Visual Encodings — What the Eye Ranks Well

P14.chart-craft.01 · Audience: guest, it-ml, language-pro

Production Case — TinyBERT, Packaged to Monitored Choosing the Chart

Two teams present the same fact — sales are three times higher in the north than in the south. One draws two bars; the reader sees "about three times" at a glance. The other draws two circles, sizing each by its value, and the reader sees "maybe twice". Same number, same honest intent, two pictures that land differently — because a bar and a circle are not two neutral ways of showing a quantity. They are two different encodings, and the human eye reads some encodings far more accurately than others. This pillar opens here, one level below the gallery of chart types, with the question every later choice depends on: which visual channels does the eye actually read well, and in what order? Get the ranking right and the rest of chart craft becomes reasoning instead of taste.

🗺️ So far you have built: the craft to get to a correct result — to analyse data, run a model, compute a number that means something (the analysis pillars, in whatever order you met them). Those tools answer 'what is true?'. This pillar adds the last mile they never cover: turning a true result into a picture an audience reads correctly. It needs no prior design background — only the willingness to notice how your own eye reads a graphic.

In this module you will: the foundation of every chart choice: the perception ranking. You will see that a chart maps data onto visual channels — position, length, angle, area, colour — and that these channels are not interchangeable, because the eye reads some almost exactly and others only vaguely. On the same two numbers drawn two ways you will watch an area encoding under-read by nearly a third, and learn the ordering (position, then length, then angle, then area, then colour) that decides which channel to reach for.

Step 1 / 4 — A chart is a mapping, and the eye is not a ruler

Every chart does the same basic thing: it takes a number and turns it into something you can see — a position along a line, the length of a bar, the size of a dot, a shade of colour. That translation is called an encoding. The data is the same whichever encoding you pick; what changes is how accurately a human reader can turn the picture back into the number. And here is the fact the whole pillar rests on: the eye is a superb judge of some encodings and a poor judge of others.

We tend to assume any reasonable-looking chart conveys its numbers faithfully. It does not. Ask someone to compare two bar lengths and they are accurate to a few percent; ask them to compare two circle areas and they are off by a quarter or more, consistently and in a predictable direction. The chart did not lie — the encoding did the reader's perception no favours. Choosing an encoding is therefore not decoration; it decides how much of your correct result survives the trip into the reader's head.

ⓘ Concept: The encoding, not the data, sets how accurately a chart is read

data value → visual channel (position | length | angle | area | colour) → reader decodes with a fidelity that depends on the channel, not the number

Why it matters — This is the reframing the rest of chart craft is built on. Beginners choose a chart by what looks good or what is conventional; the discipline is to choose by which visual channel the eye reads most accurately for the comparison the reader must make. The same value shown as a position, a length, an angle, an area or a colour arrives in the reader's mind with wildly different fidelity, so the encoding is a design decision with a measurable cost. Once you know the channels are ranked, 'which chart?' stops being a matter of preference and becomes a matter of matching the most accurate channel to the most important comparison.

✅ Key takeaways

A chart is an encoding: it maps a number onto a visual channel — position, length, angle, area or colour — and the channels are not interchangeable, because the eye reads some almost exactly and others only vaguely.
The Cleveland–McGill perception ranking, most to least accurate: position on a common scale, then length, then angle/slope, then area, then colour shade. Reach for the highest channel your comparison allows.
Area is read badly for a measurable reason — Stevens' power law says perceived area grows as (true area)^0.7, so big differences are systematically under-read. A true 3x drawn as circle area is felt as only about 2.16x, a ~28% under-reading.
The rule is a decision, not a style: put the most important comparison on position or length; spend the lower channels (area, colour) only when the data's structure forces it, knowing the fidelity you give up.
Every distortion in the later misleading-charts workshop is this ranking abused — a modest change blown up by encoding it as area, a real gap hidden the same way. Knowing the ranking is knowing both how to be honest and how to spot dishonesty.

Next: P14.chart-craft.02 — Choosing the Chart

Ask the mentor about this module

Ask a question about this content. The mentor explains and grounds its answer in what you are studying; asking is recorded as a learning signal, not a grade.

Ctrl/Cmd + Enter to send

Try it yourself

A scratch console for this page's ideas — ungraded, nothing you run here is recorded.

Scratch console

A scratch console with the scientific stack (pandas, numpy, scikit-learn). Runs on the server — no network, resource-limited and measured.

Output appears here.

Where next?

Continue

Choosing the Chart →

Chart Craft · P14

Later in Chart Craft

This module unlocks

Go up a level