Option Option Description Assumed conditions that might lead to deviations from fully informed weighted meta-analyses
10) mice: Random forest Implements a random forest algorithm (Breiman) and fills missing values with average predictions from 10 classification and regression trees that are based on 10 random subsets of the predictor variables. This method shares many features with the classification and regression tree imputation but the imputed values exhibit a larger variability. Missing values are outside the range of the reported values and/or MNAR
11) mi: Bayes predictive mean matching Fits Bayesian generalized linear models to fill missing values with those reported values that are closest to the predicted ones. Like predictive mean matching, imputed values are thereby restricted to a subset of the reported ones. Missing values are outside the range of the reported values and/or MNAR
12) Amelia: Bootstrap expectation maximization Draws multiple bootstrap samples from the supplied data and calculates separate posterior maxima. The distribution of these maxima is then used to fill the missing values. In order to yield reliable imputations this algorithm assumes multivariate normality and MCAR or MAR. Missing values are MNAR.
13) missForest: Non-parametric random forest Iterates the random forest-algorithm (Breimans) until a certain convergence criterion is fulfilled. Missing values are outside the range of the reported values and/or MNAR
14) Hmisc: Additive regression plus bootstrap predictive mean matching Draws multiple bootstrap samples from the supplied data and fits separate additive regression models to obtain averaged predictions for the missing values. These missing values are then filled with those observed values that are closest to the predicted ones. Like predictive mean matching, imputed values are thereby restricted to a subset of the reported ones. Missing values are outside the range of the reported values and/or MNAR