The Art of Statistics: How to Learn From Data

The book at hand is one of many presenting statistics to the layperson.  (You can find a great many more at Milo Schield's StatLit site at http://www.statlit.org/.) It could also serve as a secondary text or source of readings for an introductory statistics course.  A common flaw of this genre is that some authors are a bit weak in their own understanding of statistics.  This should not be a problem here: the author has served as president of the Royal Statistical Society.  He also avoids the common flaw of books written by experts, that of being incomprehensible to the layperson.  Here the author writes at an appropriate level, is usually clear, projects some personality, along with a touch of humor, and even a bit of sex.  The one problem with the writing is that the author is a bit too sketchy in a number of places where he could have been more informative without becoming too technical. 

 

Most of the book aligns nicely with the content of a first course in statistics.  It delves into some meaty issues that go beyond crunching the numbers and are very important in actually using statistics.  In some areas, it goes beyond the usual first course by discussing observational studies, "big data," and the bootstrap.  The last topic is the only one where I would question the author's statements.  On page 200 he gives bootstrap "95% uncertainly intervals" for the mean for samples of size 10.  Several years ago David Diez showed in an undated post online that such intervals only capture the true mean about 90% of the time in the best case situation of a normal population, and that the coverage depends on the population distribution so you do still have to check conditions.  Diez elaborated on his results in a much later post .  

 

For the layperson, the book is a good read, but the practical minded may feel that it is too much about what statisticians think is interesting and not enough about how statistics comes up in people's everyday lives.  For example, the first and main application discussed concerns a medical doctor in England who murdered hundreds of his own patients.  The first layer of analysis involves descriptive statistics in an effort to determine how this happened.  A second, later, layer discusses how such events could be detected earlier and so partially prevented.  It makes for a gripping detective story, but we have to hope this is not an application of statistics that many of us will ever need to use.

 

So, the book is recommended to the intellectually curious, or as a supplement to a first course.  For the very curious a website provides R code for the analyses and figures as well as a long list of errata.  The first figure in the book, concerning the murderous doctor, is strikingly revealing.

---

 

After a few years in industry, Robert W. Hayden (bob@statland.org) taught mathematics at colleges and universities for 32 years and statistics for 20 years. In 2005 he retired from full-time classroom work.