How to stack error bars in a stacked bar plot using geom_errorbar?

Question

I want to stack the error bars in a stacked histogram using geom_errorbar / ggplot.

In my ggplot statement, I have tried to used both position="stack" and position="identity". None of them worked.

Here is my ggplot statement:

ggplot(DF, aes(x=factor(year), y=proportion, fill=response)) +
        facet_grid(. ~ sex) +
        theme(legend.position="none")
        geom_bar(position="stack", stat="identity") +
        geom_errorbar(aes(ymin=ci_l, ymax=ci_u),
                      width=.2,                    # Width of the error bars
                      position="identity") +

Here is the result I'm getting, and you may notice that the error bars on the right hand-side do not follow the bar values.

enter image description here

Here is the Data Frame I've used in this example:

DF <- data.frame(sex=c("men","women","men","women","men","women"),
                  proportion=c(0.33,0.32,0.24,0.29,0.12,0.16),
                  ci_l=c(0.325,0.322,0.230,0.284,0.114,0.155),
                  ci_u=c(0.339,0.316,0.252,0.311,0.130,0.176),
                  year=c(2008,2008,2013,2013,2013,2013),
                  response=c("Yes","Yes","Yes, entire the journey","Yes, entire the journey","Yes, part of the journey","Yes, part of the journey")
                  )

Your identity and stack labels are swapped. The one on the left is what you get with with `position='stack'` — Matthew Plourde, Jun 16 '15 at 16:45
Thanks for the heads up, Matthew Plourde. I'm still hoping for a solution, though. :) — rafa.pereira, Jun 16 '15 at 17:02

score 7 · Accepted Answer · answered Jun 16 '15 at 17:01

What is happening here is that ggplot is not stacking the error bars (they would have to be summed) so you will have to do that by hand (and it seems that Hadley thinks that this is not a good idea and wil not add this functionality).

So doing by hand:

DF$ci_l[DF$response == "Yes, part of the journey"] <- with(DF,ci_l[response == "Yes, part of the journey"] +
         ci_l[response == "Yes, entire the journey"])

DF$ci_u[DF$response == "Yes, part of the journey"] <- with(DF,ci_u[response == "Yes, part of the journey"] +
                                                             ci_u[response == "Yes, entire the journey"])

Now:

ggplot(DF, aes(x=factor(year), y=proportion)) +
  facet_grid(. ~ sex) +
  geom_bar(stat="identity",aes(fill=response)) +
  geom_errorbar(aes(ymin= ci_l, 
                    ymax= ci_u),
                width=.2,                    # Width of the error bars
                position="identity")

enter image description here

This should be native, it's hard to do by hand for complex things — wolfsatthedoor, Jul 11 '20 at 21:29
I'm not quite sure this code is correct (any more): it does not reproduce your plot. — bers, Oct 22 '20 at 14:46

score 0 · Answer 2 · answered Jun 16 '15 at 17:22

The issue here is that geom_errorbar is just making nice error bars with the y values that you give it; it doesn't know anything about the geom_bar layer which has a vertical offset for some of the data. So you need to adjust for the fact that for one of your responses, the values plotted have a positive vertical offset determined by the value for another response. For the example provided, this can be accomplished by:

DF$vadj <- c(rep(0,2), rep(c(0,1,0), each=2) * DF$proportion)[1:6]

ggplot(DF, aes(x=factor(year), y=proportion, fill=response)) +
    facet_grid(. ~ sex) + geom_bar(stat='identity') + 
    geom_errorbar( aes(ymin=ci_l+vadj, ymax=ci_u+vadj), width=.2)

The technique for adjustment here is admittedly not especially elegant, and if you need to generalize, be aware that it is very much dependent on the particular structure of the dataframe (i.e. it would have to be changed if the rows were ordered differently). But it should get your error bars where you want them.

How to stack error bars in a stacked bar plot using geom_errorbar?

2 Answers2

Linked