Bests & Worsts Reviews from Amazon

Let me front-load the criticism. I wish an experienced statistics instructor had reviewed the manuscript. The book does better in its second half, where it discusses what I would call problems with empirical-research culture, than in its first half, which has more textbook statistics. The author neglects to explain the basics - things like "sample", "statistic", "sampling distribution", "conditional probability" - and often confuses matters by bringing in issue Y when setting out to discuss issue X. (Appropriately, a section named "Confounding Confounders" is itself confounded: we start talking about "coarsening" data (not what I expected based on the title, by the way; a Y-for-X switch already took place), then get into something else. I will single out the introduction to the "base-rate fallacy" as another weak spot). A choice to be non-technical means that solutions to some problems cannot be effectively presented - although sometimes they are suggested after all. The "woefully complete" part of the title is, I take it, tongue-in-cheek, so no quibbles there. A few "similar" books come to mind, including (a) the drier "Common errors in statistics" by Phillip Good, (b) the three terrific popular books by Ben Goldacre - "Bad science", "Bad pharma" and "I think you'll find it's a bit more complicated than that" - and (c) the elegant "Understanding the new statistics" by Geoff Cumming. (I have not seen "How to lie with statistics" by Huff and Geis). Reinhart's book is more "big-picture" than Good's, and broader than Goldacre's or Cumming's. (The latter is a perfect "single-issue" book; the former are not specifically about cataloging statistics errors). Statistical semi-literacy of empirical researchers is a serious problem, and any effort to improve the situation is to be lauded. Alex Reinhart's book - engagingly written, and nicely produced (and fairly cheaply sold) by No Starch Press - is a force for good, and one which can have a material impact.