20

Given a two column data frame with a categorical column of labels and a quantitative column of percent data, I can reliably produce a bar chart in ggplot that sorts the bars by value rather than by alphabetical order using the following:

ggplot(data=df, aes(x=reorder(Label, Percent), y=Percent, fill=Label)) + geom_bar()

This tells reorder to sorts the bar chart by the percent value rather than the text, making it easier to see the changes.

I cannot, however, get the legend to match: instead it persists in being sorted by the original alphabetical values. This results in a legend which doesn't match the chart, which is confusing.

I've looked on StackOverflow and elsewhere, and haven't found a fix that will work. Any advice?

Edit: per requests, this is essentially the data:

Labels <- c("G", "G", "A", "C", "M", "B", "M", "G", "A","M")
Percent <- c("-0.241","-0.046", "-0.037", "-0.024", "-0.003","0.007","0.01","0.059","0.121", "0.152")
sogrady
  • 756
  • 3
  • 8
  • 12
  • 1
    Can you provide a representative example of your `df` so we can reproduce your problem? – cdeterman Nov 11 '14 at 19:39
  • 2
    Have you tried something like `+ scale_fill_discrete(breaks=c("Top","Middle","Bottom"))`? – Jota Nov 11 '14 at 19:43
  • Data provided. As for scale_fill_discrete, that did not solve the problem. Any other questions, just let me know. – sogrady Nov 11 '14 at 20:05

2 Answers2

15

With the data you provided, the following works for me basically applying the suggestion from Frank:

Labels <- c("G", "G", "A", "C", "M", "B", "M", "G", "A","M")
Percent <- c("-0.241","-0.046", "-0.037", "-0.024", "-0.003","0.007","0.01","0.059","0.121", "0.152")
df <- data.frame(Labels, Percent)
df$Percent <- as.numeric(as.character(df$Percent))
legend_ord <- levels(with(df, reorder(Labels, Percent)))

library(ggplot2)
p <- ggplot(data=df, aes(x=reorder(Labels, Percent), y = Percent, fill=Labels)) 
p <- p + geom_bar(stat="identity")
p + scale_fill_discrete(breaks=legend_ord)

enter image description here

Regarding your followup question, if you have specific colors you can assign them within the scale_fill_manual call.

p <- ggplot(data=df, aes(x=reorder(Labels, Percent), y = Percent, fill=Labels)) 
p <- p + geom_bar(stat="identity")
p + scale_fill_manual(breaks=legend_ord, 
    values=c("G"="#0000FF", "C"="#003FBF","B"="#007F7F","A"="#00BF3F","M"="#00FF00"))

enter image description here

cdeterman
  • 19,630
  • 7
  • 76
  • 100
14

I find the easiest way is to just reorder your data before plotting. By specifying the reorder(() inside aes(), you essentially making a ordered copy of it for the plotting parts, but it's tricky internally for ggplot to pass that around, e.g. to the legend making functions.

This should work just fine:

df$Label  <- with(df, reorder(Label, Percent))
ggplot(data=df, aes(x=Label, y=Percent, fill=Label)) + geom_bar()

I do assume your Percent column is numeric, not factor or character. This isn't clear from your question. In the future, if you post dput(df) the classes will be unambiguous, as well as allowing people to copy/paste your data into R.

Gregor Thomas
  • 136,190
  • 20
  • 167
  • 294