Julia Seither

@UChi_Economics we have had incoming PhD students visit us over the past two days. Lots of great discussion and questions. One question that I really enjoyed was: what does every graduate sequence miss when it comes to training economists? In light of the fact that we live in an era of accelerating AI and automated analysis, I felt that my traditional response to this type of question has grown even stronger today: Experimental Economics belongs in every graduate sequence. Let me make my case. Many PhD programs teach econometrics as the art of extracting signal from data that already exists. That is great stuff. Yet, especially with the recent AI push, it does permit a certain dangerous habit of mind. How? Because students learn to work with the data that we have rather than asking what does an optimal data set look like? Experimental economics corrects this at the root. Designing an experiment, even hypothetically, forces a precision that reading dozens of academic papers or creating an AI prompt never will. If you cannot specify the experiment that would cleanly test your theory, then your theory may not be as sharp as you think. Experimental design is a lie detector for vague ideas. But the argument doesn't stop with theory-testing. Consider the applied scientist who wants to change the world. Their question is slightly different: what data can I generate to provide evidence that is actionable? In both cases, whether testing theory or informing optimal actions, the best experiment delivers the same two key insights: it measures whether it works and for whom (the moderators) and provides mechanisms explaining why and how it works (the mediators). A treatment effect without a mechanism is a black box. A mechanism without heterogeneity analysis is a fiction about an average person who doesn't exist. Teaching students to design experiments teaches them to demand both, always. This leads to a third, underappreciated, benefit. And, this is where the case becomes even stronger. The student who understands the value of controlling the assignment mechanism also understands exactly what assumptions their regression is making. Experimental thinking doesn't serve to replace, rather to discipline. Of course, a well-designed experiment not only explores does it work here, but asks will it work there, for them, at the scale required to matter? Experimental economics forces students to think about external validity not as an afterthought but as a design input. Such questions are under deep consideration from the beginning. That habit of mind is what we strive for in our experimental training: when fast thinking is promoted via new technologies, we need our best thinkers to slow down a bit. We teach students to be sophisticated consumers and producers of econometric methods. We should teach them with equal seriousness to be sophisticated designers of the evidence those methods analyze.

🌟 New NOVAFRICA Working Paper! "From Destination to Origin: Experimental Evidence on the International Spillovers of Migrant Integration" By @cbatista_econ @NOVAFRICA @NovaSBE , @LaraBohnet @NewYorkUni , #JulesGazeaud @CNRS & @juliaseither @UofC 🔗 tinyurl.com/mrye8mta