This Post Has 2 Comments

1. Good condition
   Received in good condition

   Reply

2. Wide-ranging examples and much detailed code, but quite who is it for?
   Nathan Yau is well known in visualization circles for his website FlowingData and his two books, “Visualize This” (2011) and “Data Points” (2013). Now we have a second edition of “Visualize This”. Second editions are often disappointing, but this one is much reworked.Executive summary. Yau is experienced and broad-minded about visualization and here covers a wide range of ideas and examples. My chief doubt is quite who this book is ideal for. It may be too code-intensive for beginners but is limited in depth for anyone who needs a sharper statistical or scientific edge.Style. Yau’s writing is informal and chatty and seasoned with small personal stories. He is frank in admitting some graphical dislikes and some areas of ignorance. Copy editing and proof-reading haven’t weeded out some awkward phrasing. Fastidious readers will note much repeated use of “a lot of”; some difficulty in matching singulars and plurals; confusion between “intuitive” and “familiar”; and a few other notable tics. There is a sprinkling of typos in the 273 numbered Figures (Mozarella; flate; distoration; incoporate). This isn’t an academic or scholarly book, and formal references are few and capricious. But friendly is a one-word summary on style.Code. A great strength of “Visualize This” is the large number of code examples, using Python, R, Adobe Illustrator, Datawrapper, RAWgraphs, and HTML, JavaScript and CSS combined. Like most authors, Yau’s recommendation is that you do what he does, in his case pick up whatever language or environment is needed to do the job. Naturally there is a big difference between being able to copy examples from a book and being able to fly, or even walk, unaided with your own code. It’s hard to know how many beginners in visualization want to hear such a message. Many books on visualization focus on one way to code, with some risk of built-in obsolescence as particular software changes, becomes unfashionable, or even disappears. (Naturally, setting coding issues aside helps to ensure that a book remains interesting years or decades after publication. Many durable visualization classics from the 1970s and 1980s did precisely that.)Statistics. In contrast, Yau is light on statistics, expecting that you have met say mean, median, minimum, maximum, percentiles (including quartiles) and mode. If you haven’t met these, or retain only fuzzy memories, there are some belated definitions on pp.297-298. Correlation is introduced gently and largely qualitatively. Attempts to go beyond this minimal core have limited success. Spline functions appear briefly with no precise hint whatsoever on how they are defined or tuned; the brief example is unconvincing (pp.109-112). The definitions of the Tukey criterion for showing distinct points on a box plot and of probability density are strictly incorrect at pp.298 and 315. Those interested in visualization linked to statistical modeling or multivariate analysis need to look elsewhere.Mathematics. Yau doesn’t go beyond what should be remembered from high school. Logarithmic scales are mentioned without being exemplified. By the way, the relationship between length and area is quadratic, not exponential (pp.152-153).Graphics. The main attraction of the book is its great variety of different kinds of graphics, some standard, some not so, some with quirky names. All books on visualization are personal ragbags, even those presenting a serious tone and an academic face. Here the most successful visualizations are variations on dot or scatter plots, line and bar charts, and simple distribution or choropleth maps. Among examples I would class as over-rated are most uses of stacked bars or areas; pie charts; tree maps; and unit charts (here called waffle charts or square pies). Sorting values, showing differences explicitly, and choosing colors carefully are vital ideas that deserved an even greater push, as was plotting groups or variables in turn with others as backdrop, a device elsewhere called front-and-back plotting. Dot charts (in Cleveland’s sense), mosaic plots, and scatter plot matrices also appear too briefly for my taste.Creativity and criticism. In visualization, as in all art or science, there is tension between encouraging creative action, such as graphs that are different or look pretty, and exercising critical discrimination. What does this visualization show us that we didn’t see before? How far does it work well, or at least better than the alternatives? These simple but also hard questions are often ducked here, despite some scepticism about say hairball representations of networks, donut charts, and some cartograms. Yau isn’t generally keen on precise rules. There is one notable exception, laying down as law that bars should always start at zero (pp.95, 341). This instruction is often given but also exaggerated. I’ve seen bars that start at parity 1 as reference, at means or other summaries, and at 32 deg F as a critical temperature. The linking thread is that bar heights or lengths encode difference from some reference level as a distance. That reference level is often zero, but could be something else. No one complains in my reading that this precept is flouted by box plots, which start by stacking paired bars around medians, not zero.What has been cut down or cut out from the first edition? Notable omissions or reductions include Chernoff faces, loess smoothing, Nightingale charts, star charts, stem and leaf plots, and stream graphs. These changes seem well judged.

   Reply