What is it about statistics?

Courtesy of Arts & Letters Daily, as is often the case around here, a review by Francis Fukuyama of a new biography of Friedrich von Hayek. Hayek was a brilliant man whose insights about spontaneous order in complex systems turned out to be as useful in neuroscience as they are in economics. However, Prof. Fukuyama manages to turn a discussion of Hayek's work into an attack on mathematical models and "deterministic, predictive outcomes" (SC thinks he meant "predictable", but no matter).

Now, SC is not a corpus fetishist, and tries not to subscribe to any corollary dogmatism about the singular acceptability of statistical methods. And he could even toss in some hoary cliches about "horses for courses". It's important to step back and try to retain a sense of why statistics are useful, especially in the social sciences. But what to make of this?

Caldwell, an economic historian at the University of North Carolina at Greensboro, ends his book by plaintively noting that the un-Hayekian agenda of turning economics into a rigorous science has driven all other approaches, including the study of economic history, out of American economics departments. But the damage done by this positivist approach is, in fact, much greater. Economic methodology has colonized political science too, eliminating individuals with knowledge of real peoples, cultures, and history—for example, experts on the Middle East—from the country’s top schools. We are thus presented with a rather depressing picture of human progress. Although the particular brand of intellectual hubris that elevated central planning over markets is gone, other forms persist, and indeed have grown stronger. Hayek’s challenge remains an open one.

Surely there are very few practitioners of economics who would boast of following a "non-rigorous" approach to the field. SC will go further and hazard a guess that "knowledge of real peoples, cultures, and history" is not incompatible with attempting to be both systematic and empirically grounded.

This sort of talk is by no means limited to economics -- just this morning, SC had an identical discussion with a colleague on statistical vs. empirical methods in artificial intelligence. At the core of objections to statistical methods are valid concerns that statistics need a framework for interpretation, and that methodological pluralism has generally been a good thing in the sciences, social or physical. And perhaps at any one moment in time, the ascendancy of a particular research method can be accompanied by both enough success and hubris that it appears destined to blot out everything else, obscuring forever what lies outside current orthodoxy. A fair argument can be made that theoretical linguistics is just now recovering from such an episode. Such critiques need to be careful, though, to avoid falling into self-parody. Appreciating the limits of statistical methods is a good thing. Denouncing rigor is not.

UPDATE: It's not just the sciences. Even baseball pitch counts can provoke vehement disagreement.

(Edited at 9:16 a.m. on 5/10/04)

Comments

[language hat]

Jim Duquette, whose Mets blew out their promising Wilson-Pulsipher-Isringhausen threesome a decade ago...

Thanks for the painful memories, pal. Hmph. Aside from that, however, an excellent article, emphasizing a point that's hard to overemphasize, especially since it's rarely even made: those old-timers had it easier because there were a lot more mediocre ballplayers, so pitchers could feast on bad hitters and vice versa. Now the average quality is far higher, which means (along with no more .400 hitters) each pitch is more difficult and pitchers burn out earlier. Some managers (hello, Whitey!) never learned this and burned out generations of young arms. I think the obsession with pitch counts is often overdone, but on the whole it's a Good Thing.