Is education a process by which individuals acquire useful skills, or one whereby individuals who already possess skills set themselves apart from others? In theory, the distinction between these two visions of the education process carry radically different implications for whether education is an enterprise worthy of public subsidy. While it’s hard to argue that either theory is completely false, much ink has been spilt in the tussle over whether one or the other predominates. Prior estimates of the proportion of returns to education attributable to signaling range from 10% to 80%.
In practice, argues Nick Huntington-Klein, these models are so difficult to disentangle from one another that there’s no point trying.
Huntington-Klein’s persuasive if somewhat depressing argument begins with a basic mediating variables framework: if there is a causal link between education and individual outcomes, signaling and human capital theories suggest that there are different mechanisms explaining the link. We might imagine a form of “horse race” where the econometrician assesses whether the apparent returns to education are better explained by mediating variables associated with signaling models or human capital models.
The problem with the “horse race” is that it’s hard to establish a one-to-one correspondence between the horses (mediating variables) and underlying models. Huntington-Klein provides examples such as “sheepskin effects,” or the tendency for returns to education to peak in years when a student earns a degree. Sheepskin effects are often associated with signaling models, by the argument that there can’t really be a spike in human capital accumulation in the year a degree is earned.
There could be, though. Those eclectic elective courses you take toward the end of a degree program might contribute more to skills than the intro courses you take early on. Or, as Huntington-Klein points out, there could be a pattern of selective attrition from education programs, such that the only ones who complete a degree are the ones who managed to learn something along the way. Which seems a quite plausible story.
The root of the problem, argues Huntington-Klein, lies in the incredible flexibility of economic theory, which renders both models entirely or nearly unfalsifiable. It’s hard to imagine a “natural experiment” that cleanly manipulates a mediating variable that has no conceivable importance in either a signaling or human capital model. Structural estimation approaches the problem by winnowing away flexibility in the model, introducing concerns that results reflect modeling assumptions rather than reality.
Huntington-Klein’s work takes on an unusual style for a modern economics paper. There is nothing in the way of formalized theory, directed acyclic graphs being the closest we get to equations or optimization problems. And there is no new empirical work here. This is a synthetic work, a literature review citing some 88 theoretical and empirical contributions to labor and the economics of education over a half century or more. While this could make it a tough article to place in a contemporary economics journal, it makes for a refreshing and thought-provoking read. It would be a very reasonable assignment for a graduate or even undergraduate course in labor or the economics of education.
My read of the paper suggested one main avenue for improvement: paying more attention to the critical question of what outcome variable we should be using in measuring returns. Labor earnings are common; Huntington-Klein cites one study where returns to teacher training are measured by the teacher’s ability to improve student test scores, which is common in that literature. Should we be studying life satisfaction? Measures of health? Of course, complicating the nature of the dependent variable does nothing to simplify the conundrum of ambiguous mediators.
In the end, Huntington-Klein suggests we shouldn’t worry so much about the impossibility of distinguishing signaling from human capital models. We should instead, he argues, focus on the question of whether education generates significant external returns, a possibility most commonly associated with human capital models but not entirely inconsistent with signaling. Were we all to agree that the measurement of external returns should be our quest, we’re still left with important questions. External returns using what outcome variable? If measured structurally, how do address concerns that modeling assumptions drive the results? If measured in natural experiments, how do we think about generalizability?
Nick Huntington-Klein has put serious effort into thinking about this question. His work, like most influential work, illustrates how much more thinking needs to be done.
This review was written by Jacob Vigdor, Professor of Public Policy and Governance at the University of Washington. Vigdor has previously received compensation as an expert witness in a school finance trial, where issues of subsidization of education figure prominently. He has no other personal or professional conflicts of interest to disclose.
Nick Huntington-Klein requested a review of this manuscript on January 21, 2019. He declared no conflicts of interest.
Comments on this post are welcome and will be moderated; commenters should identify themselves and declare any personal or professional conflicts of interest in the text of their comment.