R/centroid_variance.R

In dyerlab/popgraph: This is an R package that constructs and manipulates population graphs

#' This function takes a data set and a group set and returns centroid variances
#' 
#' This is a convience function for a reduce and distance operation based upon
#'  multivariate predictor variables.
#' @param x The NxP matrix of raw data.
#' @param grouping The Nx1 vector of grouping factors.
#' @return The sum of the within stratum variance for each group
#' @author Rodney J. Dyer <rjdyer@@vcu.edu>
#' @export
centroid_variance <- function( x, grouping ){
  
  if( !is(grouping, "factor"))
    grouping <- factor( grouping )
  N <- dim(x)[1]
  
  if( length(grouping) != N )
    stop("Your grouping and predictor variables should ',
         'have the same number of rows....  Come on now.")
  
  grps <- levels( grouping) 
  K <- length(grps)
  ret <- rep(NA,K)
  names(ret) <- grps

  
  for( i in 1:K ){
    xp <- x[ grouping == grps[i],]
    ret[i] <- sum(diag(cov(xp)))
  }

  return( ret )
}