WWID: Not growth policy?

In Chapter 6 of What Works in Development, Abhijit Banerjee of JPAL recaps the ineffectual history of research on economic growth and concludes that "it is not clear that the best way to get growth is to do growth policy of any form." Perhaps it would be better, he says, to focus on keeping our house in order - maintaining social stability and building human capital - so that we can take advantage of the growth "spark" when it occurs.

I sympathize with commentator Peter Klenow, a micro-oriented macroeconomist, who responds:

Beyond accounting, macroanalysis has contributed a slew of robust correlations that help guide microexperimental work. Correlation need not imply causality, of course, but certainly does not rule it out... It is up to theorists and microexperimentalists to flesh out the causal mechanisms behind these correlations.

Bill Easterly - editor of the book, and reserving the last word for himself - makes three main points:

1. Macroanalysis established the negative effects of extreme policies and outcomes (e.g. land expropriation and hyperinflation in Zinbabwe); it was only in cases of moderate variation where the conclusions broke down completely.


2. "Macroeconomists have earned their ignorance the hard way," largely "because the demand for explanations of growth was so intense."


3. RCTs "do not really offer a serious alternative to how to achieve 'growth success'", because the small interventions they test in specific situations could never scale up to meaningful increase growth on a macro scale.
   

I think Easterly is being too pessimistic here, and Klenow is closer to the mark - the right answer has to be a middle ground where RCTs are one of many tools in the toolkit, microexperimental work complements and is informed by macroanalysis, and we keep on trying to crack the mystery of growth, counter to Banerjee's skepticism that we ever can.

WWID: An Experimental Approach

In Chapter 2 of What Works in Development, Dani Rodrik (who has sadly stopped blogging) argues that 1), we shouldn't consider RCTs superior to other types of studies, because while they have strong internal validity they often have weak (and unacknowledgedly so) external validity (and other studies can be stronger on external validity), and 2), there is a philosophical convergence on "experimentalism" between micro development economists (e.g. the randomistas) and macro development economists, who now think much more in terms of diagnostics and testing to find binding constraints on economic growth than the previous worldview which came with an a priori vision of what works (e.g. the Washington Consensus).

Then there are two short commentaries:

• Sendhil Mullainathan sees a divergence instead of a convergence (with the macro guys willing to make sweeping recs, and the micro guys only willing to make "guarded statements" about very specific interventions)


• Martin Ravaillon argues that the internal validity of RCTs is less ironclad than advertised because of omitted variables like selective participation, other actors (e.g. gov't shifts resources toward control villages, etc.)

”Reductionist science” and development

I like two things in particular about Chris Blattman’s response to Aid Watch’s diatribe against Millenium Villages.

First, he does a great job of concisely synthesizing the opposing philosophies on development (Sachs vs. Easterly, if you will):

We wouldn’t be testing the fundamental premises: the theory of the big push; that high levels of aid simultaneously attacking many sectors and bottlenecks are needed to spur development; that there are positive interactions and externalities from multiple interventions.

The alternative hypothesis is that development is a gradual process, that marginal returns to aid may be high at low levels, and that we can also have a big impact with smaller, sector-specific interventions.

Second, he nails the flaw in the simplistic critique that MVs prove nothing because they’re not subject to rigorous evaluation (e.g. randomized impact evaluations):

To test the big push and all these externalities, we’d need to measure marginal returns to many single interventions as well as these interventions in combination (to get at the externalities). I’m not sure the sample size exists that could do it.

This reminds me of Michael Pollan’s critique of “reductionist science” in nutrition in his books – single development treatments (or nutrients) can be tested in isolation, but what if they have complex interactions with one another? The combinatorial explosion of trials you’d need to rigorously test all of these interactions renders the traditional scientific method helpless to mortals.

This, in turn, reminds me of business. Good businesspeople have no illusions that they’ll ever have rigorous certainty that they are making the right decision; but they have to make those decisions anyway.

How to apply this to development, I’m not sure (after all, “who decides who decides?”), but the reductionist science analogy was a new thought that I was excited to share.