R & ggplot2: How to get arrows under the axis label?

Question

I'm about to plot odds ratios with R/ggplot2 and I want to add two arrows underneath or next to the X-axis label. One pointing to the left, one pointing to the right, showing de-/increasing influence. I've tried lots of things, like geom_path, geom_line, without great success. This is my code:

forest <- function(d, xlab="Odds Ratio", ylab="Influencing variables"){
require(ggplot2)
p <- ggplot(d, aes(x=ordered(x, levels=rev(x)), y=y, ymin=ylo, ymax=yhi)) +
geom_pointrange() +
geom_segment(aes(x = 0, xend = 0, y= 1, yend= 2)) +
coord_flip() +
geom_hline(aes(yintercept=1), lty=2) +
ylab(xlab) +
xlab(ylab) +
scale_y_log10() 
return(p)
}
##Test Data
data <- data.frame( x   = c("A","B","C","D","E","F","G","H","I"),
                y   = c(1.782,0.136,0.978,0.645,0.518,1.474,0.855,0.673,0.369))
data <- transform(data, ylo = (0.719,0.046,0.945,0.295,0.188,0.577,0.407,0.310,0.145), 
        yhi = c(4.420,0.398,1.012,1.411,1.424,3.768,1.798,1.460,0.940))
forest(data)

Adding a line like geom_line(aes(x=1), arrow=arrow(length=unit(0.15,"cm")), colour="black", size=1) brings some arrows but they collide with my original data.
Thanks in advance for your solution, help or hint!
Marc

Answer 1

This can easily be done using mathematical annotation in R expression(x <-> y). A table of available options can be found here. expression also allows latex like math symbols.

p1 <- qplot(x = 1:10, y = 1:10, geom = "blank") 
p1 <- p1 + xlab(expression(decreasing %<->% increasing))
p1 <- p1 + ylab(expression(down %->% up))
p1 <- p1 + ggtitle(expression("Darcy's law" %->% q == -k*frac(Delta*h,Delta*l)))
p1 <- p1 + theme_bw(base_size=14)

Answer 2

This is a pretty old post, but thought I would share another option for anyone who stumbles upon this.

I have a situation which really requires that I have arrows outside of the plotting region (and don't care to involve grid). As shown below, using expressions and symbols might be a good option for some:

p1 <- qplot(x = 0:12, y = 0:12, geom = "blank") 
p1 <- p1 + ylab(expression(symbol('\256')))
p1 <- p1 + xlab(expression(symbol('\256')))
p1 <- p1 + theme(axis.title.y = element_text(size = 30))
p1 <- p1 + theme(axis.title.x = element_text(size = 30))

p1  

The theme modifications aren't absolutely necessary but you will probably want to increase the size of the arrows at least a bit.

By default, the symbol() function takes Adobe Symbol codes which are outline in this document.

If you want to add some text to the labels (just adding to the expression would probably make it too big), the following should get you started:

grid.text(label="X", x = 0.645, y = 0.02)
grid.text(label="Y", x = 0.1, y = 0.5, rot=90)
grid.gedit("GRID.text", gp=gpar(fontsize=15))

Answer 3

I'm doing a bit of guessing about what you want, exactly, as I'm not sure what you were attempting with geom_segment, whether that was part of your data, or an attempt to create an arrow. But you can build on geom_segment to get something like what you describe:

#Your test data, with a small typo repaired
dd <- data.frame(x = c("A","B","C","D","E","F","G","H","I"),
                 y = c(1.782,0.136,0.978,0.645,0.518,1.474,0.855,0.673,0.369))
dd <- transform(dd, ylo = c(0.719,0.046,0.945,0.295,0.188,0.577,0.407,0.310,0.145), 
        yhi = c(4.420,0.398,1.012,1.411,1.424,3.768,1.798,1.460,0.940))

#Create a separate data frame for the arrow labels
arrowLab <- data.frame(lab = c("Increasing","Decreasing"),
                       x = c(0.15,0.15),y = c(10^0.25,10^(-0.25)))

ggplot(data = dd, aes(x=ordered(x, levels=rev(x)), y=y)) +
        geom_pointrange(aes(ymin=ylo, ymax=yhi)) +
        geom_segment(aes(x = 0, xend = 0, y= 1, yend= 2),
                         arrow=arrow(length=unit(0.2,"cm"))) +
        geom_segment(aes(x = 0, xend = 0, y= 1, yend= 10^(-0.25)),
                         arrow=arrow(length=unit(0.2,"cm"))) +
        geom_text(data = arrowLab,aes(x=x,y=y,label = lab),size = 3) +
        coord_flip() +
        geom_hline(aes(yintercept=1), lty=2) +
        scale_y_log10() 

Note that I hard-coded much of the positioning, so you'll likely want to tinker with those values to get what you want, or you could try to choose them programmatically based on your data.