SFLWR3: shameless self-promotion

59 views
Skip to first unread message

Stefan Th. Gries

unread,
Oct 12, 2020, 9:21:21 PM10/12/20
to StatForLing with R
Hi all

In an act of shameless self-promotion, I am writing to let you know that SFLWR3 is essentially finished. I'm still working on the exercises bits (which will be HTML pages knitted from RMarkdown this time) but the book itself is 99% done: I am doing the final editing 'as we speak', the cover has been picked, and soon I'll send the whole thing to Mouton.

To let you know what to expect, you can look at the cover here, you can look at the new navigator here, and this is (the 99% version of) the preface (in RMarkdown, which is how I wrote the book):

***

When I wrote the second edition, I never ever thought there was gonna be a third one. But then, over the last few years, so much has been happening with regard to quantitative methods in linguistics, I myself have learned so many things (many of which I should probably have known way earlier ...), I have reviewed so many papers from which I learned things, methods that in 2012 were still rarer and or being debated (in particular mixed-effects models but also a variety of tree-based approaches) have become somewhat better understood and, thus, a little more mainstream, and I have taught dozens of bootcamps and workshops where people wanted to go beyond the second edition. As a result, at some point I started thinking, ok, I guess there *is* interest, let's see what De Gruyter says -- turns out they were happy to have a third edition so here we are.

While the first four chapters of course saw many changes and additions, it's actually again Chapter 5 that was revised quite a bit more: I am still using two of the same data sets as in the 2nd edition and I still place great emphasis on the coefficients etc. that regression models return, but I got rid of the sum contrasts bits and added a ton of new stuff, all of which has been used in multiple courses and bootcamps over the years; these include more discussion of curvature, *a priori* orthogonal and other contrasts, interactions, collinearity, `effects` and now also `emmeans`, autocorrelation/runs, some more bits on programming, writing statistical functions, and simulations, exploring cut-off points in logistic regression etc.; also, there's now more discussion of issues of model selection, diagnostics, and (10-fold cross-)validation. And, 'by popular demand', there's now a whole new chapter on mixed-effects modeling, with detailed discussions of all sorts of aspects that are relevant there, plus there's a whole new chapter on trees and forests. I am very grateful to (i) hundreds of bootcamp/workshop participants, who of course often requested that kind of content or invited me to teach just that, and to (ii) students who took (parts of) my 3-quarter statistics series at UC Santa Barbara or the regression block seminars at my other department at JLU Giessen for all the input/feedback I have received over the years, much of which allowed me to fine-tune things more and more. I am certain that, if any of them read this book, they would recognize examples and sections of this book as having been 'test-driven' in 'their bootcamp' and I am certain that readers of this book will benefit from the many iterations I have been able to teach this stuff till, I hope, it was structured in the best possible way -- well, at least the best possible way I was able to come up with ...

I want to close with a plea to really engage with your data. Yes, *duh!*, obviously you were planning on doing that anyway, but what I mean is, treat your data analysis like detective work. Yes, that sounds like a trite cliché, but the more/longer I've been doing this myself, the more apt it seems: The data are a suspect trying to hide something from you and it's your job to unravel whatever it is the data are hiding. As I have seen over the years, it is (still!) soooo easy to fall into traps that your data are presenting and overlook small (or even big) red flags. Always be suspicious, always ask yourself "but didn't he (STG) or the author(s) of the paper you're reading just say ...?" And always be suspicious about what you just did yourself. Always ask yourself "if I wanted to shoot my paper down as a reviewer who had access to the data, how would I do it?" and then 'insure yourself' against 'that reviewer' by checking *x*, exploring alternative *y*, testing alternative method *z* ... Bottom line: to not be led astray, you need to interrogate your data in the most sceptical of ways, and with this book I'm trying to help you do this!

***

Finally, here's the (99% version of the) table of contents (and yes, chapter 4 has changed, too, even if it doesn't look like it) and the writing style has become a bit less formal. I hope you'll like it,

STG
--
Stefan Th. Gries
------------------------------
UC Santa Barbara & JLU Giessen
http://www.stgries.info
------------------------------

Stefan Th. Gries

unread,
Feb 11, 2021, 11:26:40 PM (13 days ago) Feb 11
to StatForLing with R
What you have all been waiting for is just around the corner! No, not some vaccine - something that will make the time time waiting for the vaccine just fly by: the third edition of SFLWR!

I just sent what should be the final version of the camera-ready copy to production at De Gruyter, who should process everything while we're still waiting for 1 or 2 more back cover endorsements, and then it should be coming out very soon. Some info on 'the final product':
  • This edition is nearly 40% longer than the last one;
  • the R code version of the code file has about 5000 heavily commented lines, if you pdf the HTML version of the Rmd code file, it's 116 pages;
  • if you PDF all the HTML answer keys to the exercises, you get around 241 pages from the book plus an additional 200 pages for an answer key to one exercise from 'elsewhere';
The kind back cover endorsements we have received so far are the following:
  • A tour de force of cutting-edge statistical thought, analysis, and visualization which accomplishes the impossible: sometimes polemic, often hilariously funny, and always highly instructive and rich in detail, it is, still, easily accessible.
  • This 3rd edition of Statistics for Linguists with R is a must-read for anybody who is interested in quantitative analyses of linguistic data – from absolute beginner to expert. The explanations are written in plain, down-to-earth but at the same time extremely precise language. In addition to introducing basic concepts of data analysis and working with R, the most common statistical procedures are described and exemplified in a very reader-friendly and didactically intelligent manner. I can highly recommend this new edition of Statistics for Linguists with R to teachers and students dealing with statistics in the language sciences.
  • "The third edition of 'Statistics for Linguistics with R' gives a boost to everything that was already great about its predecessors. Its major asset continues to be the accessible hands-on approach, which is supported by a wealth of well-tested data examples. In addition to that, the new edition gives us substantial updates that present and explain recent methodological developments. The book is thus not just excellent for teaching, but also as a highly usable resource for working linguists.
  • For many researchers and students of linguistics, Gries’ Statistics for Linguistics with R is the water gear they wore when they first dipped their toes into the waters of quantitative data analysis – and many still use it when they find themselves lost at sea. With its clear (and sometimes refreshingly informal) explanations of basic concepts as well as more complex statistical models, and its abundance of demonstrations, exercises and ‘think breaks’ on linguistic case studies, this updated third edition is an ideal textbook for taught courses and self-study alike. Beyond being a ‘how to’ for statistical analyses in R, the book also provides helpful tips on how to motivate (and critically evaluate) the use of quantitative methods in Linguistics, and how to structure and report quantitative studies.
So, update your R (to at least 4.0.3), update your RStudio (to at least 1.4.1103), and pre-order your copies ;-)
Reply all
Reply to author
Forward
0 new messages