Unexpected plot using xlim() and facet_wrap(…, scales=“free_y” )

Question

I have a problem that can be reproduced using the mt data set:

require(ggplot2)
mt <- ggplot(mtcars, aes(mpg, wt, colour = factor(cyl))) + geom_point()
mt + facet_wrap(~ cyl, scales = "free_y")
mt + xlim(c(15,30)) + facet_wrap(~ cyl, scales = "free_y")

In the second plot, why does the y axis of the 3rd cell ("8") go up up to 5.5 if there are no points there? How can I get that axis to span to the actual range of the displayed points (ie, something like from 3 to 4.25)? Are not all points outside the x limits of c(15,30) being set to NA by xlim()?

Answer 1

There are some points there. You just cut them off. To get better scales, you may want to filter those points out first:

mt <- mtcars %>% 
  filter(mpg>=15, mpg<=30) %>% 
  ggplot(aes(mpg, wt, colour = factor(cyl))) + geom_point()
mt + facet_wrap(~ cyl, scales = "free_y")
mt + xlim(c(15,30)) + facet_wrap(~cyl, scales = "free_y")

You get:

Answer 2

I have understood the issue thanks to a recent discussion in ggplot google list 1

The question is that the result of xlim() is not really setting points (which imply both x and y coordinates) to NA, but only the x-coordinate is set to NA: "The corresponding y values are still available and hence used to calculate the range of y axis."

I have to add though that this behavior of xlim() is very counter-intuitive, if not illogical. If the user sets xlim(c(15,30), it is because he/she does not want to see points within that range, so automatic y limits should adapt in accordance. Or at least this should be an option, sort of na.rm=TRUE.