PLOTTING WITH GGPLOT2

Download R source file


              ## PLEASE GO THROUGH THIS PRIOR TO CLASS

# Clear the workspace
rm(list = ls())

# Load the libraries used for this lab
library(ggplot2)
library(ggthemes)
library(reshape2)
library(dplyr)
library(magrittr)
library(tidyr)

# This lab cover how to plot data.  Although it may not feel as precise, visual
# inspection of the data can reveal a lot about the structure of the data.  The
# classic example for that is Anscombe's quartet.  R includes Anscombe's quartet
# as a built-in dataset.  We will begin by plotting those data.
# Load the data.  Remember this is a built-in dataset, so we can use the data()
# command.
data(anscombe)
anscombe  # notice that it's divided into 4 sets of x & y variables.

# Plotting in base R is straightforward -- we just use the plot() command
with(anscombe, plot(x = x1, y = y1))  # it works, but it's ugly

# There are a variety of different ways to modify a plot in base R.  Instead of
# learning them, we will focus on learning a package from the Hadley-verse,
# ggplot2.  ggplot2 builds on the idea of a "grammar of graphics" (hence gg)
# allowing us to build our plot from its constituent components.  It also has
# more attractive defaults.

# ggplot has a command called qplot that does something very similar to the
# plot command in base R:
qplot(data = anscombe, x = x1, y = y1)  # prettier, right?

# The cool thing about ggplot is that it allows us to save our plot, and then
# to update pieces of it.
(p <- qplot(data = anscombe, x = x1, y = y1))  # saves the plot to the object p
# again, the () wrapper around a <- statement prints whatever was assigned
# on the left hand side of the "<-" to the screen.

# Now we can change what values are plotted:
p + aes(x = x2, y = y2)  # don't worry about what aes() means, we'll come to that
p + aes(x = x3, y = y3)
p + aes(x = x4, y = y4)

# That's all well and good, but what if we want to plot all of those on the same
# page?  To do that, we'll have to use the "grammar" for graphics.

# The basic idea behind the grammar is that you build up a plot piece by piece.
# First, you add the data, and you decide which data values get mapped to which
# dimensions on the plot (x axis, y axis, size, color, shape, etc.).  Hadley
# calls these mappings "aesthetics" and appreviates them "aes".
p1 <- ggplot(data = anscombe, aes(x = x1, y = y1))

# Next you decide what "geometry" is going to be used for your plot.  Are you
# going to just do a scatterplot, where there are just points?
p1 + geom_point()

# Or are you going to connect those points in a line graph?
p1 + geom_line()

# It's the same data, we've just changed how we're drawing the points.  Let's
# stick with points.
p1 <- p1 + geom_point()

# Next, you decide how to scale the axes.  Should everything start from 0?
p1 + xlim(c(0, NA)) + ylim(c(0, NA))  # NA in this case means calculate the
# limit automatically.  (I'm not going to save this.)

# Or, if I wanted to change the y-axis to be on a log scale, I can just add it:
p1 + scale_y_log10()  # (I'm not going to save this either.)

# Next you label the axes, and give the graph a title:
(p1 <- p1 + ggtitle("Anscombe's quartet") + xlab("X") + ylab("Y"))

# Finally, change the color scheme, if you want.
# Some themes that come with ggplot2:
p1 + theme_bw()
p1 + theme_classic()

# A few from ggthemes
p1 + theme_economist()  # looks like The Economist magazine
p1 + theme_fivethirtyeight()  #looks like plots from 538
p1 + theme_stata()  # if you really, really miss Stata plots
p1 + theme_excel()  # please don't do this.  It's so ugly.

# You can customize plots to look however you want.  And, if you save a
# customized plot, you can reuse the theme with different data.
p2 <- p1 + theme_economist()
p2 + aes(x = x2, y = y2)  #  See?  Same theme, different variables.

# One of the other powerful things that ggplot2 can do for you is create small
# multiples -- several plots that show the same thing.  Hadley calls this 
# faceting.
# To make a plot with facets, first we have to rearrange the data into columns.
# To do that, we use a new package from the Hadley-verse, tidyr.  I won't cover
# it in depth; for an intro, see:
vignette("tidy-data")

# The gather function strings the data out into two long columns.  One column
# is an indicator of what the variable name was (key), and the other is
# an indicator of what the value of that variable was (value).
anscombe %>% gather()

# This isn't quite what we want.  The separate command takes us a step closer by
# creating a column for x or y, and a column for series (1, 2, 3, or 4):
anscombe %>%
  gather() %>%
  separate(key, c("x.or.y", "series"), 1)  # the 1 means separate after the first character

  # Equivalently, we could have used extract_numeric()
  # mutate(series = extract_numeric(key))

# ASIDE: we could have created a new column using substr.
# substr stands for sub-string -- it takes a string (variable), and takes 
# characters starting at 1 and ending at 1.  For example, say I have a string
# "catinthehat".  If I just want the word "cat", I would start at the first
# character (c), and stop at the third character (t).  This is how that would
# look
substr("catinthehat", 1, 3)

# Still not quite what we're looking for.  We want a column for x values, 
# a column for y values, and a column for series.  The spread command takes care
# of this:
anscombe %>%
  gather() %>%
  separate(key, c("x.or.y", "series"), 1) %>% 
  spread(x.or.y, value)

# Whoops!  That gives us an error.  What went wrong?
# The error says "duplicate identifiers for rows".  That must mean that R
# decided it was going to try to combine all of the rows where series == 1, and
# got confused.  Let's help it out by giving it an unduplicated identifier. In
# this case, we'll have to use row numbers.
anscombe %>%
  mutate(sid = row_number()) %>%  # sid stands for subject ID
  gather(key = key, value = value, everything(), -sid) %>%
  # key = key and value = value means create two new columns named key and value
  # like we did before.
  # everything() means gather all the columns, like we did before.
  # the -sid means do NOT gather sid, but leave it as its own column
  separate(key, c("x.or.y", "series"), 1)  # same as before

# Now spread works fine.  Let's save the result.
(anscombe.ggplot <- anscombe %>%
  mutate(sid = row_number()) %>%
  gather(key = key, value = value, everything(), -sid) %>%
  separate(key, c("x.or.y", "series"), 1) %>%
  spread(x.or.y, value))

# Let's get rid of the sid column, and let's sort by series, just to make this
# cleaner to look at:
(anscombe.ggplot %<>% 
  select(-sid) %>%
  arrange(series))
  
# Now the data have one column for x, one column for y, and one column 
# indicating which series the data are from.  We can map each one of these
# columns to a dimension in the plot.  For example, let's make x and y the
# x and y dimensions, as before, but let's let the color of the points represent
# the series.
(p3 <- ggplot(data = anscombe.ggplot, aes(x = x, y = y, color = series)) +
  geom_point())

# Well that's cool, but it's pretty hard to see.  We could have changed the 
# color of the points
p3 + scale_color_brewer()  # uses colors from colorbrewer2.org -- check it out!

# Still hard to see.  Really, we want these in separate plots.  Again, those
# are called facets in ggplot
(p4 <- p3 + 
  facet_wrap(~ series) +  # could do up to two columns with facet_grid(~ col1 + col2)
  aes(color = NULL))  # removes the colored points from before

# That's more like it!  Now we could clean up the names, etc., like we did
# before.  If we just wanted to plot one, we would have to change the data:
p4 %+% filter(anscombe.ggplot, series == 1)

# The last thing we'll do in this script is show you how to add calculated
# layers.  Here I want to add a layer that shows the regression line for each
# of these.  This is fairly easy -- you just add another "geometry":
p4 + geom_smooth(method = "lm")  # lm means "linear model" -- forces the line to be straight
# (Because this is Anscombe's quartet, all of the lines are the same.)

# There's a whole book about ggplot.  It's available online from: 
# http://search.library.duke.edu/search?id=DUKE004969277
# The easiest way to learn is to start thinking of what you want to present, and
# then search for it on the internet or in the help (or come talk to me!).
# ggplot has a gallery of the different "geoms" on it's main page:
# http://docs.ggplot2.org/current/