How Humans Perceive Visuals

This is one of the most important pages in the course. If you grasp what it says, you will have a theory for why some charts feel effortless to read and others feel like work. Without the theory, chart design is guesswork. With the theory, it becomes principled.

Pre-attentive processing

Your visual system processes some properties of an image before your conscious attention gets involved. These properties are called pre-attentive features. Detecting them takes about 200 milliseconds, and crucially, it is independent of how many items are on screen — you can spot one red dot among a thousand gray dots as fast as you can spot it among ten.

Try this. Below is a paragraph. Find the bold word.

Lorem ipsum dolor sit amet consectetur adipiscing elit sed do eiusmod tempor incididunt ut labore et dolore magna aliqua.

You did it instantly. Your visual system flagged bold for you before you read a single word. Boldness (a form of weight / intensity) is a pre-attentive feature.

The same is true for many other visual properties. Here are the ones most relevant to data visualization:

The Cleveland-McGill ranking

In a famous 1984 study, William Cleveland and Robert McGill asked: which visual properties let people compare quantities most accurately? They ran experiments and produced a ranking — from most accurate to least accurate — for quantitative comparisons:

1. Position along a common scale (most accurate)
2. Position along non-aligned scales
3. Length
4. Angle / slope
5. Area
6. Volume / curvature
7. Color intensity / shading (least accurate)

This ranking has been replicated many times and is the empirical backbone of modern visualization theory.

Several practical conclusions fall out immediately:

• Bar charts beat pie charts for comparing magnitudes — bars use length and aligned position; pies use angles and areas.
• Scatter plots are extremely powerful — they use the top two positions on the ranking for both axes.
• Color is great for categories and terrible for exact magnitudes. The eye can quickly distinguish red from blue, but it cannot reliably tell you whether "this red" is 17% brighter than "that red."
• 3-D charts hurt more than they help — they swap accurate cues (position, length) for inaccurate ones (volume, angle).

See it for yourself

Look at these two charts of the same data. Which one lets you compare values more easily?

The bar chart almost forces you to see that A is biggest, then B, then C — and roughly by how much. The pie chart leaves you squinting at slice sizes that are all "about the same."

Color: powerful and dangerous

Color deserves its own paragraph because it is the most over-used and abused encoding in data visualization.

What color is good for:

• Categories. Red, blue, green for distinct groups. The eye separates them instantly.
• Diverging quantities with a meaningful midpoint. Red for "below average," blue for "above average." A diverging color scale (red→white→blue) makes the midpoint visually obvious.
• Sequential quantities when precise values aren't crucial. Light yellow → deep blue for "increasing." Good for heatmaps and maps.

What color is bad for:

• Precise quantity comparison. "Is this red 23% more than that one?" — you can't tell.
• Rainbow / jet colormaps that are perceptually nonuniform. These create false visual ridges that look like data structure but aren't. The world has largely moved on from them.
• More than ~7 categorical colors. After ~7, the eye starts confusing similar shades.

Accessibility: not everyone sees color the same way

About 8% of men and 0.5% of women have some form of color vision deficiency. The most common is red-green confusion. If your chart distinguishes "good" from "bad" only by red vs green, a significant chunk of your audience cannot read it.

The fixes are simple and free:

• Use colorblind-safe palettes (Plotly's Viridis, Cividis, and Plotly3 qualitative palettes are all good defaults).
• Double-encode important distinctions — different colors and different shapes, or different colors and labels.
• For sequential data, prefer luminance-varying palettes (light → dark), which work even in grayscale.

We will revisit this in the ethics and accessibility chapter.

A practical rule of thumb

Here is the heuristic that summarizes this whole chapter:

Use the highest-ranked encoding (position, length) for the most important variable. Use lower-ranked encodings (color, size) for secondary variables that you mostly want the reader to notice categorically, not compare precisely.

In a Plotly Express scatter plot like px.scatter(df, x="gdp", y="lifeExp", color="continent", size="pop"):

• The most accurate comparisons are along x and y.
• Color lets you spot which dots belong to the same continent.
• Size lets you notice "this is a populous country" but not reliably compare which is bigger.

That ordering matches the variables' importance for the story.

Check your understanding

QuestionSelect one

What does it mean for a visual property to be pre-attentive?

It can only be seen with conscious effort.

It requires special training to notice.

The visual system detects it before conscious attention is engaged, in roughly 200 milliseconds, regardless of how many other items are on screen.

It must be in the foreground of the chart.

QuestionSelect one

According to the Cleveland-McGill ranking, which visual encoding is most accurate for comparing quantitative values?

Color intensity (shading).

Area.

Angle.

Position along a common scale.

QuestionSelect one

Why are pie charts generally weaker than bar charts for comparing values?

Pie charts are an outdated format.

Pie charts can only show two values.

Pie charts encode quantity using angle / area, which the visual system judges much less accurately than the length / position used by bar charts.

Pie charts don't support color.

QuestionSelect one

Which is a good use of color in a chart?

Showing precise dollar amounts using shades of red.

Distinguishing 25 different product categories.

Showing a categorical grouping (e.g., continent) with 5-7 distinct hues.

Indicating the chart's title.

QuestionSelect one

Why does using only red vs green to encode "bad" vs "good" create an accessibility problem?

Red and green look identical on phone screens.

Red and green are banned by the WCAG specification.

About 8% of men have red-green color vision deficiency and cannot reliably distinguish the two colors.

Red and green are culturally inappropriate.

QuestionSelect one

A chart shows size="population" and color="continent" in a scatter plot of x="gdp", y="lifeExp". Which variables can the viewer compare most precisely?

Population (via size).

Continent (via color).

GDP and life expectancy (via x and y position).

All variables are equally precise to compare.