ABSTRACT
The two historical strands of field experiments in economics relate to two distinct problems of external validity, the problem of artificiality for lab-in-the-field experiments (LFEs) and that of generalizability for randomized field experiments (RFEs). We show this by revisiting the concept of validity in experimental social sciences and how it has been received in the two strands. We then provide a detailed analysis of how the problem of generalizability manifests in RFEs in practice in evidence-based development economics. The chapter moves beyond the general critique of RFEs by philosophers of science and economists, first by critically evaluating two recent prominent proposals by practitioners of RFEs (Machine Learning and Structured Speculations), and second by illustrating how some of the actual practices in RFEs go beyond these proposals in trying to address the problem of generalizability.
