R Graphics

Abstract

R Graphics. Paul Murrell. Boca Raton, FL: Chapman {{}{{}{}}{{}{&}{}}{{}{}}{}} Hall/CRC, 2006, xix+301 pp., {{}{{}{}}{{}{$}{}}{{}{}}{}}69.95(H), ISBN: 1-58488-486-X. 1. OVERVIEW R Graphics is an excellent technical introduction to drawing graphics with R. It pulls together information currently scattered throughout various R documents and help pages. The organization and writing is clear and coherent, which is especially welcome when dealing with the intricacies of R graphics. It serves as a very useful reference book. 2. WHAT IS IN THIS BOOK? The book starts with an introduction to R graphics including a gallery of many different graphics made using R, demonstrating the power of this system for generating static, high-quality plots. These examples range from basic statistical plots, to cleverly annotated plots, to sophisticated 3-D diagrams, and even whimsical art pieces! It is a very inspiring beginning, and one feels well motivated to do battle with R. The main material in the book is divided into two parts. The first describes the traditional, or base, graphics, while the second part introduces the newer grid graphics system, including lattice graphics. Grid was designed and written almost entirely by the author of the book, so Dr. Murrell speaks with clear authority here. The base graphics system was the first graphics system developed for R. If you have used R in the past, it is likely to be an old friend (or enemy). Base graphics has a simple metaphor: ink on paper. Just like drawing with your pen, you cannot undo your mistakes, except to start afresh. More formally, there is no representation of the graphics independent of their presence on the plot so you can only add, not edit or delete, existing output. This makes base graphics simple and easy to understand but fundamentally limited. This limitation is best seen when trying to customize graphics, where you either need to start from scratch or grapple with many arcane settings. R Graphics provides an excellent summary of these details. In terms of functionality, but not yet popularity, base graphics has been superseded by the grid graphics system. The explanation of grid is the strength of this book. The section begins with a description of lattice graphics, a system that illustrates the power of grid. Lattice is an implementation of trellis graphics (Becker, Cleveland, and Shyu 1996), which provides an easy way to produce multiple plots based on different subsets of a dataset. The plots typically share the same scales and allow one to investigate relationships between two variables conditional on one (or more) other variables. Lattice graphics present a higher level of abstraction than base graphics, but configuring lattice can be difficult due to the multitudinous (378 at last count) and repetitive options. This book provides a handy reference to some of these options as well as a brief discussion of annotating lattice plots. The description of creating new plot types is briefer still, largely a reflection of the limitations of lattice. The framework provided by grid graphics has a number of advantages over the ink on paper design of base graphics: • Grid objects have an independent representation as R objects not just pixels on a screen (described in sec. 7.3). • A system of viewports allows for extremely flexible layout (described in sec. 5.5). • A range of coordinate systems makes it easy to draw what you want where you want it (described in sec. 5.3). Although it is easy to drawdirectly to the screen with grid, its real power lies in creating objects that can be drawn at a later time. This allows much greater flexibility as objects can be extensively modified and even deleted. Unfortunately, of the few R packages that use grid, even fewer return grid objects, which makes building on top of them almost as hard as building on top of base graphics. We hope that this will change as more people read this book. With base (and even lattice) graphics, it is hard to write reusable graphical functions, and solving this problem is the promise of the grid system. An R package, ggplot, released after R Graphics, further demonstrates this potential. It builds on grid graphics using the principles described byWilkinson (2005), providing R users with a higher level compositional language and good defaults for generating both basic and sophisticated trellised plots. 3. WHAT IS NOT IN THIS BOOK? R Graphics is a technical book. It does not attempt to explain the how or why of statistical graphics. It covers the technicality of plot production, not purpose, giving the reader enough rope to be extremely creative or to fail spectacularly. It does, however, point readers to other books where they might learn about good graphics. The book does not describe interactive or dynamic graphics, such as linked brushing. Many readers may have experienced the power of a tight coupling between statistical analysis and graphics through XLispStat (Tierney 1991) or DataDesk (Velleman 1988). These systems enable the user to make plots of model diagnostics that are dynamically linked to plots of the data, updating themselves as a model changes. This is very useful for exploring data, but it remains difficult to realize within R. Base graphics and grid graphics are not designed with interaction in mind so, appropriately, R Graphics focuses exclusively on static and presentation quality graphics, and the author only points readers toward recent developments in interactive graphics in R. The book's Web site provides an electronic version of the code used in the book, but little else. Itwould be useful to add material on features of grid that have appeared subsequent to the release of the book. For example, a recent release includes facilities for converting postscript files to grid objects. The description of grid graphics is the most important part of this book, explaining a new, more powerful graphics engine. However, this system is not given sufficient emphasis and does not provide enough general examples that could be easily adapted to a reader's data. Our fear is that most readers will not get past the traditional graphics part of the book. While R Graphics provides a good summary of this system, it may slow the needful death of a primitive and out-dated system. 4. RECOMMENDATION This is an excellent book. Everyone who uses R to draw graphics (all R users, we hope!) should have it open on their desk or at least on their shelves! We especially encourage readers to get beyond base graphics and carefully study the grid graphics engine. Hadley Wickham Iowa State University Dianne Cook Iowa State University