multiple.imputation: Multiple imputation function

In mvnimpute: Simultaneously Impute the Missing and Censored Values

Multiple imputation function

Description

Multiply imputes the missing and censored values in multivariate data.

Usage

multiple.imputation(data, prior.params, initial.values, iter, verbose = TRUE)

Arguments

data	a list of data containing the lower and upper bounds information for the missing and censored values.
prior.params	list of prior parameter specifications.
initial.values	list of initial values.
iter	number of rounds for doing multiple imputation.
verbose	boolean variable indicating whether the running status is printed in the console. Default is set to TRUE.

Details

A multivariate normal model is assumed on the data, the sweep operator is adopted to calculate the parameters of the conditional models. The implemented multiple imputation algorithm is based on the data augmentation algorithm proposed by Tanner and Wong (1987). The Gibbs sampling algorithm is adopted to update the model parameters and draw imputations of the coarse data. Output is a list including the parameters of the normal models and the imputed data across different iterations of multiple imputation.

Value

A list including the simulated mean and variance values of the assumed normal model, the covariance matrix, the imputed data, and the conditional model parameters across different iterations of multiple imputation.

References

Goodnight, J. H. (1979). A tutorial on the SWEEP operator. The American Statistician, 33(3), 149-158.

Tanner, M., & Wong, W. (1987). The Calculation of Posterior Distributions by Data Augmentation. Journal of the American Statistical Association, 82(398), 528-540.

Examples

## Not run: 
### data and indicator
miss.dat <- simulated.dat[[1]]
data.ind <- simulated.dat[[2]]

### number of observations and variables
n <- nrow(miss.dat); p <- ncol(miss.dat)

#### bound matrices
b1 <- b2 <- matrix(nrow = nrow(data.ind), ncol = ncol(data.ind))

for (i in 1:nrow(b1)) {
  for (j in 1:ncol(b1)) {
    b1[i, j] <- ifelse(data.ind[i, j] != 1, NA,
                       miss.dat[i, j])
    b2[i, j] <- ifelse(data.ind[i, j] == 0, NA, miss.dat[i, j])
  }
}
colnames(b1) <- colnames(b2) <- colnames(miss.dat)

#### create a matrix for including the lower and upper bounds
bounds <- list()
bounds[[1]] <- b1; bounds[[2]] <- b2

### prior specifications
prior.param <- list(
  mu.0 = rep(0, p),
  Lambda.0 = diag(100, p),
  kappa.0 = 2,
  nu.0 = p * (p + 1) / 2
)

### starting values
start.vals <- list(
  mu = rep(0, p),
  sigma = diag(100, p)
)

### imputation
sim.res <- multiple.imputation(
   data = bounds,
   prior.params = prior.param,
   initial.values = start.vals,
   iter = 500,
   verbose = FALSE
)

## End(Not run)

mvnimpute documentation built on July 6, 2022, 5:08 p.m.