Datasaurus: the summary is not the data

Thirteen datasets. Every summary statistic below is identical to two decimal places. Switch between them and watch the numbers refuse to move — while the picture changes completely.

Dinosaur

Summary statistics

54.26
mean of x
47.83
mean of y
16.71
std dev of x
26.84
std dev of y
−0.06
correlation
142
points

Computed live from whichever dataset is plotted. They will not move. That is not a bug — that is the whole argument.

The wait, what?

A regression, an outlier hunt, and a cluster analysis would all report the same numbers for all thirteen. The mean cannot see the dinosaur. If the summary is all you keep — the report table, the dashboard tile, the model's headline metrics — you have no idea whether you are describing a tidy trend or a dinosaur. Plot the data.

Data credit: the Datasaurus Dozen (Matejka & Fitzmaurice, Autodesk Research, "Same Stats, Different Graphs"), after Alberto Cairo's original Datasaurus. The classic ancestor is Anscombe's quartet (1973). Released for teaching use; embedded here so the page works offline.