ggplot - Add regression line on a boxplot with binned (non-continuous) x-axis

Question

I have a dataset with this structure:

df<- data.frame (VPD.mean=rnorm(100,mean=2,sd=0.8), treatment=c("ambient","elevated"), variable=rnorm(100,mean=50,sd=10))
df$group <- with(df, as.factor (ifelse (VPD.mean>0 & VPD.mean<=1,"0-1",ifelse (
  VPD.mean>1 & VPD.mean<=1.5,"1-1.5",ifelse (
    VPD.mean >1.5 & VPD.mean<2, "1.5-2",ifelse (
      VPD.mean >=2 & VPD.mean<2.5, "2-2.5",ifelse (
        VPD.mean >=2.5 & VPD.mean <3,"2.5-3", ifelse(
          VPD.mean >=3,">3", NA)  
      )))))))
df$group<- factor(df$group,levels=c("0-1","1-1.5","1.5-2" ,"2-2.5","2.5-3",">3"))

I created a boxplot using the groups created after binning VPD.mean, and therefore the x-axis is non-continuous (see graph below): enter image description here

I would also like to add a regression line (smooth), and therefore I would have to use the continuous variable (VPD.mean) instead of the binned one (groups) as x-axis. The result is not nice, because the smooth line doesn't match the x-axis of the graphs. This is the code for the ggplot:

ggplot(df[!is.na(df$group),], aes(group,variable,fill=treatment)) + 
  geom_boxplot(outlier.size = 0) + geom_smooth(aes(x=VPD.mean))

What's the solution to plot the geom_smooth from a different x-axis on the same graph? Thanks

In my opinion it is not wise to plot these two different kinds of plots into one plot (if even possible). Your plot will look messy. It is therefore better to to make two seperate plots. — Jaap

jlhoward jlhoward · Accepted Answer · 2014-12-29T21:19:23

It is possible to do what you ask, but it is a stunningly bad idea.

set.seed(1)  # for reproducible example
df<- data.frame (VPD.mean=rnorm(100,mean=2,sd=0.8), treatment=c("ambient","elevated"), variable=rnorm(100,mean=50,sd=10))
df$group <- cut(df$VPD.mean,
                breaks=c(0,seq(1,3,by=0.5),Inf), 
                labels=c("0-1","1-1.5","1.5-2","2-2.5","2.5-3",">3"))
library(ggplot2)
ggplot(df[!is.na(df$group),]) +
  geom_boxplot(aes(x=factor(group),y=variable,fill=treatment),
               position=position_dodge(.7),width=.8)+
  geom_smooth(aes(x=as.integer(group),y=variable,color=treatment,fill=treatment),method=loess)

This works, more or less, because ggplot uses the factor codes for the x-axis, and the factor levels for the axis labels. as.integer(group) returns the factor codes. If your bins are not all the same size (and they are not, in your case), the plot can be misleading.

ggplot - Add regression line on a boxplot with binned (non-continuous) x-axis

1 Answers