Regression to the Mean Explained (With Examples)
Quick answer: Regression to the mean is the tendency for unusually high or low measurements to be followed by measurements closer to the average, when the two measurements are imperfectly correlated. It does not mean everyone becomes average. Students chosen for very high test scores usually score lower on a retest, though still above average, even if their ability has not changed, because temporary influences on the first score rarely repeat the same way.
Facts checked November 22, 2026.
A rookie has a spectacular first season, then a disappointing second one: the “sophomore slump.” A company has a record quarter, then a merely good one. A patient feels worst on the day they finally see a doctor, then improves. In each case people look for a cause. Often the real explanation is a statistical effect called regression to the mean, and once you see it, you will notice it everywhere.
What Regression to the Mean Means
Any measured result can be thought of as
\[ \text{result} = \text{true level} + \text{luck}. \]The true level (skill, ability, health) is fairly stable. The luck part changes from one measurement to the next. If you pick people because their result was extreme, you are selecting people whose results reflect both their underlying level and unusually favorable (or unfavorable) temporary influences. Next time those influences will probably be more ordinary, so as a group their results move back toward the average. Nothing about them has to change. This is a group tendency and a prediction, not a rule for every individual: some top scorers will do even better next time.
A Simple Simulation
We simulated one million students. Each has a true ability (average 70, standard deviation 8) and takes two tests, with random luck on each test (standard deviation 8). The two test scores then have a correlation of about 0.5. (The theoretical values for this model are 89.86 and 79.93 for the top group and 50.14 and 60.07 for the bottom group; our simulated values match.)
| Group (chosen by Test 1) | Test 1 average | Test 2 average |
|---|---|---|
| Top 10% | 89.8 | 79.9 |
| Everyone | 70.0 | 70.0 |
| Bottom 10% | 50.2 | 60.1 |
The top group fell by about 10 points and the bottom group rose by about 10 points, even though no one learned or forgot anything. Notice that the top group is still above average on Test 2: regression pulls results toward the mean, not all the way back to it.
The Regression to the Mean Formula
For two measurements with correlation \(r\), the best prediction of the second, in standard units (z-scores), is
\[ \hat{z}_2 = r \times z_1. \]With \(r = 0.5\), a student who scored 2 standard deviations above the mean on the first test is predicted to score only 1 standard deviation above it on the second. If \(r = 1\), the standardized prediction equals the first result and there is no regression. If \(r = 0\), the best prediction is simply the average. In ordinary units, the general prediction is
\[ \hat{Y} = \mu_Y + r \frac{\sigma_Y}{\sigma_X} (X – \mu_X), \]and when both measurements have the same mean \(\mu\) and spread, it simplifies to \(\hat{Y} = \mu + r(X – \mu)\). A student who scores 90 when the mean is 70 and \(r = 0.5\) has a predicted retest score of \(70 + 0.5 \times 20 = 80\): lower than 90, but still above average, because part of the high score reflects real ability. In our simulation the top group was 19.8 points above the mean on Test 1, so we predict \(0.5 \times 19.8 \approx 9.9\) points above the mean on Test 2, or about 79.9: exactly what we observed.
Try it: Retest Score Predictor
Where the Name Came From
In 1886, Francis Galton published “Regression towards mediocrity in hereditary stature.” He found that very tall parents tend to have children who are tall but, on average, less extreme than their parents, and very short parents tend to have children who are short but closer to average. The word “regression” stuck, and Galton’s work led to the statistical method now called linear regression, the same idea at the heart of many machine learning models today.
Real-World Examples
- The sophomore slump and the “cover jinx.” Athletes often appear on magazine covers or win rookie awards after an exceptional run, and a weaker stretch tends to follow. Part of the exceptional run was favorable variation. Regression explains part of the pattern, but not every decline: injuries, opponents adjusting and role changes are real too, and studies do not find a universal second-year slump.
- Praise and punishment. In Thinking, Fast and Slow (2011), psychologist Daniel Kahneman described flight instructors who believed that praising a great landing made the next one worse and criticizing a bad landing made the next one better. Regression alone predicts exactly that pattern, whatever the instructor says.
- Medicine. People tend to seek treatment when symptoms are at their worst, so many would improve anyway. That is one reason clinical trials need control groups.
- Business and investing. Top-performing funds or salespeople in one year often look more ordinary the next. Our article on whether AI can predict the stock market shows how rarely top performance persists.
- Schools and policy. If the lowest-scoring schools get a new program, their scores will probably rise the next year even if the program does nothing.
How Not to Be Fooled
- Ask how the group was chosen. If it was chosen for an extreme result, expect regression.
- Use a control group. Compare with a similar group that got no treatment, chosen the same way.
- Look at more than one measurement. Averages over many games, tests or quarters contain less luck.
- Estimate the correlation. The lower the correlation between repeated measurements, the stronger the pull toward the mean.
Regression to the Mean and AI
AI benchmark leaderboards show a closely related effect. When many AI models are compared on a limited set of questions, the highest-ranked model may benefit partly from which questions happened to be included. Retest the same models on fresh questions of similar difficulty and the leader’s margin often shrinks. That is why careful evaluations use large, held-out test sets and report uncertainty; see how to read AI benchmarks.
Regression to the Mean vs the Gambler’s Fallacy
| Regression to the mean | Gambler’s fallacy | |
|---|---|---|
| Claim | Extreme results predict less extreme next results when measurements are imperfectly correlated | A random outcome is “due” after a streak |
| Example | A top test score predicts a lower, still above-average retest | Five heads supposedly make tails more likely |
| Correct? | Yes, as a group prediction | No: a fair, independent coin stays 50/50 |
Common Misconceptions
“Regression means things always even out.” No. It is not a force that balances luck; it simply means the next result does not inherit the previous luck. That is different from the gambler’s fallacy.
“Regression brings results all the way back to average.” Only part of the way, depending on the correlation.
“Regression to the mean means everyone becomes average over time.” No. In our simulation the overall score distribution stays unchanged even though selected groups move toward the average.
“If results improve after an intervention, the intervention worked.” Not necessarily, if the group was chosen for being extreme.
Watch: Regression to the Mean Explained
If you learn better by watching, these videos are worth your time.
Check Your Understanding
Five quick questions to see how much of this has stuck. Pick an answer and press Submit; you will see a short explanation either way.
Answer each question and press Submit to see your result right away.
Frequently Asked Questions
What is regression to the mean in simple terms?
It is the tendency for an extreme result to be followed by one closer to average, because the extreme result usually included unusual luck that does not repeat.
What is an example of regression to the mean?
Students who score in the top 10% on one test usually score lower, though still above average, on a second test, even if their ability has not changed.
Is regression to the mean the same as the gambler’s fallacy?
No. The gambler’s fallacy wrongly assumes luck will balance out. Regression to the mean only says that future luck is ordinary, so extreme results are not likely to repeat.
Who discovered regression to the mean?
Francis Galton described it in 1886 in a study of the heights of parents and their adult children.
How do you avoid being fooled by regression to the mean?
Ask how the group was selected, compare with a control group chosen the same way, and look at averages over several measurements.
Keep Exploring
- Survivorship Bias Explained: Examples and WWII Planes
- Friendship Paradox Explained: Why Friends Have More Friends
- How to Read AI Benchmarks: 7 Statistical Checks
Sources and Further Reading
Related to This Article
More math articles
- Cholesterol Ratio & LDL Calculator
- Political Ideology and Social Policy: Values and Program Design
- Riemann Sum Calculator (Left, Right, Midpoint, Trapezoid)
- Monroe Doctrine and Hemispheric Diplomacy: Principles, Context and Power
- Brick Calculator
- Human Migration, Culture, and Population
- 3rd Grade PARCC Math Practice Test Questions
- The Best Grade 5 Math Book for Kentucky Students
- Measurement, Instruments, and Safety
- Framing


What people say about "Regression to the Mean Explained (With Examples) - Effortless Math"?
No one replied yet.