Different fill orders in geom_bar per discrete x-value

Question

I have this ggplot:

What I want to do is change the order of cat2 in each distinct cat1-category in decreasing order of share:

For cat1 == Z: i would come first, because it has the largest share in that cat1, followed by k, then j etc.
For cat1 == Y: j would come first, because it has the largest share in that cat1, followed by h, k, f, g
and so on for all cat1-categories.

I know I can control the order by defining factor levels, but since I need a different order in different cat1-categories, I'm not sure how to do this.

Can someone enlighten me?

df <- structure(list(cat2 = c("i", "k", "j", "h", "g", "j", "h", "k", 
"f", "g", "j", "h", "k", "f", "g", "i", "k", "j", "h", "f"), 
    cat1 = c("Z", "Z", "Z", "Z", "Z", "Y", "Y", "Y", "Y", "Y", 
    "X", "X", "X", "X", "X", "W", "W", "W", "W", "W"), share = c(0.254318086442458, 
    0.217254586570476, 0.13303361614456, 0.107317457057957, 0.0796390207719751, 
    0.255762968963295, 0.198921069216629, 0.177295815678624, 
    0.100770340584133, 0.0971042138291394, 0.222007778896866, 
    0.177174367182501, 0.156891912894117, 0.097677432116308, 
    0.0975337223454565, 0.503295513011454, 0.154491050393999, 
    0.114284166914891, 0.0802892036069214, 0.0549053589320047
    ), pos = c(0.477447245936265, 0.108627293285238, 0.283771394642756, 
    0.658265017686473, 0.751743256601439, 0.305177300160272, 
    0.532519319250233, 0.088647907839312, 0.779469237979754, 
    0.680531960773118, 0.26789580234255, 0.467486875382234, 0.0784459564470585, 
    0.702446497377094, 0.604840920146212, 0.520422973814617, 
    0.0772455251969995, 0.211633133851444, 0.812215332123805, 
    0.879812613393268)), .Names = c("cat2", "cat1", "share", 
"pos"), class = "data.frame", row.names = c(NA, -20L))

library(ggplot2)
ggplot(df, aes(cat1, share)) + 
  geom_bar(stat = "identity", aes(fill = cat2)) + 
  geom_text(aes(label = cat2, y = pos)) +
  coord_flip()

Note: my current thinking is that perhaps it could be done using a new category with the levels set as 1st, 2nd, 3rd, 4th, 5th but if I use that for filling, different cat2-values for example i for Z and j for Y would have the same fill-color because they are 1st in the respective cat1-category which I wouldn't want.

Possible duplicate of [Ordering stacks by size in a ggplot2 stacked bar graph](https://stackoverflow.com/questions/9227389/ordering-stacks-by-size-in-a-ggplot2-stacked-bar-graph) — Mako212, Sep 13 '17 at 15:26
@Mako212, thanks for that link. I may be wrong but wouldn't that define a "static" order that will be applied in all stacked bars the same way? — talat, Sep 13 '17 at 15:32

eipi10 · Accepted Answer · 2017-09-13T16:02:51.790

You can do this by creating a new factor column, let's call it cat3, based on both cat1 and cat2. Set the factor ordering of this column based on cat1 and share. Then use this new cat3 factor column as a group aesthetic to set the stacking order for the bars.

In the code below, arrange sets the ordering. We sort the data frame by share separately within each level of cat1. Now the data frame is in the order we want. Then we create cat3 by pasting cat1 and cat2 together and set the order of the levels to be the current sorted order of the data frame. Finally, within ggplot we use cat3 as a group aesthetic to set the stacking order in geom_bar.

Another addition is the use of position_stack(vjust=0.5) to set the label locations, rather than a y aesthetic.

library(tidyverse)

df %>% 
  arrange(cat1, share) %>% 
  mutate(cat3 = factor(paste(cat1, cat2), levels=paste(cat1, cat2))) %>%
  ggplot(aes(cat1, share, group=cat3)) + 
    geom_bar(stat = "identity", aes(fill = cat2)) + 
    geom_text(aes(label = cat2), position=position_stack(vjust=0.5)) +
    coord_flip()

Perfect, thanks! (also for introducing `position_stack`) – talat Sep 13 '17 at 15:44 — talat, Sep 13 '17 at 15:44

Different fill orders in geom_bar per discrete x-value

1 Answers1

Linked