Us psychology types are constantly reading and thinking about things like logic, experimental design, and statistics. I recently came across a nice little article, Mistakes in Experimental Design and Interpretation, that summarizes a bunch of issues in designing and interpreting science experiments.
I found the last point the most interesting:
Mistake I9: Being Too Clever
Sir R. A. Fisher (1890-1962) was one of the greatest statisticians of all time, perhaps most noted for the idea of analysis of variance. But he sullied his reputation by arguing strongly that smoking does not cause cancer. He had some sensible arguments. First, he rightfully pointed out our Mistake I7, correlation is not causation. He was clever at coming up with alternative scenarios: perhaps lung cancer causes an irritation that the patient can feel long before it can be diagnosed, such that the irritation is alleviated by smoking. Or perhaps there is some unknown common cause that leads to both cancer and a tendency to smoke. Fisher was also correct in pointing out Mistake D1, lack of randomized trials: we can’t randomly separate children at birth and force one group to smoke and the other not to. (Although we can do that with animal studies.) But he was wrong to be so dismissive of reproducible studies, in humans and animals, that showed a strong correlation, with clear medical theories explaining why smoking could cause cancer, and no good theories explaining the correlation any other way. He was wrong not to see that he may have been influenced by his own fondness for smoking a pipe, or by his libertarian objections to any interference with personal liberties, or by his employ as a consultant for the tobacco industry. Fisher died in 1962 of colon cancer (a disease that is 30% more prevalent in smokers than non-smokers). It is sad that the disease took Fisher’s life, but it is a tragedy that Fisher’s stuborness provided encouragement for the 5 million people a year who kill themselves through smoking.
It’s a nice reminder that sometimes, knowing too much can get in the way of seeing the truth that’s right in front of us, and can even be deadly. If we don’t agree with some conclusion, we can whip out all the “correlation does not equal causation”, “research is still inconclusive”, and “there was no control group” we want, but that doesn’t make the conclusion false. Deep issues concerning statistics and scientific reasoning are important, sure, but sometimes we just need to look past these trees and see the giant fucking forest that’s been there all along.