This thesis focuses on the experimental approach used by J-PAL's researchers in development economics. The goal of these researchers is twofold: (1) producing evidence concerning the efficiency of development programs, (2) in order to guide policy decisions. The aim of this thesis is to conduct an epistemological analysis of the J-PAL's approach. This analysis studies two dimensions: a methodological one and a theoretical one. The methodological dimension aims to question the method used by the J-PAL: randomization. Two main questions guide this methodological analysis: (1) randomization's methodological status of «gold standard », (2) and the possibility to transpose the results obtained by the J-P AL in the political area. The second, theoretical, dimension questions the J-PAL's contributions to the (theoretical) debates within development economics over the past ten years. The study of these two dimensions allows an examination of the J-PAL's approach as a whole. The thesis shows that the results of this approach have a strong internal validity at the expense of a weak external validity. This makes the use of such results in the political area difficult. This tension between internal validity and external validity points out the antagonism of the J-PAL's two objectives and defines an epistemological issue: the refusal of a theory along with the lack of emphasis on the underlying mechanisms of the J-PAL's results prevents it from producing clear political recommendations.
