Regression to the mean is the statistical tendency for extreme measurements to move closer to the average when measured again. If you score unusually high on a test one day, you’ll likely score closer to your typical average the next time, not because anything changed, but because the extreme result was partly driven by chance. The same works in reverse: an unusually low result tends to be followed by a higher one. This pattern shows up everywhere, from medical trials to sports performance to student test scores, and misunderstanding it leads people to see causes where none exist.
How the Effect Works
Any measurement you take reflects two things: the true underlying value and some amount of random variation. That variation might come from measurement error, natural day-to-day fluctuations, or just luck. When someone records an extreme score, it’s statistically likely that the random component pushed the result further from the average than usual. On a second measurement, that random push is unlikely to repeat in the same direction with the same intensity, so the new result lands closer to the person’s true average.
Think of it this way. If you flip a coin ten times and get nine heads, that’s an extreme outcome. On your next ten flips, you’re overwhelmingly likely to get fewer than nine heads. Nothing about the coin changed. The first result was just unusually far from what you’d expect, and the next one probably won’t be.
The strength of the effect depends on how much random noise is in the measurement. In statistical terms, the lower the correlation between two rounds of measurement, the stronger regression to the mean becomes. When measurements are perfectly correlated (meaning no randomness at all), there’s zero regression. When the correlation is weak, extreme scores on the first measurement snap back dramatically toward the average on the second. Francis Galton, who discovered the phenomenon in the 1880s, found this pattern in human height: children of very tall parents tended to be tall themselves but not quite as tall as their parents, with the data showing a correlation of about 0.50 between parent and child heights. The children’s heights “regressed” partway back toward the population average.
Why It Fools Us in Medicine
Regression to the mean creates one of the most persistent illusions in healthcare: the belief that a treatment worked when the patient simply returned to their baseline. This happens because people typically seek treatment when their symptoms are at their worst. A person visits the doctor on a day when their pain, blood pressure, or anxiety is unusually severe. Whatever they do next, whether it’s taking a new supplement, trying acupuncture, or starting medication, their symptoms are statistically likely to improve simply because the extreme reading was partly driven by random fluctuation.
Research has shown this effect can be surprisingly large. In a series of 15 biochemical blood tests, selecting patients with abnormally high values (three standard deviations above the mean) produced estimated improvements on retesting ranging from about 2.5% to over 37%, with a median around 10%, entirely from regression. No treatment needed. One landmark analysis of clinical trials argued that most improvements attributed to the placebo effect are actually regression to the mean in disguise. In older trial designs that didn’t protect against this bias, placebo groups showed marked improvement. In modern trials designed to account for regression, placebo-treated patients showed a median change of just 0.3%, which wasn’t statistically significant.
This has real consequences. Patients and doctors alike can become convinced that an ineffective treatment is working, simply because symptoms naturally drifted back toward their average level.
The Flight Instructor Illusion
One of the most famous examples of regression to the mean comes from the psychologist Daniel Kahneman, who described his experience with Israeli flight instructors. The instructors had noticed a pattern: when they praised a cadet for an excellent maneuver, the cadet’s next attempt was usually worse. When they criticized a cadet for a poor maneuver, the next attempt was usually better. The instructors concluded that praise hurts performance and criticism helps.
The real explanation was pure statistics. A cadet who performed exceptionally well had likely benefited from some favorable randomness, good wind conditions, a moment of unusual focus, simple luck. On the next attempt, that randomness wouldn’t repeat, so performance would drop back toward average. A cadet who performed terribly had the opposite experience, and their next flight would likely be closer to their norm. The instructors’ praise and criticism just happened to fall between these natural swings, creating the illusion of cause and effect.
This pattern generalizes far beyond flight school. Teachers may believe strict discipline improves grades because they intervene after a student’s worst performance. Coaches bench a player after a terrible game, then credit the benching when performance bounces back. In each case, the improvement was likely to happen regardless.
Sports and the “Sophomore Slump”
Regression to the mean explains many patterns that sports fans attribute to psychology or motivation. A rookie who has a breakout season is statistically likely to perform closer to average the following year. Fans call it a sophomore slump or say the player got complacent, but the simpler explanation is that the outstanding first season included an above-average share of lucky bounces, favorable matchups, or unsustainably hot streaks. The talent is real, but the extreme peak wasn’t entirely representative.
The same logic applies to the “Sports Illustrated cover jinx,” the observation that athletes tend to perform worse after appearing on the magazine’s cover. They made the cover because they were having an unusually great stretch. What followed was, predictably, something closer to normal.
How It Differs From the Gambler’s Fallacy
People often confuse regression to the mean with the idea that outcomes need to “balance out.” They’re different concepts. The gambler’s fallacy says that after flipping ten heads in a row, tails is now “due,” as if the coin remembers its history and compensates. That’s wrong. The coin has no memory, and each flip is independent.
Regression to the mean says something much more modest: the next run of heads is unlikely to be as extreme as ten in a row, simply because ten in a row is rare. It doesn’t predict tails. It doesn’t require any compensating force. It just recognizes that extreme outcomes are unlikely to repeat at the same level of extremity. The previous extreme result stays in the record. Nothing cancels it out. Future results are simply likely to be less extreme on their own terms.
How Researchers Control for It
Because regression to the mean can make useless treatments look effective, researchers have developed ways to separate real effects from statistical drift. The most important is the randomized controlled trial: by randomly assigning patients to a treatment group and a control group, both groups experience the same regression to the mean, so any difference between them reflects the actual treatment effect.
For studies that compare a group’s scores before and after an intervention, a statistical method called analysis of covariance (ANCOVA) helps adjust for baseline differences. Research has confirmed that this approach produces unbiased estimates of treatment effects and is particularly valuable when groups start at different levels. Simply comparing the change in scores from before to after treatment, without this kind of adjustment, leaves the results vulnerable to regression artifacts.
The practical takeaway is straightforward. Whenever you see a dramatic improvement following an extreme starting point, whether it’s your blood pressure dropping after a scary reading, a student’s grades improving after a terrible semester, or a team bouncing back after a historically bad losing streak, regression to the mean is one of the most likely explanations. It doesn’t rule out that something real also happened. It just means the improvement alone isn’t proof that anything you did made the difference.