Skip to content
All apps

Regression to the mean

About this app

A sales team across two quarters: the best fall back and the weakest catch up without anyone changing anything. A slider for the share of luck shows when this happens and how strongly.

With explain mode: an AI tutor walks you through the app, points out what matters in the scene and answers your questions.

The core idea

Regression to the mean appears whenever a measurement combines a stable part and a random part. Someone who scores extremely well once usually had both high skill and good luck. At the next measurement luck is drawn again, so the result tends to lie closer to the average. The lower the correlation r between two measurements, the stronger the effect. Francis Galton described the pattern in 1886 using the heights of parents and their children.

What the app shows

A sales team of 60 people is shown as a ranking and as a scatter plot of two quarters. Arrows show where the top 10 % and the bottom 10 % move in the following quarter. Next to the diagonal y = x sits the regression line, which is flatter. The model is revenue = 100 + 20 · (√(1 − z) · skill + √z · luck), where z is the share of chance. Slope and correlation are both equal to r = 1 − z.

Things to try

Move the “share of chance” slider from “skill only” to “luck only” and step through quarter after quarter. At 0 % everyone keeps their place; at 100 % one quarter says nothing about the next. Then switch on praise for the best and criticism for the weakest: the arrows do not change. This shows why the effect of praise and blame is easily misjudged.

Frequently asked questions

What does regression to the mean mean?
Regression to the mean means that an extreme measurement is, on average, followed by a less extreme one whenever chance plays a part. This is not because the people changed but because the random component is drawn anew. The effect is stronger the weaker two measurements correlate.
How does regression to the mean relate to praise and punishment?
Daniel Kahneman tells of flight instructors who believed praise made trainees careless and criticism helped. After a very good landing, a weaker one usually did follow, and after a bad one a better one. Regression to the mean alone explains this, without praise or criticism having any effect.
How do you spot regression to the mean in data?
A typical sign is that groups selected for an extreme value move towards the average at the next measurement. In a scatter plot of two measurements, the regression line is then flatter than the diagonal. To test a real intervention you need a comparison group without it.
What can I change in the app?
You set the share of chance, play further quarters and can focus on the top 10 %, the bottom 10 % or both. A switch hands out praise and criticism. The numbers panel shows the correlation, the slope and the group averages.

Subject: Data & decisions

More from Data & decisions

A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps