💡 Today I Learned

All Models are Wrong, Some Models are Useful

The saying

All models are wrong, some are useful

...is from British statistician, George Box

What does it mean?

When you set out to "model" or "simulate" some system or process in the real world, your model will start off as a drastic oversimplification of the thing you are modelling.

Then you may refine the model and it may be a little more similar to the thing you are modelling.

Stop right there.

The goal of modelling is to get useful results. It is not to make a perfect recreation of the thing that is being modelled.

If you attempt to make a perfect model, you will never achieve it. You could spend your life trying to make a "perfect" model of the interactions between the atoms within a single grain of sand.

Only two things are certain about your modelling effort:

  1. You will die before you correct the errors in the model
  2. That is all we know.

So the goal must never be to make a perfect "thing" -- but rather to stop when the thing is useful, usable. Ideally we stop before we reach "the point of diminishing returns." If we find we have blown right past the point of diminishing returns, then stop right now.

Extrapolating the rule everywhere

The rule doesn't just apply to a statistician's attempt at modelling a process. I'm happy to extend it to every other domain of life...

e.g.

(Never listen to anyone who gives you advice, particularly me. (See also Paradox))

References

See also