The principles of effective visualization are the small set of rules that distinguish a chart that communicates honestly from one that misleads or fails. They’re simple to state and easy to violate.
Choose the appropriate type of plot. If we want to show the gender breakdown of a class — Female 60%, Male 40%, Undeclared 0%, Non-binary 0% — and we draw it as a Line graph, we’ve made a category error. The connecting line implies the categories are ordered along a continuous axis, which they aren’t. A Bar chart or Pie chart is the right tool for categorical data; a Line graph is for continuous-in-time data; a Scatter plot is for showing the relationship between two numerical variables.
Label everything. Axis labels, units, legends for multiple series, titles, annotations marking notable points. A chart of historical Bitcoin price that has axes but no labels — no scale, no units, no time markers — fails as communication. We can guess what the chart is because we recognize it, but a chart that requires the reader to recognize it has already failed.
Keep it simple. Cluttered charts hide their own message. A line chart with eight time-series stacked on top of each other and a legend taking up a third of the figure is hard to read no matter how good the data is. Often the right move is to thin the visualization: show the two or three series that matter, dim or remove the rest, annotate the key moments rather than expecting the reader to find them. Edward Tufte calls the underlying rule the data-ink ratio — the share of the chart’s ink that actually encodes data should be as high as possible. A useful rule from Effective Data Storytelling: instead of Top Products by Monthly Sales with eight overlapping lines, redraw the same data with the focal series in bold color, the rest in light gray, and a caption — Product C’s update drove a $90K sales increase (88% ↑) but Product G’s update has slowed sales (7% ↓).
Use color and other visual elements purposefully. Color is a powerful tool and an easy one to abuse. A bar chart comparing April and May temperatures, with one bar colored brown for wood texture and the other colored solid blue, is using color decoratively rather than meaningfully — and the visual difference between the bars now suggests something other than the actual temperature difference. Color should encode information the reader needs to read off the chart. If two bars are different colors, the reader will look for what that color difference means; if it doesn’t mean anything, we’ve misled them.
Use appropriate scales. Linear scales are right for most data. Logarithmic scales are right for data that spans many orders of magnitude. The chart of world population over the last 12,000 years is a good example: from roughly 4 million people in 10,000 BCE to nearly 8 billion today. On a linear scale, all the action is in the last few hundred years, and the previous nine and a half millennia are flat against the x-axis. On a log scale, the gradual exponential growth is visible across the whole history. The point isn’t that log scales are always right; the point is that axis scales should be chosen to make the structure of the data visible. A bad axis hides the story; a good axis reveals it.
Provide context. A chart is rarely self-sufficient. Annotations naming key events, captions explaining what’s shown, multiple complementary plots showing different aspects of the same data — all of these turn a chart from a record of numbers into a piece of communication.
Be accurate. This is the most-violated principle and the one that matters most. A bar chart comparing our number-of-views (18,250) and a competitor’s (18,258) with a y-axis starting at 18,245 instead of 0 makes a difference of 8 views look like a 2:1 ratio. The chart is technically correct and grossly misleading. The same data on a y-axis starting at 0 — or simply reported as numbers — would correctly suggest they’re virtually identical. Tufte’s lie factor — the size of the effect shown divided by the size of the effect in the data — measures exactly this kind of distortion; honest charts have a lie factor near 1. Distorting axes, omitting data points that don’t fit the narrative, choosing a chart type that exaggerates a small difference: all of these are forms of dishonesty. They erode trust the moment the reader notices them, and the reader does notice.
Get creative when it helps. Visualization sits at the intersection of data science and design. Interactive visualizations let the reader explore the data themselves; striking visual metaphors (Our World in Data’s hourglass of humanity’s past and present, with grains of sand representing 10 million people each) can make a quantitative argument that a Bar chart couldn’t.
The principles work together. A chart that is the right type, well labelled, simple, sensibly colored, accurately scaled, contextualized, and honestly drawn is a chart the reader trusts and learns from. Every principle violated takes a small piece of that trust away. By the time several are violated, the chart has become a piece of advertising rather than a piece of communication.