Longtime readers may recall this farce of a post, which hinted at a crucial lesson from this book - good data visualization is based on just a few basic principles. My post from three years ago mentioned a basic rule of thumb for preferring charts to graphics. It also shared Tufte's concern regarding the problem of variation in the design instead of in the data. I returned to this book late in 2019 when I suspected that internalizing its lessons would prove hugely beneficial in my new role.
The Visual Display of Quantitative Information by Edward Tufte (October 2019)
My book notes serve as a decent shortlist for good data visualization principles, but let's highlight a couple of the points I've found most helpful over the past eighteen months.
First, a good graphic encourages the eye to make data comparisons. Like with much of the advice in the book this sounds simple enough, but I often find myself failing to meet this standard. One obstacle is replication, in the sense that a graphic which was appropriate at one point in the past may lose its relevance over time as the underlying dataset changes. The solution is to make adjustments according to how the data changes, which is particularly challenging when the changes happen subtly over a prolonged period or there is significant pressure to present a consistent visual representation. However, without being open to change it becomes impossible to meet this particular standard. The best way to allow for eyeball comparisons is to use a visual that takes on the shape of the data - for example, by using a tall rather than wide graphic in the case of visualizing rapid growth. It's also advised to remain open to redundancy (that is, repeat data) if the duplication simplifies the eyeball comparison.
Another useful idea was to rely on labeling to defeat ambiguity, going so far as to write explanations on the graphic if necessary. This seems to run counter to another highly emphasized point in the book - maximize the data-to-ink ratio - but the extra investment is often worth the price if it reduces or eliminates ambiguity. As it relates to the question about ink, the best advice may be in the clarification of the above rule - add new ink if it presents new information.
The overall lesson in The Visual Display of Quantitative Information is hard to pin down, but over a number of readings I gathered some basic concepts that came up time and again - simplicity, clarity, and common sense make up the critical foundation for any good visual. Does this rule out the value of knowing how to make an aesthetically impressive visual? There is always a time and place for adorning a graphic, but having the skill is no reason for using it. This works in some ways like knowing how to swim - it's indisputably a valuable ability, but in life it's generally easier to travel by land or air. The key is to have the confidence that when the data calls for complexity, you know how to do it. The rest of the time, it's best to understand the dataset first, then create a simple graphic that harnesses all of the insights mentioned in these reading reviews - show data (not design) variation, enable eyeball comparisons, and add ink to defeat ambiguity. Of course, let's not forget the most important rule of all - know when to use a chart instead.
TOA Rating: Three histograms out of four.
A parting thought - think of a graphic like a paragraph about the data.
