MICE: Multivariate Imputation by Chained Equations

MICE stands for Multivariate Imputation by Chained Equations. The software mice version 1.0 appeared in the year 2000 as an S-PLUS library, and in 2001 as an R package. The package mice 1.0 introduced predictor selection, passive imputation and automatic pooling. The package mice 2.0 added new functionality for imputing multilevel data, automatic predictor selection, data handling, post-processing imputed values, specialized pooling routines, model selection tools, and diagnostic graphs.

Specific features of MICE 2.12 include:

  • fully condition specification of the imputation model;
  • arbitrary patterns of missing data;
  • transformations, sum scores, indices by passive imputation;
  • subset selection of predictors;
  • quick predictor selection by quickpred();
  • multilevel imputation by mice.impute.2l.norm();
  • post-processing imputations through the post argument;
  • supports for arbitrary complete-data methods using with();
  • model testing by;
  • pooling by Rubin's rules, Wald test, and likelihood ratio test;
  • callable user-written imputation functions;
  • combining imputed data by cbind.mids(), rbind.mids() and ibind();
  • graphic diagnostics of imputations.



Original MICE reports

Van Buuren S, Oudshoorn CGM (2000). Multivariate Imputation by Chained Equations: MICE V1.0 User's manual. Leiden: TNO Prevention and Health. TNO Report PG/VGZ/00.038.
Oudshoorn CGM, van Buuren S, van Rijckevorsel JLA (1999). Flexible multiple imputation by chained equations of the AVO-95 Survey. Leiden: TNO Prevention and Health. Report PG/VGZ/99.045.
Van Buuren S, Oudshoorn CGM (1999). Flexible multivariate imputation by MICE. Leiden: TNO Prevention and Health, Report PG/VGZ/99.054.
Note: The syntax used in the 1999 reports is based on unpublished versions of the MICE library. The appropriate syntax is given in de MICE V1.0 documentation.