See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/315993111

R for Data Science

Article in Journal of statistical software · April 2017

DOI: 10.18637/jss.v077.b01

CITATIONS READS

0 5,977

1 author:

Christopher J. Lortie
York University
187 PUBLICATIONS 7,250 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Direct and indirect consequences of dominant plants in arid environments View project

The importance of non-independence in social media science amplification. View project

All content following this page was uploaded by Christopher J. Lortie on 14 August 2017.

The user has requested enhancement of the downloaded file.

JSS Journal of Statistical Software
April 2017, Volume 77, Book Review 1. doi: 10.18637/jss.v077.b01

Reviewer: Christopher J. Lortie

York University and NCEAS

R for Data Science

Hadley Wickham, Garrett Grolemund

O’Reilly, Canada, 2016.
ISBN 978-1-4919-1039-9. 522 pp. USD 39.11 (P).
http://r4ds.had.co.nz/

Data science is a complex domain, and decisions associated with wrangling big and little data
are non-trivial (Gandomi and Haider 2015; Peters, Havstad, Cushing, Tweedie, Fuentes, and
Villanueva-Rosales 2014; Marx 2013). This book is written as a general resource for R by
providing a complete data science workflow, i.e., a set of steps for specific packages. The
workflow or set of steps is the anchor for the book and is developed immediately within the
preface. Import, tidy, transform-visualize-model iteratively, followed by communicate. The
workflow is described in text, illustrated, and sets of chapters are linked to the workflow
throughout the book. This structure provides an excellent backbone to the content and
facilitates its use as a resource because one can easily revisit a specific chapter for reference
when working through a real problem. The meaning of each step is self-explanatory but
nonetheless well defined and demonstrated by worked examples throughout the book.
The efficacy of communication for this statistical software and implementation book was eval-
uated using the following criteria: clarity of writing, supporting visuals that make complex
data science concepts accessible, and an appropriate balance between detail and general un-
derstanding of process. ‘R for Data Science’ was successful in all three potential dimensions
of communication. The writing is direct. Most chapters lead with code, examples, then the
description follows. This exposes the reader more rapidly to the relevant material needed to
grasp and do the data science. The book is primarily written in a show-then-tell format, and
this approach reduces the need for the reader to process large chunks of description (intro-
ductions are very brief in each chapter). Telling one how to do something versus showing it
directly can of course be appropriate in some contexts, and readers have different learning
styles. Nonetheless, showing the data science first engages and challenges the reader to read
the R code and learn the grammar. Reading code others have written is an important skill
and considering a problem before seeing the solution stimulates deeper learning. If anything,
there could have been even more development of the problem-solution model in the writing,
but I recognize that this can sometimes come at the cost of clarity and can tax the patience of
readers at different levels. There are exercises provided to consolidate learning and they are
pitched at the right level consistent with each chapter. The supporting visuals excel (but not
2 R for Data Science

Excel, pardon the pun) at visualizing the layered grammar of graphics in ggplot2, relational
data with dplyr, and subsetting with vectors. Visual learners will appreciate the concepts
illustrated, use of color, and a certain to be favorite – the pepper shaker, with pepper packet
in it, with pepper in the packet – to illustrate subsetting of lists of lists. Most chapters bal-
ance detail and general understanding of process well. This it not to say that the details of
coding were never a challenge to reconcile with the big picture. Many data science and coding
concepts are complex. The ‘Iteration with purr’ chapter was a challenge in merging and con-
trasting the details between different options such as ‘for loops’ versus functionals. However,
later chapters such as those in the model section struck a better balance. This difference can
in part be due to an audience experience bias such as one’s background in statistics versus
data science. This suggests that different audiences will be able to better capitalize on the
show-then-tell approach depending on their experience. The book is thus well pitched for
beginner to intermediate data scientists and likely for statisticians with an intermediate level
of experience with data science concepts and approaches. The communication and writing
style is accessible and not unduly technical for all readers.
There is extensive support for R available in the form of documentation (documentation for
R directly and reference manuals and vignettes for CRAN packages), FAQs, StackOverflow,
blogs, webinars, workshops, and many books (and many are also free). Too much informa-
tion, not too little is most likely the challenge for data scientists and statisticians working
in R. For the R community in particular, the breadth and scope of packages, discussion, and
documentation are unparalleled. Typically, this is a benefit in solving a problem, and fre-
quently, there is no one single solution but many. However, processing and parsing responses,
solutions, and code from different sources is time consuming and, at times, overwhelming. ‘R
for Data Science’ is a logical, contemporary entry point that compiles a relatively consistent
set of current R packages together into a clean data science workflow appropriate for many
purposes. The book is built up from extensive package development, and both R and its
packages will continue to evolve. The book reframes and updates a ggplot2 book (Wickham
2009) and complements the updated book (Wickham 2016). It explains the philosophy and
grammar of this package succinctly. It also further develops the concept of ‘tidydata’ (i.e.,
columns as variables, rows as observations, Wickham 2014). The concept of this mapping of
data is not unique to the ‘tidyverse’, but this ecosystem offers functions to easily deal with
some frequent types of inconvenient data and to readily wrangle and specify what constitutes
a variable and an observation so that the concept of tidydata makes sense. Tidydata thus
set up dataframes for more efficient processing. This ecosystem of packages, its grammar,
and the thinking are better situated within the domain of data science through the book.
The novelty in this book is a coherent workflow across different concepts and packages. It
is a solid foundation for the statistician interested in learning and improving data handling
skills. For the data scientist versed in the extensive resources distributed online for R, it is
an integrated set of resources and sample code that can readily provide and affirm a literate,
reproducible philosophy of data science. It is not about efficient programming or coding in
R, it is about efficient data science.
‘R for Data Science’ is an excellent resource. If you are already familiar with this ecosystem
of packages and ideas, it is nonetheless still valuable. You may be reading about many of
the approaches and tools you already use or have seen, but in seeing them organized and
described, in many instances by the authors of the packages, one gains novel insights. Even if
you do not agree with the assumptions in full, the documentation and logic described provides
 Journal of Statistical Software – Book Reviews 3

a more complete sense of how data science needs, package development in R, and the goal of
integration are useful for statistical languages. Open science development can rapidly provide
us with new packages, but sometimes connecting and understanding them is a challenge.
This book is thus an excellent example of the value of documentation beyond vignettes that
facilitates deeper learning and appreciation of the landscape and not just the details of the
moment. When using R, it is not uncommon to be in the midst of a problem, rapidly look
up a solution online (from whatever resource works), and move on. The solution may or may
not come from a book, and if it does, one captures the relevant code or explanation from
the snippet only. This begs the question of investing in a complete book. For this book, I
recommend the investment: time you enjoy wasting (on a technical book like this one) is not
wasted.

References

Gandomi A, Haider M (2015). “Beyond the Hype: Big Data Concepts, Methods, and
Analytics.” International Journal of Information Management, 35(2), 137–144. doi:
10.1016/j.ijinfomgt.2014.10.007.

Marx V (2013). “Biology: The Big Challenges of Big Data.” Nature, 498(7453), 255–260.
doi:10.1038/498255a.

Peters DPC, Havstad KM, Cushing J, Tweedie C, Fuentes O, Villanueva-Rosales N (2014).

“Harnessing the Power of Big Data: Infusing the Scientific Method with Machine Learning
to Transform Ecology.” Ecosphere, 5(6), 1–15. doi:10.1890/es13-00359.1.

Wickham H (2009). ggplot2: Elegant Graphics for Data Analysis. Springer-Verlag, New York.

Wickham H (2014). “Tidy Data.” Journal of Statistical Software, 59, 1–23. doi:10.18637/
jss.v059.i10.

Wickham H (2016). ggplot2: Elegant Graphics for Data Analysis. 2nd edition. Springer-
Verlag, New York.

Reviewer:
Christopher J. Lortie
York University and NCEAS
Biology
Toronto, Canada, M3J1P3
E-mail: [email protected]
URL: http://www.christopherlortie.info/

Journal of Statistical Software http://www.jstatsoft.org/

published by the Foundation for Open Access Statistics http://www.foastat.org/
April 2017, Volume 77, Book Review 1 Published: 2017-04-03
doi:10.18637/jss.v077.b01

View publication stats