Book Review: Statistics Done Wrong
Send us your comments.
I sadly cannot say, that I have established the habit of writing reviews on everything I read, but I wish that I had. But it probably is not to late to start. Especially as the main benefit is not that others may read this review, but as it is with so many things, the one writing the review gains much more from reading, if he is able to rephrase and evaluate the primary statements from an article or book.
“Statistics Done Wrong” by Alex Reinhart (no starch press, 2015) seems to be fueled by the most productive kind of frustration there is. The frustration of someone who is is deep love with his field of research seing the importance of his work while observing that its contributions are severly misunderstood in areas where they are fundamental ingrediences.
For many years I despised statistics, although I have been one of two out of 120 students in my cohort sitting through the voluntary one-semester course on statistic in the incredible stressful 4th semester. I cannot remember anything from that course, and have avoided statistics up until a few years ago, when I could not but accept that my research very much was founded on stochastic models. Statistics and stochastic are now the single most dominant topic in my personal computer-science library. I am now very much astonished how this field wrangles clarity and knowledge from overwhelmingly large, confusing sets of data.
But knowing models and algorithms doesn’t mean actually understanding them. I seem to have missed the opportunity to get deep intuition from my maths teacher when I had the chance and deriving it from text-books has been a struggle, with nobody giving you feedback when you run into dead ends. (I am not sure whether I am still running into the wall at the end of a wrong turn.) Well, nothing can replace a good teacher, and I still am not sure whether I grasped anything correctly.
But this book did something might not even have been intended. By discussing shortcomings of methods by way of low complexity, mostly informal discussion of examples, both from actual scientific publications and toy scenarios, by venting the frustration about failures that could easily have been prevented, I have a feeling of understanding. It probably helps to be an academic with similar experiences from academic work to feel a good connection to the author.
You should read this book if you want to grasp statistics and to avoid (some of the) traps of empirical studies. But you will be enlighted when you see how the author derives a wholehearted criticism on the current science industry from this topic. This turns the endless laments we share in campus halls into well founded criticism without needing to resolve to the boring contradictory guidelines of science philosophers on how to conduct science. Alex Reinhart is focussing on the field he probably knows best and that is sufficient to find an expression for things going dramatically wrong in contemporary science.
I will be going to read it again, as soon as I have clarified some things in the textbooks mentioned above.