I have a thought about the new econ/epi debate spurred by my colleague @AnupamBJena's recent NY times op-ed. Thought comes from my last few years as econ faculty in a soft-money department at a med school, a similar environment to that of most epis but different from most econs
Let's get one thing out of the way quickly here: While both econ and epi have some natural experiments, the *representative* (applied micro) econ paper uses a natural experiment and the *representative* epi paper does not.
So the question is, why? Some have taken offense at Bapu's op-ed because they seem to have felt he was telling epis that they're dumb. That's not my interpretation of either reality or Bapu's view.
My interpretation of how things look on the ground is that the two fields have developed differently due to different incentive structures. More horizontal differentiation rather than vertical.
In econ, almost all observational papers without a clean natural experiment are basically unpublishable at the small set of journals that actually count toward tenure and toward respect in the field.
This has led to econs focusing heavily on this type of work, at the cost of (potentially valuable) other types of work. Such highly incentivized specialization seems pretty likely to lead econs to be better at this type of work than people in other fields due to 2 factors:
(1) selection into econ - people who are good at identifying and implementing quasi-experimental studies are potentially more likely to select into econ vs. other fields
(2) technology - econs focus much more heavily on natural quasi-experiments than other fields, so in time they get better at those types of studies, both individually and as a field. See all the development around RD, DiD, IV that have come out of econ.
In epi, things are quite different. Many epis are in soft money departments, so getting grants and publishing lots of papers are the currency of the field.
This results in a very different set of incentives. From my own experience, when you write a grant that says you're going to study X, you have to write a paper studying X, EVEN IF YOU CAN'T FIND A NATURAL EXPERIMENT.
Thus, epis often have to write the paper even if there's no natural experiment. Thus, epi methods have developed to *do the best you can* in the absence of a natural experiment/RCT. Again, this affects epi methods in 2 ways
(1) selection into epi - people select into epi who are interested in/good at answering questions, no matter whether you have a nat exp/RCT
(2) technology - epis have to answer the question no matter what, so in time they focus much of their methods development on that
(2) technology - epis have to answer the question no matter what, so in time they focus much of their methods development on that
Each approach has advantages and disadvantages:
Econ - Uses really cool quasi-experimental methods that the gen public understand and can interpret easily w/o any understanding of "selection on unobservables".
Econ - Uses really cool quasi-experimental methods that the gen public understand and can interpret easily w/o any understanding of "selection on unobservables".
BUT econs risk missing out on important questions where there is no good quasi-experiment.
Epi - Can provide "best possible" answers to many questions. Policymakers often need to know the "best evidence" available, and there is only evidence if someone actually writes the paper, which epi always does and econ sometimes doesn't.
BUT epi risks the possibility that they miss great natural experiments because they actually CAN publish a paper without one so they aren't as incentivized as econs to hunt down that elusive clean quasi-experiment.
(epis also have to adhere to grant timelines which forces them to move faster and move on to the next fund-able topic rather than keep hunting for that quasi-exp)
Basically, this is a long way of saying that the two fields have specialized the way that they were incentivized to specialize. Both equilibrium outcomes have their advantages and disadvantages, and we can probably learn a lot from each other.
These outcomes are not driven by one field being inherently "smarter" or "better" than the other, I would call this horizontal differentiation rather than vertical.
Ultimately, the primary implication of this theory is that it's not at all clear that one field *should* look more like the other. It also seems clear that if one wants to shift one field toward the other, it's the incentives that need to change, not the people. [END]