Designing Experiments to Measure Spillover and Threshold Effects
Craig McIntosh
This paper formalizes the design of experiments intended specifically to study spillover effects. By first randomizing the intensity of treatment within clusters and then randomizing individual treatment conditional on this cluster-level intensity, a novel set of research questions can be addressed. Not only do we gain direct evidence as the impact of the treatment on untreated units, but the experimental variation in the treatment intensity allows the researcher a straightforward way to observe saturation and threshold effects among treated and untreated units alike. We present a framework in which to back out a rich set of treatment effects from such an experiment, and provide formulae for power calculations with two-level randomization. The technique is implemented using a Cash Transfer program in Malawi; we find spillover effects to be relatively muted at the cluster level but more potent within households. Conditional cash transfers appear to exert a broad positive educational spillover in treated villages, and we find no evidence that the program protects beneficiaries from HIV by diverting harm onto others.
Craig McIntosh is a development economist whose work focuses on program evaluation. His main research interest is the design of institutions which promote the provision of financial services to micro-entrepreneurs. He has conducted field evaluations of innovative anti-poverty policies in Mexico, Guatemala, Malawi, Rwanda, Uganda, and Tanzania. He is currently working on research projects investigating how to boost savings among the poor, on whether schooling can be used as a tool to fight HIV/AIDS in Sub-Saharan Africa, and on mechanisms to improve the long-term viability of Fair Trade markets. Craif McIntosh is an Associate Professor of Economics at UCSD.