User:Brozo stan/sandbox

= The Deaton - Banerjee debate = Two of the most prominent development economists, Angus Deaton and Abhijit Banerjee are on opposite sides of the academic debate about the role of Randomized Control Trials, in particular in the discipline of development economics.

The prompt for the discussion was given by the Debates in development conference, organized by New York University after the publication of Banerjee's and Esther Duflo's book "Poor Economics".

Banerjee's position
Banerjee argued in favor of the heuristic value of Randomized Control Trials (RCTs) by focusing on the subject of microcredit.

The strength of controlled experiments- he argued- relies on their ability to establish causality by ruling out confounding factors.

Moreover, they can be tailored on the specific research question of interest- they are not limited by the actual implementation of a policy to be analyzed.

This also means that the scholar can test different questions at the same time, can replicate the experiment and thus embrace a 'dynamic learning agenda'- in a way that is hard to reach with other research methods.

Microcredit is a particularly good subject of research in which to implement RCTs- Banerjee argues- and this can be seen in the results of an experiment in Hyderabad, India. The study showed an ineffectiveness of microcredit in stimulating small businesses growth, despite the previously established high marginal returns on small investments for these businesses. The puzzle presented by these results and the will to solve them, Banerjee argues, is what moves research forward and RCTs play a key role in that.

Deaton's position
Deaton's reply and critique of RCTs is articulated in two main points. The first point states that often RCTs are not actually capable of removing selection bias and thus may suffer from the same limitations of quasi-experimental research methods. The second point, on the other hand, is that- even when RCTs do manage to rigorously establish causality- they only establish an INUS - Insufficient but Non-redundant part of a Condition, which is itself Unnecessary but Sufficient- form of causality.

With respect to the former argument, Deaton points out that there are two stages of randomization in an RCT- the former selects from the population the units of analysis that will be later divided in treatment and control groups. It is precisely this passage that- he states- tends not to be immune to selection bias, as convenience and politics may affect the selection. This can inficiate the external validity of the study. Moreover, randomness requires large samples to eliminate selection bias. But samples used for RCTs are often not large enough- Deaton points out- and thus outliers can greatly affect the results, compromising the internal validity of the study.

The second point he raises can be understood through the example of a kite sharpening a pencil through a sophisticated device - even if causality is established, the notion is of limited heuristic value. In general, Deaton argues in favor of a better understanding of the theoretical mechanism behind the results of a policy, something that an RCT is not really fit to unearth, in his mind.

Instead of RCTs, Trial and Error should be- according to Deaton - the way to proceed in order to uncover the reasons behind econometric results.

A final critique to RCTs brought forth by Deaton concentrates on the fact that this research method yields result on the average effect of the treatment of interest, without regard for the effects on the tails of the distribution.

In sum, RCTs are extremely dependent on the specific contest in which they are implemented- just like all other research methods- and this is the reason not to hold them in a special place vis a vis other econometric specifications, according to Deaton.