Because I’m currently sitting in the same lecture room in Strasbourg as Steve Pischke and yet another paper on labor markets is presented, I feel inspired to comment on the newest Angrist and Pischke piece on econometrics education. Furthermore, my own graduation doesn’t lie too much in the past, so I might still be part of the target group for an improved coursework in quantitative methods.
In general, I’m very positive about their suggestions. Much of the econometric courses you get as an undergrad are horribly outdated. At the beginning it’s still fine. Almost always you start with the simple linear regression model in your introductory class. This is reasonable because it’s still the workhorse model of the profession. And many of the concepts which are also relevant in more advanced models can be easily grasped in the linear case. Moreover, to introduce a bit of theory here cannot hurt. As an undergrad, you might not take away too much for your future life by going through the proof of the Gauss-Markov theorem. But introducing the model thoroughly in matrix notation helps to excel later courses.
I think, the real mess starts when you go further. If you’re interested in macro stuff, you will still see a lot of time series material and unit root testing. In micro, you will hear a lot about things like GLS, Limited Dependent Variable Models, or Simultaneous Equation Models. Even the hugely important instrumental variables techniques are usually presented in an GMM setting. All these model frameworks are useful and you definitely want to understand them if you plan a career in academia. But unfortunately, they often mask the real struggle of, e.g., implementing an instrumental variable method in practice in order to establish a causal relationship between two variables.
Both authors, judging from their two textbooks, seem to pursue an agenda of taking a step back from the vast amount of statistical models to analyze economic data that were developed until now. Then, they revisit what is really useful in practice (and with practice I mean applied research, policy advice or quantitative analysis in industry) and how you can teach “best-practices” of implementation. I like that idea. And the huge success of their first book “Mostly harmless econometrics” (MHE), as a manual (others would say cookbook) for an applied crowd (although theorists hate it), proves them right
Their new book “Mastering Metrics” is even more boiled down, written in a witty and informal style, and only focussing on five core methods which you’ll meet everywhere in micro econometrics. All these methods stand in the so-called experimentalist tradition, trying to emulate an ideal randomized experimental setting with non-experimental data without relying on a specific economic model.
That’s indeed the stuff I’d like to make economics majors understand if they later in their life try to analyze energy markets or want to understand studies about the gender gap or the influence of marijuana consumption on education outcomes. But if you really want to understand what’s going on, you definitely have to put a lot of more effort in it. And, I’m afraid, not only when you want work on theoretical econometrics.
Teaching econometrics without models won’t get you very far. You might understand the exclusion restrictions that are discussed in MHE, but already the next application will leave you puzzled if you don’t know the basics of Simultaneous Equation Models. You also won’t understand why panel data helps to get rid of confounding unobserved heterogeneity but that simple Fixed Effects Estimation breaks down if your past performance matters and you suddenly have to specify a Dynamic Panel Data Model. This list goes on.
There is a lot of debate between the structuralist and experimentalist camp in econometrics. The latter approach is easier to grasp, that is probably why it gained so much traction in the last years. But the struggles that people still have after having read MHE (which is the even more “theory-heavy” of the two books) shows how confusing verbal reasoning can be without having the chance to come back to a (sufficiently abstract) model. To sum up, a more applied focus in econometrics education is definitely desirable. But, as usual…