Automatic dodging or position_dodge not enforced on boxplots with missing values

[apontejosea]

I started this issue as a question in StackOverflow:
http://stackoverflow.com/questions/12806260/how-to-enforce-ggplots-position-dodge-on-categories-with-no-data

I'm trying to obtain boxplots of two different signals (ind) sharing the same categories (cat). When there is a category with data for one signal but not for the other one, the boxplot for the signal with data covers all the horizontal spacing, and does not respect the position_dodge instruction for that particular category. As you can see on the example below, the signal x has no data for category B, so it loses the space reserved by position_dodge.

Thanks in advance.

data<-data.frame(cat=c('A','A','A','A','B','B','A','A','A','A','B','B'), 
                 values=c(3,2,1,4,NA,NA,4,5,6,7,8,9), 
                 ind=c('x','x','x','x','x','x','y','y','y','y','y','y'))

print(ggplot() +
        scale_colour_hue(guide='none') +
      geom_boxplot(
           aes(x=as.factor(cat), y=values, 
               fill=ind), 
           position=position_dodge(width=.60), 
           data=data,
           outlier.size = 1.2,
           na.rm=T))

Here is what I have attempted so far:

1. I first tried to change the color of boxes with missing data. I couldn't figure out how to directly inject the desired color for a single box, as color information is automatically generated from signals (fill = data$ind in this case).
2. Since that did not work, I exaggerated the dummy values out of the normal y-axis range (-1000), and I found something interesting happening that looks like a ggplot2 bug:

Here is the example data with a dummy value for the category with missing data:

data            <- data.frame(
cat=c('A','A','A','A','B','B','A','A','A','A','B','B','B'), 
values=c(3,2,1,4,NA,NA,4,5,6,7,8,9, -1000), 
ind=c('x','x','x','x','x','x','y','y','y','y','y','y','x'))

p  <- ggplot() +
      scale_colour_hue(guide='none') +
      geom_boxplot(aes(x=as.factor(cat), y=values, fill=ind),
      position=position_dodge(width=.60), 
      data=data,
      outlier.size = 1.2,
      na.rm=T)

As shown below, printing the plot as it is without defining y-axis limits, we get the correct behavior on the x-axis. Of course the undesired dummy boxplot will also show up.

print(p)

But if we add the y limits, we still get the original undesired behavior.

print(p + ylim(0, 10))

[hadley]

That's by design of position_dodge. Try using facetting for the behaviour you want.

[apontejosea]

After some workarounds, I came up with the outcome I was looking for... (kind of)

data            <- data.frame(
cat=c('A','A','A','A','B','B','A','A','A','A','B','B','B'), 
values=c(3,2,1,4,NA,NA,4,5,6,7,8,9, 0), 
ind=c('x','x','x','x','x','x','y','y','y','y','y','y','x'))

p  <- ggplot() +
      scale_colour_hue(guide='none') +
      geom_boxplot(aes(x=as.factor(cat), y=values, fill=ind),
      position=position_dodge(width=.60), 
      data=data,
      outlier.size = 1.2,
      na.rm=T) +
      geom_line(aes(x=x, y=y), 
                data=data.frame(x=c(0,3),y=rep(0,2)), 
                size = 1, 
                col='white')
print(p)

Unfortunately, faceting doesn't give the effect I'm looking for. The final graph I was looking for is shown below:

If you notice, the white major tick mark at y = 10 is thicker than the other tick marks. This thicker line is the geom_line with size=1 that hides unwanted boxplots.

I wish we could combine different geom objects more seamlessly. I guess I'm using ggplot2 in a non-standard way and workarounds are the way to go on these kind of issues. Anyhow, thank you for your fast response, Hadley. And thanks for putting together ggplot2, wish I love.

[hadley]

Hmmm, maybe we should have an option to position_dodge (maybe drop = FALSE) that would support this behaviour. It is a fairly common request.

[hadley]

This sounds like a great feature, but unfortunately we don't currently have the development bandwidth to support it. If you'd like to submit a pull request that implements this feature, please follow the instructions in the development vignette.

[larry77]

Any progress on this? I am not so technically competent to implement this myself, but it would be a nice feature to have!