Nonparametric Analysis of Randomized Experiments With Missing Covariate and Outcome Data

Analysis of randomized experiments with missing covariate and outcome data is problematic because the population parameters of interest are not identified unless one makes untestable assumptions about the distribution of the missing data. This paper shows how population parameters can be bounded without making untestable distributional assumptions. The bounds are sharp; that is, they exhaust all of the information that is available from the data. The bounds are illustrated with an application to data obtained from a clinical trial of treatments for hypertension. The bounds are sufficiently narrow to permit substantive conclusions to be drawn about the effects of different treatments.