The existing literature on treatment e¤ects assumes perfect observability of the treatments received by the population of interest. Even in cases of imperfect compliance, it is usually as- sumed that both the assigned and administered treatment are observed (or missing completely at random). This paper abandons such assumptions. Imperfect observability of the received treatment can arise as a result of survey nonresponse in observational studies, or noncompliance with randomly assigned treatments that are not directly monitored. I study the problem in the context of observational studies. I derive sharp worst case bounds without assuming anything about treatment selection, and I show that the bounds are a function of the available prior information on the distribution of the missing treatments. Under the maintained assumption of monotone treatment response, I show that no prior information on the distribution of missing treatments is necessary to get sharp informative bounds. I apply the methodologies recently proposed by Imbens and Manski (2004) and Chernozhukov, Hong, and Tamer (2004) to derive two types of confidence intervals for the partially identi.ed parameters. The results are illustrated with an empirical analysis of drug use and employment using data from the National Longitudinal Survey of Youth.
