Saturday, September 06, 2008
Correlation and causation
Everyone examines data for evidence that their side is right. If a statistical morsel seems to our taste, the temptation is to swallow it.
Such zeal produces, in this case, the following logic:
Here's event A - a smoking ban.
Here's event B - a selective fall in heart attacks.
Therefore, A caused B.
And if the earlier months for which data is available show an inconvenient rise, the logic presumably staggers on - why, it simply proves that it took a few months for the effect to appear, surely?
It's the kind of logic that ought to wilt with embarrassment. In the event, no one even blushed. Is it serious? Well, it passed for broadsheet news, political and medical establishment opinion.
