Jun 1, 2016
Published in: Journal of Experimental Criminology (2018). doi: 10.1007/s11292-018-9344-4
Posted on RAND.org on December 27, 2018
A central issue in experiments is protecting the integrity of causal identification from treatment spillover effects. The objective of this article is to demonstrate a bright line beyond which spillover of treatment renders experimental results misleading. We focus on a highly publicized recent test of police body cameras that violated the key assumption of a valid experiment: independence of treatment conditions for each unit of analysis.
In this article, we set out arguments for and against particular units of random assignment in relation to protecting against spillover effects that violate the Stable Unit Treatment Value Assumption (SUTVA).
Comparisons to methodological solutions from other disciplines demonstrate several ways of dealing with interference in experiments, all of which give priority to causal identification over sample size as the best pathway to statistical power.
Researchers contemplating which units of analysis to randomize can use the case of police body-worn cameras to argue against research designs that guarantee large spillover effects.