Horizontal Sliding Bar Chart in ggplot2

Question

I've created the plot I need, but I'd like to slide the bars so that the center neutral category evenly straddles zero (x=0) for each subplot. Any ideas? Perhaps I'm not using the right geometric construct here?

library(ggplot2)
survey_data <- data.frame(gender=rep(c("Unreported","Female","Male"),7),
                      feel_job=c(1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6, 7, 7, 7),
                      Freq=c(0, 0, 0, 1, 3, 5, 0, 4, 4, 0, 7, 15, 3, 28, 35, 3, 35, 80, 1, 52, 108))
p <- ggplot(survey_data, aes(gender)) + 
  geom_bar(aes(y = Freq, fill = factor(feel_job)), stat = "identity") +
  coord_flip()
p

Apparently these type of plots are also called "Diverging Stacked Bar Charts" in some circles.

Have a look at [this](http://stackoverflow.com/questions/13734368/ggplot2-and-a-stacked-bar-chart-with-negative-values) answer, which does something very similar, but in vertical rather than horizontal mode. — SlowLearner, Jul 11 '13 at 22:37
I would think the problem with that plot is it screws up the ordering of the factors in the bar. I also don't like stacking as that doesn't let you see the actual size of the different bars. — Andy Clifton, Jul 11 '13 at 23:14
Sorry for the delay. The plots used in the literature are usually stacked, but I see the value in leaving them unstacked. The HH package mentioned below will yield exactly what I wanted (if you can get your data in the format their package will easily accept, which is another task), but I'd rather use ggplot2 at this point. — Sterling, Jul 23 '13 at 16:26

Andy Clifton · Accepted Answer · 2013-07-11T23:55:09.537

I'm guessing that because you have 7 levels in the factor feel_job that you want level = 4 to be over the y axis. From the examples at http://learnr.wordpress.com/2009/09/24/ggplot2-back-to-back-bar-charts/, I figured there might be a way to cheat.

The key seems to be not to rely on ggplot to do everything. Instead, you have to create your statistics and then fiddle with them. I decided to try using geom_rect() rather than the vanilla geom_bar(), which means I need to give each bar values for xmin, xmax, ymin and ymax. The rest of this answer is going to show how to do that.

# save the data we were given
a.survey.data <-survey_data

# going to plot this as rectangles
a.survey.data$xmin[a.survey.data$feel_job < 4] = -a.survey.data$Freq[a.survey.data$feel_job < 4]
a.survey.data$xmin[a.survey.data$feel_job == 4] = -a.survey.data$Freq[a.survey.data$feel_job == 4]/2
a.survey.data$xmin[a.survey.data$feel_job > 4] = 0

a.survey.data$xmax[a.survey.data$feel_job < 4] = 0
a.survey.data$xmax[a.survey.data$feel_job == 4] = a.survey.data$Freq[a.survey.data$feel_job == 4]/2
a.survey.data$xmax[a.survey.data$feel_job > 4] = a.survey.data$Freq[a.survey.data$feel_job > 4]

# assign values to ymin and ymax based on gender and 
y.base <- NA
y.base[a.survey.data$gender == "Female"] = 1
y.base[a.survey.data$gender == "Male"] = 2
y.base[a.survey.data$gender == "Unreported"] = 3

a.survey.data$ymin <- y.base + (a.survey.data$feel_job-4)*0.1 - 0.05
a.survey.data$ymax <- y.base + (a.survey.data$feel_job-4)*0.1 + 0.05

# set the labels
a.survey.data$feel_job.cut <- factor(cut(a.survey.data$feel_job,
                                         breaks = c(0.5,1.5,2.5,3.5,4.5,5.5,6.5,7.5),
                                         labels = c("1",
                                                    "2",
                                                    "3",
                                                    "neutral",
                                                    "5",
                                                    "6",
                                                    "7"),
                                         ordered = TRUE))

p2 <- ggplot(data = a.survey.data,
             aes(xmax = xmax,
                 xmin = xmin,
                 ymax = ymax,
                 ymin = ymin)) +
  geom_rect(aes(fill = feel_job.cut)) +  
  scale_y_continuous(limits = c(0.5,3.5),
                     breaks=c(1,2,3), 
                     labels=c("Female","Male","Unreported"))
print(p2)

and then we en up with... enter image description here

I'd like to add that this is a bad hack even by my standards. — Andy Clifton, Jul 11 '13 at 23:03
Not exactly what I was looking for, but close enough. This is strong work. From my perspective, though, the best tip you gave was to not rely so much on ggplot to do everything. — Sterling, Jul 23 '13 at 16:28

score 0 · Answer 2 · answered Jul 11 '13 at 21:24

You can do it using the likert function from the HH library:

library(HH)
survey_data <- data.frame(gender=rep(c("Unreported","Female","Male"),7),
                  feel_job=c(1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4, 4, 5, 5, 5, 6, 6, 6, 7, 7, 7),
                  Freq=c(0, 0, 0, 1, 3, 5, 0, 4, 4, 0, 7, 15, 3, 28, 35, 3, 35, 80, 1, 52, 108))
likert(Freq ~ gender + feel_job, survey_data)

I'm not sure if this is exactly what you're going for, but it looks to me like this (or something very similar) should work.

It's close. The code you posted doesn't quite work for what I need, but it looks like that package might be helpful. I'd like to stick with ggplot2, if possible, though. — Sterling, Jul 11 '13 at 21:37

Horizontal Sliding Bar Chart in ggplot2

2 Answers2