Skip to contents

Multivariate imputation by chained equations

mice_impute.Rd

Minimal, dependency-free re-implementation of the subset of the mice package (van Buuren & Groothuis-Oudshoorn, 2011) that is used by sample_covariates_mice(). Missing values are imputed variable-by-variable using a Gibbs sampler: each incomplete variable is regressed on all other (current, completed) predictors, and new imputations are drawn from the posterior predictive distribution of that model. This is repeated for maxit iterations, independently for each of the m replicates.

Usage

mice_impute(
  data,
  method = NULL,
  predictors = names(data),
  m = 1,
  maxit = 5,
  donors = 5L,
  ridge = 1e-05
)

Arguments

data

data.frame with missing values coded as NA. Categorical variables must be factors.

method

named character vector with the imputation method per variable. Variables not in method get a default method based on their type: pmm for numeric, logreg for factors with 2 levels, and polyreg for factors with more than 2 levels.

predictors

character vector with names of variables that may be used as predictors. Default is all variables in data.

m

number of imputed data sets (replicates).

maxit

number of iterations of the Gibbs sampler.

donors

number of donors to sample from in predictive mean matching.

ridge

ridge penalty used in the linear regression draws, to stabilize the estimation in case of collinear predictors.

Value

list of m completed data.frames.

Details

Supported imputation methods:

  • pmm: predictive mean matching (type 1 matching, as in mice). Imputed values are always values that are observed in the data.

  • norm: Bayesian linear regression (Rubin, 1987, p. 167).

  • logreg: Bayesian logistic regression, for binary factors. Uses data augmentation (White, Daniel & Royston, 2010) to deal with perfect prediction.

  • polyreg: polytomous regression for unordered factors with more than two levels, approximated by separate baseline-category binary logistic regressions (Begg & Gray, 1984).

  • "": variable is not imputed.

References

van Buuren S, Groothuis-Oudshoorn K (2011). mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software, 45(3), 1-67.

Rubin DB (1987). Multiple Imputation for Nonresponse in Surveys. Wiley.

White IR, Daniel R, Royston P (2010). Avoiding bias due to perfect prediction in multiple imputation of incomplete categorical variables. Computational Statistics & Data Analysis, 54(10), 2267-2275.

Begg CB, Gray R (1984). Calculation of polychotomous logistic regression parameters using individualized regressions. Biometrika, 71(1), 11-18.