The primary concern in such situations is that distribution of the outcome variable may vary in important ways between individuals whose outcomes are observed and individuals whose outcomes are missing. Since the data cannot be used to resolve this concern, it is necessary to take the uncertainty created by the missing data into account when developing inferences about the magnitude of the treatment effect. This article considers a modeling framework that accomplishes that objective. It then applies the framework to a study of the effectiveness of different types of police responses to spouse assault incidents in Charlotte, NC. The major premise underlying this work is that the investigation of treatment effects in experiments with partially missing outcome data can benefit from the adoption of middle-ground methods for confronting the uncertainty created by the missing data. Notes, tables, figures, appendix, references