As was correctly pointed out by the Rigotnomic's observers (here and here), the debate about empirical methods in development economics and policy evaluation is on fire, and the flames were lit by Deaton and Heckman-Urzua (the good chilean econometrician), deeply questioning the main methodologies of impact evaluation like randomization and instrumental variables ....
As expected, the aswer did not take much time to come. Mr. Guido Imbens took the lead with the motto "better LATE than nothing" . Some highlights:
- "The fact that two such distinguished economists so forcefully question trends in current practice, may suggest to those not familiar with this literature that it is going seriously awry. In these comments I will argue that this is not the case".
- "Curiously, Deaton exempts the leaders of this movement from these charges, by declaring them 'too talented to be bound by their own methodological prescriptions' ”
- "conditional on the question of interest being one for which randomized experiment is feasible, randomized experiments are superior to all other designs in terms of statistical reliability".
- "But of course Deaton’s statements are wrong. Deaton is both formally wrong, and wrong in spirit"
- "The causal literature has emphasized internal validity over external validity, with the view that a credible estimate of the average effect for a subpopulation is preferred to an estimate of the average for the overall population with little credibility."
LATE = Local Average Treatment Effect