Statistikken biter tilbake

Malcolm Gladwell har en morsom liten filosofering om pitbull-hunder. Som en av kommentatorene sier: Problemet er vel snarere at man har ustabilt aggressive eiere men stabilt aggressive hunder.

Måleproblemet – utvalgsskjevhet – er imidlertid høyst reellt. Minner om den klassiske historien om kulehullene i bombefly under annen verdenskrig:

Statistician Abraham Wald’s research on where to put extra armor on warplanes during World War II is a wonderful example. The British and U.S. air forces were concerned because many of their planes were being shot down. They wanted to use more armor, but were not quite sure where to put it. Wald put a mark on every bullet hole in the airplanes that returned from battle. He found that two major sections of the fuselage — one between the wings and the other between the tails — had far fewer bullet holes. He decided to put the armor in these places, where he saw fewer, not more, holes. Why? Because it stood to reason that the planes were hit randomly. The planes he analyzed had not been shot down! So it was the holes he wasn’t seeing — in the planes that weren’t returning — that needed extra protection.