More statistics booksreviewed by T. Nelson
Reviewed by T. Nelson
This one has a lot of valuable information for the R user that you won't find in most other books. Even though they're not statisticians, the authors did a fantastic job in finding a variety of different statistical functions, many of which are not in the base R distribution. You need more than one function for each task because lots of them have parameters that books like this don't describe but which will change the result, making the function, for all practical purposes, useless. There are also chapters on how to reformat data and so forth.
For example, I know of no other R books that describe how to do a one-way repeated measures ANOVA; this book has four different ways. Unfortunately, the term “concise” is not in the authors' vocabulary. If the information is in here, you have to read page after page of bad jokes to find it. If they had written this book while sober it would have been half the size—or they would have space for more thorough explanations.
As an example, on page 809, their pretend questionnaire that they use to teach exploratory factor analysis has questions like:
and on and on like that. I got this book for one specific task, and it saved my
butt, but even with nearly 1000 pages I still couldn't get enough explanation here
of important functions like
lme() to get them to work. The
nlme manual alone on CRAN
is 336 pages long. So one way or the other, I'm going to have to spend my next
vacation plowing through a bunch of useless prose. Rememmmmmber meee .....
dec 10, 2016
Disclaimer: No way am I going to read this whole book.
Reviewed by T. Nelson
It is said you can never have too many statistics books. But is that really true? Suppose each statistics book teaches 50% of what you need to know, and their coverage is perfectly random. With this book you can calculate how many statistics books you will need to achieve a given level of proficiency.
The goal of this one is to teach the basic concepts of statistics. It's very clearly written and it doesn't try to teach every conceivable statistical test. Wilcox just says: this is the best software, this is the best test to use, and this is what it does. Although the book uses R, it's also not intended to teach R.
Wilcox is an advocate of what is known as robust statistical testing, which is
a modification of standard tests to reduce the effect of outliers. Wilcox's WRS
library incorporates these robust tests, and they're easier to use than the
standard ones. For example, for a robust one-way repeated measures ANOVA you
only have to type
tr causes it to 'trim' 20% of your data as outliers.
To get a standard test you just set
What you get here are just the basic stuff like ANOVAs, multiple comparisons, and regression. Each test is introduced by the relevant equations and accompanied by sample data and several example calculations. Tables are included to allow hand calculations.
The WRS package is welcome for R users, since many of the non-base packages on CRAN nowadays either don't compile or refuse to install. The packages have many dependencies, and there always seems to be one that gives problems. So you end up trying half a dozen different tests before finding one that actually works. The WRS package is superior in that regard, though even it doesn't install on every machine.
I'm still coming to grips, though, with the fact that different statistical tests often give conflicting answers as to whether something is significant. This is a big problem when dealing with collaborators who insist on trying one test after another until they get that p-value below 0.05. Robust tests add another layer to this complexity. This book and Wilcox's more advanced book Introduction to Robust Estimation and Hypothesis Testing provide the necessary background to help you deal with all the fallout.
Make sure you can get WRS or WRS2 to install before getting this book. WRS is better. A few PCisms and an incomplete index but an otherwise well written and very useful textbook.
dec 25, 2016