Thanks Roy for posing a very stimulating question and putting the forum on a very fruitful trajectory.

Clearly, as Roy and David Colander have argued, there have been some important changes in the profession over the last 25 years or so which have led academic economics departments to place far more emphasis on applied micro.  However, I think we are missing two factors that work to temper the apparent changes we observe.  Moreover, each of these two factors provide (relatively) comfortable entry points for historians of economics to start thinking about contemporary economics.

The first factor is the one that Roger Backhouse has already raised.  Namely, as a group we have tended to write our histories around the theoretical constructs.  As Roy says, hardly any economist today "does theory."  But what he has called a "massive retreat" from theory may only look like one when we've over-emphasized the extent to which economists were ever doing theory.  If we are looking for the historical roots of some of today's applied micro, we may be looking in the wrong places:  not all economists were ever at top-10 economics departments.   Much applied work was done, as we know, in governments and think tanks, but also in university ag econ departments.  Pursuing Roger's agenda of writing our histories around applied issues may provide insights into themes which are continuing today.  (And I recognize that many are doing just that.)

Second, at the same time I think we may be overstating the extent to which theory has now been banished from empirical work.  Without a doubt, it is absolutely striking how, when one picks up a copy of the AER, one is now confronted with work on very narrow questions: the effect of teen alcohol consumption on labor earnings, the labor supply of taxi drivers, the response of charitable giving to matching grants, the optimal airport congestion tax, the behavior on some TV game show or another, and so forth.  But I would disagree with David Colander when he says that most of this is atheoretical empirical work with perhaps a little theory tacked on as an afterthought.  Rather, I think there remains a broad spectrum for the roles given to "theory" and measurement."

Really, embedded in this literature is yet another renewal of the old "measurement without theory debate," this time in the heart of applied micro.  On the "measurement" side we have the kind of reduced form work associated with "program evaluation," e.g. the labor economics of J. Angrist and G. Imbens (and D. Rubin outside economics) -- the work that David was highlighting.  But on the "theory" side we still have the structural modelers.  While this is empirical work, its practitioners are quite proud of the way the econometric model is wedded to an economic model of dynamic optimization and/or equilibrium.  The model generates properties for, say, a Nash equilibrium, which in turn are written as moment conditions that can be taken to the data.  This line of work is probably most associated with empirical IO (see e.g. the work by S. Berry or V. Aguirregabiria) but also is represented in labor economics (e.g. work by J. Rust) or public economics (e.g. the work by D. Epple or Roy's junior colleagues).  [And of course, there are the similar developments in macro that Kevin Hoover and others have written about.]  Charges that structural models are "merely" calibration of a model that the economist is imposing, or that the quasi-experimental results have no interesting or relevant counterfactual are levied back and forth.  Little of this debate makes it into the journals, but for one example see the fruitful exchange between Heckman and Angrist & Imbens in the JHR from '97 to '99 and the recent NBER paper by Heckman on "Econometric Causality" available on the web.

Interestingly, while the debate over the roles of measurement and theory live on, it has been reconstituted in ways that make it impossible to think in terms of some sort of stable Cowles-v-NBER debate over time.  What I've called the measurement side is now probably most aligned with the NBER, but the relentless pursuit of the best instrumental variables can also be seen as a continuation of Cowles' one-time agenda.  What I've called the theory side blends the structural econometrics of the early Cowles with a reinterpretation of the term "structure" through the lense of Phil Mirowski's cyborgs from Cowles Mark II, plus a dash of the old NBER-like taste for practical relevance.  (And in all of this, of course, one must be wary of over-generalizing:  today's NBER has fellows representing the entire spectrum and, as Kevin Hoover has pointed out, not everybody in yesterday's Cowles was as adamant as Koopmans about measurement without theory.)  

In short, I think there are some great stories to be told about how some research programs have reconstituted themselves over time.  But I also think a lot of the fundamental building blocks remain familiar to us and that, indeed, we have a lot of insights to bring to bear on contemporary economics and that we needn't be shy to do so.

H. Spencer Banzhaf