John HilbertPortfolioNo comments
     

Ever want to combine multiple plots into a consistent file easily without having to open up Inkscape or GIMP?

Perhaps you never thought you would need a garden of pie charts but rather:

  • Revenue of multiple products over the fiscal year.
  • Price (line) and volume (bar) of a stock over time.
  • Income distribution (pie) per state (map).
All these and more can be yours with a few simple key strokes (and a hack or two) in R.  We will also build them with one of my all time favorite packages: Hadley Wickham’s ggplot2.  Sure you could do much of this with other packages, but why? ggplot2 is an excellent piece of machinery for graphics; it just needs a little nudge.

Stacking Similar Plots: Facets

ggplot2, unbeknownst to some, does multiple plots already, they are just called facets.  Facets basically allow you to stack or grid subsets of your data.  For instance the following code
qplot(data = mtcars, 
      x = mpg, 
      y = wt, 
      color = factor(am))   +  
  facet_grid(am ~ .)
yields the MPG vs weight of automatic (0) and manual (1) cars as seen below
What is nice about facet is that it keeps everything consistent.  The MPG axis matches on both plots and if we other aesthetics they would also agree.  This is due to the fact that the data is the same, it is just split between two plots.  So you can see if you needed to plot your revenue from particular products over the course of the year you could simply do it with this.
What facet does not allow you to do is change the plot type or data of the subplots.  This is somewhat obvious since the two plots are not actually subplots, they are facets of the data.  To do stacked subplots we need something more.

Stacked Different Plots: gridExtra

To advance this into true subplots we need the help of an addition package: gridExtra.  This package allows you to make separate plots and arrange them on a canvas.

a <- qplot(data = mtcars,
           x = mpg,
           y = wt,
           color = factor(am)) +
  theme(legend.position="none",
        axis.title.x = element_blank(),
        axis.text.x = element_blank(),
        axis.ticks.x = element_blank()) +
  xlim(c(10,35))

b <- qplot(data = mtcars,
           x = mpg,
           fill = factor(am),
           geom = "histogram",
           binwidth = 5) +
  xlim(c(10,35)) +
  theme(legend.position="none")

grid.arrange(a, b, heights = c(0.5, 0.250))

which yields a plot of MPG & weight (the two previous facets combined) and a histogram of the counts 

Wonderful, however it appears that I did some work that needs some explanation.  Notice that lines 1-5 are simply creating a scatter plot, nothing new.  Next we remove the color legend and the X axis (the title, text, and ticks).  We do this primarily for aesthetics.  Similarly we create a histogram in lines 11-15 and remove the legend on line 17.  Finally we take these two plots and send it to grid.arrange.  Notice we can set the height proportions via an array of values (which will automatically normalize itself, i.e. the upper plot is 66% of the canvas and the second is 33%, or 2:1).Simple, right?  Well not quite.  I purposely neglected to mention the manual setting of the limits of the two plots on lines 9 & 16.  This typically needs to be done to get the two X-axes to line up.  The issue with gridExtra is, well, it is a grid.  The width of the two subplot areas are the same and when plots a and b are drawn ggplot takes up the whole width.  When this happens they become askew.

This is the primary issue with gridExtra.  If the subplots do not need to be aligned, you are good to go.  If not, you may be able to shift them into place via the limits, however in some cases it may be hopeless.  What would be needed is manually placing graphic objects on a canvas, not constrained to a grid, but anywhere.  Don’t worry, I found a back-door solution and will be covered in Part 2.

Leave a comment