Title: The not at random fully conditional specification procedure
Authors: Daniel Tompsett - MRC Biostatistics Unit:University of Cambridge (United Kingdom) [presenting]
Abstract: The Not at Random Fully Conditional Specification Procedure (NARFCS) for imputing multivariate missing data under the Missing Not at Random (MNAR) assumption will be described. Methods exist to impute MNAR data under the Fully Conditional Specification procedure (FCS). The NARFCS procedure represents a formalisation of such methods, which fully defines the imputation models for each variable with missing data. We will first outline the imputation procedure, and offer general advice as to how to construct it's imputation models. We will then show how to elicit values for the sensitivity parameters of the procedure, which are conditional on the remaining variables of the data. The procedure will be demonstrated on an example dataset using functions for R, specifically designed for NARFCS.