← The Casebook

Send station · the claim, dispatched

Clustering Illusion

The clustering illusion is the tendency to perceive meaningful streaks or clumps in small samples of data that are in fact fully consistent with a random process.

Supports

Mis-route → belongs in Refutes

The bias files you catch yourself explaining a run of outcomes ('three plane crashes this month as confirmation.

Subjective randomness / alternation bias

Perceived randomness is best predicted by how hard a sequence is to encode/memorize

qual.

Refutes / corrects

Hot hand in basketball (canonical)

No statistically significant positive dependence between consecutive shots

n≈76

Streak selection bias correction (reversal)

Correcting the bias reverses GVT's conclusion: the data show meaningful hot-hand shooting

n≈26

In the wild

V-1 flying-bomb hits on London modeled as Poisson (no clustering)

Actuary R

1946

The cancer-cluster myth

Reported neighborhood 'cancer clusters' overwhelmingly fail to show a true environmental cause

1999

Hot-hand belief in casino betting

Analysis of real online casino betting records shows gamblers act on perceived streaks: after wins they bet as if a hot…

2014

Intermittent jam

Dispatch log · replication mixed. The general phenomenon that people misperceive randomness, expecting too few streaks and too much alternation, is robust and repeatedly demonstrated (Falk & Konold 1997; large subjective-randomness literature). But the single most famous instance, the basketball hot hand as a 'cognitive illusion,' is genuinely contested: Miller & Sanjurjo (2018, Econometrica) identified a streak selection bias that, once corrected, reverses GVT's original conclusion and a close replication of it, indicating a real hot hand in shooting. So 'people see clusters that aren't there' holds broadly, while 'the basketball streaks definitely weren't there' does not.

Pressure diagram · why it sorts this way

The dominant account is the representativeness heuristic plus the 'belief in the law of small numbers' (Tversky & Kahneman 1971-1972): people hold a mental prototype of randomness that is too uniform and too alternating, so when a real random sequence produces the runs and clumps it inevitably must, those clumps violate the prototype and get attributed to a cause (a trend, a hot shooter, an environmental hazard). Falk & Konold (1997) refine this: perceived randomness tracks subjective encoding difficulty, so streaky sequences (easy to encode, e.g. HHHHH) read as 'non-random' while high-alternation sequences read as 'random.'

An evolutionary/adaptationist account frames over-detection of pattern as the cheaper error (missing a real pattern was costlier than a false alarm). A pointed competing claim comes from the hot-hand reversal literature: Miller & Sanjurjo (2018) argue that in the canonical basketball case the 'illusion' was partly a statistical artifact in the researchers' estimator, not (only) a bias in the observers, so what looked like misperception of randomness was in part mis-measurement of real dependence.

Manual override

Compute or simulate the null: ask what a genuinely random process of the same size would produce (e.g. runs of 5+ heads are near-certain in 100 flips), pre-specify the hypothesis and population before looking at the data to avoid drawing the target around the cluster, and use larger samples since the illusion shrinks as n grows. Where the 'pattern' is a performance streak, beware that the naive streak estimator is itself biased (Miller & Sanjurjo), so both over- and under-detection are possible.

Clarke (1946) defused perceived V-1 clustering by fitting a Poisson model; epidemiologists routinely find reported cancer clusters dissolve once the base population and expected counts are specified in advance; Miller & Sanjurjo (2018) show why the streak statistic must be bias-corrected before concluding anything.

Routed from Thomas Gilovich, Robert Vallone, Amos Tversky, 1985 — The Hot Hand in Basketball: On the Misperception of Random Sequences.

Adjacent stations

File your own case

Open the same case on your own draft.

Paste a memo, a research draft, or a strategy argument. It is scored against all 175 cards, and the strongest two or three risks come back with the evidence quoted and one practical next check.

Open a case on your draft →