1

Let say I have below ggplot to draw a stacked bar chart

library(ggplot2)

col_define = c('red', 'orange', 'blue', 'lightblue')
names(col_define) = c('A', 'B', 'C', 'D')
data = rbind(data.frame('grp1' = 'X', 'grp2' = c('A', 'B', 'C', 'D'), 'val' = c(1,2,3,4)), data.frame('grp1' = 'Y', 'grp2' = c('A', 'B', 'C', 'D'), 'val' = c(1,2,3,4)+2))
ggplot(data, aes(x = grp1, fill = grp2, y = val)) +
    geom_bar(stat = 'identity', position = 'stack') +
    scale_fill_manual(aesthetics = "fill", values = col_define,
                    breaks = names(col_define))

It is placing all colour in a single legend. However in my case, I basically have 2 groups of colour i.e. one for A & B and second for C & D

I was looking into a similar discussion in ggplot2: Divide Legend into Two Columns, Each with Its Own Title, where there is an approach to group colours of legend using package ggnewscale or relayer

However it looks like, this approach can only be applied in ordinary bar chart, where geom_bar can be called multiple times.

On the contrary, geom_bar, in my case, can't be called multiple times, as it is an whole object

I am looking for some way to use ggnewscale or relayer package in my stack bar chart to group colours in the legend.

As @stefan suggested in one of the answers, a possible way to use geom_col.

However I found that this approach is fairly restrictive, as I cant apply this method for alluvial plot with the same data as below

library(ggalluvial)
ggplot(data,
       aes(x = grp1, stratum = grp2, alluvium = grp2,
           y = val,
           fill = grp2)) +
  geom_flow(aes(fill = grp2), alpha = .3) +
  geom_stratum(aes(color = grp2), alpha = .9) +
  scale_fill_manual(values = col_define, breaks = names(col_define))

Is there a more general approach to group colours in legend?

Bogaso
  • 2,838
  • 3
  • 24
  • 54
  • "On the contrary, `geom_bar`, in my case, can't be called multiple times, as it is an whole object" I don't think thats correct, the `new_scale_fill()` allows for you to call it more than once, even with it being "a whole object", as @stefan answer showed. – Ricardo Semião e Castro Oct 24 '22 at 23:03
  • If I call twice e.g. `scale_fill_manual(aesthetics = "fill", values = col_define, breaks = names(col_define)[1:2], name = '1') + scale_fill_manual(aesthetics = "fill", values = col_define, breaks = names(col_define)[3:4], name = '2')` then I get warning `Scale for 'fill' is already present. Adding another scale for 'fill', which will replace the existing scale.` and only second instance is displayed – Bogaso Oct 24 '22 at 23:14

2 Answers2

2

As the additional information provided asks for a different problem IMHO a second answer is appropriate. Basically it builds on my first answer in that I use ggnewscale to create a grouped legend for a barchart. In second step I extract this legend via cowplot::get_legend to add it to the alluvial plot via patchwork. Once more, this it not elegant but IMHO the easiest way to achieve the desired result:

Note: I tried using ´ggnewscalewithggalluvial` but it seems that the latter is special and a bit stubborn. (; That's why I switched to a different approach.

library(ggplot2)
library(ggnewscale)
library(cowplot)
library(ggalluvial)
library(patchwork)

# Create a grouped legend
p <- ggplot(data, aes(x = grp1, group = grp2, y = val)) +
  geom_col(aes(fill = grp2)) +
  scale_fill_manual(values = col_define, breaks = c("A", "B"), name = "1") +
  new_scale_fill() +
  geom_col(aes(fill = grp2)) +
  scale_fill_manual(values = col_define, breaks = c("C", "D"), name = "2")

p_legend <- cowplot::get_legend(p)

# Alluvial plot without legend
p_alluvial <- ggplot(data,
       aes(x = grp1, stratum = grp2, alluvium = grp2,
           y = val,
           fill = grp2)) +
  geom_flow(aes(fill = grp2), alpha = .3) +
  geom_stratum(aes(color = grp2), alpha = .9) +
  scale_fill_manual(values = col_define, breaks = names(col_define), aesthetics = c("color", "fill")) +
  guides(color = "none", fill = "none")

# Alluvial plot with legend via patchwork
p_alluvial + p_legend + plot_layout(widths = c(10, 1))
#> Warning: `spread_()` was deprecated in tidyr 1.2.0.
#> ℹ Please use `spread()` instead.
#> ℹ The deprecated feature was likely used in the ggalluvial package.
#>   Please report the issue at
#>   <https://github.com/corybrunson/ggalluvial/issues>.
#> Warning: The `.dots` argument of `group_by()` is deprecated as of dplyr 1.0.0.
#> ℹ The deprecated feature was likely used in the dplyr package.
#>   Please report the issue at <https://github.com/tidyverse/dplyr/issues>.

stefan
  • 90,330
  • 6
  • 25
  • 51
1

Not elegant but of course could you simply add a second geom_col on top of the first to get two legends:

library(ggplot2)
library(ggnewscale)

ggplot(data, aes(x = grp1, group = grp2, y = val)) +
  geom_col(aes(fill = grp2)) +
  scale_fill_manual(values = col_define, breaks = c("A", "B"), name = "1") +
  new_scale_fill() +
  geom_col(aes(fill = grp2)) +
  scale_fill_manual(values = col_define, breaks = c("C", "D"), name = "2")

stefan
  • 90,330
  • 6
  • 25
  • 51
  • Thanks for your solution. However I believe this fairly restrictive approach, as I can't use this on `alluvial` plot using `library(ggalluvial) `. I have modified original post with this information – Bogaso Oct 25 '22 at 06:07
  • Hm. First. You have to be fair. My answer gives an answer to your original question which was about the legend for a stacked bar chart. Not meant bad but it's sometimes a bit annoying when people complain about answers not working for their real case when the real case is different from what they asked for and the information provided. While that could happen in such cases I would suggest to simply ask a new question which building on the first one refines the question and clarifies the issue. – stefan Oct 25 '22 at 06:45
  • I really understand that you provided an workable solution. I very much appreciate that. But shouldn't the solution be general so that it can be applied on fairly broad range of possibilities? – Bogaso Oct 25 '22 at 06:49
  • One always tries. But you are expecting too much. Working out general solutions which work for each circumstance and for each extension package is impossible and would require much more effort and time. Put differently the preferred option would be a package which allows to create a grouped legend without using ggnewscale. But quite often answers are simply "hacks" which show a way to achieve a desired result with the least effort and which unfortunately are not always generalizable. But see my second answer for an another hack. (; – stefan Oct 25 '22 at 07:07
  • 1
    Many thanks for your second answer. I think this is perfect solution to my original problem (I provided only dummy data as original data is restrictive to share) – Bogaso Oct 25 '22 at 07:12