← Reads
Reads · Charts

Good Charts

Scott Berinato · HBR Press 2016 · 4 min

Most books on data visualization treat charts as illustrations — decorations bolted onto numbers to make them readable. Berinato's thesis is that charts are arguments. Every chart is making a case. The case might be honest or dishonest, sharp or muddy, persuasive or boring, but the chart is always doing rhetorical work, and the people who make charts well are people who understand they're writing in a visual grammar that has its own rules of emphasis, omission, and tone.

The most useful single thing in the book is a 2×2 typology that Berinato uses to sort what kind of chart you are actually trying to make. The two axes are: conceptual versus data-driven, and declarative versus exploratory. A conceptual + declarative chart is a slide explaining how a process works. A conceptual + exploratory chart is a whiteboard sketch of an idea you don't yet understand. A data-driven + declarative chart is the one with bars and a headline; this is what most people mean when they say "chart". A data-driven + exploratory chart is the analyst's notebook — many of them, ugly, fast, in service of finding a thing.

Knowing which quadrant you're in changes what tool you pick, how much polish you owe the work, and who the audience is. The mistake Berinato sees most often is people in the data-driven + declarative quadrant making charts that look like they came from the data-driven + exploratory quadrant — twelve series on one axis, no headline, no annotation, and a request that the reader figure out what they were supposed to see. A chart that doesn't tell you what it's saying is a chart that hasn't decided what it's saying.

Berinato is an editor at the Harvard Business Review. The book is written for managers who have to make charts but don't think of themselves as designers, and that posture is what makes it a useful read for engineers and operators too. He's not trying to teach you how to recreate a New York Times graphic. He's trying to teach you how to know what you're saying before you reach for matplotlib.

For PointCast the lesson is the same one McLuhan was teaching: the form is doing rhetorical work whether or not you noticed it. A small, declarative chart on a wire post is making an argument. So is a coffee mug count. So is a daily race chip. The work is to know which argument and to mean it.