Performs multiple imputation using the mice package.
Usage
perform_multiple_imputation(
data,
m = 5L,
maxit = 5L,
method = "pmm",
predictorMatrix = NULL,
seed = NULL,
print = FALSE,
...
)Arguments
- data
A data frame with missing values
- m
Integer. Number of imputations to perform (default: 5)
- maxit
Integer. Number of iterations for mice algorithm (default: 5)
- method
Character. Imputation method(s). Can be a single method applied to all variables or a named vector specifying method per variable. Common methods include:
"pmm": Predictive mean matching (default, for numeric)
"logreg": Logistic regression (for binary)
"polyreg": Polytomous regression (for unordered factors)
"polr": Proportional odds model (for ordered factors)
"norm": Bayesian linear regression
"rf": Random forest
- predictorMatrix
Matrix specifying which variables predict which. If NULL (default), mice determines automatically.
- seed
Integer. Random seed for reproducibility
Logical. Whether to print mice progress (default: FALSE)
- ...
Additional arguments passed to mice::mice()
Details
This function wraps mice::mice() to perform multiple imputation under the Missing at Random (MAR) assumption. The mice algorithm uses chained equations (MICE/FCS) to impute missing values.
For clinical trial data, predictive mean matching ("pmm") is often recommended as it preserves the distribution of observed values and works well with continuous variables.
References
van Buuren, S. and Groothuis-Oudshoorn, K. (2011). mice: Multivariate Imputation by Chained Equations in R. Journal of Statistical Software, 45(3), 1-67.
Examples
if (FALSE) { # \dontrun{
# Basic multiple imputation
data <- data.frame(
age = c(45, NA, 52, 38, NA),
bmi = c(25, 28, NA, 31, 27),
treatment = c("A", "B", "A", NA, "B")
)
result <- perform_multiple_imputation(data, m = 5, seed = 123)
# Access completed datasets
completed <- get_complete_data(result)
# View imputation summary
result@imputed_vars
result@n_missing
} # }