R/tab2way.R

In rnitulescu/munge: Efficient Summary Statistics

# Two-way tabulation
#
# Returns a two-way table counting the number (and percentage) of observations per
# category combination across two vectors.
# @param x vector of factor or logical values.
# @param byvar vector of factor or logical values (stratification).
# @param FACLST character vector of valid class types.
# @keywords Two-way
#' @import data.table
#' @importFrom stats reshape
# @examples
# tab2way(x=factor(c("A","B","B","A")), byvar=factor(c("C","D","C","D")))
tab2way <- function(x, byvar, FACLST=MYFACLST) {
	## Don't run if variables are not of logical or factor type
    ## (Added "flatten" here to get rid of "ordered" component to some factors' classes)
	if ( !(flatten(class(x)) %in% FACLST) | !(flatten(class(byvar)) %in% FACLST) ) {
		stop("Invalid variable type")
	## Don't run if there are missing values for the byvar, since this leads to warning in reshape
	} else if ( length(byvar[is.na(byvar)]) > 0 ) {
		stop("<NA> values in byvar")
	} else {
		## Compute frequency and proportion
		freq <- as.data.table(table(x, byvar, exclude="useNA"))
		perc <- as.data.table(prop.table(table(x, byvar, exclude="useNA"), 2))
		## Rename columns so we can merge freq and perc
		setnames(freq, old=c("x","N"), new=c("CAT","freq"))
		setnames(perc, old=c("x","N"), new=c("CAT","prop"))
		## Merge freq and perc on CAT and byvar (changed from << on=.(CAT, byvar) >>)
        y <- freq[perc, on=c("CAT", "byvar")]
		## Convert proportion to percentage and drop prop column (forcing use of standard evaluation)
		set(y, j="perc", value = 100 * y[["prop"]])
		set(y, j="prop", value = NULL)
		## Restructure to have columns for each byvar level
		yt <- reshape(y, v.names=c("freq","perc"), timevar="byvar", idvar="CAT", direction="wide", sep="_")
		return(yt)
	}
}