Calculates variable-wise proportion of usable cases (missing and observed) as in Molenberghs et al. (2014).
prop_usable_cases(data)a list with two elements: p_obs and p_miss
p_obsthe proportion of missing \(Y_k\) among observed \(Y_j\); j in rows, k in columns
p_missthe proportion of observed \(Y_k\) among missing \(Y_j\); j in rows, k in columns
dataframe to be imputed
missForest builds models for each variable using the observed values of that variable as outcome of a random forest model. It then imputes the missing part of the variable using the learned models.
If all values of a predictor are missing among the observed value of the outcome,
the value of p_obs will be 1 and the model built will rely heavily on the initialized values.
If all values of a predictor are observed among the observed values of the outcome, p_obs will be 0
and the model will rely on observed values. Low values of p_obs are preferred.
Similarly, if all values of a predictor are missing among the missing values of the outcome,
p_miss will have a value of 0 and the imputations (predictions) will heavily rely on the initialized values.
If all values of a predictor are observed among the missing value of the outcome, p_miss will have a value of 1
and the imputations (predictions) will rely on real values. High values of p_miss are preferred.
Each row represents a variable to be imputed and each column a predictor.
Molenberghs, G., Fitzmaurice, G., Kenward, M. G., Tsiatis, A., & Verbeke, G. (Eds.). (2014). Handbook of missing data methodology. CRC Press. Chapter "Multiple Imputation"