Theoretical and computational issues when making causal inference in randomized experiments with imperfect compliance and missing data are discussed. The analysis of the complications from which a randomized experiment can suffer is attractive, because it is not only limited to a randomized trial analysis. The template of a randomized experiment with imperfect compliance can be adopted in fact to identify and estimate causal effects also in observational studies. More, some complications like the presence of non-responses in the treatment and/or in the assignment to treatment are more plausible in the context of an observational Study than in a randomized trial. From the theoretical point of view the complications in expliciting the likelihood function when the inference is based on ignorability conditions for the missing data mechanism are discussed. Then the inputs for implementing the EM algorithm and examples based on artificial data are presented. (C) 2003 Elsevier B.V. All rights reserved.
Analyzing a randomized experiment with imperfect compliance and ignorable conditions for missing data: theoretical and computational issues / Mercatanti, Andrea. - In: COMPUTATIONAL STATISTICS & DATA ANALYSIS. - ISSN 0167-9473. - 46:3(2004), pp. 493-509. [10.1016/j.csda.2003.09.003]
Analyzing a randomized experiment with imperfect compliance and ignorable conditions for missing data: theoretical and computational issues
Andrea Mercatanti
2004
Abstract
Theoretical and computational issues when making causal inference in randomized experiments with imperfect compliance and missing data are discussed. The analysis of the complications from which a randomized experiment can suffer is attractive, because it is not only limited to a randomized trial analysis. The template of a randomized experiment with imperfect compliance can be adopted in fact to identify and estimate causal effects also in observational studies. More, some complications like the presence of non-responses in the treatment and/or in the assignment to treatment are more plausible in the context of an observational Study than in a randomized trial. From the theoretical point of view the complications in expliciting the likelihood function when the inference is based on ignorability conditions for the missing data mechanism are discussed. Then the inputs for implementing the EM algorithm and examples based on artificial data are presented. (C) 2003 Elsevier B.V. All rights reserved.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.