Show the percentage instead of count in histogram using ggplot2 | R

Question

I'm using a histogram to plot my 3 groups data. But as histogram do, it counts how much each group have those values (in x-axis) and what I want is to the percentage of how much (in %) this value appears/occurs.

Here is my generated figure, I use this regular code to plot the histogram:

ggplot2.histogram(data=dat, xName='dens',
                  groupName='lines', legendPosition="top",
                  alpha=0.1) + 
  labs(x="X", y="Count") +
  theme(panel.border = element_rect(colour = "black"),
        panel.grid.minor = element_blank(), 
        axis.line = element_line(colour = "black")) + 
  theme_bw()+
  theme(legend.title=element_blank())

Any ideas/suggestions?

try adding `+ scale_y_continuous(labels = scales::percent)` ? (better to give reproducible data!) — moodymudskipper, Oct 07 '18 at 01:16
No, it seems to multiply y-axis* 100 and it's changed to 300,000% from my above figure @Moody_Mudskipper — LamaMo, Oct 07 '18 at 03:08
indeed... well you're using a custom function `ggplot2.histogram` and we don't have its code so we can't do much, again, better to give reproducible data — moodymudskipper, Oct 07 '18 at 08:57
would it work if you add `+ aes(y=stat(count)/sum(stat(count))) + scale_y_continuous(labels = scales::percent)` ? — moodymudskipper, Oct 07 '18 at 09:09
cool! I added a proper answer, please consider accepting it if it solves your issue — moodymudskipper, Oct 07 '18 at 15:07

score 4 · Accepted Answer · answered Oct 07 '18 at 15:06

We can replace the y aesthetic by the relative value of the count computed statistic, and set the scale to show percentages :

ggplot2.histogram(data=dat, xName='dens',
                  groupName='lines', legendPosition="top",
                  alpha=0.1) + 
  labs(x="X", y="Count") +
  theme(panel.border = element_rect(colour = "black"),
        panel.grid.minor = element_blank(), 
        axis.line = element_line(colour = "black")) + 
  theme_bw()+
  theme(legend.title=element_blank()) + 
  aes(y=stat(count)/sum(stat(count))) + 
  scale_y_continuous(labels = scales::percent)

y=stat(count)/sum(stat(count), this make the different. Thanks a lot — LamaMo, Oct 07 '18 at 15:22

coffeinjunky · Answer 2 · 2018-10-07T11:31:55.077

If I understand you correctly, then fill would answer your question?

For instance,

mtcars %>% 
ggplot(aes(x = factor(gear), group = factor(cyl), fill = factor(cyl))) + 
geom_bar(position = "fill")

Here, you don't have the counts any longer, but for each value along the x-axis, you have the percentage of each group (here: cylinder) plotted.

If this is not what you want, a general recommendation is to compute the data that you want to be plotted first, and then to plot it. That is, many people think it is generally advisable to separate computation/transformation/aggregation from plotting.

To follow up on my suggestion to separate computation from visualisation, let's consider the mtcars dataset and focus on gear and carb.

with(mtcars, table(gear, carb))
    carb
gear 1 2 3 4 6 8
   3 3 4 3 5 0 0
   4 4 4 0 4 0 0
   5 0 2 0 1 1 1

For instance, you see that 3 (out of 32) observations have gear = 3, carb = 1, which is a bit less than 10%. Similarly, 4 observations have gear = 4, carb = 1, which is a bit more than 10%. Let's get the percentages directly:

with(mtcars, prop.table(table(gear, carb)))
    carb
gear       1       2       3       4       6       8
   3 0.09375 0.12500 0.09375 0.15625 0.00000 0.00000
   4 0.12500 0.12500 0.00000 0.12500 0.00000 0.00000
   5 0.00000 0.06250 0.00000 0.03125 0.03125 0.03125

I have used prop.table here which also has a margin argument. That is, if you wanted to know conditional percentages, you could easily adjust this (see below). Let's stay with this for the time being, though. Let's say we want to visualize this now after we have computed the numbers, we could simply call the following:

with(mtcars, prop.table(table(gear, carb))) %>% 
             as.data.frame() %>% 
             ggplot(aes(x = factor(carb), y = Freq, group = factor(gear), fill = factor(gear))) + 
             geom_bar(stat = "identity")

which would give us:

Now imagine you want to get the conditional version, e.g.

with(mtcars, prop.table(table(gear, carb), margin = 1))
    carb
gear         1         2         3         4         6         8
   3 0.2000000 0.2666667 0.2000000 0.3333333 0.0000000 0.0000000
   4 0.3333333 0.3333333 0.0000000 0.3333333 0.0000000 0.0000000
   5 0.0000000 0.4000000 0.0000000 0.2000000 0.2000000 0.2000000

Notice how each row sums up to 1. This can be plotted in the same way:

with(mtcars, prop.table(table(gear, carb), margin = 1)) %>% 
as.data.frame() %>% 
ggplot(aes(x = factor(carb), y = Freq, group = factor(gear), fill = factor(gear))) + 
geom_bar(stat = "identity")

Note the similarity to the smoothed version produced by:

mtcars %>% 
ggplot(aes(x = factor(carb), group = factor(gear), fill = factor(gear))) + 
geom_density(alpha = 0.5)

No, it really differs, my x-axis has many values within each of these 3 groups (not exactly the same example you showed) — LamaMo, Oct 07 '18 at 03:10
@SaraWasl I am not sure i fully understand. Can you please post some sample data and describe a bit more what your desired graph should look like? You can edit your post for that, but leave me a comment here. — coffeinjunky, Oct 07 '18 at 09:26

Show the percentage instead of count in histogram using ggplot2 | R

2 Answers2