R/sort.associationMatrix.R

In userfriendlyscience: Quantitative Analysis Made Accessible

#' sort.associationMatrix
#' 
#' This function sorts an \code{\link{associationMatrix}} ascendingly or
#' descendingly by one of its columns.
#' 
#' Note that if the \code{\link{associationMatrix}} contains values of
#' different effectsizes, the sorting may be misleading. For example, a value
#' of Cohen's d of .45 is higher than a value of Pearson's r of .35, and so
#' will end up higher in a 'decreasing' sort - even though the association
#' represented by an r of .35 is stronger than that represented by a d of .45.
#' 
#' Furthermore, only asymmetrical associationMatrices can be sorted; sorting a
#' symmetrical association matrix would also change the order of the columns,
#' after all.
#' 
#' @param x The \code{\link{associationMatrix}} object to sort.
#' @param decreasing Whether to sort ascendingly (FALSE) or descending (TRUE).
#' @param byColumn Which column to sort the matrix by, as an index.
#' @param \dots Passed on to \code{\link{sort}}.
#' @return The \code{\link{associationMatrix}}, but sorted.
#' @author Gjalt-Jorn Peters
#' 
#' Maintainer: Gjalt-Jorn Peters <gjalt-jorn@@userfriendlyscience.com>
#' @seealso \code{\link{associationMatrix}}
#' @keywords ~kwd1 ~kwd2
#' @examples
#' 
#' sort(associationMatrix(infert, y=c("parity", "age"),
#'                        x=c("induced", "case", "spontaneous"), colNames=TRUE));
#' 
sort.associationMatrix <- function(x, decreasing = TRUE, byColumn = 1, ...) {
  
  if (is.null(x$input$y)) {
    stop("You can only sort asymmetrical associationMatrices, i.e., ",
         "those where you explicitely specified row and column variables. ",
         "for symmetrical association matrices, sorting would not make much ",
         "sense.");
  }

  ### First generate a numeric table with effectsizes and one with their types
  res <- x;
  res$intermediate$sorting <- list();
  res$intermediate$sorting$esValues <- data.frame();
  res$intermediate$sorting$esTypes <- data.frame();
  
  for (currentRowVar in 1:length(x$intermediate$effectSizes)) {
    for (currentColVar in 1:length(x$intermediate$effectSizes[[currentRowVar]])) {
      res$intermediate$sorting$esValues[currentRowVar, currentColVar] <-
        x$intermediate$effectSizes[[currentRowVar]][[currentColVar]]$es;
      res$intermediate$sorting$esTypes[currentRowVar, currentColVar] <-
        x$intermediate$effectSizes[[currentRowVar]][[currentColVar]]$es.type;
    }
  }
  
  if (length(unique(unlist(res$intermediate$sorting$esTypes))) > 1) {
    warning("The associationMatrix you provided contains ",
            length(unique(unlist(res$intermediate$sorting$esTypes))),
            " different effect size types (specifically ",
            vecTxt(unique(unlist(res$intermediate$sorting$esTypes)), useQuote="'"),
            ")! Sorting may be unreliable.");
  }

  res$intermediate$sorting$order <- order(res$intermediate$sorting$esValues[, byColumn],
                                          decreasing = decreasing);
  res$intermediate$sorting$orderForFull <-
    c(rbind((2*res$intermediate$sorting$order) - 1, 2*res$intermediate$sorting$order));
    
  res$output$matrix.raw <- res$output$matrix;
  
  ### Some measures to make sure we actually get a matrix back - apparently,
  ### reordering a matrix also simplifies it to a vector :-/
  nCol <- ncol(res$output$matrix.raw$es);
  colNames <- colnames(res$output$matrix.raw$es);
  rowNames <- rownames(res$output$matrix.raw$es)[res$intermediate$sorting$order];
  rowNamesForFull <- rownames(res$output$matrix.raw$full)[res$intermediate$sorting$orderForFull];

  res$output$matrix$es <- matrix(res$output$matrix$es[res$intermediate$sorting$order, ],
                                 ncol=nCol, dimnames=list(rowNames, colNames));
  res$output$matrix$ci <- matrix(res$output$matrix$ci[res$intermediate$sorting$order, ],
                                 ncol=nCol, dimnames=list(rowNames, colNames));
  res$output$matrix$full <- matrix(res$output$matrix$full[res$intermediate$sorting$orderForFull, ],
                                   ncol=nCol, dimnames=list(rowNamesForFull, colNames));

  return(res);
  
}