[mcguire]

Exactly. With a causal model (which can be validated independently) you have a principled reason for choosing which variables to control for.