Was the hot hand a fallacy after all?
The idea
In 1985 Thomas Gilovich, Robert Vallone and Amos Tversky published a paper in Cognitive Psychology asking whether basketball players really get a “hot handhot hand The idea that someone who has just succeeded several times in a row is more likely than usual to succeed on the next try.”, a spell in which a run of made shots makes the next one more likely. Their cleanest test was a controlled shooting experiment with 26 players from Cornell’s men’s and women’s teams, each taking 100 shots. Players hit 49 percent of shots after three hits in a row and 45 percent after three misses, a small gap that was not statistically significant. The authors concluded that belief in the hot hand was a widely shared cognitive illusion. The result became a standard teaching example of people seeing patterns in random sequences.
In 2018 Joshua Miller and Adam Sanjurjo showed in Econometrica that the standard test had a flaw. The usual method takes each player’s sequence, finds the shots that follow a streak, computes the hit rate on those shots, and compares it with the rate after a streak of misses. Miller and Sanjurjo proved that, in a finite sequence, this measure is biased downwards even for a perfectly random shooter. They called it streak selection biasstreak selection bias A downward distortion that appears when you measure what happens right after a run of outcomes within a short sequence, even when every outcome is pure chance..
The coin version makes it concrete. Flip a fair coin three times. There are eight equally likely sequences. Drop the two (TTH, TTT) where no flip follows a head. For each of the other six, record the share of heads among flips that come right after a head: HHH gives 1, HHT gives 1/2, HTH gives 0, HTT gives 0, THH gives 1, THT gives 0. The average is 2.5/6, or 5/12, about 0.42, not 0.5. The coin has no memory; the shortfall comes from the way flips are selected. Within a short sequence, a head that follows a head uses up one of a limited number of heads, so the flips that follow are slightly more likely to be tails.
Why it might change a decision
The bias predicts that a perfectly random shooter, measured this way, will look anti-streaky. So finding no difference does not mean no hot hand; it means performance after hits beat a baseline that should have been below zero. For any analysis of streaks in sports, markets or workplace performance, the testable implication is that a raw “after-streak minus after-non-streak” gap near zero should be read as weak evidence for streakiness, and should be compared against the expected bias for that sequence and streak length rather than against zero.
How solid is this?
The mathematical result is a proof, not an empirical finding, and the paper states that the bias shrinks as sequences get longer but grows as the streak being conditioned on gets longer. For 100 shots, a 50 percent shooter and streaks of three, the expected after-hits minus after-misses difference for a random shooter is about minus 8 percentage pointspercentage point The plain difference between two percentages, so a move from 40 to 45 percent is a rise of 5 percentage points.. Applied to GVT’s Cornell data, the correction raises the average difference from about plus 3 to about plus 13 percentage points, significant at the 1 percent level. Miller and Sanjurjo describe this as reversing the canonical study’s conclusion.
That reversal rests on a single small experiment with college players, so its size should be held loosely. The figures here come from the authors’ arXiv version of the paper, which carries the same title as the published one.
Game data are harder, because defenders and shooters react. Andrew Bocskocsky, John Ezekowitz and Carolyn Stein, at the 2014 MIT Sloan Sports Analytics Conference, used optical tracking of more than 83,000 shots from the 2012-2013 NBA season. Players who had been outperforming took harder shots, from farther out and against tighter defence. After modelling shot difficulty they estimated a hot hand of 1.2 to 2.4 percentage points. This is a conference paper rather than a journal article, and the estimate depends on how well the difficulty model captures what defenders see.
What the newer work does not show is that fans and players judge streaks accurately. A real effect of a few percentage points is compatible with people perceiving a much larger one, and the popular claim that “the hot hand is real” can outrun the data just as the old fallacy claim did. The strongest competing explanation for residual streakiness in games is unmeasured shot difficulty; in controlled shooting it is that a small sample of players can produce a large average by chance.
Sources
- Gilovich, T., Vallone, R., & Tversky, A. (1985). The hot hand in basketball: On the misperception of random sequences. Cognitive Psychology, 17(3), 295-314. doi:10.1016/0010-0285(85)90010-6
- Miller, J. B., & Sanjurjo, A. (2018). Surprised by the hot hand fallacy? A truth in the law of small numberslaw of small numbers The habit of expecting small samples to look as balanced as the long-run averages they come from.. Econometrica, 86(6), 2019-2047. doi:10.3982/ECTA14943. Author version: arXiv:1902.01265
- Bocskocsky, A., Ezekowitz, J., & Stein, C. (2014). The hot hand: A new approach to an old “fallacy”. MIT Sloan Sports Analytics Conference. sloansportsconference.com