When random assignment fails: Some lessons from the Minneapolis Spouse Abuse Experiment

In this paper, we consider what may be done when researchers anticipate that in the implementation of field experiments, random assignment to experimental and control groups is likely to be flawed. We then reanalyze data from the Minneapolis Spouse Abuse Experiment in a manner that explicitly models violations of random assignment. As anticipated, we find far larger treatment effects than previously reported. The techniques developed should be useful in a wide variety of settings when random assignment is implemented imperfectly.