Pass by value in R

Question

When trying to call grid.arrange to put multiple plots on a same ggplot2 graph, I first build a list of the plots I want. Then I build the corresponding argument list to call grid.arrange, as was explained in a previous question. This is my code (my dataframe is called manip):

args.list <- NULL;
plot.list <- NULL;
for (m in names(manip[2:10])) {
  plot.list <- c(plot.list, list(qplot(manip$side, y=manip[,m],ylab=m))
}
args.list <- c(plot.list, 1, 9)
names(args.list) <- c(names(manip)[2:10], list("nrow","ncol"))
do.call(grid.arrange, args.list)

This works, except that the 9 graphs are exactly the same! After checking, it turns out that the data is always the one corresponding to m=10. So my guess was that the value of m is not assigned in the loop, but evaluated later. However, the label ylab=m is assigned correctly and is different for all the graphs.

So I don't really get what the difference is and how the interpreter chooses when to evaluate m for the plots. Can someone explain?

Answer 1

The behavior is due to the lazy evaluation of R.

Here is a minimal(?) example:

d <- 1:3

args.list <- NULL;
plot.list <- NULL;
for (m in 1:3) {
 plot.list <- c(plot.list, list(qplot(d[m], d[m], ylab=letters[m])))
}

args.list <- c(plot.list, nrow=1, ncol=3)
do.call(grid.arrange, args.list)

in this case, d[m] is evaluated at the call of do.call. so m is 3 for all panel.

here is a workaround:

d <- 1:3

args.list <- NULL;
plot.list <- NULL;
for (m in 1:3) {
  plot.list <- c(plot.list,
    list(qplot(d, d, data=data.frame(d=d[m]), ylab=letters[m])))
}

args.list <- c(plot.list, nrow=1, ncol=3)
do.call(grid.arrange, args.list)

in this case, d[m] is evaluated at the call of qplot, and the d[m] is stored in the output object of qplot.

so, the simple solution is to pass data to qplot() or ggplot().

Answer 2

I will first answer your question and then show an alternative using a facet plot.

Edited

The following, much simplified, code seems to work:

library(gridExtra)
manip <- mtcars
plot.list <- lapply(2:11, 
                    function(x)qplot(manip$mpg, y=manip[, x], 
                    ylab=names(manip)[x]))
do.call(grid.arrange, c(plot.list, nrow=10))

It produces this ugly plot:

---

Without knowing your objectives, it is dangerous to try and give advice, I know. Nonetheless, have you considered using facets for your plot instead?

The following code is much simpler, executes quiker and produces a graph that is easier to interpret:

library(reshape2)
manip <- mtcars
mmanip <- melt(manip, id.vars="mpg")
str(mmanip)
ggplot(mmanip, aes(x=mpg, y=value)) + 
    geom_point(stat="identity") + 
    facet_grid(.~variable, scales="free")