The Datasaurus Dozen

Summary statistics can look reassuringly “normal” while the underlying points tell a completely different story. Alberto Cairo’s Datasaurus made that point with a scatter plot that is literally a dinosaur. Justin Matejka and George Fitzmaurice (Autodesk Research) then built twelve more shapes that keep the same means, standard deviations, and correlation — the Datasaurus Dozen .

Below, each panel is the same 142 points rearranged into a different picture. The table of summary stats (to two decimal places) barely moves; the charts do.

Loading…

Why this matters

• Never stop at mean / sd / correlation — plot the data.
• Anscombe’s Quartet made the same point with four datasets; this is that idea turned up to eleven (plus a dinosaur).
• Identical summary stats do not imply identical structure.

        /js/datasaurus-dozen.js