Skip to content
All apps

Anscombe's quartet

About this app

Four datasets with the same means, variances, r and regression line but completely different shapes. With residual plot, influence measures, a morph that keeps the statistics fixed, and a build mode for your own cloud.

The statistics

Anscombe’s quartet (Francis J. Anscombe, 1973) consists of four data sets with 11 points each. Means, variances, correlation r ≈ 0.82, regression line ŷ ≈ 3 + 0.5·x and R² ≈ 0.67 agree to two or three decimal places. Yet the scatter plots look completely different: linear with scatter, curved, linear with one outlier, and a vertical column with a single leverage point. The lesson: summary statistics condense data and lose its shape. Always look at the scatter plot before any correlation or regression.

What you see

Four cards show data sets I to IV with points, regression line and a band ŷ ± sₑ. Layers for residuals, means, residual plot and influence measures (leverage h, Cook’s distance) can be switched on. A table sets the nearly identical statistics against those that tell the sets apart: rank correlation ρ, largest residual, largest leverage and largest Cook’s distance. In morph mode the points glide from one data set to the next, and every in-between frame keeps the same statistics and the same line. In build mode you create your own point cloud as a fifth member.

Things to try

Tap a card to enlarge it and read the finding. Start the morph and notice that the numbers in the card header do not move while the shape changes completely. Switch to build mode and move points until the intercept a = 3.00 ± 0.10 and slope b = 0.500 ± 0.010 fall within the corridor; as a bonus, hit r = 0.82 too. Countless shapes manage this, because the line depends only on the normal equations.

Frequently asked questions

What is Anscombe’s quartet?
Anscombe’s quartet is a set of four small data sets published by the statistician Francis Anscombe in 1973. They have almost identical means, variances, correlations and regression lines but look completely different in a scatter plot. They show why data should always be checked graphically.
How can different data share the same regression line?
The least-squares line depends on only two conditions: the residuals sum to zero, and they also sum to zero when weighted by x (the normal equations). Beyond that the line cannot see where the points lie. That is why countless point clouds fit the same line.
What does Anscombe’s quartet teach?
Statistics such as mean, variance, correlation and R² are not enough to describe data. A scatter plot reveals curvature, outliers and leverage points hidden in the numbers. Diagnosis also needs residual plots and influence measures such as Cook’s distance.
What can I do in the Anscombe app?
You compare the four data sets with residuals, residual plots and influence measures, morph them into one another while the statistics stay constant, and in build mode create your own point cloud with the same line, starting from one of the sets, a random cloud or an empty canvas.

Subject: Statistics

More from Statistics

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