I think I first learned about this idea from Andrew Gelman: that a separate legend requires a lot of back-and-forth glances, so it’s better to put the labels right by the relevant bits. For example, like this: rather than this:
I’ve adopted this approach as much as possible, though it often requires a bit of work (and thought) to get the labels in just the right place. Here’s the code I used for the first of those pictures. (It was relatively easy here.)
load(con <- url("http://www.biostat.wisc.edu/~kbroman/blog/mapexpansion.RData")) close(con) plot(dat[,1], dat[,4], xlab=expression(paste("Generation ", F[k])), ylab="Map expansion", type="l", lwd=2, col="black", las=1, xaxs="i", yaxs="i", ylim=c(0, 7)) for(i in 2:3) lines(dat[,1], dat[,i], lwd=2, col=c("blue", "red")[i-1]) text(17, dat[18,-1]-0.1, paste(colnames(dat)[-1], "-way", sep=""), adj=c(0,1), col=c("blue","red","black"))
The aim of the directlabels package is get this without effort.
I need to switch to either
ggplot2 (vs. base graphics). But I should be doing that anyway. I’ll try
lattice for this. I rearrange the data a bit, call
xyplot and then use
direct.labels to make the actual plot, as follows. (Note the use of
gl, which I just learned about from Richie Cotton.)
library(lattice) library(directlabels) dat2 <- with(dat, data.frame(gen=rep(gen,3), mapexpansion=c(two, four, eight), cross=gl(3, nrow(dat), labels=c("two","four","eight")))) p <- xyplot(mapexpansion ~ gen, data=dat2, groups=cross, type="l", lwd=2) direct.label(p)
And here’s what I get:
For a final figure for publication, one will want to do some editing by hand, but for day-to-day graphics, this looks really useful. The following is the “real” version of the above figure, from a paper under review, using a mixture of legend and direct labels:
Here’s another figure I’m quite proud of, from a paper nearing submission.