Consistent legend colors with qplot

Question

Consider the following data frame:

x = read.table(text = 'Lo Re Pe
1 T 33
1 F 22
1 H 11
2 T 22
2 F 22', header = TRUE)

and the following plot:

qplot(factor(Lo), data=x, geom='bar', fill=Re, weight=Pe, 
      xlab='L', main='Title', ylab='Pe')

Now consider this data frame:

x <- read.table(text = 'Lo Re Pe
1 D 33
1 K 22
2 D 22
2 K 22', header=TRUE)

with the same qplot statement.

The colors assigned to each Re value are not consistent between the plots, so it is difficult to compare the plots directly.

How do I specify that Re value T should always be "Red", for example, and that Re value F should always be "Blue", for example, so that the qplot command always uses consistent colors for each Re value, regardless of the contents of the data frame? There are a finite and known number of values for Re, so I could specify them all.

I tried the following when the data frame contained values T, F and H:

qplot(factor(Lo), data=x, geom='bar', fill=Re, weight=Pe, 
      xlab='Loci', main='Title', ylab='Pe', 
      scale_fill_manual(values=c("Blue","Red","Green"),labels=c("T","F","H")))

but R reports an error about incorrect length and does not produce a plot.

The solution would ideally allow me to specify colors for all possible values of Re, even though all of these may not be present in the data frame.

Answer 1

This is perfectly possible using the modular nature of ggplot. I'm going to recommend that you drop qplot, though, and switch to using ggplot(). It will cost you nothing and will be more convenient in the long run, as it is more suited to doing "complicated" things.

Let's start with your two data sets:

x1 = read.table(text = 'Lo Re Pe
1 T 33
1 F 22
1 H 11
2 T 22
2 F 22', header = TRUE)

x2 <- read.table(text = 'Lo Re Pe
1 D 33
1 K 22
2 D 22
2 K 22', header=TRUE)

Now here's your first plot, but translated into ggplot():

p <- ggplot(x1,aes(x = factor(Lo))) + 
        geom_bar(aes(fill = Re,weight = Pe)) + 
        labs(x = 'L',y = 'Pe') + 
        opts(title = 'Title')

To keep the color consistent across plots, and to prevent unused colors from appearing in the legend, we will simply create a master color key, and pass only the needed subset of it to our scale:

color_key <- c('red','blue','green','black','orange')
#If Re is a character variable:
names(color_key) <- unique(c(x1$Re,x2$Re))
#If Re is a factor:
names(color_key) <- unique(c(as.character(x1$Re),as.character(x2$Re)))

(You could also do something similar using the levels function, but I wanted to guard against including levels that do not appear in the data set.)

Obviously, you can choose whatever colors you like. Now I can customize the fill scale for our plot p by passing only that segment of color_key that is relevant to scale_fill_manual:

p + scale_fill_manual(values = color_key[names(color_key) %in% x1$Re])

Additionally, if your plots all really do have the same structure, we don't even need to replicate the ggplot call over and over. We can simply apply our plot p to a new data set:

p1 <- p %+% x2

And then add the fill scale in the same manner:

p1 + scale_fill_manual(values = color_key[names(color_key) %in% x2$Re])

Finally, let's mix and match ourselves a new data set:

x3 <- rbind(x1[1:2,],x2[3:4,])

Same process works again:

p3 <- p %+% x3
p3 + scale_fill_manual(values = color_key[names(color_key) %in% x3$Re])