IV with Endogenous Controls

Applied papers that rely on instrumental variables are often focused solely on a single parameter of interest. This parameter is the coefficient on an endogenous variable. Using an IV for this variable, one obtains the IV estimate of this coefficient. But what about all those other "exogenous" vars in the model?

 Usually the coefficient estimates are ignored because they are not "of interest." What if one of those exogenous control variable is really endogenous as well? The typical response: "I don't care about that coefficient, so it doesn't matter." But does it matter? Um, perhaps yes!

Let's say the true model is

Y=a+b1*x1+b2*x2+e

 and both x1 and x2 are endogenous. You only care about b1. You have an IV for x1, call it z. For z to be valid, we know it must be correlated with x1 and uncorrelated with e. But ... it also needs to be (in all likelihood) uncorrelated with x2!

 If your IV is correlated with an endogenous regressor incorrectly treated as exogenous, then your IV estimate of b1 will likely be inconsistent (remember, IV is always biased). So, if z and x2 are correlated, IV is unlikely to work!


Consider the following simulations. The DGP has 2 controls, both "equally" endogenous and uncorrelated with each other. Z is correlated with x1 and not with e. Everything is standard normal and the true value of b1=1.

First set of results (based on 250,000 sims) has Corr(z,x2)=0. IV is consistent even though x2 is incorrectly treated as exo.


Now, let's have z be correlated with x2 as well (albeit less strongly than z is with x1). IV is inconsistent (although less so than OLS in this particular DGP).


Uh oh. This is not looking good. What if z is only weakly correlated with x1, but still correlated with x2? Then, you might want to look away.


 So, do NOT ignore the endogeneity of controls just because you don't care about them.
One final thought. At this point, you might be thinking you can game the system by just dropping x2 from the model since you don't care about it anyway.

If you omit x2 from the model and it belongs, then the error term becomes e'=(e+b2*x2). Since x2 is part of the error term, z is no longer a valid instrument since it is likely correlated with e' through x2. And, once again, that's why should not ask how the sausage is made.

Note: Code is available here: http://faculty.smu.edu/millimet/blog.html

Popular posts from this blog

The Great Divide

Image
I applaud the fact that the American Economic Association, the European Economic Association, the Econometric Society, and the Royal Economic Society have established an ad-hoc joint committee tasked with proposing ways to improve the publication process in economics. I encourage everyone to fill out the survey  (limited to members of the AEA, EEA, ES, and RES, however) even though I am quite cynical about what may come of it. Although only tangentially related, I decided to write a post about a topic that vexes me (and others to whom I have spoken, but I will leave it to them to out themselves) at the intersection of the Venn diagram containing publishing (editing and refereeing), teaching, and econometrics.   What lies at this intersection is the great divide between theoretical and applied econometrics, both in the classroom and in the wild.  As I have progressed in my career, I have enjoyed writing more econometrics papers and fewer economics papers per ...
Read more

There is Exogeneity, and Then There is Strict Exogeneity

Image
Panel data has become quite abundant. As a result, fixed effects models have become prevalent in applied research. But, this week I handled a paper at one of the journals for which I am on the editorial board and was reminded of a mistake that I have seen all too frequently over the years. Most researchers are probably aware of the issue I am highlighting, but indulge me. In a cross-sectional regression model, we have y_i = a + b*x_i + e_i OLS is unbiased if x is exogenous , which requires that Cov(x_i,e_i) = 0. In other words, the covariates need to be uncorrelated with the error term from the same time period. In contrast, in a panel regression model with individual effects, we have y_it = a_i + b*x_it + e_it If we wish to allow for the possibility that Cov(a_i,x_it) differs from zero, then a_i is a "fixed" effect instead of a "random" effect. To understand what we need to assume to obtain unbiased estimates of b, we need to understand how the model i...
Read more

Black Magic

Image
The world is a very different place from when I started this blog. The pandemic is over ... but I'm not sure we are in a better place. Different, not necessarily better. This brave new world has made it hard for me to find the time and mental energy to continue this blog. Moreover, with the changes at #EconTwitter ( X ) , no one may read this anyway. Nonetheless, I move forward and we shall see if this post (and any future posts) provide utility to anyone beyond myself.  So, I have spent some time thus far this summer revising my lecture notes (available here  ... although it is still in progress for some). I still stand firm in my belief that all empirical researchers ought to teach econometrics (if for no other reason as a constant reminder that there is more to the field than diff-in-diff). As I did so, I added two new extensions to my discussion of -- what else -- measurement error. But, this post is not about measurement error per se. Really. No, I mean it. Instead, it is...
Read more