Regression Neglect: Why Extreme Performance Doesn't Last

A flight instructor watches a trainee execute a perfect landing. He praises the student lavishly. On the next attempt, the landing is mediocre. Another trainee makes a terrible landing; the instructor delivers sharp criticism. The next attempt is better. The instructor concludes, from repeated experience, that praise leads to deterioration and punishment leads to improvement. He is wrong. What he is observing is regression to the mean — and his misinterpretation of it has led him to a false theory of human learning that may be harming his students.

What Regression to the Mean Is

Regression to the mean is one of the most important and most neglected concepts in statistics. It was first described by Francis Galton in the 1880s, who noticed that tall fathers tended to have sons who were tall but less tall than their fathers — and that short fathers tended to have sons who were short but less short than their fathers. The population didn't become more extreme over time; it regressed toward the average.

The mechanism is simple: any measurement or performance that is influenced by both genuine ability (or trait) and random variation (luck, noise, measurement error) will tend to show extreme values when the random component happens to be extreme in the same direction as the underlying trait. An exceptionally good score likely reflects both genuine skill and a lucky day. An exceptionally bad score likely reflects genuine limitations plus an unlucky day. On the next measurement, the luck component will be closer to average — producing a result closer to the mean, regardless of any intervention.

Regression neglect is the failure to account for this. When performance improves after criticism or deteriorates after praise, we attribute the change to the intervention rather than to the statistical return to average. This is the error Kahneman's flight instructor made — and it is one of the most consequential cognitive mistakes in applied settings.

Kahneman's Flight Instructor Story

Daniel Kahneman recounts the flight instructor episode in Thinking, Fast and Slow as one of the most important insights of his career. He was teaching a seminar on behaviour modification to Israeli Air Force instructors when one of them delivered a pushback: in his experience, praise reliably caused deterioration (the trainee got complacent) while criticism reliably caused improvement (the trainee tried harder).

Kahneman's response was to explain regression to the mean. When a trainee executes an unusually good landing, that performance is, by definition, above their normal level — it includes a lucky component. The next performance is more likely to be closer to average: worse. The instructor praises, then observes "deterioration" — and attributes it to the praise. When a trainee executes an unusually poor landing, the next one is more likely to be better — closer to average — regardless of whether the instructor says anything. The criticism appears to "work" because the regression was coming anyway.

The instructor had trained himself, through repeated experience, to believe that criticism improves and praise harms — the precise opposite of what a century of behaviour science suggests about reinforcement. The statistical artefact of regression had generated a plausible causal story, consistently reinforced by experience, that was entirely wrong.

The Sports Illustrated Cover Jinx

The "Sports Illustrated cover jinx" is the popular belief that athletes who appear on the cover of Sports Illustrated subsequently suffer a decline in performance. The belief is widespread enough to be treated as folk wisdom in American sports culture — some athletes reportedly decline cover invitations to avoid the curse.

The statistical explanation is straightforward: athletes appear on the magazine's cover when they are performing at peak levels — an exceptional season, a record-breaking game, a championship win. These peaks, by the logic of regression, are likely to be followed by performance closer to average. The cover didn't cause the decline; the cover was offered precisely because the athlete was at a peak, and peaks regress. The "jinx" is regression to the mean wearing a supernatural costume.

The same effect drives a dozen similar myths: the "sophomore slump" (a rookie of the year performer whose second season disappoints), the "Super Bowl hangover" (champions underperforming the following season), the "hot new CEO" whose second year fails to match the first. In every case, the outstanding initial performance that created the reputation included a component of variance that is not replicable on command.

Medicine: The Most Dangerous Application

Regression to the mean is perhaps most dangerous in medicine, because it can make ineffective treatments appear effective and effective ones appear useless.

Consider a patient with high blood pressure. Blood pressure fluctuates considerably from day to day and even from measurement to measurement. A patient whose blood pressure is measured when it is exceptionally high will, on remeasurement, likely show a lower reading — not because of any treatment, but because the first reading was at the high end of their natural variance. If they were started on a new medication between measurements, the medication appears to work. If they received a placebo, the placebo appears to work. If they ate a special diet, the diet appears to work. Regression to the mean is the single biggest reason that poorly designed medical studies produce false positive results.

This is why randomised controlled trials (RCTs) with control groups are essential. The control group, receiving no intervention, also shows regression toward the mean — capturing the "natural improvement" that would have occurred anyway. Only by comparing treated and untreated groups can you isolate the effect of the treatment from the effect of regression.

Alternative medicine is particularly susceptible to exploitation by this effect. People typically seek treatment when they are at their worst — at the peak of their symptoms. Whatever intervention they receive, natural regression means they will tend to feel better afterward. The practitioner and patient both observe "improvement following treatment" and attribute causation to the treatment. The regression was coming regardless. This is not a claim that all alternative therapies are useless — it is an observation that improvement following treatment is extremely weak evidence of efficacy without a control group.

Management and Performance Evaluation

Kahneman argues that regression to the mean is among the most important concepts for managers to understand, and among the least understood. The practical implications are counterintuitive:

• Rewarding good performance appears to produce worse subsequent performance — not because reward is harmful, but because good performance was already at a peak, and regression brings it down. Managers who observe this pattern can incorrectly conclude that reward is counterproductive.
• Punishing bad performance appears to produce better subsequent performance — not because punishment is beneficial, but because bad performance was at a trough, and regression brings it up. Managers who observe this can incorrectly develop an authoritarian management theory built on a statistical artefact.
• Employee evaluations are noisy. An employee who has an exceptional quarter is likely to have a worse subsequent quarter; one who has a terrible quarter is likely to improve. Performance management systems that respond aggressively to variance (bonuses for exceptional quarters, performance improvement plans for poor ones) may be measuring noise, not signal.

The managerial insight that follows: distinguish between genuine changes in underlying capability and regression from variance. The former warrants intervention; the latter will self-correct. The difficulty is that, in real time, you can almost never be certain which you're observing.

Regression Neglect in Finance and Research

Investment fund performance is a textbook example. Funds that outperform in one year significantly disproportionately attract investor inflows for the following year — precisely when, due to regression, their outperformance is least likely to continue. Studies by Carhart (1997) and others show that past mutual fund performance is a poor predictor of future performance, yet fund flows respond strongly to recent results. Investors systematically buy performance at its peak, just as regression is about to pull it toward average.

In academic research, "winner's curse" effects operate similarly: studies with dramatic, unexpected results that pass significance thresholds are more likely to contain chance variations that will not replicate. This is a major driver of the replication crisis in psychology and medicine — not fraud, not deliberate bias, but regression to the mean operating on the selective publication of significant findings.

How to Account for Regression

Regression to the mean cannot be wished away — it is a mathematical fact about measurements that contain noise. But several practices reduce its misinterpretation:

• Multiple measurements: Base assessments on averages of multiple observations rather than single data points. Average performance is less susceptible to regression than individual extreme readings.
• Control groups: Whenever evaluating an intervention, compare against an untreated control group that will show the same regression pattern without the intervention.
• Lengthen the time horizon: Single-period evaluation windows capture maximum variance; longer evaluation periods increasingly reflect underlying signal rather than noise.
• Ask: "Is this extreme performance repeatable?": When assessing someone on the basis of an exceptional performance — a single quarter, a single game, a single test — explicitly ask whether it reflects persistent skill or peak variance.

Galton noticed regression to the mean in heights; Kahneman's flight instructor had to have it explained to him despite decades of firsthand experience. The lesson is not that our brains are broken — they are doing something entirely natural, finding causal stories for observed changes. The lesson is that in a world full of noise, extreme values are routinely followed by less extreme ones, and we need an explicit corrective to stop ourselves from finding meaning in the regression.

Sources & Further Reading

• Kahneman, D. Thinking, Fast and Slow. Farrar, Straus and Giroux, 2011. Chapter 17: "Regression to the Mean."
• Galton, F. "Regression Towards Mediocrity in Hereditary Stature." Journal of the Anthropological Institute 15 (1886): 246–263.
• Carhart, M. M. "On Persistence in Mutual Fund Performance." Journal of Finance 52, no. 1 (1997): 57–82.
• Morton, V., & Torgerson, D. J. "Effect of Regression to the Mean on Decision Making in Health Care." British Medical Journal 326 (2003): 1083–1084.
• Barnett, A. G., van der Pols, J. C., & Dobson, A. J. "Regression to the Mean: What It Is and How to Deal with It." International Journal of Epidemiology 34, no. 1 (2005): 215–220.
• Wikipedia: Regression toward the mean