0
$\begingroup$

enter image description here

if the double arrows show that X and the error term are correlated, but that neither variable affects Y, is endogeneity a problem in this scenario? Why or why not?

$\endgroup$
  • $\begingroup$ Google the formula for omitted variable bias. $\endgroup$ – Dimitriy V. Masterov Feb 8 '18 at 23:33
  • $\begingroup$ Is that figure a formal DAG? $\endgroup$ – Carlos Cinelli Feb 16 '18 at 6:55
0
$\begingroup$

Endogeneity is a problem when you have a regression model

$$y=X\beta+\epsilon$$

and

$$E(\epsilon | X) \neq 0$$

But if you assume that neither $X$ nor $\epsilon$ affects $y$ then the above model is mis-specified. In other words, $y$ is not a function of $X$ or $\epsilon$ but rather of something else. So there is no point in discussing $X$ or $\epsilon$ when talking about $y$.

Endogeneity can only be discussed conditional on the above relationship ($y=X\beta+\epsilon$) being true. Otherwise (for example, if $y=W\gamma+\nu$), then you need to discuss endogeneity in terms of $W$ and $\nu$.

When you say a variable is endogenous, you need to specify "endogeneous with respect to what other variable?". I in case of $y=W\gamma+\nu$, if $E(\nu|W)\neq0$, i.e. $W$ is endogenous, then $W$ is endogenous with respect to $y$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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