Multiple imputation by chained equations (MICE): Implementation in Stata

Abstract

Missing data are a common occurrence in real datasets. For epidemiological and prognostic factors studies in medicine, multiple imputation is becoming the standard route to estimating models with missing covariate data under a missing-at-random assumption. We describe ice, an implementation in Stata of the MICE approach to multiple imputation. Real data from an observational study in ovarian cancer are used to illustrate the most important of the many options available with ice. We remark briefly on the new database architecture and procedures for multiple imputation introduced in releases 11 and 12 of Stata.