This delightful and informative guide from my friends at No Starch Press comes with the following cover blurb: “Statistics Done Wrong is a pithy, essential guide to statistical blunders in modern Science that will show you how to keep your research blunder-free.” It is somewhat pithy, but as to blunder free, I will quote the old maxim that “nothing is foolproof, as fools are so very clever.”
Still, the book has much to recommend it. It devotes time to asking the proper questions and their subsequent statistical analysis. For those who choke up when they see equations, the book has ample graphics, a few cartoons, but no math (at least any couched in the symbolic shorthand that mathematicians love). The author also covers the increasingly important concepts of power and sample size, as well as the core area of misinterpretation of p-values. The book assumes no formal statistical training on the part of the reader so the language is everyday plain. It seeks to clarify basic concepts and NOT teach the intricacies of the mathematics. Lastly, it offers up gems such as ‘studies with statistical and logical errors are not necessarily wrong, just poorly supported.’ Let me now leave you with another gem discovered in this book that was new to your editor. I quote the author “…Hanlon’s Razor directs us to ‘never attribute to malice that which is adequately explained by incompetence’…”. Now to the pithy parts…
Throughout the text there are several “markers” to draw the reader’s attention to the core names and subject matter, usually bold section headers and italics within the body of the text. The incredibly lucid explanations are always followed by relevant examples, usually medically related but at both the research and clinical levels.
His constant warnings concerning the evils of low power, and equating statistical with practical significance are priceless. The novice data analyst cannot hear this enough. Another rather large statistical gem is the admonition against reading too much into a correlation. In these days of big data, where multiple hypotheses are tested at once, we expect to see false positives and have developed methods to address the problems. Similarly for correlations. They may be of statistical significance but have no relevant, practical bearing on the underlying science.
The reference list yields a treasure of the most up-to-date references and several of the older, classic ones. In addition, the author evidences a sense of humor when, for example, in describing the physicist’s job of discovering ever more elementary particles and very small differences, the author suggests that confidence intervals be constructed on the uppers bounds of particle production and then used to compare competing theories with the data. The humor? He concludes that this is one way of justification for building bigger instruments to actually confirm the effects!
This short text is a compilation of clarity and can easily be read in a single sitting. Just remember this is a book of concepts, not a manual on calculations.
Availability
Statistics Done Wrong: The Woefully Complete Guide, by Alex Reinhart. No Starch Press, San Francisco. 152pp + xviii (2015). ISBN: 1593276206. $24.95 (ppb)
John Wass is a statistician based in Chicago, IL. He may be reached at editor@ScientificComputing.com.