Suppose a linear predictor of the form $a + b'X$. To find estimators for a and b, should we minimize $E[Y-a-b'X]^2$ or $E[(Y-a-b'X)^2|X]$.

Former gives $\hat{a} = E[Y] - b'E[X]$ and latter gives $\hat{a} = E[Y|X] - b'X$


1 Answer 1


You most likely want the former.

  • You probably don't want your $\hat{a}$ to depend on the value of $X$.
  • Even if you do want a dependence between the two, computing $E[Y|X]$ is something you would have to model again (possibly using a linear estimate) so you probably haven't gained much.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.