Economics as an Experimental Science: Using Field Experiments to Test Models of Economic Behavior

Abstract

Economics is limited by the fact that it is not an experimental science: our ability to experimentally test models of economic behavior is constrained by logistics and ethics. However, this view has been reshaped by the credibility revolution of the last two decades and the recent growth of randomized trials in developing countries, which have shown that many economic models can indeed be tested experimentally. Critics of this movement claim that experimental and quasi-experimental results do not generalize, and that such studies focus on convenient rather than important topics.

It is possible to overcome this critique by running so-called "mechanism experiments" that are designed to test economic models. Economic theory can also help uncover the drivers of specific results, shedding light on whether they will generalize to other settings and why. This approach can address topics that are not directly amenable to experimentation, by running experiments that capture the same theoretical object in another context. This dissertation applies this approach to three first-order questions in development economics.

The first chapter reassesses the standard economic model of risk compensation, which assumes that increases in risk cause people to become more careful. I show that risk-seeking, or fatalistic, behavior, can also be rational. I then use a randomized experiment to show that some southern Malawians respond fatalistically to HIV risks.

The second chapter, written with Lasse Brune, examines how random variations in income timing affect expenditure and savings in southern Malawi. We show that lump-sum payments lead to increased savings, but, contrary to existing theoretical work and empirical evidence from developed countries, we find no evidence that exposure to tempting goods affects this result.

The third chapter, written with Rebecca Thornton, studies a literacy program in Northern Uganda. When implemented at full cost, the program strongly improves learning. A reduced-cost version of the program also improves learning, but only for the most basic reading and writing skills – and actually harms the development of advanced writing skills. Our results cast doubt on typical cost-effectiveness calculations: the cheaper version is more cost-effective for basic skills, but the opposite is true for overall learning.