The Regression Fallacy: Why Things Seem to Get Better After Bad Luck

A batter who hits four home runs in a week makes the cover of a sports magazine. The following week, he goes hitless in six games. Fans whisper about the "Sports Illustrated jinx." A patient with chronic back pain tries acupuncture during a particularly bad flare-up and feels significantly better afterward. She recommends it to everyone she knows. A child does terribly on an exam; her parents enrol her in an intensive tutoring programme; she does much better next time. The tutors are praised. In each case, something plausible-sounding is credited with an outcome that statistics would have predicted anyway. This is the regression fallacy.

What Is Regression to the Mean?

Before understanding the fallacy, you need to understand the statistical phenomenon it misidentifies. Regression to the mean is the tendency for extreme measurements of a variable to be followed by measurements closer to the long-run average, simply because extreme outcomes partly reflect temporary deviations rather than stable underlying conditions.

The concept was first formalised by Francis Galton in 1886, studying how the heights of children related to those of their parents. He noticed that very tall parents tended to have children who were tall, but not as tall as they were; very short parents had children who were short, but not as short. The children "regressed" toward the population mean. He initially called this "regression toward mediocrity" — a term that has, unfortunately, been lost to time.

Any measurement that contains both a stable component and a random one will show this pattern. When the random component is extreme in one direction, the next measurement will likely be less extreme — not because anything changed, but because extreme randomness doesn't typically repeat. This is a mathematical certainty, not a physical law about reversion or balance.

The Fallacy: Crediting an Intervention

The regression fallacy occurs when someone observes an improvement (or decline) following an extreme measurement and attributes it to an intervention — a treatment, a policy, a coaching session, a ritual — when the statistical baseline alone would predict the same outcome.

The logical structure is:

1. Variable X is measured at an extreme value.
2. Intervention Y is applied.
3. Variable X returns toward its average.
4. Conclusion: Y caused the improvement.

The problem is that step 4 doesn't follow. The return toward average was statistically expected regardless of step 2. The intervention gets credited for something that was going to happen anyway.

The Sports Illustrated Cover Jinx

Few illustrations are as vivid — or as culturally embedded — as the supposed "Sports Illustrated cover curse." Athletes featured on the cover are said to suffer a subsequent decline in performance. The magazine has been blamed for injuries, losing streaks, and slumps spanning decades of sports history.

The explanation, of course, is regression to the mean. An athlete appears on a magazine cover precisely because they just had an exceptional period of performance. Exceptional performance partly reflects genuine skill and partly reflects luck running hot. After the exceptional period, luck reverts toward average — which looks, from the outside, like a mysterious curse. There is no curse. There is only the baseline nature of extreme outcomes.

The same dynamic explains the "rookie of the year curse" in baseball — players who win the award in their first season often underperform in their second — and the general phenomenon that teams who make breakthrough seasons sometimes disappoint the following year. It's not psychological pressure, it's not complacency, it's not a jinx. It's statistics.

Medical Consequences: When the Fallacy Harms

In medicine, the regression fallacy is not just intellectually interesting — it has real consequences for what treatments people adopt and what gets funded.

Consider the pain cycle. Chronic conditions like back pain, migraines, or arthritis typically fluctuate — they worsen and improve in patterns that are partly predictable and partly random. Patients, naturally, seek treatment when the condition is at its worst. If they try a new therapy during a flare-up and the pain subsides afterward, the therapy gets credit — even though the subsidence was the statistically expected next step in the cycle regardless of treatment.

This is the mechanism by which many alternative and complementary therapies accumulate testimonial evidence. Healing crystals, homeopathy, magnet therapy, and various other interventions tend to be tried when things are at their worst, and when things improve — as they statistically tend to — the intervention is credited. Controlled clinical trials exist precisely to rule out this effect: if a treatment is no better than regression alone would predict, it has not demonstrated efficacy.

Daniel Kahneman, in Thinking, Fast and Slow, gives a striking military example. Flight instructors in the Israeli Air Force observed that trainees who were praised after a good performance often did worse next time, while those who were criticised after a bad performance often improved. This seemed to confirm that praise was counterproductive and criticism was effective. The actual explanation: both were regression to the mean. Average performance followed both exceptional success and exceptional failure, regardless of the feedback given. The instructors had convinced themselves of a causal theory built on a statistical artefact.

The Hot Hand — Real or Fallacy?

A fascinating complication: the famous "hot hand fallacy" — the belief that an athlete who has made several shots in a row is more likely to make the next one — was long considered a textbook example of misperceiving random sequences. A landmark 1985 study by Gilovich, Vallone, and Tversky found no statistical evidence for the hot hand in basketball.

But more recent analyses (Miller and Sanjurjo, 2018) have complicated this picture, suggesting that the original study had a methodological flaw related to how sequences were sampled, and that there may be genuine streakiness in some athletic performance. The debate continues. What remains clear is that naive observation of streaks and slumps is unreliable as evidence for either a "hot hand" or a "curse" — precise statistical analysis is required, and the default prior should account for regression.

Economic and Policy Implications

Regression to the mean creates systematic distortions in how we evaluate policies and interventions.

If a city implements a new traffic safety programme in the year following an unusually high number of accidents, and accidents then decline, how should we interpret the decline? Regression to the mean suggests we'd expect some decline anyway. Only a properly controlled study — comparing the city's experience to similar cities without the programme, or to a baseline trend — can isolate the programme's actual effect.

Similarly, schools or hospitals identified as "poorly performing" and subjected to intensive interventions often show improvement afterward — improvement that would require careful decomposition to attribute properly. How much is the intervention? How much is the statistical tendency of poor performers to look slightly better in a subsequent measurement? How much is simply noise reverting?

This doesn't mean interventions don't work. It means that naive before-and-after comparisons, applied specifically to extreme cases, systematically overestimate intervention effectiveness. This is why randomised controlled trials randomise: so that regression effects apply equally to treatment and control groups and therefore cancel out.

Recognising the Fallacy in the Wild

The regression fallacy is easy to commit because it requires no motivated reasoning — it follows naturally from normal human pattern recognition. We see sequence, we infer cause. A few markers to watch for:

• Interventions applied at extreme moments: If the treatment was tried precisely because things were unusually bad (or unusually good), the subsequent change may partly or wholly reflect regression.
• No control group: Without comparison to a similar case that didn't receive the intervention, separating regression from genuine effect is impossible.
• Testimonials rather than statistics: "It worked for me when I was at my worst" is precisely the pattern regression predicts, regardless of treatment.

Related Concepts

The regression fallacy is distinct from — but related to — several other reasoning errors. False cause and correlation-to-causation errors are the broader family to which it belongs: inferring a causal relationship from a sequential or correlated pattern. Confirmation bias amplifies it: we notice and remember the cases where the intervention seemed to work; we forget or rationalise away the cases where it didn't. And the closely named regression to the mean entry on this platform covers the underlying statistical phenomenon in more detail.

Summary

Sources & Further Reading

• Galton, Francis. "Regression Towards Mediocrity in Hereditary Stature." Journal of the Anthropological Institute, 1886.
• Kahneman, Daniel. Thinking, Fast and Slow. Farrar, Straus and Giroux, 2011. (Chapter 17: "Regression to the Mean")
• Gilovich, Thomas, Robert Vallone, and Amos Tversky. "The Hot Hand in Basketball." Cognitive Psychology 17(3), 1985.
• Miller, Joshua B. and Adam Sanjurjo. "Surprised by the Hot Hand Fallacy? A Truth in the Law of Small Numbers." Econometrica 86(6), 2018.
• Wikipedia: Regression fallacy