At the next Perspectives on Science seminar (Monday 16.9.), Carlo Martini (San Raffaele/Helsinki) will present his paper entitled “Here, There, and Everywhere: Extrapolation and Policy Validity in Randomized Field Experiments in Economics”.
Perspectives on Science is a weekly research seminar which brings together experts from science studies and philosophy of science. It is organized by TINT, the Centre for Philosophy of Social Science at the University of Helsinki.
The seminar will take place in room P673 in Porthania (Yliopistonkatu 3), from 14-16.
In the past few decades the use of randomized field experiments (RFEs) in Economics has increased dramatically and is still increasing. The high internal validity of such experiments is undoubtedly their most attractive feature; however, their external validity is still problematic. RFEs are considered to have strong internal validity, allowing them to produce what is labelled as evidence, but strong internal validity may be coupled with weak external validity. There are a number of issues related to external validity: one is the problem of ”global external validity” – i.e., how do we generalise from the test-population to the whole population. Another issue is ”local external validity” – i.e., how to infer, from the success of an experiment in a test-population, to some conclusions about a different but equivalent (for the purposes of the test) population. Both issues are discussed extensively in the literature, and RFEs are problematic under both aspects. In this paper, however, we claim that the subdivision does not capture all the possible ways in RFEs can go wrong. We postulate and explain the concept of policy validity. Policy validity is validity for the purposes of policy making. Policy validity may still fail when both global and local external valid are carefully controlled for. We present a case study about the use of fertilizers: The study assesses the impact of a development program (SAFI) on the use of fertilizers by Kenyan farmers. We conclude that policy validity requires a high degree of expert judgment, and cannot be easily managed through the standard techniques that are sometimes employed for other types of validity.