3

facet_grid and facet_wrap have the scales parameter, which as far as I know allows each plot to adjust the scales of the x and/or y axis to the data being plotted. Since according to the grammar of ggplot x and y are just two among many aesthetics, and there's a scale for each aesthetic, I figured it would be reasonable to have the option of letting each aesthetic be free, but so far I didn't find a way to do it.

I was trying to set it in particular for the Size, since sometimes a variables lives in a different order of magnitude depending on the group I'm using for the facet, and having the same scale for every group blocks the possibility of seeing within-group variation.

A reproducible example:

set.seed(1)
x <- runif(20,0,1)
y <- runif(20,0,1)
groups <- c(rep('small', 10), rep('big', 10))
size_small <- runif(10,0,1)
size_big   <- runif(10,0,1) * 1000

df <- data.frame(x, y, groups, sizes = c(size_small, size_big))

And an auxiliary function for plotting:

basic_plot <- function(df) ggplot(df) + 
  geom_point(aes(x, y, size = sizes, color = groups)) + 
  scale_color_manual(values = c('big' = 'red', 'small' = 'blue')) +
  coord_cartesian(xlim=c(0,1), ylim=c(0,1))

If I we plot the data as is, we get the following:

basic_plot(df)

Non faceted plot

The blue dots are relatively small, but there is nothing we can do. If we add the facet:

basic_plot(df) +
  facet_grid(~groups, scales = 'free')

Faceted plot

The blue dots continue being small. But I would like to take advantage of the fact that I'm dividing the data in two, and allow the size scale to adjust to the data of each plot. I would like to have something like the following:

plot_big <- basic_plot(df[df$groups == 'big',])
plot_small <- basic_plot(df[df$groups == 'small',])

grid.arrange(plot_big, plot_small, ncol = 2)

What I want

Can it be done without resorting to this kind of micromanaging, or a manual rescaling of the sizes like the following?

df %>% 
  group_by(groups) %>% 
  mutate(maximo = max(sizes),
         sizes = scale(sizes, center = F)) %>% 
  basic_plot() +
  facet_grid(~groups)

I can manage to do those things, I'm just trying to see if I'm not missing another option, or if I'm misunderstanding the grammar of graphics.

Thank you for your time!

mgiormenti
  • 793
  • 6
  • 14
  • 2
    Not really, without creating separate plots with separate legends, as you've done. – joran Feb 07 '19 at 16:58
  • 4
    When you facet a plot, you only get one legend. With scales like size, you wouldn't be able to raw just one legend to represent the values for all facets if they each used a different range. With the x and y scales, it's just easier because those scales are rendered as axis rather than put in legends. You could add a column to your name which normalizes the size scale variable within each group so they are directly comparable perhaps. – MrFlick Feb 07 '19 at 17:04
  • That makes a lot of sense! Thank you very much. – mgiormenti Feb 07 '19 at 19:45

1 Answers1

2

As mentioned, original plot aesthetics are maintained when calling facet_wrap. Since you need grouped graphs, consider base::by (the subsetting data frame function) wrapped in do.call:

do.call(grid.arrange, 
        args=list(grobs=by(df, df$groups, basic_plot), 
                  ncol=2,
                  top="Grouped Point Plots"))

Grouped Plot Output

Should you need to share a legend, I always use this wrapper from @Steven Lockton's answer

do.call(grid_arrange_shared_legend, by(df, df$groups, basic_plot))

Shared Legend Plot Output

Parfait
  • 104,375
  • 17
  • 94
  • 125
  • That's somewhat more elegant than what I wrote, thank you. And grid_arrange_shared_legend is a lovely function! – mgiormenti Feb 07 '19 at 19:48