R/delete_MCAR.R

In torockel/missMethods: Methods for Missing Data

#' Create MCAR values
#'
#' Create missing completely at random (MCAR) values in a data frame or a matrix
#'
#' @template delete
#' @template delete-stochastic
#'
#' @details This function creates missing completely at random (MCAR) values in
#' the dataset \code{ds}.
#' The proportion of missing values is specified with \code{p}.
#' The columns in which missing values are created can be set via \code{cols_mis}.
#' If \code{cols_mis} is not specified, then missing values are created in
#' all columns of \code{ds}.
#'
#  part about p copied from man-roxygen/MAR.R !!!!!!
#
#' The probability for missing values is controlled by \code{p}. If \code{p} is
#' a single number, then the overall probability for a value to be missing will
#' be \code{p} in all columns of \code{cols_mis}. (Internally \code{p} will be
#' replicated to a vector of the same length as \code{cols_mis}. So, all
#' \code{p[i]} in the following sections will be equal to the given single
#' number \code{p}.) Otherwise, \code{p} must be of the same length as
#' \code{cols_mis}. In this case, the overall probability for a value to be
#' missing will be \code{p[i]} in the column \code{cols_mis[i]}.
#'
#' If \code{n_mis_stochastic = FALSE} and \code{p_overall = FALSE} (the default), then
#' exactly \code{round(nrow(ds) * p[i])} values will be set \code{NA} in column
#' \code{cols_mis[i]}. If \code{n_mis_stochastic = FALSE} and \code{p_overall =
#' TRUE}, then \code{p} must be of length one and exactly \code{round(nrow(ds) *
#' p * length(cols_mis))} values will be set NA (over all columns in
#' \code{cols_mis}). This can result in a proportion of missing values in every
#' \code{miss_col} unequal to \code{p}, but the proportion of missing values in
#' all columns together will be close to \code{p}.
#'
#' If \code{n_mis_stochastic = TRUE}, then each value in column
#' \code{cols_mis[i]} has probability \code{p[i]} to be missing (independently
#' of all other values). Therefore, the number of missing values in
#' \code{cols_mis[i]} is a random variable with a binomial distribution
#' \emph{B}(\code{nrow(ds)}, \code{p[i]}). This can (and will most of the time)
#' lead to more or less missing values than \code{round(nrow(ds) * p[i])} in
#' column \code{cols_mis[i]}. If \code{n_mis_stochastic = TRUE}, then the
#' argument \code{p_overall} is ignored because it is superfluous.
#'
#' @param p_overall Logical; see details.
#'
#' @export
#'
#' @examples
#' ds <- data.frame(X = 1:20, Y = 101:120)
#' delete_MCAR(ds, 0.2)
delete_MCAR <- function(ds, p, cols_mis = seq_len(ncol(ds)),
                        n_mis_stochastic = FALSE, p_overall = FALSE, miss_cols, stochastic) {
  # The real work is done inside of .delete_MCAR()
  do.call(delete_values, c(
    list(mech_type = "MCAR"),
    as.list(environment())
  ))
}

.delete_MCAR <- function(ds, p, cols_mis = seq_len(ncol(ds)),
                         n_mis_stochastic = FALSE, p_overall = FALSE, miss_cols) {
  n <- nrow(ds)
  if (!p_overall || n_mis_stochastic) {
    for (i in seq_along((cols_mis))) {
      ds[, cols_mis[i]] <- delete_MCAR_vec(
        ds[, cols_mis[i], drop = TRUE],
        p[i], n_mis_stochastic
      )
    }
  } else { # p_overall = FALSE && n_mis_stochastich = FALSE
    na_indices_overall <- sample.int(
      n * length(cols_mis),
      round(n * length(cols_mis) * p[1])
    )
    for (i in seq_along(cols_mis)) {
      na_indices <- na_indices_overall[(na_indices_overall - 1L) %/% n == (i - 1L)]
      na_indices <- ((na_indices - 1L) %% n) + 1L
      ds[na_indices, cols_mis[i]] <- NA
    }
  }
  ds
}

# delete values MCAR in a single vector
delete_MCAR_vec <- function(x, p, n_mis_stochastic) {
  na_indices <- get_NA_indices(n_mis_stochastic, n = length(x), p = p)
  x[na_indices] <- NA
  x
}