Question

I’ve run two experiments, some replicating the conditions of a previous set. I have a column for the unique lot ID as well as another column containing the ID from experiment 1 which the lots from experiment 2 replicates. Here’s some example data stored like this:

test <- data.frame(var1=c(rep("A",4), rep("B",4), rep("C",4), rep("D",4)),
                   var2=rep(c(rep("A",4), rep("B",4)),2),
                   value=runif(16,1,5))

Here’s my ggplot code:

ggplot(test, aes(x=var1, y=value, fill=var2)) + geom_boxplot()

This gives me the lot IDs arranged according to var1‘s factor order.

I’d like the A’s from var2 side by side and the B’s from var2 side by side. Is the only way to do this by using facet_grid or facet_wrap?

 ggplot(test, aes(x=var1, y=value, fill=var2)) + geom_boxplot() +
        facet_grid(. ~ var2, scales="free_x")

I tried adding group=var2 but that gives me overlapping and very wide boxplots, which I also don’t understand:

ggplot(test, aes(x=var1, y=value, group=var2, fill=var2)) + geom_boxplot()
Warning message:
position_dodge requires non-overlapping x intervals

I can use facetting; I’m mostly asking the question as I was surprised when I couldn’t group how I expected. I also looked in the examples for geom_bar() for more grouping approaches but it seems most people aren’t grouping things this way.

Feel free to provide other input on how one might approach this. I’m simply looking to compare pairs of test results to make it easy to see whether the replicate matches the original trial.

Answer 1

A cheats method using interaction and adjusting the scale_x_discrete labels

 ggplot(test, aes(x=interaction(var1,var2), y=value, fill=var2)) + 
  geom_boxplot() +
  scale_x_discrete(name = 'var1',breaks = c('A.A','C.A','B.B','D.B'), 
                   labels = c('A','C','B','D')) 

EDIT Thanks to @Andrie’s comments

Or you can create an appropriately ordered factor, or (unordered) factor with levels specified in the correct order.

• ggplot2 respects the order of the factor levels (whether in an ordered factor or not).
• R will order the levels a factor lexicographically by default, so for any other order, you will need to specify it.

.

library(plyr) # for arrange 
var1_order <- unique(as.character(arrange(test,var2)[['var1']]))

test$var1_order <- ordered(test$var1, levels = var1_order)
## or
test$var1_order_2 <- factor(test$var1, levels = var1_order)
## so that 
ggplot(test, aes(x=var1_order_2, y=value, fill=var2)) + 
 geom_boxplot() + 
 xlab('var1')
## or 
ggplot(test, aes(x=var1_order, y=value, fill=var2)) + 
 geom_boxplot() + 
 xlab('var1')

Will both give the same result

EDIT — Another approach

You could have your x axis as var2 and fill by var1, which will order by var2 then dodge and fill by var1

  ggplot(test, aes(x=var2,y=value)) + 
   geom_boxplot(aes(fill = var1))