Monday 4 April 2011

Concerning Correlation

I write this post with a certain trepidation, since my purpose is to explain to an economics professor how tricky it is to judge correlation by eye from a scatterplot. This calls to mind an old New Statesman competition, in which readers were invited to express a familiar proverb in verse. The winning entry was:
Teach not thy parent’s mother to extract
The embryo juices of the bird by suction.
The good old lady can that feat enact,
Quite irrespective of your kind instruction.
Anyway, here’s my illustration of the problem. Take a look at the far-from-random scatterplot below. (There are 301 observations.) Now – no peeking – guess the correlation coefficient.




If you correctly guessed that the answer is 0.93 then I suspect your peripheral vision kicked in.