One day workshop on Instrumental variables
Taught by Dr. Peter Hull
Instrumental variables was invented, discovered, or dreamed up by Philip Wright, an early 20th century economist. Philip definitely has one of the more intriguing vitas of an economist I’ve ever seen. James Stock found it and posted it online. He published such luminaries as:
The Cannibal Converts: A Human Roast in Three Cuts, which according to what I could find online was an opera
a lot of poetry
he published Carl Sandburg’s very first book of poems, which is mind blowing
and 9 Top 5 publications (QJE, AER and JPE)
He also wrote some books, one of which was on tariffs and animal oils, and in the appendix, he lays a proof of sorts using directed acyclic graphs (DAGs) of all things of an early IV estimator. The Cowles Commission would around the same time also invent/discover/dream up the IV estimator, and it’s wild to think that IV was “in the air” like that. It wouldn’t be the first time scientists at the same time independently figure out the same thing, more or less. After all, with enough trials, anything that can happen will happen no matter how improbable it may seem.
The instrumental variables method is neat because economists actually dreamed it up. Regression discontinuity, for instance, was the brain child of Donald Campbell, an educational psychologist working in the 1960s. It wouldn’t be until Josh Angrist, Victor Lavy and Sandra Black re-discovered RDD in their original research (Angrist and Lavy used it for a fuzzy IV paper on class sizes, and Sandy for a study evaluating the impact of desirable schools on housing prices using a kind of spatial fixed effects design) that we saw that design become “born again”.
Difference-in-differences became popular because of the work of the Princeton labor economists in the 1970s and 1980s, like Orley Ashenfelter, David Card and Alan Krueger. Card coins the term at some point in the mid-1980s. But of course, Princeton did not create difference-in-differences — that goes back at least 130 years to John Snow, that wild eyed anesthesiologist who during a raging cholera epidemic fruitlessly argues with policy makers and scientists that the disease was coming from the river, and not the smelly air engulfing London.
A lot of the work that economists do was born in another land. We, being a curious group, noticed and obsessively tinkered with them until our fingerprints covered them. But IV? IV was born in the brain lab of economists, and for that reason, I tend to love it a little more than the others. At the end of the day, I am and always will be an economist.
But instrumental variables is a maligned, much hated class of estimators when you think about it — secondly only to the propensity score in fact. How we love to eat our young. I recently had a funny joke I told myself — so funny I will tell it again. Instrumental variables is the Justin Bieber of econometrics: it is tragically misunderstood, falsely maligned by everyone, and yet arguably the greatest of its generation and will be remembered fondly for hundreds if not thousands of years. Bet me on it!
And so it is only fitting that I have managed to persuade one of the best and the brightest young applied econometricians I know to teach a workshop on instrumental variables at my “Mixtape Sessions” platform — Peter Hull. Peter (seen below), is a wonderful person I have gotten to know on Twitter, as well as someone I have occasionally played putt putt golf with on a Virtual Reality headset called Oculus Rift 2. Never in a million years would I have thought any of those words would come out of my mouth. Here’s a picture of him smiling.
Peter is an exceptional mind. He is an assistant professor of economics at Brown University. Prior to that, he was an assistant professor of University of Chicago. And prior to that, he was a PhD student at MIT and an advisee of recent Nobel Prize winner, Josh Angrist. Peter, along with a small group of young econometricians like Michal Kolesar, Paul Goldsmith-Pinkham, and Kirill Borusyak, as well as applied microeconomists like Will Dobbie, Crystal Yang and many others have been refocusing our attention to what I call “sub IV designs” like the leniency design and the shift-share design, deepening our understanding both of these particular applications of IV, but also helping us to better understand the class of estimators themselves. It is a fun time for those of us who find econometrics fun, funny and interesting.
And so, why am I writing all this to you? I am writing this to tell you that over at Mixtape Sessions — a platform I created in order “to democratize causal inference” by offering workshops over Zoom and Discord — Peter will be teaching a one day workshop on contemporary IV on March 12th. And I would like to give you some information about the workshop. Here is Peter’s description:
Instrumental variables (IV) is a powerful tool for leveraging external (“exogenous”) variation to estimate the causal effects of otherwise confounded (“endogenous”) variables. This one-day workshop will introduce the basics of IV through different practical examples, formalize the requirements of a valid and powerful IV, and discuss the mechanics of the two-stage least squares (2SLS) estimator. Special focus will be paid on interpreting linear IV under heterogeneous treatment effects and recent advances in judge leniency designs, shift-share IV, and more. The course will include substantial group programming exercises, where different IV techniques will be illustrated in real-world applications.
He gives us a little taste of what we might expect by listing a few topics. They are:
Regression Review and Regression Endogeneity
Introduction to IV
Understanding Instrument Validity; 2SLS Mechanics; Applications
Coding Lab
Heterogeneous Treatment Effects; Characterizing Compliers; MTEs
Judge Leniency Designs; Shift-Share IV; New IV Frontiers
Coding Lab
Mixtape Sessions uses a simple price discrimination model in order to democratize causal inference. If you are an enrolled student, predoc RA, postdoc, or a resident of a low or middle income country, then the class is only $50 plus fees (I use eventbrite to do all this and they charge fees). For everyone else, be it faculty, employee of the government or in industry, the price is $595 plus fees.
So, is this worth it? In my opinion, yes. I think even at $595, the consumer surplus is positive, but it is definitely positive for anyone who fits that student description. IV is a powerful tool. It can recover local average treatment effects, and using the tools of marginal treatment effects, can go even further than that to more general policy relevant treatment parameters. The increased digitization of all of human life has caused an explosion in data opportunities, but it has also caused an explosion in naturally occurring variation that can easily be used to identify causal effects from observational data. The IV class of estimators is and will always be at the front of the line of being well equipped to help us learn about the world. You just need to know what it is, how to use it, and how to learn from it. Peter Hull will for one day help you do that.
It is not every day that most of us have the opportunity to study econometrics with a professor at Brown, a former MIT student who studied under Nobel Laureate Josh Angrist, a researcher who has already stacked up multiple high profile, high impact important papers in both econometric theory as well as applied work more generally. It is not everyday. But COVID has pushed education into new modalities, over Zoom and through slack channels, and we have all learned and continue to learn that while not perfect, online education can work for the highly motivated student and teacher. Mixtape Sessions is a matching function that bridges students to econometricians like Peter by scaling up their teaching, making them feel closer to us, and answering our questions. I do not think you will regret taking a class on IV from Peter Hull. So I highly encourage you to walk over to the website, and register.
If you need a discount code because you are a student, predoc, postdoc, or resident of low and middle income country, please email me at causalinf@mixtape.consulting. Just send me a screen shot of your photo ID with a description of where you are a student or your residence, and I’ll send you the code. And if you want to take another class, perhaps, on causal inference (March) or difference-in-differences and synthetic control (April), just ask about those too, because I will also be teaching courses on those again in the spring (the first two filled up). So enjoy your winter break. I am hoping to start back up the substack explainers in the new year soon
As the rare person to openly write in favor of price discrimination, thank you!