Multiple Imputation for Incomplete Data With Semicontinuous Variables

• DOI: 10.1198/016214503000000611
• Publication Date: 2004-11-25T03:01:28Z
• Authors: Kristin N Javaras; David A Van Dyk
• Keywords: Imputation (statistics); Categorical variable

AbstractWe consider the application of multiple imputation to data containing not only partially missing categorical and continuous variables, but also 'semicontinuous' variables (variables that take on a single discrete value with positive probability are otherwise continuously distributed). As an model for sets this type, we introduce extension standard general location proposed by Olkin Tate; our extension, blocked model, provides robust strategy handling observed semicontinuous variables. In particular, incorporate two-level into model. The first level models variable takes its point mass value, second distribution given it is at mass. addition, EM augmentation algorithms data; these can be used generate imputations under have been implemented in publicly available software. We illustrate computational methods via simulation study analysis survey Massachusetts Megabucks Lottery winners.KEY WORDS: Data augmentationEM algorithmGeneral modelMissing dataSurvey