In the early 90s, I saw Ken Arrow informally debate Murray Rothbard.  Arrow was not impressive; all he did was repeat tired textbook arguments about market failure.  My subsequent encounters with Arrow’s thought were no better.  Early this year, however, I read Arrow’s 1973 “Higher Education as a Filter,” and decided I was completely wrong about the man.  He wasn’t just a pioneer of the mathematics of the signaling model; he foresaw and answered many misinterpretations of the model that still plague us today.

The highlights begin on page one, when Arrow describes the many ways that education could increase productivity – and, by implication, the many traits that education could signal:

From the viewpoint of formal theory, it does not matter how the student’s productivity is increased, but implicitly it is assumed that the student receives cognitive skills through his education. Educators on the other hand, have long felt that the activity of education is a process of socialization; the latent content of the process, the acquisition of skills such as the carrying out of assigned tasks, getting along with others, regularity, punctuality, and the like, being at least as important as the manifest objectives of conveying information… [F]rom the viewpoint of economic theory, the socialization hypothesis is just as much a human capital theory as the cognitive skill acquisition hypothesis. Both hypotheses imply that education supplies skills that lead to higher productivity.

Only then does Arrow introduce his alternative:

I would like to present a very different view. Higher education, in this model, contributes in no way to superior economic performance; it increases neither cognition nor socialization. Instead, higher education serves as a screening device, in that it sorts out individuals of differing abilities, thereby conveying information to the purchasers of labor.

On page two, Arrow immediately disavows the silly “100% signaling” story so popular with critics of the model, then acknowledges the consumption benefits of education:

Perhaps I should make clear that I personally do not believe that higher education performs only a screening purpose. Clearly professional schools impart real skills valued in the market and so do undergraduate courses in the sciences. The case is considerably less clear with regard to the bulk of liberal arts courses. But in any case I think it better to make a dramatic and one-sided presentation of the screening model in order to develop it than to produce a premature synthesis. It should also be understood that I am speaking only about the contribution of higher education to production; the consumption aspects are real and important, but they are irrelevant to the points being made here.

On page three, Arrow considers and rejects the “employer learning” critique of the model:

It will probably be argued that this description [that “the buyer has very good statistical information but nothing more”] is valid enough at the time of hiring but that after a period of time the employer will know his workers and their productivities on an individual basis. No doubt there is something to this viewpoint but not as much as may be thought. After all, what is needed for allocative efficiency is the marginal productivity of each individual. But in a complex production process, the employer has simply no way of determining that. All he can do is act like an ideal econometrician, relating his output to the numbers of different kinds of workers (and other inputs, from which I am abstracting in this

In the conclusion, Arrow specifically rejects the view that IQ tests are a sufficient statistic for the “ability” that educational success supposedly signals to employers.  He admits that this is unsatisfying, but prefers to be vaguely right than clearly wrong:

[T]he model of this paper depends upon an unmeasured and unmeasurable variable, “ability”. There may be no way of ever achieving a direct measurement; after all, a premise of the model is that employers cannot measure ability directly, and there is no reason to suppose that the economist is going to do better. It remains to be seen if the theory can be made to yield interesting and testable implications in the absence of direct measurements of ability.

He even anticipates my point that in a signaling model, national educational success remains a symptom – though not a cause – of economic success:

[T]he filter model has some implications for macroeconomic observations. It says that an
increase in the resources devoted to college education will have no positive effect on output in the non-educational sector, if all other variables are controlled for. This is indeed a strong inference, but its usefulness in making intertemporal or international comparisons is limited by the need to hold the statistical distribution of ability constant. If “ability” is influenced by cultural factors, then it will certainly vary internationally and may also be thought to vary over time.

The signaling model of education still needs improvement.  I wouldn’t be writing a book about the topic if I thought otherwise.  What “Higher Education as a Signal” shows, however, is that the model hit the ground running.  Almost four decades ago, Ken Arrow foresaw and answered many of the objections that human capital extremists have never stopped repeating.