\documentclass[11pt]{article}

\setlength{\oddsidemargin}{0in}
\setlength{\evensidemargin}{0in}
\setlength{\topmargin}{0.0in}
% \setlength{\headheight}{0in}
% \setlength{\headsep}{0in}
\setlength{\textwidth}{6.5in}
\setlength{\textheight}{9.0in}
\setlength{\parindent}{0in}
\setlength{\parskip}{0.15in}

\usepackage{times}
\usepackage{url}

\usepackage{listings}

\usepackage{graphicx}

\begin{document}

\title{Introduction to ggplot2}

\author{N. Matloff}

\date{January 11, 2013}

\maketitle

\section{Introduction}

Hadley Wickham's {\bf ggplot2} package is a very popular alternative to
R's base graphics package.  (Others include {\bf lattice}, {\bf ggobi}
and so on.)  

The {\bf ggplot2} pacakge is an implementation of the ideas in the book,
{\it The Grammar of Graphics}, by Leland Wilkison, whose goal was to set
out a set of general unifying principles for the visualization of data.
For this reason, {\bf ggplot2} offers a more elegant and arguably more
natural approach than does the base R graphics package.

The package has a relatively small number of primitive functions, making
it relatively easy to master.  But through combining these functions in
various ways, a very large number of types of graphs may be produced.
It is considered especially good in setting reasonable default values of
parameters, and much is done without the user's asking.  Legends are
automatically added to graphs, for instance.

The package is quite extensive (only a few functions, but lots of
options), and thus this document is merely a brief introduction.

\section{Installation and Use}

Download and install {\bf ggplot2} with the usual {\bf
install.packages()} function, and then at each usage, load via {\bf
library()}.  Here's what I did on my netbook:

\begin{lstlisting}
# did once:
> install.packages("ggplot2","/home/nm/R")
# do each time I use the package (or set in .Rprofile)
> .libPaths("/home/nm/R")
> library(ggplot2)
\end{lstlisting}

\section{Basic Structures}

One operates in the following pattern:

\begin{itemize}

\item One begins with a call to {\bf ggplot()}: 

\begin{lstlisting}
> p <- ggplot(yourdataframe)
\end{lstlisting}

or 

\begin{lstlisting}
> p <- ggplot(yourdataframe,aes(yourargs))
\end{lstlisting}

Here {\bf yourdataframe} could have been read from a file, say using
{\bf read.table()}, or generated within the program.

The result {\bf p} is an R S3 object of class {\bf "ggplot"}, consisting
of a component named {\bf data}, and other components containing
information about the plot.  

Note that at this point, though, there is no plot; {\bf p} won't be
displayed until we print it.\footnote{In R, {\bf print()} is a {\bf
generic} function, i.e. a stub for a class-specific one.  In this case
the latter does a plot.}

\item One adds features to---or even changes---the plot via the {\bf +}
operator, which of course is an overloaded version of R's built-in {\bf
+}, the function {\bf "+.ggplot"()}.  One can do this repeatedly, e.g.
to superimpose several curves on the same graph.

Each invocation of {\bf +} adds a new {\bf layer} to the graph.

\item The function {\bf aes()} (``aesthetics'') is used to map
your data variables to graph attributes.  

Its first one or two arguments specify the data variables to be
plotted, in one dimension (say for a histogram) or two dimensions (say
for a scatterplot).

The remaining arguments are name/value pairs, where the name is an
attribute such as color and the value is a data variable.  This could
be, for example, to indicate a variable that will determine the color of
any given point in the graph.

One can set attributes in this way at various levels:

   \begin{itemize}

   \item We can set attributes for the entire graph by calling {\bf
   aes()} within our call to {\bf ggplot()}, e.g. to specify which
   variables from our dataset we wish to plot.

   \item We can set attributes specific to one of the + actions,
   such as specifying point color in a scatterplot.

   \end{itemize}

So for instance we could use {\bf aes()} to specify our data variables
either when we call {\bf ggplot()}, so these variables will be used in all
operations, or when we call, say, {\bf geom\_point()}, to indicate data
variables for this specific operation.

\end{itemize}

There are various types of objects that can be used as the second
operand for the + operator.  Examples are:

\begin{itemize}

\item {\bf geoms} (``geometrics''):  Geometric objects to be drawn, such
as points, lines, bars, polygons and text.

\item {\bf position adjustments}:  For instance, in a bar graph, this
controls whether bars should be side by side, or stacked on top of each
other.

\item {\bf facets}:  Specifications to draw many graphs together, as
panels in a large graph.  You can have rows of panels, columns of
panels, and rows and columns of panels.

\item {\bf themes}:  Don't like the gray background in a graph?  Want
nicer labeling, etc.?  You can set each of these individually, but one
of the built-in themes, or a user-contributed one, can save you the
trouble, or you can write one that you anticipate using a lot.

\end{itemize}

\section{Example:  Simple Line Graphs}

\begin{lstlisting}
> df1
  u v
1 0 3
2 2 4
3 5 5
> ggplot(df1) + geom_line(aes(x=u,y=v))
\end{lstlisting}

Here {\bf aes()} was called from {\bf geom\_line()} rather than from
{\bf ggplot()}, so as to apply just to this line.  The result is

\includegraphics[width=4.0in]{lines1.pdf}

Now let's add a second line, from a {\it different} data frame:

\begin{lstlisting}
> df2
  w z
1 1 4
2 2 5
3 3 9
ggplot(df1) + geom_line(aes(x=u,y=v)) + geom_line(data=df2,aes(x=w,y=z))
\end{lstlisting}

Here is the result:

\includegraphics[width=4.0in]{lines2.pdf}

It worked as long as we specified {\bf data} in the second line.

Note that {\bf ggplot2} automatically adjusted that second graph, to
make room for the ``taller'' second line.

\section{Example:  Census Data}

The data set here consists of programmers (software engineers, etc.) and
electrical engineers in Silicon Valley, in the 2000 Census.  I've
removed those with less than a Bachelor's degree.  The R object was a
data frame named {\bf pm}.

I first ran

\begin{lstlisting}
p <- ggplot(pm)
\end{lstlisting}

to set up the {\bf ggplot} object.  Next, I typed

\begin{lstlisting}
p + geom_histogram(aes(HiDeg))
\end{lstlisting}

which produced a histogram of the highest-degree values of the workers:

\includegraphics[bb=0 0 504 504,width=4.0in]{edhist.pdf}

Note that the + operation yields a new object of class {\bf "ggplot"}.
Since the generic print function for that class actually plots the
graph, the graph did appear on the screen.  I could have saved the new
object in a variable if needed.

I then decided to do a scatter plot of salary versus age:

\begin{lstlisting}
> p + geom_point(aes(x=Age,y=WgInc))
\end{lstlisting}

Note the roles of {\bf aes()} here and above; I used it to tell the geom
what variable(s) to use, for instance telling {\bf geom\_point()} in the
second exampl,e which of my data variables would correspond to the X-
and Y-axes.

This gave me this graph:

\includegraphics[bb=0 0 504 504,width=3.5in]{ageinc.pdf}

(As is often the case with large data sets, the points tend to ``fill
in'' entire regions, one solution to which is to graph a random subset
of the data, not done here.  Similar comments apply to some of the
graphs below.)

However, I wanted to separate the points according to highest degree
level:

\begin{lstlisting}
> p + geom_point(aes(x=Age,y=WgInc,color=HiDeg))
\end{lstlisting}

Here I have three data variables informing {\bf aes()}:  Age, wage
income and highest degree.  The argument {\bf color} here means that I
want the degree to be used for color coding the points:\footnote{Note
that if I had wanted the same color to be set for {\it all} points, I'd
set the {\bf color} option \underline{outside} the {\bf aes()} call, as
my second argument to {\bf geom\_point()}.}

\includegraphics[bb=0 0 504 504,width=3.5in]{ageinccolor.pdf}

Note the legend that was automatically included on the right.

Since some people might be viewing a black-and-white version of this
document, I ran the command again, specifying point shape instead of
point color:

\begin{lstlisting}
p + geom_point(aes(x=Age,y=WgInc,shape=HiDeg))
\end{lstlisting}

Here {\bf ggplot2} decided to use a circle, a triangle and a square to
represent Bachelor's, Master's and PhD workers:

\includegraphics[bb=0 0 504 504,width=4.5in]{ageincshape.pdf}

Since I'm interested in age discrimination in the industry, I decided to
restrict my graph to those over age 40.  The {\bf ggplot2} package
cleverly exploits the R {\bf subset()} function, allowing me to write

\begin{lstlisting}
p %+% subset(pm,Age > 40) + geom_point(aes(x=Age,y=WgInc,color=HiDeg))
\end{lstlisting}

The new operator {\bf \%+\%} is again mapped to {\bf "+.ggplot"()}.  The
result was

\includegraphics[bb=0 0 504 504,width=4.5in]{40ageinccolor.pdf}

Look at all those 0-income PhDs!  (There was also a business income
variable, which I did not pursue here, so maybe some had high incomes
after all.)

Even with color, it's a little hard to compare the three degree levels,
so let's try faceting:

\begin{lstlisting}
> pt <- p + geom_point(aes(x=Age, y=WgInc))
> pt + facet_grid(HiDeg ~ .)
\end{lstlisting}

Here I saved an overall graph of wage income versus age to {\bf pt}, and
then added faceting by degree.  This instructed {\bf ggplot2} to draw
three graphs, one for each degree level, as three panels in the same
large graph.  The \lstinline{HiDeg ~ .} argument (which you may
recognize as similar notation to R's {\bf lm()} function for linear
models) states that I want HiDeg to define the rows of panels, without
any variable defining columns.  The result was:

\includegraphics[bb=0 0 504 504,width=4.5in]{facetvertageinc.pdf}

I tried a horizontal presentation too:

\includegraphics[bb=0 0 504 504,width=4.5in]{facethorizageinc.pdf}

Note that I could have combined operations, e.g.

\begin{lstlisting}
> p + geom_point(aes(x=Age, y=WgInc)) + facet_grid(HiDeg ~ .)
\end{lstlisting}

to get the vertical set of panels, without directly going through the
intermediate step of drawing the nonfaceted graph.

If there had been a Gender variable in the data, I could have defined
rows and columns of panels:

\begin{lstlisting}
> p + geom_point(aes(x=Age, y=WgInc)) + facet_grid(Gender ~ HiDeg)
\end{lstlisting}

\section{For Further Information}

Just plugging ``ggplot2 tutorial,'' ``ggplot2 introduction,'' ``ggplot2
examples'' and so on into your favorite search engine will give you tons
of information.

Hadley's book, {\it ggplot2: Elegant Graphics for Data Analysis}, is of
course the definitive source, but also try his pictorial reference
manual, at \url{http://had.co.nz/ggplot2/}.

\end{document}

pm <- pums[pums$WgInc < 300000,]
pm$Mas <- pm$Educ == 14
pm$Bach <- pm$Educ == 13
pm$PhD <- pm$Educ == 16
pm$EdCode <- ifelse(pm$Bach,1,ifelse(pm$Mas,2,3))
pm$HiDeg <- ifelse(pm$EdCode == 1,"BS",ifelse(pm$EdCode == 2,"MS","PhD"))