Noncompliance and missing data often occur in randomized trials, whichcomplicate the inference of causal effects. When both noncompliance and missingdata are present, previous papers proposed moment and maximum likelihoodestimators for binary and normally distributed continuous outcomes under thelatent ignorable missing data mechanism. However, the latent ignorable missingdata mechanism may be violated in practice, because the missing data mechanismmay depend directly on the missing outcome itself. Under noncompliance and anoutcome-dependent nonignorable missing data mechanism, previous studies showedthe identifiability of complier average causal effect for discrete outcomes. Inthis paper, we study the semiparametric identifiability and estimation ofcomplier average causal effect in randomized clinical trials with bothall-or-none noncompliance and the outcome-dependent nonignorable missingcontinuous outcomes, and propose a two-step maximum likelihood estimator inorder to eliminate the infinite dimensional nuisance parameter. Our method doesnot need to specify a parametric form for the missing data mechanism. We alsoevaluate the finite sample property of our method via extensive simulationstudies and sensitivity analysis, with an application to a double-blindedpsychiatric clinical trial.
展开▼
