R ggplot2 stacked barplot by percentage with several categorical variables

Question

This is a simple question, but I'm having difficulty understanding the format required by ggplot2:

I have the following data.table in R,

print(dt)
    ID       category      A    B   C     totalABC                                                                                                                                                                                                                                         
1:  10      group1        1    3   0      4                                                                                                                                                                                                                                         
2:  11      group1        1   11   1      13                                                                                                                                                                                                                                         
3:  12      group2        15  20   2      37                                                                                                                                                                                                                                         
4:  13      group2        6   12   2      20                                                                                                                                                                                                                                         
5:  14      group2        17  83   6      106   
...

My goal is to create a proportional stacked bar graph as in this example: https://rpubs.com/escott8908/RGC_Ch3_Gar_Graphs

where the percentages of X/totalABC, where X is category_type either A, B, or C. I would also like to perform this by category, e.g. the x-axis values should be group1, group2, etc.

As a concrete example, in the case of group1, there are 4+13=17 total elements.

The percentages would be percent_A = 11.7%, percent_B = 82.3%, percent_C = 5.9%

The correct ggplot2 solution appears to be:

library(ggplot2)
pp = ggplot(dt, aes(x=category, y=percentage, fill=category_type)) +                                                                                                                                                                                                                               
          geom_bar(position="dodge", stat="identity")

My confusion: how would I create a single percentage column that corresponds to three categorical values?

If the above is incorrect, how would I format my data.table to create the stacked barplot?

use `position="fill"` instead of `position="dodge"` – A Gore Jun 26 '17 at 16:41 — A Gore, Jun 26 '17 at 16:41

M-- · Answer 1 · 2017-06-26T18:55:25.860

3

You can use the following code:

melt(data.frame( #melt to get each variable (i.e. A, B, C) in a single row
     dt[,-1] %>% #get rid of ID
            group_by(category) %>% #group by category
                  summarise_each(funs(sum))), #get the summation for each variable
                  id.vars=c("category", "totalABC")) %>% 
ggplot(aes(x=category,y=value/totalABC,fill=variable))+ #define the x and y 
       geom_bar(stat = "identity",position="fill") + #make the stacked bars
                scale_y_continuous(labels = scales::percent) #change y axis to % format

which will plot:

enter image description here

Data:

dt <- structure(list(ID = 10:14, category = structure(c(1L, 1L, 2L, 
    2L, 2L), .Label = c("group1", "group2"), class = "factor"), A = c(1L, 
    1L, 15L, 6L, 17L), B = c(3L, 11L, 20L, 12L, 83L), C = c(0L, 1L, 
    2L, 2L, 6L), totalABC = c(4L, 13L, 37L, 20L, 106L)), .Names = c("ID", 
    "category", "A", "B", "C", "totalABC"), row.names = c(NA, -5L
    ), class = c("data.table", "data.frame"), .internal.selfref = <pointer: 0x0000000000100788>)

What if you want to stick to the code that you had for plotting?

In that case, you can use this to get the percentage:

df <- melt(data.frame( #melt to get each variable (i.e. A, B, C) in a single row
        dt[,-1] %>% #get rid of ID
          group_by(category) %>% #group by category
            summarise_each(funs(sum))), #get the summation for each variable
              id.vars=c("category", "totalABC")) %>% 
                mutate(percentage = dtf$value*100/dtf$totalABC)

But need to modify your ggplot to get the stacked bars correctly:

#variable is the column carrying category_type
#position dodge make the bars to be plotted next to each other 
#while fill makes the stacked bars
ggplot(df, aes(x=category, y=percentage, fill=variable)) +           
       geom_bar(position="fill", stat="identity")

edited Jun 26 '17 at 18:55

answered Jun 26 '17 at 17:31

M--

25,431
8
61
93

1

Thank you for the explanatory comments! – ShanZhengYang Jun 26 '17 at 17:32
Oh, I've defined the data.table as `dt`, not `df`. Just to be consistent for future readers – ShanZhengYang Jun 26 '17 at 17:41
What does `fill=variable` mean in this context? – ShanZhengYang Jun 26 '17 at 17:43
@ShanZhengYang after you perform melt function a column called variable would be filled with A, B or C in each row. So, `fill=variable` defines the category. – M-- Jun 26 '17 at 17:45
I understand now. Thank you. One last question regarding this problem: let's say that I wanted to sort the stacked barplots by the size of the percentages, `A`. At the moment, the larger percentage of A/total of group 2 is on the right. How could I place this on the left, i.e. decreasing % of A? – ShanZhengYang Jun 26 '17 at 17:52
@ShanZhengYang That's a bit hard question. Each group goes upto 100 percent. if it wasn't like that I am pretty sure the're is a thread that can help you but in this case, let me give it some thought. – M-- Jun 26 '17 at 18:03
I think you might have misunderstood me. The percentage A/totalABC is always less than 100 – ShanZhengYang Jun 26 '17 at 18:07
@ShanZhengYang No. I got your question.But you want to order group 1 and 2 and by looking at the structure of the data you'd figure out that it's not that straightforward. – M-- Jun 26 '17 at 18:24
Let us [continue this discussion in chat](http://chat.stackoverflow.com/rooms/147649/discussion-between-masoud-and-shanzhengyang). – M-- Jun 26 '17 at 18:42

score 1 · Accepted Answer · edited Apr 05 '19 at 16:26

1

Here's a solution:

require(data.table)
require(ggplot2)
require(dplyr)

melt(dt,measure.vars = c("A","B","C"),
     variable.name = "groups",value.name = "nobs") %>%
 ggplot(aes(x=category,y=nobs,fill=groups)) + 
  geom_bar(stat = "identity",position="fill")

edited Apr 05 '19 at 16:26

M--

25,431
8
61
93

answered Jun 26 '17 at 16:47

A Gore

1,870
2
15
26

R ggplot2 stacked barplot by percentage with several categorical variables

2 Answers2

What if you want to stick to the code that you had for plotting?

Linked