The reference book for statistics with R – does it exist and what should it contain?

Background

There is a lot of discussion around this, so I thought that I could find my answer from earlier treads on StackExchange and by googling furiously. After using half a day trying to find only one reference book for (bio)statistics with R, I got utterly confused and had to give up. Maybe the free material combined is actually better than any of the books you can buy at the moment. Let’s it find out.

The internet is full of good free literature for R language, so there is really no point paying for a mediocre book, which ends up being used as an office decoration most of the time. The R home site lists books related to R and there are a lot of them. To be more exact: 115. Only one of them is advertised with words “standalone statistics reference book”. It is 8 years old now and may be outdated. The fourth edition of Modern Applied Statistics with S is even older. The R Book is often chewed out as too basic and not recommended because of lack of references, poorly formatted code and sloppy finish.

However, I am looking for one book, which I could use as a standalone reference to practical statistics (first and foremost) with R (secondary). The book should live on my office desk collecting annotations, coffee stains and greasy finger prints instead of dust on the book shelf. It should replace the collection of free pdf’s I have been using so far, not forgetting that R comes with an excellent reference library. “What is the right approach?”, “why?" and “technically, how does it work?” are often more burning questions than “how to do it with R?

Since I am an ecologist, I am mostly interested about applications to biostatistics. However, since these things are often connected, an interdisciplinary general reference would be the most valuable for me.

If such a book exists (I doubt it), please provide the name of the book (only one per answer) and a short review of the book explaining why it should be named as the reference book for the topic. Since this question is not very different than the existing ones, please use this tread for your answer. You can also list flaws of the book so that we can list those as the features for the ideal reference book.

My question is what should the reference book for statistics (of most used kinds) with R contain?

Some initial thoughts are following general features (please, update):

• Thick as a brick
• Concise, but understandable
• Filled with figures (with the R code provided)
• Easy to understand tables and diagrams describing the most important details from the text
• Easy to understand, descriptive text about the statistics / methods containing the most important equations.
• Good examples for each approach (with R code)
• Broad and up-to-date list of references
• Minimal number of typos

1. Getting Started
2. Essentials of the R Language
3. Data Input
4. Dataframes
5. Graphics
6. Tables
7. Mathematics
8. Classical Tests
9. Statistical Modelling
10. Regression
11. Analysis of Variance
12. Analysis of Covariance
13. Generalized Linear Models
14. Count Data
15. Count Data in Tables
16. Proportion Data
17. Binary Response Variables
19. Mixed-Effects Models
20. Non-linear Regression
21. Tree Models
22. Time Series Analysis
23. Multivariate Statistics
24. Spatial Statistics
25. Survival Analysis
26. Simulation Models
27. Changing the Look of Graphics
29. Index

What has been said earlier?

StackExhange contains several treads asking statistics and R book suggestions. Books for learning the R language asks about a reference book learning R language without statistics aspect. The Art of R Programming is ranked out as the best single suggestion. Book to Learn Statistics using R asks for an ideal introductory book to statistics, which is really not the same thing than a reference book. Open Source statistical textbooks ranks Multivariate statistics with R as the best alternative. What book would you recommend for non-statistician scientists? asks about the best statistics reference book without specifying the program of choice. Reference or book on simulation of experimental design data in R scores perhaps closest to my question. Introduction to Scientific Programming and Simulation Using R is the most recommended book here and might be close to what I am looking for. However, this book either won't suffice as a single reference book to statistics with R.

Some suggestions for the reference book and their flaws

R in Action has received better reviews than The R Book, yet it is apparently rather introductory.

Biostatistical design and analysis using R: a practical guide is perhaps close to what I am looking for. It has received a good review, but apparently also this one contains many typos. In addition, this book does not concentrate on explaining statistics, but rather gives statistical analyses as readymade recipes for researchers to use.

Ecological Models and Data in R skips the introductory level. This is a very useful feature seeing that word "introduction", scores 43 occurrences in the R book list, but perhaps not entirely satisfying, if we’re after the reference book for statistics…?

Introduction to Scientific Programming and Simulation Using R received very positive review, but is limited to data simulation.

Richiemorrisroe suggests that Modern Applied Statistics with S is sufficient for a standalone statistics reference book with R. This book has received excellent reviews (1,2) and is probably the best candidate for the title at the moment? The most recent version came out 10 years ago, which is quite a long time considering program development.

Dimitriy V. Masterov suggests Data Analysis Using Regression and Multilevel/Hierarchical Models. Haven't checked this book out yet.

After reading lots of book reviews, it seems apparent that the perfect book asked here does not exist yet. However, it is perhaps possible to choose one that is pretty close. This tread is intended as a community wiki for statistics users to find the best existing reference book and as a motivation for the new and old book writers to improve their work.

• (+1) for the good review! However, it seems that you have answered your own question within your own question... – ocram May 2 '12 at 13:46
• If you spent so much time figuring this out, coming up with a long list of your own, and even an outline of such a book, may be you should write one. This is a recommendation I often give on statistics and econometrics lists when somebody asks for a good review paper on [BLAH] and discusses what they don't like about the five or ten existing review papers -- write your own paper on it. – StasK May 2 '12 at 19:58

I personally thought that Modern Applied Statistics with S-Plus ticks all of the boxes you have outlined. Every example has R code, they give good references to other sources, and Venables and Ripley have a wonderfully terse and explanatory writing style which I really appreciated. I tend to re-read the book every so often, and each time I get more from it. Of course, your mileage may vary.

• I agree. I have many statistics books that are R based, and MASS4 is probably the closest to what you are looking for, but in places "terse" becomes un-readably terse and requires most statistical background knowledge than I have. That said, I have the book nearly 10 years and I keep going back to it and learning new stuff. I wouldn't let its' age put you off. Oh, and I'm now doing a stats phd :-) – Sean May 3 '12 at 18:04
• I also go back and back to MASS, which sounds like revealed preference for it as a reference book. – Peter Ellis May 8 '12 at 21:01
• Is the 1998 version of MASS much different to the 2003? Wondering if the content difference is sufficient to shell out about £50 more for it. – conjectures Apr 21 '13 at 9:38

Thanks for such a good question, and especially compiling all of that information. Unfortunately, the book you're describing doesn't exist, and to be honest, it couldn't possibly exist. If what you primarily want is a reference book for statistics, I would start with a really good book on linear models. My recommendation is Kutner et al, it meets the criteria of being greater than a brick in both volume and mass, is very comprehensive, clear, and with lots of examples. In fact, if you eliminate the R requirement, it pretty much ticks off your whole list. I refer back to it often. However, in ~1500 pages, it pretty much only covers linear models--i.e., regression, and ANOVA--there are some brief chapters on a couple of other topics, but you'll really want other books for that. Next, I would get a top-notch statistical reference book, at the level appropriate for you, for whatever other techniques you may need to work with (e.g., survival analysis, spatial analysis, etc.). If those books don't use R for their examples, you may want to get an R specific book, like one of the use-R! books, but between the documentation, the vignettes, the R-help mailing lists, StackOverflow, and CV, you may not need to. If you want to learn to program in R the right way, you should get one of those books, too. At this point, you have at least 4 books. I'm sorry, but that's the way it is. No one who works extensively with statistics has just one book that covers everything.

I don't think a book like this exists. The book that I think comes closest is Gelman and Hill's Data Analysis Using Regression and Multilevel/Hierarchical Models.

Cons:

• It's ~5 old and aimed at social scientists.

• It does not have everything on your TOC list (nothing spatial, basically nothing on time series, etc.)

Pros:

• Well-written

• It's got a list of errata and a TOC at the link

• It covers key things like missing data, which is not on your numbered list.

• It does hit most items on your bullet list.

• Lots of graphs and R code (some Bugs code for the multi-level).

I am working my way through Elements of Statistical Learning. This book covers an incredible range of techniques (so is 700+ pages) but each approach is explained clearly in a very practical, rather than highly theoretical way. It doesn't explicitly contain anything about R, however the plots and graphs are all clearly made with R and there are packages on CRAN for all the topics discussed. The authors have all been involved with the development of R (as well as a fair chunk of modern machine learning techniques).

• There's even an R package for that book: ElemStatLearn :-) – chl May 27 '12 at 22:21

I agreed with the currently top-voted answer that MASS4 was a pretty good fit to the request and have the same experience as another respondent with difficulty meeting its requirement of a fairly high level of statistical sophistication. MASS3 was in fact my first "Rbook" and it served me fairly well in that capacity. I did buy Crawley's "The R Book" and found it unsatisfactory for both an inaccurate description of the R language and being little more than a set of worked examples that seemed to lack depth of statistical theory.

However, with the passage of time, I have found Harrell's "Regression Modeling Strategies" (RMS) a better fit for the "biostatistical" focus of this question as well as having good depth. It's not an introductory text on R. For that one needs to look elsewhere and for that I recommend one of Introduction to Scientific Programming and Simulation Using R [http://www.crcpress.com/product/isbn/9781420068726] or (despite its name) "R for Dummies" written by a couple of long time contributors to StackOverflow's R posting tags. I only have RMS in its first edition when it was more focused on S, but since that time Harrell has switched over to R and fully supports the rms/Hmisc R package duo. I believe it satisfies @gung's suggestion for specialty coverage in several of the listed domains, although not for spatial analysis or mixed models.

• I'd highly recommend both RMS and MASS. I'm not in biostatistics, but most of the advice in Harrell is useful much more generally. I often ask prospective research students to read Harrell, or at the very least chapter 4, and then often recommend MASS as a good general book to make sure they have familiarity with. – Glen_b Feb 16 '15 at 0:02
• For general self-study I nominate Cox and Hinkleys' "Theoretical Statistics" and Feller's 2 volume "Introduction to Probability Theory". But that is obviously not addressing the R-part of this question. – DWin Feb 16 '15 at 0:22
• [The students I supervise are in areas outside of statistics, even though their work involves quite a lot of it... MASS and RMS are more often helpful to them than Cox and Hinkley and Feller Vol 2, though both those -- along with Kendall and Stuart -- were very valuable to my own background] – Glen_b Feb 16 '15 at 0:38

If you want to translate... (this a companion book of a 4,900 page theoretical book):

Big R Book

This book (of which I am a co-author) is a compilation of 15 years of consulting experience and teaching at undergraduate and graduate level and show only examples of R stuff for whose the details of mathematics (proofs) are given in my 4,900 pages companion books where calculations are also made by hand with numerical values (+500 pages that will be available in the next edition). This book also gives the possibility to check that the software gives the right values and it is much more fun than making calculations by hand or in MS Excel about subjects that are normally taught in graduate courses in European schools. The purpose of this book is also to show that you can use 1 software instead of many for the same results without cost (instead of using JMP + Minitab + SPSS + SAS + MATLAB together). This book also shows the weaknesses of R (package maintenance not guaranteed). It is also a compendium of highly valuable questions on various R forums and blogs. It is free and in color!

• Could you also provide the requested "short review"? Why are you recommending this book? What are the good (and bad) things about it? – whuber May 14 '15 at 2:11
• I'm one of the co-auhtor... not very neutral for a short review... – Vincent ISOZ May 15 '15 at 11:05
• That's ok--we would be grateful to hear from you what you think are the strengths of your book or a characterization of who would benefit from it. By disclosing your connection to the book (which is essential), you enable readers to account for that in evaluating what you say. I suspect that many readers would understand that your are intimately knowledgeable about this book and would appreciate what you say. Without providing some kind of review, your answer would have to relegated to a mere comment which would get relatively little attention. – whuber May 15 '15 at 15:51