I've been reading Tukey's book "Exploratory Data Analysis". Being written in 1977, the book emphasizes paper/pencil methods. Is there a more 'modern' successor which takes into account that we can now instantaneosly plot large data sets?
$\begingroup$
$\endgroup$
5
-
$\begingroup$ should this be community wiki? $\endgroup$– richiemorrisroeFeb 8, 2012 at 10:35
-
$\begingroup$ It's not clear to me whether this ought to be CW. There may be no good answers; there may be one clear outstanding answer; we might generate a long list of effective answers. Let's see what happens. $\endgroup$– whuber ♦Feb 8, 2012 at 17:17
-
4$\begingroup$ This is a good question, biofreezer. I just wanted to remark that there are close analogies to other methods of work. My favorite is, pen & paper EDA is to modern stats as hand tools are to modern woodworking. ("Modern" woodworking employs many power tools like tablesaws and routers that enable even beginners to turn out acceptable results in much less time. However, these tools also account for thousands of missing digits and limbs every year. People who learn to use hand tools generally learn to work better and more efficiently even when they employ power tools.) $\endgroup$– whuber ♦Feb 8, 2012 at 17:21
-
4$\begingroup$ Yes, woodworking is a nice analogy (missing digits, missing digits). See also software-carpentry.org. $\endgroup$– denisDec 13, 2012 at 15:46
-
$\begingroup$ Update ten years later: IMHO, most answers, helpful though they may be, implicitly misunderstand Tukey's EDA. They all focus on visualization, which is the most minor part of the book. AFAIK, there is no modern successor, because few people have learned or apply the creative, insightful exploratory methods described there. The most difficult and perhaps the most underappreciated parts of the book concern techniques of creating associated datasets whose analysis permits you to discover patterns in the original data. One example: log-log regression of spread vs. level plots. $\endgroup$– whuber ♦Apr 6 at 13:52
Add a comment
|
9 Answers
$\begingroup$
$\endgroup$
1
The closest thing is Cleveland's Visualizing Data. It's about Exploratory Data Analysis, it's about computer-generated visualizations, it's profound, it's a classic.
-
2$\begingroup$ The same also applies to the book The Elements of Graphing Data by the same author. Buy both of them; they are both excellent. $\endgroup$ Feb 2, 2014 at 8:19
$\begingroup$
$\endgroup$
Well, its not an exact replica, but I found tons of useful plotting advice (and R code) in Gelman and Hill's Data Analysis using Regression and Multilevel/Hierarchical Models
In addition, his blog is often full of useful graphics advice.
$\begingroup$
$\endgroup$
Interactive Graphics for Data Analysis: Principles and Examples is one I like; the book description says it "discusses exploratory data analysis (EDA) and how interactive graphical methods can help gain insights as well as generate new questions and hypotheses from datasets."
$\begingroup$
$\endgroup$
Hadley Wickham's ggplot2 book is interesting because it teaches both the Grammar of Graphics and how to use the ggplot2 software.
$\begingroup$
$\endgroup$
Ronald Pearson's Exploring Data in Engineering, the Sciences, and Medicine is worth mentioning here. Its main target readership seems to be scientists not afraid of a little mathematics who wish they knew more statistics. That is quite a large group, and one well represented here. It's a little quirky and offbeat, but it covers a lot of ground and it includes much sensible advice. It's not Tukey revisited in the sense that it offers many new ideas, but it can be rewarding to study, even when you think it is a little wrong-headed.
This book seems to have attracted very little notice, quite possibly because it is very expensive, not obviously suitable as a course text, and as yet only available in hardback. But it is intelligent and readable and free of the garbage of modern introductory textbooks (pages and pages of elementary exercises, silly icons, gratuitous photos of happy young people, fussy layout with boxes, whatever, etc.).
$\begingroup$
$\endgroup$
Also Interactive and Dynamic Graphics for Data Analysis: With Examples Using R and GGobi, Cook and Swayne
This has two chapters publicly available on the web that describe the process of data analysis, and handling missing values. There's a new book coming out by Antony Unwin soon.
$\begingroup$
$\endgroup$
Claus Wilke's 2019 book "Fundamentals of Data Visualization" is another possible "modern successor." The book's preprint is still freely available online.
Like Tukey's EDA, Wilke's book is focused on exploring your data using graphs while keeping in mind the things that matter to statisticians: thinking in terms of distributions, thinking about precision & uncertainty in our estimates, thinking about bias-variance tradeoffs when smoothing a trend or choosing a histogram bin size, and so on.
Wilke assumes you'll be making your graphs on the computer and provides the code for all his graphs (mostly in R's ggplot2) on GitHub. But the book itself is written in a software-agnostic way: the text is about best practices, not about how to implement them in a specific software tool. There's a brief chapter on choosing the right viz software tool for your needs.
He also concisely introduces concepts like Wilkinson's Grammar of Graphics; recommends best practices in line with folks like Cleveland and Tufte; and discusses how to make effective graphics for communication, not just exploration. Wilke's book does not break new ground on these fronts (unlike the Tukey or Cleveland books mentioned in other answers), but rather does a great job of distilling it and putting it all in one place, illustrated with good/bad/ugly examples using real datasets. It's become my go-to book for introducing data visualization to statisticians.
$\begingroup$
$\endgroup$
5
Another couple of good books to read are Beautiful Visualization and Beautiful Data. These are edited books, there are amazingly good examples of exploring data with plots, and some absolutely appalling chapters.
Another book that has some good examples of using ggplot2 is a new one by Winston Chang
-
1$\begingroup$ I just want to double-check, Di, in case a subtle typo crept in: did you perhaps mean to write "appealing" instead of "appalling"? Although both make sense in this context, the appearance of the latter--without any further explanation--is rather a surprise! $\endgroup$– whuber ♦Sep 8, 2014 at 15:31
-
2$\begingroup$ appalling was correct - it is a mixed bag - edited volumes often are $\endgroup$ Sep 8, 2014 at 22:32
-
$\begingroup$ I'm surprised at these recommendations. I found both books mostly disappointing (long on guff, short on graphics). Unfortunately O'Reilly, which I first encountered as the publisher of spectacularly good Unix books, seems to have very uneven quality control for books on anything even remotely statistical. $\endgroup$– Nick CoxSep 9, 2014 at 12:45
-
$\begingroup$ I like both books, and really feel that they are substantial contributions. Winston Chang's has a lot of basic details on plotting with ggplot2. It is a good beginners reference. It does not tell you much about why you would make these plots, but most make good sense for the purpose, from the pieces that I have read. The Beautiful Visualization has some very impressive chapters, tackling difficult problems like visualizing wikipedia, massive data, many complexities, and it goes through the thinking process/decisions taken to make the plots. $\endgroup$ Sep 9, 2014 at 19:47
-
$\begingroup$ Just in case my comment is ambiguous: I was referring to the "Beautiful" books. Winston Chang's book is nice and helpful. $\endgroup$– Nick CoxSep 9, 2014 at 22:52
$\begingroup$
$\endgroup$
I think of Understanding robust and exploratory analysis by Hoaglin, Mosteller and Tukey an the companion volume on Exploring data tables and shapes as the technical follow-up to EDA. I also see data analysis and regression, a second course in statistics by Mosteller and Tukey as follow-up to EDA. The various Cleveland books mentioned above are treasures.