I have often wondered at linearity problems. E.g., why is the BMI formula based on weight divided by a square of height when geometry says that weight is propo

I have often wondered at linearity problems. E.g., why is the BMI formula based on weight divided by a square of height when geometry says that weight is proportional to the cube of (similarly proportioned) height? Probably, in the region of interest, the one curve approximates the other, but why fight locally Euclidean fidelity? Feedback mechanisms are common; unless you look for the correct function (whether polynomial, exponential, factorial, etc.), as implied by the underlying mechanism, throwing regression at the problem is going to obfuscate.