Ben Goldacre
Thursday July 14, 2005
The Guardian

Talk about bad science here.
· And so our highly improbable Popular Statistics with Sperm series continues: I’ve locked myself in the Bad Science office with two weeks’ worth of baked beans and they haven’t been able to sack me yet. This week we look at cognitive illusions: they’re a major reason we use statistics, to avoid falling prey to our tendency to detect patterns and causality in random patterns. Studying cognitive illusions – a bit like optical illusions, only for beliefs – can help us understand why people can wind up believing weird things such as, for example, that alternative therapies work, even when they don’t.

· One important contributor to this is our failure to spot something called “regression to the mean”. We live in a state of constant flux. You can see, now, how easily I could pass over to the dark side, and start writing self-help books. Anyway, we live in a state of flux, as in, things get better, then they get worse, often randomly. Maybe you have back pain. It comes and goes, sometimes it’s very bad, sometimes it’s not. If you roll a six, the chances are your next roll will be closer to three and a half, the average or “median” score on a dice: the same if you roll a one. That’s regression to the mean.

· Sporting chaps in America talk about the “Sports Illustrated Jinx”: if you appear on the cover of the magazine, bad things happen to your game. Bless them. If you appear on the cover of the magazine, it’s because you’re performing extremely well, unexpectedly well perhaps. This will be through a mix of talent and luck. Luck, or noise, generally passes, or regresses to the mean, by itself, just like with the dice. If you fail to understand that, you start looking for another cause for that regression, and you find … the Sports Illustrated Jinx.

· Let’s say someone has a theory that rubbing semen into your face is good for spots. Again. Just go with it. So for various reasons, you might only cave in and rub it on when your spots are really bad. They were probably going to get better anyway, from their worst point, by regression to the mean, but you don’t read Bad Science, so you didn’t know about that, so you ascribe causal significance to the semen. In fact, you become an evangelist for the stuff. Or alternatively you can learn about statistics: because if we use a sample size of greater than one, and we have a control group to compare with, we can avoid this kind of sticky mess.