R/MFClus.r

In ABS-dev/MF: Mitigated Fraction

#' Estimates mitigated fraction from clustered or stratified data.
#'
#' Averages the U statistic over the clusters and computes MF from it. Clusters
#' are excluded if they do not include both treatments.
#'
#' @title Clustered mitigated fraction
#' @param formula Formula of the form \code{y ~ x + cluster(w)}, where y is a
#'   continuous response, x is a factor with two levels of treatment, and w is a
#'   factor indicating the clusters.
#' @param data Data frame.  See \code{Note} for handling of input data with more
#'   than two levels.
#' @param compare Text vector stating the factor levels - \code{compare[1]} is
#'   the control or reference group to which \code{compare[2]} is compared
#' @param trace.it Verbose tracking of the cycles? Default FALSE.
#' @return a \code{\link{mfcluster-class}} data object
#' @note If input data contains more than two levels of treatment, rows
#' associated with unused treatment levels will be removed. \cr Factor levels
#' for treatments not present in the input data will be ignored. \cr Clusters
#' with missing treatments will be excluded. See
#' \code{\link{mfbootcluster-class}} or use \code{trace.it} to identify excluded
#' clusters.
#' @export
#' @references Siev D. (2005). An estimator of intervention effect on disease
#'   severity. \emph{Journal of Modern Applied Statistical Methods.}
#'   \bold{4:500--508}
#' @author \link{MF-package}
#' @seealso \code{\link{mfcluster-class}}
#' @examples
#' \dontrun{
#' MFClus(lesion ~ group + cluster(litter), piglung)
#'
#' #  Comparing vac to con
#' #
#' #  MF = 0.3533835
#' #
#' #  By Cluster
#' #     w  u         r n1 n2         mf
#' #  U 25 10 0.4000000  5  5 -0.2000000
#' #  K 12  2 0.2500000  4  2 -0.5000000
#' #  Z 16 10 0.8333333  3  4  0.6666667
#' #  D  3  2 1.0000000  1  2  1.0000000
#' #  N  1  0 0.0000000  1  3 -1.0000000
#' #  T  8  5 0.8333333  2  3  0.6666667
#' #  P  4  1 0.5000000  2  1  0.0000000
#' #  L  3  2 0.6666667  1  3  0.3333333
#' #  G 15  9 0.7500000  3  4  0.5000000
#' #  J 15  9 1.0000000  3  3  1.0000000
#' #  W  6  3 0.7500000  2  2  0.5000000
#' #  A  9  3 0.3333333  3  3 -0.3333333
#' #  X 12  6 1.0000000  3  2  1.0000000
#' #  F 13  7 0.7777778  3  3  0.5555556
#' #  S 21 11 0.9166667  4  3  0.8333333
#' #  H 14  8 0.8888889  3  3  0.7777778
#' #  Y  2  1 1.0000000  1  1  1.0000000
#' #  E  2  1 1.0000000  1  1  1.0000000
#' #
#' #  All
#' #        w  u         r n1 n2        mf
#' #  All 181 90 0.6766917 50 52 0.3533835
#' #
#' #  Excluded Clusters
#' #  [1] M, Q, R, B, O, V, I, C
#' }

#--------------------------------------------------------------------
# Clustered or Stratified MF
#--------------------------------------------------------------------
#
MFClus <- function(formula, data, compare = c("con", "vac"), trace.it = FALSE) {
  # formula of the form response ~ treatment + cluster(clustername)
  # based on prob{F(y) < F(x)}
  # within-cluster ranking only
  # 3/19/01 initial coding
  # revised 10/3/06 to eliminate clusters without both treatments represented
  # revised 8/27/13 - remove group levels if no observations from that level
  #     are present in original data
  # revised 9/03/13 - subset initial data by comparison group levels
  # revised 9/03/13 - move data reshaping shared by MFClusBoot and MFClus to
  #     external function
  dat <- NULL
  group <- NULL
  clusters <- NULL
  strat <- NULL
  reshape_cluster(data = data, formula = formula, compare = compare,
                  envir = environment())
  id <- compare
  out <- matrix(NA, length(strat), 6,
                dimnames = list(strat, c("w", "u", "r", "n1", "n2", "mf")))
  excluded.clusters <- NULL
  for (stratum in strat) {
    x <- dat[group == id[1] & as.character(clusters) == stratum]
    y <- dat[group == id[2] & as.character(clusters) == stratum]
    n.x <- length(x)
    n.y <- length(y)
    if (n.x > 0 && n.y > 0) {
      x.y <- c(x, y)
      w <- sum(rank(x.y)[1:n.x])
      u <- w - (n.x * (n.x + 1)) / 2
      r <- u / (n.x * n.y)
      mf <- 2 * r - 1
      out[stratum, ] <- c(w, u, r, n.x, n.y, mf)
    } else {
      if (trace.it) {
        cat("Cluster", stratum, "missing a treatment\n")
      }
      out[stratum, ] <- c(NA, NA, NA, n.x, n.y, NA)
      excluded.clusters <- c(excluded.clusters, stratum)
    }
  }
  All <- apply(out, 2, sum, na.rm = TRUE)
  R <- All["u"] / sum(out[, "n1"] * out[, "n2"])
  MF <- 2 * R - 1
  All["r"] <- R
  All["mf"] <- MF
  All <- data.frame(t(All))
  dimnames(All)[[1]] <- "All"

  if (round(All$mf, digits = 1) == 1.0) {
    message("Complete separation observed.")
  }
  return(mfcluster$new(All = All, byCluster = out,
                       excludedClusters = excluded.clusters,
                       call = match.call(),
                       compare = compare))
}