Data Visualization Best Practices

A chart has one job: to help someone understand something about the data faster than they could from a table. Most bad charts fail that job in ordinary ways. The message is buried under a legend with eight colors. The bars start at 40, so a 5% difference looks like a doubling. Two lines that matter are red and green, and one reader in your audience cannot tell them apart. The title says "Revenue by month" when what you wanted people to notice is that revenue fell after the price change. These data visualization best practices are about avoiding those failures.

The guide leans on the UK Government Analysis Function's chart guidance, written for analysts who publish official statistics, and on the Web Content Accessibility Guidelines (WCAG) 2.2, the standard that public sector charts in many countries must meet. Code examples use matplotlib, Plotly and Vega-Lite, because each publishes detailed documentation and between them they cover static charts, interactive charts and declarative specs. The principles apply equally to Excel, Tableau, Power BI or D3.

Work through the chapters in order the first time. The first chapter starts with the question, not the chart type, because a chart chosen before you know your message is usually the wrong chart. The middle chapters cover the details that decide whether people can read it: encoding, color, axes and annotation. The last two cover dashboards and the ways a technically correct chart can still mislead.

Two habits are worth taking from this guide even if you read nothing else. Write your chart's message as one sentence before you draw it, and never let color be the only way to tell two things apart.