title:         Statistical Analysis of Randomized Experiments with Nonignorable Missing Outcomes
authors:       Kosuke Imai
entrydate:     2006-07-24 21:14:13
keywords:      Causal Inference, Instrumental Variables, Intention-to-Treat Effect, Latent Ignorability, Noncompliance, Treatment Effect.
abstract:      Missing data are frequently encountered in the statistical analysis of randomized experiments. In this paper, I study the problem of nonignorable missing outcomes where the missing-data mechanism of an outcome variable may depend on the values of the variable itself. I first consider standard randomized experiments, and propose an identification strategy for the average treatment effect in the presence of nonignorable missing outcomes. The maximum likelihood and Bayes estimators are derived, and a simulation study is conducted to examine their finite-sample properties. I then extend the proposed methodology to identify the intention-to-treat effect and complier average causal effect in randomized experiments with noncompliance. The proposed approach is compared it with the existing approaches commonly used in the literature. The maximum likelihood and Bayes estimators are derived, and a Monte Carlo experiment is conducted to compare the two estimators. I find that when the sample size is small, the Bayes estimator outperforms the maximum likelihood estimator. Finally, I apply the proposed methodology to analyze data from a German election experiment and an influenza vaccination study, which motivated the methodological questions studied in this paper.

http://polmeth.wustl.edu/retrieve.php?id=631

**********************************************************
             Political Methodology E-Mail List
        Editor: Karen Long Jusko <[log in to unmask]>
**********************************************************
        Send messages to [log in to unmask]
  To join the list, cancel your subscription, or modify
           your subscription settings visit:

          http://polmeth.wustl.edu/polmeth.php

**********************************************************