Data Visualization

This is a valiant attempt to draw together statistical and presentational methods based on the author's experience in medical research and thinking about communicating findings. There are many valid points, useful observations and clever examples; I approve the advice to use a variety of tools, including both software and sketching with pens and paper. Unfortunately, most of the examples are not fully developed and the text is uneven in tone and level: data transformations are dismissed as ‘sticking them through some mathematical function’ and comparing distributions from two samples is introduced as ‘some data on one variable and some on another’.

The book divides into four sections. Part 1 (there are inconsistent references to parts and sections) comprises ‘two chapters that everyone should read’ but should also include the preface which introduces his terminology. Part 2 ‘can be skipped if you have already studied statistics’, and this may be good advice as I frequently pulled myself up to mutter that I saw what he was getting at but felt that a beginner would be confused. Part 3 contains 10 chapters on statistical topics and their relation to visual presentations. Text boxes of ‘How to read a …’ are scattered arbitrarily through the pages, and the final section's summary of overarching ideas could be sorted and augmented.

The example graphs are often placed far from the text describing them and displayed with no context. Repeated measurements are introduced (as ‘matched data’) on page 48 with a discussion of shopping surveys but illustrated on page 52 with a multiline chart of patients’ blood pressure over time. The text then switches within the page to cross-sectional data, illustrated with Fisher's iris data before referring to data on train delays and gross domestic product. An accompanying web site of ‘code to generate my own images in the book’ contains brief comments rather than the instructions to recreate examples.

Many of the author's own graphs fail as exemplars of good practice. He advises several times to seek user feedback of drafts and to revise graphs. Unfortunately, in my experience most people do not give thoughtful analysis of graphs; they either ‘like’ or ‘don’t like’. The train delay data are plotted against month (Figs 2.6 and 16.2) showing that delays are worst in December and January, so why not plot on a scale April–March to bring the winter months adjacent? (See the example in Reese (2017a).) The same data are plotted by season (Fig. 3.6)—in the order autumn, spring, summer, winter! (See the example in Reese (2017b).) Histograms are drawn with the end value of each bin centred under its bar, and a kernel density for commuting times shows positive density for times less than 0.

The author has been badly served by any editor and by the publisher. Apart from the problems of positioning and sizing figures and the distant cross-referencing, there is an ugly page design with an inadequate bottom margin and weirdly misplaced drop capitals to introduce each chapter.

It is good that more people are thinking about ‘dataviz’ (ugh!; though Wikipedia gives this as only a company name). Grant states, ‘Scientific training discourages analysts from telling people what to think’ [really?], and Grant (2018) implies a disconnect between ‘dataviz’ operators and statistical analysts: anathema to, for example, Tukey's exploratory data analysis approach. At the price this is an interesting book that I hope can prompt users of graphics to look twice and to think for themselves, and then to use charts to display justified conclusions from data.

References