Counts & Percentages in xTable, Sweave, R, cross tabulations

Question

Edit: Building off of aL3xa's answer below, I've modified his syntax below. Not perfect, but getting closer. I still haven't found a way to make xtable accept \multicolumn{} arguments for columns or rows. It also appears that Hmisc handles some of these type of tasks behind the scenes, but it looks like a bit of an undertaking to understand what's going on there. Does anyone have experience with the latex function in Hmisc?

ctab <- function(tab, dec = 2, margin = NULL) {
    tab <- as.table(tab)
    ptab <- paste(round(prop.table(tab, margin = margin) * 100, dec), "%", sep = "")
    res <- matrix(NA, nrow = nrow(tab) , ncol = ncol(tab) * 2, byrow = TRUE)
    oddc <- 1:ncol(tab) %% 2 == 1
    evenc <- 1:ncol(tab) %% 2 == 0
    res[,oddc ] <- tab
    res[,evenc ] <- ptab
    res <- as.table(res)
    colnames(res) <- rep(colnames(tab), each = 2)
    rownames(res) <- rownames(tab)
    return(res)
}

I would like to create a table formatted for LaTeX output that contains both the counts and percentages for each column or variable. I have not found a ready made solution to this problem, but feel I must be recreating the wheel to some extent.

I have developed a solution for straight tabulations, but am struggling with adopting something for a cross tabulation.

First some sample data:

#Generate sample data
dow <- sample(1:7, 100, replace=TRUE)
purp <- sample(1:4, 100, replace=TRUE)
dow <- factor(dow, 1:7, c("Mon", "Tues", "Wed", "Thurs", "Fri", "Sat", "Sun"))
purp <- factor(purp, 1:4, c("Business", "Commute", "Vacation", "Other"))

And now the working straight tab function:

customTable <- function(var, capt = NULL){
    counts <- table(var)
    percs <- 100 * prop.table(counts)       

    print(
        xtable(
            cbind(
                Count = counts
                , Percent = percs
            )
        , caption = capt
        , digits = c(0,0,2)
        )
    , caption.placement="top"
    )
}

#Usage
customTable(dow, capt="Day of Week")
customTable(purp, capt="Trip Pupose")

Does anyone have any suggestions for adopting this for cross tabulations (i.e. day of week BY trip purpose)? Here is what I've currently written, which does NOT use the xtable library and ALMOST works, but is not dynamic and is quite ugly to work with:

#Create table and percentages
a <- table(dow, purp)
b <- round(prop.table(a, 1),2)

#Column bind all of the counts & percentages together, this SHOULD become dynamic in future
d <- cbind( cbind(Count = a[,1],Percent =  b[,1])
        , cbind(Count = a[,2], Percent = b[,2])
        , cbind(Count = a[,3], Percent = b[,3])
        , cbind(Count = a[,4], Percent = b[,4])
)

#Ugly function that needs help, or scrapped for something else
crossTab <- function(title){
    cat("\\begin{table}[ht]\n")
    cat("\\begin{center}\n")
    cat("\\caption{", title, "}\n", sep="") 

    cat("\\begin{tabular}{rllllllll}\n")
    cat("\\hline\n")

    cat("", cat("", paste("&\\multicolumn{2}{c}{",colnames(a), "}"), sep = ""), "\\\\\n", sep="")
    c("&", cat("", colnames(d), "\\\\\n", sep=" & "))
    cat("\\hline\n")
    c("&", write.table(d, sep = " & ", eol="\\\\\n", quote=FALSE, col.names=FALSE))

    cat("\\hline\n")
    cat("\\end{tabular}\n")
    cat("\\end{center}\n")
    cat("\\end{table}\n")   
}   

crossTab(title = "Day of week BY Trip Purpose")

Answer 1

How would this work for you?

library(reshape)
library(plyr)
df <- data.frame(dow = dow, purp = purp)

df.count <- count(df)
df.count <- ddply(df.count, .(dow), transform, p = round(freq / sum(freq),2))

df.m <- melt(df.count)

df.print <- cast(df.m, dow ~ purp + variable)

library(xtable)
xtable(df.print)

It doesn't give you nice multicolumns, and I don't have enough experience with xtable to figure out if that's possible. However, if you're going to be writing custom functions, you might try one which operates over the column names of df.print. You might be even able to write one sufficiently general to take all manner of recast data frames as input.

Edit: Just thought of a good solution to get you closer. After creating df.m

df.preprint <- ddply(df.m, .(dow, purp), function(x){
        x <- cast(x, dow ~ variable)
        x$value <- paste(x$freq, x$p, sep = " / ")
        return(c(value = x$value))
     }
)

df.print <- cast(df.preprint, dow ~ purp)

print(xtable(df.print), include.rownames = F)

Now, every cell will contain N / percent values

Answer 2

Great question, this one's bothering me for a while (it's not that hard, it's just me being lazy as hell... as usual). However... though the question's great, your approach, I'm afraid, isn't. There's priceless package called xtable that you can (mis)use. Besides, this issue is too common - there's a great chance that there's already some ready-made solution sitting somewhere on the Internets.

One of these days I'm about to work it out once and for all (I'll post the code on GitHub). The main idea goes a little bit like this: would you like frequency and/or percentage values within one cell (separated by \) or rows with absolute and relative frequencies (or %) in succession? I'd go with the 2^nd one, so I'll post a "first-aid" solution for now:

ctab <- function(tab, dec = 2, ...) {
  tab <- as.table(tab)
  ptab <- paste(round(prop.table(tab) * 100, dec), "%", sep = "")
  res <- matrix(NA, nrow = nrow(tab) * 2, ncol = ncol(tab), byrow = TRUE)
  oddr <- 1:nrow(tab) %% 2 == 1
  evenr <- 1:nrow(tab) %% 2 == 0
  res[oddr, ] <- tab
  res[evenr, ] <- ptab
  res <- as.table(res)
  colnames(res) <- colnames(tab)
  rownames(res) <- rep(rownames(tab), each = 2)
  return(res)
}

Now try something like:

data(HairEyeColor)           # load an appropriate dataset
tb <- HairEyeColor[, , 1]    # choose only male respondents
ctab(tb)
      Brown  Blue   Hazel Green
Black 32     11     10    3    
Black 11.47% 3.94%  3.58% 1.08%
Brown 53     50     25    15   
Brown 19%    17.92% 8.96% 5.38%
Red   10     10     7     7    
Red   3.58%  3.58%  2.51% 2.51%
Blond 3      30     5     8    
Blond 1.08%  10.75% 1.79% 2.87%

Make sure you loaded xtable package and use print (it's a generic function, so you must pass a xtable classed object). It's important that you suppress the row names. I'll optimize this one tomorrow - it should be xtable compatible. It's 3AM in my time zone, so with these lines I'll end my answer:

print(xtable(ctab(tb)), include.rownames = FALSE)

Cheers!

Answer 3

I wasn't able to figure out how to generate a multi column header using xtable, but I did realize that i could concatenate my counts & percentages into the same column for printing purposes. Not ideal, but seems to get the job done. Here's the function I've written:

ctab3 <- function(row, col, margin = 1, dec = 2, percs = FALSE, total = FALSE, tex = FALSE, caption = NULL){
    tab <- as.table(table(row,col))
    ptab <- signif(prop.table(tab, margin = margin), dec)

    if (percs){

        z <- matrix(NA, nrow = nrow(tab), ncol = ncol(tab), byrow = TRUE) 
        for (i in 1:ncol(tab)) z[,i] <- paste(tab[,i], ptab[,i], sep = " ")
        rownames(z) <- rownames(tab)
        colnames(z) <- colnames(tab)

        if (margin == 1 & total){
            rowTot <- paste(apply(tab, 1, sum), apply(ptab, 1, sum), sep = " ")
            z <- cbind(z, Total = rowTot)
        } else if (margin == 2 & total) {
            colTot <- paste(apply(tab, 2, sum), apply(ptab, 2, sum), sep = " ")
            z <- rbind(z,Total = colTot)
        }
    } else {
        z <- table(row, col)    
    }
ifelse(tex, return(xtable(z, caption)), return(z))
}

Probably not the final product, but does allow for some flexibility in parameters. At the most basic level, is only a wrapper of table() but can also generate LaTeX formatted output as well. Here is what I ended up using in a Sweave document:

<<echo = FALSE>>=
for (i in 1:ncol(df)){
    print(ctab3(
        col = df[,1]
        , row = df[,i]
        , margin = 2
        , total = TRUE
        , tex = TRUE
        , caption = paste("Dow by", colnames(df[i]), sep = " ")
    ))
}
@