Question

I’m trying to do a plot that has most data points drawn normally, but one set of data points with a different sized symbol. I want the legend to show the same: most points shown normally, but the exception drawn with a different sized symbol. Here is a short bit of code:

library(ggplot2)
x = c(1,2,1,2,3)
y = c(1,2,3,4,3)
vendor = c("x", "x", "y", "y", "z")
df = data.frame(x,y,vendor)

p <- ggplot(df) +
     aes_string(x="x", y="y", color="vendor") +
     geom_point(size=3, data=subset(df, vendor!="z")) +
     geom_point(size=5, data=subset(df, vendor=="z"))
ggsave("foo.pdf")

The problem is that in the resulting legend, all points are drawn with the larger (size=5) symbol, not just those with vendor z. I want vendor z drawn with the larger point in the legend, and the others drawn with size=3.

(Bonus question: What I really want is a larger thick outlined symbol: instead of a circle, I want a donut. I realize that shape=2 will draw an outlined circle, but it is very thin. I’d rather have a thicker outlined circle. I want to do the same with a triangle. Any easy way to do this?)

Maybe I applied it wrong, but following this advice:

ggplot2: Making changes to symbols in the legend

with the addition of the “guides” line did not help:

guides(size = guide_legend(override.aes = list(shape = 1)))

i.e. same output, with size=5 symbols for all three vendors in the legend.

EDITED: Fantastic answer, which I quickly implemented. Now I’ve added lines:

library(ggplot2)
x = c(1,2,1,2,3)
y = c(1,2,3,4,3)
vendor = c("x", "x", "y", "y", "z")
df = data.frame(x,y,vendor)

df$vendor_z <- df$vendor=="z"     # create a new column 

ggplot(df) +
  aes_string(x = "x", y = "y", color = "vendor", size = "vendor_z") +
  geom_point() +
  geom_line(size=1.5) +   # this is the only difference
  scale_size_manual(values = c(3, 5), guide = FALSE) 
  guides(colour = guide_legend(override.aes = list(size = c(3, 3, 5))))

ggsave("foo.pdf")

and now the size of the legend is back down to 3 again for all dots, including the ones with vendor z. Any ideas on how to fix this?

Answer 1

The size is not applied to the legend since size is outside aes_string. Furtermore, the work with ggplot will be much easier if you create an additional column indicating whether vendor == "z".

Here’s a solution for part 1:

df$vendor_z <- df$vendor=="z"     # create a new column 

ggplot(df) +
  aes_string(x = "x", y = "y", color = "vendor", size = "vendor_z") +
  geom_point() +
  scale_size_manual(values = c(3, 5), guide = FALSE) + 
  guides(colour = guide_legend(override.aes = list(size = c(3, 3, 5))))

Note that vendor_z is as argument of aes_string. This will tell ggplot to create a legend for the size characteristic. In the function scale_size_manual, the values for size are set. Furthermore, guide = FALSE avoids a second legend for size only. Finally, the size values are applied to the color legend.

Part2: a “donut” symbol

The size of the lines for circles cannot be modified in ggplot. Here is a workaround:

ggplot(df) +
  aes_string(x = "x", y = "y", color = "vendor", size = "vendor_z") +
  geom_point() +
  geom_point(data = df[df$vendor_z, ], aes(x = x, y = y),
             size = 3, shape = 21, fill = "white", show_guide = FALSE) +
  scale_size_manual(values = c(3, 5), guide = FALSE) + 
  guides(colour = guide_legend(override.aes = list(size = c(3, 3, 5))))

Here, a single point is drawn using geom_point and a subset of the data (df[df$vendor_z, ]). I chose a size of 3 since this is the value of the smaller circles. The shape 21 is a circle for which a fill colour could be specified. Finally, show_guide = FALSE avoids that the legend characteristics are overwritten by the new shape.

Edit: part 3: Add lines

You could suppress the legend for geom_line with the argument show_guide = FALSE:

ggplot(df) +
  aes_string(x = "x", y = "y", color = "vendor", size = "vendor_z") +
  geom_point() +
  geom_line(size=1.5, show_guide = FALSE) +   # this is the only difference
  scale_size_manual(values = c(3, 5), guide = FALSE) +
  guides(colour = guide_legend(override.aes = list(size = c(3, 3, 5))))