Abstract

This chapter discusses several important topics related to randomization in field experiments. In the field, researchers face constraints in the design they can implement and/or the type of data that can be collected. The chapter looks at such constraints from the point of view of randomization. We discuss challenges posed by these constraints and possible ways to tackle them, illustrated with examples from articles published in top economics journals. We review how randomization can help to estimate treatment effects. We discuss how treatment spillovers that invalidate stable unit treatment value assumption (SUTVA) can be addressed by choosing different levels of randomization. We pay attention to implementation and practical issues that matter in selecting the level of randomization, and to the use of covariates in the design and analysis stage. We also discuss how insufficient control of the assignment mechanism or partial treatment compliance may lead to failures, and how such failures can be prevented.