RCT
The rise of a methodology
Randomized control studies are designed to compare the outcome of a project (program or policy)
with what would have happened without the intervention in order to measure its net impact, ie less
all changes occurring elsewhere. The challenge is to build the basic scenario (the counterfactual
without a project) which, by definition, is never observed. The solution proposed by randomized
control studies consists in taking two random samples from a population that could benefit from the
intervention. The project is assigned to only one of the groups, but surveys are conducted on both
groups before and after the project. The statistical properties of sampling theory ensure that, on
average, the observed differences between beneficiaries and non-beneficiaries can be attributed to
the project. As with all probabilistic methods, the results are reported with a margin of error
(confidence interval), which depends on the sampling characteristics (size, method, friction, etc.).
The randomized control studies then try to formally establish a causal link between an intervention
and a certain number of outcome variables. From the scientific and theoretical point of view, one
could legitimately claim to be the most convincing option available to identify existence and
quantify the magnitude of the observed impact. In quantitative evaluations, they are probably more
robust than other methods: when a control group is not set up before the fact, successive and con-
dependent approaches can not control changes in the context; The quasi-experimental comparison
methods - which correspond to beneficiaries and non-beneficiaries based on shared observable
characteristics - partially cancel this constraint. Still in quantitative assessments, the RCTs in
principle meet the methodological challenge of demonstrating the causality direction without
relying on complex econometric and yet refutable assumptions. Last and more classically, they
differ from qualitative methods (case studies, monographs, interviews and observation of
participants) in their quantitative measurement of impact, which is outside the scope (and scope) of
qualitative methods.
Supporters of the use of RCTs in the development economy have imported the method from the
medical world without due consideration of the critical discussions, the conditions for their use and
the questions already raised about them in the field of public health (Labrousse, 2010; Eble et al.,
2014). They also passed the controversy that marked decades of debates on the development
economy.
The internal validity of the RCTs (ie the reliability of the results obtained) should be the main
strength of the method. Yet the RCTs are anything but without limits. Although their limits are
essentially empirical, in the way that RCTs are conducted in the field, some are more theoretical.
The main criticism in this regard concerns the tension between prejudices (to be minimized) and
precision (to be maximized), in which the RCTs fail to make an optimal compromise in line with
the basic principles of decision theory.
The question of external validity, on the other hand, is by far the most discussed in the literature.
The focus on a "medium" impact, beyond the statistical problems mentioned above, says nothing
about the heterogeneity of the impacts and their distribution (Ravallion, 2009, DFID, 2012). The
limitation to short-term impact (for cost and attrition reasons) often means that the mid-point
indicators are studied, which can be very different from the final results (Eble et al., 2014) if not the
opposite, since many project trajectories are non-linear
