Go clean

Liam Julian

Noam Scheiber, senior editor at The New Republic, is none too pleased about what he calls the "cleverness problem" bedeviling top economic graduate schools. According to him, today's students and professors are far less interested in using the dismal science to investigate important issues (poverty, inequality, etc.) than in finding cute, clever ways to show causation between situations that have little or no practical consequence.

Demonstrating compelling causation--called clean identifying--is the holy grail of economic studies. And Scheiber isn't necessarily against it. He thinks clean identifying is well and good when applied to areas that deserve study (he gives as an example this paper about the correlation between education and future wages). But he also thinks the clean identification fetish for showing causation between unimportant, everyday occurrences has gone too far. Freakonomics might be to blame.  

He's correct in a sense. The economics that garners headlines is that of the "cute" variety. Does it really matter if diplomats' parking tickets are correlated to their country's level of corruption (see here), or that Mexican men pay prostitutes a premium for unprotected sex (see here)? Economists are supposed to be solving problems, not noting largely worthless causations. 

But in another sense, Scheiber gets it wrong, as noted by MIT economics professor Joshua Angrist. Angrist, who coauthored the aforementioned study of education's correlation to earnings, writes that he is "especially pleased when students manage to come up with clean identification." He continues: "Clean identification is not a fetish; without it, little of value is learned."

That's especially true in education, an area that has too long been plagued by studies with shoddy methodologies. (Still true, alas; see here and here.) But a substantial push is underway to make education studies more rigorous, and thus, make their results more useful and conclusive. 

Randomized trials are the gold standard for controlling causality, and often they allow for clean identification. Some education problems, however, cannot be researched using randomized trials (see here). When Angrist and coauthor Alan Krueger decided to examine whether going to school longer increased future wages, they couldn't very well compare the salaries of Ph.Ds to those of dropouts--there's no way to establish causation with such a coarse comparison. But nor could they devise a randomized trial in which they selected a group of high school students, forced half to drop out and the other to attend college and grad schools. Besides being impossible, it would be unethical.

So they had to look for a "clever" way in which natural divisions would set up the functional equivalent of a randomized trial. What they found was that, because of school attendance rules, if student A was born in January and student B in December of the same year, and both dropped out as 16-year-olds, student B was in school nearly a year longer than student A. Thus, a natural, randomized trial could be studied. And the study was clever, if not cute.

If an urge to dig for natural clean identification yields more such studies in education, and if it raises the standards of education research as a whole, so much the better. Moreover, as Freakonomics demonstrated, clever clean identification piques people's interests. And if more Americans become interested in education policy, we're sure that bodes well for reformers.

Should we care whether drug dealers still live with their mothers? If it will eventually help us build better schools, absolutely.