Oral History Interviews

Interviews that offer unique insights into the lives, works, and personalities of modern scientists

Philip Anderson on the bias that both experimental and theoretical physicists can place on their data sets.

Oral history audio excerpt

Philip Anderson on the bias that both experimental and theoretical physicists can place on their data sets.

Anderson:

(I don't think so ???. Probably.) Or people came to call it the Mott transition. Mott needed something to distinguish the Anderson transition from the Mott transition. He called the mobility edge and the metal insulator transition associated with the mobility edge, the Anderson transition. So he accepted the name that the rest of us had been using.

It requires taste and judgement, but it also requires knowing that a lot can go wrong with experiments. When you see one substance is a metal and another is an insulator, that difference isn't something that experimentalists can make a mistake on. I mean it's clear. If it's shiny it's a metal; if it's transparent it's an insulator. It isn't quite like that in the case of these substances but it was like black and white. In the modern era, when I'm worrying about high Tc super conductors, these are very complex materials. The number of mistakes you can make analyzing them is infinite. You need a great deal more experience and subtlety in understanding what experimentalists are telling you. There are other problems. Thanks to the success of the Bell Labs style, the Bell Labs method of working in teams of experimentalists and theorists together, that there is a tendency for every paper to have its own theoretical interpretation. You have to be very watchful to make sure that either the experimentalist himself or the theorist hasn't biased the presentation of the data. Very few people fake data, but very many people bias the presentation to bring out the points of the particular interpretation they make. Always my advice to young theorists is always don't look at what the experimentalist says about his data. Look at the actual points. Look at the actual experimental data. And then you'll, more than half the time, or a very appreciable fraction of the time, you will find that the points on his graphs don't say what the experimentalists says they say. They do not. They're interpretable in some entirely different way.