Can Four Parameters Really Model an Elephant?

[Whirlwindx]

There is a famous saying in physics: "With four parameters I can fit an elephant and with five I can make him wiggle his trunk."

Firstly, does anyone know who definitely said it?

Secondly, what exactly does it mean in Layman's terms? I have a rough idea, which goes as follows:

Researcher: I have a mathematical model, and with the parameters it has it fits the publishes data very well.

Supervisor: Your model fits the data. So what? With 4 parameters I can fit an elephant...

 

[lpfr]

There is a similar version: There is always a straight line that passes by three points.

All this means that accepting experimental errors, and/or statistical fluctuations you can always fit any rotten model to the experimental data (sometimes also rotten).

Better still, always introduce adjustable parameters in your model. This will help to fit experimental data.

If you have scruples you can always do a test like "Student test" or "\chi^2" that always remove all scruples.

 

[Meir Achuz]

mahalanobis.twoday.net/stories/264091/

 

[cesiumfrog]

Better still, always introduce adjustable parameters in your model. This will help to fit experimental data.

Uh.. sarcasm? I think the point of the story is to mock the fact that if *any* model has "lots" of unknown parameters, those parameters can be adjusted to fit absolutely whatever data you already have (even data not corresponding to what the model was intended for). Hence, models with lots of adjustable parameters aren't very useful in either understanding the underlying physics or predicting future results; a good model has very few parameters (and preferably, even those can be independently measured rather than being arbitrarily adjusted), so at the very least, any good model can be proven false (no elephant would fit to a good model of a skateboard).