03 SEEING · How eyes read a picture

Line things up to compare.

You will learn which kinds of charts people read accurately, and how to pick one when the comparison matters. It takes 3 minutes.

Take the test first

Four quick tests. Answer each one in your head before you open the answer. Don't measure. Just look.

EXAMPLE DATA

Test 1. Rank A to E, smallest to largest.

EXAMPLE DATA

Test 2. Same five numbers as areas. Rank them.

EXAMPLE DATA

Test 3. Same five numbers as bars on one baseline. Rank them.

Show the answers to tests 1 to 3

Answer to all three: C (38), A (41), E (44), B (47), D (50).

The numbers are close together on purpose: 38 to 50. The pie and the bubbles made that hard. The bars made it easy. Same data, three encodings.

EXAMPLE DATA

Test 4. B is twice as wide as A. How many times bigger is its area: 2×, 3×, 4×?

Show the answer to test 4

4×. Twice as wide means four times the area. Most people guess low.

Watch the same tests

The clip runs the same tests at video speed. Notice how your confidence changes from one chart to the next.

30 seconds · lesson 3 of 20 from the full video

The name for it

THE NAME FOR IT

Graphical perception

William Cleveland and Robert McGill · 1984

People judge some encodings far more accurately than others. Position on a common scale beats length, angle, area and volume.

In 1984, Cleveland and McGill tested how accurately people judge different encodings. The results form a ladder.

Position on a common scale
Position on unaligned scales
Length, direction, angle
Area
Volume, curvature
Shading, color saturation

More accurate at the top. Less accurate at the bottom.

You just felt the top and the bottom of the ladder. Bars on one baseline are near the top. Areas and angles are near the bottom.

TAKEAWAY

To compare precisely,
line things up
on one shared scale.

Pick the right chart

Ask one question first: does the reader need to compare values precisely?

Yes. Use position on a common scale. That means bars from one baseline, dots on one axis, or lines on one shared axis. Put the things to compare side by side.
No, they need the rough share of a whole. A pie is acceptable with two or three slices. With more, use a sorted bar chart.
Never use area, volume or 3D when the exact difference matters. People guess low on area, every time.

Ask your agent

PROMPT

Pick a chart type for the data below. First, state the one comparison the reader must make.
Then rank the candidate encodings using Cleveland and McGill's graphical perception ladder:
position on a common scale, position on unaligned scales, length, angle, area, volume, color saturation.
Choose the highest-ranked encoding that fits the comparison. Sort the categories by value unless they have a natural order.
Explain the choice in one sentence, then draw the chart.

Data:
[paste here]

What good output looks like: a bar or dot chart on one shared axis, sorted, with a one-sentence reason that names the comparison.

Sources

William S. Cleveland and Robert McGill, "Graphical Perception: Theory, Experimentation, and Application to the Development of Graphical Methods," Journal of the American Statistical Association, 1984.
The five values are invented example data.