gifford (1976) showed people sentences about a big Group A and a small Group B, keeping the ratio of desirable to undesirable behaviors identical for both. Because minority membership and negative behavior were both rare, their co-occurrence was distinctive, and subjects reliably rated the smaller group more negatively. No real difference existed. It is a lab-clean demonstration of how prejudice can form with zero actual evidence.
  • **The full moon in the ER.** Nurses and cops routinely insist chaos spikes on full moons. The meta-analysis by Rotton and Kelly (1985) pooled 37 studies of psychiatric admissions, crisis calls, homicides, and other crime, and found the lunar phase explained no more than 1 percent of the variance. But a wild shift under a full moon is a memorable pairing, and every calm full-moon night gets forgotten, so the belief survives.
  • **Arthritis and the weather.** Patients are certain their joints predict rain. Redelmeier and Tversky (1996) tracked arthritis sufferers for over a year and found essentially no link between reported pain and local weather, yet patients stayed convinced. The painful day that was also rainy sticks; the painful dry days and the rainy pain-free days do not get counted.

Where it comes from

Loren J. Chapman and Jean P. Chapman coined the term and ran the seminal work in "Genesis of Popular but Erroneous Psychodiagnostic Observations" (Journal of Abnormal Psychology, 1967). They showed that naive students and experienced clinicians alike reliably 'discovered' correlations between Draw-A-Person signs and symptoms that were completely absent in randomly paired materials, driven by prior semantic association. Two years later, in "Illusory Correlation as an Obstacle to the Use of Valid Psychodiagnostic Signs" (1969), they extended the finding to the Rorschach and showed the illusion even overrode genuinely valid signs. In 1976, David Hamilton and Robert Gifford carried the idea into social psychology, demonstrating that stereotypes of minority groups can emerge purely from the distinctiveness of rare-plus-rare co-occurrences, with no real group difference present.

How to counter it

Fill in the whole 2x2 table. A correlation lives in four cells, not one. Before you trust "X goes with Y," deliberately count the other three: X without Y, Y without X, and neither happening. The vivid cell is the one your memory hands you for free; the non-events are the data you have to go dig for.

Log it, then trust the log over your recall. Your memory is biased toward the distinctive, so stop sampling it. Write events down as they happen (supplement-and-symptom, ER-shift-and-moon-phase) and let the tally decide. If you cannot produce a number, you do not actually know the correlation, you just remember the highlight reel.

Ask what a fake link would look like. If a full-moon chaos night and a normal chaos night would feel identical to you, your gut cannot tell the two worlds apart. Default to "no correlation" until real frequency data forces you off it.

The tell

You catch yourself saying "it always happens when..." or "you can just tell" while reaching for one or two unforgettable stories instead of a count. The giveaway is that you can vividly name the hits but go blank when asked how many times the same setup produced nothing.

Related biases

References

  1. Chapman, L. J., & Chapman, J. P. (1967). Genesis of popular but erroneous psychodiagnostic observations. Journal of Abnormal Psychology, 72(3), 193-204
  2. Chapman, L. J., & Chapman, J. P. (1969). Illusory correlation as an obstacle to the use of valid psychodiagnostic signs. Journal of Abnormal Psychology, 74(3), 271-280
  3. Hamilton, D. L., & Gifford, R. K. (1976). Illusory correlation in interpersonal perception: A cognitive basis of stereotypic judgments. Journal of Experimental Social Psychology, 12(4), 392-407
  4. Mullen, B., & Johnson, C. (1990). Distinctiveness-based illusory correlations and stereotyping: A meta-analytic integration. British Journal of Social Psychology, 29(1), 11-28
  5. Rotton, J., & Kelly, I. W. (1985). Much ado about the full moon: A meta-analysis of lunar-lunacy research. Psychological Bulletin, 97(2), 286-306