The Correlation
Files
Three numbers that moved in perfect sync for over a decade, and had nothing whatsoever to do with each other. Scroll to review the evidence.
Exhibit A
Nicolas Cage films vs. accidental pool drownings
Every year the actor released more films, more Americans drowned in swimming pools. Fewer films, fewer drownings. For twelve straight years, the two lines barely left each other's side.
Trend shapes stylized for illustration, in the spirit of the widely-cited Vigen (2014) spurious-correlations dataset. Not exact published figures.
Exhibit B
US margarine consumption vs. divorce rate in Maine
As Americans quietly spread less margarine on their toast each year, marriages in Maine quietly got a little sturdier too. Nobody has ever explained the mechanism. Nobody ever will.
Trend shapes stylized for illustration, in the spirit of the widely-cited Vigen (2014) spurious-correlations dataset. Not exact published figures.
Exhibit C
Arcade revenue vs. computer science doctorates
Fewer quarters dropped into arcade machines lined up almost exactly with fewer people finishing a PhD in computer science. The overlap is uncanny. The explanation is nothing.
Trend shapes stylized for illustration, in the spirit of the widely-cited Vigen (2014) spurious-correlations dataset. Not exact published figures.
Act two: Find your own
Go hunting for a coincidence
Pick any two of these six unrelated series. Some pairs won't look alike at all. At least one pair will move in near-perfect sync, purely by chance.
Why does this keep happening?
Track enough variables for enough years and some pairs will drift together by pure chance. This isn't a flaw in statistics: it's an expected feature of it, sometimes called the multiple comparisons problem.
Picture 64 random variable pairs, tracked over time. Almost none of them are related. But a handful will still line up well enough to look damning:
Three pairs out of sixty-four line up by chance alone: no relationship required, just enough attempts.
The lesson isn't that correlation is useless. It's that a strong correlation is an invitation to ask why, never proof that you already know.