Evidence and Evolution: The logic behind the science

Evidence and Evolution was written by Elliott Sober, a philosopher of science who has worked on the notion of evidence, in the statistical meaning of the word. All chapters relate to papers that he has authored or co-authored over the past few years. Evidence and Evolution does not aim to demonstrate the validity of one evolution theory against another, but rather at validating statistical ways of testing such hypotheses. While Sober establishes that creationism cannot be analysed within this framework because it fails to make predictions, subsequent chapters consider ways to evaluate evidence about natural selection (against the drift alternative) and about common ancestry, without drawing conclusions. Sober also relates very much to the original works of Charles Darwin — sometimes placing too much emphasis on his over-generalisations — but this historical touch nonetheless adds to the already considerable appeal of the book. First, an acknowledgement of my limitations is in order. As a statistician, I cannot evaluate the philosophical relevance of the book, even though the arguments are fairly accessible to a layman like me; however, I appreciate the critical assessment of Popper’s testability criterion when applied to creationism, as well as the extensive coverage of statistical principles for testing. A philosophical perspective on Evidence and Evolution is given by Pfeifer. Furthermore, being equally a layman in population genetics and evolutionary biology, I have difficulties in assessing the impact on biologists of the debate about the construction of tests, as the examples given seem to be too formalised and simplistic to be realistic. My review is therefore necessarily biased towards a statistician’s perspective and hence maybe unnecessarily critical in terms of what I perceive as a lack of proper modelling. Indeed, I bemoan the absence throughout the book of a genuine statistical framework that would allow for a complete statistical analysis of even one real dataset, including the estimation aspects that are bypassed, as this would illustrate much more clearly the concepts at work. In addition, while I understand the historical and philosophical appeal of discussing creationism, since Darwin was subjected to many attacks on this very issue, I am sceptical about the impact that the book could have on the current debate. Unsurprisingly (and I will explain why below), the book discusses creationism in very general, and hence vague, terms. There are so many possible accounts of the intervention of a god or another supernatural being in the management of the world, that to pick one in particular would be like grabbing at water — all remaining versions emerging unscathed from a detailed criticism of the chosen one. But to maintain that creationism is testable (in Popper’s sense), as Sober does, is to open a similar ‘Pandora’s box’ about which version he is considering. I thus personally deplore the inclusion of Chapter 2 in the book, even though this position is not taken on statistical grounds. This review proceeds linearly through the four chapters of the book: ‘Evidence’, ‘Intelligent design’, ‘Natural selection’ and ‘Common ancestry’. To reiterate, it focuses primarily on the statistical aspects of the debate initiated by Sober, without discussing the biological or philosophical consequences. Although a hard read at times, proceeding through Evidence and Evolution was a worthy and rewarding experience for me that led me to re-think the terms and objects of statistical testing when applied to a scientific theory. The book is accessible to the layman (in any of the three fields — philosophy, biology, statistics) and I thus encourage readers to persevere with it.