Why Data People Should Think Like Physicists

Physicists run experiments so they can create and refine simple heuristics that summarize their understanding of the world. They don’t just say, “acceleration varies significantly with mass (p<.01).” They build a model: F = MA. They test the model too, which makes it predictive.

You can object that physicists work in a special domain, amenable to precise experimentation. But that’s beside the point. Even F = MA is not a perfectly accurate model: it ignores lots of messy details, like friction and time dilation, and therefore doesn’t make perfect predictions. But it isolates and describes a real-world relationship. That’s how physicists can go beyond individual experiments, and lots of data, to understanding. We can all emulate this.

How? Don’t let data do all the talking: let your own understanding of the situation have equal input. Formalize that understanding with an equation, or sketches, or verbal relationships. Then look at the data to see if you’re right; revise if necessary. Think about:

  • What values are even remotely conceivable? Which aren’t?
  • What kind of effect should one variable have on another? Will it logically increase it or decrease? Will this change be additive, or multiplicative?
  • How will one effect compare, in magnitude, to other effects in the same situation? Which are the largest? Which are so small that they barely matter?

Then you can start to go beyond describing the cases that currently exist in the world, and begin to describe what’s going on to create them.