Multiple imputation by chained equations for systematically and sporadically missing multilevel data

• DOI: 10.1177/0962280216666564
• Publication Date: 2016-09-20T00:49:08Z
• Authors: Matthieu Resche-R...; Ian R White
• Keywords: Imputation (statistics)

In multilevel settings such as individual participant data meta-analysis, a variable is ‘systematically missing’ if it wholly missing in some clusters and ‘sporadically partly clusters. Previously proposed methods to impute incomplete handle either systematically or sporadically data, but frequently both patterns are observed. We describe new multiple imputation by chained equations (MICE) algorithm for with arbitrary of variables. The described normal can easily be extended other types. first propose two imputing single variable: an extension existing method two-stage which conveniently allows heteroscedastic data. then discuss the difficulties values several variables using MICE, show that even simplest joint model implies conditional models involve cluster means heteroscedasticity. However, simulation study finds successfully combined MICE procedure, when not included models.