I want to plot data for a two-factorial experiment in a simple boxplot. The attached diagram shows an example. The grouping structure of the second factor should be indicated by horizontal lines under the axis labels (red lines in the example).
I can solve the problem as shown in the code snipped below. But there is a problem:
When the plot is (vertically) resized, the red lines are not anymore drawn at the desired vertical position. I am looking for a solution for this problem. The text can be placed at a particular "line" in the margin. Isn't there such a possibility to draw lines at a particular "line" instead of giving the user coordinates?
Code snipped used to generate the plot including the red lines:
# generate some values of a two-factorial experiment
values = rnorm(6*10)
lvl1 = c("one","two","three")
lvl2 = c("control","treated")
factor1 = rep(gl(3,10,labels=lvl1),2)
factor2 = gl(2,30,labels=lvl2)
plotgrp = factor(paste(factor1, factor2), levels=c(sapply(lvl1,paste,lvl2)))
# boxplot
at=c(1:3,5:7)
boxplot(values~plotgrp,at=at,xaxt="n",xlab="",las=2,ylab="",main="Example")
axis(1,at=at,labels=rep(lvl1,2),tick=FALSE)
mtext(lvl2,1,line=3,at=c(2,6))
# add the horizontal lines
cxy = par("cxy")
d = 3*cxy[1]
ypos = par("usr")[3] - 2.75*cxy[2]
segments(c(1,5)-d,c(ypos,ypos),c(3,7)+d,c(ypos,ypos),xpd=NA,lwd=2,col="red")
I found drawing into the margins troublesome, too. For a solution that works under arbitrary rescaling from the R Mac GUI, I only found solutions to work that allow the specification of positions in line coordinates but not in user coordinates. For your particular case, I can merely offer an ad hoc hack based on an abuse of additional axes. Try plotting your lines with these two commands:
axis(1,at=c(0.5,1,2,3,3.5),col="red",line=2.5,tick=T,labels=rep("",5),lwd=2,lwd.ticks=0)
axis(1,at=4+c(0.5,1,2,3,3.5),col="red",line=2.5,tick=T,labels=rep("",5),lwd=2,lwd.ticks=0)
Do you really need the lines - simply having the titles Control and Treatment are sufficient, surely?
Sure I do need these lines :) There are cases where such lines considerably improve the clarity of the plot. There should be a proper solution to such problems.
Such things are requested from co-authors and reviewers, and you find a lot of examples in the literature (e.g. see attached link).
http://ajpgi.physiology.org/content/ajpgi/301/5/G887/F8.large.jpg
I found drawing into the margins troublesome, too. For a solution that works under arbitrary rescaling from the R Mac GUI, I only found solutions to work that allow the specification of positions in line coordinates but not in user coordinates. For your particular case, I can merely offer an ad hoc hack based on an abuse of additional axes. Try plotting your lines with these two commands:
axis(1,at=c(0.5,1,2,3,3.5),col="red",line=2.5,tick=T,labels=rep("",5),lwd=2,lwd.ticks=0)
axis(1,at=4+c(0.5,1,2,3,3.5),col="red",line=2.5,tick=T,labels=rep("",5),lwd=2,lwd.ticks=0)
Yep, Andreas, that's it. It works well. I also had the idea of using axes but I did not know how to limit the x-range of the axis line. Thank you!
Also, if you want to get rid of the dotted lines for the whiskers...set the lty=1. Sorry for the unsolicited comment...for some reason, I hate the dotted lines. Not sure why that was ever set as default in the boxplot() function.
Yes, Patrick, thanks for the comment, and I in fact share your opinion :) However, this was just a made-up example to be re-used without much effort. To present the data (with typically n
@Jochen. Yes, I figured this was just an example! I just could not resist. And good to know you have the same strong feelings against such dotted lines!
Re: the beeswarm plot. Thanks for sharing. This is very interesting! Great for seeing the underlying structure of data.
Thanks guys, this info has been immensely helpful. Good question Jochen and great answer Patrick. Also thanks @jochen for letting me know about the beeswarm package.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R