Question on endogeneity and bias in OLS when the independent variables are causally related

[Leon Schmidt]

Dear Statalist,

My question does not concern Stata specifically but rather the estimation procedure of linear regression in general. Please forgive me for posting it here as I did not know of another place to ask it.

I read in the SAGE Encyclopedia of Social Science Research Methods that the independent variables are said to be partly endogenous when they influence each other. Their example was the regression of crop output on variables such as pest level, fertilizers, etc since within this system the pest level is at least partly endogenous as it is influenced by the fertilizers used.

My question is: If I estimate such an equation will this problem bias the OLS estimates in the same way as the more general endogeneity problem (omitted variables, measurement errors, reverse causation)? Do you know of any book where to read up on that?

Thank you very much.

All the best,
Leon

[Marcos Almeida]

If you have the latest release (Stata 15). you may start by reading the "Extended Regression Models" manual.

[Phil Bromiley]

In general, the x's can be related as long as they're not correlated with the error term. This is why we can impose x values in experiments. It may be that some mixes of values of the x's don't make sense in margins. Alternatively, it may be you want to look into endogeneity, look at the extended regression and ivreg documentation.

[Leon Schmidt]

Thank you both very much for your feedback!