Combining violin plot with box plot

Question

I have a very simple dataset (2 groups, n=15 per group). Using ggplot2, I can easily plot a violin plot or a box plot of the two groups. However, I would like to graph a violin plot, but have the fill shade correspond to the 3 quartiles of my data.

There is an example done in SAS here but I would like to do this in R.

Answer 1

Like this using ggplot_build() to get the outlines?

EDIT UPDATED TO SHOW QUANTILE OF THE ORIGINAL DATA

require(ggplot2) # for ggplot
require(dplyr) # for mutation 
df<-data.frame( # make sample data 
 grp=rep(1:6,each=15),
 val=c(rnorm(15,runif(1)*5,runif(1)),
 rnorm(15,runif(1)*5,runif(1)),
 rnorm(15,runif(1)*5,runif(1)),
 rnorm(15,runif(1)*5,runif(1)),
 rnorm(15,runif(1)*5,runif(1)),
 rnorm(15,runif(1)*5,runif(1))
 )
 )
g<-ggplot(df)+geom_violin(aes(x=grp,y=val,group=grp),color="darkred",fill="darkred",size=2) # build the base violins
coords<-ggplot_build(g)$data # use ggbuild to get the outline co-ords
d<-coords[[1]] # this gets the df in a usable form
groups<-unique(d$group) # get the unique "violin" ids
# function to create geom_ploygon calls
fill_viol<-function(v,gr){
 quants<-mutate(v,x.l=x-violinwidth/2,x.r=x+violinwidth/2,cuts=cut(y,quantile(df[df$grp==gr,"val"]))) # add 1/2 width each way to each x value
 plotquants<-data.frame(x=c(quants$x.l,rev(quants$x.r)), # left x bottom to top, then right x top to bottom
 y=c(quants$y,rev(quants$y)), # double up the y values to match
 id=c(quants$cuts,rev(quants$cuts)))# cut by quantile to create polygon id
 geom_polygon(aes(x,y,fill=as.factor(id)),data=plotquants) # return the geom_ploygon object
}
g + # plot g
 lapply(groups,function(x)fill_viol(d[d$group==x,],x)) + # plus polygon objects for each violin
 scale_fill_brewer(palette="Reds", # plus fill
 name="Quantile\n",
 labels=c("25","50","75","100")) +
 coord_flip()

enter image description here

• Josh: Yes, thank you! This is what I was looking for. I'm not so sure how the code works exactly, but is it possible to break up the polygons by quantile(df$grp) instead of just breaking up the vertical space in 25% blocks?
• Troy: Yes - I'd actually meant to do this, but calculated the quantile of the violin coords which are regularly spaced, rather than the original values - oops. I've corrected the code, and flipped the chart so it looks more like your example.
• Jan Stanstrup: Any idea of how to get this to work with facets? It seems quite strambled if I just add a facet_grid or facet_wrap.