r - How to scale the size of line and point separately in ggplot2

Question

The code is as follows:

set.seed(123)
d1=data.frame(x=runif(10),y=runif(10),z=runif(10,1,10))
d2=data.frame(x=runif(10),y=runif(10),z=runif(10,100,1000))
ggplot()+geom_point(aes(x,y,size=z),data=d1)+
geom_line(aes(x,y,size=z),data=d2)

And the result is like this:

The size of points are too small so I want to change its size by scale_size. However, it seems both lines and points are influenced. So I wonder if there is a way to scale lines and points separately with a separate legend?

Answer 1

The two ways I can think of are 1) combining two legend grobs or 2) hacking another legend aesthetic. Both of these were mentioned by @Mike Wise in the comments above.

Approach #1: combining 2 separate legends in the same plot using grobs.

I used code from this answer to grab the legend. Baptiste's arrangeGrob vignette is a useful reference.

library(grid); library(gridExtra)

#Function to extract legend grob
g_legend <- function(a.gplot){
  tmp <- ggplot_gtable(ggplot_build(a.gplot))
  leg <- which(sapply(tmp$grobs, function(x) x$name) == "guide-box")
  legend <- tmp$grobs[[leg]]
  legend
}

#Create plots
p1 <- ggplot()+  geom_point(aes(x,y,size=z),data=d1) + scale_size(name = "point")
p2 <- ggplot()+  geom_line(aes(x,y,size=z),data=d2) + scale_size(name = "line")
p3 <- ggplot()+  geom_line(aes(x,y,size=z),data=d2) + 
        geom_point(aes(x,y, size=z * 100),data=d1)  # Combined plot
legend1 <- g_legend(p1)
legend2 <- g_legend(p2)
legend.width <- sum(legend2$width)  

gplot <- grid.arrange(p3 +theme(legend.position = "none"), legend1, legend2,
             ncol = 2, nrow = 2,
             layout_matrix = rbind(c(1,2 ),  
                                   c(1,3 )), 
             widths = unit.c(unit(1, "npc") - legend.width, legend.width))
grid.draw(gplot)

Note for printing: use arrangeGrob() instead of grid.arrange(). I had to use png; grid.draw; dev.off to save the (arrangeGrob) plot.

Approach #2: hacking another aesthetic legend.

MilanoR has a great post on this, focusing on colour instead of size. More SO examples: 1) discrete colour and 2) colour gradient.

#Create discrete levels for point sizes (because points will be mapped to fill)
d1$z.bin <- findInterval(d1$z, c(0,2,4,6,8,10), all.inside= TRUE)  #Create bins

#Scale the points to the same size as the lines (points * 100).  
#Map points to a dummy aesthetic (fill)
#Hack the fill properties.
ggplot()+  geom_line(aes(x,y,size=z),data=d2) + 
  geom_point(aes(x,y, size=z * 100, fill = as.character(z.bin)),data=d1) +
  scale_size("line", range = c(1,5)) + 
  scale_fill_manual("points", values = rep(1, 10) , 
                    guide = guide_legend(override.aes = 
                              list(colour = "black", 
                              size = sort(unique(d1$z.bin)) )))